2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE
= 0,
51 CHUNK_ALLOC_FORCE
= 1,
52 CHUNK_ALLOC_LIMITED
= 2,
55 static int update_block_group(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 bytenr
, u64 num_bytes
, int alloc
);
58 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
59 struct btrfs_root
*root
,
60 u64 bytenr
, u64 num_bytes
, u64 parent
,
61 u64 root_objectid
, u64 owner_objectid
,
62 u64 owner_offset
, int refs_to_drop
,
63 struct btrfs_delayed_extent_op
*extra_op
);
64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
65 struct extent_buffer
*leaf
,
66 struct btrfs_extent_item
*ei
);
67 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
68 struct btrfs_root
*root
,
69 u64 parent
, u64 root_objectid
,
70 u64 flags
, u64 owner
, u64 offset
,
71 struct btrfs_key
*ins
, int ref_mod
);
72 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
73 struct btrfs_root
*root
,
74 u64 parent
, u64 root_objectid
,
75 u64 flags
, struct btrfs_disk_key
*key
,
76 int level
, struct btrfs_key
*ins
);
77 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
79 u64 flags
, int force
);
80 static int find_next_key(struct btrfs_path
*path
, int level
,
81 struct btrfs_key
*key
);
82 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
83 int dump_block_groups
);
86 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
89 return cache
->cached
== BTRFS_CACHE_FINISHED
;
92 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
94 return (cache
->flags
& bits
) == bits
;
97 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
99 atomic_inc(&cache
->count
);
102 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
104 if (atomic_dec_and_test(&cache
->count
)) {
105 WARN_ON(cache
->pinned
> 0);
106 WARN_ON(cache
->reserved
> 0);
107 WARN_ON(cache
->reserved_pinned
> 0);
113 * this adds the block group to the fs_info rb tree for the block group
116 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
117 struct btrfs_block_group_cache
*block_group
)
120 struct rb_node
*parent
= NULL
;
121 struct btrfs_block_group_cache
*cache
;
123 spin_lock(&info
->block_group_cache_lock
);
124 p
= &info
->block_group_cache_tree
.rb_node
;
128 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
130 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
132 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
135 spin_unlock(&info
->block_group_cache_lock
);
140 rb_link_node(&block_group
->cache_node
, parent
, p
);
141 rb_insert_color(&block_group
->cache_node
,
142 &info
->block_group_cache_tree
);
143 spin_unlock(&info
->block_group_cache_lock
);
149 * This will return the block group at or after bytenr if contains is 0, else
150 * it will return the block group that contains the bytenr
152 static struct btrfs_block_group_cache
*
153 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
156 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
160 spin_lock(&info
->block_group_cache_lock
);
161 n
= info
->block_group_cache_tree
.rb_node
;
164 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
166 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
167 start
= cache
->key
.objectid
;
169 if (bytenr
< start
) {
170 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
173 } else if (bytenr
> start
) {
174 if (contains
&& bytenr
<= end
) {
185 btrfs_get_block_group(ret
);
186 spin_unlock(&info
->block_group_cache_lock
);
191 static int add_excluded_extent(struct btrfs_root
*root
,
192 u64 start
, u64 num_bytes
)
194 u64 end
= start
+ num_bytes
- 1;
195 set_extent_bits(&root
->fs_info
->freed_extents
[0],
196 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
197 set_extent_bits(&root
->fs_info
->freed_extents
[1],
198 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
202 static void free_excluded_extents(struct btrfs_root
*root
,
203 struct btrfs_block_group_cache
*cache
)
207 start
= cache
->key
.objectid
;
208 end
= start
+ cache
->key
.offset
- 1;
210 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
211 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
212 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
213 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
216 static int exclude_super_stripes(struct btrfs_root
*root
,
217 struct btrfs_block_group_cache
*cache
)
224 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
225 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
226 cache
->bytes_super
+= stripe_len
;
227 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
232 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
233 bytenr
= btrfs_sb_offset(i
);
234 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
235 cache
->key
.objectid
, bytenr
,
236 0, &logical
, &nr
, &stripe_len
);
240 cache
->bytes_super
+= stripe_len
;
241 ret
= add_excluded_extent(root
, logical
[nr
],
251 static struct btrfs_caching_control
*
252 get_caching_control(struct btrfs_block_group_cache
*cache
)
254 struct btrfs_caching_control
*ctl
;
256 spin_lock(&cache
->lock
);
257 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
258 spin_unlock(&cache
->lock
);
262 /* We're loading it the fast way, so we don't have a caching_ctl. */
263 if (!cache
->caching_ctl
) {
264 spin_unlock(&cache
->lock
);
268 ctl
= cache
->caching_ctl
;
269 atomic_inc(&ctl
->count
);
270 spin_unlock(&cache
->lock
);
274 static void put_caching_control(struct btrfs_caching_control
*ctl
)
276 if (atomic_dec_and_test(&ctl
->count
))
281 * this is only called by cache_block_group, since we could have freed extents
282 * we need to check the pinned_extents for any extents that can't be used yet
283 * since their free space will be released as soon as the transaction commits.
285 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
286 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
288 u64 extent_start
, extent_end
, size
, total_added
= 0;
291 while (start
< end
) {
292 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
293 &extent_start
, &extent_end
,
294 EXTENT_DIRTY
| EXTENT_UPTODATE
);
298 if (extent_start
<= start
) {
299 start
= extent_end
+ 1;
300 } else if (extent_start
> start
&& extent_start
< end
) {
301 size
= extent_start
- start
;
303 ret
= btrfs_add_free_space(block_group
, start
,
306 start
= extent_end
+ 1;
315 ret
= btrfs_add_free_space(block_group
, start
, size
);
322 static int caching_kthread(void *data
)
324 struct btrfs_block_group_cache
*block_group
= data
;
325 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
326 struct btrfs_caching_control
*caching_ctl
= block_group
->caching_ctl
;
327 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
328 struct btrfs_path
*path
;
329 struct extent_buffer
*leaf
;
330 struct btrfs_key key
;
336 path
= btrfs_alloc_path();
340 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
343 * We don't want to deadlock with somebody trying to allocate a new
344 * extent for the extent root while also trying to search the extent
345 * root to add free space. So we skip locking and search the commit
346 * root, since its read-only
348 path
->skip_locking
= 1;
349 path
->search_commit_root
= 1;
354 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
356 mutex_lock(&caching_ctl
->mutex
);
357 /* need to make sure the commit_root doesn't disappear */
358 down_read(&fs_info
->extent_commit_sem
);
360 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
364 leaf
= path
->nodes
[0];
365 nritems
= btrfs_header_nritems(leaf
);
369 if (fs_info
->closing
> 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 caching_ctl
->progress
= last
;
382 btrfs_release_path(extent_root
, path
);
383 up_read(&fs_info
->extent_commit_sem
);
384 mutex_unlock(&caching_ctl
->mutex
);
385 if (btrfs_transaction_in_commit(fs_info
))
392 if (key
.objectid
< block_group
->key
.objectid
) {
397 if (key
.objectid
>= block_group
->key
.objectid
+
398 block_group
->key
.offset
)
401 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
402 total_found
+= add_new_free_space(block_group
,
405 last
= key
.objectid
+ key
.offset
;
407 if (total_found
> (1024 * 1024 * 2)) {
409 wake_up(&caching_ctl
->wait
);
416 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
417 block_group
->key
.objectid
+
418 block_group
->key
.offset
);
419 caching_ctl
->progress
= (u64
)-1;
421 spin_lock(&block_group
->lock
);
422 block_group
->caching_ctl
= NULL
;
423 block_group
->cached
= BTRFS_CACHE_FINISHED
;
424 spin_unlock(&block_group
->lock
);
427 btrfs_free_path(path
);
428 up_read(&fs_info
->extent_commit_sem
);
430 free_excluded_extents(extent_root
, block_group
);
432 mutex_unlock(&caching_ctl
->mutex
);
433 wake_up(&caching_ctl
->wait
);
435 put_caching_control(caching_ctl
);
436 atomic_dec(&block_group
->space_info
->caching_threads
);
437 btrfs_put_block_group(block_group
);
442 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
443 struct btrfs_trans_handle
*trans
,
444 struct btrfs_root
*root
,
447 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
448 struct btrfs_caching_control
*caching_ctl
;
449 struct task_struct
*tsk
;
453 if (cache
->cached
!= BTRFS_CACHE_NO
)
457 * We can't do the read from on-disk cache during a commit since we need
458 * to have the normal tree locking. Also if we are currently trying to
459 * allocate blocks for the tree root we can't do the fast caching since
460 * we likely hold important locks.
462 if (trans
&& (!trans
->transaction
->in_commit
) &&
463 (root
&& root
!= root
->fs_info
->tree_root
)) {
464 spin_lock(&cache
->lock
);
465 if (cache
->cached
!= BTRFS_CACHE_NO
) {
466 spin_unlock(&cache
->lock
);
469 cache
->cached
= BTRFS_CACHE_STARTED
;
470 spin_unlock(&cache
->lock
);
472 ret
= load_free_space_cache(fs_info
, cache
);
474 spin_lock(&cache
->lock
);
476 cache
->cached
= BTRFS_CACHE_FINISHED
;
477 cache
->last_byte_to_unpin
= (u64
)-1;
479 cache
->cached
= BTRFS_CACHE_NO
;
481 spin_unlock(&cache
->lock
);
483 free_excluded_extents(fs_info
->extent_root
, cache
);
491 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
492 BUG_ON(!caching_ctl
);
494 INIT_LIST_HEAD(&caching_ctl
->list
);
495 mutex_init(&caching_ctl
->mutex
);
496 init_waitqueue_head(&caching_ctl
->wait
);
497 caching_ctl
->block_group
= cache
;
498 caching_ctl
->progress
= cache
->key
.objectid
;
499 /* one for caching kthread, one for caching block group list */
500 atomic_set(&caching_ctl
->count
, 2);
502 spin_lock(&cache
->lock
);
503 if (cache
->cached
!= BTRFS_CACHE_NO
) {
504 spin_unlock(&cache
->lock
);
508 cache
->caching_ctl
= caching_ctl
;
509 cache
->cached
= BTRFS_CACHE_STARTED
;
510 spin_unlock(&cache
->lock
);
512 down_write(&fs_info
->extent_commit_sem
);
513 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
514 up_write(&fs_info
->extent_commit_sem
);
516 atomic_inc(&cache
->space_info
->caching_threads
);
517 btrfs_get_block_group(cache
);
519 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
520 cache
->key
.objectid
);
523 printk(KERN_ERR
"error running thread %d\n", ret
);
531 * return the block group that starts at or after bytenr
533 static struct btrfs_block_group_cache
*
534 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
536 struct btrfs_block_group_cache
*cache
;
538 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
544 * return the block group that contains the given bytenr
546 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
547 struct btrfs_fs_info
*info
,
550 struct btrfs_block_group_cache
*cache
;
552 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
557 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
560 struct list_head
*head
= &info
->space_info
;
561 struct btrfs_space_info
*found
;
563 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
564 BTRFS_BLOCK_GROUP_METADATA
;
567 list_for_each_entry_rcu(found
, head
, list
) {
568 if (found
->flags
& flags
) {
578 * after adding space to the filesystem, we need to clear the full flags
579 * on all the space infos.
581 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
583 struct list_head
*head
= &info
->space_info
;
584 struct btrfs_space_info
*found
;
587 list_for_each_entry_rcu(found
, head
, list
)
592 static u64
div_factor(u64 num
, int factor
)
601 static u64
div_factor_fine(u64 num
, int factor
)
610 u64
btrfs_find_block_group(struct btrfs_root
*root
,
611 u64 search_start
, u64 search_hint
, int owner
)
613 struct btrfs_block_group_cache
*cache
;
615 u64 last
= max(search_hint
, search_start
);
622 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
626 spin_lock(&cache
->lock
);
627 last
= cache
->key
.objectid
+ cache
->key
.offset
;
628 used
= btrfs_block_group_used(&cache
->item
);
630 if ((full_search
|| !cache
->ro
) &&
631 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
632 if (used
+ cache
->pinned
+ cache
->reserved
<
633 div_factor(cache
->key
.offset
, factor
)) {
634 group_start
= cache
->key
.objectid
;
635 spin_unlock(&cache
->lock
);
636 btrfs_put_block_group(cache
);
640 spin_unlock(&cache
->lock
);
641 btrfs_put_block_group(cache
);
649 if (!full_search
&& factor
< 10) {
659 /* simple helper to search for an existing extent at a given offset */
660 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
663 struct btrfs_key key
;
664 struct btrfs_path
*path
;
666 path
= btrfs_alloc_path();
668 key
.objectid
= start
;
670 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
671 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
673 btrfs_free_path(path
);
678 * helper function to lookup reference count and flags of extent.
680 * the head node for delayed ref is used to store the sum of all the
681 * reference count modifications queued up in the rbtree. the head
682 * node may also store the extent flags to set. This way you can check
683 * to see what the reference count and extent flags would be if all of
684 * the delayed refs are not processed.
686 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
687 struct btrfs_root
*root
, u64 bytenr
,
688 u64 num_bytes
, u64
*refs
, u64
*flags
)
690 struct btrfs_delayed_ref_head
*head
;
691 struct btrfs_delayed_ref_root
*delayed_refs
;
692 struct btrfs_path
*path
;
693 struct btrfs_extent_item
*ei
;
694 struct extent_buffer
*leaf
;
695 struct btrfs_key key
;
701 path
= btrfs_alloc_path();
705 key
.objectid
= bytenr
;
706 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
707 key
.offset
= num_bytes
;
709 path
->skip_locking
= 1;
710 path
->search_commit_root
= 1;
713 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
719 leaf
= path
->nodes
[0];
720 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
721 if (item_size
>= sizeof(*ei
)) {
722 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
723 struct btrfs_extent_item
);
724 num_refs
= btrfs_extent_refs(leaf
, ei
);
725 extent_flags
= btrfs_extent_flags(leaf
, ei
);
727 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
728 struct btrfs_extent_item_v0
*ei0
;
729 BUG_ON(item_size
!= sizeof(*ei0
));
730 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
731 struct btrfs_extent_item_v0
);
732 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
733 /* FIXME: this isn't correct for data */
734 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
739 BUG_ON(num_refs
== 0);
749 delayed_refs
= &trans
->transaction
->delayed_refs
;
750 spin_lock(&delayed_refs
->lock
);
751 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
753 if (!mutex_trylock(&head
->mutex
)) {
754 atomic_inc(&head
->node
.refs
);
755 spin_unlock(&delayed_refs
->lock
);
757 btrfs_release_path(root
->fs_info
->extent_root
, path
);
759 mutex_lock(&head
->mutex
);
760 mutex_unlock(&head
->mutex
);
761 btrfs_put_delayed_ref(&head
->node
);
764 if (head
->extent_op
&& head
->extent_op
->update_flags
)
765 extent_flags
|= head
->extent_op
->flags_to_set
;
767 BUG_ON(num_refs
== 0);
769 num_refs
+= head
->node
.ref_mod
;
770 mutex_unlock(&head
->mutex
);
772 spin_unlock(&delayed_refs
->lock
);
774 WARN_ON(num_refs
== 0);
778 *flags
= extent_flags
;
780 btrfs_free_path(path
);
785 * Back reference rules. Back refs have three main goals:
787 * 1) differentiate between all holders of references to an extent so that
788 * when a reference is dropped we can make sure it was a valid reference
789 * before freeing the extent.
791 * 2) Provide enough information to quickly find the holders of an extent
792 * if we notice a given block is corrupted or bad.
794 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
795 * maintenance. This is actually the same as #2, but with a slightly
796 * different use case.
798 * There are two kinds of back refs. The implicit back refs is optimized
799 * for pointers in non-shared tree blocks. For a given pointer in a block,
800 * back refs of this kind provide information about the block's owner tree
801 * and the pointer's key. These information allow us to find the block by
802 * b-tree searching. The full back refs is for pointers in tree blocks not
803 * referenced by their owner trees. The location of tree block is recorded
804 * in the back refs. Actually the full back refs is generic, and can be
805 * used in all cases the implicit back refs is used. The major shortcoming
806 * of the full back refs is its overhead. Every time a tree block gets
807 * COWed, we have to update back refs entry for all pointers in it.
809 * For a newly allocated tree block, we use implicit back refs for
810 * pointers in it. This means most tree related operations only involve
811 * implicit back refs. For a tree block created in old transaction, the
812 * only way to drop a reference to it is COW it. So we can detect the
813 * event that tree block loses its owner tree's reference and do the
814 * back refs conversion.
816 * When a tree block is COW'd through a tree, there are four cases:
818 * The reference count of the block is one and the tree is the block's
819 * owner tree. Nothing to do in this case.
821 * The reference count of the block is one and the tree is not the
822 * block's owner tree. In this case, full back refs is used for pointers
823 * in the block. Remove these full back refs, add implicit back refs for
824 * every pointers in the new block.
826 * The reference count of the block is greater than one and the tree is
827 * the block's owner tree. In this case, implicit back refs is used for
828 * pointers in the block. Add full back refs for every pointers in the
829 * block, increase lower level extents' reference counts. The original
830 * implicit back refs are entailed to the new block.
832 * The reference count of the block is greater than one and the tree is
833 * not the block's owner tree. Add implicit back refs for every pointer in
834 * the new block, increase lower level extents' reference count.
836 * Back Reference Key composing:
838 * The key objectid corresponds to the first byte in the extent,
839 * The key type is used to differentiate between types of back refs.
840 * There are different meanings of the key offset for different types
843 * File extents can be referenced by:
845 * - multiple snapshots, subvolumes, or different generations in one subvol
846 * - different files inside a single subvolume
847 * - different offsets inside a file (bookend extents in file.c)
849 * The extent ref structure for the implicit back refs has fields for:
851 * - Objectid of the subvolume root
852 * - objectid of the file holding the reference
853 * - original offset in the file
854 * - how many bookend extents
856 * The key offset for the implicit back refs is hash of the first
859 * The extent ref structure for the full back refs has field for:
861 * - number of pointers in the tree leaf
863 * The key offset for the implicit back refs is the first byte of
866 * When a file extent is allocated, The implicit back refs is used.
867 * the fields are filled in:
869 * (root_key.objectid, inode objectid, offset in file, 1)
871 * When a file extent is removed file truncation, we find the
872 * corresponding implicit back refs and check the following fields:
874 * (btrfs_header_owner(leaf), inode objectid, offset in file)
876 * Btree extents can be referenced by:
878 * - Different subvolumes
880 * Both the implicit back refs and the full back refs for tree blocks
881 * only consist of key. The key offset for the implicit back refs is
882 * objectid of block's owner tree. The key offset for the full back refs
883 * is the first byte of parent block.
885 * When implicit back refs is used, information about the lowest key and
886 * level of the tree block are required. These information are stored in
887 * tree block info structure.
890 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
891 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
892 struct btrfs_root
*root
,
893 struct btrfs_path
*path
,
894 u64 owner
, u32 extra_size
)
896 struct btrfs_extent_item
*item
;
897 struct btrfs_extent_item_v0
*ei0
;
898 struct btrfs_extent_ref_v0
*ref0
;
899 struct btrfs_tree_block_info
*bi
;
900 struct extent_buffer
*leaf
;
901 struct btrfs_key key
;
902 struct btrfs_key found_key
;
903 u32 new_size
= sizeof(*item
);
907 leaf
= path
->nodes
[0];
908 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
910 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
911 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
912 struct btrfs_extent_item_v0
);
913 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
915 if (owner
== (u64
)-1) {
917 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
918 ret
= btrfs_next_leaf(root
, path
);
922 leaf
= path
->nodes
[0];
924 btrfs_item_key_to_cpu(leaf
, &found_key
,
926 BUG_ON(key
.objectid
!= found_key
.objectid
);
927 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
931 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
932 struct btrfs_extent_ref_v0
);
933 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
937 btrfs_release_path(root
, path
);
939 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
940 new_size
+= sizeof(*bi
);
942 new_size
-= sizeof(*ei0
);
943 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
944 new_size
+ extra_size
, 1);
949 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
952 leaf
= path
->nodes
[0];
953 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
954 btrfs_set_extent_refs(leaf
, item
, refs
);
955 /* FIXME: get real generation */
956 btrfs_set_extent_generation(leaf
, item
, 0);
957 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
958 btrfs_set_extent_flags(leaf
, item
,
959 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
960 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
961 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
962 /* FIXME: get first key of the block */
963 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
964 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
966 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
968 btrfs_mark_buffer_dirty(leaf
);
973 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
975 u32 high_crc
= ~(u32
)0;
976 u32 low_crc
= ~(u32
)0;
979 lenum
= cpu_to_le64(root_objectid
);
980 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
981 lenum
= cpu_to_le64(owner
);
982 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
983 lenum
= cpu_to_le64(offset
);
984 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
986 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
989 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
990 struct btrfs_extent_data_ref
*ref
)
992 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
993 btrfs_extent_data_ref_objectid(leaf
, ref
),
994 btrfs_extent_data_ref_offset(leaf
, ref
));
997 static int match_extent_data_ref(struct extent_buffer
*leaf
,
998 struct btrfs_extent_data_ref
*ref
,
999 u64 root_objectid
, u64 owner
, u64 offset
)
1001 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1002 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1003 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1008 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1009 struct btrfs_root
*root
,
1010 struct btrfs_path
*path
,
1011 u64 bytenr
, u64 parent
,
1013 u64 owner
, u64 offset
)
1015 struct btrfs_key key
;
1016 struct btrfs_extent_data_ref
*ref
;
1017 struct extent_buffer
*leaf
;
1023 key
.objectid
= bytenr
;
1025 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1026 key
.offset
= parent
;
1028 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1029 key
.offset
= hash_extent_data_ref(root_objectid
,
1034 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1043 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1044 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1045 btrfs_release_path(root
, path
);
1046 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1057 leaf
= path
->nodes
[0];
1058 nritems
= btrfs_header_nritems(leaf
);
1060 if (path
->slots
[0] >= nritems
) {
1061 ret
= btrfs_next_leaf(root
, path
);
1067 leaf
= path
->nodes
[0];
1068 nritems
= btrfs_header_nritems(leaf
);
1072 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1073 if (key
.objectid
!= bytenr
||
1074 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1077 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1078 struct btrfs_extent_data_ref
);
1080 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1083 btrfs_release_path(root
, path
);
1095 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1096 struct btrfs_root
*root
,
1097 struct btrfs_path
*path
,
1098 u64 bytenr
, u64 parent
,
1099 u64 root_objectid
, u64 owner
,
1100 u64 offset
, int refs_to_add
)
1102 struct btrfs_key key
;
1103 struct extent_buffer
*leaf
;
1108 key
.objectid
= bytenr
;
1110 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1111 key
.offset
= parent
;
1112 size
= sizeof(struct btrfs_shared_data_ref
);
1114 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1115 key
.offset
= hash_extent_data_ref(root_objectid
,
1117 size
= sizeof(struct btrfs_extent_data_ref
);
1120 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1121 if (ret
&& ret
!= -EEXIST
)
1124 leaf
= path
->nodes
[0];
1126 struct btrfs_shared_data_ref
*ref
;
1127 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1128 struct btrfs_shared_data_ref
);
1130 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1132 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1133 num_refs
+= refs_to_add
;
1134 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1137 struct btrfs_extent_data_ref
*ref
;
1138 while (ret
== -EEXIST
) {
1139 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1140 struct btrfs_extent_data_ref
);
1141 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1144 btrfs_release_path(root
, path
);
1146 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1148 if (ret
&& ret
!= -EEXIST
)
1151 leaf
= path
->nodes
[0];
1153 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1154 struct btrfs_extent_data_ref
);
1156 btrfs_set_extent_data_ref_root(leaf
, ref
,
1158 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1159 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1160 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1162 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1163 num_refs
+= refs_to_add
;
1164 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1167 btrfs_mark_buffer_dirty(leaf
);
1170 btrfs_release_path(root
, path
);
1174 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1175 struct btrfs_root
*root
,
1176 struct btrfs_path
*path
,
1179 struct btrfs_key key
;
1180 struct btrfs_extent_data_ref
*ref1
= NULL
;
1181 struct btrfs_shared_data_ref
*ref2
= NULL
;
1182 struct extent_buffer
*leaf
;
1186 leaf
= path
->nodes
[0];
1187 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1189 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1190 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1191 struct btrfs_extent_data_ref
);
1192 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1193 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1194 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1195 struct btrfs_shared_data_ref
);
1196 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1197 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1198 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1199 struct btrfs_extent_ref_v0
*ref0
;
1200 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1201 struct btrfs_extent_ref_v0
);
1202 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1208 BUG_ON(num_refs
< refs_to_drop
);
1209 num_refs
-= refs_to_drop
;
1211 if (num_refs
== 0) {
1212 ret
= btrfs_del_item(trans
, root
, path
);
1214 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1215 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1216 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1217 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1218 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1220 struct btrfs_extent_ref_v0
*ref0
;
1221 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1222 struct btrfs_extent_ref_v0
);
1223 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1226 btrfs_mark_buffer_dirty(leaf
);
1231 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1232 struct btrfs_path
*path
,
1233 struct btrfs_extent_inline_ref
*iref
)
1235 struct btrfs_key key
;
1236 struct extent_buffer
*leaf
;
1237 struct btrfs_extent_data_ref
*ref1
;
1238 struct btrfs_shared_data_ref
*ref2
;
1241 leaf
= path
->nodes
[0];
1242 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1244 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1245 BTRFS_EXTENT_DATA_REF_KEY
) {
1246 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1247 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1249 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1250 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1252 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1253 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1254 struct btrfs_extent_data_ref
);
1255 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1256 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1257 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1258 struct btrfs_shared_data_ref
);
1259 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1262 struct btrfs_extent_ref_v0
*ref0
;
1263 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1264 struct btrfs_extent_ref_v0
);
1265 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1273 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1274 struct btrfs_root
*root
,
1275 struct btrfs_path
*path
,
1276 u64 bytenr
, u64 parent
,
1279 struct btrfs_key key
;
1282 key
.objectid
= bytenr
;
1284 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1285 key
.offset
= parent
;
1287 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1288 key
.offset
= root_objectid
;
1291 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295 if (ret
== -ENOENT
&& parent
) {
1296 btrfs_release_path(root
, path
);
1297 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1298 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1306 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1307 struct btrfs_root
*root
,
1308 struct btrfs_path
*path
,
1309 u64 bytenr
, u64 parent
,
1312 struct btrfs_key key
;
1315 key
.objectid
= bytenr
;
1317 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1318 key
.offset
= parent
;
1320 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1321 key
.offset
= root_objectid
;
1324 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1325 btrfs_release_path(root
, path
);
1329 static inline int extent_ref_type(u64 parent
, u64 owner
)
1332 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1334 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1336 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1339 type
= BTRFS_SHARED_DATA_REF_KEY
;
1341 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1346 static int find_next_key(struct btrfs_path
*path
, int level
,
1347 struct btrfs_key
*key
)
1350 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1351 if (!path
->nodes
[level
])
1353 if (path
->slots
[level
] + 1 >=
1354 btrfs_header_nritems(path
->nodes
[level
]))
1357 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1358 path
->slots
[level
] + 1);
1360 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1361 path
->slots
[level
] + 1);
1368 * look for inline back ref. if back ref is found, *ref_ret is set
1369 * to the address of inline back ref, and 0 is returned.
1371 * if back ref isn't found, *ref_ret is set to the address where it
1372 * should be inserted, and -ENOENT is returned.
1374 * if insert is true and there are too many inline back refs, the path
1375 * points to the extent item, and -EAGAIN is returned.
1377 * NOTE: inline back refs are ordered in the same way that back ref
1378 * items in the tree are ordered.
1380 static noinline_for_stack
1381 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1382 struct btrfs_root
*root
,
1383 struct btrfs_path
*path
,
1384 struct btrfs_extent_inline_ref
**ref_ret
,
1385 u64 bytenr
, u64 num_bytes
,
1386 u64 parent
, u64 root_objectid
,
1387 u64 owner
, u64 offset
, int insert
)
1389 struct btrfs_key key
;
1390 struct extent_buffer
*leaf
;
1391 struct btrfs_extent_item
*ei
;
1392 struct btrfs_extent_inline_ref
*iref
;
1403 key
.objectid
= bytenr
;
1404 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1405 key
.offset
= num_bytes
;
1407 want
= extent_ref_type(parent
, owner
);
1409 extra_size
= btrfs_extent_inline_ref_size(want
);
1410 path
->keep_locks
= 1;
1413 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1420 leaf
= path
->nodes
[0];
1421 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1422 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1423 if (item_size
< sizeof(*ei
)) {
1428 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1434 leaf
= path
->nodes
[0];
1435 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1438 BUG_ON(item_size
< sizeof(*ei
));
1440 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1441 flags
= btrfs_extent_flags(leaf
, ei
);
1443 ptr
= (unsigned long)(ei
+ 1);
1444 end
= (unsigned long)ei
+ item_size
;
1446 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1447 ptr
+= sizeof(struct btrfs_tree_block_info
);
1450 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1459 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1460 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1464 ptr
+= btrfs_extent_inline_ref_size(type
);
1468 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1469 struct btrfs_extent_data_ref
*dref
;
1470 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1471 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1476 if (hash_extent_data_ref_item(leaf
, dref
) <
1477 hash_extent_data_ref(root_objectid
, owner
, offset
))
1481 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1483 if (parent
== ref_offset
) {
1487 if (ref_offset
< parent
)
1490 if (root_objectid
== ref_offset
) {
1494 if (ref_offset
< root_objectid
)
1498 ptr
+= btrfs_extent_inline_ref_size(type
);
1500 if (err
== -ENOENT
&& insert
) {
1501 if (item_size
+ extra_size
>=
1502 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1507 * To add new inline back ref, we have to make sure
1508 * there is no corresponding back ref item.
1509 * For simplicity, we just do not add new inline back
1510 * ref if there is any kind of item for this block
1512 if (find_next_key(path
, 0, &key
) == 0 &&
1513 key
.objectid
== bytenr
&&
1514 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1519 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1522 path
->keep_locks
= 0;
1523 btrfs_unlock_up_safe(path
, 1);
1529 * helper to add new inline back ref
1531 static noinline_for_stack
1532 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1533 struct btrfs_root
*root
,
1534 struct btrfs_path
*path
,
1535 struct btrfs_extent_inline_ref
*iref
,
1536 u64 parent
, u64 root_objectid
,
1537 u64 owner
, u64 offset
, int refs_to_add
,
1538 struct btrfs_delayed_extent_op
*extent_op
)
1540 struct extent_buffer
*leaf
;
1541 struct btrfs_extent_item
*ei
;
1544 unsigned long item_offset
;
1550 leaf
= path
->nodes
[0];
1551 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1552 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1554 type
= extent_ref_type(parent
, owner
);
1555 size
= btrfs_extent_inline_ref_size(type
);
1557 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1560 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1561 refs
= btrfs_extent_refs(leaf
, ei
);
1562 refs
+= refs_to_add
;
1563 btrfs_set_extent_refs(leaf
, ei
, refs
);
1565 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1567 ptr
= (unsigned long)ei
+ item_offset
;
1568 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1569 if (ptr
< end
- size
)
1570 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1573 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1574 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1575 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1576 struct btrfs_extent_data_ref
*dref
;
1577 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1578 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1579 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1580 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1581 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1582 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1583 struct btrfs_shared_data_ref
*sref
;
1584 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1585 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1586 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1587 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1588 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1590 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1592 btrfs_mark_buffer_dirty(leaf
);
1596 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1597 struct btrfs_root
*root
,
1598 struct btrfs_path
*path
,
1599 struct btrfs_extent_inline_ref
**ref_ret
,
1600 u64 bytenr
, u64 num_bytes
, u64 parent
,
1601 u64 root_objectid
, u64 owner
, u64 offset
)
1605 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1606 bytenr
, num_bytes
, parent
,
1607 root_objectid
, owner
, offset
, 0);
1611 btrfs_release_path(root
, path
);
1614 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1615 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1618 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1619 root_objectid
, owner
, offset
);
1625 * helper to update/remove inline back ref
1627 static noinline_for_stack
1628 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1629 struct btrfs_root
*root
,
1630 struct btrfs_path
*path
,
1631 struct btrfs_extent_inline_ref
*iref
,
1633 struct btrfs_delayed_extent_op
*extent_op
)
1635 struct extent_buffer
*leaf
;
1636 struct btrfs_extent_item
*ei
;
1637 struct btrfs_extent_data_ref
*dref
= NULL
;
1638 struct btrfs_shared_data_ref
*sref
= NULL
;
1647 leaf
= path
->nodes
[0];
1648 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1649 refs
= btrfs_extent_refs(leaf
, ei
);
1650 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1651 refs
+= refs_to_mod
;
1652 btrfs_set_extent_refs(leaf
, ei
, refs
);
1654 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1656 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1658 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1659 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1660 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1661 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1662 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1663 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1666 BUG_ON(refs_to_mod
!= -1);
1669 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1670 refs
+= refs_to_mod
;
1673 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1674 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1676 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1678 size
= btrfs_extent_inline_ref_size(type
);
1679 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1680 ptr
= (unsigned long)iref
;
1681 end
= (unsigned long)ei
+ item_size
;
1682 if (ptr
+ size
< end
)
1683 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1686 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1689 btrfs_mark_buffer_dirty(leaf
);
1693 static noinline_for_stack
1694 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1695 struct btrfs_root
*root
,
1696 struct btrfs_path
*path
,
1697 u64 bytenr
, u64 num_bytes
, u64 parent
,
1698 u64 root_objectid
, u64 owner
,
1699 u64 offset
, int refs_to_add
,
1700 struct btrfs_delayed_extent_op
*extent_op
)
1702 struct btrfs_extent_inline_ref
*iref
;
1705 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1706 bytenr
, num_bytes
, parent
,
1707 root_objectid
, owner
, offset
, 1);
1709 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1710 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1711 refs_to_add
, extent_op
);
1712 } else if (ret
== -ENOENT
) {
1713 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1714 parent
, root_objectid
,
1715 owner
, offset
, refs_to_add
,
1721 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1722 struct btrfs_root
*root
,
1723 struct btrfs_path
*path
,
1724 u64 bytenr
, u64 parent
, u64 root_objectid
,
1725 u64 owner
, u64 offset
, int refs_to_add
)
1728 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1729 BUG_ON(refs_to_add
!= 1);
1730 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1731 parent
, root_objectid
);
1733 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1734 parent
, root_objectid
,
1735 owner
, offset
, refs_to_add
);
1740 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1741 struct btrfs_root
*root
,
1742 struct btrfs_path
*path
,
1743 struct btrfs_extent_inline_ref
*iref
,
1744 int refs_to_drop
, int is_data
)
1748 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1750 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1751 -refs_to_drop
, NULL
);
1752 } else if (is_data
) {
1753 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1755 ret
= btrfs_del_item(trans
, root
, path
);
1760 static int btrfs_issue_discard(struct block_device
*bdev
,
1763 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1766 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1767 u64 num_bytes
, u64
*actual_bytes
)
1770 u64 discarded_bytes
= 0;
1771 struct btrfs_multi_bio
*multi
= NULL
;
1774 /* Tell the block device(s) that the sectors can be discarded */
1775 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1776 bytenr
, &num_bytes
, &multi
, 0);
1778 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1782 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1783 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1787 discarded_bytes
+= stripe
->length
;
1788 else if (ret
!= -EOPNOTSUPP
)
1793 if (discarded_bytes
&& ret
== -EOPNOTSUPP
)
1797 *actual_bytes
= discarded_bytes
;
1803 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1804 struct btrfs_root
*root
,
1805 u64 bytenr
, u64 num_bytes
, u64 parent
,
1806 u64 root_objectid
, u64 owner
, u64 offset
)
1809 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1810 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1812 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1813 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1814 parent
, root_objectid
, (int)owner
,
1815 BTRFS_ADD_DELAYED_REF
, NULL
);
1817 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1818 parent
, root_objectid
, owner
, offset
,
1819 BTRFS_ADD_DELAYED_REF
, NULL
);
1824 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1825 struct btrfs_root
*root
,
1826 u64 bytenr
, u64 num_bytes
,
1827 u64 parent
, u64 root_objectid
,
1828 u64 owner
, u64 offset
, int refs_to_add
,
1829 struct btrfs_delayed_extent_op
*extent_op
)
1831 struct btrfs_path
*path
;
1832 struct extent_buffer
*leaf
;
1833 struct btrfs_extent_item
*item
;
1838 path
= btrfs_alloc_path();
1843 path
->leave_spinning
= 1;
1844 /* this will setup the path even if it fails to insert the back ref */
1845 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1846 path
, bytenr
, num_bytes
, parent
,
1847 root_objectid
, owner
, offset
,
1848 refs_to_add
, extent_op
);
1852 if (ret
!= -EAGAIN
) {
1857 leaf
= path
->nodes
[0];
1858 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1859 refs
= btrfs_extent_refs(leaf
, item
);
1860 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1862 __run_delayed_extent_op(extent_op
, leaf
, item
);
1864 btrfs_mark_buffer_dirty(leaf
);
1865 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1868 path
->leave_spinning
= 1;
1870 /* now insert the actual backref */
1871 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1872 path
, bytenr
, parent
, root_objectid
,
1873 owner
, offset
, refs_to_add
);
1876 btrfs_free_path(path
);
1880 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1881 struct btrfs_root
*root
,
1882 struct btrfs_delayed_ref_node
*node
,
1883 struct btrfs_delayed_extent_op
*extent_op
,
1884 int insert_reserved
)
1887 struct btrfs_delayed_data_ref
*ref
;
1888 struct btrfs_key ins
;
1893 ins
.objectid
= node
->bytenr
;
1894 ins
.offset
= node
->num_bytes
;
1895 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1897 ref
= btrfs_delayed_node_to_data_ref(node
);
1898 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1899 parent
= ref
->parent
;
1901 ref_root
= ref
->root
;
1903 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1905 BUG_ON(extent_op
->update_key
);
1906 flags
|= extent_op
->flags_to_set
;
1908 ret
= alloc_reserved_file_extent(trans
, root
,
1909 parent
, ref_root
, flags
,
1910 ref
->objectid
, ref
->offset
,
1911 &ins
, node
->ref_mod
);
1912 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1913 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1914 node
->num_bytes
, parent
,
1915 ref_root
, ref
->objectid
,
1916 ref
->offset
, node
->ref_mod
,
1918 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1919 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1920 node
->num_bytes
, parent
,
1921 ref_root
, ref
->objectid
,
1922 ref
->offset
, node
->ref_mod
,
1930 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1931 struct extent_buffer
*leaf
,
1932 struct btrfs_extent_item
*ei
)
1934 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1935 if (extent_op
->update_flags
) {
1936 flags
|= extent_op
->flags_to_set
;
1937 btrfs_set_extent_flags(leaf
, ei
, flags
);
1940 if (extent_op
->update_key
) {
1941 struct btrfs_tree_block_info
*bi
;
1942 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1943 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1944 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1948 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1949 struct btrfs_root
*root
,
1950 struct btrfs_delayed_ref_node
*node
,
1951 struct btrfs_delayed_extent_op
*extent_op
)
1953 struct btrfs_key key
;
1954 struct btrfs_path
*path
;
1955 struct btrfs_extent_item
*ei
;
1956 struct extent_buffer
*leaf
;
1961 path
= btrfs_alloc_path();
1965 key
.objectid
= node
->bytenr
;
1966 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1967 key
.offset
= node
->num_bytes
;
1970 path
->leave_spinning
= 1;
1971 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1982 leaf
= path
->nodes
[0];
1983 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1984 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1985 if (item_size
< sizeof(*ei
)) {
1986 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1992 leaf
= path
->nodes
[0];
1993 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1996 BUG_ON(item_size
< sizeof(*ei
));
1997 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1998 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2000 btrfs_mark_buffer_dirty(leaf
);
2002 btrfs_free_path(path
);
2006 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2007 struct btrfs_root
*root
,
2008 struct btrfs_delayed_ref_node
*node
,
2009 struct btrfs_delayed_extent_op
*extent_op
,
2010 int insert_reserved
)
2013 struct btrfs_delayed_tree_ref
*ref
;
2014 struct btrfs_key ins
;
2018 ins
.objectid
= node
->bytenr
;
2019 ins
.offset
= node
->num_bytes
;
2020 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2022 ref
= btrfs_delayed_node_to_tree_ref(node
);
2023 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2024 parent
= ref
->parent
;
2026 ref_root
= ref
->root
;
2028 BUG_ON(node
->ref_mod
!= 1);
2029 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2030 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2031 !extent_op
->update_key
);
2032 ret
= alloc_reserved_tree_block(trans
, root
,
2034 extent_op
->flags_to_set
,
2037 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2038 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2039 node
->num_bytes
, parent
, ref_root
,
2040 ref
->level
, 0, 1, extent_op
);
2041 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2042 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2043 node
->num_bytes
, parent
, ref_root
,
2044 ref
->level
, 0, 1, extent_op
);
2051 /* helper function to actually process a single delayed ref entry */
2052 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2053 struct btrfs_root
*root
,
2054 struct btrfs_delayed_ref_node
*node
,
2055 struct btrfs_delayed_extent_op
*extent_op
,
2056 int insert_reserved
)
2059 if (btrfs_delayed_ref_is_head(node
)) {
2060 struct btrfs_delayed_ref_head
*head
;
2062 * we've hit the end of the chain and we were supposed
2063 * to insert this extent into the tree. But, it got
2064 * deleted before we ever needed to insert it, so all
2065 * we have to do is clean up the accounting
2068 head
= btrfs_delayed_node_to_head(node
);
2069 if (insert_reserved
) {
2070 btrfs_pin_extent(root
, node
->bytenr
,
2071 node
->num_bytes
, 1);
2072 if (head
->is_data
) {
2073 ret
= btrfs_del_csums(trans
, root
,
2079 mutex_unlock(&head
->mutex
);
2083 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2084 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2085 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2087 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2088 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2089 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2096 static noinline
struct btrfs_delayed_ref_node
*
2097 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2099 struct rb_node
*node
;
2100 struct btrfs_delayed_ref_node
*ref
;
2101 int action
= BTRFS_ADD_DELAYED_REF
;
2104 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2105 * this prevents ref count from going down to zero when
2106 * there still are pending delayed ref.
2108 node
= rb_prev(&head
->node
.rb_node
);
2112 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2114 if (ref
->bytenr
!= head
->node
.bytenr
)
2116 if (ref
->action
== action
)
2118 node
= rb_prev(node
);
2120 if (action
== BTRFS_ADD_DELAYED_REF
) {
2121 action
= BTRFS_DROP_DELAYED_REF
;
2127 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2128 struct btrfs_root
*root
,
2129 struct list_head
*cluster
)
2131 struct btrfs_delayed_ref_root
*delayed_refs
;
2132 struct btrfs_delayed_ref_node
*ref
;
2133 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2134 struct btrfs_delayed_extent_op
*extent_op
;
2137 int must_insert_reserved
= 0;
2139 delayed_refs
= &trans
->transaction
->delayed_refs
;
2142 /* pick a new head ref from the cluster list */
2143 if (list_empty(cluster
))
2146 locked_ref
= list_entry(cluster
->next
,
2147 struct btrfs_delayed_ref_head
, cluster
);
2149 /* grab the lock that says we are going to process
2150 * all the refs for this head */
2151 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2154 * we may have dropped the spin lock to get the head
2155 * mutex lock, and that might have given someone else
2156 * time to free the head. If that's true, it has been
2157 * removed from our list and we can move on.
2159 if (ret
== -EAGAIN
) {
2167 * record the must insert reserved flag before we
2168 * drop the spin lock.
2170 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2171 locked_ref
->must_insert_reserved
= 0;
2173 extent_op
= locked_ref
->extent_op
;
2174 locked_ref
->extent_op
= NULL
;
2177 * locked_ref is the head node, so we have to go one
2178 * node back for any delayed ref updates
2180 ref
= select_delayed_ref(locked_ref
);
2182 /* All delayed refs have been processed, Go ahead
2183 * and send the head node to run_one_delayed_ref,
2184 * so that any accounting fixes can happen
2186 ref
= &locked_ref
->node
;
2188 if (extent_op
&& must_insert_reserved
) {
2194 spin_unlock(&delayed_refs
->lock
);
2196 ret
= run_delayed_extent_op(trans
, root
,
2202 spin_lock(&delayed_refs
->lock
);
2206 list_del_init(&locked_ref
->cluster
);
2211 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2212 delayed_refs
->num_entries
--;
2214 spin_unlock(&delayed_refs
->lock
);
2216 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2217 must_insert_reserved
);
2220 btrfs_put_delayed_ref(ref
);
2225 spin_lock(&delayed_refs
->lock
);
2231 * this starts processing the delayed reference count updates and
2232 * extent insertions we have queued up so far. count can be
2233 * 0, which means to process everything in the tree at the start
2234 * of the run (but not newly added entries), or it can be some target
2235 * number you'd like to process.
2237 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2238 struct btrfs_root
*root
, unsigned long count
)
2240 struct rb_node
*node
;
2241 struct btrfs_delayed_ref_root
*delayed_refs
;
2242 struct btrfs_delayed_ref_node
*ref
;
2243 struct list_head cluster
;
2245 int run_all
= count
== (unsigned long)-1;
2248 if (root
== root
->fs_info
->extent_root
)
2249 root
= root
->fs_info
->tree_root
;
2251 delayed_refs
= &trans
->transaction
->delayed_refs
;
2252 INIT_LIST_HEAD(&cluster
);
2254 spin_lock(&delayed_refs
->lock
);
2256 count
= delayed_refs
->num_entries
* 2;
2260 if (!(run_all
|| run_most
) &&
2261 delayed_refs
->num_heads_ready
< 64)
2265 * go find something we can process in the rbtree. We start at
2266 * the beginning of the tree, and then build a cluster
2267 * of refs to process starting at the first one we are able to
2270 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2271 delayed_refs
->run_delayed_start
);
2275 ret
= run_clustered_refs(trans
, root
, &cluster
);
2278 count
-= min_t(unsigned long, ret
, count
);
2285 node
= rb_first(&delayed_refs
->root
);
2288 count
= (unsigned long)-1;
2291 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2293 if (btrfs_delayed_ref_is_head(ref
)) {
2294 struct btrfs_delayed_ref_head
*head
;
2296 head
= btrfs_delayed_node_to_head(ref
);
2297 atomic_inc(&ref
->refs
);
2299 spin_unlock(&delayed_refs
->lock
);
2300 mutex_lock(&head
->mutex
);
2301 mutex_unlock(&head
->mutex
);
2303 btrfs_put_delayed_ref(ref
);
2307 node
= rb_next(node
);
2309 spin_unlock(&delayed_refs
->lock
);
2310 schedule_timeout(1);
2314 spin_unlock(&delayed_refs
->lock
);
2318 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2319 struct btrfs_root
*root
,
2320 u64 bytenr
, u64 num_bytes
, u64 flags
,
2323 struct btrfs_delayed_extent_op
*extent_op
;
2326 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2330 extent_op
->flags_to_set
= flags
;
2331 extent_op
->update_flags
= 1;
2332 extent_op
->update_key
= 0;
2333 extent_op
->is_data
= is_data
? 1 : 0;
2335 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2341 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2342 struct btrfs_root
*root
,
2343 struct btrfs_path
*path
,
2344 u64 objectid
, u64 offset
, u64 bytenr
)
2346 struct btrfs_delayed_ref_head
*head
;
2347 struct btrfs_delayed_ref_node
*ref
;
2348 struct btrfs_delayed_data_ref
*data_ref
;
2349 struct btrfs_delayed_ref_root
*delayed_refs
;
2350 struct rb_node
*node
;
2354 delayed_refs
= &trans
->transaction
->delayed_refs
;
2355 spin_lock(&delayed_refs
->lock
);
2356 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2360 if (!mutex_trylock(&head
->mutex
)) {
2361 atomic_inc(&head
->node
.refs
);
2362 spin_unlock(&delayed_refs
->lock
);
2364 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2366 mutex_lock(&head
->mutex
);
2367 mutex_unlock(&head
->mutex
);
2368 btrfs_put_delayed_ref(&head
->node
);
2372 node
= rb_prev(&head
->node
.rb_node
);
2376 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2378 if (ref
->bytenr
!= bytenr
)
2382 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2385 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2387 node
= rb_prev(node
);
2389 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2390 if (ref
->bytenr
== bytenr
)
2394 if (data_ref
->root
!= root
->root_key
.objectid
||
2395 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2400 mutex_unlock(&head
->mutex
);
2402 spin_unlock(&delayed_refs
->lock
);
2406 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2407 struct btrfs_root
*root
,
2408 struct btrfs_path
*path
,
2409 u64 objectid
, u64 offset
, u64 bytenr
)
2411 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2412 struct extent_buffer
*leaf
;
2413 struct btrfs_extent_data_ref
*ref
;
2414 struct btrfs_extent_inline_ref
*iref
;
2415 struct btrfs_extent_item
*ei
;
2416 struct btrfs_key key
;
2420 key
.objectid
= bytenr
;
2421 key
.offset
= (u64
)-1;
2422 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2424 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2430 if (path
->slots
[0] == 0)
2434 leaf
= path
->nodes
[0];
2435 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2437 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2441 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2442 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2443 if (item_size
< sizeof(*ei
)) {
2444 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2448 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2450 if (item_size
!= sizeof(*ei
) +
2451 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2454 if (btrfs_extent_generation(leaf
, ei
) <=
2455 btrfs_root_last_snapshot(&root
->root_item
))
2458 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2459 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2460 BTRFS_EXTENT_DATA_REF_KEY
)
2463 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2464 if (btrfs_extent_refs(leaf
, ei
) !=
2465 btrfs_extent_data_ref_count(leaf
, ref
) ||
2466 btrfs_extent_data_ref_root(leaf
, ref
) !=
2467 root
->root_key
.objectid
||
2468 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2469 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2477 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2478 struct btrfs_root
*root
,
2479 u64 objectid
, u64 offset
, u64 bytenr
)
2481 struct btrfs_path
*path
;
2485 path
= btrfs_alloc_path();
2490 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2492 if (ret
&& ret
!= -ENOENT
)
2495 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2497 } while (ret2
== -EAGAIN
);
2499 if (ret2
&& ret2
!= -ENOENT
) {
2504 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2507 btrfs_free_path(path
);
2508 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2514 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2515 struct extent_buffer
*buf
, u32 nr_extents
)
2517 struct btrfs_key key
;
2518 struct btrfs_file_extent_item
*fi
;
2526 if (!root
->ref_cows
)
2529 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2531 root_gen
= root
->root_key
.offset
;
2534 root_gen
= trans
->transid
- 1;
2537 level
= btrfs_header_level(buf
);
2538 nritems
= btrfs_header_nritems(buf
);
2541 struct btrfs_leaf_ref
*ref
;
2542 struct btrfs_extent_info
*info
;
2544 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2550 ref
->root_gen
= root_gen
;
2551 ref
->bytenr
= buf
->start
;
2552 ref
->owner
= btrfs_header_owner(buf
);
2553 ref
->generation
= btrfs_header_generation(buf
);
2554 ref
->nritems
= nr_extents
;
2555 info
= ref
->extents
;
2557 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2559 btrfs_item_key_to_cpu(buf
, &key
, i
);
2560 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2562 fi
= btrfs_item_ptr(buf
, i
,
2563 struct btrfs_file_extent_item
);
2564 if (btrfs_file_extent_type(buf
, fi
) ==
2565 BTRFS_FILE_EXTENT_INLINE
)
2567 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2568 if (disk_bytenr
== 0)
2571 info
->bytenr
= disk_bytenr
;
2573 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2574 info
->objectid
= key
.objectid
;
2575 info
->offset
= key
.offset
;
2579 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2580 if (ret
== -EEXIST
&& shared
) {
2581 struct btrfs_leaf_ref
*old
;
2582 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2584 btrfs_remove_leaf_ref(root
, old
);
2585 btrfs_free_leaf_ref(root
, old
);
2586 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2589 btrfs_free_leaf_ref(root
, ref
);
2595 /* when a block goes through cow, we update the reference counts of
2596 * everything that block points to. The internal pointers of the block
2597 * can be in just about any order, and it is likely to have clusters of
2598 * things that are close together and clusters of things that are not.
2600 * To help reduce the seeks that come with updating all of these reference
2601 * counts, sort them by byte number before actual updates are done.
2603 * struct refsort is used to match byte number to slot in the btree block.
2604 * we sort based on the byte number and then use the slot to actually
2607 * struct refsort is smaller than strcut btrfs_item and smaller than
2608 * struct btrfs_key_ptr. Since we're currently limited to the page size
2609 * for a btree block, there's no way for a kmalloc of refsorts for a
2610 * single node to be bigger than a page.
2618 * for passing into sort()
2620 static int refsort_cmp(const void *a_void
, const void *b_void
)
2622 const struct refsort
*a
= a_void
;
2623 const struct refsort
*b
= b_void
;
2625 if (a
->bytenr
< b
->bytenr
)
2627 if (a
->bytenr
> b
->bytenr
)
2633 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2634 struct btrfs_root
*root
,
2635 struct extent_buffer
*buf
,
2636 int full_backref
, int inc
)
2643 struct btrfs_key key
;
2644 struct btrfs_file_extent_item
*fi
;
2648 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2649 u64
, u64
, u64
, u64
, u64
, u64
);
2651 ref_root
= btrfs_header_owner(buf
);
2652 nritems
= btrfs_header_nritems(buf
);
2653 level
= btrfs_header_level(buf
);
2655 if (!root
->ref_cows
&& level
== 0)
2659 process_func
= btrfs_inc_extent_ref
;
2661 process_func
= btrfs_free_extent
;
2664 parent
= buf
->start
;
2668 for (i
= 0; i
< nritems
; i
++) {
2670 btrfs_item_key_to_cpu(buf
, &key
, i
);
2671 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2673 fi
= btrfs_item_ptr(buf
, i
,
2674 struct btrfs_file_extent_item
);
2675 if (btrfs_file_extent_type(buf
, fi
) ==
2676 BTRFS_FILE_EXTENT_INLINE
)
2678 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2682 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2683 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2684 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2685 parent
, ref_root
, key
.objectid
,
2690 bytenr
= btrfs_node_blockptr(buf
, i
);
2691 num_bytes
= btrfs_level_size(root
, level
- 1);
2692 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2693 parent
, ref_root
, level
- 1, 0);
2704 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2705 struct extent_buffer
*buf
, int full_backref
)
2707 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2710 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2711 struct extent_buffer
*buf
, int full_backref
)
2713 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2716 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2717 struct btrfs_root
*root
,
2718 struct btrfs_path
*path
,
2719 struct btrfs_block_group_cache
*cache
)
2722 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2724 struct extent_buffer
*leaf
;
2726 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2731 leaf
= path
->nodes
[0];
2732 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2733 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2734 btrfs_mark_buffer_dirty(leaf
);
2735 btrfs_release_path(extent_root
, path
);
2743 static struct btrfs_block_group_cache
*
2744 next_block_group(struct btrfs_root
*root
,
2745 struct btrfs_block_group_cache
*cache
)
2747 struct rb_node
*node
;
2748 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2749 node
= rb_next(&cache
->cache_node
);
2750 btrfs_put_block_group(cache
);
2752 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2754 btrfs_get_block_group(cache
);
2757 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2761 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2762 struct btrfs_trans_handle
*trans
,
2763 struct btrfs_path
*path
)
2765 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2766 struct inode
*inode
= NULL
;
2768 int dcs
= BTRFS_DC_ERROR
;
2774 * If this block group is smaller than 100 megs don't bother caching the
2777 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2778 spin_lock(&block_group
->lock
);
2779 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2780 spin_unlock(&block_group
->lock
);
2785 inode
= lookup_free_space_inode(root
, block_group
, path
);
2786 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2787 ret
= PTR_ERR(inode
);
2788 btrfs_release_path(root
, path
);
2792 if (IS_ERR(inode
)) {
2796 if (block_group
->ro
)
2799 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2806 * We want to set the generation to 0, that way if anything goes wrong
2807 * from here on out we know not to trust this cache when we load up next
2810 BTRFS_I(inode
)->generation
= 0;
2811 ret
= btrfs_update_inode(trans
, root
, inode
);
2814 if (i_size_read(inode
) > 0) {
2815 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2821 spin_lock(&block_group
->lock
);
2822 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2823 /* We're not cached, don't bother trying to write stuff out */
2824 dcs
= BTRFS_DC_WRITTEN
;
2825 spin_unlock(&block_group
->lock
);
2828 spin_unlock(&block_group
->lock
);
2830 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2835 * Just to make absolutely sure we have enough space, we're going to
2836 * preallocate 12 pages worth of space for each block group. In
2837 * practice we ought to use at most 8, but we need extra space so we can
2838 * add our header and have a terminator between the extents and the
2842 num_pages
*= PAGE_CACHE_SIZE
;
2844 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2848 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2849 num_pages
, num_pages
,
2852 dcs
= BTRFS_DC_SETUP
;
2853 btrfs_free_reserved_data_space(inode
, num_pages
);
2857 btrfs_release_path(root
, path
);
2859 spin_lock(&block_group
->lock
);
2860 block_group
->disk_cache_state
= dcs
;
2861 spin_unlock(&block_group
->lock
);
2866 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2867 struct btrfs_root
*root
)
2869 struct btrfs_block_group_cache
*cache
;
2871 struct btrfs_path
*path
;
2874 path
= btrfs_alloc_path();
2880 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2882 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2884 cache
= next_block_group(root
, cache
);
2892 err
= cache_save_setup(cache
, trans
, path
);
2893 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2894 btrfs_put_block_group(cache
);
2899 err
= btrfs_run_delayed_refs(trans
, root
,
2904 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2906 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2907 btrfs_put_block_group(cache
);
2913 cache
= next_block_group(root
, cache
);
2922 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2923 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2925 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2927 err
= write_one_cache_group(trans
, root
, path
, cache
);
2929 btrfs_put_block_group(cache
);
2934 * I don't think this is needed since we're just marking our
2935 * preallocated extent as written, but just in case it can't
2939 err
= btrfs_run_delayed_refs(trans
, root
,
2944 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2947 * Really this shouldn't happen, but it could if we
2948 * couldn't write the entire preallocated extent and
2949 * splitting the extent resulted in a new block.
2952 btrfs_put_block_group(cache
);
2955 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2957 cache
= next_block_group(root
, cache
);
2966 btrfs_write_out_cache(root
, trans
, cache
, path
);
2969 * If we didn't have an error then the cache state is still
2970 * NEED_WRITE, so we can set it to WRITTEN.
2972 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2973 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2974 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2975 btrfs_put_block_group(cache
);
2978 btrfs_free_path(path
);
2982 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2984 struct btrfs_block_group_cache
*block_group
;
2987 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2988 if (!block_group
|| block_group
->ro
)
2991 btrfs_put_block_group(block_group
);
2995 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2996 u64 total_bytes
, u64 bytes_used
,
2997 struct btrfs_space_info
**space_info
)
2999 struct btrfs_space_info
*found
;
3003 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3004 BTRFS_BLOCK_GROUP_RAID10
))
3009 found
= __find_space_info(info
, flags
);
3011 spin_lock(&found
->lock
);
3012 found
->total_bytes
+= total_bytes
;
3013 found
->disk_total
+= total_bytes
* factor
;
3014 found
->bytes_used
+= bytes_used
;
3015 found
->disk_used
+= bytes_used
* factor
;
3017 spin_unlock(&found
->lock
);
3018 *space_info
= found
;
3021 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3025 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3026 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3027 init_rwsem(&found
->groups_sem
);
3028 spin_lock_init(&found
->lock
);
3029 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
3030 BTRFS_BLOCK_GROUP_SYSTEM
|
3031 BTRFS_BLOCK_GROUP_METADATA
);
3032 found
->total_bytes
= total_bytes
;
3033 found
->disk_total
= total_bytes
* factor
;
3034 found
->bytes_used
= bytes_used
;
3035 found
->disk_used
= bytes_used
* factor
;
3036 found
->bytes_pinned
= 0;
3037 found
->bytes_reserved
= 0;
3038 found
->bytes_readonly
= 0;
3039 found
->bytes_may_use
= 0;
3041 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3042 *space_info
= found
;
3043 list_add_rcu(&found
->list
, &info
->space_info
);
3044 atomic_set(&found
->caching_threads
, 0);
3048 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3050 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
3051 BTRFS_BLOCK_GROUP_RAID1
|
3052 BTRFS_BLOCK_GROUP_RAID10
|
3053 BTRFS_BLOCK_GROUP_DUP
);
3055 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3056 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3057 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3058 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3059 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3060 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3064 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3067 * we add in the count of missing devices because we want
3068 * to make sure that any RAID levels on a degraded FS
3069 * continue to be honored.
3071 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3072 root
->fs_info
->fs_devices
->missing_devices
;
3074 if (num_devices
== 1)
3075 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3076 if (num_devices
< 4)
3077 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3079 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3080 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3081 BTRFS_BLOCK_GROUP_RAID10
))) {
3082 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3085 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3086 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3087 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3090 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3091 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3092 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3093 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3094 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3098 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3100 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3101 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3102 root
->fs_info
->data_alloc_profile
;
3103 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3104 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3105 root
->fs_info
->system_alloc_profile
;
3106 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3107 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3108 root
->fs_info
->metadata_alloc_profile
;
3109 return btrfs_reduce_alloc_profile(root
, flags
);
3112 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3117 flags
= BTRFS_BLOCK_GROUP_DATA
;
3118 else if (root
== root
->fs_info
->chunk_root
)
3119 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3121 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3123 return get_alloc_profile(root
, flags
);
3126 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3128 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3129 BTRFS_BLOCK_GROUP_DATA
);
3133 * This will check the space that the inode allocates from to make sure we have
3134 * enough space for bytes.
3136 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3138 struct btrfs_space_info
*data_sinfo
;
3139 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3141 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3143 /* make sure bytes are sectorsize aligned */
3144 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3146 if (root
== root
->fs_info
->tree_root
) {
3151 data_sinfo
= BTRFS_I(inode
)->space_info
;
3156 /* make sure we have enough space to handle the data first */
3157 spin_lock(&data_sinfo
->lock
);
3158 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3159 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3160 data_sinfo
->bytes_may_use
;
3162 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3163 struct btrfs_trans_handle
*trans
;
3166 * if we don't have enough free bytes in this space then we need
3167 * to alloc a new chunk.
3169 if (!data_sinfo
->full
&& alloc_chunk
) {
3172 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3173 spin_unlock(&data_sinfo
->lock
);
3175 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3176 trans
= btrfs_join_transaction(root
, 1);
3178 return PTR_ERR(trans
);
3180 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3181 bytes
+ 2 * 1024 * 1024,
3183 CHUNK_ALLOC_NO_FORCE
);
3184 btrfs_end_transaction(trans
, root
);
3193 btrfs_set_inode_space_info(root
, inode
);
3194 data_sinfo
= BTRFS_I(inode
)->space_info
;
3198 spin_unlock(&data_sinfo
->lock
);
3200 /* commit the current transaction and try again */
3202 if (!committed
&& !root
->fs_info
->open_ioctl_trans
) {
3204 trans
= btrfs_join_transaction(root
, 1);
3206 return PTR_ERR(trans
);
3207 ret
= btrfs_commit_transaction(trans
, root
);
3213 #if 0 /* I hope we never need this code again, just in case */
3214 printk(KERN_ERR
"no space left, need %llu, %llu bytes_used, "
3215 "%llu bytes_reserved, " "%llu bytes_pinned, "
3216 "%llu bytes_readonly, %llu may use %llu total\n",
3217 (unsigned long long)bytes
,
3218 (unsigned long long)data_sinfo
->bytes_used
,
3219 (unsigned long long)data_sinfo
->bytes_reserved
,
3220 (unsigned long long)data_sinfo
->bytes_pinned
,
3221 (unsigned long long)data_sinfo
->bytes_readonly
,
3222 (unsigned long long)data_sinfo
->bytes_may_use
,
3223 (unsigned long long)data_sinfo
->total_bytes
);
3227 data_sinfo
->bytes_may_use
+= bytes
;
3228 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3229 spin_unlock(&data_sinfo
->lock
);
3235 * called when we are clearing an delalloc extent from the
3236 * inode's io_tree or there was an error for whatever reason
3237 * after calling btrfs_check_data_free_space
3239 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3241 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3242 struct btrfs_space_info
*data_sinfo
;
3244 /* make sure bytes are sectorsize aligned */
3245 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3247 data_sinfo
= BTRFS_I(inode
)->space_info
;
3248 spin_lock(&data_sinfo
->lock
);
3249 data_sinfo
->bytes_may_use
-= bytes
;
3250 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3251 spin_unlock(&data_sinfo
->lock
);
3254 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3256 struct list_head
*head
= &info
->space_info
;
3257 struct btrfs_space_info
*found
;
3260 list_for_each_entry_rcu(found
, head
, list
) {
3261 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3262 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3267 static int should_alloc_chunk(struct btrfs_root
*root
,
3268 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3271 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3272 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3275 if (force
== CHUNK_ALLOC_FORCE
)
3279 * in limited mode, we want to have some free space up to
3280 * about 1% of the FS size.
3282 if (force
== CHUNK_ALLOC_LIMITED
) {
3283 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3284 thresh
= max_t(u64
, 64 * 1024 * 1024,
3285 div_factor_fine(thresh
, 1));
3287 if (num_bytes
- num_allocated
< thresh
)
3292 * we have two similar checks here, one based on percentage
3293 * and once based on a hard number of 256MB. The idea
3294 * is that if we have a good amount of free
3295 * room, don't allocate a chunk. A good mount is
3296 * less than 80% utilized of the chunks we have allocated,
3297 * or more than 256MB free
3299 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3302 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3305 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3307 /* 256MB or 5% of the FS */
3308 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3310 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3315 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3316 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3317 u64 flags
, int force
)
3319 struct btrfs_space_info
*space_info
;
3320 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3323 mutex_lock(&fs_info
->chunk_mutex
);
3325 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3327 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3329 ret
= update_space_info(extent_root
->fs_info
, flags
,
3333 BUG_ON(!space_info
);
3335 spin_lock(&space_info
->lock
);
3336 if (space_info
->force_alloc
)
3337 force
= space_info
->force_alloc
;
3338 if (space_info
->full
) {
3339 spin_unlock(&space_info
->lock
);
3343 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3344 spin_unlock(&space_info
->lock
);
3348 spin_unlock(&space_info
->lock
);
3351 * If we have mixed data/metadata chunks we want to make sure we keep
3352 * allocating mixed chunks instead of individual chunks.
3354 if (btrfs_mixed_space_info(space_info
))
3355 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3358 * if we're doing a data chunk, go ahead and make sure that
3359 * we keep a reasonable number of metadata chunks allocated in the
3362 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3363 fs_info
->data_chunk_allocations
++;
3364 if (!(fs_info
->data_chunk_allocations
%
3365 fs_info
->metadata_ratio
))
3366 force_metadata_allocation(fs_info
);
3369 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3370 spin_lock(&space_info
->lock
);
3372 space_info
->full
= 1;
3375 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3376 spin_unlock(&space_info
->lock
);
3378 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3383 * shrink metadata reservation for delalloc
3385 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3386 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3388 struct btrfs_block_rsv
*block_rsv
;
3389 struct btrfs_space_info
*space_info
;
3394 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3396 unsigned long progress
;
3398 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3399 space_info
= block_rsv
->space_info
;
3402 reserved
= space_info
->bytes_reserved
;
3403 progress
= space_info
->reservation_progress
;
3408 max_reclaim
= min(reserved
, to_reclaim
);
3410 while (loops
< 1024) {
3411 /* have the flusher threads jump in and do some IO */
3413 nr_pages
= min_t(unsigned long, nr_pages
,
3414 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3415 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3417 spin_lock(&space_info
->lock
);
3418 if (reserved
> space_info
->bytes_reserved
)
3419 reclaimed
+= reserved
- space_info
->bytes_reserved
;
3420 reserved
= space_info
->bytes_reserved
;
3421 spin_unlock(&space_info
->lock
);
3425 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3428 if (trans
&& trans
->transaction
->blocked
)
3431 time_left
= schedule_timeout_interruptible(1);
3433 /* We were interrupted, exit */
3437 /* we've kicked the IO a few times, if anything has been freed,
3438 * exit. There is no sense in looping here for a long time
3439 * when we really need to commit the transaction, or there are
3440 * just too many writers without enough free space
3445 if (progress
!= space_info
->reservation_progress
)
3450 return reclaimed
>= to_reclaim
;
3454 * Retries tells us how many times we've called reserve_metadata_bytes. The
3455 * idea is if this is the first call (retries == 0) then we will add to our
3456 * reserved count if we can't make the allocation in order to hold our place
3457 * while we go and try and free up space. That way for retries > 1 we don't try
3458 * and add space, we just check to see if the amount of unused space is >= the
3459 * total space, meaning that our reservation is valid.
3461 * However if we don't intend to retry this reservation, pass -1 as retries so
3462 * that it short circuits this logic.
3464 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3465 struct btrfs_root
*root
,
3466 struct btrfs_block_rsv
*block_rsv
,
3467 u64 orig_bytes
, int flush
)
3469 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3471 u64 num_bytes
= orig_bytes
;
3474 bool reserved
= false;
3475 bool committed
= false;
3482 spin_lock(&space_info
->lock
);
3483 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3484 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3485 space_info
->bytes_may_use
;
3488 * The idea here is that we've not already over-reserved the block group
3489 * then we can go ahead and save our reservation first and then start
3490 * flushing if we need to. Otherwise if we've already overcommitted
3491 * lets start flushing stuff first and then come back and try to make
3494 if (unused
<= space_info
->total_bytes
) {
3495 unused
= space_info
->total_bytes
- unused
;
3496 if (unused
>= num_bytes
) {
3498 space_info
->bytes_reserved
+= orig_bytes
;
3502 * Ok set num_bytes to orig_bytes since we aren't
3503 * overocmmitted, this way we only try and reclaim what
3506 num_bytes
= orig_bytes
;
3510 * Ok we're over committed, set num_bytes to the overcommitted
3511 * amount plus the amount of bytes that we need for this
3514 num_bytes
= unused
- space_info
->total_bytes
+
3515 (orig_bytes
* (retries
+ 1));
3519 * Couldn't make our reservation, save our place so while we're trying
3520 * to reclaim space we can actually use it instead of somebody else
3521 * stealing it from us.
3523 if (ret
&& !reserved
) {
3524 space_info
->bytes_reserved
+= orig_bytes
;
3528 spin_unlock(&space_info
->lock
);
3537 * We do synchronous shrinking since we don't actually unreserve
3538 * metadata until after the IO is completed.
3540 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3547 * So if we were overcommitted it's possible that somebody else flushed
3548 * out enough space and we simply didn't have enough space to reclaim,
3549 * so go back around and try again.
3556 spin_lock(&space_info
->lock
);
3558 * Not enough space to be reclaimed, don't bother committing the
3561 if (space_info
->bytes_pinned
< orig_bytes
)
3563 spin_unlock(&space_info
->lock
);
3568 if (trans
|| committed
)
3572 trans
= btrfs_join_transaction(root
, 1);
3575 ret
= btrfs_commit_transaction(trans
, root
);
3584 spin_lock(&space_info
->lock
);
3585 space_info
->bytes_reserved
-= orig_bytes
;
3586 spin_unlock(&space_info
->lock
);
3592 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3593 struct btrfs_root
*root
)
3595 struct btrfs_block_rsv
*block_rsv
;
3597 block_rsv
= trans
->block_rsv
;
3599 block_rsv
= root
->block_rsv
;
3602 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3607 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3611 spin_lock(&block_rsv
->lock
);
3612 if (block_rsv
->reserved
>= num_bytes
) {
3613 block_rsv
->reserved
-= num_bytes
;
3614 if (block_rsv
->reserved
< block_rsv
->size
)
3615 block_rsv
->full
= 0;
3618 spin_unlock(&block_rsv
->lock
);
3622 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3623 u64 num_bytes
, int update_size
)
3625 spin_lock(&block_rsv
->lock
);
3626 block_rsv
->reserved
+= num_bytes
;
3628 block_rsv
->size
+= num_bytes
;
3629 else if (block_rsv
->reserved
>= block_rsv
->size
)
3630 block_rsv
->full
= 1;
3631 spin_unlock(&block_rsv
->lock
);
3634 void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3635 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3637 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3639 spin_lock(&block_rsv
->lock
);
3640 if (num_bytes
== (u64
)-1)
3641 num_bytes
= block_rsv
->size
;
3642 block_rsv
->size
-= num_bytes
;
3643 if (block_rsv
->reserved
>= block_rsv
->size
) {
3644 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3645 block_rsv
->reserved
= block_rsv
->size
;
3646 block_rsv
->full
= 1;
3650 spin_unlock(&block_rsv
->lock
);
3652 if (num_bytes
> 0) {
3654 spin_lock(&dest
->lock
);
3658 bytes_to_add
= dest
->size
- dest
->reserved
;
3659 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3660 dest
->reserved
+= bytes_to_add
;
3661 if (dest
->reserved
>= dest
->size
)
3663 num_bytes
-= bytes_to_add
;
3665 spin_unlock(&dest
->lock
);
3668 spin_lock(&space_info
->lock
);
3669 space_info
->bytes_reserved
-= num_bytes
;
3670 space_info
->reservation_progress
++;
3671 spin_unlock(&space_info
->lock
);
3676 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3677 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3681 ret
= block_rsv_use_bytes(src
, num_bytes
);
3685 block_rsv_add_bytes(dst
, num_bytes
, 1);
3689 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3691 memset(rsv
, 0, sizeof(*rsv
));
3692 spin_lock_init(&rsv
->lock
);
3693 atomic_set(&rsv
->usage
, 1);
3695 INIT_LIST_HEAD(&rsv
->list
);
3698 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3700 struct btrfs_block_rsv
*block_rsv
;
3701 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3703 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3707 btrfs_init_block_rsv(block_rsv
);
3708 block_rsv
->space_info
= __find_space_info(fs_info
,
3709 BTRFS_BLOCK_GROUP_METADATA
);
3713 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3714 struct btrfs_block_rsv
*rsv
)
3716 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3717 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3724 * make the block_rsv struct be able to capture freed space.
3725 * the captured space will re-add to the the block_rsv struct
3726 * after transaction commit
3728 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3729 struct btrfs_block_rsv
*block_rsv
)
3731 block_rsv
->durable
= 1;
3732 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3733 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3734 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3737 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3738 struct btrfs_root
*root
,
3739 struct btrfs_block_rsv
*block_rsv
,
3747 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3749 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3756 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3757 struct btrfs_root
*root
,
3758 struct btrfs_block_rsv
*block_rsv
,
3759 u64 min_reserved
, int min_factor
)
3762 int commit_trans
= 0;
3768 spin_lock(&block_rsv
->lock
);
3770 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3771 if (min_reserved
> num_bytes
)
3772 num_bytes
= min_reserved
;
3774 if (block_rsv
->reserved
>= num_bytes
) {
3777 num_bytes
-= block_rsv
->reserved
;
3778 if (block_rsv
->durable
&&
3779 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3782 spin_unlock(&block_rsv
->lock
);
3786 if (block_rsv
->refill_used
) {
3787 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3790 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3799 trans
= btrfs_join_transaction(root
, 1);
3800 BUG_ON(IS_ERR(trans
));
3801 ret
= btrfs_commit_transaction(trans
, root
);
3808 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3809 struct btrfs_block_rsv
*dst_rsv
,
3812 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3815 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3816 struct btrfs_block_rsv
*block_rsv
,
3819 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3820 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3821 block_rsv
->space_info
!= global_rsv
->space_info
)
3823 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3827 * helper to calculate size of global block reservation.
3828 * the desired value is sum of space used by extent tree,
3829 * checksum tree and root tree
3831 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3833 struct btrfs_space_info
*sinfo
;
3837 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3840 * per tree used space accounting can be inaccuracy, so we
3843 spin_lock(&fs_info
->extent_root
->accounting_lock
);
3844 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
);
3845 spin_unlock(&fs_info
->extent_root
->accounting_lock
);
3847 spin_lock(&fs_info
->csum_root
->accounting_lock
);
3848 num_bytes
+= btrfs_root_used(&fs_info
->csum_root
->root_item
);
3849 spin_unlock(&fs_info
->csum_root
->accounting_lock
);
3851 spin_lock(&fs_info
->tree_root
->accounting_lock
);
3852 num_bytes
+= btrfs_root_used(&fs_info
->tree_root
->root_item
);
3853 spin_unlock(&fs_info
->tree_root
->accounting_lock
);
3855 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3856 spin_lock(&sinfo
->lock
);
3857 data_used
= sinfo
->bytes_used
;
3858 spin_unlock(&sinfo
->lock
);
3860 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3861 spin_lock(&sinfo
->lock
);
3862 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3864 meta_used
= sinfo
->bytes_used
;
3865 spin_unlock(&sinfo
->lock
);
3867 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3869 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3871 if (num_bytes
* 3 > meta_used
)
3872 num_bytes
= div64_u64(meta_used
, 3);
3874 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3877 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3879 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3880 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3883 num_bytes
= calc_global_metadata_size(fs_info
);
3885 spin_lock(&block_rsv
->lock
);
3886 spin_lock(&sinfo
->lock
);
3888 block_rsv
->size
= num_bytes
;
3890 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3891 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3892 sinfo
->bytes_may_use
;
3894 if (sinfo
->total_bytes
> num_bytes
) {
3895 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3896 block_rsv
->reserved
+= num_bytes
;
3897 sinfo
->bytes_reserved
+= num_bytes
;
3900 if (block_rsv
->reserved
>= block_rsv
->size
) {
3901 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3902 sinfo
->bytes_reserved
-= num_bytes
;
3903 sinfo
->reservation_progress
++;
3904 block_rsv
->reserved
= block_rsv
->size
;
3905 block_rsv
->full
= 1;
3908 printk(KERN_INFO
"global block rsv size %llu reserved %llu\n",
3909 block_rsv
->size
, block_rsv
->reserved
);
3911 spin_unlock(&sinfo
->lock
);
3912 spin_unlock(&block_rsv
->lock
);
3915 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3917 struct btrfs_space_info
*space_info
;
3919 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3920 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3921 fs_info
->chunk_block_rsv
.priority
= 10;
3923 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3924 fs_info
->global_block_rsv
.space_info
= space_info
;
3925 fs_info
->global_block_rsv
.priority
= 10;
3926 fs_info
->global_block_rsv
.refill_used
= 1;
3927 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3928 fs_info
->trans_block_rsv
.space_info
= space_info
;
3929 fs_info
->empty_block_rsv
.space_info
= space_info
;
3930 fs_info
->empty_block_rsv
.priority
= 10;
3932 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3933 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3934 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3935 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3936 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3938 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3940 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3942 update_global_block_rsv(fs_info
);
3945 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3947 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3948 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3949 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3950 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3951 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3952 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3953 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3956 static u64
calc_trans_metadata_size(struct btrfs_root
*root
, int num_items
)
3958 return (root
->leafsize
+ root
->nodesize
* (BTRFS_MAX_LEVEL
- 1)) *
3962 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle
*trans
,
3963 struct btrfs_root
*root
,
3969 if (num_items
== 0 || root
->fs_info
->chunk_root
== root
)
3972 num_bytes
= calc_trans_metadata_size(root
, num_items
);
3973 ret
= btrfs_block_rsv_add(trans
, root
, &root
->fs_info
->trans_block_rsv
,
3976 trans
->bytes_reserved
+= num_bytes
;
3977 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3982 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3983 struct btrfs_root
*root
)
3985 if (!trans
->bytes_reserved
)
3988 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3989 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3990 trans
->bytes_reserved
);
3991 trans
->bytes_reserved
= 0;
3994 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3995 struct inode
*inode
)
3997 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3998 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3999 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4002 * one for deleting orphan item, one for updating inode and
4003 * two for calling btrfs_truncate_inode_items.
4005 * btrfs_truncate_inode_items is a delete operation, it frees
4006 * more space than it uses in most cases. So two units of
4007 * metadata space should be enough for calling it many times.
4008 * If all of the metadata space is used, we can commit
4009 * transaction and use space it freed.
4011 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
4012 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4015 void btrfs_orphan_release_metadata(struct inode
*inode
)
4017 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4018 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
4019 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4022 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
4023 struct btrfs_pending_snapshot
*pending
)
4025 struct btrfs_root
*root
= pending
->root
;
4026 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4027 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
4029 * two for root back/forward refs, two for directory entries
4030 * and one for root of the snapshot.
4032 u64 num_bytes
= calc_trans_metadata_size(root
, 5);
4033 dst_rsv
->space_info
= src_rsv
->space_info
;
4034 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4037 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
4039 return num_bytes
>>= 3;
4042 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4044 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4045 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4048 int reserved_extents
;
4051 if (btrfs_transaction_in_commit(root
->fs_info
))
4052 schedule_timeout(1);
4054 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4056 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
) + 1;
4057 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4059 if (nr_extents
> reserved_extents
) {
4060 nr_extents
-= reserved_extents
;
4061 to_reserve
= calc_trans_metadata_size(root
, nr_extents
);
4067 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
4068 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
4072 atomic_add(nr_extents
, &BTRFS_I(inode
)->reserved_extents
);
4073 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
4075 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4077 if (block_rsv
->size
> 512 * 1024 * 1024)
4078 shrink_delalloc(NULL
, root
, to_reserve
, 0);
4083 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4085 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4088 int reserved_extents
;
4090 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4091 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
4092 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
) < 0);
4094 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4098 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
);
4099 if (nr_extents
>= reserved_extents
) {
4103 old
= reserved_extents
;
4104 nr_extents
= reserved_extents
- nr_extents
;
4105 new = reserved_extents
- nr_extents
;
4106 old
= atomic_cmpxchg(&BTRFS_I(inode
)->reserved_extents
,
4107 reserved_extents
, new);
4108 if (likely(old
== reserved_extents
))
4110 reserved_extents
= old
;
4113 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4115 to_free
+= calc_trans_metadata_size(root
, nr_extents
);
4117 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4121 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4125 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4129 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4131 btrfs_free_reserved_data_space(inode
, num_bytes
);
4138 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4140 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4141 btrfs_free_reserved_data_space(inode
, num_bytes
);
4144 static int update_block_group(struct btrfs_trans_handle
*trans
,
4145 struct btrfs_root
*root
,
4146 u64 bytenr
, u64 num_bytes
, int alloc
)
4148 struct btrfs_block_group_cache
*cache
= NULL
;
4149 struct btrfs_fs_info
*info
= root
->fs_info
;
4150 u64 total
= num_bytes
;
4155 /* block accounting for super block */
4156 spin_lock(&info
->delalloc_lock
);
4157 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4159 old_val
+= num_bytes
;
4161 old_val
-= num_bytes
;
4162 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4163 spin_unlock(&info
->delalloc_lock
);
4166 cache
= btrfs_lookup_block_group(info
, bytenr
);
4169 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4170 BTRFS_BLOCK_GROUP_RAID1
|
4171 BTRFS_BLOCK_GROUP_RAID10
))
4176 * If this block group has free space cache written out, we
4177 * need to make sure to load it if we are removing space. This
4178 * is because we need the unpinning stage to actually add the
4179 * space back to the block group, otherwise we will leak space.
4181 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4182 cache_block_group(cache
, trans
, NULL
, 1);
4184 byte_in_group
= bytenr
- cache
->key
.objectid
;
4185 WARN_ON(byte_in_group
> cache
->key
.offset
);
4187 spin_lock(&cache
->space_info
->lock
);
4188 spin_lock(&cache
->lock
);
4190 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4191 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4192 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4195 old_val
= btrfs_block_group_used(&cache
->item
);
4196 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4198 old_val
+= num_bytes
;
4199 btrfs_set_block_group_used(&cache
->item
, old_val
);
4200 cache
->reserved
-= num_bytes
;
4201 cache
->space_info
->bytes_reserved
-= num_bytes
;
4202 cache
->space_info
->reservation_progress
++;
4203 cache
->space_info
->bytes_used
+= num_bytes
;
4204 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4205 spin_unlock(&cache
->lock
);
4206 spin_unlock(&cache
->space_info
->lock
);
4208 old_val
-= num_bytes
;
4209 btrfs_set_block_group_used(&cache
->item
, old_val
);
4210 cache
->pinned
+= num_bytes
;
4211 cache
->space_info
->bytes_pinned
+= num_bytes
;
4212 cache
->space_info
->bytes_used
-= num_bytes
;
4213 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4214 spin_unlock(&cache
->lock
);
4215 spin_unlock(&cache
->space_info
->lock
);
4217 set_extent_dirty(info
->pinned_extents
,
4218 bytenr
, bytenr
+ num_bytes
- 1,
4219 GFP_NOFS
| __GFP_NOFAIL
);
4221 btrfs_put_block_group(cache
);
4223 bytenr
+= num_bytes
;
4228 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4230 struct btrfs_block_group_cache
*cache
;
4233 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4237 bytenr
= cache
->key
.objectid
;
4238 btrfs_put_block_group(cache
);
4243 static int pin_down_extent(struct btrfs_root
*root
,
4244 struct btrfs_block_group_cache
*cache
,
4245 u64 bytenr
, u64 num_bytes
, int reserved
)
4247 spin_lock(&cache
->space_info
->lock
);
4248 spin_lock(&cache
->lock
);
4249 cache
->pinned
+= num_bytes
;
4250 cache
->space_info
->bytes_pinned
+= num_bytes
;
4252 cache
->reserved
-= num_bytes
;
4253 cache
->space_info
->bytes_reserved
-= num_bytes
;
4254 cache
->space_info
->reservation_progress
++;
4256 spin_unlock(&cache
->lock
);
4257 spin_unlock(&cache
->space_info
->lock
);
4259 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4260 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4265 * this function must be called within transaction
4267 int btrfs_pin_extent(struct btrfs_root
*root
,
4268 u64 bytenr
, u64 num_bytes
, int reserved
)
4270 struct btrfs_block_group_cache
*cache
;
4272 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4275 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4277 btrfs_put_block_group(cache
);
4282 * update size of reserved extents. this function may return -EAGAIN
4283 * if 'reserve' is true or 'sinfo' is false.
4285 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4286 u64 num_bytes
, int reserve
, int sinfo
)
4290 struct btrfs_space_info
*space_info
= cache
->space_info
;
4291 spin_lock(&space_info
->lock
);
4292 spin_lock(&cache
->lock
);
4297 cache
->reserved
+= num_bytes
;
4298 space_info
->bytes_reserved
+= num_bytes
;
4302 space_info
->bytes_readonly
+= num_bytes
;
4303 cache
->reserved
-= num_bytes
;
4304 space_info
->bytes_reserved
-= num_bytes
;
4305 space_info
->reservation_progress
++;
4307 spin_unlock(&cache
->lock
);
4308 spin_unlock(&space_info
->lock
);
4310 spin_lock(&cache
->lock
);
4315 cache
->reserved
+= num_bytes
;
4317 cache
->reserved
-= num_bytes
;
4319 spin_unlock(&cache
->lock
);
4324 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4325 struct btrfs_root
*root
)
4327 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4328 struct btrfs_caching_control
*next
;
4329 struct btrfs_caching_control
*caching_ctl
;
4330 struct btrfs_block_group_cache
*cache
;
4332 down_write(&fs_info
->extent_commit_sem
);
4334 list_for_each_entry_safe(caching_ctl
, next
,
4335 &fs_info
->caching_block_groups
, list
) {
4336 cache
= caching_ctl
->block_group
;
4337 if (block_group_cache_done(cache
)) {
4338 cache
->last_byte_to_unpin
= (u64
)-1;
4339 list_del_init(&caching_ctl
->list
);
4340 put_caching_control(caching_ctl
);
4342 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4346 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4347 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4349 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4351 up_write(&fs_info
->extent_commit_sem
);
4353 update_global_block_rsv(fs_info
);
4357 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4359 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4360 struct btrfs_block_group_cache
*cache
= NULL
;
4363 while (start
<= end
) {
4365 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4367 btrfs_put_block_group(cache
);
4368 cache
= btrfs_lookup_block_group(fs_info
, start
);
4372 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4373 len
= min(len
, end
+ 1 - start
);
4375 if (start
< cache
->last_byte_to_unpin
) {
4376 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4377 btrfs_add_free_space(cache
, start
, len
);
4382 spin_lock(&cache
->space_info
->lock
);
4383 spin_lock(&cache
->lock
);
4384 cache
->pinned
-= len
;
4385 cache
->space_info
->bytes_pinned
-= len
;
4387 cache
->space_info
->bytes_readonly
+= len
;
4388 } else if (cache
->reserved_pinned
> 0) {
4389 len
= min(len
, cache
->reserved_pinned
);
4390 cache
->reserved_pinned
-= len
;
4391 cache
->space_info
->bytes_reserved
+= len
;
4393 spin_unlock(&cache
->lock
);
4394 spin_unlock(&cache
->space_info
->lock
);
4398 btrfs_put_block_group(cache
);
4402 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4403 struct btrfs_root
*root
)
4405 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4406 struct extent_io_tree
*unpin
;
4407 struct btrfs_block_rsv
*block_rsv
;
4408 struct btrfs_block_rsv
*next_rsv
;
4414 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4415 unpin
= &fs_info
->freed_extents
[1];
4417 unpin
= &fs_info
->freed_extents
[0];
4420 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4425 if (btrfs_test_opt(root
, DISCARD
))
4426 ret
= btrfs_discard_extent(root
, start
,
4427 end
+ 1 - start
, NULL
);
4429 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4430 unpin_extent_range(root
, start
, end
);
4434 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4435 list_for_each_entry_safe(block_rsv
, next_rsv
,
4436 &fs_info
->durable_block_rsv_list
, list
) {
4438 idx
= trans
->transid
& 0x1;
4439 if (block_rsv
->freed
[idx
] > 0) {
4440 block_rsv_add_bytes(block_rsv
,
4441 block_rsv
->freed
[idx
], 0);
4442 block_rsv
->freed
[idx
] = 0;
4444 if (atomic_read(&block_rsv
->usage
) == 0) {
4445 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4447 if (block_rsv
->freed
[0] == 0 &&
4448 block_rsv
->freed
[1] == 0) {
4449 list_del_init(&block_rsv
->list
);
4453 btrfs_block_rsv_release(root
, block_rsv
, 0);
4456 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4461 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4462 struct btrfs_root
*root
,
4463 u64 bytenr
, u64 num_bytes
, u64 parent
,
4464 u64 root_objectid
, u64 owner_objectid
,
4465 u64 owner_offset
, int refs_to_drop
,
4466 struct btrfs_delayed_extent_op
*extent_op
)
4468 struct btrfs_key key
;
4469 struct btrfs_path
*path
;
4470 struct btrfs_fs_info
*info
= root
->fs_info
;
4471 struct btrfs_root
*extent_root
= info
->extent_root
;
4472 struct extent_buffer
*leaf
;
4473 struct btrfs_extent_item
*ei
;
4474 struct btrfs_extent_inline_ref
*iref
;
4477 int extent_slot
= 0;
4478 int found_extent
= 0;
4483 path
= btrfs_alloc_path();
4488 path
->leave_spinning
= 1;
4490 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4491 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4493 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4494 bytenr
, num_bytes
, parent
,
4495 root_objectid
, owner_objectid
,
4498 extent_slot
= path
->slots
[0];
4499 while (extent_slot
>= 0) {
4500 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4502 if (key
.objectid
!= bytenr
)
4504 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4505 key
.offset
== num_bytes
) {
4509 if (path
->slots
[0] - extent_slot
> 5)
4513 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4514 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4515 if (found_extent
&& item_size
< sizeof(*ei
))
4518 if (!found_extent
) {
4520 ret
= remove_extent_backref(trans
, extent_root
, path
,
4524 btrfs_release_path(extent_root
, path
);
4525 path
->leave_spinning
= 1;
4527 key
.objectid
= bytenr
;
4528 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4529 key
.offset
= num_bytes
;
4531 ret
= btrfs_search_slot(trans
, extent_root
,
4534 printk(KERN_ERR
"umm, got %d back from search"
4535 ", was looking for %llu\n", ret
,
4536 (unsigned long long)bytenr
);
4537 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4540 extent_slot
= path
->slots
[0];
4543 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4545 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4546 "parent %llu root %llu owner %llu offset %llu\n",
4547 (unsigned long long)bytenr
,
4548 (unsigned long long)parent
,
4549 (unsigned long long)root_objectid
,
4550 (unsigned long long)owner_objectid
,
4551 (unsigned long long)owner_offset
);
4554 leaf
= path
->nodes
[0];
4555 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4556 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4557 if (item_size
< sizeof(*ei
)) {
4558 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4559 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4563 btrfs_release_path(extent_root
, path
);
4564 path
->leave_spinning
= 1;
4566 key
.objectid
= bytenr
;
4567 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4568 key
.offset
= num_bytes
;
4570 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4573 printk(KERN_ERR
"umm, got %d back from search"
4574 ", was looking for %llu\n", ret
,
4575 (unsigned long long)bytenr
);
4576 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4579 extent_slot
= path
->slots
[0];
4580 leaf
= path
->nodes
[0];
4581 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4584 BUG_ON(item_size
< sizeof(*ei
));
4585 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4586 struct btrfs_extent_item
);
4587 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4588 struct btrfs_tree_block_info
*bi
;
4589 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4590 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4591 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4594 refs
= btrfs_extent_refs(leaf
, ei
);
4595 BUG_ON(refs
< refs_to_drop
);
4596 refs
-= refs_to_drop
;
4600 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4602 * In the case of inline back ref, reference count will
4603 * be updated by remove_extent_backref
4606 BUG_ON(!found_extent
);
4608 btrfs_set_extent_refs(leaf
, ei
, refs
);
4609 btrfs_mark_buffer_dirty(leaf
);
4612 ret
= remove_extent_backref(trans
, extent_root
, path
,
4619 BUG_ON(is_data
&& refs_to_drop
!=
4620 extent_data_ref_count(root
, path
, iref
));
4622 BUG_ON(path
->slots
[0] != extent_slot
);
4624 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4625 path
->slots
[0] = extent_slot
;
4630 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4633 btrfs_release_path(extent_root
, path
);
4636 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4639 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4640 bytenr
>> PAGE_CACHE_SHIFT
,
4641 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4644 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4647 btrfs_free_path(path
);
4652 * when we free an block, it is possible (and likely) that we free the last
4653 * delayed ref for that extent as well. This searches the delayed ref tree for
4654 * a given extent, and if there are no other delayed refs to be processed, it
4655 * removes it from the tree.
4657 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4658 struct btrfs_root
*root
, u64 bytenr
)
4660 struct btrfs_delayed_ref_head
*head
;
4661 struct btrfs_delayed_ref_root
*delayed_refs
;
4662 struct btrfs_delayed_ref_node
*ref
;
4663 struct rb_node
*node
;
4666 delayed_refs
= &trans
->transaction
->delayed_refs
;
4667 spin_lock(&delayed_refs
->lock
);
4668 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4672 node
= rb_prev(&head
->node
.rb_node
);
4676 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4678 /* there are still entries for this ref, we can't drop it */
4679 if (ref
->bytenr
== bytenr
)
4682 if (head
->extent_op
) {
4683 if (!head
->must_insert_reserved
)
4685 kfree(head
->extent_op
);
4686 head
->extent_op
= NULL
;
4690 * waiting for the lock here would deadlock. If someone else has it
4691 * locked they are already in the process of dropping it anyway
4693 if (!mutex_trylock(&head
->mutex
))
4697 * at this point we have a head with no other entries. Go
4698 * ahead and process it.
4700 head
->node
.in_tree
= 0;
4701 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4703 delayed_refs
->num_entries
--;
4706 * we don't take a ref on the node because we're removing it from the
4707 * tree, so we just steal the ref the tree was holding.
4709 delayed_refs
->num_heads
--;
4710 if (list_empty(&head
->cluster
))
4711 delayed_refs
->num_heads_ready
--;
4713 list_del_init(&head
->cluster
);
4714 spin_unlock(&delayed_refs
->lock
);
4716 BUG_ON(head
->extent_op
);
4717 if (head
->must_insert_reserved
)
4720 mutex_unlock(&head
->mutex
);
4721 btrfs_put_delayed_ref(&head
->node
);
4724 spin_unlock(&delayed_refs
->lock
);
4728 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4729 struct btrfs_root
*root
,
4730 struct extent_buffer
*buf
,
4731 u64 parent
, int last_ref
)
4733 struct btrfs_block_rsv
*block_rsv
;
4734 struct btrfs_block_group_cache
*cache
= NULL
;
4737 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4738 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4739 parent
, root
->root_key
.objectid
,
4740 btrfs_header_level(buf
),
4741 BTRFS_DROP_DELAYED_REF
, NULL
);
4748 block_rsv
= get_block_rsv(trans
, root
);
4749 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4750 if (block_rsv
->space_info
!= cache
->space_info
)
4753 if (btrfs_header_generation(buf
) == trans
->transid
) {
4754 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4755 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4760 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4761 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4765 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4767 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4768 ret
= btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 0);
4769 if (ret
== -EAGAIN
) {
4770 /* block group became read-only */
4771 btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 1);
4776 spin_lock(&block_rsv
->lock
);
4777 if (block_rsv
->reserved
< block_rsv
->size
) {
4778 block_rsv
->reserved
+= buf
->len
;
4781 spin_unlock(&block_rsv
->lock
);
4784 spin_lock(&cache
->space_info
->lock
);
4785 cache
->space_info
->bytes_reserved
-= buf
->len
;
4786 cache
->space_info
->reservation_progress
++;
4787 spin_unlock(&cache
->space_info
->lock
);
4792 if (block_rsv
->durable
&& !cache
->ro
) {
4794 spin_lock(&cache
->lock
);
4796 cache
->reserved_pinned
+= buf
->len
;
4799 spin_unlock(&cache
->lock
);
4802 spin_lock(&block_rsv
->lock
);
4803 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4804 spin_unlock(&block_rsv
->lock
);
4809 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4812 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4813 btrfs_put_block_group(cache
);
4816 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4817 struct btrfs_root
*root
,
4818 u64 bytenr
, u64 num_bytes
, u64 parent
,
4819 u64 root_objectid
, u64 owner
, u64 offset
)
4824 * tree log blocks never actually go into the extent allocation
4825 * tree, just update pinning info and exit early.
4827 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4828 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4829 /* unlocks the pinned mutex */
4830 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4832 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4833 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4834 parent
, root_objectid
, (int)owner
,
4835 BTRFS_DROP_DELAYED_REF
, NULL
);
4838 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4839 parent
, root_objectid
, owner
,
4840 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4846 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4848 u64 mask
= ((u64
)root
->stripesize
- 1);
4849 u64 ret
= (val
+ mask
) & ~mask
;
4854 * when we wait for progress in the block group caching, its because
4855 * our allocation attempt failed at least once. So, we must sleep
4856 * and let some progress happen before we try again.
4858 * This function will sleep at least once waiting for new free space to
4859 * show up, and then it will check the block group free space numbers
4860 * for our min num_bytes. Another option is to have it go ahead
4861 * and look in the rbtree for a free extent of a given size, but this
4865 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4868 struct btrfs_caching_control
*caching_ctl
;
4871 caching_ctl
= get_caching_control(cache
);
4875 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4876 (cache
->free_space
>= num_bytes
));
4878 put_caching_control(caching_ctl
);
4883 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4885 struct btrfs_caching_control
*caching_ctl
;
4888 caching_ctl
= get_caching_control(cache
);
4892 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4894 put_caching_control(caching_ctl
);
4898 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4901 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4903 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4905 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4907 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4914 enum btrfs_loop_type
{
4915 LOOP_FIND_IDEAL
= 0,
4916 LOOP_CACHING_NOWAIT
= 1,
4917 LOOP_CACHING_WAIT
= 2,
4918 LOOP_ALLOC_CHUNK
= 3,
4919 LOOP_NO_EMPTY_SIZE
= 4,
4923 * walks the btree of allocated extents and find a hole of a given size.
4924 * The key ins is changed to record the hole:
4925 * ins->objectid == block start
4926 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4927 * ins->offset == number of blocks
4928 * Any available blocks before search_start are skipped.
4930 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4931 struct btrfs_root
*orig_root
,
4932 u64 num_bytes
, u64 empty_size
,
4933 u64 search_start
, u64 search_end
,
4934 u64 hint_byte
, struct btrfs_key
*ins
,
4938 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4939 struct btrfs_free_cluster
*last_ptr
= NULL
;
4940 struct btrfs_block_group_cache
*block_group
= NULL
;
4941 int empty_cluster
= 2 * 1024 * 1024;
4942 int allowed_chunk_alloc
= 0;
4943 int done_chunk_alloc
= 0;
4944 struct btrfs_space_info
*space_info
;
4945 int last_ptr_loop
= 0;
4948 bool found_uncached_bg
= false;
4949 bool failed_cluster_refill
= false;
4950 bool failed_alloc
= false;
4951 bool use_cluster
= true;
4952 u64 ideal_cache_percent
= 0;
4953 u64 ideal_cache_offset
= 0;
4955 WARN_ON(num_bytes
< root
->sectorsize
);
4956 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4960 space_info
= __find_space_info(root
->fs_info
, data
);
4962 printk(KERN_ERR
"No space info for %d\n", data
);
4967 * If the space info is for both data and metadata it means we have a
4968 * small filesystem and we can't use the clustering stuff.
4970 if (btrfs_mixed_space_info(space_info
))
4971 use_cluster
= false;
4973 if (orig_root
->ref_cows
|| empty_size
)
4974 allowed_chunk_alloc
= 1;
4976 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4977 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4978 if (!btrfs_test_opt(root
, SSD
))
4979 empty_cluster
= 64 * 1024;
4982 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4983 btrfs_test_opt(root
, SSD
)) {
4984 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4988 spin_lock(&last_ptr
->lock
);
4989 if (last_ptr
->block_group
)
4990 hint_byte
= last_ptr
->window_start
;
4991 spin_unlock(&last_ptr
->lock
);
4994 search_start
= max(search_start
, first_logical_byte(root
, 0));
4995 search_start
= max(search_start
, hint_byte
);
5000 if (search_start
== hint_byte
) {
5002 block_group
= btrfs_lookup_block_group(root
->fs_info
,
5005 * we don't want to use the block group if it doesn't match our
5006 * allocation bits, or if its not cached.
5008 * However if we are re-searching with an ideal block group
5009 * picked out then we don't care that the block group is cached.
5011 if (block_group
&& block_group_bits(block_group
, data
) &&
5012 (block_group
->cached
!= BTRFS_CACHE_NO
||
5013 search_start
== ideal_cache_offset
)) {
5014 down_read(&space_info
->groups_sem
);
5015 if (list_empty(&block_group
->list
) ||
5018 * someone is removing this block group,
5019 * we can't jump into the have_block_group
5020 * target because our list pointers are not
5023 btrfs_put_block_group(block_group
);
5024 up_read(&space_info
->groups_sem
);
5026 index
= get_block_group_index(block_group
);
5027 goto have_block_group
;
5029 } else if (block_group
) {
5030 btrfs_put_block_group(block_group
);
5034 down_read(&space_info
->groups_sem
);
5035 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
5040 btrfs_get_block_group(block_group
);
5041 search_start
= block_group
->key
.objectid
;
5044 * this can happen if we end up cycling through all the
5045 * raid types, but we want to make sure we only allocate
5046 * for the proper type.
5048 if (!block_group_bits(block_group
, data
)) {
5049 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5050 BTRFS_BLOCK_GROUP_RAID1
|
5051 BTRFS_BLOCK_GROUP_RAID10
;
5054 * if they asked for extra copies and this block group
5055 * doesn't provide them, bail. This does allow us to
5056 * fill raid0 from raid1.
5058 if ((data
& extra
) && !(block_group
->flags
& extra
))
5063 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
5066 ret
= cache_block_group(block_group
, trans
,
5068 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5069 goto have_block_group
;
5071 free_percent
= btrfs_block_group_used(&block_group
->item
);
5072 free_percent
*= 100;
5073 free_percent
= div64_u64(free_percent
,
5074 block_group
->key
.offset
);
5075 free_percent
= 100 - free_percent
;
5076 if (free_percent
> ideal_cache_percent
&&
5077 likely(!block_group
->ro
)) {
5078 ideal_cache_offset
= block_group
->key
.objectid
;
5079 ideal_cache_percent
= free_percent
;
5083 * We only want to start kthread caching if we are at
5084 * the point where we will wait for caching to make
5085 * progress, or if our ideal search is over and we've
5086 * found somebody to start caching.
5088 if (loop
> LOOP_CACHING_NOWAIT
||
5089 (loop
> LOOP_FIND_IDEAL
&&
5090 atomic_read(&space_info
->caching_threads
) < 2)) {
5091 ret
= cache_block_group(block_group
, trans
,
5095 found_uncached_bg
= true;
5098 * If loop is set for cached only, try the next block
5101 if (loop
== LOOP_FIND_IDEAL
)
5105 cached
= block_group_cache_done(block_group
);
5106 if (unlikely(!cached
))
5107 found_uncached_bg
= true;
5109 if (unlikely(block_group
->ro
))
5113 * Ok we want to try and use the cluster allocator, so lets look
5114 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5115 * have tried the cluster allocator plenty of times at this
5116 * point and not have found anything, so we are likely way too
5117 * fragmented for the clustering stuff to find anything, so lets
5118 * just skip it and let the allocator find whatever block it can
5121 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5123 * the refill lock keeps out other
5124 * people trying to start a new cluster
5126 spin_lock(&last_ptr
->refill_lock
);
5127 if (last_ptr
->block_group
&&
5128 (last_ptr
->block_group
->ro
||
5129 !block_group_bits(last_ptr
->block_group
, data
))) {
5131 goto refill_cluster
;
5134 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5135 num_bytes
, search_start
);
5137 /* we have a block, we're done */
5138 spin_unlock(&last_ptr
->refill_lock
);
5142 spin_lock(&last_ptr
->lock
);
5144 * whoops, this cluster doesn't actually point to
5145 * this block group. Get a ref on the block
5146 * group is does point to and try again
5148 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5149 last_ptr
->block_group
!= block_group
) {
5151 btrfs_put_block_group(block_group
);
5152 block_group
= last_ptr
->block_group
;
5153 btrfs_get_block_group(block_group
);
5154 spin_unlock(&last_ptr
->lock
);
5155 spin_unlock(&last_ptr
->refill_lock
);
5158 search_start
= block_group
->key
.objectid
;
5160 * we know this block group is properly
5161 * in the list because
5162 * btrfs_remove_block_group, drops the
5163 * cluster before it removes the block
5164 * group from the list
5166 goto have_block_group
;
5168 spin_unlock(&last_ptr
->lock
);
5171 * this cluster didn't work out, free it and
5174 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5178 /* allocate a cluster in this block group */
5179 ret
= btrfs_find_space_cluster(trans
, root
,
5180 block_group
, last_ptr
,
5182 empty_cluster
+ empty_size
);
5185 * now pull our allocation out of this
5188 offset
= btrfs_alloc_from_cluster(block_group
,
5189 last_ptr
, num_bytes
,
5192 /* we found one, proceed */
5193 spin_unlock(&last_ptr
->refill_lock
);
5196 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5197 && !failed_cluster_refill
) {
5198 spin_unlock(&last_ptr
->refill_lock
);
5200 failed_cluster_refill
= true;
5201 wait_block_group_cache_progress(block_group
,
5202 num_bytes
+ empty_cluster
+ empty_size
);
5203 goto have_block_group
;
5207 * at this point we either didn't find a cluster
5208 * or we weren't able to allocate a block from our
5209 * cluster. Free the cluster we've been trying
5210 * to use, and go to the next block group
5212 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5213 spin_unlock(&last_ptr
->refill_lock
);
5217 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5218 num_bytes
, empty_size
);
5220 * If we didn't find a chunk, and we haven't failed on this
5221 * block group before, and this block group is in the middle of
5222 * caching and we are ok with waiting, then go ahead and wait
5223 * for progress to be made, and set failed_alloc to true.
5225 * If failed_alloc is true then we've already waited on this
5226 * block group once and should move on to the next block group.
5228 if (!offset
&& !failed_alloc
&& !cached
&&
5229 loop
> LOOP_CACHING_NOWAIT
) {
5230 wait_block_group_cache_progress(block_group
,
5231 num_bytes
+ empty_size
);
5232 failed_alloc
= true;
5233 goto have_block_group
;
5234 } else if (!offset
) {
5238 search_start
= stripe_align(root
, offset
);
5239 /* move on to the next group */
5240 if (search_start
+ num_bytes
>= search_end
) {
5241 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5245 /* move on to the next group */
5246 if (search_start
+ num_bytes
>
5247 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5248 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5252 ins
->objectid
= search_start
;
5253 ins
->offset
= num_bytes
;
5255 if (offset
< search_start
)
5256 btrfs_add_free_space(block_group
, offset
,
5257 search_start
- offset
);
5258 BUG_ON(offset
> search_start
);
5260 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
, 1,
5261 (data
& BTRFS_BLOCK_GROUP_DATA
));
5262 if (ret
== -EAGAIN
) {
5263 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5267 /* we are all good, lets return */
5268 ins
->objectid
= search_start
;
5269 ins
->offset
= num_bytes
;
5271 if (offset
< search_start
)
5272 btrfs_add_free_space(block_group
, offset
,
5273 search_start
- offset
);
5274 BUG_ON(offset
> search_start
);
5277 failed_cluster_refill
= false;
5278 failed_alloc
= false;
5279 BUG_ON(index
!= get_block_group_index(block_group
));
5280 btrfs_put_block_group(block_group
);
5282 up_read(&space_info
->groups_sem
);
5284 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5287 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5288 * for them to make caching progress. Also
5289 * determine the best possible bg to cache
5290 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5291 * caching kthreads as we move along
5292 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5293 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5294 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5297 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
5298 (found_uncached_bg
|| empty_size
|| empty_cluster
||
5299 allowed_chunk_alloc
)) {
5301 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5302 found_uncached_bg
= false;
5304 if (!ideal_cache_percent
&&
5305 atomic_read(&space_info
->caching_threads
))
5309 * 1 of the following 2 things have happened so far
5311 * 1) We found an ideal block group for caching that
5312 * is mostly full and will cache quickly, so we might
5313 * as well wait for it.
5315 * 2) We searched for cached only and we didn't find
5316 * anything, and we didn't start any caching kthreads
5317 * either, so chances are we will loop through and
5318 * start a couple caching kthreads, and then come back
5319 * around and just wait for them. This will be slower
5320 * because we will have 2 caching kthreads reading at
5321 * the same time when we could have just started one
5322 * and waited for it to get far enough to give us an
5323 * allocation, so go ahead and go to the wait caching
5326 loop
= LOOP_CACHING_WAIT
;
5327 search_start
= ideal_cache_offset
;
5328 ideal_cache_percent
= 0;
5330 } else if (loop
== LOOP_FIND_IDEAL
) {
5332 * Didn't find a uncached bg, wait on anything we find
5335 loop
= LOOP_CACHING_WAIT
;
5339 if (loop
< LOOP_CACHING_WAIT
) {
5344 if (loop
== LOOP_ALLOC_CHUNK
) {
5349 if (allowed_chunk_alloc
) {
5350 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5351 2 * 1024 * 1024, data
,
5352 CHUNK_ALLOC_LIMITED
);
5353 allowed_chunk_alloc
= 0;
5354 done_chunk_alloc
= 1;
5355 } else if (!done_chunk_alloc
&&
5356 space_info
->force_alloc
== CHUNK_ALLOC_NO_FORCE
) {
5357 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5360 if (loop
< LOOP_NO_EMPTY_SIZE
) {
5365 } else if (!ins
->objectid
) {
5369 /* we found what we needed */
5370 if (ins
->objectid
) {
5371 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
5372 trans
->block_group
= block_group
->key
.objectid
;
5374 btrfs_put_block_group(block_group
);
5381 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5382 int dump_block_groups
)
5384 struct btrfs_block_group_cache
*cache
;
5387 spin_lock(&info
->lock
);
5388 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
5389 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5390 info
->bytes_pinned
- info
->bytes_reserved
-
5391 info
->bytes_readonly
),
5392 (info
->full
) ? "" : "not ");
5393 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5394 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5395 (unsigned long long)info
->total_bytes
,
5396 (unsigned long long)info
->bytes_used
,
5397 (unsigned long long)info
->bytes_pinned
,
5398 (unsigned long long)info
->bytes_reserved
,
5399 (unsigned long long)info
->bytes_may_use
,
5400 (unsigned long long)info
->bytes_readonly
);
5401 spin_unlock(&info
->lock
);
5403 if (!dump_block_groups
)
5406 down_read(&info
->groups_sem
);
5408 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5409 spin_lock(&cache
->lock
);
5410 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5411 "%llu pinned %llu reserved\n",
5412 (unsigned long long)cache
->key
.objectid
,
5413 (unsigned long long)cache
->key
.offset
,
5414 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5415 (unsigned long long)cache
->pinned
,
5416 (unsigned long long)cache
->reserved
);
5417 btrfs_dump_free_space(cache
, bytes
);
5418 spin_unlock(&cache
->lock
);
5420 if (++index
< BTRFS_NR_RAID_TYPES
)
5422 up_read(&info
->groups_sem
);
5425 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5426 struct btrfs_root
*root
,
5427 u64 num_bytes
, u64 min_alloc_size
,
5428 u64 empty_size
, u64 hint_byte
,
5429 u64 search_end
, struct btrfs_key
*ins
,
5433 u64 search_start
= 0;
5435 data
= btrfs_get_alloc_profile(root
, data
);
5438 * the only place that sets empty_size is btrfs_realloc_node, which
5439 * is not called recursively on allocations
5441 if (empty_size
|| root
->ref_cows
)
5442 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5443 num_bytes
+ 2 * 1024 * 1024, data
,
5444 CHUNK_ALLOC_NO_FORCE
);
5446 WARN_ON(num_bytes
< root
->sectorsize
);
5447 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5448 search_start
, search_end
, hint_byte
,
5451 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5452 num_bytes
= num_bytes
>> 1;
5453 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5454 num_bytes
= max(num_bytes
, min_alloc_size
);
5455 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5456 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5459 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5460 struct btrfs_space_info
*sinfo
;
5462 sinfo
= __find_space_info(root
->fs_info
, data
);
5463 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5464 "wanted %llu\n", (unsigned long long)data
,
5465 (unsigned long long)num_bytes
);
5466 dump_space_info(sinfo
, num_bytes
, 1);
5469 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5474 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5476 struct btrfs_block_group_cache
*cache
;
5479 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5481 printk(KERN_ERR
"Unable to find block group for %llu\n",
5482 (unsigned long long)start
);
5486 if (btrfs_test_opt(root
, DISCARD
))
5487 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5489 btrfs_add_free_space(cache
, start
, len
);
5490 btrfs_update_reserved_bytes(cache
, len
, 0, 1);
5491 btrfs_put_block_group(cache
);
5493 trace_btrfs_reserved_extent_free(root
, start
, len
);
5498 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5499 struct btrfs_root
*root
,
5500 u64 parent
, u64 root_objectid
,
5501 u64 flags
, u64 owner
, u64 offset
,
5502 struct btrfs_key
*ins
, int ref_mod
)
5505 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5506 struct btrfs_extent_item
*extent_item
;
5507 struct btrfs_extent_inline_ref
*iref
;
5508 struct btrfs_path
*path
;
5509 struct extent_buffer
*leaf
;
5514 type
= BTRFS_SHARED_DATA_REF_KEY
;
5516 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5518 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5520 path
= btrfs_alloc_path();
5524 path
->leave_spinning
= 1;
5525 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5529 leaf
= path
->nodes
[0];
5530 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5531 struct btrfs_extent_item
);
5532 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5533 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5534 btrfs_set_extent_flags(leaf
, extent_item
,
5535 flags
| BTRFS_EXTENT_FLAG_DATA
);
5537 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5538 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5540 struct btrfs_shared_data_ref
*ref
;
5541 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5542 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5543 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5545 struct btrfs_extent_data_ref
*ref
;
5546 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5547 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5548 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5549 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5550 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5553 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5554 btrfs_free_path(path
);
5556 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5558 printk(KERN_ERR
"btrfs update block group failed for %llu "
5559 "%llu\n", (unsigned long long)ins
->objectid
,
5560 (unsigned long long)ins
->offset
);
5566 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5567 struct btrfs_root
*root
,
5568 u64 parent
, u64 root_objectid
,
5569 u64 flags
, struct btrfs_disk_key
*key
,
5570 int level
, struct btrfs_key
*ins
)
5573 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5574 struct btrfs_extent_item
*extent_item
;
5575 struct btrfs_tree_block_info
*block_info
;
5576 struct btrfs_extent_inline_ref
*iref
;
5577 struct btrfs_path
*path
;
5578 struct extent_buffer
*leaf
;
5579 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5581 path
= btrfs_alloc_path();
5584 path
->leave_spinning
= 1;
5585 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5589 leaf
= path
->nodes
[0];
5590 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5591 struct btrfs_extent_item
);
5592 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5593 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5594 btrfs_set_extent_flags(leaf
, extent_item
,
5595 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5596 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5598 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5599 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5601 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5603 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5604 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5605 BTRFS_SHARED_BLOCK_REF_KEY
);
5606 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5608 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5609 BTRFS_TREE_BLOCK_REF_KEY
);
5610 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5613 btrfs_mark_buffer_dirty(leaf
);
5614 btrfs_free_path(path
);
5616 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5618 printk(KERN_ERR
"btrfs update block group failed for %llu "
5619 "%llu\n", (unsigned long long)ins
->objectid
,
5620 (unsigned long long)ins
->offset
);
5626 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5627 struct btrfs_root
*root
,
5628 u64 root_objectid
, u64 owner
,
5629 u64 offset
, struct btrfs_key
*ins
)
5633 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5635 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5636 0, root_objectid
, owner
, offset
,
5637 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5642 * this is used by the tree logging recovery code. It records that
5643 * an extent has been allocated and makes sure to clear the free
5644 * space cache bits as well
5646 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5647 struct btrfs_root
*root
,
5648 u64 root_objectid
, u64 owner
, u64 offset
,
5649 struct btrfs_key
*ins
)
5652 struct btrfs_block_group_cache
*block_group
;
5653 struct btrfs_caching_control
*caching_ctl
;
5654 u64 start
= ins
->objectid
;
5655 u64 num_bytes
= ins
->offset
;
5657 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5658 cache_block_group(block_group
, trans
, NULL
, 0);
5659 caching_ctl
= get_caching_control(block_group
);
5662 BUG_ON(!block_group_cache_done(block_group
));
5663 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5666 mutex_lock(&caching_ctl
->mutex
);
5668 if (start
>= caching_ctl
->progress
) {
5669 ret
= add_excluded_extent(root
, start
, num_bytes
);
5671 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5672 ret
= btrfs_remove_free_space(block_group
,
5676 num_bytes
= caching_ctl
->progress
- start
;
5677 ret
= btrfs_remove_free_space(block_group
,
5681 start
= caching_ctl
->progress
;
5682 num_bytes
= ins
->objectid
+ ins
->offset
-
5683 caching_ctl
->progress
;
5684 ret
= add_excluded_extent(root
, start
, num_bytes
);
5688 mutex_unlock(&caching_ctl
->mutex
);
5689 put_caching_control(caching_ctl
);
5692 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5694 btrfs_put_block_group(block_group
);
5695 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5696 0, owner
, offset
, ins
, 1);
5700 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5701 struct btrfs_root
*root
,
5702 u64 bytenr
, u32 blocksize
,
5705 struct extent_buffer
*buf
;
5707 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5709 return ERR_PTR(-ENOMEM
);
5710 btrfs_set_header_generation(buf
, trans
->transid
);
5711 btrfs_set_buffer_lockdep_class(buf
, level
);
5712 btrfs_tree_lock(buf
);
5713 clean_tree_block(trans
, root
, buf
);
5715 btrfs_set_lock_blocking(buf
);
5716 btrfs_set_buffer_uptodate(buf
);
5718 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5720 * we allow two log transactions at a time, use different
5721 * EXENT bit to differentiate dirty pages.
5723 if (root
->log_transid
% 2 == 0)
5724 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5725 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5727 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5728 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5730 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5731 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5733 trans
->blocks_used
++;
5734 /* this returns a buffer locked for blocking */
5738 static struct btrfs_block_rsv
*
5739 use_block_rsv(struct btrfs_trans_handle
*trans
,
5740 struct btrfs_root
*root
, u32 blocksize
)
5742 struct btrfs_block_rsv
*block_rsv
;
5743 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5746 block_rsv
= get_block_rsv(trans
, root
);
5748 if (block_rsv
->size
== 0) {
5749 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5752 * If we couldn't reserve metadata bytes try and use some from
5753 * the global reserve.
5755 if (ret
&& block_rsv
!= global_rsv
) {
5756 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5759 return ERR_PTR(ret
);
5761 return ERR_PTR(ret
);
5766 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5771 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, blocksize
,
5774 spin_lock(&block_rsv
->lock
);
5775 block_rsv
->size
+= blocksize
;
5776 spin_unlock(&block_rsv
->lock
);
5778 } else if (ret
&& block_rsv
!= global_rsv
) {
5779 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5785 return ERR_PTR(-ENOSPC
);
5788 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5790 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5791 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5795 * finds a free extent and does all the dirty work required for allocation
5796 * returns the key for the extent through ins, and a tree buffer for
5797 * the first block of the extent through buf.
5799 * returns the tree buffer or NULL.
5801 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5802 struct btrfs_root
*root
, u32 blocksize
,
5803 u64 parent
, u64 root_objectid
,
5804 struct btrfs_disk_key
*key
, int level
,
5805 u64 hint
, u64 empty_size
)
5807 struct btrfs_key ins
;
5808 struct btrfs_block_rsv
*block_rsv
;
5809 struct extent_buffer
*buf
;
5814 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5815 if (IS_ERR(block_rsv
))
5816 return ERR_CAST(block_rsv
);
5818 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5819 empty_size
, hint
, (u64
)-1, &ins
, 0);
5821 unuse_block_rsv(block_rsv
, blocksize
);
5822 return ERR_PTR(ret
);
5825 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5827 BUG_ON(IS_ERR(buf
));
5829 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5831 parent
= ins
.objectid
;
5832 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5836 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5837 struct btrfs_delayed_extent_op
*extent_op
;
5838 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5841 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5843 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5844 extent_op
->flags_to_set
= flags
;
5845 extent_op
->update_key
= 1;
5846 extent_op
->update_flags
= 1;
5847 extent_op
->is_data
= 0;
5849 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5850 ins
.offset
, parent
, root_objectid
,
5851 level
, BTRFS_ADD_DELAYED_EXTENT
,
5858 struct walk_control
{
5859 u64 refs
[BTRFS_MAX_LEVEL
];
5860 u64 flags
[BTRFS_MAX_LEVEL
];
5861 struct btrfs_key update_progress
;
5871 #define DROP_REFERENCE 1
5872 #define UPDATE_BACKREF 2
5874 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5875 struct btrfs_root
*root
,
5876 struct walk_control
*wc
,
5877 struct btrfs_path
*path
)
5885 struct btrfs_key key
;
5886 struct extent_buffer
*eb
;
5891 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5892 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5893 wc
->reada_count
= max(wc
->reada_count
, 2);
5895 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5896 wc
->reada_count
= min_t(int, wc
->reada_count
,
5897 BTRFS_NODEPTRS_PER_BLOCK(root
));
5900 eb
= path
->nodes
[wc
->level
];
5901 nritems
= btrfs_header_nritems(eb
);
5902 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5904 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5905 if (nread
>= wc
->reada_count
)
5909 bytenr
= btrfs_node_blockptr(eb
, slot
);
5910 generation
= btrfs_node_ptr_generation(eb
, slot
);
5912 if (slot
== path
->slots
[wc
->level
])
5915 if (wc
->stage
== UPDATE_BACKREF
&&
5916 generation
<= root
->root_key
.offset
)
5919 /* We don't lock the tree block, it's OK to be racy here */
5920 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5925 if (wc
->stage
== DROP_REFERENCE
) {
5929 if (wc
->level
== 1 &&
5930 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5932 if (!wc
->update_ref
||
5933 generation
<= root
->root_key
.offset
)
5935 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5936 ret
= btrfs_comp_cpu_keys(&key
,
5937 &wc
->update_progress
);
5941 if (wc
->level
== 1 &&
5942 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5946 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5952 wc
->reada_slot
= slot
;
5956 * hepler to process tree block while walking down the tree.
5958 * when wc->stage == UPDATE_BACKREF, this function updates
5959 * back refs for pointers in the block.
5961 * NOTE: return value 1 means we should stop walking down.
5963 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5964 struct btrfs_root
*root
,
5965 struct btrfs_path
*path
,
5966 struct walk_control
*wc
, int lookup_info
)
5968 int level
= wc
->level
;
5969 struct extent_buffer
*eb
= path
->nodes
[level
];
5970 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5973 if (wc
->stage
== UPDATE_BACKREF
&&
5974 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5978 * when reference count of tree block is 1, it won't increase
5979 * again. once full backref flag is set, we never clear it.
5982 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5983 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5984 BUG_ON(!path
->locks
[level
]);
5985 ret
= btrfs_lookup_extent_info(trans
, root
,
5990 BUG_ON(wc
->refs
[level
] == 0);
5993 if (wc
->stage
== DROP_REFERENCE
) {
5994 if (wc
->refs
[level
] > 1)
5997 if (path
->locks
[level
] && !wc
->keep_locks
) {
5998 btrfs_tree_unlock(eb
);
5999 path
->locks
[level
] = 0;
6004 /* wc->stage == UPDATE_BACKREF */
6005 if (!(wc
->flags
[level
] & flag
)) {
6006 BUG_ON(!path
->locks
[level
]);
6007 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
6009 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6011 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
6014 wc
->flags
[level
] |= flag
;
6018 * the block is shared by multiple trees, so it's not good to
6019 * keep the tree lock
6021 if (path
->locks
[level
] && level
> 0) {
6022 btrfs_tree_unlock(eb
);
6023 path
->locks
[level
] = 0;
6029 * hepler to process tree block pointer.
6031 * when wc->stage == DROP_REFERENCE, this function checks
6032 * reference count of the block pointed to. if the block
6033 * is shared and we need update back refs for the subtree
6034 * rooted at the block, this function changes wc->stage to
6035 * UPDATE_BACKREF. if the block is shared and there is no
6036 * need to update back, this function drops the reference
6039 * NOTE: return value 1 means we should stop walking down.
6041 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
6042 struct btrfs_root
*root
,
6043 struct btrfs_path
*path
,
6044 struct walk_control
*wc
, int *lookup_info
)
6050 struct btrfs_key key
;
6051 struct extent_buffer
*next
;
6052 int level
= wc
->level
;
6056 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6057 path
->slots
[level
]);
6059 * if the lower level block was created before the snapshot
6060 * was created, we know there is no need to update back refs
6063 if (wc
->stage
== UPDATE_BACKREF
&&
6064 generation
<= root
->root_key
.offset
) {
6069 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6070 blocksize
= btrfs_level_size(root
, level
- 1);
6072 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6074 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6079 btrfs_tree_lock(next
);
6080 btrfs_set_lock_blocking(next
);
6082 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6083 &wc
->refs
[level
- 1],
6084 &wc
->flags
[level
- 1]);
6086 BUG_ON(wc
->refs
[level
- 1] == 0);
6089 if (wc
->stage
== DROP_REFERENCE
) {
6090 if (wc
->refs
[level
- 1] > 1) {
6092 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6095 if (!wc
->update_ref
||
6096 generation
<= root
->root_key
.offset
)
6099 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6100 path
->slots
[level
]);
6101 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6105 wc
->stage
= UPDATE_BACKREF
;
6106 wc
->shared_level
= level
- 1;
6110 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6114 if (!btrfs_buffer_uptodate(next
, generation
)) {
6115 btrfs_tree_unlock(next
);
6116 free_extent_buffer(next
);
6122 if (reada
&& level
== 1)
6123 reada_walk_down(trans
, root
, wc
, path
);
6124 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6127 btrfs_tree_lock(next
);
6128 btrfs_set_lock_blocking(next
);
6132 BUG_ON(level
!= btrfs_header_level(next
));
6133 path
->nodes
[level
] = next
;
6134 path
->slots
[level
] = 0;
6135 path
->locks
[level
] = 1;
6141 wc
->refs
[level
- 1] = 0;
6142 wc
->flags
[level
- 1] = 0;
6143 if (wc
->stage
== DROP_REFERENCE
) {
6144 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6145 parent
= path
->nodes
[level
]->start
;
6147 BUG_ON(root
->root_key
.objectid
!=
6148 btrfs_header_owner(path
->nodes
[level
]));
6152 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6153 root
->root_key
.objectid
, level
- 1, 0);
6156 btrfs_tree_unlock(next
);
6157 free_extent_buffer(next
);
6163 * hepler to process tree block while walking up the tree.
6165 * when wc->stage == DROP_REFERENCE, this function drops
6166 * reference count on the block.
6168 * when wc->stage == UPDATE_BACKREF, this function changes
6169 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6170 * to UPDATE_BACKREF previously while processing the block.
6172 * NOTE: return value 1 means we should stop walking up.
6174 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6175 struct btrfs_root
*root
,
6176 struct btrfs_path
*path
,
6177 struct walk_control
*wc
)
6180 int level
= wc
->level
;
6181 struct extent_buffer
*eb
= path
->nodes
[level
];
6184 if (wc
->stage
== UPDATE_BACKREF
) {
6185 BUG_ON(wc
->shared_level
< level
);
6186 if (level
< wc
->shared_level
)
6189 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6193 wc
->stage
= DROP_REFERENCE
;
6194 wc
->shared_level
= -1;
6195 path
->slots
[level
] = 0;
6198 * check reference count again if the block isn't locked.
6199 * we should start walking down the tree again if reference
6202 if (!path
->locks
[level
]) {
6204 btrfs_tree_lock(eb
);
6205 btrfs_set_lock_blocking(eb
);
6206 path
->locks
[level
] = 1;
6208 ret
= btrfs_lookup_extent_info(trans
, root
,
6213 BUG_ON(wc
->refs
[level
] == 0);
6214 if (wc
->refs
[level
] == 1) {
6215 btrfs_tree_unlock(eb
);
6216 path
->locks
[level
] = 0;
6222 /* wc->stage == DROP_REFERENCE */
6223 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6225 if (wc
->refs
[level
] == 1) {
6227 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6228 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6230 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6233 /* make block locked assertion in clean_tree_block happy */
6234 if (!path
->locks
[level
] &&
6235 btrfs_header_generation(eb
) == trans
->transid
) {
6236 btrfs_tree_lock(eb
);
6237 btrfs_set_lock_blocking(eb
);
6238 path
->locks
[level
] = 1;
6240 clean_tree_block(trans
, root
, eb
);
6243 if (eb
== root
->node
) {
6244 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6247 BUG_ON(root
->root_key
.objectid
!=
6248 btrfs_header_owner(eb
));
6250 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6251 parent
= path
->nodes
[level
+ 1]->start
;
6253 BUG_ON(root
->root_key
.objectid
!=
6254 btrfs_header_owner(path
->nodes
[level
+ 1]));
6257 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6259 wc
->refs
[level
] = 0;
6260 wc
->flags
[level
] = 0;
6264 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6265 struct btrfs_root
*root
,
6266 struct btrfs_path
*path
,
6267 struct walk_control
*wc
)
6269 int level
= wc
->level
;
6270 int lookup_info
= 1;
6273 while (level
>= 0) {
6274 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6281 if (path
->slots
[level
] >=
6282 btrfs_header_nritems(path
->nodes
[level
]))
6285 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6287 path
->slots
[level
]++;
6296 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6297 struct btrfs_root
*root
,
6298 struct btrfs_path
*path
,
6299 struct walk_control
*wc
, int max_level
)
6301 int level
= wc
->level
;
6304 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6305 while (level
< max_level
&& path
->nodes
[level
]) {
6307 if (path
->slots
[level
] + 1 <
6308 btrfs_header_nritems(path
->nodes
[level
])) {
6309 path
->slots
[level
]++;
6312 ret
= walk_up_proc(trans
, root
, path
, wc
);
6316 if (path
->locks
[level
]) {
6317 btrfs_tree_unlock(path
->nodes
[level
]);
6318 path
->locks
[level
] = 0;
6320 free_extent_buffer(path
->nodes
[level
]);
6321 path
->nodes
[level
] = NULL
;
6329 * drop a subvolume tree.
6331 * this function traverses the tree freeing any blocks that only
6332 * referenced by the tree.
6334 * when a shared tree block is found. this function decreases its
6335 * reference count by one. if update_ref is true, this function
6336 * also make sure backrefs for the shared block and all lower level
6337 * blocks are properly updated.
6339 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6340 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6342 struct btrfs_path
*path
;
6343 struct btrfs_trans_handle
*trans
;
6344 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6345 struct btrfs_root_item
*root_item
= &root
->root_item
;
6346 struct walk_control
*wc
;
6347 struct btrfs_key key
;
6352 path
= btrfs_alloc_path();
6355 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6358 trans
= btrfs_start_transaction(tree_root
, 0);
6359 BUG_ON(IS_ERR(trans
));
6362 trans
->block_rsv
= block_rsv
;
6364 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6365 level
= btrfs_header_level(root
->node
);
6366 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6367 btrfs_set_lock_blocking(path
->nodes
[level
]);
6368 path
->slots
[level
] = 0;
6369 path
->locks
[level
] = 1;
6370 memset(&wc
->update_progress
, 0,
6371 sizeof(wc
->update_progress
));
6373 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6374 memcpy(&wc
->update_progress
, &key
,
6375 sizeof(wc
->update_progress
));
6377 level
= root_item
->drop_level
;
6379 path
->lowest_level
= level
;
6380 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6381 path
->lowest_level
= 0;
6389 * unlock our path, this is safe because only this
6390 * function is allowed to delete this snapshot
6392 btrfs_unlock_up_safe(path
, 0);
6394 level
= btrfs_header_level(root
->node
);
6396 btrfs_tree_lock(path
->nodes
[level
]);
6397 btrfs_set_lock_blocking(path
->nodes
[level
]);
6399 ret
= btrfs_lookup_extent_info(trans
, root
,
6400 path
->nodes
[level
]->start
,
6401 path
->nodes
[level
]->len
,
6405 BUG_ON(wc
->refs
[level
] == 0);
6407 if (level
== root_item
->drop_level
)
6410 btrfs_tree_unlock(path
->nodes
[level
]);
6411 WARN_ON(wc
->refs
[level
] != 1);
6417 wc
->shared_level
= -1;
6418 wc
->stage
= DROP_REFERENCE
;
6419 wc
->update_ref
= update_ref
;
6421 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6424 ret
= walk_down_tree(trans
, root
, path
, wc
);
6430 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6437 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6441 if (wc
->stage
== DROP_REFERENCE
) {
6443 btrfs_node_key(path
->nodes
[level
],
6444 &root_item
->drop_progress
,
6445 path
->slots
[level
]);
6446 root_item
->drop_level
= level
;
6449 BUG_ON(wc
->level
== 0);
6450 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6451 ret
= btrfs_update_root(trans
, tree_root
,
6456 btrfs_end_transaction_throttle(trans
, tree_root
);
6457 trans
= btrfs_start_transaction(tree_root
, 0);
6458 BUG_ON(IS_ERR(trans
));
6460 trans
->block_rsv
= block_rsv
;
6463 btrfs_release_path(root
, path
);
6466 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6469 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6470 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6474 /* if we fail to delete the orphan item this time
6475 * around, it'll get picked up the next time.
6477 * The most common failure here is just -ENOENT.
6479 btrfs_del_orphan_item(trans
, tree_root
,
6480 root
->root_key
.objectid
);
6484 if (root
->in_radix
) {
6485 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6487 free_extent_buffer(root
->node
);
6488 free_extent_buffer(root
->commit_root
);
6492 btrfs_end_transaction_throttle(trans
, tree_root
);
6494 btrfs_free_path(path
);
6499 * drop subtree rooted at tree block 'node'.
6501 * NOTE: this function will unlock and release tree block 'node'
6503 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6504 struct btrfs_root
*root
,
6505 struct extent_buffer
*node
,
6506 struct extent_buffer
*parent
)
6508 struct btrfs_path
*path
;
6509 struct walk_control
*wc
;
6515 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6517 path
= btrfs_alloc_path();
6521 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6523 btrfs_free_path(path
);
6527 btrfs_assert_tree_locked(parent
);
6528 parent_level
= btrfs_header_level(parent
);
6529 extent_buffer_get(parent
);
6530 path
->nodes
[parent_level
] = parent
;
6531 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6533 btrfs_assert_tree_locked(node
);
6534 level
= btrfs_header_level(node
);
6535 path
->nodes
[level
] = node
;
6536 path
->slots
[level
] = 0;
6537 path
->locks
[level
] = 1;
6539 wc
->refs
[parent_level
] = 1;
6540 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6542 wc
->shared_level
= -1;
6543 wc
->stage
= DROP_REFERENCE
;
6546 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6549 wret
= walk_down_tree(trans
, root
, path
, wc
);
6555 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6563 btrfs_free_path(path
);
6568 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
6571 return min(last
, start
+ nr
- 1);
6574 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
6579 unsigned long first_index
;
6580 unsigned long last_index
;
6583 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6584 struct file_ra_state
*ra
;
6585 struct btrfs_ordered_extent
*ordered
;
6586 unsigned int total_read
= 0;
6587 unsigned int total_dirty
= 0;
6590 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
6594 mutex_lock(&inode
->i_mutex
);
6595 first_index
= start
>> PAGE_CACHE_SHIFT
;
6596 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
6598 /* make sure the dirty trick played by the caller work */
6599 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6600 first_index
, last_index
);
6604 file_ra_state_init(ra
, inode
->i_mapping
);
6606 for (i
= first_index
; i
<= last_index
; i
++) {
6607 if (total_read
% ra
->ra_pages
== 0) {
6608 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
6609 calc_ra(i
, last_index
, ra
->ra_pages
));
6613 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
6615 page
= grab_cache_page(inode
->i_mapping
, i
);
6620 if (!PageUptodate(page
)) {
6621 btrfs_readpage(NULL
, page
);
6623 if (!PageUptodate(page
)) {
6625 page_cache_release(page
);
6630 wait_on_page_writeback(page
);
6632 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
6633 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6634 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6636 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6638 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6640 page_cache_release(page
);
6641 btrfs_start_ordered_extent(inode
, ordered
, 1);
6642 btrfs_put_ordered_extent(ordered
);
6645 set_page_extent_mapped(page
);
6647 if (i
== first_index
)
6648 set_extent_bits(io_tree
, page_start
, page_end
,
6649 EXTENT_BOUNDARY
, GFP_NOFS
);
6650 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
6652 set_page_dirty(page
);
6655 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6657 page_cache_release(page
);
6662 mutex_unlock(&inode
->i_mutex
);
6663 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
6667 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
6668 struct btrfs_key
*extent_key
,
6671 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6672 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
6673 struct extent_map
*em
;
6674 u64 start
= extent_key
->objectid
- offset
;
6675 u64 end
= start
+ extent_key
->offset
- 1;
6677 em
= alloc_extent_map(GFP_NOFS
);
6681 em
->len
= extent_key
->offset
;
6682 em
->block_len
= extent_key
->offset
;
6683 em
->block_start
= extent_key
->objectid
;
6684 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6685 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6687 /* setup extent map to cheat btrfs_readpage */
6688 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6691 write_lock(&em_tree
->lock
);
6692 ret
= add_extent_mapping(em_tree
, em
);
6693 write_unlock(&em_tree
->lock
);
6694 if (ret
!= -EEXIST
) {
6695 free_extent_map(em
);
6698 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
6700 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6702 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
6705 struct btrfs_ref_path
{
6707 u64 nodes
[BTRFS_MAX_LEVEL
];
6709 u64 root_generation
;
6716 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
6717 u64 new_nodes
[BTRFS_MAX_LEVEL
];
6720 struct disk_extent
{
6731 static int is_cowonly_root(u64 root_objectid
)
6733 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
6734 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
6735 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
6736 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
6737 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6738 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
6743 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
6744 struct btrfs_root
*extent_root
,
6745 struct btrfs_ref_path
*ref_path
,
6748 struct extent_buffer
*leaf
;
6749 struct btrfs_path
*path
;
6750 struct btrfs_extent_ref
*ref
;
6751 struct btrfs_key key
;
6752 struct btrfs_key found_key
;
6758 path
= btrfs_alloc_path();
6763 ref_path
->lowest_level
= -1;
6764 ref_path
->current_level
= -1;
6765 ref_path
->shared_level
= -1;
6769 level
= ref_path
->current_level
- 1;
6770 while (level
>= -1) {
6772 if (level
< ref_path
->lowest_level
)
6776 bytenr
= ref_path
->nodes
[level
];
6778 bytenr
= ref_path
->extent_start
;
6779 BUG_ON(bytenr
== 0);
6781 parent
= ref_path
->nodes
[level
+ 1];
6782 ref_path
->nodes
[level
+ 1] = 0;
6783 ref_path
->current_level
= level
;
6784 BUG_ON(parent
== 0);
6786 key
.objectid
= bytenr
;
6787 key
.offset
= parent
+ 1;
6788 key
.type
= BTRFS_EXTENT_REF_KEY
;
6790 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6795 leaf
= path
->nodes
[0];
6796 nritems
= btrfs_header_nritems(leaf
);
6797 if (path
->slots
[0] >= nritems
) {
6798 ret
= btrfs_next_leaf(extent_root
, path
);
6803 leaf
= path
->nodes
[0];
6806 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6807 if (found_key
.objectid
== bytenr
&&
6808 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
6809 if (level
< ref_path
->shared_level
)
6810 ref_path
->shared_level
= level
;
6815 btrfs_release_path(extent_root
, path
);
6818 /* reached lowest level */
6822 level
= ref_path
->current_level
;
6823 while (level
< BTRFS_MAX_LEVEL
- 1) {
6827 bytenr
= ref_path
->nodes
[level
];
6829 bytenr
= ref_path
->extent_start
;
6831 BUG_ON(bytenr
== 0);
6833 key
.objectid
= bytenr
;
6835 key
.type
= BTRFS_EXTENT_REF_KEY
;
6837 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6841 leaf
= path
->nodes
[0];
6842 nritems
= btrfs_header_nritems(leaf
);
6843 if (path
->slots
[0] >= nritems
) {
6844 ret
= btrfs_next_leaf(extent_root
, path
);
6848 /* the extent was freed by someone */
6849 if (ref_path
->lowest_level
== level
)
6851 btrfs_release_path(extent_root
, path
);
6854 leaf
= path
->nodes
[0];
6857 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6858 if (found_key
.objectid
!= bytenr
||
6859 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
6860 /* the extent was freed by someone */
6861 if (ref_path
->lowest_level
== level
) {
6865 btrfs_release_path(extent_root
, path
);
6869 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
6870 struct btrfs_extent_ref
);
6871 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
6872 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
6874 level
= (int)ref_objectid
;
6875 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
6876 ref_path
->lowest_level
= level
;
6877 ref_path
->current_level
= level
;
6878 ref_path
->nodes
[level
] = bytenr
;
6880 WARN_ON(ref_objectid
!= level
);
6883 WARN_ON(level
!= -1);
6887 if (ref_path
->lowest_level
== level
) {
6888 ref_path
->owner_objectid
= ref_objectid
;
6889 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
6893 * the block is tree root or the block isn't in reference
6896 if (found_key
.objectid
== found_key
.offset
||
6897 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
6898 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6899 ref_path
->root_generation
=
6900 btrfs_ref_generation(leaf
, ref
);
6902 /* special reference from the tree log */
6903 ref_path
->nodes
[0] = found_key
.offset
;
6904 ref_path
->current_level
= 0;
6911 BUG_ON(ref_path
->nodes
[level
] != 0);
6912 ref_path
->nodes
[level
] = found_key
.offset
;
6913 ref_path
->current_level
= level
;
6916 * the reference was created in the running transaction,
6917 * no need to continue walking up.
6919 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
6920 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6921 ref_path
->root_generation
=
6922 btrfs_ref_generation(leaf
, ref
);
6927 btrfs_release_path(extent_root
, path
);
6930 /* reached max tree level, but no tree root found. */
6933 btrfs_free_path(path
);
6937 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
6938 struct btrfs_root
*extent_root
,
6939 struct btrfs_ref_path
*ref_path
,
6942 memset(ref_path
, 0, sizeof(*ref_path
));
6943 ref_path
->extent_start
= extent_start
;
6945 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
6948 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
6949 struct btrfs_root
*extent_root
,
6950 struct btrfs_ref_path
*ref_path
)
6952 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
6955 static noinline
int get_new_locations(struct inode
*reloc_inode
,
6956 struct btrfs_key
*extent_key
,
6957 u64 offset
, int no_fragment
,
6958 struct disk_extent
**extents
,
6961 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6962 struct btrfs_path
*path
;
6963 struct btrfs_file_extent_item
*fi
;
6964 struct extent_buffer
*leaf
;
6965 struct disk_extent
*exts
= *extents
;
6966 struct btrfs_key found_key
;
6971 int max
= *nr_extents
;
6974 WARN_ON(!no_fragment
&& *extents
);
6977 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6982 path
= btrfs_alloc_path();
6984 if (exts
!= *extents
)
6989 cur_pos
= extent_key
->objectid
- offset
;
6990 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
6991 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
7001 leaf
= path
->nodes
[0];
7002 nritems
= btrfs_header_nritems(leaf
);
7003 if (path
->slots
[0] >= nritems
) {
7004 ret
= btrfs_next_leaf(root
, path
);
7009 leaf
= path
->nodes
[0];
7012 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7013 if (found_key
.offset
!= cur_pos
||
7014 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
7015 found_key
.objectid
!= reloc_inode
->i_ino
)
7018 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7019 struct btrfs_file_extent_item
);
7020 if (btrfs_file_extent_type(leaf
, fi
) !=
7021 BTRFS_FILE_EXTENT_REG
||
7022 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
7026 struct disk_extent
*old
= exts
;
7028 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
7033 memcpy(exts
, old
, sizeof(*exts
) * nr
);
7034 if (old
!= *extents
)
7038 exts
[nr
].disk_bytenr
=
7039 btrfs_file_extent_disk_bytenr(leaf
, fi
);
7040 exts
[nr
].disk_num_bytes
=
7041 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7042 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
7043 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7044 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7045 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
7046 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
7047 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
7049 BUG_ON(exts
[nr
].offset
> 0);
7050 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
7051 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
7053 cur_pos
+= exts
[nr
].num_bytes
;
7056 if (cur_pos
+ offset
>= last_byte
)
7066 BUG_ON(cur_pos
+ offset
> last_byte
);
7067 if (cur_pos
+ offset
< last_byte
) {
7073 btrfs_free_path(path
);
7075 if (exts
!= *extents
)
7084 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
7085 struct btrfs_root
*root
,
7086 struct btrfs_path
*path
,
7087 struct btrfs_key
*extent_key
,
7088 struct btrfs_key
*leaf_key
,
7089 struct btrfs_ref_path
*ref_path
,
7090 struct disk_extent
*new_extents
,
7093 struct extent_buffer
*leaf
;
7094 struct btrfs_file_extent_item
*fi
;
7095 struct inode
*inode
= NULL
;
7096 struct btrfs_key key
;
7101 u64 search_end
= (u64
)-1;
7104 int extent_locked
= 0;
7108 memcpy(&key
, leaf_key
, sizeof(key
));
7109 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7110 if (key
.objectid
< ref_path
->owner_objectid
||
7111 (key
.objectid
== ref_path
->owner_objectid
&&
7112 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
7113 key
.objectid
= ref_path
->owner_objectid
;
7114 key
.type
= BTRFS_EXTENT_DATA_KEY
;
7120 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
7124 leaf
= path
->nodes
[0];
7125 nritems
= btrfs_header_nritems(leaf
);
7127 if (extent_locked
&& ret
> 0) {
7129 * the file extent item was modified by someone
7130 * before the extent got locked.
7132 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7133 lock_end
, GFP_NOFS
);
7137 if (path
->slots
[0] >= nritems
) {
7138 if (++nr_scaned
> 2)
7141 BUG_ON(extent_locked
);
7142 ret
= btrfs_next_leaf(root
, path
);
7147 leaf
= path
->nodes
[0];
7148 nritems
= btrfs_header_nritems(leaf
);
7151 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
7153 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7154 if ((key
.objectid
> ref_path
->owner_objectid
) ||
7155 (key
.objectid
== ref_path
->owner_objectid
&&
7156 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
7157 key
.offset
>= search_end
)
7161 if (inode
&& key
.objectid
!= inode
->i_ino
) {
7162 BUG_ON(extent_locked
);
7163 btrfs_release_path(root
, path
);
7164 mutex_unlock(&inode
->i_mutex
);
7170 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7175 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7176 struct btrfs_file_extent_item
);
7177 extent_type
= btrfs_file_extent_type(leaf
, fi
);
7178 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
7179 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
7180 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
7181 extent_key
->objectid
)) {
7187 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7188 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
7190 if (search_end
== (u64
)-1) {
7191 search_end
= key
.offset
- ext_offset
+
7192 btrfs_file_extent_ram_bytes(leaf
, fi
);
7195 if (!extent_locked
) {
7196 lock_start
= key
.offset
;
7197 lock_end
= lock_start
+ num_bytes
- 1;
7199 if (lock_start
> key
.offset
||
7200 lock_end
+ 1 < key
.offset
+ num_bytes
) {
7201 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7202 lock_start
, lock_end
, GFP_NOFS
);
7208 btrfs_release_path(root
, path
);
7210 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
7211 key
.objectid
, root
);
7212 if (inode
->i_state
& I_NEW
) {
7213 BTRFS_I(inode
)->root
= root
;
7214 BTRFS_I(inode
)->location
.objectid
=
7216 BTRFS_I(inode
)->location
.type
=
7217 BTRFS_INODE_ITEM_KEY
;
7218 BTRFS_I(inode
)->location
.offset
= 0;
7219 btrfs_read_locked_inode(inode
);
7220 unlock_new_inode(inode
);
7223 * some code call btrfs_commit_transaction while
7224 * holding the i_mutex, so we can't use mutex_lock
7227 if (is_bad_inode(inode
) ||
7228 !mutex_trylock(&inode
->i_mutex
)) {
7231 key
.offset
= (u64
)-1;
7236 if (!extent_locked
) {
7237 struct btrfs_ordered_extent
*ordered
;
7239 btrfs_release_path(root
, path
);
7241 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7242 lock_end
, GFP_NOFS
);
7243 ordered
= btrfs_lookup_first_ordered_extent(inode
,
7246 ordered
->file_offset
<= lock_end
&&
7247 ordered
->file_offset
+ ordered
->len
> lock_start
) {
7248 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7249 lock_start
, lock_end
, GFP_NOFS
);
7250 btrfs_start_ordered_extent(inode
, ordered
, 1);
7251 btrfs_put_ordered_extent(ordered
);
7252 key
.offset
+= num_bytes
;
7256 btrfs_put_ordered_extent(ordered
);
7262 if (nr_extents
== 1) {
7263 /* update extent pointer in place */
7264 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7265 new_extents
[0].disk_bytenr
);
7266 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7267 new_extents
[0].disk_num_bytes
);
7268 btrfs_mark_buffer_dirty(leaf
);
7270 btrfs_drop_extent_cache(inode
, key
.offset
,
7271 key
.offset
+ num_bytes
- 1, 0);
7273 ret
= btrfs_inc_extent_ref(trans
, root
,
7274 new_extents
[0].disk_bytenr
,
7275 new_extents
[0].disk_num_bytes
,
7277 root
->root_key
.objectid
,
7282 ret
= btrfs_free_extent(trans
, root
,
7283 extent_key
->objectid
,
7286 btrfs_header_owner(leaf
),
7287 btrfs_header_generation(leaf
),
7291 btrfs_release_path(root
, path
);
7292 key
.offset
+= num_bytes
;
7300 * drop old extent pointer at first, then insert the
7301 * new pointers one bye one
7303 btrfs_release_path(root
, path
);
7304 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
7305 key
.offset
+ num_bytes
,
7306 key
.offset
, &alloc_hint
);
7309 for (i
= 0; i
< nr_extents
; i
++) {
7310 if (ext_offset
>= new_extents
[i
].num_bytes
) {
7311 ext_offset
-= new_extents
[i
].num_bytes
;
7314 extent_len
= min(new_extents
[i
].num_bytes
-
7315 ext_offset
, num_bytes
);
7317 ret
= btrfs_insert_empty_item(trans
, root
,
7322 leaf
= path
->nodes
[0];
7323 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7324 struct btrfs_file_extent_item
);
7325 btrfs_set_file_extent_generation(leaf
, fi
,
7327 btrfs_set_file_extent_type(leaf
, fi
,
7328 BTRFS_FILE_EXTENT_REG
);
7329 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7330 new_extents
[i
].disk_bytenr
);
7331 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7332 new_extents
[i
].disk_num_bytes
);
7333 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
7334 new_extents
[i
].ram_bytes
);
7336 btrfs_set_file_extent_compression(leaf
, fi
,
7337 new_extents
[i
].compression
);
7338 btrfs_set_file_extent_encryption(leaf
, fi
,
7339 new_extents
[i
].encryption
);
7340 btrfs_set_file_extent_other_encoding(leaf
, fi
,
7341 new_extents
[i
].other_encoding
);
7343 btrfs_set_file_extent_num_bytes(leaf
, fi
,
7345 ext_offset
+= new_extents
[i
].offset
;
7346 btrfs_set_file_extent_offset(leaf
, fi
,
7348 btrfs_mark_buffer_dirty(leaf
);
7350 btrfs_drop_extent_cache(inode
, key
.offset
,
7351 key
.offset
+ extent_len
- 1, 0);
7353 ret
= btrfs_inc_extent_ref(trans
, root
,
7354 new_extents
[i
].disk_bytenr
,
7355 new_extents
[i
].disk_num_bytes
,
7357 root
->root_key
.objectid
,
7358 trans
->transid
, key
.objectid
);
7360 btrfs_release_path(root
, path
);
7362 inode_add_bytes(inode
, extent_len
);
7365 num_bytes
-= extent_len
;
7366 key
.offset
+= extent_len
;
7371 BUG_ON(i
>= nr_extents
);
7375 if (extent_locked
) {
7376 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7377 lock_end
, GFP_NOFS
);
7381 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
7382 key
.offset
>= search_end
)
7389 btrfs_release_path(root
, path
);
7391 mutex_unlock(&inode
->i_mutex
);
7392 if (extent_locked
) {
7393 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7394 lock_end
, GFP_NOFS
);
7401 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
7402 struct btrfs_root
*root
,
7403 struct extent_buffer
*buf
, u64 orig_start
)
7408 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
7409 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7411 level
= btrfs_header_level(buf
);
7413 struct btrfs_leaf_ref
*ref
;
7414 struct btrfs_leaf_ref
*orig_ref
;
7416 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
7420 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
7422 btrfs_free_leaf_ref(root
, orig_ref
);
7426 ref
->nritems
= orig_ref
->nritems
;
7427 memcpy(ref
->extents
, orig_ref
->extents
,
7428 sizeof(ref
->extents
[0]) * ref
->nritems
);
7430 btrfs_free_leaf_ref(root
, orig_ref
);
7432 ref
->root_gen
= trans
->transid
;
7433 ref
->bytenr
= buf
->start
;
7434 ref
->owner
= btrfs_header_owner(buf
);
7435 ref
->generation
= btrfs_header_generation(buf
);
7437 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
7439 btrfs_free_leaf_ref(root
, ref
);
7444 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
7445 struct extent_buffer
*leaf
,
7446 struct btrfs_block_group_cache
*group
,
7447 struct btrfs_root
*target_root
)
7449 struct btrfs_key key
;
7450 struct inode
*inode
= NULL
;
7451 struct btrfs_file_extent_item
*fi
;
7452 struct extent_state
*cached_state
= NULL
;
7454 u64 skip_objectid
= 0;
7458 nritems
= btrfs_header_nritems(leaf
);
7459 for (i
= 0; i
< nritems
; i
++) {
7460 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7461 if (key
.objectid
== skip_objectid
||
7462 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7464 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7465 if (btrfs_file_extent_type(leaf
, fi
) ==
7466 BTRFS_FILE_EXTENT_INLINE
)
7468 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
7470 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
7472 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
7473 key
.objectid
, target_root
, 1);
7476 skip_objectid
= key
.objectid
;
7479 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7481 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7482 key
.offset
+ num_bytes
- 1, 0, &cached_state
,
7484 btrfs_drop_extent_cache(inode
, key
.offset
,
7485 key
.offset
+ num_bytes
- 1, 1);
7486 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7487 key
.offset
+ num_bytes
- 1, &cached_state
,
7495 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
7496 struct btrfs_root
*root
,
7497 struct extent_buffer
*leaf
,
7498 struct btrfs_block_group_cache
*group
,
7499 struct inode
*reloc_inode
)
7501 struct btrfs_key key
;
7502 struct btrfs_key extent_key
;
7503 struct btrfs_file_extent_item
*fi
;
7504 struct btrfs_leaf_ref
*ref
;
7505 struct disk_extent
*new_extent
;
7514 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
7518 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
7522 nritems
= btrfs_header_nritems(leaf
);
7523 for (i
= 0; i
< nritems
; i
++) {
7524 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7525 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
7527 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7528 if (btrfs_file_extent_type(leaf
, fi
) ==
7529 BTRFS_FILE_EXTENT_INLINE
)
7531 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7532 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7537 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
7538 bytenr
+ num_bytes
<= group
->key
.objectid
)
7541 extent_key
.objectid
= bytenr
;
7542 extent_key
.offset
= num_bytes
;
7543 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7545 ret
= get_new_locations(reloc_inode
, &extent_key
,
7546 group
->key
.objectid
, 1,
7547 &new_extent
, &nr_extent
);
7552 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
7553 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
7554 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
7555 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
7557 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7558 new_extent
->disk_bytenr
);
7559 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7560 new_extent
->disk_num_bytes
);
7561 btrfs_mark_buffer_dirty(leaf
);
7563 ret
= btrfs_inc_extent_ref(trans
, root
,
7564 new_extent
->disk_bytenr
,
7565 new_extent
->disk_num_bytes
,
7567 root
->root_key
.objectid
,
7568 trans
->transid
, key
.objectid
);
7571 ret
= btrfs_free_extent(trans
, root
,
7572 bytenr
, num_bytes
, leaf
->start
,
7573 btrfs_header_owner(leaf
),
7574 btrfs_header_generation(leaf
),
7580 BUG_ON(ext_index
+ 1 != ref
->nritems
);
7581 btrfs_free_leaf_ref(root
, ref
);
7585 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
7586 struct btrfs_root
*root
)
7588 struct btrfs_root
*reloc_root
;
7591 if (root
->reloc_root
) {
7592 reloc_root
= root
->reloc_root
;
7593 root
->reloc_root
= NULL
;
7594 list_add(&reloc_root
->dead_list
,
7595 &root
->fs_info
->dead_reloc_roots
);
7597 btrfs_set_root_bytenr(&reloc_root
->root_item
,
7598 reloc_root
->node
->start
);
7599 btrfs_set_root_level(&root
->root_item
,
7600 btrfs_header_level(reloc_root
->node
));
7601 memset(&reloc_root
->root_item
.drop_progress
, 0,
7602 sizeof(struct btrfs_disk_key
));
7603 reloc_root
->root_item
.drop_level
= 0;
7605 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
7606 &reloc_root
->root_key
,
7607 &reloc_root
->root_item
);
7613 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
7615 struct btrfs_trans_handle
*trans
;
7616 struct btrfs_root
*reloc_root
;
7617 struct btrfs_root
*prev_root
= NULL
;
7618 struct list_head dead_roots
;
7622 INIT_LIST_HEAD(&dead_roots
);
7623 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
7625 while (!list_empty(&dead_roots
)) {
7626 reloc_root
= list_entry(dead_roots
.prev
,
7627 struct btrfs_root
, dead_list
);
7628 list_del_init(&reloc_root
->dead_list
);
7630 BUG_ON(reloc_root
->commit_root
!= NULL
);
7632 trans
= btrfs_join_transaction(root
, 1);
7633 BUG_ON(IS_ERR(trans
));
7635 mutex_lock(&root
->fs_info
->drop_mutex
);
7636 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
7639 mutex_unlock(&root
->fs_info
->drop_mutex
);
7641 nr
= trans
->blocks_used
;
7642 ret
= btrfs_end_transaction(trans
, root
);
7644 btrfs_btree_balance_dirty(root
, nr
);
7647 free_extent_buffer(reloc_root
->node
);
7649 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
7650 &reloc_root
->root_key
);
7652 mutex_unlock(&root
->fs_info
->drop_mutex
);
7654 nr
= trans
->blocks_used
;
7655 ret
= btrfs_end_transaction(trans
, root
);
7657 btrfs_btree_balance_dirty(root
, nr
);
7660 prev_root
= reloc_root
;
7663 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
7669 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
7671 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
7675 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
7677 struct btrfs_root
*reloc_root
;
7678 struct btrfs_trans_handle
*trans
;
7679 struct btrfs_key location
;
7683 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7684 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
7686 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
7687 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7690 trans
= btrfs_start_transaction(root
, 1);
7691 BUG_ON(IS_ERR(trans
));
7692 ret
= btrfs_commit_transaction(trans
, root
);
7696 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
7697 location
.offset
= (u64
)-1;
7698 location
.type
= BTRFS_ROOT_ITEM_KEY
;
7700 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
7701 BUG_ON(!reloc_root
);
7702 ret
= btrfs_orphan_cleanup(reloc_root
);
7707 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
7708 struct btrfs_root
*root
)
7710 struct btrfs_root
*reloc_root
;
7711 struct extent_buffer
*eb
;
7712 struct btrfs_root_item
*root_item
;
7713 struct btrfs_key root_key
;
7716 BUG_ON(!root
->ref_cows
);
7717 if (root
->reloc_root
)
7720 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
7724 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
7725 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
7728 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
7729 root_key
.offset
= root
->root_key
.objectid
;
7730 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7732 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
7733 btrfs_set_root_refs(root_item
, 0);
7734 btrfs_set_root_bytenr(root_item
, eb
->start
);
7735 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
7736 btrfs_set_root_generation(root_item
, trans
->transid
);
7738 btrfs_tree_unlock(eb
);
7739 free_extent_buffer(eb
);
7741 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
7742 &root_key
, root_item
);
7746 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
7748 BUG_ON(IS_ERR(reloc_root
));
7749 reloc_root
->last_trans
= trans
->transid
;
7750 reloc_root
->commit_root
= NULL
;
7751 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
7753 root
->reloc_root
= reloc_root
;
7758 * Core function of space balance.
7760 * The idea is using reloc trees to relocate tree blocks in reference
7761 * counted roots. There is one reloc tree for each subvol, and all
7762 * reloc trees share same root key objectid. Reloc trees are snapshots
7763 * of the latest committed roots of subvols (root->commit_root).
7765 * To relocate a tree block referenced by a subvol, there are two steps.
7766 * COW the block through subvol's reloc tree, then update block pointer
7767 * in the subvol to point to the new block. Since all reloc trees share
7768 * same root key objectid, doing special handing for tree blocks owned
7769 * by them is easy. Once a tree block has been COWed in one reloc tree,
7770 * we can use the resulting new block directly when the same block is
7771 * required to COW again through other reloc trees. By this way, relocated
7772 * tree blocks are shared between reloc trees, so they are also shared
7775 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
7776 struct btrfs_root
*root
,
7777 struct btrfs_path
*path
,
7778 struct btrfs_key
*first_key
,
7779 struct btrfs_ref_path
*ref_path
,
7780 struct btrfs_block_group_cache
*group
,
7781 struct inode
*reloc_inode
)
7783 struct btrfs_root
*reloc_root
;
7784 struct extent_buffer
*eb
= NULL
;
7785 struct btrfs_key
*keys
;
7789 int lowest_level
= 0;
7792 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
7793 lowest_level
= ref_path
->owner_objectid
;
7795 if (!root
->ref_cows
) {
7796 path
->lowest_level
= lowest_level
;
7797 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
7799 path
->lowest_level
= 0;
7800 btrfs_release_path(root
, path
);
7804 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7805 ret
= init_reloc_tree(trans
, root
);
7807 reloc_root
= root
->reloc_root
;
7809 shared_level
= ref_path
->shared_level
;
7810 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
7812 keys
= ref_path
->node_keys
;
7813 nodes
= ref_path
->new_nodes
;
7814 memset(&keys
[shared_level
+ 1], 0,
7815 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7816 memset(&nodes
[shared_level
+ 1], 0,
7817 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7819 if (nodes
[lowest_level
] == 0) {
7820 path
->lowest_level
= lowest_level
;
7821 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7824 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
7825 eb
= path
->nodes
[level
];
7826 if (!eb
|| eb
== reloc_root
->node
)
7828 nodes
[level
] = eb
->start
;
7830 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
7832 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
7835 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7836 eb
= path
->nodes
[0];
7837 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
7838 group
, reloc_inode
);
7841 btrfs_release_path(reloc_root
, path
);
7843 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
7849 * replace tree blocks in the fs tree with tree blocks in
7852 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
7855 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7856 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7859 extent_buffer_get(path
->nodes
[0]);
7860 eb
= path
->nodes
[0];
7861 btrfs_release_path(reloc_root
, path
);
7862 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
7864 free_extent_buffer(eb
);
7867 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7868 path
->lowest_level
= 0;
7872 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
7873 struct btrfs_root
*root
,
7874 struct btrfs_path
*path
,
7875 struct btrfs_key
*first_key
,
7876 struct btrfs_ref_path
*ref_path
)
7880 ret
= relocate_one_path(trans
, root
, path
, first_key
,
7881 ref_path
, NULL
, NULL
);
7887 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
7888 struct btrfs_root
*extent_root
,
7889 struct btrfs_path
*path
,
7890 struct btrfs_key
*extent_key
)
7894 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
7897 ret
= btrfs_del_item(trans
, extent_root
, path
);
7899 btrfs_release_path(extent_root
, path
);
7903 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
7904 struct btrfs_ref_path
*ref_path
)
7906 struct btrfs_key root_key
;
7908 root_key
.objectid
= ref_path
->root_objectid
;
7909 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7910 if (is_cowonly_root(ref_path
->root_objectid
))
7911 root_key
.offset
= 0;
7913 root_key
.offset
= (u64
)-1;
7915 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
7918 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
7919 struct btrfs_path
*path
,
7920 struct btrfs_key
*extent_key
,
7921 struct btrfs_block_group_cache
*group
,
7922 struct inode
*reloc_inode
, int pass
)
7924 struct btrfs_trans_handle
*trans
;
7925 struct btrfs_root
*found_root
;
7926 struct btrfs_ref_path
*ref_path
= NULL
;
7927 struct disk_extent
*new_extents
= NULL
;
7932 struct btrfs_key first_key
;
7936 trans
= btrfs_start_transaction(extent_root
, 1);
7937 BUG_ON(IS_ERR(trans
));
7939 if (extent_key
->objectid
== 0) {
7940 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
7944 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
7950 for (loops
= 0; ; loops
++) {
7952 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
7953 extent_key
->objectid
);
7955 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
7962 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
7963 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
7966 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
7967 BUG_ON(!found_root
);
7969 * for reference counted tree, only process reference paths
7970 * rooted at the latest committed root.
7972 if (found_root
->ref_cows
&&
7973 ref_path
->root_generation
!= found_root
->root_key
.offset
)
7976 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7979 * copy data extents to new locations
7981 u64 group_start
= group
->key
.objectid
;
7982 ret
= relocate_data_extent(reloc_inode
,
7991 level
= ref_path
->owner_objectid
;
7994 if (prev_block
!= ref_path
->nodes
[level
]) {
7995 struct extent_buffer
*eb
;
7996 u64 block_start
= ref_path
->nodes
[level
];
7997 u64 block_size
= btrfs_level_size(found_root
, level
);
7999 eb
= read_tree_block(found_root
, block_start
,
8005 btrfs_tree_lock(eb
);
8006 BUG_ON(level
!= btrfs_header_level(eb
));
8009 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
8011 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
8013 btrfs_tree_unlock(eb
);
8014 free_extent_buffer(eb
);
8015 prev_block
= block_start
;
8018 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
8019 btrfs_record_root_in_trans(found_root
);
8020 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
8021 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
8023 * try to update data extent references while
8024 * keeping metadata shared between snapshots.
8027 ret
= relocate_one_path(trans
, found_root
,
8028 path
, &first_key
, ref_path
,
8029 group
, reloc_inode
);
8035 * use fallback method to process the remaining
8039 u64 group_start
= group
->key
.objectid
;
8040 new_extents
= kmalloc(sizeof(*new_extents
),
8043 ret
= get_new_locations(reloc_inode
,
8051 ret
= replace_one_extent(trans
, found_root
,
8053 &first_key
, ref_path
,
8054 new_extents
, nr_extents
);
8056 ret
= relocate_tree_block(trans
, found_root
, path
,
8057 &first_key
, ref_path
);
8064 btrfs_end_transaction(trans
, extent_root
);
8071 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8074 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8075 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8078 * we add in the count of missing devices because we want
8079 * to make sure that any RAID levels on a degraded FS
8080 * continue to be honored.
8082 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
8083 root
->fs_info
->fs_devices
->missing_devices
;
8085 if (num_devices
== 1) {
8086 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8087 stripped
= flags
& ~stripped
;
8089 /* turn raid0 into single device chunks */
8090 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8093 /* turn mirroring into duplication */
8094 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8095 BTRFS_BLOCK_GROUP_RAID10
))
8096 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8099 /* they already had raid on here, just return */
8100 if (flags
& stripped
)
8103 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8104 stripped
= flags
& ~stripped
;
8106 /* switch duplicated blocks with raid1 */
8107 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8108 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8110 /* turn single device chunks into raid0 */
8111 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
8116 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
)
8118 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8125 spin_lock(&sinfo
->lock
);
8126 spin_lock(&cache
->lock
);
8127 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8128 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8130 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8131 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
8132 cache
->reserved_pinned
+ num_bytes
<= sinfo
->total_bytes
) {
8133 sinfo
->bytes_readonly
+= num_bytes
;
8134 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
8135 cache
->reserved_pinned
= 0;
8140 spin_unlock(&cache
->lock
);
8141 spin_unlock(&sinfo
->lock
);
8145 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8146 struct btrfs_block_group_cache
*cache
)
8149 struct btrfs_trans_handle
*trans
;
8155 trans
= btrfs_join_transaction(root
, 1);
8156 BUG_ON(IS_ERR(trans
));
8158 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8159 if (alloc_flags
!= cache
->flags
)
8160 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
8163 ret
= set_block_group_ro(cache
);
8166 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8167 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
8171 ret
= set_block_group_ro(cache
);
8173 btrfs_end_transaction(trans
, root
);
8177 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8178 struct btrfs_root
*root
, u64 type
)
8180 u64 alloc_flags
= get_alloc_profile(root
, type
);
8181 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
8186 * helper to account the unused space of all the readonly block group in the
8187 * list. takes mirrors into account.
8189 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
8191 struct btrfs_block_group_cache
*block_group
;
8195 list_for_each_entry(block_group
, groups_list
, list
) {
8196 spin_lock(&block_group
->lock
);
8198 if (!block_group
->ro
) {
8199 spin_unlock(&block_group
->lock
);
8203 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8204 BTRFS_BLOCK_GROUP_RAID10
|
8205 BTRFS_BLOCK_GROUP_DUP
))
8210 free_bytes
+= (block_group
->key
.offset
-
8211 btrfs_block_group_used(&block_group
->item
)) *
8214 spin_unlock(&block_group
->lock
);
8221 * helper to account the unused space of all the readonly block group in the
8222 * space_info. takes mirrors into account.
8224 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8229 spin_lock(&sinfo
->lock
);
8231 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
8232 if (!list_empty(&sinfo
->block_groups
[i
]))
8233 free_bytes
+= __btrfs_get_ro_block_group_free_space(
8234 &sinfo
->block_groups
[i
]);
8236 spin_unlock(&sinfo
->lock
);
8241 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
8242 struct btrfs_block_group_cache
*cache
)
8244 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8249 spin_lock(&sinfo
->lock
);
8250 spin_lock(&cache
->lock
);
8251 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8252 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8253 sinfo
->bytes_readonly
-= num_bytes
;
8255 spin_unlock(&cache
->lock
);
8256 spin_unlock(&sinfo
->lock
);
8261 * checks to see if its even possible to relocate this block group.
8263 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8264 * ok to go ahead and try.
8266 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8268 struct btrfs_block_group_cache
*block_group
;
8269 struct btrfs_space_info
*space_info
;
8270 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8271 struct btrfs_device
*device
;
8275 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8277 /* odd, couldn't find the block group, leave it alone */
8281 /* no bytes used, we're good */
8282 if (!btrfs_block_group_used(&block_group
->item
))
8285 space_info
= block_group
->space_info
;
8286 spin_lock(&space_info
->lock
);
8288 full
= space_info
->full
;
8291 * if this is the last block group we have in this space, we can't
8292 * relocate it unless we're able to allocate a new chunk below.
8294 * Otherwise, we need to make sure we have room in the space to handle
8295 * all of the extents from this block group. If we can, we're good
8297 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8298 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8299 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8300 btrfs_block_group_used(&block_group
->item
) <
8301 space_info
->total_bytes
)) {
8302 spin_unlock(&space_info
->lock
);
8305 spin_unlock(&space_info
->lock
);
8308 * ok we don't have enough space, but maybe we have free space on our
8309 * devices to allocate new chunks for relocation, so loop through our
8310 * alloc devices and guess if we have enough space. However, if we
8311 * were marked as full, then we know there aren't enough chunks, and we
8318 mutex_lock(&root
->fs_info
->chunk_mutex
);
8319 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8320 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
8324 * check to make sure we can actually find a chunk with enough
8325 * space to fit our block group in.
8327 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
8328 ret
= find_free_dev_extent(NULL
, device
, min_free
,
8335 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8337 btrfs_put_block_group(block_group
);
8341 static int find_first_block_group(struct btrfs_root
*root
,
8342 struct btrfs_path
*path
, struct btrfs_key
*key
)
8345 struct btrfs_key found_key
;
8346 struct extent_buffer
*leaf
;
8349 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8354 slot
= path
->slots
[0];
8355 leaf
= path
->nodes
[0];
8356 if (slot
>= btrfs_header_nritems(leaf
)) {
8357 ret
= btrfs_next_leaf(root
, path
);
8364 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8366 if (found_key
.objectid
>= key
->objectid
&&
8367 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8377 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8379 struct btrfs_block_group_cache
*block_group
;
8383 struct inode
*inode
;
8385 block_group
= btrfs_lookup_first_block_group(info
, last
);
8386 while (block_group
) {
8387 spin_lock(&block_group
->lock
);
8388 if (block_group
->iref
)
8390 spin_unlock(&block_group
->lock
);
8391 block_group
= next_block_group(info
->tree_root
,
8401 inode
= block_group
->inode
;
8402 block_group
->iref
= 0;
8403 block_group
->inode
= NULL
;
8404 spin_unlock(&block_group
->lock
);
8406 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8407 btrfs_put_block_group(block_group
);
8411 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8413 struct btrfs_block_group_cache
*block_group
;
8414 struct btrfs_space_info
*space_info
;
8415 struct btrfs_caching_control
*caching_ctl
;
8418 down_write(&info
->extent_commit_sem
);
8419 while (!list_empty(&info
->caching_block_groups
)) {
8420 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8421 struct btrfs_caching_control
, list
);
8422 list_del(&caching_ctl
->list
);
8423 put_caching_control(caching_ctl
);
8425 up_write(&info
->extent_commit_sem
);
8427 spin_lock(&info
->block_group_cache_lock
);
8428 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8429 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8431 rb_erase(&block_group
->cache_node
,
8432 &info
->block_group_cache_tree
);
8433 spin_unlock(&info
->block_group_cache_lock
);
8435 down_write(&block_group
->space_info
->groups_sem
);
8436 list_del(&block_group
->list
);
8437 up_write(&block_group
->space_info
->groups_sem
);
8439 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8440 wait_block_group_cache_done(block_group
);
8443 * We haven't cached this block group, which means we could
8444 * possibly have excluded extents on this block group.
8446 if (block_group
->cached
== BTRFS_CACHE_NO
)
8447 free_excluded_extents(info
->extent_root
, block_group
);
8449 btrfs_remove_free_space_cache(block_group
);
8450 btrfs_put_block_group(block_group
);
8452 spin_lock(&info
->block_group_cache_lock
);
8454 spin_unlock(&info
->block_group_cache_lock
);
8456 /* now that all the block groups are freed, go through and
8457 * free all the space_info structs. This is only called during
8458 * the final stages of unmount, and so we know nobody is
8459 * using them. We call synchronize_rcu() once before we start,
8460 * just to be on the safe side.
8464 release_global_block_rsv(info
);
8466 while(!list_empty(&info
->space_info
)) {
8467 space_info
= list_entry(info
->space_info
.next
,
8468 struct btrfs_space_info
,
8470 if (space_info
->bytes_pinned
> 0 ||
8471 space_info
->bytes_reserved
> 0) {
8473 dump_space_info(space_info
, 0, 0);
8475 list_del(&space_info
->list
);
8481 static void __link_block_group(struct btrfs_space_info
*space_info
,
8482 struct btrfs_block_group_cache
*cache
)
8484 int index
= get_block_group_index(cache
);
8486 down_write(&space_info
->groups_sem
);
8487 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8488 up_write(&space_info
->groups_sem
);
8491 int btrfs_read_block_groups(struct btrfs_root
*root
)
8493 struct btrfs_path
*path
;
8495 struct btrfs_block_group_cache
*cache
;
8496 struct btrfs_fs_info
*info
= root
->fs_info
;
8497 struct btrfs_space_info
*space_info
;
8498 struct btrfs_key key
;
8499 struct btrfs_key found_key
;
8500 struct extent_buffer
*leaf
;
8504 root
= info
->extent_root
;
8507 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8508 path
= btrfs_alloc_path();
8512 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
8513 if (cache_gen
!= 0 &&
8514 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
8516 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8518 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
8519 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
8522 ret
= find_first_block_group(root
, path
, &key
);
8527 leaf
= path
->nodes
[0];
8528 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8529 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8535 atomic_set(&cache
->count
, 1);
8536 spin_lock_init(&cache
->lock
);
8537 spin_lock_init(&cache
->tree_lock
);
8538 cache
->fs_info
= info
;
8539 INIT_LIST_HEAD(&cache
->list
);
8540 INIT_LIST_HEAD(&cache
->cluster_list
);
8543 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8546 * we only want to have 32k of ram per block group for keeping
8547 * track of free space, and if we pass 1/2 of that we want to
8548 * start converting things over to using bitmaps
8550 cache
->extents_thresh
= ((1024 * 32) / 2) /
8551 sizeof(struct btrfs_free_space
);
8553 read_extent_buffer(leaf
, &cache
->item
,
8554 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8555 sizeof(cache
->item
));
8556 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
8558 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8559 btrfs_release_path(root
, path
);
8560 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8561 cache
->sectorsize
= root
->sectorsize
;
8564 * We need to exclude the super stripes now so that the space
8565 * info has super bytes accounted for, otherwise we'll think
8566 * we have more space than we actually do.
8568 exclude_super_stripes(root
, cache
);
8571 * check for two cases, either we are full, and therefore
8572 * don't need to bother with the caching work since we won't
8573 * find any space, or we are empty, and we can just add all
8574 * the space in and be done with it. This saves us _alot_ of
8575 * time, particularly in the full case.
8577 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8578 cache
->last_byte_to_unpin
= (u64
)-1;
8579 cache
->cached
= BTRFS_CACHE_FINISHED
;
8580 free_excluded_extents(root
, cache
);
8581 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8582 cache
->last_byte_to_unpin
= (u64
)-1;
8583 cache
->cached
= BTRFS_CACHE_FINISHED
;
8584 add_new_free_space(cache
, root
->fs_info
,
8586 found_key
.objectid
+
8588 free_excluded_extents(root
, cache
);
8591 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8592 btrfs_block_group_used(&cache
->item
),
8595 cache
->space_info
= space_info
;
8596 spin_lock(&cache
->space_info
->lock
);
8597 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8598 spin_unlock(&cache
->space_info
->lock
);
8600 __link_block_group(space_info
, cache
);
8602 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8605 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8606 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8607 set_block_group_ro(cache
);
8610 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8611 if (!(get_alloc_profile(root
, space_info
->flags
) &
8612 (BTRFS_BLOCK_GROUP_RAID10
|
8613 BTRFS_BLOCK_GROUP_RAID1
|
8614 BTRFS_BLOCK_GROUP_DUP
)))
8617 * avoid allocating from un-mirrored block group if there are
8618 * mirrored block groups.
8620 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
8621 set_block_group_ro(cache
);
8622 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
8623 set_block_group_ro(cache
);
8626 init_global_block_rsv(info
);
8629 btrfs_free_path(path
);
8633 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8634 struct btrfs_root
*root
, u64 bytes_used
,
8635 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8639 struct btrfs_root
*extent_root
;
8640 struct btrfs_block_group_cache
*cache
;
8642 extent_root
= root
->fs_info
->extent_root
;
8644 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8646 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8650 cache
->key
.objectid
= chunk_offset
;
8651 cache
->key
.offset
= size
;
8652 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8653 cache
->sectorsize
= root
->sectorsize
;
8654 cache
->fs_info
= root
->fs_info
;
8657 * we only want to have 32k of ram per block group for keeping track
8658 * of free space, and if we pass 1/2 of that we want to start
8659 * converting things over to using bitmaps
8661 cache
->extents_thresh
= ((1024 * 32) / 2) /
8662 sizeof(struct btrfs_free_space
);
8663 atomic_set(&cache
->count
, 1);
8664 spin_lock_init(&cache
->lock
);
8665 spin_lock_init(&cache
->tree_lock
);
8666 INIT_LIST_HEAD(&cache
->list
);
8667 INIT_LIST_HEAD(&cache
->cluster_list
);
8669 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8670 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8671 cache
->flags
= type
;
8672 btrfs_set_block_group_flags(&cache
->item
, type
);
8674 cache
->last_byte_to_unpin
= (u64
)-1;
8675 cache
->cached
= BTRFS_CACHE_FINISHED
;
8676 exclude_super_stripes(root
, cache
);
8678 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8679 chunk_offset
+ size
);
8681 free_excluded_extents(root
, cache
);
8683 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8684 &cache
->space_info
);
8687 spin_lock(&cache
->space_info
->lock
);
8688 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8689 spin_unlock(&cache
->space_info
->lock
);
8691 __link_block_group(cache
->space_info
, cache
);
8693 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8696 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
8697 sizeof(cache
->item
));
8700 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8705 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8706 struct btrfs_root
*root
, u64 group_start
)
8708 struct btrfs_path
*path
;
8709 struct btrfs_block_group_cache
*block_group
;
8710 struct btrfs_free_cluster
*cluster
;
8711 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8712 struct btrfs_key key
;
8713 struct inode
*inode
;
8717 root
= root
->fs_info
->extent_root
;
8719 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8720 BUG_ON(!block_group
);
8721 BUG_ON(!block_group
->ro
);
8724 * Free the reserved super bytes from this block group before
8727 free_excluded_extents(root
, block_group
);
8729 memcpy(&key
, &block_group
->key
, sizeof(key
));
8730 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8731 BTRFS_BLOCK_GROUP_RAID1
|
8732 BTRFS_BLOCK_GROUP_RAID10
))
8737 /* make sure this block group isn't part of an allocation cluster */
8738 cluster
= &root
->fs_info
->data_alloc_cluster
;
8739 spin_lock(&cluster
->refill_lock
);
8740 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8741 spin_unlock(&cluster
->refill_lock
);
8744 * make sure this block group isn't part of a metadata
8745 * allocation cluster
8747 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8748 spin_lock(&cluster
->refill_lock
);
8749 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8750 spin_unlock(&cluster
->refill_lock
);
8752 path
= btrfs_alloc_path();
8755 inode
= lookup_free_space_inode(root
, block_group
, path
);
8756 if (!IS_ERR(inode
)) {
8757 btrfs_orphan_add(trans
, inode
);
8759 /* One for the block groups ref */
8760 spin_lock(&block_group
->lock
);
8761 if (block_group
->iref
) {
8762 block_group
->iref
= 0;
8763 block_group
->inode
= NULL
;
8764 spin_unlock(&block_group
->lock
);
8767 spin_unlock(&block_group
->lock
);
8769 /* One for our lookup ref */
8773 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8774 key
.offset
= block_group
->key
.objectid
;
8777 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8781 btrfs_release_path(tree_root
, path
);
8783 ret
= btrfs_del_item(trans
, tree_root
, path
);
8786 btrfs_release_path(tree_root
, path
);
8789 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8790 rb_erase(&block_group
->cache_node
,
8791 &root
->fs_info
->block_group_cache_tree
);
8792 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8794 down_write(&block_group
->space_info
->groups_sem
);
8796 * we must use list_del_init so people can check to see if they
8797 * are still on the list after taking the semaphore
8799 list_del_init(&block_group
->list
);
8800 up_write(&block_group
->space_info
->groups_sem
);
8802 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8803 wait_block_group_cache_done(block_group
);
8805 btrfs_remove_free_space_cache(block_group
);
8807 spin_lock(&block_group
->space_info
->lock
);
8808 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8809 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8810 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8811 spin_unlock(&block_group
->space_info
->lock
);
8813 memcpy(&key
, &block_group
->key
, sizeof(key
));
8815 btrfs_clear_space_info_full(root
->fs_info
);
8817 btrfs_put_block_group(block_group
);
8818 btrfs_put_block_group(block_group
);
8820 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8826 ret
= btrfs_del_item(trans
, root
, path
);
8828 btrfs_free_path(path
);
8832 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
8834 struct btrfs_space_info
*space_info
;
8837 ret
= update_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
, 0, 0,
8842 ret
= update_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
, 0, 0,
8847 ret
= update_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
, 0, 0,
8855 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8857 return unpin_extent_range(root
, start
, end
);
8860 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8861 u64 num_bytes
, u64
*actual_bytes
)
8863 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
8866 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
8868 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8869 struct btrfs_block_group_cache
*cache
= NULL
;
8876 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
8879 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
8880 btrfs_put_block_group(cache
);
8884 start
= max(range
->start
, cache
->key
.objectid
);
8885 end
= min(range
->start
+ range
->len
,
8886 cache
->key
.objectid
+ cache
->key
.offset
);
8888 if (end
- start
>= range
->minlen
) {
8889 if (!block_group_cache_done(cache
)) {
8890 ret
= cache_block_group(cache
, NULL
, root
, 0);
8892 wait_block_group_cache_done(cache
);
8894 ret
= btrfs_trim_block_group(cache
,
8900 trimmed
+= group_trimmed
;
8902 btrfs_put_block_group(cache
);
8907 cache
= next_block_group(fs_info
->tree_root
, cache
);
8910 range
->len
= trimmed
;