2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 static int update_block_group(struct btrfs_trans_handle
*trans
,
37 struct btrfs_root
*root
,
38 u64 bytenr
, u64 num_bytes
, int alloc
);
39 static int update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
40 u64 num_bytes
, int reserve
, int sinfo
);
41 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
42 struct btrfs_root
*root
,
43 u64 bytenr
, u64 num_bytes
, u64 parent
,
44 u64 root_objectid
, u64 owner_objectid
,
45 u64 owner_offset
, int refs_to_drop
,
46 struct btrfs_delayed_extent_op
*extra_op
);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
48 struct extent_buffer
*leaf
,
49 struct btrfs_extent_item
*ei
);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
51 struct btrfs_root
*root
,
52 u64 parent
, u64 root_objectid
,
53 u64 flags
, u64 owner
, u64 offset
,
54 struct btrfs_key
*ins
, int ref_mod
);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 parent
, u64 root_objectid
,
58 u64 flags
, struct btrfs_disk_key
*key
,
59 int level
, struct btrfs_key
*ins
);
60 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
61 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
62 u64 flags
, int force
);
63 static int find_next_key(struct btrfs_path
*path
, int level
,
64 struct btrfs_key
*key
);
65 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
66 int dump_block_groups
);
69 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
72 return cache
->cached
== BTRFS_CACHE_FINISHED
;
75 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
77 return (cache
->flags
& bits
) == bits
;
80 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
82 atomic_inc(&cache
->count
);
85 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
87 if (atomic_dec_and_test(&cache
->count
)) {
88 WARN_ON(cache
->pinned
> 0);
89 WARN_ON(cache
->reserved
> 0);
90 WARN_ON(cache
->reserved_pinned
> 0);
96 * this adds the block group to the fs_info rb tree for the block group
99 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
100 struct btrfs_block_group_cache
*block_group
)
103 struct rb_node
*parent
= NULL
;
104 struct btrfs_block_group_cache
*cache
;
106 spin_lock(&info
->block_group_cache_lock
);
107 p
= &info
->block_group_cache_tree
.rb_node
;
111 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
113 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
115 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
118 spin_unlock(&info
->block_group_cache_lock
);
123 rb_link_node(&block_group
->cache_node
, parent
, p
);
124 rb_insert_color(&block_group
->cache_node
,
125 &info
->block_group_cache_tree
);
126 spin_unlock(&info
->block_group_cache_lock
);
132 * This will return the block group at or after bytenr if contains is 0, else
133 * it will return the block group that contains the bytenr
135 static struct btrfs_block_group_cache
*
136 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
139 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
143 spin_lock(&info
->block_group_cache_lock
);
144 n
= info
->block_group_cache_tree
.rb_node
;
147 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
149 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
150 start
= cache
->key
.objectid
;
152 if (bytenr
< start
) {
153 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
156 } else if (bytenr
> start
) {
157 if (contains
&& bytenr
<= end
) {
168 btrfs_get_block_group(ret
);
169 spin_unlock(&info
->block_group_cache_lock
);
174 static int add_excluded_extent(struct btrfs_root
*root
,
175 u64 start
, u64 num_bytes
)
177 u64 end
= start
+ num_bytes
- 1;
178 set_extent_bits(&root
->fs_info
->freed_extents
[0],
179 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
180 set_extent_bits(&root
->fs_info
->freed_extents
[1],
181 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
185 static void free_excluded_extents(struct btrfs_root
*root
,
186 struct btrfs_block_group_cache
*cache
)
190 start
= cache
->key
.objectid
;
191 end
= start
+ cache
->key
.offset
- 1;
193 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
194 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
195 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
196 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
199 static int exclude_super_stripes(struct btrfs_root
*root
,
200 struct btrfs_block_group_cache
*cache
)
207 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
208 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
209 cache
->bytes_super
+= stripe_len
;
210 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
215 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
216 bytenr
= btrfs_sb_offset(i
);
217 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
218 cache
->key
.objectid
, bytenr
,
219 0, &logical
, &nr
, &stripe_len
);
223 cache
->bytes_super
+= stripe_len
;
224 ret
= add_excluded_extent(root
, logical
[nr
],
234 static struct btrfs_caching_control
*
235 get_caching_control(struct btrfs_block_group_cache
*cache
)
237 struct btrfs_caching_control
*ctl
;
239 spin_lock(&cache
->lock
);
240 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
241 spin_unlock(&cache
->lock
);
245 ctl
= cache
->caching_ctl
;
246 atomic_inc(&ctl
->count
);
247 spin_unlock(&cache
->lock
);
251 static void put_caching_control(struct btrfs_caching_control
*ctl
)
253 if (atomic_dec_and_test(&ctl
->count
))
258 * this is only called by cache_block_group, since we could have freed extents
259 * we need to check the pinned_extents for any extents that can't be used yet
260 * since their free space will be released as soon as the transaction commits.
262 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
263 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
265 u64 extent_start
, extent_end
, size
, total_added
= 0;
268 while (start
< end
) {
269 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
270 &extent_start
, &extent_end
,
271 EXTENT_DIRTY
| EXTENT_UPTODATE
);
275 if (extent_start
<= start
) {
276 start
= extent_end
+ 1;
277 } else if (extent_start
> start
&& extent_start
< end
) {
278 size
= extent_start
- start
;
280 ret
= btrfs_add_free_space(block_group
, start
,
283 start
= extent_end
+ 1;
292 ret
= btrfs_add_free_space(block_group
, start
, size
);
299 static int caching_kthread(void *data
)
301 struct btrfs_block_group_cache
*block_group
= data
;
302 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
303 struct btrfs_caching_control
*caching_ctl
= block_group
->caching_ctl
;
304 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
305 struct btrfs_path
*path
;
306 struct extent_buffer
*leaf
;
307 struct btrfs_key key
;
313 path
= btrfs_alloc_path();
317 exclude_super_stripes(extent_root
, block_group
);
318 spin_lock(&block_group
->space_info
->lock
);
319 block_group
->space_info
->bytes_readonly
+= block_group
->bytes_super
;
320 spin_unlock(&block_group
->space_info
->lock
);
322 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
325 * We don't want to deadlock with somebody trying to allocate a new
326 * extent for the extent root while also trying to search the extent
327 * root to add free space. So we skip locking and search the commit
328 * root, since its read-only
330 path
->skip_locking
= 1;
331 path
->search_commit_root
= 1;
336 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
338 mutex_lock(&caching_ctl
->mutex
);
339 /* need to make sure the commit_root doesn't disappear */
340 down_read(&fs_info
->extent_commit_sem
);
342 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
346 leaf
= path
->nodes
[0];
347 nritems
= btrfs_header_nritems(leaf
);
351 if (fs_info
->closing
> 1) {
356 if (path
->slots
[0] < nritems
) {
357 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
359 ret
= find_next_key(path
, 0, &key
);
363 caching_ctl
->progress
= last
;
364 btrfs_release_path(extent_root
, path
);
365 up_read(&fs_info
->extent_commit_sem
);
366 mutex_unlock(&caching_ctl
->mutex
);
367 if (btrfs_transaction_in_commit(fs_info
))
374 if (key
.objectid
< block_group
->key
.objectid
) {
379 if (key
.objectid
>= block_group
->key
.objectid
+
380 block_group
->key
.offset
)
383 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
384 total_found
+= add_new_free_space(block_group
,
387 last
= key
.objectid
+ key
.offset
;
389 if (total_found
> (1024 * 1024 * 2)) {
391 wake_up(&caching_ctl
->wait
);
398 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
399 block_group
->key
.objectid
+
400 block_group
->key
.offset
);
401 caching_ctl
->progress
= (u64
)-1;
403 spin_lock(&block_group
->lock
);
404 block_group
->caching_ctl
= NULL
;
405 block_group
->cached
= BTRFS_CACHE_FINISHED
;
406 spin_unlock(&block_group
->lock
);
409 btrfs_free_path(path
);
410 up_read(&fs_info
->extent_commit_sem
);
412 free_excluded_extents(extent_root
, block_group
);
414 mutex_unlock(&caching_ctl
->mutex
);
415 wake_up(&caching_ctl
->wait
);
417 put_caching_control(caching_ctl
);
418 atomic_dec(&block_group
->space_info
->caching_threads
);
419 btrfs_put_block_group(block_group
);
424 static int cache_block_group(struct btrfs_block_group_cache
*cache
)
426 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
427 struct btrfs_caching_control
*caching_ctl
;
428 struct task_struct
*tsk
;
432 if (cache
->cached
!= BTRFS_CACHE_NO
)
435 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_KERNEL
);
436 BUG_ON(!caching_ctl
);
438 INIT_LIST_HEAD(&caching_ctl
->list
);
439 mutex_init(&caching_ctl
->mutex
);
440 init_waitqueue_head(&caching_ctl
->wait
);
441 caching_ctl
->block_group
= cache
;
442 caching_ctl
->progress
= cache
->key
.objectid
;
443 /* one for caching kthread, one for caching block group list */
444 atomic_set(&caching_ctl
->count
, 2);
446 spin_lock(&cache
->lock
);
447 if (cache
->cached
!= BTRFS_CACHE_NO
) {
448 spin_unlock(&cache
->lock
);
452 cache
->caching_ctl
= caching_ctl
;
453 cache
->cached
= BTRFS_CACHE_STARTED
;
454 spin_unlock(&cache
->lock
);
456 down_write(&fs_info
->extent_commit_sem
);
457 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
458 up_write(&fs_info
->extent_commit_sem
);
460 atomic_inc(&cache
->space_info
->caching_threads
);
461 btrfs_get_block_group(cache
);
463 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
464 cache
->key
.objectid
);
467 printk(KERN_ERR
"error running thread %d\n", ret
);
475 * return the block group that starts at or after bytenr
477 static struct btrfs_block_group_cache
*
478 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
480 struct btrfs_block_group_cache
*cache
;
482 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
488 * return the block group that contains the given bytenr
490 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
491 struct btrfs_fs_info
*info
,
494 struct btrfs_block_group_cache
*cache
;
496 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
501 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
504 struct list_head
*head
= &info
->space_info
;
505 struct btrfs_space_info
*found
;
507 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
508 BTRFS_BLOCK_GROUP_METADATA
;
511 list_for_each_entry_rcu(found
, head
, list
) {
512 if (found
->flags
== flags
) {
522 * after adding space to the filesystem, we need to clear the full flags
523 * on all the space infos.
525 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
527 struct list_head
*head
= &info
->space_info
;
528 struct btrfs_space_info
*found
;
531 list_for_each_entry_rcu(found
, head
, list
)
536 static u64
div_factor(u64 num
, int factor
)
545 u64
btrfs_find_block_group(struct btrfs_root
*root
,
546 u64 search_start
, u64 search_hint
, int owner
)
548 struct btrfs_block_group_cache
*cache
;
550 u64 last
= max(search_hint
, search_start
);
557 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
561 spin_lock(&cache
->lock
);
562 last
= cache
->key
.objectid
+ cache
->key
.offset
;
563 used
= btrfs_block_group_used(&cache
->item
);
565 if ((full_search
|| !cache
->ro
) &&
566 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
567 if (used
+ cache
->pinned
+ cache
->reserved
<
568 div_factor(cache
->key
.offset
, factor
)) {
569 group_start
= cache
->key
.objectid
;
570 spin_unlock(&cache
->lock
);
571 btrfs_put_block_group(cache
);
575 spin_unlock(&cache
->lock
);
576 btrfs_put_block_group(cache
);
584 if (!full_search
&& factor
< 10) {
594 /* simple helper to search for an existing extent at a given offset */
595 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
598 struct btrfs_key key
;
599 struct btrfs_path
*path
;
601 path
= btrfs_alloc_path();
603 key
.objectid
= start
;
605 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
606 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
608 btrfs_free_path(path
);
613 * helper function to lookup reference count and flags of extent.
615 * the head node for delayed ref is used to store the sum of all the
616 * reference count modifications queued up in the rbtree. the head
617 * node may also store the extent flags to set. This way you can check
618 * to see what the reference count and extent flags would be if all of
619 * the delayed refs are not processed.
621 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
622 struct btrfs_root
*root
, u64 bytenr
,
623 u64 num_bytes
, u64
*refs
, u64
*flags
)
625 struct btrfs_delayed_ref_head
*head
;
626 struct btrfs_delayed_ref_root
*delayed_refs
;
627 struct btrfs_path
*path
;
628 struct btrfs_extent_item
*ei
;
629 struct extent_buffer
*leaf
;
630 struct btrfs_key key
;
636 path
= btrfs_alloc_path();
640 key
.objectid
= bytenr
;
641 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
642 key
.offset
= num_bytes
;
644 path
->skip_locking
= 1;
645 path
->search_commit_root
= 1;
648 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
654 leaf
= path
->nodes
[0];
655 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
656 if (item_size
>= sizeof(*ei
)) {
657 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
658 struct btrfs_extent_item
);
659 num_refs
= btrfs_extent_refs(leaf
, ei
);
660 extent_flags
= btrfs_extent_flags(leaf
, ei
);
662 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
663 struct btrfs_extent_item_v0
*ei0
;
664 BUG_ON(item_size
!= sizeof(*ei0
));
665 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
666 struct btrfs_extent_item_v0
);
667 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
668 /* FIXME: this isn't correct for data */
669 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
674 BUG_ON(num_refs
== 0);
684 delayed_refs
= &trans
->transaction
->delayed_refs
;
685 spin_lock(&delayed_refs
->lock
);
686 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
688 if (!mutex_trylock(&head
->mutex
)) {
689 atomic_inc(&head
->node
.refs
);
690 spin_unlock(&delayed_refs
->lock
);
692 btrfs_release_path(root
->fs_info
->extent_root
, path
);
694 mutex_lock(&head
->mutex
);
695 mutex_unlock(&head
->mutex
);
696 btrfs_put_delayed_ref(&head
->node
);
699 if (head
->extent_op
&& head
->extent_op
->update_flags
)
700 extent_flags
|= head
->extent_op
->flags_to_set
;
702 BUG_ON(num_refs
== 0);
704 num_refs
+= head
->node
.ref_mod
;
705 mutex_unlock(&head
->mutex
);
707 spin_unlock(&delayed_refs
->lock
);
709 WARN_ON(num_refs
== 0);
713 *flags
= extent_flags
;
715 btrfs_free_path(path
);
720 * Back reference rules. Back refs have three main goals:
722 * 1) differentiate between all holders of references to an extent so that
723 * when a reference is dropped we can make sure it was a valid reference
724 * before freeing the extent.
726 * 2) Provide enough information to quickly find the holders of an extent
727 * if we notice a given block is corrupted or bad.
729 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
730 * maintenance. This is actually the same as #2, but with a slightly
731 * different use case.
733 * There are two kinds of back refs. The implicit back refs is optimized
734 * for pointers in non-shared tree blocks. For a given pointer in a block,
735 * back refs of this kind provide information about the block's owner tree
736 * and the pointer's key. These information allow us to find the block by
737 * b-tree searching. The full back refs is for pointers in tree blocks not
738 * referenced by their owner trees. The location of tree block is recorded
739 * in the back refs. Actually the full back refs is generic, and can be
740 * used in all cases the implicit back refs is used. The major shortcoming
741 * of the full back refs is its overhead. Every time a tree block gets
742 * COWed, we have to update back refs entry for all pointers in it.
744 * For a newly allocated tree block, we use implicit back refs for
745 * pointers in it. This means most tree related operations only involve
746 * implicit back refs. For a tree block created in old transaction, the
747 * only way to drop a reference to it is COW it. So we can detect the
748 * event that tree block loses its owner tree's reference and do the
749 * back refs conversion.
751 * When a tree block is COW'd through a tree, there are four cases:
753 * The reference count of the block is one and the tree is the block's
754 * owner tree. Nothing to do in this case.
756 * The reference count of the block is one and the tree is not the
757 * block's owner tree. In this case, full back refs is used for pointers
758 * in the block. Remove these full back refs, add implicit back refs for
759 * every pointers in the new block.
761 * The reference count of the block is greater than one and the tree is
762 * the block's owner tree. In this case, implicit back refs is used for
763 * pointers in the block. Add full back refs for every pointers in the
764 * block, increase lower level extents' reference counts. The original
765 * implicit back refs are entailed to the new block.
767 * The reference count of the block is greater than one and the tree is
768 * not the block's owner tree. Add implicit back refs for every pointer in
769 * the new block, increase lower level extents' reference count.
771 * Back Reference Key composing:
773 * The key objectid corresponds to the first byte in the extent,
774 * The key type is used to differentiate between types of back refs.
775 * There are different meanings of the key offset for different types
778 * File extents can be referenced by:
780 * - multiple snapshots, subvolumes, or different generations in one subvol
781 * - different files inside a single subvolume
782 * - different offsets inside a file (bookend extents in file.c)
784 * The extent ref structure for the implicit back refs has fields for:
786 * - Objectid of the subvolume root
787 * - objectid of the file holding the reference
788 * - original offset in the file
789 * - how many bookend extents
791 * The key offset for the implicit back refs is hash of the first
794 * The extent ref structure for the full back refs has field for:
796 * - number of pointers in the tree leaf
798 * The key offset for the implicit back refs is the first byte of
801 * When a file extent is allocated, The implicit back refs is used.
802 * the fields are filled in:
804 * (root_key.objectid, inode objectid, offset in file, 1)
806 * When a file extent is removed file truncation, we find the
807 * corresponding implicit back refs and check the following fields:
809 * (btrfs_header_owner(leaf), inode objectid, offset in file)
811 * Btree extents can be referenced by:
813 * - Different subvolumes
815 * Both the implicit back refs and the full back refs for tree blocks
816 * only consist of key. The key offset for the implicit back refs is
817 * objectid of block's owner tree. The key offset for the full back refs
818 * is the first byte of parent block.
820 * When implicit back refs is used, information about the lowest key and
821 * level of the tree block are required. These information are stored in
822 * tree block info structure.
825 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
826 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
827 struct btrfs_root
*root
,
828 struct btrfs_path
*path
,
829 u64 owner
, u32 extra_size
)
831 struct btrfs_extent_item
*item
;
832 struct btrfs_extent_item_v0
*ei0
;
833 struct btrfs_extent_ref_v0
*ref0
;
834 struct btrfs_tree_block_info
*bi
;
835 struct extent_buffer
*leaf
;
836 struct btrfs_key key
;
837 struct btrfs_key found_key
;
838 u32 new_size
= sizeof(*item
);
842 leaf
= path
->nodes
[0];
843 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
845 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
846 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
847 struct btrfs_extent_item_v0
);
848 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
850 if (owner
== (u64
)-1) {
852 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
853 ret
= btrfs_next_leaf(root
, path
);
857 leaf
= path
->nodes
[0];
859 btrfs_item_key_to_cpu(leaf
, &found_key
,
861 BUG_ON(key
.objectid
!= found_key
.objectid
);
862 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
866 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
867 struct btrfs_extent_ref_v0
);
868 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
872 btrfs_release_path(root
, path
);
874 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
875 new_size
+= sizeof(*bi
);
877 new_size
-= sizeof(*ei0
);
878 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
879 new_size
+ extra_size
, 1);
884 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
887 leaf
= path
->nodes
[0];
888 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
889 btrfs_set_extent_refs(leaf
, item
, refs
);
890 /* FIXME: get real generation */
891 btrfs_set_extent_generation(leaf
, item
, 0);
892 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
893 btrfs_set_extent_flags(leaf
, item
,
894 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
895 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
896 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
897 /* FIXME: get first key of the block */
898 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
899 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
901 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
903 btrfs_mark_buffer_dirty(leaf
);
908 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
910 u32 high_crc
= ~(u32
)0;
911 u32 low_crc
= ~(u32
)0;
914 lenum
= cpu_to_le64(root_objectid
);
915 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
916 lenum
= cpu_to_le64(owner
);
917 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
918 lenum
= cpu_to_le64(offset
);
919 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
921 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
924 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
925 struct btrfs_extent_data_ref
*ref
)
927 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
928 btrfs_extent_data_ref_objectid(leaf
, ref
),
929 btrfs_extent_data_ref_offset(leaf
, ref
));
932 static int match_extent_data_ref(struct extent_buffer
*leaf
,
933 struct btrfs_extent_data_ref
*ref
,
934 u64 root_objectid
, u64 owner
, u64 offset
)
936 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
937 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
938 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
943 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
944 struct btrfs_root
*root
,
945 struct btrfs_path
*path
,
946 u64 bytenr
, u64 parent
,
948 u64 owner
, u64 offset
)
950 struct btrfs_key key
;
951 struct btrfs_extent_data_ref
*ref
;
952 struct extent_buffer
*leaf
;
958 key
.objectid
= bytenr
;
960 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
963 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
964 key
.offset
= hash_extent_data_ref(root_objectid
,
969 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
980 btrfs_release_path(root
, path
);
981 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
992 leaf
= path
->nodes
[0];
993 nritems
= btrfs_header_nritems(leaf
);
995 if (path
->slots
[0] >= nritems
) {
996 ret
= btrfs_next_leaf(root
, path
);
1002 leaf
= path
->nodes
[0];
1003 nritems
= btrfs_header_nritems(leaf
);
1007 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1008 if (key
.objectid
!= bytenr
||
1009 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1012 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1013 struct btrfs_extent_data_ref
);
1015 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1018 btrfs_release_path(root
, path
);
1030 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1031 struct btrfs_root
*root
,
1032 struct btrfs_path
*path
,
1033 u64 bytenr
, u64 parent
,
1034 u64 root_objectid
, u64 owner
,
1035 u64 offset
, int refs_to_add
)
1037 struct btrfs_key key
;
1038 struct extent_buffer
*leaf
;
1043 key
.objectid
= bytenr
;
1045 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1046 key
.offset
= parent
;
1047 size
= sizeof(struct btrfs_shared_data_ref
);
1049 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1050 key
.offset
= hash_extent_data_ref(root_objectid
,
1052 size
= sizeof(struct btrfs_extent_data_ref
);
1055 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1056 if (ret
&& ret
!= -EEXIST
)
1059 leaf
= path
->nodes
[0];
1061 struct btrfs_shared_data_ref
*ref
;
1062 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1063 struct btrfs_shared_data_ref
);
1065 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1067 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1068 num_refs
+= refs_to_add
;
1069 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1072 struct btrfs_extent_data_ref
*ref
;
1073 while (ret
== -EEXIST
) {
1074 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1075 struct btrfs_extent_data_ref
);
1076 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1079 btrfs_release_path(root
, path
);
1081 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1083 if (ret
&& ret
!= -EEXIST
)
1086 leaf
= path
->nodes
[0];
1088 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1089 struct btrfs_extent_data_ref
);
1091 btrfs_set_extent_data_ref_root(leaf
, ref
,
1093 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1094 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1095 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1097 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1098 num_refs
+= refs_to_add
;
1099 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1102 btrfs_mark_buffer_dirty(leaf
);
1105 btrfs_release_path(root
, path
);
1109 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1110 struct btrfs_root
*root
,
1111 struct btrfs_path
*path
,
1114 struct btrfs_key key
;
1115 struct btrfs_extent_data_ref
*ref1
= NULL
;
1116 struct btrfs_shared_data_ref
*ref2
= NULL
;
1117 struct extent_buffer
*leaf
;
1121 leaf
= path
->nodes
[0];
1122 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1124 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1125 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1126 struct btrfs_extent_data_ref
);
1127 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1128 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1129 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1130 struct btrfs_shared_data_ref
);
1131 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1132 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1133 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1134 struct btrfs_extent_ref_v0
*ref0
;
1135 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1136 struct btrfs_extent_ref_v0
);
1137 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1143 BUG_ON(num_refs
< refs_to_drop
);
1144 num_refs
-= refs_to_drop
;
1146 if (num_refs
== 0) {
1147 ret
= btrfs_del_item(trans
, root
, path
);
1149 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1150 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1151 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1152 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1155 struct btrfs_extent_ref_v0
*ref0
;
1156 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1157 struct btrfs_extent_ref_v0
);
1158 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1161 btrfs_mark_buffer_dirty(leaf
);
1166 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1167 struct btrfs_path
*path
,
1168 struct btrfs_extent_inline_ref
*iref
)
1170 struct btrfs_key key
;
1171 struct extent_buffer
*leaf
;
1172 struct btrfs_extent_data_ref
*ref1
;
1173 struct btrfs_shared_data_ref
*ref2
;
1176 leaf
= path
->nodes
[0];
1177 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1179 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1180 BTRFS_EXTENT_DATA_REF_KEY
) {
1181 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1182 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1184 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1185 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1187 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1188 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1189 struct btrfs_extent_data_ref
);
1190 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1191 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1192 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1193 struct btrfs_shared_data_ref
);
1194 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1196 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1197 struct btrfs_extent_ref_v0
*ref0
;
1198 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1199 struct btrfs_extent_ref_v0
);
1200 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1208 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1209 struct btrfs_root
*root
,
1210 struct btrfs_path
*path
,
1211 u64 bytenr
, u64 parent
,
1214 struct btrfs_key key
;
1217 key
.objectid
= bytenr
;
1219 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1220 key
.offset
= parent
;
1222 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1223 key
.offset
= root_objectid
;
1226 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1229 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1230 if (ret
== -ENOENT
&& parent
) {
1231 btrfs_release_path(root
, path
);
1232 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1233 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1241 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1242 struct btrfs_root
*root
,
1243 struct btrfs_path
*path
,
1244 u64 bytenr
, u64 parent
,
1247 struct btrfs_key key
;
1250 key
.objectid
= bytenr
;
1252 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1253 key
.offset
= parent
;
1255 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1256 key
.offset
= root_objectid
;
1259 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1260 btrfs_release_path(root
, path
);
1264 static inline int extent_ref_type(u64 parent
, u64 owner
)
1267 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1269 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1271 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1274 type
= BTRFS_SHARED_DATA_REF_KEY
;
1276 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1281 static int find_next_key(struct btrfs_path
*path
, int level
,
1282 struct btrfs_key
*key
)
1285 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1286 if (!path
->nodes
[level
])
1288 if (path
->slots
[level
] + 1 >=
1289 btrfs_header_nritems(path
->nodes
[level
]))
1292 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1293 path
->slots
[level
] + 1);
1295 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1296 path
->slots
[level
] + 1);
1303 * look for inline back ref. if back ref is found, *ref_ret is set
1304 * to the address of inline back ref, and 0 is returned.
1306 * if back ref isn't found, *ref_ret is set to the address where it
1307 * should be inserted, and -ENOENT is returned.
1309 * if insert is true and there are too many inline back refs, the path
1310 * points to the extent item, and -EAGAIN is returned.
1312 * NOTE: inline back refs are ordered in the same way that back ref
1313 * items in the tree are ordered.
1315 static noinline_for_stack
1316 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1317 struct btrfs_root
*root
,
1318 struct btrfs_path
*path
,
1319 struct btrfs_extent_inline_ref
**ref_ret
,
1320 u64 bytenr
, u64 num_bytes
,
1321 u64 parent
, u64 root_objectid
,
1322 u64 owner
, u64 offset
, int insert
)
1324 struct btrfs_key key
;
1325 struct extent_buffer
*leaf
;
1326 struct btrfs_extent_item
*ei
;
1327 struct btrfs_extent_inline_ref
*iref
;
1338 key
.objectid
= bytenr
;
1339 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1340 key
.offset
= num_bytes
;
1342 want
= extent_ref_type(parent
, owner
);
1344 extra_size
= btrfs_extent_inline_ref_size(want
);
1345 path
->keep_locks
= 1;
1348 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1355 leaf
= path
->nodes
[0];
1356 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 if (item_size
< sizeof(*ei
)) {
1363 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1369 leaf
= path
->nodes
[0];
1370 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1373 BUG_ON(item_size
< sizeof(*ei
));
1375 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1376 flags
= btrfs_extent_flags(leaf
, ei
);
1378 ptr
= (unsigned long)(ei
+ 1);
1379 end
= (unsigned long)ei
+ item_size
;
1381 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1382 ptr
+= sizeof(struct btrfs_tree_block_info
);
1385 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1394 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1395 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1399 ptr
+= btrfs_extent_inline_ref_size(type
);
1403 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1404 struct btrfs_extent_data_ref
*dref
;
1405 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1406 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1411 if (hash_extent_data_ref_item(leaf
, dref
) <
1412 hash_extent_data_ref(root_objectid
, owner
, offset
))
1416 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1418 if (parent
== ref_offset
) {
1422 if (ref_offset
< parent
)
1425 if (root_objectid
== ref_offset
) {
1429 if (ref_offset
< root_objectid
)
1433 ptr
+= btrfs_extent_inline_ref_size(type
);
1435 if (err
== -ENOENT
&& insert
) {
1436 if (item_size
+ extra_size
>=
1437 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1442 * To add new inline back ref, we have to make sure
1443 * there is no corresponding back ref item.
1444 * For simplicity, we just do not add new inline back
1445 * ref if there is any kind of item for this block
1447 if (find_next_key(path
, 0, &key
) == 0 &&
1448 key
.objectid
== bytenr
&&
1449 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1454 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1457 path
->keep_locks
= 0;
1458 btrfs_unlock_up_safe(path
, 1);
1464 * helper to add new inline back ref
1466 static noinline_for_stack
1467 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1468 struct btrfs_root
*root
,
1469 struct btrfs_path
*path
,
1470 struct btrfs_extent_inline_ref
*iref
,
1471 u64 parent
, u64 root_objectid
,
1472 u64 owner
, u64 offset
, int refs_to_add
,
1473 struct btrfs_delayed_extent_op
*extent_op
)
1475 struct extent_buffer
*leaf
;
1476 struct btrfs_extent_item
*ei
;
1479 unsigned long item_offset
;
1485 leaf
= path
->nodes
[0];
1486 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1487 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1489 type
= extent_ref_type(parent
, owner
);
1490 size
= btrfs_extent_inline_ref_size(type
);
1492 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1495 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1496 refs
= btrfs_extent_refs(leaf
, ei
);
1497 refs
+= refs_to_add
;
1498 btrfs_set_extent_refs(leaf
, ei
, refs
);
1500 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1502 ptr
= (unsigned long)ei
+ item_offset
;
1503 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1504 if (ptr
< end
- size
)
1505 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1508 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1509 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1510 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1511 struct btrfs_extent_data_ref
*dref
;
1512 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1513 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1514 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1515 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1516 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1517 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1518 struct btrfs_shared_data_ref
*sref
;
1519 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1520 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1521 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1522 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1523 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1525 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1527 btrfs_mark_buffer_dirty(leaf
);
1531 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1532 struct btrfs_root
*root
,
1533 struct btrfs_path
*path
,
1534 struct btrfs_extent_inline_ref
**ref_ret
,
1535 u64 bytenr
, u64 num_bytes
, u64 parent
,
1536 u64 root_objectid
, u64 owner
, u64 offset
)
1540 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1541 bytenr
, num_bytes
, parent
,
1542 root_objectid
, owner
, offset
, 0);
1546 btrfs_release_path(root
, path
);
1549 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1550 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1553 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1554 root_objectid
, owner
, offset
);
1560 * helper to update/remove inline back ref
1562 static noinline_for_stack
1563 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1564 struct btrfs_root
*root
,
1565 struct btrfs_path
*path
,
1566 struct btrfs_extent_inline_ref
*iref
,
1568 struct btrfs_delayed_extent_op
*extent_op
)
1570 struct extent_buffer
*leaf
;
1571 struct btrfs_extent_item
*ei
;
1572 struct btrfs_extent_data_ref
*dref
= NULL
;
1573 struct btrfs_shared_data_ref
*sref
= NULL
;
1582 leaf
= path
->nodes
[0];
1583 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1584 refs
= btrfs_extent_refs(leaf
, ei
);
1585 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1586 refs
+= refs_to_mod
;
1587 btrfs_set_extent_refs(leaf
, ei
, refs
);
1589 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1591 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1593 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1594 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1595 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1596 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1597 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1598 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1601 BUG_ON(refs_to_mod
!= -1);
1604 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1605 refs
+= refs_to_mod
;
1608 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1609 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1611 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1613 size
= btrfs_extent_inline_ref_size(type
);
1614 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1615 ptr
= (unsigned long)iref
;
1616 end
= (unsigned long)ei
+ item_size
;
1617 if (ptr
+ size
< end
)
1618 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1621 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1624 btrfs_mark_buffer_dirty(leaf
);
1628 static noinline_for_stack
1629 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1630 struct btrfs_root
*root
,
1631 struct btrfs_path
*path
,
1632 u64 bytenr
, u64 num_bytes
, u64 parent
,
1633 u64 root_objectid
, u64 owner
,
1634 u64 offset
, int refs_to_add
,
1635 struct btrfs_delayed_extent_op
*extent_op
)
1637 struct btrfs_extent_inline_ref
*iref
;
1640 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1641 bytenr
, num_bytes
, parent
,
1642 root_objectid
, owner
, offset
, 1);
1644 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1645 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1646 refs_to_add
, extent_op
);
1647 } else if (ret
== -ENOENT
) {
1648 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1649 parent
, root_objectid
,
1650 owner
, offset
, refs_to_add
,
1656 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1657 struct btrfs_root
*root
,
1658 struct btrfs_path
*path
,
1659 u64 bytenr
, u64 parent
, u64 root_objectid
,
1660 u64 owner
, u64 offset
, int refs_to_add
)
1663 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1664 BUG_ON(refs_to_add
!= 1);
1665 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1666 parent
, root_objectid
);
1668 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1669 parent
, root_objectid
,
1670 owner
, offset
, refs_to_add
);
1675 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1676 struct btrfs_root
*root
,
1677 struct btrfs_path
*path
,
1678 struct btrfs_extent_inline_ref
*iref
,
1679 int refs_to_drop
, int is_data
)
1683 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1685 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1686 -refs_to_drop
, NULL
);
1687 } else if (is_data
) {
1688 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1690 ret
= btrfs_del_item(trans
, root
, path
);
1695 static void btrfs_issue_discard(struct block_device
*bdev
,
1698 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
,
1699 DISCARD_FL_BARRIER
);
1702 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1706 u64 map_length
= num_bytes
;
1707 struct btrfs_multi_bio
*multi
= NULL
;
1709 if (!btrfs_test_opt(root
, DISCARD
))
1712 /* Tell the block device(s) that the sectors can be discarded */
1713 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1714 bytenr
, &map_length
, &multi
, 0);
1716 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1719 if (map_length
> num_bytes
)
1720 map_length
= num_bytes
;
1722 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1723 btrfs_issue_discard(stripe
->dev
->bdev
,
1733 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1734 struct btrfs_root
*root
,
1735 u64 bytenr
, u64 num_bytes
, u64 parent
,
1736 u64 root_objectid
, u64 owner
, u64 offset
)
1739 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1740 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1742 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1743 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1744 parent
, root_objectid
, (int)owner
,
1745 BTRFS_ADD_DELAYED_REF
, NULL
);
1747 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1748 parent
, root_objectid
, owner
, offset
,
1749 BTRFS_ADD_DELAYED_REF
, NULL
);
1754 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1755 struct btrfs_root
*root
,
1756 u64 bytenr
, u64 num_bytes
,
1757 u64 parent
, u64 root_objectid
,
1758 u64 owner
, u64 offset
, int refs_to_add
,
1759 struct btrfs_delayed_extent_op
*extent_op
)
1761 struct btrfs_path
*path
;
1762 struct extent_buffer
*leaf
;
1763 struct btrfs_extent_item
*item
;
1768 path
= btrfs_alloc_path();
1773 path
->leave_spinning
= 1;
1774 /* this will setup the path even if it fails to insert the back ref */
1775 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1776 path
, bytenr
, num_bytes
, parent
,
1777 root_objectid
, owner
, offset
,
1778 refs_to_add
, extent_op
);
1782 if (ret
!= -EAGAIN
) {
1787 leaf
= path
->nodes
[0];
1788 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1789 refs
= btrfs_extent_refs(leaf
, item
);
1790 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1792 __run_delayed_extent_op(extent_op
, leaf
, item
);
1794 btrfs_mark_buffer_dirty(leaf
);
1795 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1798 path
->leave_spinning
= 1;
1800 /* now insert the actual backref */
1801 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1802 path
, bytenr
, parent
, root_objectid
,
1803 owner
, offset
, refs_to_add
);
1806 btrfs_free_path(path
);
1810 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1811 struct btrfs_root
*root
,
1812 struct btrfs_delayed_ref_node
*node
,
1813 struct btrfs_delayed_extent_op
*extent_op
,
1814 int insert_reserved
)
1817 struct btrfs_delayed_data_ref
*ref
;
1818 struct btrfs_key ins
;
1823 ins
.objectid
= node
->bytenr
;
1824 ins
.offset
= node
->num_bytes
;
1825 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1827 ref
= btrfs_delayed_node_to_data_ref(node
);
1828 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1829 parent
= ref
->parent
;
1831 ref_root
= ref
->root
;
1833 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1835 BUG_ON(extent_op
->update_key
);
1836 flags
|= extent_op
->flags_to_set
;
1838 ret
= alloc_reserved_file_extent(trans
, root
,
1839 parent
, ref_root
, flags
,
1840 ref
->objectid
, ref
->offset
,
1841 &ins
, node
->ref_mod
);
1842 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1843 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1844 node
->num_bytes
, parent
,
1845 ref_root
, ref
->objectid
,
1846 ref
->offset
, node
->ref_mod
,
1848 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1849 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1850 node
->num_bytes
, parent
,
1851 ref_root
, ref
->objectid
,
1852 ref
->offset
, node
->ref_mod
,
1860 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1861 struct extent_buffer
*leaf
,
1862 struct btrfs_extent_item
*ei
)
1864 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1865 if (extent_op
->update_flags
) {
1866 flags
|= extent_op
->flags_to_set
;
1867 btrfs_set_extent_flags(leaf
, ei
, flags
);
1870 if (extent_op
->update_key
) {
1871 struct btrfs_tree_block_info
*bi
;
1872 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1873 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1874 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1878 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1879 struct btrfs_root
*root
,
1880 struct btrfs_delayed_ref_node
*node
,
1881 struct btrfs_delayed_extent_op
*extent_op
)
1883 struct btrfs_key key
;
1884 struct btrfs_path
*path
;
1885 struct btrfs_extent_item
*ei
;
1886 struct extent_buffer
*leaf
;
1891 path
= btrfs_alloc_path();
1895 key
.objectid
= node
->bytenr
;
1896 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1897 key
.offset
= node
->num_bytes
;
1900 path
->leave_spinning
= 1;
1901 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1912 leaf
= path
->nodes
[0];
1913 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1914 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1915 if (item_size
< sizeof(*ei
)) {
1916 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1922 leaf
= path
->nodes
[0];
1923 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1926 BUG_ON(item_size
< sizeof(*ei
));
1927 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1928 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1930 btrfs_mark_buffer_dirty(leaf
);
1932 btrfs_free_path(path
);
1936 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1937 struct btrfs_root
*root
,
1938 struct btrfs_delayed_ref_node
*node
,
1939 struct btrfs_delayed_extent_op
*extent_op
,
1940 int insert_reserved
)
1943 struct btrfs_delayed_tree_ref
*ref
;
1944 struct btrfs_key ins
;
1948 ins
.objectid
= node
->bytenr
;
1949 ins
.offset
= node
->num_bytes
;
1950 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1952 ref
= btrfs_delayed_node_to_tree_ref(node
);
1953 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1954 parent
= ref
->parent
;
1956 ref_root
= ref
->root
;
1958 BUG_ON(node
->ref_mod
!= 1);
1959 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1960 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
1961 !extent_op
->update_key
);
1962 ret
= alloc_reserved_tree_block(trans
, root
,
1964 extent_op
->flags_to_set
,
1967 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1968 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1969 node
->num_bytes
, parent
, ref_root
,
1970 ref
->level
, 0, 1, extent_op
);
1971 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1972 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1973 node
->num_bytes
, parent
, ref_root
,
1974 ref
->level
, 0, 1, extent_op
);
1981 /* helper function to actually process a single delayed ref entry */
1982 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
1983 struct btrfs_root
*root
,
1984 struct btrfs_delayed_ref_node
*node
,
1985 struct btrfs_delayed_extent_op
*extent_op
,
1986 int insert_reserved
)
1989 if (btrfs_delayed_ref_is_head(node
)) {
1990 struct btrfs_delayed_ref_head
*head
;
1992 * we've hit the end of the chain and we were supposed
1993 * to insert this extent into the tree. But, it got
1994 * deleted before we ever needed to insert it, so all
1995 * we have to do is clean up the accounting
1998 head
= btrfs_delayed_node_to_head(node
);
1999 if (insert_reserved
) {
2000 btrfs_pin_extent(root
, node
->bytenr
,
2001 node
->num_bytes
, 1);
2002 if (head
->is_data
) {
2003 ret
= btrfs_del_csums(trans
, root
,
2009 mutex_unlock(&head
->mutex
);
2013 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2014 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2015 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2017 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2018 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2019 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2026 static noinline
struct btrfs_delayed_ref_node
*
2027 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2029 struct rb_node
*node
;
2030 struct btrfs_delayed_ref_node
*ref
;
2031 int action
= BTRFS_ADD_DELAYED_REF
;
2034 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2035 * this prevents ref count from going down to zero when
2036 * there still are pending delayed ref.
2038 node
= rb_prev(&head
->node
.rb_node
);
2042 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2044 if (ref
->bytenr
!= head
->node
.bytenr
)
2046 if (ref
->action
== action
)
2048 node
= rb_prev(node
);
2050 if (action
== BTRFS_ADD_DELAYED_REF
) {
2051 action
= BTRFS_DROP_DELAYED_REF
;
2057 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2058 struct btrfs_root
*root
,
2059 struct list_head
*cluster
)
2061 struct btrfs_delayed_ref_root
*delayed_refs
;
2062 struct btrfs_delayed_ref_node
*ref
;
2063 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2064 struct btrfs_delayed_extent_op
*extent_op
;
2067 int must_insert_reserved
= 0;
2069 delayed_refs
= &trans
->transaction
->delayed_refs
;
2072 /* pick a new head ref from the cluster list */
2073 if (list_empty(cluster
))
2076 locked_ref
= list_entry(cluster
->next
,
2077 struct btrfs_delayed_ref_head
, cluster
);
2079 /* grab the lock that says we are going to process
2080 * all the refs for this head */
2081 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2084 * we may have dropped the spin lock to get the head
2085 * mutex lock, and that might have given someone else
2086 * time to free the head. If that's true, it has been
2087 * removed from our list and we can move on.
2089 if (ret
== -EAGAIN
) {
2097 * record the must insert reserved flag before we
2098 * drop the spin lock.
2100 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2101 locked_ref
->must_insert_reserved
= 0;
2103 extent_op
= locked_ref
->extent_op
;
2104 locked_ref
->extent_op
= NULL
;
2107 * locked_ref is the head node, so we have to go one
2108 * node back for any delayed ref updates
2110 ref
= select_delayed_ref(locked_ref
);
2112 /* All delayed refs have been processed, Go ahead
2113 * and send the head node to run_one_delayed_ref,
2114 * so that any accounting fixes can happen
2116 ref
= &locked_ref
->node
;
2118 if (extent_op
&& must_insert_reserved
) {
2124 spin_unlock(&delayed_refs
->lock
);
2126 ret
= run_delayed_extent_op(trans
, root
,
2132 spin_lock(&delayed_refs
->lock
);
2136 list_del_init(&locked_ref
->cluster
);
2141 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2142 delayed_refs
->num_entries
--;
2144 spin_unlock(&delayed_refs
->lock
);
2146 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2147 must_insert_reserved
);
2150 btrfs_put_delayed_ref(ref
);
2155 spin_lock(&delayed_refs
->lock
);
2161 * this starts processing the delayed reference count updates and
2162 * extent insertions we have queued up so far. count can be
2163 * 0, which means to process everything in the tree at the start
2164 * of the run (but not newly added entries), or it can be some target
2165 * number you'd like to process.
2167 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2168 struct btrfs_root
*root
, unsigned long count
)
2170 struct rb_node
*node
;
2171 struct btrfs_delayed_ref_root
*delayed_refs
;
2172 struct btrfs_delayed_ref_node
*ref
;
2173 struct list_head cluster
;
2175 int run_all
= count
== (unsigned long)-1;
2178 if (root
== root
->fs_info
->extent_root
)
2179 root
= root
->fs_info
->tree_root
;
2181 delayed_refs
= &trans
->transaction
->delayed_refs
;
2182 INIT_LIST_HEAD(&cluster
);
2184 spin_lock(&delayed_refs
->lock
);
2186 count
= delayed_refs
->num_entries
* 2;
2190 if (!(run_all
|| run_most
) &&
2191 delayed_refs
->num_heads_ready
< 64)
2195 * go find something we can process in the rbtree. We start at
2196 * the beginning of the tree, and then build a cluster
2197 * of refs to process starting at the first one we are able to
2200 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2201 delayed_refs
->run_delayed_start
);
2205 ret
= run_clustered_refs(trans
, root
, &cluster
);
2208 count
-= min_t(unsigned long, ret
, count
);
2215 node
= rb_first(&delayed_refs
->root
);
2218 count
= (unsigned long)-1;
2221 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2223 if (btrfs_delayed_ref_is_head(ref
)) {
2224 struct btrfs_delayed_ref_head
*head
;
2226 head
= btrfs_delayed_node_to_head(ref
);
2227 atomic_inc(&ref
->refs
);
2229 spin_unlock(&delayed_refs
->lock
);
2230 mutex_lock(&head
->mutex
);
2231 mutex_unlock(&head
->mutex
);
2233 btrfs_put_delayed_ref(ref
);
2237 node
= rb_next(node
);
2239 spin_unlock(&delayed_refs
->lock
);
2240 schedule_timeout(1);
2244 spin_unlock(&delayed_refs
->lock
);
2248 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2249 struct btrfs_root
*root
,
2250 u64 bytenr
, u64 num_bytes
, u64 flags
,
2253 struct btrfs_delayed_extent_op
*extent_op
;
2256 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2260 extent_op
->flags_to_set
= flags
;
2261 extent_op
->update_flags
= 1;
2262 extent_op
->update_key
= 0;
2263 extent_op
->is_data
= is_data
? 1 : 0;
2265 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2271 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2272 struct btrfs_root
*root
,
2273 struct btrfs_path
*path
,
2274 u64 objectid
, u64 offset
, u64 bytenr
)
2276 struct btrfs_delayed_ref_head
*head
;
2277 struct btrfs_delayed_ref_node
*ref
;
2278 struct btrfs_delayed_data_ref
*data_ref
;
2279 struct btrfs_delayed_ref_root
*delayed_refs
;
2280 struct rb_node
*node
;
2284 delayed_refs
= &trans
->transaction
->delayed_refs
;
2285 spin_lock(&delayed_refs
->lock
);
2286 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2290 if (!mutex_trylock(&head
->mutex
)) {
2291 atomic_inc(&head
->node
.refs
);
2292 spin_unlock(&delayed_refs
->lock
);
2294 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2296 mutex_lock(&head
->mutex
);
2297 mutex_unlock(&head
->mutex
);
2298 btrfs_put_delayed_ref(&head
->node
);
2302 node
= rb_prev(&head
->node
.rb_node
);
2306 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2308 if (ref
->bytenr
!= bytenr
)
2312 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2315 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2317 node
= rb_prev(node
);
2319 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2320 if (ref
->bytenr
== bytenr
)
2324 if (data_ref
->root
!= root
->root_key
.objectid
||
2325 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2330 mutex_unlock(&head
->mutex
);
2332 spin_unlock(&delayed_refs
->lock
);
2336 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2337 struct btrfs_root
*root
,
2338 struct btrfs_path
*path
,
2339 u64 objectid
, u64 offset
, u64 bytenr
)
2341 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2342 struct extent_buffer
*leaf
;
2343 struct btrfs_extent_data_ref
*ref
;
2344 struct btrfs_extent_inline_ref
*iref
;
2345 struct btrfs_extent_item
*ei
;
2346 struct btrfs_key key
;
2350 key
.objectid
= bytenr
;
2351 key
.offset
= (u64
)-1;
2352 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2354 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2360 if (path
->slots
[0] == 0)
2364 leaf
= path
->nodes
[0];
2365 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2367 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2371 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2372 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2373 if (item_size
< sizeof(*ei
)) {
2374 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2378 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2380 if (item_size
!= sizeof(*ei
) +
2381 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2384 if (btrfs_extent_generation(leaf
, ei
) <=
2385 btrfs_root_last_snapshot(&root
->root_item
))
2388 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2389 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2390 BTRFS_EXTENT_DATA_REF_KEY
)
2393 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2394 if (btrfs_extent_refs(leaf
, ei
) !=
2395 btrfs_extent_data_ref_count(leaf
, ref
) ||
2396 btrfs_extent_data_ref_root(leaf
, ref
) !=
2397 root
->root_key
.objectid
||
2398 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2399 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2407 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2408 struct btrfs_root
*root
,
2409 u64 objectid
, u64 offset
, u64 bytenr
)
2411 struct btrfs_path
*path
;
2415 path
= btrfs_alloc_path();
2420 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2422 if (ret
&& ret
!= -ENOENT
)
2425 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2427 } while (ret2
== -EAGAIN
);
2429 if (ret2
&& ret2
!= -ENOENT
) {
2434 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2437 btrfs_free_path(path
);
2438 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2444 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2445 struct extent_buffer
*buf
, u32 nr_extents
)
2447 struct btrfs_key key
;
2448 struct btrfs_file_extent_item
*fi
;
2456 if (!root
->ref_cows
)
2459 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2461 root_gen
= root
->root_key
.offset
;
2464 root_gen
= trans
->transid
- 1;
2467 level
= btrfs_header_level(buf
);
2468 nritems
= btrfs_header_nritems(buf
);
2471 struct btrfs_leaf_ref
*ref
;
2472 struct btrfs_extent_info
*info
;
2474 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2480 ref
->root_gen
= root_gen
;
2481 ref
->bytenr
= buf
->start
;
2482 ref
->owner
= btrfs_header_owner(buf
);
2483 ref
->generation
= btrfs_header_generation(buf
);
2484 ref
->nritems
= nr_extents
;
2485 info
= ref
->extents
;
2487 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2489 btrfs_item_key_to_cpu(buf
, &key
, i
);
2490 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2492 fi
= btrfs_item_ptr(buf
, i
,
2493 struct btrfs_file_extent_item
);
2494 if (btrfs_file_extent_type(buf
, fi
) ==
2495 BTRFS_FILE_EXTENT_INLINE
)
2497 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2498 if (disk_bytenr
== 0)
2501 info
->bytenr
= disk_bytenr
;
2503 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2504 info
->objectid
= key
.objectid
;
2505 info
->offset
= key
.offset
;
2509 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2510 if (ret
== -EEXIST
&& shared
) {
2511 struct btrfs_leaf_ref
*old
;
2512 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2514 btrfs_remove_leaf_ref(root
, old
);
2515 btrfs_free_leaf_ref(root
, old
);
2516 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2519 btrfs_free_leaf_ref(root
, ref
);
2525 /* when a block goes through cow, we update the reference counts of
2526 * everything that block points to. The internal pointers of the block
2527 * can be in just about any order, and it is likely to have clusters of
2528 * things that are close together and clusters of things that are not.
2530 * To help reduce the seeks that come with updating all of these reference
2531 * counts, sort them by byte number before actual updates are done.
2533 * struct refsort is used to match byte number to slot in the btree block.
2534 * we sort based on the byte number and then use the slot to actually
2537 * struct refsort is smaller than strcut btrfs_item and smaller than
2538 * struct btrfs_key_ptr. Since we're currently limited to the page size
2539 * for a btree block, there's no way for a kmalloc of refsorts for a
2540 * single node to be bigger than a page.
2548 * for passing into sort()
2550 static int refsort_cmp(const void *a_void
, const void *b_void
)
2552 const struct refsort
*a
= a_void
;
2553 const struct refsort
*b
= b_void
;
2555 if (a
->bytenr
< b
->bytenr
)
2557 if (a
->bytenr
> b
->bytenr
)
2563 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2564 struct btrfs_root
*root
,
2565 struct extent_buffer
*buf
,
2566 int full_backref
, int inc
)
2573 struct btrfs_key key
;
2574 struct btrfs_file_extent_item
*fi
;
2578 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2579 u64
, u64
, u64
, u64
, u64
, u64
);
2581 ref_root
= btrfs_header_owner(buf
);
2582 nritems
= btrfs_header_nritems(buf
);
2583 level
= btrfs_header_level(buf
);
2585 if (!root
->ref_cows
&& level
== 0)
2589 process_func
= btrfs_inc_extent_ref
;
2591 process_func
= btrfs_free_extent
;
2594 parent
= buf
->start
;
2598 for (i
= 0; i
< nritems
; i
++) {
2600 btrfs_item_key_to_cpu(buf
, &key
, i
);
2601 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2603 fi
= btrfs_item_ptr(buf
, i
,
2604 struct btrfs_file_extent_item
);
2605 if (btrfs_file_extent_type(buf
, fi
) ==
2606 BTRFS_FILE_EXTENT_INLINE
)
2608 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2612 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2613 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2614 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2615 parent
, ref_root
, key
.objectid
,
2620 bytenr
= btrfs_node_blockptr(buf
, i
);
2621 num_bytes
= btrfs_level_size(root
, level
- 1);
2622 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2623 parent
, ref_root
, level
- 1, 0);
2634 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2635 struct extent_buffer
*buf
, int full_backref
)
2637 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2640 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2641 struct extent_buffer
*buf
, int full_backref
)
2643 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2646 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2647 struct btrfs_root
*root
,
2648 struct btrfs_path
*path
,
2649 struct btrfs_block_group_cache
*cache
)
2652 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2654 struct extent_buffer
*leaf
;
2656 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2661 leaf
= path
->nodes
[0];
2662 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2663 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2664 btrfs_mark_buffer_dirty(leaf
);
2665 btrfs_release_path(extent_root
, path
);
2673 static struct btrfs_block_group_cache
*
2674 next_block_group(struct btrfs_root
*root
,
2675 struct btrfs_block_group_cache
*cache
)
2677 struct rb_node
*node
;
2678 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2679 node
= rb_next(&cache
->cache_node
);
2680 btrfs_put_block_group(cache
);
2682 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2684 btrfs_get_block_group(cache
);
2687 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2691 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2692 struct btrfs_root
*root
)
2694 struct btrfs_block_group_cache
*cache
;
2696 struct btrfs_path
*path
;
2699 path
= btrfs_alloc_path();
2705 err
= btrfs_run_delayed_refs(trans
, root
,
2710 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2714 cache
= next_block_group(root
, cache
);
2724 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2726 err
= write_one_cache_group(trans
, root
, path
, cache
);
2728 btrfs_put_block_group(cache
);
2731 btrfs_free_path(path
);
2735 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2737 struct btrfs_block_group_cache
*block_group
;
2740 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2741 if (!block_group
|| block_group
->ro
)
2744 btrfs_put_block_group(block_group
);
2748 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2749 u64 total_bytes
, u64 bytes_used
,
2750 struct btrfs_space_info
**space_info
)
2752 struct btrfs_space_info
*found
;
2756 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2757 BTRFS_BLOCK_GROUP_RAID10
))
2762 found
= __find_space_info(info
, flags
);
2764 spin_lock(&found
->lock
);
2765 found
->total_bytes
+= total_bytes
;
2766 found
->bytes_used
+= bytes_used
;
2767 found
->disk_used
+= bytes_used
* factor
;
2769 spin_unlock(&found
->lock
);
2770 *space_info
= found
;
2773 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2777 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
2778 INIT_LIST_HEAD(&found
->block_groups
[i
]);
2779 init_rwsem(&found
->groups_sem
);
2780 spin_lock_init(&found
->lock
);
2781 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
2782 BTRFS_BLOCK_GROUP_SYSTEM
|
2783 BTRFS_BLOCK_GROUP_METADATA
);
2784 found
->total_bytes
= total_bytes
;
2785 found
->bytes_used
= bytes_used
;
2786 found
->disk_used
= bytes_used
* factor
;
2787 found
->bytes_pinned
= 0;
2788 found
->bytes_reserved
= 0;
2789 found
->bytes_readonly
= 0;
2790 found
->bytes_may_use
= 0;
2792 found
->force_alloc
= 0;
2793 *space_info
= found
;
2794 list_add_rcu(&found
->list
, &info
->space_info
);
2795 atomic_set(&found
->caching_threads
, 0);
2799 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2801 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2802 BTRFS_BLOCK_GROUP_RAID1
|
2803 BTRFS_BLOCK_GROUP_RAID10
|
2804 BTRFS_BLOCK_GROUP_DUP
);
2806 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2807 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2808 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2809 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2810 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2811 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2815 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2817 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
2819 if (num_devices
== 1)
2820 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2821 if (num_devices
< 4)
2822 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2824 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2825 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2826 BTRFS_BLOCK_GROUP_RAID10
))) {
2827 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
2830 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
2831 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
2832 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
2835 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
2836 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
2837 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
2838 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
2839 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
2843 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2845 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2846 flags
|= root
->fs_info
->avail_data_alloc_bits
&
2847 root
->fs_info
->data_alloc_profile
;
2848 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2849 flags
|= root
->fs_info
->avail_system_alloc_bits
&
2850 root
->fs_info
->system_alloc_profile
;
2851 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2852 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
2853 root
->fs_info
->metadata_alloc_profile
;
2854 return btrfs_reduce_alloc_profile(root
, flags
);
2857 static u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
2862 flags
= BTRFS_BLOCK_GROUP_DATA
;
2863 else if (root
== root
->fs_info
->chunk_root
)
2864 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
2866 flags
= BTRFS_BLOCK_GROUP_METADATA
;
2868 return get_alloc_profile(root
, flags
);
2871 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
2873 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
2874 BTRFS_BLOCK_GROUP_DATA
);
2878 * This will check the space that the inode allocates from to make sure we have
2879 * enough space for bytes.
2881 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
2883 struct btrfs_space_info
*data_sinfo
;
2884 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2886 int ret
= 0, committed
= 0;
2888 /* make sure bytes are sectorsize aligned */
2889 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
2891 data_sinfo
= BTRFS_I(inode
)->space_info
;
2896 /* make sure we have enough space to handle the data first */
2897 spin_lock(&data_sinfo
->lock
);
2898 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
2899 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
2900 data_sinfo
->bytes_may_use
;
2902 if (used
+ bytes
> data_sinfo
->total_bytes
) {
2903 struct btrfs_trans_handle
*trans
;
2906 * if we don't have enough free bytes in this space then we need
2907 * to alloc a new chunk.
2909 if (!data_sinfo
->full
) {
2912 data_sinfo
->force_alloc
= 1;
2913 spin_unlock(&data_sinfo
->lock
);
2915 alloc_target
= btrfs_get_alloc_profile(root
, 1);
2916 trans
= btrfs_join_transaction(root
, 1);
2918 return PTR_ERR(trans
);
2920 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2921 bytes
+ 2 * 1024 * 1024,
2923 btrfs_end_transaction(trans
, root
);
2928 btrfs_set_inode_space_info(root
, inode
);
2929 data_sinfo
= BTRFS_I(inode
)->space_info
;
2933 spin_unlock(&data_sinfo
->lock
);
2935 /* commit the current transaction and try again */
2936 if (!committed
&& !root
->fs_info
->open_ioctl_trans
) {
2938 trans
= btrfs_join_transaction(root
, 1);
2940 return PTR_ERR(trans
);
2941 ret
= btrfs_commit_transaction(trans
, root
);
2947 printk(KERN_ERR
"no space left, need %llu, %llu bytes_used, "
2948 "%llu bytes_reserved, " "%llu bytes_pinned, "
2949 "%llu bytes_readonly, %llu may use %llu total\n",
2950 (unsigned long long)bytes
,
2951 (unsigned long long)data_sinfo
->bytes_used
,
2952 (unsigned long long)data_sinfo
->bytes_reserved
,
2953 (unsigned long long)data_sinfo
->bytes_pinned
,
2954 (unsigned long long)data_sinfo
->bytes_readonly
,
2955 (unsigned long long)data_sinfo
->bytes_may_use
,
2956 (unsigned long long)data_sinfo
->total_bytes
);
2959 data_sinfo
->bytes_may_use
+= bytes
;
2960 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
2961 spin_unlock(&data_sinfo
->lock
);
2967 * called when we are clearing an delalloc extent from the
2968 * inode's io_tree or there was an error for whatever reason
2969 * after calling btrfs_check_data_free_space
2971 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
2973 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2974 struct btrfs_space_info
*data_sinfo
;
2976 /* make sure bytes are sectorsize aligned */
2977 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
2979 data_sinfo
= BTRFS_I(inode
)->space_info
;
2980 spin_lock(&data_sinfo
->lock
);
2981 data_sinfo
->bytes_may_use
-= bytes
;
2982 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
2983 spin_unlock(&data_sinfo
->lock
);
2986 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
2988 struct list_head
*head
= &info
->space_info
;
2989 struct btrfs_space_info
*found
;
2992 list_for_each_entry_rcu(found
, head
, list
) {
2993 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
2994 found
->force_alloc
= 1;
2999 static int should_alloc_chunk(struct btrfs_space_info
*sinfo
,
3002 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3004 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3005 alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3008 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3009 alloc_bytes
< div_factor(num_bytes
, 8))
3015 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3016 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3017 u64 flags
, int force
)
3019 struct btrfs_space_info
*space_info
;
3020 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3023 mutex_lock(&fs_info
->chunk_mutex
);
3025 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3027 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3029 ret
= update_space_info(extent_root
->fs_info
, flags
,
3033 BUG_ON(!space_info
);
3035 spin_lock(&space_info
->lock
);
3036 if (space_info
->force_alloc
)
3038 if (space_info
->full
) {
3039 spin_unlock(&space_info
->lock
);
3043 if (!force
&& !should_alloc_chunk(space_info
, alloc_bytes
)) {
3044 spin_unlock(&space_info
->lock
);
3047 spin_unlock(&space_info
->lock
);
3050 * if we're doing a data chunk, go ahead and make sure that
3051 * we keep a reasonable number of metadata chunks allocated in the
3054 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3055 fs_info
->data_chunk_allocations
++;
3056 if (!(fs_info
->data_chunk_allocations
%
3057 fs_info
->metadata_ratio
))
3058 force_metadata_allocation(fs_info
);
3061 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3062 spin_lock(&space_info
->lock
);
3064 space_info
->full
= 1;
3067 space_info
->force_alloc
= 0;
3068 spin_unlock(&space_info
->lock
);
3070 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3074 static int maybe_allocate_chunk(struct btrfs_trans_handle
*trans
,
3075 struct btrfs_root
*root
,
3076 struct btrfs_space_info
*sinfo
, u64 num_bytes
)
3084 spin_lock(&sinfo
->lock
);
3085 ret
= should_alloc_chunk(sinfo
, num_bytes
+ 2 * 1024 * 1024);
3086 spin_unlock(&sinfo
->lock
);
3091 trans
= btrfs_join_transaction(root
, 1);
3092 BUG_ON(IS_ERR(trans
));
3096 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3097 num_bytes
+ 2 * 1024 * 1024,
3098 get_alloc_profile(root
, sinfo
->flags
), 0);
3101 btrfs_end_transaction(trans
, root
);
3103 return ret
== 1 ? 1 : 0;
3107 * shrink metadata reservation for delalloc
3109 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3110 struct btrfs_root
*root
, u64 to_reclaim
)
3112 struct btrfs_block_rsv
*block_rsv
;
3119 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3120 spin_lock(&block_rsv
->lock
);
3121 reserved
= block_rsv
->reserved
;
3122 spin_unlock(&block_rsv
->lock
);
3127 max_reclaim
= min(reserved
, to_reclaim
);
3130 ret
= btrfs_start_one_delalloc_inode(root
, trans
? 1 : 0);
3132 __set_current_state(TASK_INTERRUPTIBLE
);
3133 schedule_timeout(pause
);
3135 if (pause
> HZ
/ 10)
3141 spin_lock(&block_rsv
->lock
);
3142 if (reserved
> block_rsv
->reserved
)
3143 reclaimed
= reserved
- block_rsv
->reserved
;
3144 reserved
= block_rsv
->reserved
;
3145 spin_unlock(&block_rsv
->lock
);
3147 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3150 if (trans
&& trans
->transaction
->blocked
)
3153 return reclaimed
>= to_reclaim
;
3156 static int should_retry_reserve(struct btrfs_trans_handle
*trans
,
3157 struct btrfs_root
*root
,
3158 struct btrfs_block_rsv
*block_rsv
,
3159 u64 num_bytes
, int *retries
)
3161 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3167 ret
= maybe_allocate_chunk(trans
, root
, space_info
, num_bytes
);
3171 if (trans
&& trans
->transaction
->in_commit
)
3174 ret
= shrink_delalloc(trans
, root
, num_bytes
);
3178 spin_lock(&space_info
->lock
);
3179 if (space_info
->bytes_pinned
< num_bytes
)
3181 spin_unlock(&space_info
->lock
);
3190 trans
= btrfs_join_transaction(root
, 1);
3191 BUG_ON(IS_ERR(trans
));
3192 ret
= btrfs_commit_transaction(trans
, root
);
3198 static int reserve_metadata_bytes(struct btrfs_block_rsv
*block_rsv
,
3201 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3205 spin_lock(&space_info
->lock
);
3206 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3207 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3209 if (unused
< space_info
->total_bytes
)
3210 unused
= space_info
->total_bytes
- unused
;
3214 if (unused
>= num_bytes
) {
3215 if (block_rsv
->priority
>= 10) {
3216 space_info
->bytes_reserved
+= num_bytes
;
3219 if ((unused
+ block_rsv
->reserved
) *
3220 block_rsv
->priority
>=
3221 (num_bytes
+ block_rsv
->reserved
) * 10) {
3222 space_info
->bytes_reserved
+= num_bytes
;
3227 spin_unlock(&space_info
->lock
);
3232 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3233 struct btrfs_root
*root
)
3235 struct btrfs_block_rsv
*block_rsv
;
3237 block_rsv
= trans
->block_rsv
;
3239 block_rsv
= root
->block_rsv
;
3242 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3247 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3251 spin_lock(&block_rsv
->lock
);
3252 if (block_rsv
->reserved
>= num_bytes
) {
3253 block_rsv
->reserved
-= num_bytes
;
3254 if (block_rsv
->reserved
< block_rsv
->size
)
3255 block_rsv
->full
= 0;
3258 spin_unlock(&block_rsv
->lock
);
3262 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3263 u64 num_bytes
, int update_size
)
3265 spin_lock(&block_rsv
->lock
);
3266 block_rsv
->reserved
+= num_bytes
;
3268 block_rsv
->size
+= num_bytes
;
3269 else if (block_rsv
->reserved
>= block_rsv
->size
)
3270 block_rsv
->full
= 1;
3271 spin_unlock(&block_rsv
->lock
);
3274 void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3275 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3277 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3279 spin_lock(&block_rsv
->lock
);
3280 if (num_bytes
== (u64
)-1)
3281 num_bytes
= block_rsv
->size
;
3282 block_rsv
->size
-= num_bytes
;
3283 if (block_rsv
->reserved
>= block_rsv
->size
) {
3284 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3285 block_rsv
->reserved
= block_rsv
->size
;
3286 block_rsv
->full
= 1;
3290 spin_unlock(&block_rsv
->lock
);
3292 if (num_bytes
> 0) {
3294 block_rsv_add_bytes(dest
, num_bytes
, 0);
3296 spin_lock(&space_info
->lock
);
3297 space_info
->bytes_reserved
-= num_bytes
;
3298 spin_unlock(&space_info
->lock
);
3303 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3304 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3308 ret
= block_rsv_use_bytes(src
, num_bytes
);
3312 block_rsv_add_bytes(dst
, num_bytes
, 1);
3316 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3318 memset(rsv
, 0, sizeof(*rsv
));
3319 spin_lock_init(&rsv
->lock
);
3320 atomic_set(&rsv
->usage
, 1);
3322 INIT_LIST_HEAD(&rsv
->list
);
3325 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3327 struct btrfs_block_rsv
*block_rsv
;
3328 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3331 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3335 btrfs_init_block_rsv(block_rsv
);
3337 alloc_target
= btrfs_get_alloc_profile(root
, 0);
3338 block_rsv
->space_info
= __find_space_info(fs_info
,
3339 BTRFS_BLOCK_GROUP_METADATA
);
3344 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3345 struct btrfs_block_rsv
*rsv
)
3347 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3348 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3355 * make the block_rsv struct be able to capture freed space.
3356 * the captured space will re-add to the the block_rsv struct
3357 * after transaction commit
3359 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3360 struct btrfs_block_rsv
*block_rsv
)
3362 block_rsv
->durable
= 1;
3363 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3364 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3365 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3368 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3369 struct btrfs_root
*root
,
3370 struct btrfs_block_rsv
*block_rsv
,
3371 u64 num_bytes
, int *retries
)
3378 ret
= reserve_metadata_bytes(block_rsv
, num_bytes
);
3380 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3384 ret
= should_retry_reserve(trans
, root
, block_rsv
, num_bytes
, retries
);
3391 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3392 struct btrfs_root
*root
,
3393 struct btrfs_block_rsv
*block_rsv
,
3394 u64 min_reserved
, int min_factor
)
3397 int commit_trans
= 0;
3403 spin_lock(&block_rsv
->lock
);
3405 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3406 if (min_reserved
> num_bytes
)
3407 num_bytes
= min_reserved
;
3409 if (block_rsv
->reserved
>= num_bytes
) {
3412 num_bytes
-= block_rsv
->reserved
;
3413 if (block_rsv
->durable
&&
3414 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3417 spin_unlock(&block_rsv
->lock
);
3421 if (block_rsv
->refill_used
) {
3422 ret
= reserve_metadata_bytes(block_rsv
, num_bytes
);
3424 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3433 trans
= btrfs_join_transaction(root
, 1);
3434 BUG_ON(IS_ERR(trans
));
3435 ret
= btrfs_commit_transaction(trans
, root
);
3440 printk(KERN_INFO
"block_rsv size %llu reserved %llu freed %llu %llu\n",
3441 block_rsv
->size
, block_rsv
->reserved
,
3442 block_rsv
->freed
[0], block_rsv
->freed
[1]);
3447 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3448 struct btrfs_block_rsv
*dst_rsv
,
3451 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3454 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3455 struct btrfs_block_rsv
*block_rsv
,
3458 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3459 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3460 block_rsv
->space_info
!= global_rsv
->space_info
)
3462 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3466 * helper to calculate size of global block reservation.
3467 * the desired value is sum of space used by extent tree,
3468 * checksum tree and root tree
3470 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3472 struct btrfs_space_info
*sinfo
;
3476 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3479 * per tree used space accounting can be inaccuracy, so we
3482 spin_lock(&fs_info
->extent_root
->accounting_lock
);
3483 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
);
3484 spin_unlock(&fs_info
->extent_root
->accounting_lock
);
3486 spin_lock(&fs_info
->csum_root
->accounting_lock
);
3487 num_bytes
+= btrfs_root_used(&fs_info
->csum_root
->root_item
);
3488 spin_unlock(&fs_info
->csum_root
->accounting_lock
);
3490 spin_lock(&fs_info
->tree_root
->accounting_lock
);
3491 num_bytes
+= btrfs_root_used(&fs_info
->tree_root
->root_item
);
3492 spin_unlock(&fs_info
->tree_root
->accounting_lock
);
3494 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3495 spin_lock(&sinfo
->lock
);
3496 data_used
= sinfo
->bytes_used
;
3497 spin_unlock(&sinfo
->lock
);
3499 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3500 spin_lock(&sinfo
->lock
);
3501 meta_used
= sinfo
->bytes_used
;
3502 spin_unlock(&sinfo
->lock
);
3504 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3506 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3508 if (num_bytes
* 3 > meta_used
)
3509 num_bytes
= div64_u64(meta_used
, 3);
3511 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3514 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3516 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3517 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3520 num_bytes
= calc_global_metadata_size(fs_info
);
3522 spin_lock(&block_rsv
->lock
);
3523 spin_lock(&sinfo
->lock
);
3525 block_rsv
->size
= num_bytes
;
3527 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3528 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
;
3530 if (sinfo
->total_bytes
> num_bytes
) {
3531 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3532 block_rsv
->reserved
+= num_bytes
;
3533 sinfo
->bytes_reserved
+= num_bytes
;
3536 if (block_rsv
->reserved
>= block_rsv
->size
) {
3537 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3538 sinfo
->bytes_reserved
-= num_bytes
;
3539 block_rsv
->reserved
= block_rsv
->size
;
3540 block_rsv
->full
= 1;
3543 printk(KERN_INFO
"global block rsv size %llu reserved %llu\n",
3544 block_rsv
->size
, block_rsv
->reserved
);
3546 spin_unlock(&sinfo
->lock
);
3547 spin_unlock(&block_rsv
->lock
);
3550 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3552 struct btrfs_space_info
*space_info
;
3554 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3555 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3556 fs_info
->chunk_block_rsv
.priority
= 10;
3558 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3559 fs_info
->global_block_rsv
.space_info
= space_info
;
3560 fs_info
->global_block_rsv
.priority
= 10;
3561 fs_info
->global_block_rsv
.refill_used
= 1;
3562 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3563 fs_info
->trans_block_rsv
.space_info
= space_info
;
3564 fs_info
->empty_block_rsv
.space_info
= space_info
;
3565 fs_info
->empty_block_rsv
.priority
= 10;
3567 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3568 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3569 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3570 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3571 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3573 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3575 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3577 update_global_block_rsv(fs_info
);
3580 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3582 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3583 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3584 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3585 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3586 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3587 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3588 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3591 static u64
calc_trans_metadata_size(struct btrfs_root
*root
, int num_items
)
3593 return (root
->leafsize
+ root
->nodesize
* (BTRFS_MAX_LEVEL
- 1)) *
3597 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle
*trans
,
3598 struct btrfs_root
*root
,
3599 int num_items
, int *retries
)
3604 if (num_items
== 0 || root
->fs_info
->chunk_root
== root
)
3607 num_bytes
= calc_trans_metadata_size(root
, num_items
);
3608 ret
= btrfs_block_rsv_add(trans
, root
, &root
->fs_info
->trans_block_rsv
,
3609 num_bytes
, retries
);
3611 trans
->bytes_reserved
+= num_bytes
;
3612 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3617 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3618 struct btrfs_root
*root
)
3620 if (!trans
->bytes_reserved
)
3623 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3624 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3625 trans
->bytes_reserved
);
3626 trans
->bytes_reserved
= 0;
3629 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3630 struct inode
*inode
)
3632 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3633 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3634 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3637 * one for deleting orphan item, one for updating inode and
3638 * two for calling btrfs_truncate_inode_items.
3640 * btrfs_truncate_inode_items is a delete operation, it frees
3641 * more space than it uses in most cases. So two units of
3642 * metadata space should be enough for calling it many times.
3643 * If all of the metadata space is used, we can commit
3644 * transaction and use space it freed.
3646 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3647 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3650 void btrfs_orphan_release_metadata(struct inode
*inode
)
3652 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3653 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3654 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3657 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3658 struct btrfs_pending_snapshot
*pending
)
3660 struct btrfs_root
*root
= pending
->root
;
3661 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3662 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3664 * two for root back/forward refs, two for directory entries
3665 * and one for root of the snapshot.
3667 u64 num_bytes
= calc_trans_metadata_size(root
, 5);
3668 dst_rsv
->space_info
= src_rsv
->space_info
;
3669 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3672 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
3674 return num_bytes
>>= 3;
3677 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
3679 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3680 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3686 if (btrfs_transaction_in_commit(root
->fs_info
))
3687 schedule_timeout(1);
3689 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
3691 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
3692 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
) + 1;
3693 if (nr_extents
> BTRFS_I(inode
)->reserved_extents
) {
3694 nr_extents
-= BTRFS_I(inode
)->reserved_extents
;
3695 to_reserve
= calc_trans_metadata_size(root
, nr_extents
);
3701 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
3702 ret
= reserve_metadata_bytes(block_rsv
, to_reserve
);
3704 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
3705 ret
= should_retry_reserve(NULL
, root
, block_rsv
, to_reserve
,
3712 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
3713 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
3714 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
3716 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
3718 if (block_rsv
->size
> 512 * 1024 * 1024)
3719 shrink_delalloc(NULL
, root
, to_reserve
);
3724 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
3726 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3730 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
3731 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
3733 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
3734 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
);
3735 if (nr_extents
< BTRFS_I(inode
)->reserved_extents
) {
3736 nr_extents
= BTRFS_I(inode
)->reserved_extents
- nr_extents
;
3737 BTRFS_I(inode
)->reserved_extents
-= nr_extents
;
3741 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
3743 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
3745 to_free
+= calc_trans_metadata_size(root
, nr_extents
);
3747 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
3751 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
3755 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
3759 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
3761 btrfs_free_reserved_data_space(inode
, num_bytes
);
3768 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
3770 btrfs_delalloc_release_metadata(inode
, num_bytes
);
3771 btrfs_free_reserved_data_space(inode
, num_bytes
);
3774 static int update_block_group(struct btrfs_trans_handle
*trans
,
3775 struct btrfs_root
*root
,
3776 u64 bytenr
, u64 num_bytes
, int alloc
)
3778 struct btrfs_block_group_cache
*cache
;
3779 struct btrfs_fs_info
*info
= root
->fs_info
;
3781 u64 total
= num_bytes
;
3785 /* block accounting for super block */
3786 spin_lock(&info
->delalloc_lock
);
3787 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
3789 old_val
+= num_bytes
;
3791 old_val
-= num_bytes
;
3792 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
3793 spin_unlock(&info
->delalloc_lock
);
3796 cache
= btrfs_lookup_block_group(info
, bytenr
);
3799 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
3800 BTRFS_BLOCK_GROUP_RAID1
|
3801 BTRFS_BLOCK_GROUP_RAID10
))
3805 byte_in_group
= bytenr
- cache
->key
.objectid
;
3806 WARN_ON(byte_in_group
> cache
->key
.offset
);
3808 spin_lock(&cache
->space_info
->lock
);
3809 spin_lock(&cache
->lock
);
3811 old_val
= btrfs_block_group_used(&cache
->item
);
3812 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
3814 old_val
+= num_bytes
;
3815 btrfs_set_block_group_used(&cache
->item
, old_val
);
3816 cache
->reserved
-= num_bytes
;
3817 cache
->space_info
->bytes_reserved
-= num_bytes
;
3818 cache
->space_info
->bytes_used
+= num_bytes
;
3819 cache
->space_info
->disk_used
+= num_bytes
* factor
;
3820 spin_unlock(&cache
->lock
);
3821 spin_unlock(&cache
->space_info
->lock
);
3823 old_val
-= num_bytes
;
3824 btrfs_set_block_group_used(&cache
->item
, old_val
);
3825 cache
->pinned
+= num_bytes
;
3826 cache
->space_info
->bytes_pinned
+= num_bytes
;
3827 cache
->space_info
->bytes_used
-= num_bytes
;
3828 cache
->space_info
->disk_used
-= num_bytes
* factor
;
3829 spin_unlock(&cache
->lock
);
3830 spin_unlock(&cache
->space_info
->lock
);
3832 set_extent_dirty(info
->pinned_extents
,
3833 bytenr
, bytenr
+ num_bytes
- 1,
3834 GFP_NOFS
| __GFP_NOFAIL
);
3836 btrfs_put_block_group(cache
);
3838 bytenr
+= num_bytes
;
3843 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
3845 struct btrfs_block_group_cache
*cache
;
3848 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
3852 bytenr
= cache
->key
.objectid
;
3853 btrfs_put_block_group(cache
);
3858 static int pin_down_extent(struct btrfs_root
*root
,
3859 struct btrfs_block_group_cache
*cache
,
3860 u64 bytenr
, u64 num_bytes
, int reserved
)
3862 spin_lock(&cache
->space_info
->lock
);
3863 spin_lock(&cache
->lock
);
3864 cache
->pinned
+= num_bytes
;
3865 cache
->space_info
->bytes_pinned
+= num_bytes
;
3867 cache
->reserved
-= num_bytes
;
3868 cache
->space_info
->bytes_reserved
-= num_bytes
;
3870 spin_unlock(&cache
->lock
);
3871 spin_unlock(&cache
->space_info
->lock
);
3873 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
3874 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
3879 * this function must be called within transaction
3881 int btrfs_pin_extent(struct btrfs_root
*root
,
3882 u64 bytenr
, u64 num_bytes
, int reserved
)
3884 struct btrfs_block_group_cache
*cache
;
3886 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3889 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
3891 btrfs_put_block_group(cache
);
3896 * update size of reserved extents. this function may return -EAGAIN
3897 * if 'reserve' is true or 'sinfo' is false.
3899 static int update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
3900 u64 num_bytes
, int reserve
, int sinfo
)
3904 struct btrfs_space_info
*space_info
= cache
->space_info
;
3905 spin_lock(&space_info
->lock
);
3906 spin_lock(&cache
->lock
);
3911 cache
->reserved
+= num_bytes
;
3912 space_info
->bytes_reserved
+= num_bytes
;
3916 space_info
->bytes_readonly
+= num_bytes
;
3917 cache
->reserved
-= num_bytes
;
3918 space_info
->bytes_reserved
-= num_bytes
;
3920 spin_unlock(&cache
->lock
);
3921 spin_unlock(&space_info
->lock
);
3923 spin_lock(&cache
->lock
);
3928 cache
->reserved
+= num_bytes
;
3930 cache
->reserved
-= num_bytes
;
3932 spin_unlock(&cache
->lock
);
3937 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
3938 struct btrfs_root
*root
)
3940 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3941 struct btrfs_caching_control
*next
;
3942 struct btrfs_caching_control
*caching_ctl
;
3943 struct btrfs_block_group_cache
*cache
;
3945 down_write(&fs_info
->extent_commit_sem
);
3947 list_for_each_entry_safe(caching_ctl
, next
,
3948 &fs_info
->caching_block_groups
, list
) {
3949 cache
= caching_ctl
->block_group
;
3950 if (block_group_cache_done(cache
)) {
3951 cache
->last_byte_to_unpin
= (u64
)-1;
3952 list_del_init(&caching_ctl
->list
);
3953 put_caching_control(caching_ctl
);
3955 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
3959 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
3960 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
3962 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
3964 up_write(&fs_info
->extent_commit_sem
);
3966 update_global_block_rsv(fs_info
);
3970 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
3972 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3973 struct btrfs_block_group_cache
*cache
= NULL
;
3976 while (start
<= end
) {
3978 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
3980 btrfs_put_block_group(cache
);
3981 cache
= btrfs_lookup_block_group(fs_info
, start
);
3985 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
3986 len
= min(len
, end
+ 1 - start
);
3988 if (start
< cache
->last_byte_to_unpin
) {
3989 len
= min(len
, cache
->last_byte_to_unpin
- start
);
3990 btrfs_add_free_space(cache
, start
, len
);
3995 spin_lock(&cache
->space_info
->lock
);
3996 spin_lock(&cache
->lock
);
3997 cache
->pinned
-= len
;
3998 cache
->space_info
->bytes_pinned
-= len
;
4000 cache
->space_info
->bytes_readonly
+= len
;
4001 } else if (cache
->reserved_pinned
> 0) {
4002 len
= min(len
, cache
->reserved_pinned
);
4003 cache
->reserved_pinned
-= len
;
4004 cache
->space_info
->bytes_reserved
+= len
;
4006 spin_unlock(&cache
->lock
);
4007 spin_unlock(&cache
->space_info
->lock
);
4011 btrfs_put_block_group(cache
);
4015 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4016 struct btrfs_root
*root
)
4018 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4019 struct extent_io_tree
*unpin
;
4020 struct btrfs_block_rsv
*block_rsv
;
4021 struct btrfs_block_rsv
*next_rsv
;
4027 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4028 unpin
= &fs_info
->freed_extents
[1];
4030 unpin
= &fs_info
->freed_extents
[0];
4033 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4038 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
4040 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4041 unpin_extent_range(root
, start
, end
);
4045 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4046 list_for_each_entry_safe(block_rsv
, next_rsv
,
4047 &fs_info
->durable_block_rsv_list
, list
) {
4049 idx
= trans
->transid
& 0x1;
4050 if (block_rsv
->freed
[idx
] > 0) {
4051 block_rsv_add_bytes(block_rsv
,
4052 block_rsv
->freed
[idx
], 0);
4053 block_rsv
->freed
[idx
] = 0;
4055 if (atomic_read(&block_rsv
->usage
) == 0) {
4056 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4058 if (block_rsv
->freed
[0] == 0 &&
4059 block_rsv
->freed
[1] == 0) {
4060 list_del_init(&block_rsv
->list
);
4064 btrfs_block_rsv_release(root
, block_rsv
, 0);
4067 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4072 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4073 struct btrfs_root
*root
,
4074 u64 bytenr
, u64 num_bytes
, u64 parent
,
4075 u64 root_objectid
, u64 owner_objectid
,
4076 u64 owner_offset
, int refs_to_drop
,
4077 struct btrfs_delayed_extent_op
*extent_op
)
4079 struct btrfs_key key
;
4080 struct btrfs_path
*path
;
4081 struct btrfs_fs_info
*info
= root
->fs_info
;
4082 struct btrfs_root
*extent_root
= info
->extent_root
;
4083 struct extent_buffer
*leaf
;
4084 struct btrfs_extent_item
*ei
;
4085 struct btrfs_extent_inline_ref
*iref
;
4088 int extent_slot
= 0;
4089 int found_extent
= 0;
4094 path
= btrfs_alloc_path();
4099 path
->leave_spinning
= 1;
4101 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4102 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4104 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4105 bytenr
, num_bytes
, parent
,
4106 root_objectid
, owner_objectid
,
4109 extent_slot
= path
->slots
[0];
4110 while (extent_slot
>= 0) {
4111 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4113 if (key
.objectid
!= bytenr
)
4115 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4116 key
.offset
== num_bytes
) {
4120 if (path
->slots
[0] - extent_slot
> 5)
4124 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4125 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4126 if (found_extent
&& item_size
< sizeof(*ei
))
4129 if (!found_extent
) {
4131 ret
= remove_extent_backref(trans
, extent_root
, path
,
4135 btrfs_release_path(extent_root
, path
);
4136 path
->leave_spinning
= 1;
4138 key
.objectid
= bytenr
;
4139 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4140 key
.offset
= num_bytes
;
4142 ret
= btrfs_search_slot(trans
, extent_root
,
4145 printk(KERN_ERR
"umm, got %d back from search"
4146 ", was looking for %llu\n", ret
,
4147 (unsigned long long)bytenr
);
4148 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4151 extent_slot
= path
->slots
[0];
4154 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4156 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4157 "parent %llu root %llu owner %llu offset %llu\n",
4158 (unsigned long long)bytenr
,
4159 (unsigned long long)parent
,
4160 (unsigned long long)root_objectid
,
4161 (unsigned long long)owner_objectid
,
4162 (unsigned long long)owner_offset
);
4165 leaf
= path
->nodes
[0];
4166 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4167 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4168 if (item_size
< sizeof(*ei
)) {
4169 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4170 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4174 btrfs_release_path(extent_root
, path
);
4175 path
->leave_spinning
= 1;
4177 key
.objectid
= bytenr
;
4178 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4179 key
.offset
= num_bytes
;
4181 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4184 printk(KERN_ERR
"umm, got %d back from search"
4185 ", was looking for %llu\n", ret
,
4186 (unsigned long long)bytenr
);
4187 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4190 extent_slot
= path
->slots
[0];
4191 leaf
= path
->nodes
[0];
4192 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4195 BUG_ON(item_size
< sizeof(*ei
));
4196 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4197 struct btrfs_extent_item
);
4198 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4199 struct btrfs_tree_block_info
*bi
;
4200 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4201 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4202 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4205 refs
= btrfs_extent_refs(leaf
, ei
);
4206 BUG_ON(refs
< refs_to_drop
);
4207 refs
-= refs_to_drop
;
4211 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4213 * In the case of inline back ref, reference count will
4214 * be updated by remove_extent_backref
4217 BUG_ON(!found_extent
);
4219 btrfs_set_extent_refs(leaf
, ei
, refs
);
4220 btrfs_mark_buffer_dirty(leaf
);
4223 ret
= remove_extent_backref(trans
, extent_root
, path
,
4230 BUG_ON(is_data
&& refs_to_drop
!=
4231 extent_data_ref_count(root
, path
, iref
));
4233 BUG_ON(path
->slots
[0] != extent_slot
);
4235 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4236 path
->slots
[0] = extent_slot
;
4241 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4244 btrfs_release_path(extent_root
, path
);
4247 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4250 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4251 bytenr
>> PAGE_CACHE_SHIFT
,
4252 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4255 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4258 btrfs_free_path(path
);
4263 * when we free an block, it is possible (and likely) that we free the last
4264 * delayed ref for that extent as well. This searches the delayed ref tree for
4265 * a given extent, and if there are no other delayed refs to be processed, it
4266 * removes it from the tree.
4268 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4269 struct btrfs_root
*root
, u64 bytenr
)
4271 struct btrfs_delayed_ref_head
*head
;
4272 struct btrfs_delayed_ref_root
*delayed_refs
;
4273 struct btrfs_delayed_ref_node
*ref
;
4274 struct rb_node
*node
;
4277 delayed_refs
= &trans
->transaction
->delayed_refs
;
4278 spin_lock(&delayed_refs
->lock
);
4279 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4283 node
= rb_prev(&head
->node
.rb_node
);
4287 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4289 /* there are still entries for this ref, we can't drop it */
4290 if (ref
->bytenr
== bytenr
)
4293 if (head
->extent_op
) {
4294 if (!head
->must_insert_reserved
)
4296 kfree(head
->extent_op
);
4297 head
->extent_op
= NULL
;
4301 * waiting for the lock here would deadlock. If someone else has it
4302 * locked they are already in the process of dropping it anyway
4304 if (!mutex_trylock(&head
->mutex
))
4308 * at this point we have a head with no other entries. Go
4309 * ahead and process it.
4311 head
->node
.in_tree
= 0;
4312 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4314 delayed_refs
->num_entries
--;
4317 * we don't take a ref on the node because we're removing it from the
4318 * tree, so we just steal the ref the tree was holding.
4320 delayed_refs
->num_heads
--;
4321 if (list_empty(&head
->cluster
))
4322 delayed_refs
->num_heads_ready
--;
4324 list_del_init(&head
->cluster
);
4325 spin_unlock(&delayed_refs
->lock
);
4327 BUG_ON(head
->extent_op
);
4328 if (head
->must_insert_reserved
)
4331 mutex_unlock(&head
->mutex
);
4332 btrfs_put_delayed_ref(&head
->node
);
4335 spin_unlock(&delayed_refs
->lock
);
4339 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4340 struct btrfs_root
*root
,
4341 struct extent_buffer
*buf
,
4342 u64 parent
, int last_ref
)
4344 struct btrfs_block_rsv
*block_rsv
;
4345 struct btrfs_block_group_cache
*cache
= NULL
;
4348 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4349 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4350 parent
, root
->root_key
.objectid
,
4351 btrfs_header_level(buf
),
4352 BTRFS_DROP_DELAYED_REF
, NULL
);
4359 block_rsv
= get_block_rsv(trans
, root
);
4360 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4361 BUG_ON(block_rsv
->space_info
!= cache
->space_info
);
4363 if (btrfs_header_generation(buf
) == trans
->transid
) {
4364 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4365 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4370 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4371 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4375 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4377 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4378 ret
= update_reserved_bytes(cache
, buf
->len
, 0, 0);
4379 if (ret
== -EAGAIN
) {
4380 /* block group became read-only */
4381 update_reserved_bytes(cache
, buf
->len
, 0, 1);
4386 spin_lock(&block_rsv
->lock
);
4387 if (block_rsv
->reserved
< block_rsv
->size
) {
4388 block_rsv
->reserved
+= buf
->len
;
4391 spin_unlock(&block_rsv
->lock
);
4394 spin_lock(&cache
->space_info
->lock
);
4395 cache
->space_info
->bytes_reserved
-= buf
->len
;
4396 spin_unlock(&cache
->space_info
->lock
);
4401 if (block_rsv
->durable
&& !cache
->ro
) {
4403 spin_lock(&cache
->lock
);
4405 cache
->reserved_pinned
+= buf
->len
;
4408 spin_unlock(&cache
->lock
);
4411 spin_lock(&block_rsv
->lock
);
4412 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4413 spin_unlock(&block_rsv
->lock
);
4417 btrfs_put_block_group(cache
);
4420 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4421 struct btrfs_root
*root
,
4422 u64 bytenr
, u64 num_bytes
, u64 parent
,
4423 u64 root_objectid
, u64 owner
, u64 offset
)
4428 * tree log blocks never actually go into the extent allocation
4429 * tree, just update pinning info and exit early.
4431 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4432 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4433 /* unlocks the pinned mutex */
4434 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4436 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4437 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4438 parent
, root_objectid
, (int)owner
,
4439 BTRFS_DROP_DELAYED_REF
, NULL
);
4442 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4443 parent
, root_objectid
, owner
,
4444 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4450 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4452 u64 mask
= ((u64
)root
->stripesize
- 1);
4453 u64 ret
= (val
+ mask
) & ~mask
;
4458 * when we wait for progress in the block group caching, its because
4459 * our allocation attempt failed at least once. So, we must sleep
4460 * and let some progress happen before we try again.
4462 * This function will sleep at least once waiting for new free space to
4463 * show up, and then it will check the block group free space numbers
4464 * for our min num_bytes. Another option is to have it go ahead
4465 * and look in the rbtree for a free extent of a given size, but this
4469 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4472 struct btrfs_caching_control
*caching_ctl
;
4475 caching_ctl
= get_caching_control(cache
);
4479 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4480 (cache
->free_space
>= num_bytes
));
4482 put_caching_control(caching_ctl
);
4487 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4489 struct btrfs_caching_control
*caching_ctl
;
4492 caching_ctl
= get_caching_control(cache
);
4496 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4498 put_caching_control(caching_ctl
);
4502 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4505 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4507 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4509 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4511 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4518 enum btrfs_loop_type
{
4519 LOOP_FIND_IDEAL
= 0,
4520 LOOP_CACHING_NOWAIT
= 1,
4521 LOOP_CACHING_WAIT
= 2,
4522 LOOP_ALLOC_CHUNK
= 3,
4523 LOOP_NO_EMPTY_SIZE
= 4,
4527 * walks the btree of allocated extents and find a hole of a given size.
4528 * The key ins is changed to record the hole:
4529 * ins->objectid == block start
4530 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4531 * ins->offset == number of blocks
4532 * Any available blocks before search_start are skipped.
4534 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4535 struct btrfs_root
*orig_root
,
4536 u64 num_bytes
, u64 empty_size
,
4537 u64 search_start
, u64 search_end
,
4538 u64 hint_byte
, struct btrfs_key
*ins
,
4542 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4543 struct btrfs_free_cluster
*last_ptr
= NULL
;
4544 struct btrfs_block_group_cache
*block_group
= NULL
;
4545 int empty_cluster
= 2 * 1024 * 1024;
4546 int allowed_chunk_alloc
= 0;
4547 int done_chunk_alloc
= 0;
4548 struct btrfs_space_info
*space_info
;
4549 int last_ptr_loop
= 0;
4552 bool found_uncached_bg
= false;
4553 bool failed_cluster_refill
= false;
4554 bool failed_alloc
= false;
4555 u64 ideal_cache_percent
= 0;
4556 u64 ideal_cache_offset
= 0;
4558 WARN_ON(num_bytes
< root
->sectorsize
);
4559 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4563 space_info
= __find_space_info(root
->fs_info
, data
);
4565 printk(KERN_ERR
"No space info for %d\n", data
);
4569 if (orig_root
->ref_cows
|| empty_size
)
4570 allowed_chunk_alloc
= 1;
4572 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
4573 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4574 if (!btrfs_test_opt(root
, SSD
))
4575 empty_cluster
= 64 * 1024;
4578 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
4579 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4583 spin_lock(&last_ptr
->lock
);
4584 if (last_ptr
->block_group
)
4585 hint_byte
= last_ptr
->window_start
;
4586 spin_unlock(&last_ptr
->lock
);
4589 search_start
= max(search_start
, first_logical_byte(root
, 0));
4590 search_start
= max(search_start
, hint_byte
);
4595 if (search_start
== hint_byte
) {
4597 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4600 * we don't want to use the block group if it doesn't match our
4601 * allocation bits, or if its not cached.
4603 * However if we are re-searching with an ideal block group
4604 * picked out then we don't care that the block group is cached.
4606 if (block_group
&& block_group_bits(block_group
, data
) &&
4607 (block_group
->cached
!= BTRFS_CACHE_NO
||
4608 search_start
== ideal_cache_offset
)) {
4609 down_read(&space_info
->groups_sem
);
4610 if (list_empty(&block_group
->list
) ||
4613 * someone is removing this block group,
4614 * we can't jump into the have_block_group
4615 * target because our list pointers are not
4618 btrfs_put_block_group(block_group
);
4619 up_read(&space_info
->groups_sem
);
4621 index
= get_block_group_index(block_group
);
4622 goto have_block_group
;
4624 } else if (block_group
) {
4625 btrfs_put_block_group(block_group
);
4629 down_read(&space_info
->groups_sem
);
4630 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
4635 btrfs_get_block_group(block_group
);
4636 search_start
= block_group
->key
.objectid
;
4639 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
4642 free_percent
= btrfs_block_group_used(&block_group
->item
);
4643 free_percent
*= 100;
4644 free_percent
= div64_u64(free_percent
,
4645 block_group
->key
.offset
);
4646 free_percent
= 100 - free_percent
;
4647 if (free_percent
> ideal_cache_percent
&&
4648 likely(!block_group
->ro
)) {
4649 ideal_cache_offset
= block_group
->key
.objectid
;
4650 ideal_cache_percent
= free_percent
;
4654 * We only want to start kthread caching if we are at
4655 * the point where we will wait for caching to make
4656 * progress, or if our ideal search is over and we've
4657 * found somebody to start caching.
4659 if (loop
> LOOP_CACHING_NOWAIT
||
4660 (loop
> LOOP_FIND_IDEAL
&&
4661 atomic_read(&space_info
->caching_threads
) < 2)) {
4662 ret
= cache_block_group(block_group
);
4665 found_uncached_bg
= true;
4668 * If loop is set for cached only, try the next block
4671 if (loop
== LOOP_FIND_IDEAL
)
4675 cached
= block_group_cache_done(block_group
);
4676 if (unlikely(!cached
))
4677 found_uncached_bg
= true;
4679 if (unlikely(block_group
->ro
))
4683 * Ok we want to try and use the cluster allocator, so lets look
4684 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4685 * have tried the cluster allocator plenty of times at this
4686 * point and not have found anything, so we are likely way too
4687 * fragmented for the clustering stuff to find anything, so lets
4688 * just skip it and let the allocator find whatever block it can
4691 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
4693 * the refill lock keeps out other
4694 * people trying to start a new cluster
4696 spin_lock(&last_ptr
->refill_lock
);
4697 if (last_ptr
->block_group
&&
4698 (last_ptr
->block_group
->ro
||
4699 !block_group_bits(last_ptr
->block_group
, data
))) {
4701 goto refill_cluster
;
4704 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
4705 num_bytes
, search_start
);
4707 /* we have a block, we're done */
4708 spin_unlock(&last_ptr
->refill_lock
);
4712 spin_lock(&last_ptr
->lock
);
4714 * whoops, this cluster doesn't actually point to
4715 * this block group. Get a ref on the block
4716 * group is does point to and try again
4718 if (!last_ptr_loop
&& last_ptr
->block_group
&&
4719 last_ptr
->block_group
!= block_group
) {
4721 btrfs_put_block_group(block_group
);
4722 block_group
= last_ptr
->block_group
;
4723 btrfs_get_block_group(block_group
);
4724 spin_unlock(&last_ptr
->lock
);
4725 spin_unlock(&last_ptr
->refill_lock
);
4728 search_start
= block_group
->key
.objectid
;
4730 * we know this block group is properly
4731 * in the list because
4732 * btrfs_remove_block_group, drops the
4733 * cluster before it removes the block
4734 * group from the list
4736 goto have_block_group
;
4738 spin_unlock(&last_ptr
->lock
);
4741 * this cluster didn't work out, free it and
4744 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
4748 /* allocate a cluster in this block group */
4749 ret
= btrfs_find_space_cluster(trans
, root
,
4750 block_group
, last_ptr
,
4752 empty_cluster
+ empty_size
);
4755 * now pull our allocation out of this
4758 offset
= btrfs_alloc_from_cluster(block_group
,
4759 last_ptr
, num_bytes
,
4762 /* we found one, proceed */
4763 spin_unlock(&last_ptr
->refill_lock
);
4766 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
4767 && !failed_cluster_refill
) {
4768 spin_unlock(&last_ptr
->refill_lock
);
4770 failed_cluster_refill
= true;
4771 wait_block_group_cache_progress(block_group
,
4772 num_bytes
+ empty_cluster
+ empty_size
);
4773 goto have_block_group
;
4777 * at this point we either didn't find a cluster
4778 * or we weren't able to allocate a block from our
4779 * cluster. Free the cluster we've been trying
4780 * to use, and go to the next block group
4782 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
4783 spin_unlock(&last_ptr
->refill_lock
);
4787 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
4788 num_bytes
, empty_size
);
4790 * If we didn't find a chunk, and we haven't failed on this
4791 * block group before, and this block group is in the middle of
4792 * caching and we are ok with waiting, then go ahead and wait
4793 * for progress to be made, and set failed_alloc to true.
4795 * If failed_alloc is true then we've already waited on this
4796 * block group once and should move on to the next block group.
4798 if (!offset
&& !failed_alloc
&& !cached
&&
4799 loop
> LOOP_CACHING_NOWAIT
) {
4800 wait_block_group_cache_progress(block_group
,
4801 num_bytes
+ empty_size
);
4802 failed_alloc
= true;
4803 goto have_block_group
;
4804 } else if (!offset
) {
4808 search_start
= stripe_align(root
, offset
);
4809 /* move on to the next group */
4810 if (search_start
+ num_bytes
>= search_end
) {
4811 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4815 /* move on to the next group */
4816 if (search_start
+ num_bytes
>
4817 block_group
->key
.objectid
+ block_group
->key
.offset
) {
4818 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4822 ins
->objectid
= search_start
;
4823 ins
->offset
= num_bytes
;
4825 if (offset
< search_start
)
4826 btrfs_add_free_space(block_group
, offset
,
4827 search_start
- offset
);
4828 BUG_ON(offset
> search_start
);
4830 ret
= update_reserved_bytes(block_group
, num_bytes
, 1,
4831 (data
& BTRFS_BLOCK_GROUP_DATA
));
4832 if (ret
== -EAGAIN
) {
4833 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4837 /* we are all good, lets return */
4838 ins
->objectid
= search_start
;
4839 ins
->offset
= num_bytes
;
4841 if (offset
< search_start
)
4842 btrfs_add_free_space(block_group
, offset
,
4843 search_start
- offset
);
4844 BUG_ON(offset
> search_start
);
4847 failed_cluster_refill
= false;
4848 failed_alloc
= false;
4849 BUG_ON(index
!= get_block_group_index(block_group
));
4850 btrfs_put_block_group(block_group
);
4852 up_read(&space_info
->groups_sem
);
4854 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
4857 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4858 * for them to make caching progress. Also
4859 * determine the best possible bg to cache
4860 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4861 * caching kthreads as we move along
4862 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4863 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4864 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4867 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
4868 (found_uncached_bg
|| empty_size
|| empty_cluster
||
4869 allowed_chunk_alloc
)) {
4871 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
4872 found_uncached_bg
= false;
4874 if (!ideal_cache_percent
&&
4875 atomic_read(&space_info
->caching_threads
))
4879 * 1 of the following 2 things have happened so far
4881 * 1) We found an ideal block group for caching that
4882 * is mostly full and will cache quickly, so we might
4883 * as well wait for it.
4885 * 2) We searched for cached only and we didn't find
4886 * anything, and we didn't start any caching kthreads
4887 * either, so chances are we will loop through and
4888 * start a couple caching kthreads, and then come back
4889 * around and just wait for them. This will be slower
4890 * because we will have 2 caching kthreads reading at
4891 * the same time when we could have just started one
4892 * and waited for it to get far enough to give us an
4893 * allocation, so go ahead and go to the wait caching
4896 loop
= LOOP_CACHING_WAIT
;
4897 search_start
= ideal_cache_offset
;
4898 ideal_cache_percent
= 0;
4900 } else if (loop
== LOOP_FIND_IDEAL
) {
4902 * Didn't find a uncached bg, wait on anything we find
4905 loop
= LOOP_CACHING_WAIT
;
4909 if (loop
< LOOP_CACHING_WAIT
) {
4914 if (loop
== LOOP_ALLOC_CHUNK
) {
4919 if (allowed_chunk_alloc
) {
4920 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
4921 2 * 1024 * 1024, data
, 1);
4922 allowed_chunk_alloc
= 0;
4923 done_chunk_alloc
= 1;
4924 } else if (!done_chunk_alloc
) {
4925 space_info
->force_alloc
= 1;
4928 if (loop
< LOOP_NO_EMPTY_SIZE
) {
4933 } else if (!ins
->objectid
) {
4937 /* we found what we needed */
4938 if (ins
->objectid
) {
4939 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
4940 trans
->block_group
= block_group
->key
.objectid
;
4942 btrfs_put_block_group(block_group
);
4949 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
4950 int dump_block_groups
)
4952 struct btrfs_block_group_cache
*cache
;
4955 spin_lock(&info
->lock
);
4956 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
4957 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
4958 info
->bytes_pinned
- info
->bytes_reserved
-
4959 info
->bytes_readonly
),
4960 (info
->full
) ? "" : "not ");
4961 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
4962 "reserved=%llu, may_use=%llu, readonly=%llu\n",
4963 (unsigned long long)info
->total_bytes
,
4964 (unsigned long long)info
->bytes_used
,
4965 (unsigned long long)info
->bytes_pinned
,
4966 (unsigned long long)info
->bytes_reserved
,
4967 (unsigned long long)info
->bytes_may_use
,
4968 (unsigned long long)info
->bytes_readonly
);
4969 spin_unlock(&info
->lock
);
4971 if (!dump_block_groups
)
4974 down_read(&info
->groups_sem
);
4976 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
4977 spin_lock(&cache
->lock
);
4978 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
4979 "%llu pinned %llu reserved\n",
4980 (unsigned long long)cache
->key
.objectid
,
4981 (unsigned long long)cache
->key
.offset
,
4982 (unsigned long long)btrfs_block_group_used(&cache
->item
),
4983 (unsigned long long)cache
->pinned
,
4984 (unsigned long long)cache
->reserved
);
4985 btrfs_dump_free_space(cache
, bytes
);
4986 spin_unlock(&cache
->lock
);
4988 if (++index
< BTRFS_NR_RAID_TYPES
)
4990 up_read(&info
->groups_sem
);
4993 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
4994 struct btrfs_root
*root
,
4995 u64 num_bytes
, u64 min_alloc_size
,
4996 u64 empty_size
, u64 hint_byte
,
4997 u64 search_end
, struct btrfs_key
*ins
,
5001 u64 search_start
= 0;
5003 data
= btrfs_get_alloc_profile(root
, data
);
5006 * the only place that sets empty_size is btrfs_realloc_node, which
5007 * is not called recursively on allocations
5009 if (empty_size
|| root
->ref_cows
)
5010 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5011 num_bytes
+ 2 * 1024 * 1024, data
, 0);
5013 WARN_ON(num_bytes
< root
->sectorsize
);
5014 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5015 search_start
, search_end
, hint_byte
,
5018 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5019 num_bytes
= num_bytes
>> 1;
5020 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5021 num_bytes
= max(num_bytes
, min_alloc_size
);
5022 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5023 num_bytes
, data
, 1);
5026 if (ret
== -ENOSPC
) {
5027 struct btrfs_space_info
*sinfo
;
5029 sinfo
= __find_space_info(root
->fs_info
, data
);
5030 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5031 "wanted %llu\n", (unsigned long long)data
,
5032 (unsigned long long)num_bytes
);
5033 dump_space_info(sinfo
, num_bytes
, 1);
5039 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5041 struct btrfs_block_group_cache
*cache
;
5044 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5046 printk(KERN_ERR
"Unable to find block group for %llu\n",
5047 (unsigned long long)start
);
5051 ret
= btrfs_discard_extent(root
, start
, len
);
5053 btrfs_add_free_space(cache
, start
, len
);
5054 update_reserved_bytes(cache
, len
, 0, 1);
5055 btrfs_put_block_group(cache
);
5060 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5061 struct btrfs_root
*root
,
5062 u64 parent
, u64 root_objectid
,
5063 u64 flags
, u64 owner
, u64 offset
,
5064 struct btrfs_key
*ins
, int ref_mod
)
5067 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5068 struct btrfs_extent_item
*extent_item
;
5069 struct btrfs_extent_inline_ref
*iref
;
5070 struct btrfs_path
*path
;
5071 struct extent_buffer
*leaf
;
5076 type
= BTRFS_SHARED_DATA_REF_KEY
;
5078 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5080 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5082 path
= btrfs_alloc_path();
5085 path
->leave_spinning
= 1;
5086 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5090 leaf
= path
->nodes
[0];
5091 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5092 struct btrfs_extent_item
);
5093 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5094 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5095 btrfs_set_extent_flags(leaf
, extent_item
,
5096 flags
| BTRFS_EXTENT_FLAG_DATA
);
5098 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5099 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5101 struct btrfs_shared_data_ref
*ref
;
5102 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5103 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5104 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5106 struct btrfs_extent_data_ref
*ref
;
5107 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5108 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5109 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5110 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5111 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5114 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5115 btrfs_free_path(path
);
5117 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5119 printk(KERN_ERR
"btrfs update block group failed for %llu "
5120 "%llu\n", (unsigned long long)ins
->objectid
,
5121 (unsigned long long)ins
->offset
);
5127 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5128 struct btrfs_root
*root
,
5129 u64 parent
, u64 root_objectid
,
5130 u64 flags
, struct btrfs_disk_key
*key
,
5131 int level
, struct btrfs_key
*ins
)
5134 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5135 struct btrfs_extent_item
*extent_item
;
5136 struct btrfs_tree_block_info
*block_info
;
5137 struct btrfs_extent_inline_ref
*iref
;
5138 struct btrfs_path
*path
;
5139 struct extent_buffer
*leaf
;
5140 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5142 path
= btrfs_alloc_path();
5145 path
->leave_spinning
= 1;
5146 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5150 leaf
= path
->nodes
[0];
5151 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5152 struct btrfs_extent_item
);
5153 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5154 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5155 btrfs_set_extent_flags(leaf
, extent_item
,
5156 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5157 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5159 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5160 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5162 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5164 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5165 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5166 BTRFS_SHARED_BLOCK_REF_KEY
);
5167 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5169 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5170 BTRFS_TREE_BLOCK_REF_KEY
);
5171 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5174 btrfs_mark_buffer_dirty(leaf
);
5175 btrfs_free_path(path
);
5177 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5179 printk(KERN_ERR
"btrfs update block group failed for %llu "
5180 "%llu\n", (unsigned long long)ins
->objectid
,
5181 (unsigned long long)ins
->offset
);
5187 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5188 struct btrfs_root
*root
,
5189 u64 root_objectid
, u64 owner
,
5190 u64 offset
, struct btrfs_key
*ins
)
5194 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5196 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5197 0, root_objectid
, owner
, offset
,
5198 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5203 * this is used by the tree logging recovery code. It records that
5204 * an extent has been allocated and makes sure to clear the free
5205 * space cache bits as well
5207 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5208 struct btrfs_root
*root
,
5209 u64 root_objectid
, u64 owner
, u64 offset
,
5210 struct btrfs_key
*ins
)
5213 struct btrfs_block_group_cache
*block_group
;
5214 struct btrfs_caching_control
*caching_ctl
;
5215 u64 start
= ins
->objectid
;
5216 u64 num_bytes
= ins
->offset
;
5218 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5219 cache_block_group(block_group
);
5220 caching_ctl
= get_caching_control(block_group
);
5223 BUG_ON(!block_group_cache_done(block_group
));
5224 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5227 mutex_lock(&caching_ctl
->mutex
);
5229 if (start
>= caching_ctl
->progress
) {
5230 ret
= add_excluded_extent(root
, start
, num_bytes
);
5232 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5233 ret
= btrfs_remove_free_space(block_group
,
5237 num_bytes
= caching_ctl
->progress
- start
;
5238 ret
= btrfs_remove_free_space(block_group
,
5242 start
= caching_ctl
->progress
;
5243 num_bytes
= ins
->objectid
+ ins
->offset
-
5244 caching_ctl
->progress
;
5245 ret
= add_excluded_extent(root
, start
, num_bytes
);
5249 mutex_unlock(&caching_ctl
->mutex
);
5250 put_caching_control(caching_ctl
);
5253 ret
= update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5255 btrfs_put_block_group(block_group
);
5256 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5257 0, owner
, offset
, ins
, 1);
5261 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5262 struct btrfs_root
*root
,
5263 u64 bytenr
, u32 blocksize
,
5266 struct extent_buffer
*buf
;
5268 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5270 return ERR_PTR(-ENOMEM
);
5271 btrfs_set_header_generation(buf
, trans
->transid
);
5272 btrfs_set_buffer_lockdep_class(buf
, level
);
5273 btrfs_tree_lock(buf
);
5274 clean_tree_block(trans
, root
, buf
);
5276 btrfs_set_lock_blocking(buf
);
5277 btrfs_set_buffer_uptodate(buf
);
5279 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5281 * we allow two log transactions at a time, use different
5282 * EXENT bit to differentiate dirty pages.
5284 if (root
->log_transid
% 2 == 0)
5285 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5286 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5288 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5289 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5291 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5292 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5294 trans
->blocks_used
++;
5295 /* this returns a buffer locked for blocking */
5299 static struct btrfs_block_rsv
*
5300 use_block_rsv(struct btrfs_trans_handle
*trans
,
5301 struct btrfs_root
*root
, u32 blocksize
)
5303 struct btrfs_block_rsv
*block_rsv
;
5306 block_rsv
= get_block_rsv(trans
, root
);
5308 if (block_rsv
->size
== 0) {
5309 ret
= reserve_metadata_bytes(block_rsv
, blocksize
);
5311 return ERR_PTR(ret
);
5315 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5320 printk(KERN_INFO
"block_rsv size %llu reserved %llu freed %llu %llu\n",
5321 block_rsv
->size
, block_rsv
->reserved
,
5322 block_rsv
->freed
[0], block_rsv
->freed
[1]);
5324 return ERR_PTR(-ENOSPC
);
5327 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5329 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5330 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5334 * finds a free extent and does all the dirty work required for allocation
5335 * returns the key for the extent through ins, and a tree buffer for
5336 * the first block of the extent through buf.
5338 * returns the tree buffer or NULL.
5340 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5341 struct btrfs_root
*root
, u32 blocksize
,
5342 u64 parent
, u64 root_objectid
,
5343 struct btrfs_disk_key
*key
, int level
,
5344 u64 hint
, u64 empty_size
)
5346 struct btrfs_key ins
;
5347 struct btrfs_block_rsv
*block_rsv
;
5348 struct extent_buffer
*buf
;
5353 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5354 if (IS_ERR(block_rsv
))
5355 return ERR_CAST(block_rsv
);
5357 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5358 empty_size
, hint
, (u64
)-1, &ins
, 0);
5360 unuse_block_rsv(block_rsv
, blocksize
);
5361 return ERR_PTR(ret
);
5364 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5366 BUG_ON(IS_ERR(buf
));
5368 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5370 parent
= ins
.objectid
;
5371 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5375 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5376 struct btrfs_delayed_extent_op
*extent_op
;
5377 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5380 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5382 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5383 extent_op
->flags_to_set
= flags
;
5384 extent_op
->update_key
= 1;
5385 extent_op
->update_flags
= 1;
5386 extent_op
->is_data
= 0;
5388 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5389 ins
.offset
, parent
, root_objectid
,
5390 level
, BTRFS_ADD_DELAYED_EXTENT
,
5397 struct walk_control
{
5398 u64 refs
[BTRFS_MAX_LEVEL
];
5399 u64 flags
[BTRFS_MAX_LEVEL
];
5400 struct btrfs_key update_progress
;
5410 #define DROP_REFERENCE 1
5411 #define UPDATE_BACKREF 2
5413 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5414 struct btrfs_root
*root
,
5415 struct walk_control
*wc
,
5416 struct btrfs_path
*path
)
5425 struct btrfs_key key
;
5426 struct extent_buffer
*eb
;
5431 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5432 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5433 wc
->reada_count
= max(wc
->reada_count
, 2);
5435 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5436 wc
->reada_count
= min_t(int, wc
->reada_count
,
5437 BTRFS_NODEPTRS_PER_BLOCK(root
));
5440 eb
= path
->nodes
[wc
->level
];
5441 nritems
= btrfs_header_nritems(eb
);
5442 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5444 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5445 if (nread
>= wc
->reada_count
)
5449 bytenr
= btrfs_node_blockptr(eb
, slot
);
5450 generation
= btrfs_node_ptr_generation(eb
, slot
);
5452 if (slot
== path
->slots
[wc
->level
])
5455 if (wc
->stage
== UPDATE_BACKREF
&&
5456 generation
<= root
->root_key
.offset
)
5459 /* We don't lock the tree block, it's OK to be racy here */
5460 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5465 if (wc
->stage
== DROP_REFERENCE
) {
5469 if (wc
->level
== 1 &&
5470 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5472 if (!wc
->update_ref
||
5473 generation
<= root
->root_key
.offset
)
5475 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5476 ret
= btrfs_comp_cpu_keys(&key
,
5477 &wc
->update_progress
);
5481 if (wc
->level
== 1 &&
5482 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5486 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5490 last
= bytenr
+ blocksize
;
5493 wc
->reada_slot
= slot
;
5497 * hepler to process tree block while walking down the tree.
5499 * when wc->stage == UPDATE_BACKREF, this function updates
5500 * back refs for pointers in the block.
5502 * NOTE: return value 1 means we should stop walking down.
5504 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5505 struct btrfs_root
*root
,
5506 struct btrfs_path
*path
,
5507 struct walk_control
*wc
, int lookup_info
)
5509 int level
= wc
->level
;
5510 struct extent_buffer
*eb
= path
->nodes
[level
];
5511 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5514 if (wc
->stage
== UPDATE_BACKREF
&&
5515 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5519 * when reference count of tree block is 1, it won't increase
5520 * again. once full backref flag is set, we never clear it.
5523 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5524 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5525 BUG_ON(!path
->locks
[level
]);
5526 ret
= btrfs_lookup_extent_info(trans
, root
,
5531 BUG_ON(wc
->refs
[level
] == 0);
5534 if (wc
->stage
== DROP_REFERENCE
) {
5535 if (wc
->refs
[level
] > 1)
5538 if (path
->locks
[level
] && !wc
->keep_locks
) {
5539 btrfs_tree_unlock(eb
);
5540 path
->locks
[level
] = 0;
5545 /* wc->stage == UPDATE_BACKREF */
5546 if (!(wc
->flags
[level
] & flag
)) {
5547 BUG_ON(!path
->locks
[level
]);
5548 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5550 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5552 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5555 wc
->flags
[level
] |= flag
;
5559 * the block is shared by multiple trees, so it's not good to
5560 * keep the tree lock
5562 if (path
->locks
[level
] && level
> 0) {
5563 btrfs_tree_unlock(eb
);
5564 path
->locks
[level
] = 0;
5570 * hepler to process tree block pointer.
5572 * when wc->stage == DROP_REFERENCE, this function checks
5573 * reference count of the block pointed to. if the block
5574 * is shared and we need update back refs for the subtree
5575 * rooted at the block, this function changes wc->stage to
5576 * UPDATE_BACKREF. if the block is shared and there is no
5577 * need to update back, this function drops the reference
5580 * NOTE: return value 1 means we should stop walking down.
5582 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5583 struct btrfs_root
*root
,
5584 struct btrfs_path
*path
,
5585 struct walk_control
*wc
, int *lookup_info
)
5591 struct btrfs_key key
;
5592 struct extent_buffer
*next
;
5593 int level
= wc
->level
;
5597 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
5598 path
->slots
[level
]);
5600 * if the lower level block was created before the snapshot
5601 * was created, we know there is no need to update back refs
5604 if (wc
->stage
== UPDATE_BACKREF
&&
5605 generation
<= root
->root_key
.offset
) {
5610 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
5611 blocksize
= btrfs_level_size(root
, level
- 1);
5613 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
5615 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5620 btrfs_tree_lock(next
);
5621 btrfs_set_lock_blocking(next
);
5623 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5624 &wc
->refs
[level
- 1],
5625 &wc
->flags
[level
- 1]);
5627 BUG_ON(wc
->refs
[level
- 1] == 0);
5630 if (wc
->stage
== DROP_REFERENCE
) {
5631 if (wc
->refs
[level
- 1] > 1) {
5633 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5636 if (!wc
->update_ref
||
5637 generation
<= root
->root_key
.offset
)
5640 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
5641 path
->slots
[level
]);
5642 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
5646 wc
->stage
= UPDATE_BACKREF
;
5647 wc
->shared_level
= level
- 1;
5651 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5655 if (!btrfs_buffer_uptodate(next
, generation
)) {
5656 btrfs_tree_unlock(next
);
5657 free_extent_buffer(next
);
5663 if (reada
&& level
== 1)
5664 reada_walk_down(trans
, root
, wc
, path
);
5665 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
5666 btrfs_tree_lock(next
);
5667 btrfs_set_lock_blocking(next
);
5671 BUG_ON(level
!= btrfs_header_level(next
));
5672 path
->nodes
[level
] = next
;
5673 path
->slots
[level
] = 0;
5674 path
->locks
[level
] = 1;
5680 wc
->refs
[level
- 1] = 0;
5681 wc
->flags
[level
- 1] = 0;
5682 if (wc
->stage
== DROP_REFERENCE
) {
5683 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
5684 parent
= path
->nodes
[level
]->start
;
5686 BUG_ON(root
->root_key
.objectid
!=
5687 btrfs_header_owner(path
->nodes
[level
]));
5691 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
5692 root
->root_key
.objectid
, level
- 1, 0);
5695 btrfs_tree_unlock(next
);
5696 free_extent_buffer(next
);
5702 * hepler to process tree block while walking up the tree.
5704 * when wc->stage == DROP_REFERENCE, this function drops
5705 * reference count on the block.
5707 * when wc->stage == UPDATE_BACKREF, this function changes
5708 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5709 * to UPDATE_BACKREF previously while processing the block.
5711 * NOTE: return value 1 means we should stop walking up.
5713 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
5714 struct btrfs_root
*root
,
5715 struct btrfs_path
*path
,
5716 struct walk_control
*wc
)
5719 int level
= wc
->level
;
5720 struct extent_buffer
*eb
= path
->nodes
[level
];
5723 if (wc
->stage
== UPDATE_BACKREF
) {
5724 BUG_ON(wc
->shared_level
< level
);
5725 if (level
< wc
->shared_level
)
5728 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
5732 wc
->stage
= DROP_REFERENCE
;
5733 wc
->shared_level
= -1;
5734 path
->slots
[level
] = 0;
5737 * check reference count again if the block isn't locked.
5738 * we should start walking down the tree again if reference
5741 if (!path
->locks
[level
]) {
5743 btrfs_tree_lock(eb
);
5744 btrfs_set_lock_blocking(eb
);
5745 path
->locks
[level
] = 1;
5747 ret
= btrfs_lookup_extent_info(trans
, root
,
5752 BUG_ON(wc
->refs
[level
] == 0);
5753 if (wc
->refs
[level
] == 1) {
5754 btrfs_tree_unlock(eb
);
5755 path
->locks
[level
] = 0;
5761 /* wc->stage == DROP_REFERENCE */
5762 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
5764 if (wc
->refs
[level
] == 1) {
5766 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5767 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
5769 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5772 /* make block locked assertion in clean_tree_block happy */
5773 if (!path
->locks
[level
] &&
5774 btrfs_header_generation(eb
) == trans
->transid
) {
5775 btrfs_tree_lock(eb
);
5776 btrfs_set_lock_blocking(eb
);
5777 path
->locks
[level
] = 1;
5779 clean_tree_block(trans
, root
, eb
);
5782 if (eb
== root
->node
) {
5783 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5786 BUG_ON(root
->root_key
.objectid
!=
5787 btrfs_header_owner(eb
));
5789 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5790 parent
= path
->nodes
[level
+ 1]->start
;
5792 BUG_ON(root
->root_key
.objectid
!=
5793 btrfs_header_owner(path
->nodes
[level
+ 1]));
5796 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
5798 wc
->refs
[level
] = 0;
5799 wc
->flags
[level
] = 0;
5803 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
5804 struct btrfs_root
*root
,
5805 struct btrfs_path
*path
,
5806 struct walk_control
*wc
)
5808 int level
= wc
->level
;
5809 int lookup_info
= 1;
5812 while (level
>= 0) {
5813 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
5820 if (path
->slots
[level
] >=
5821 btrfs_header_nritems(path
->nodes
[level
]))
5824 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
5826 path
->slots
[level
]++;
5835 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
5836 struct btrfs_root
*root
,
5837 struct btrfs_path
*path
,
5838 struct walk_control
*wc
, int max_level
)
5840 int level
= wc
->level
;
5843 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
5844 while (level
< max_level
&& path
->nodes
[level
]) {
5846 if (path
->slots
[level
] + 1 <
5847 btrfs_header_nritems(path
->nodes
[level
])) {
5848 path
->slots
[level
]++;
5851 ret
= walk_up_proc(trans
, root
, path
, wc
);
5855 if (path
->locks
[level
]) {
5856 btrfs_tree_unlock(path
->nodes
[level
]);
5857 path
->locks
[level
] = 0;
5859 free_extent_buffer(path
->nodes
[level
]);
5860 path
->nodes
[level
] = NULL
;
5868 * drop a subvolume tree.
5870 * this function traverses the tree freeing any blocks that only
5871 * referenced by the tree.
5873 * when a shared tree block is found. this function decreases its
5874 * reference count by one. if update_ref is true, this function
5875 * also make sure backrefs for the shared block and all lower level
5876 * blocks are properly updated.
5878 int btrfs_drop_snapshot(struct btrfs_root
*root
, int update_ref
)
5880 struct btrfs_path
*path
;
5881 struct btrfs_trans_handle
*trans
;
5882 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
5883 struct btrfs_root_item
*root_item
= &root
->root_item
;
5884 struct walk_control
*wc
;
5885 struct btrfs_key key
;
5890 path
= btrfs_alloc_path();
5893 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5896 trans
= btrfs_start_transaction(tree_root
, 0);
5898 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
5899 level
= btrfs_header_level(root
->node
);
5900 path
->nodes
[level
] = btrfs_lock_root_node(root
);
5901 btrfs_set_lock_blocking(path
->nodes
[level
]);
5902 path
->slots
[level
] = 0;
5903 path
->locks
[level
] = 1;
5904 memset(&wc
->update_progress
, 0,
5905 sizeof(wc
->update_progress
));
5907 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
5908 memcpy(&wc
->update_progress
, &key
,
5909 sizeof(wc
->update_progress
));
5911 level
= root_item
->drop_level
;
5913 path
->lowest_level
= level
;
5914 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5915 path
->lowest_level
= 0;
5923 * unlock our path, this is safe because only this
5924 * function is allowed to delete this snapshot
5926 btrfs_unlock_up_safe(path
, 0);
5928 level
= btrfs_header_level(root
->node
);
5930 btrfs_tree_lock(path
->nodes
[level
]);
5931 btrfs_set_lock_blocking(path
->nodes
[level
]);
5933 ret
= btrfs_lookup_extent_info(trans
, root
,
5934 path
->nodes
[level
]->start
,
5935 path
->nodes
[level
]->len
,
5939 BUG_ON(wc
->refs
[level
] == 0);
5941 if (level
== root_item
->drop_level
)
5944 btrfs_tree_unlock(path
->nodes
[level
]);
5945 WARN_ON(wc
->refs
[level
] != 1);
5951 wc
->shared_level
= -1;
5952 wc
->stage
= DROP_REFERENCE
;
5953 wc
->update_ref
= update_ref
;
5955 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
5958 ret
= walk_down_tree(trans
, root
, path
, wc
);
5964 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
5971 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
5975 if (wc
->stage
== DROP_REFERENCE
) {
5977 btrfs_node_key(path
->nodes
[level
],
5978 &root_item
->drop_progress
,
5979 path
->slots
[level
]);
5980 root_item
->drop_level
= level
;
5983 BUG_ON(wc
->level
== 0);
5984 if (trans
->transaction
->in_commit
||
5985 trans
->transaction
->delayed_refs
.flushing
) {
5986 ret
= btrfs_update_root(trans
, tree_root
,
5991 btrfs_end_transaction(trans
, tree_root
);
5992 trans
= btrfs_start_transaction(tree_root
, 0);
5994 return PTR_ERR(trans
);
5996 unsigned long update
;
5997 update
= trans
->delayed_ref_updates
;
5998 trans
->delayed_ref_updates
= 0;
6000 btrfs_run_delayed_refs(trans
, tree_root
,
6004 btrfs_release_path(root
, path
);
6007 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6010 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6011 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6015 ret
= btrfs_del_orphan_item(trans
, tree_root
,
6016 root
->root_key
.objectid
);
6021 if (root
->in_radix
) {
6022 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6024 free_extent_buffer(root
->node
);
6025 free_extent_buffer(root
->commit_root
);
6029 btrfs_end_transaction(trans
, tree_root
);
6031 btrfs_free_path(path
);
6036 * drop subtree rooted at tree block 'node'.
6038 * NOTE: this function will unlock and release tree block 'node'
6040 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6041 struct btrfs_root
*root
,
6042 struct extent_buffer
*node
,
6043 struct extent_buffer
*parent
)
6045 struct btrfs_path
*path
;
6046 struct walk_control
*wc
;
6052 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6054 path
= btrfs_alloc_path();
6057 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6060 btrfs_assert_tree_locked(parent
);
6061 parent_level
= btrfs_header_level(parent
);
6062 extent_buffer_get(parent
);
6063 path
->nodes
[parent_level
] = parent
;
6064 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6066 btrfs_assert_tree_locked(node
);
6067 level
= btrfs_header_level(node
);
6068 path
->nodes
[level
] = node
;
6069 path
->slots
[level
] = 0;
6070 path
->locks
[level
] = 1;
6072 wc
->refs
[parent_level
] = 1;
6073 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6075 wc
->shared_level
= -1;
6076 wc
->stage
= DROP_REFERENCE
;
6079 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6082 wret
= walk_down_tree(trans
, root
, path
, wc
);
6088 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6096 btrfs_free_path(path
);
6101 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
6104 return min(last
, start
+ nr
- 1);
6107 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
6112 unsigned long first_index
;
6113 unsigned long last_index
;
6116 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6117 struct file_ra_state
*ra
;
6118 struct btrfs_ordered_extent
*ordered
;
6119 unsigned int total_read
= 0;
6120 unsigned int total_dirty
= 0;
6123 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
6125 mutex_lock(&inode
->i_mutex
);
6126 first_index
= start
>> PAGE_CACHE_SHIFT
;
6127 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
6129 /* make sure the dirty trick played by the caller work */
6130 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6131 first_index
, last_index
);
6135 file_ra_state_init(ra
, inode
->i_mapping
);
6137 for (i
= first_index
; i
<= last_index
; i
++) {
6138 if (total_read
% ra
->ra_pages
== 0) {
6139 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
6140 calc_ra(i
, last_index
, ra
->ra_pages
));
6144 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
6146 page
= grab_cache_page(inode
->i_mapping
, i
);
6151 if (!PageUptodate(page
)) {
6152 btrfs_readpage(NULL
, page
);
6154 if (!PageUptodate(page
)) {
6156 page_cache_release(page
);
6161 wait_on_page_writeback(page
);
6163 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
6164 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6165 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6167 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6169 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6171 page_cache_release(page
);
6172 btrfs_start_ordered_extent(inode
, ordered
, 1);
6173 btrfs_put_ordered_extent(ordered
);
6176 set_page_extent_mapped(page
);
6178 if (i
== first_index
)
6179 set_extent_bits(io_tree
, page_start
, page_end
,
6180 EXTENT_BOUNDARY
, GFP_NOFS
);
6181 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
6183 set_page_dirty(page
);
6186 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6188 page_cache_release(page
);
6193 mutex_unlock(&inode
->i_mutex
);
6194 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
6198 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
6199 struct btrfs_key
*extent_key
,
6202 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6203 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
6204 struct extent_map
*em
;
6205 u64 start
= extent_key
->objectid
- offset
;
6206 u64 end
= start
+ extent_key
->offset
- 1;
6208 em
= alloc_extent_map(GFP_NOFS
);
6209 BUG_ON(!em
|| IS_ERR(em
));
6212 em
->len
= extent_key
->offset
;
6213 em
->block_len
= extent_key
->offset
;
6214 em
->block_start
= extent_key
->objectid
;
6215 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6216 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6218 /* setup extent map to cheat btrfs_readpage */
6219 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6222 write_lock(&em_tree
->lock
);
6223 ret
= add_extent_mapping(em_tree
, em
);
6224 write_unlock(&em_tree
->lock
);
6225 if (ret
!= -EEXIST
) {
6226 free_extent_map(em
);
6229 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
6231 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6233 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
6236 struct btrfs_ref_path
{
6238 u64 nodes
[BTRFS_MAX_LEVEL
];
6240 u64 root_generation
;
6247 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
6248 u64 new_nodes
[BTRFS_MAX_LEVEL
];
6251 struct disk_extent
{
6262 static int is_cowonly_root(u64 root_objectid
)
6264 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
6265 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
6266 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
6267 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
6268 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6269 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
6274 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
6275 struct btrfs_root
*extent_root
,
6276 struct btrfs_ref_path
*ref_path
,
6279 struct extent_buffer
*leaf
;
6280 struct btrfs_path
*path
;
6281 struct btrfs_extent_ref
*ref
;
6282 struct btrfs_key key
;
6283 struct btrfs_key found_key
;
6289 path
= btrfs_alloc_path();
6294 ref_path
->lowest_level
= -1;
6295 ref_path
->current_level
= -1;
6296 ref_path
->shared_level
= -1;
6300 level
= ref_path
->current_level
- 1;
6301 while (level
>= -1) {
6303 if (level
< ref_path
->lowest_level
)
6307 bytenr
= ref_path
->nodes
[level
];
6309 bytenr
= ref_path
->extent_start
;
6310 BUG_ON(bytenr
== 0);
6312 parent
= ref_path
->nodes
[level
+ 1];
6313 ref_path
->nodes
[level
+ 1] = 0;
6314 ref_path
->current_level
= level
;
6315 BUG_ON(parent
== 0);
6317 key
.objectid
= bytenr
;
6318 key
.offset
= parent
+ 1;
6319 key
.type
= BTRFS_EXTENT_REF_KEY
;
6321 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6326 leaf
= path
->nodes
[0];
6327 nritems
= btrfs_header_nritems(leaf
);
6328 if (path
->slots
[0] >= nritems
) {
6329 ret
= btrfs_next_leaf(extent_root
, path
);
6334 leaf
= path
->nodes
[0];
6337 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6338 if (found_key
.objectid
== bytenr
&&
6339 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
6340 if (level
< ref_path
->shared_level
)
6341 ref_path
->shared_level
= level
;
6346 btrfs_release_path(extent_root
, path
);
6349 /* reached lowest level */
6353 level
= ref_path
->current_level
;
6354 while (level
< BTRFS_MAX_LEVEL
- 1) {
6358 bytenr
= ref_path
->nodes
[level
];
6360 bytenr
= ref_path
->extent_start
;
6362 BUG_ON(bytenr
== 0);
6364 key
.objectid
= bytenr
;
6366 key
.type
= BTRFS_EXTENT_REF_KEY
;
6368 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6372 leaf
= path
->nodes
[0];
6373 nritems
= btrfs_header_nritems(leaf
);
6374 if (path
->slots
[0] >= nritems
) {
6375 ret
= btrfs_next_leaf(extent_root
, path
);
6379 /* the extent was freed by someone */
6380 if (ref_path
->lowest_level
== level
)
6382 btrfs_release_path(extent_root
, path
);
6385 leaf
= path
->nodes
[0];
6388 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6389 if (found_key
.objectid
!= bytenr
||
6390 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
6391 /* the extent was freed by someone */
6392 if (ref_path
->lowest_level
== level
) {
6396 btrfs_release_path(extent_root
, path
);
6400 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
6401 struct btrfs_extent_ref
);
6402 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
6403 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
6405 level
= (int)ref_objectid
;
6406 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
6407 ref_path
->lowest_level
= level
;
6408 ref_path
->current_level
= level
;
6409 ref_path
->nodes
[level
] = bytenr
;
6411 WARN_ON(ref_objectid
!= level
);
6414 WARN_ON(level
!= -1);
6418 if (ref_path
->lowest_level
== level
) {
6419 ref_path
->owner_objectid
= ref_objectid
;
6420 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
6424 * the block is tree root or the block isn't in reference
6427 if (found_key
.objectid
== found_key
.offset
||
6428 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
6429 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6430 ref_path
->root_generation
=
6431 btrfs_ref_generation(leaf
, ref
);
6433 /* special reference from the tree log */
6434 ref_path
->nodes
[0] = found_key
.offset
;
6435 ref_path
->current_level
= 0;
6442 BUG_ON(ref_path
->nodes
[level
] != 0);
6443 ref_path
->nodes
[level
] = found_key
.offset
;
6444 ref_path
->current_level
= level
;
6447 * the reference was created in the running transaction,
6448 * no need to continue walking up.
6450 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
6451 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6452 ref_path
->root_generation
=
6453 btrfs_ref_generation(leaf
, ref
);
6458 btrfs_release_path(extent_root
, path
);
6461 /* reached max tree level, but no tree root found. */
6464 btrfs_free_path(path
);
6468 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
6469 struct btrfs_root
*extent_root
,
6470 struct btrfs_ref_path
*ref_path
,
6473 memset(ref_path
, 0, sizeof(*ref_path
));
6474 ref_path
->extent_start
= extent_start
;
6476 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
6479 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
6480 struct btrfs_root
*extent_root
,
6481 struct btrfs_ref_path
*ref_path
)
6483 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
6486 static noinline
int get_new_locations(struct inode
*reloc_inode
,
6487 struct btrfs_key
*extent_key
,
6488 u64 offset
, int no_fragment
,
6489 struct disk_extent
**extents
,
6492 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6493 struct btrfs_path
*path
;
6494 struct btrfs_file_extent_item
*fi
;
6495 struct extent_buffer
*leaf
;
6496 struct disk_extent
*exts
= *extents
;
6497 struct btrfs_key found_key
;
6502 int max
= *nr_extents
;
6505 WARN_ON(!no_fragment
&& *extents
);
6508 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6513 path
= btrfs_alloc_path();
6516 cur_pos
= extent_key
->objectid
- offset
;
6517 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
6518 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
6528 leaf
= path
->nodes
[0];
6529 nritems
= btrfs_header_nritems(leaf
);
6530 if (path
->slots
[0] >= nritems
) {
6531 ret
= btrfs_next_leaf(root
, path
);
6536 leaf
= path
->nodes
[0];
6539 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6540 if (found_key
.offset
!= cur_pos
||
6541 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
6542 found_key
.objectid
!= reloc_inode
->i_ino
)
6545 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6546 struct btrfs_file_extent_item
);
6547 if (btrfs_file_extent_type(leaf
, fi
) !=
6548 BTRFS_FILE_EXTENT_REG
||
6549 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6553 struct disk_extent
*old
= exts
;
6555 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6556 memcpy(exts
, old
, sizeof(*exts
) * nr
);
6557 if (old
!= *extents
)
6561 exts
[nr
].disk_bytenr
=
6562 btrfs_file_extent_disk_bytenr(leaf
, fi
);
6563 exts
[nr
].disk_num_bytes
=
6564 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6565 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
6566 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6567 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6568 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
6569 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
6570 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
6572 BUG_ON(exts
[nr
].offset
> 0);
6573 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
6574 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
6576 cur_pos
+= exts
[nr
].num_bytes
;
6579 if (cur_pos
+ offset
>= last_byte
)
6589 BUG_ON(cur_pos
+ offset
> last_byte
);
6590 if (cur_pos
+ offset
< last_byte
) {
6596 btrfs_free_path(path
);
6598 if (exts
!= *extents
)
6607 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
6608 struct btrfs_root
*root
,
6609 struct btrfs_path
*path
,
6610 struct btrfs_key
*extent_key
,
6611 struct btrfs_key
*leaf_key
,
6612 struct btrfs_ref_path
*ref_path
,
6613 struct disk_extent
*new_extents
,
6616 struct extent_buffer
*leaf
;
6617 struct btrfs_file_extent_item
*fi
;
6618 struct inode
*inode
= NULL
;
6619 struct btrfs_key key
;
6624 u64 search_end
= (u64
)-1;
6627 int extent_locked
= 0;
6631 memcpy(&key
, leaf_key
, sizeof(key
));
6632 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
6633 if (key
.objectid
< ref_path
->owner_objectid
||
6634 (key
.objectid
== ref_path
->owner_objectid
&&
6635 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
6636 key
.objectid
= ref_path
->owner_objectid
;
6637 key
.type
= BTRFS_EXTENT_DATA_KEY
;
6643 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
6647 leaf
= path
->nodes
[0];
6648 nritems
= btrfs_header_nritems(leaf
);
6650 if (extent_locked
&& ret
> 0) {
6652 * the file extent item was modified by someone
6653 * before the extent got locked.
6655 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6656 lock_end
, GFP_NOFS
);
6660 if (path
->slots
[0] >= nritems
) {
6661 if (++nr_scaned
> 2)
6664 BUG_ON(extent_locked
);
6665 ret
= btrfs_next_leaf(root
, path
);
6670 leaf
= path
->nodes
[0];
6671 nritems
= btrfs_header_nritems(leaf
);
6674 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
6676 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
6677 if ((key
.objectid
> ref_path
->owner_objectid
) ||
6678 (key
.objectid
== ref_path
->owner_objectid
&&
6679 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
6680 key
.offset
>= search_end
)
6684 if (inode
&& key
.objectid
!= inode
->i_ino
) {
6685 BUG_ON(extent_locked
);
6686 btrfs_release_path(root
, path
);
6687 mutex_unlock(&inode
->i_mutex
);
6693 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6698 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6699 struct btrfs_file_extent_item
);
6700 extent_type
= btrfs_file_extent_type(leaf
, fi
);
6701 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
6702 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
6703 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
6704 extent_key
->objectid
)) {
6710 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6711 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
6713 if (search_end
== (u64
)-1) {
6714 search_end
= key
.offset
- ext_offset
+
6715 btrfs_file_extent_ram_bytes(leaf
, fi
);
6718 if (!extent_locked
) {
6719 lock_start
= key
.offset
;
6720 lock_end
= lock_start
+ num_bytes
- 1;
6722 if (lock_start
> key
.offset
||
6723 lock_end
+ 1 < key
.offset
+ num_bytes
) {
6724 unlock_extent(&BTRFS_I(inode
)->io_tree
,
6725 lock_start
, lock_end
, GFP_NOFS
);
6731 btrfs_release_path(root
, path
);
6733 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
6734 key
.objectid
, root
);
6735 if (inode
->i_state
& I_NEW
) {
6736 BTRFS_I(inode
)->root
= root
;
6737 BTRFS_I(inode
)->location
.objectid
=
6739 BTRFS_I(inode
)->location
.type
=
6740 BTRFS_INODE_ITEM_KEY
;
6741 BTRFS_I(inode
)->location
.offset
= 0;
6742 btrfs_read_locked_inode(inode
);
6743 unlock_new_inode(inode
);
6746 * some code call btrfs_commit_transaction while
6747 * holding the i_mutex, so we can't use mutex_lock
6750 if (is_bad_inode(inode
) ||
6751 !mutex_trylock(&inode
->i_mutex
)) {
6754 key
.offset
= (u64
)-1;
6759 if (!extent_locked
) {
6760 struct btrfs_ordered_extent
*ordered
;
6762 btrfs_release_path(root
, path
);
6764 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6765 lock_end
, GFP_NOFS
);
6766 ordered
= btrfs_lookup_first_ordered_extent(inode
,
6769 ordered
->file_offset
<= lock_end
&&
6770 ordered
->file_offset
+ ordered
->len
> lock_start
) {
6771 unlock_extent(&BTRFS_I(inode
)->io_tree
,
6772 lock_start
, lock_end
, GFP_NOFS
);
6773 btrfs_start_ordered_extent(inode
, ordered
, 1);
6774 btrfs_put_ordered_extent(ordered
);
6775 key
.offset
+= num_bytes
;
6779 btrfs_put_ordered_extent(ordered
);
6785 if (nr_extents
== 1) {
6786 /* update extent pointer in place */
6787 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6788 new_extents
[0].disk_bytenr
);
6789 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6790 new_extents
[0].disk_num_bytes
);
6791 btrfs_mark_buffer_dirty(leaf
);
6793 btrfs_drop_extent_cache(inode
, key
.offset
,
6794 key
.offset
+ num_bytes
- 1, 0);
6796 ret
= btrfs_inc_extent_ref(trans
, root
,
6797 new_extents
[0].disk_bytenr
,
6798 new_extents
[0].disk_num_bytes
,
6800 root
->root_key
.objectid
,
6805 ret
= btrfs_free_extent(trans
, root
,
6806 extent_key
->objectid
,
6809 btrfs_header_owner(leaf
),
6810 btrfs_header_generation(leaf
),
6814 btrfs_release_path(root
, path
);
6815 key
.offset
+= num_bytes
;
6823 * drop old extent pointer at first, then insert the
6824 * new pointers one bye one
6826 btrfs_release_path(root
, path
);
6827 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
6828 key
.offset
+ num_bytes
,
6829 key
.offset
, &alloc_hint
);
6832 for (i
= 0; i
< nr_extents
; i
++) {
6833 if (ext_offset
>= new_extents
[i
].num_bytes
) {
6834 ext_offset
-= new_extents
[i
].num_bytes
;
6837 extent_len
= min(new_extents
[i
].num_bytes
-
6838 ext_offset
, num_bytes
);
6840 ret
= btrfs_insert_empty_item(trans
, root
,
6845 leaf
= path
->nodes
[0];
6846 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6847 struct btrfs_file_extent_item
);
6848 btrfs_set_file_extent_generation(leaf
, fi
,
6850 btrfs_set_file_extent_type(leaf
, fi
,
6851 BTRFS_FILE_EXTENT_REG
);
6852 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6853 new_extents
[i
].disk_bytenr
);
6854 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6855 new_extents
[i
].disk_num_bytes
);
6856 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
6857 new_extents
[i
].ram_bytes
);
6859 btrfs_set_file_extent_compression(leaf
, fi
,
6860 new_extents
[i
].compression
);
6861 btrfs_set_file_extent_encryption(leaf
, fi
,
6862 new_extents
[i
].encryption
);
6863 btrfs_set_file_extent_other_encoding(leaf
, fi
,
6864 new_extents
[i
].other_encoding
);
6866 btrfs_set_file_extent_num_bytes(leaf
, fi
,
6868 ext_offset
+= new_extents
[i
].offset
;
6869 btrfs_set_file_extent_offset(leaf
, fi
,
6871 btrfs_mark_buffer_dirty(leaf
);
6873 btrfs_drop_extent_cache(inode
, key
.offset
,
6874 key
.offset
+ extent_len
- 1, 0);
6876 ret
= btrfs_inc_extent_ref(trans
, root
,
6877 new_extents
[i
].disk_bytenr
,
6878 new_extents
[i
].disk_num_bytes
,
6880 root
->root_key
.objectid
,
6881 trans
->transid
, key
.objectid
);
6883 btrfs_release_path(root
, path
);
6885 inode_add_bytes(inode
, extent_len
);
6888 num_bytes
-= extent_len
;
6889 key
.offset
+= extent_len
;
6894 BUG_ON(i
>= nr_extents
);
6898 if (extent_locked
) {
6899 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6900 lock_end
, GFP_NOFS
);
6904 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
6905 key
.offset
>= search_end
)
6912 btrfs_release_path(root
, path
);
6914 mutex_unlock(&inode
->i_mutex
);
6915 if (extent_locked
) {
6916 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6917 lock_end
, GFP_NOFS
);
6924 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
6925 struct btrfs_root
*root
,
6926 struct extent_buffer
*buf
, u64 orig_start
)
6931 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
6932 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6934 level
= btrfs_header_level(buf
);
6936 struct btrfs_leaf_ref
*ref
;
6937 struct btrfs_leaf_ref
*orig_ref
;
6939 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
6943 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
6945 btrfs_free_leaf_ref(root
, orig_ref
);
6949 ref
->nritems
= orig_ref
->nritems
;
6950 memcpy(ref
->extents
, orig_ref
->extents
,
6951 sizeof(ref
->extents
[0]) * ref
->nritems
);
6953 btrfs_free_leaf_ref(root
, orig_ref
);
6955 ref
->root_gen
= trans
->transid
;
6956 ref
->bytenr
= buf
->start
;
6957 ref
->owner
= btrfs_header_owner(buf
);
6958 ref
->generation
= btrfs_header_generation(buf
);
6960 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
6962 btrfs_free_leaf_ref(root
, ref
);
6967 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
6968 struct extent_buffer
*leaf
,
6969 struct btrfs_block_group_cache
*group
,
6970 struct btrfs_root
*target_root
)
6972 struct btrfs_key key
;
6973 struct inode
*inode
= NULL
;
6974 struct btrfs_file_extent_item
*fi
;
6975 struct extent_state
*cached_state
= NULL
;
6977 u64 skip_objectid
= 0;
6981 nritems
= btrfs_header_nritems(leaf
);
6982 for (i
= 0; i
< nritems
; i
++) {
6983 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6984 if (key
.objectid
== skip_objectid
||
6985 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6987 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6988 if (btrfs_file_extent_type(leaf
, fi
) ==
6989 BTRFS_FILE_EXTENT_INLINE
)
6991 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6993 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
6995 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
6996 key
.objectid
, target_root
, 1);
6999 skip_objectid
= key
.objectid
;
7002 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7004 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7005 key
.offset
+ num_bytes
- 1, 0, &cached_state
,
7007 btrfs_drop_extent_cache(inode
, key
.offset
,
7008 key
.offset
+ num_bytes
- 1, 1);
7009 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7010 key
.offset
+ num_bytes
- 1, &cached_state
,
7018 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
7019 struct btrfs_root
*root
,
7020 struct extent_buffer
*leaf
,
7021 struct btrfs_block_group_cache
*group
,
7022 struct inode
*reloc_inode
)
7024 struct btrfs_key key
;
7025 struct btrfs_key extent_key
;
7026 struct btrfs_file_extent_item
*fi
;
7027 struct btrfs_leaf_ref
*ref
;
7028 struct disk_extent
*new_extent
;
7037 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
7038 BUG_ON(!new_extent
);
7040 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
7044 nritems
= btrfs_header_nritems(leaf
);
7045 for (i
= 0; i
< nritems
; i
++) {
7046 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7047 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
7049 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7050 if (btrfs_file_extent_type(leaf
, fi
) ==
7051 BTRFS_FILE_EXTENT_INLINE
)
7053 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7054 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7059 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
7060 bytenr
+ num_bytes
<= group
->key
.objectid
)
7063 extent_key
.objectid
= bytenr
;
7064 extent_key
.offset
= num_bytes
;
7065 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7067 ret
= get_new_locations(reloc_inode
, &extent_key
,
7068 group
->key
.objectid
, 1,
7069 &new_extent
, &nr_extent
);
7074 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
7075 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
7076 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
7077 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
7079 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7080 new_extent
->disk_bytenr
);
7081 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7082 new_extent
->disk_num_bytes
);
7083 btrfs_mark_buffer_dirty(leaf
);
7085 ret
= btrfs_inc_extent_ref(trans
, root
,
7086 new_extent
->disk_bytenr
,
7087 new_extent
->disk_num_bytes
,
7089 root
->root_key
.objectid
,
7090 trans
->transid
, key
.objectid
);
7093 ret
= btrfs_free_extent(trans
, root
,
7094 bytenr
, num_bytes
, leaf
->start
,
7095 btrfs_header_owner(leaf
),
7096 btrfs_header_generation(leaf
),
7102 BUG_ON(ext_index
+ 1 != ref
->nritems
);
7103 btrfs_free_leaf_ref(root
, ref
);
7107 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
7108 struct btrfs_root
*root
)
7110 struct btrfs_root
*reloc_root
;
7113 if (root
->reloc_root
) {
7114 reloc_root
= root
->reloc_root
;
7115 root
->reloc_root
= NULL
;
7116 list_add(&reloc_root
->dead_list
,
7117 &root
->fs_info
->dead_reloc_roots
);
7119 btrfs_set_root_bytenr(&reloc_root
->root_item
,
7120 reloc_root
->node
->start
);
7121 btrfs_set_root_level(&root
->root_item
,
7122 btrfs_header_level(reloc_root
->node
));
7123 memset(&reloc_root
->root_item
.drop_progress
, 0,
7124 sizeof(struct btrfs_disk_key
));
7125 reloc_root
->root_item
.drop_level
= 0;
7127 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
7128 &reloc_root
->root_key
,
7129 &reloc_root
->root_item
);
7135 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
7137 struct btrfs_trans_handle
*trans
;
7138 struct btrfs_root
*reloc_root
;
7139 struct btrfs_root
*prev_root
= NULL
;
7140 struct list_head dead_roots
;
7144 INIT_LIST_HEAD(&dead_roots
);
7145 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
7147 while (!list_empty(&dead_roots
)) {
7148 reloc_root
= list_entry(dead_roots
.prev
,
7149 struct btrfs_root
, dead_list
);
7150 list_del_init(&reloc_root
->dead_list
);
7152 BUG_ON(reloc_root
->commit_root
!= NULL
);
7154 trans
= btrfs_join_transaction(root
, 1);
7157 mutex_lock(&root
->fs_info
->drop_mutex
);
7158 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
7161 mutex_unlock(&root
->fs_info
->drop_mutex
);
7163 nr
= trans
->blocks_used
;
7164 ret
= btrfs_end_transaction(trans
, root
);
7166 btrfs_btree_balance_dirty(root
, nr
);
7169 free_extent_buffer(reloc_root
->node
);
7171 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
7172 &reloc_root
->root_key
);
7174 mutex_unlock(&root
->fs_info
->drop_mutex
);
7176 nr
= trans
->blocks_used
;
7177 ret
= btrfs_end_transaction(trans
, root
);
7179 btrfs_btree_balance_dirty(root
, nr
);
7182 prev_root
= reloc_root
;
7185 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
7191 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
7193 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
7197 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
7199 struct btrfs_root
*reloc_root
;
7200 struct btrfs_trans_handle
*trans
;
7201 struct btrfs_key location
;
7205 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7206 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
7208 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
7209 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7212 trans
= btrfs_start_transaction(root
, 1);
7214 ret
= btrfs_commit_transaction(trans
, root
);
7218 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
7219 location
.offset
= (u64
)-1;
7220 location
.type
= BTRFS_ROOT_ITEM_KEY
;
7222 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
7223 BUG_ON(!reloc_root
);
7224 btrfs_orphan_cleanup(reloc_root
);
7228 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
7229 struct btrfs_root
*root
)
7231 struct btrfs_root
*reloc_root
;
7232 struct extent_buffer
*eb
;
7233 struct btrfs_root_item
*root_item
;
7234 struct btrfs_key root_key
;
7237 BUG_ON(!root
->ref_cows
);
7238 if (root
->reloc_root
)
7241 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
7244 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
7245 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
7248 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
7249 root_key
.offset
= root
->root_key
.objectid
;
7250 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7252 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
7253 btrfs_set_root_refs(root_item
, 0);
7254 btrfs_set_root_bytenr(root_item
, eb
->start
);
7255 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
7256 btrfs_set_root_generation(root_item
, trans
->transid
);
7258 btrfs_tree_unlock(eb
);
7259 free_extent_buffer(eb
);
7261 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
7262 &root_key
, root_item
);
7266 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
7268 BUG_ON(!reloc_root
);
7269 reloc_root
->last_trans
= trans
->transid
;
7270 reloc_root
->commit_root
= NULL
;
7271 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
7273 root
->reloc_root
= reloc_root
;
7278 * Core function of space balance.
7280 * The idea is using reloc trees to relocate tree blocks in reference
7281 * counted roots. There is one reloc tree for each subvol, and all
7282 * reloc trees share same root key objectid. Reloc trees are snapshots
7283 * of the latest committed roots of subvols (root->commit_root).
7285 * To relocate a tree block referenced by a subvol, there are two steps.
7286 * COW the block through subvol's reloc tree, then update block pointer
7287 * in the subvol to point to the new block. Since all reloc trees share
7288 * same root key objectid, doing special handing for tree blocks owned
7289 * by them is easy. Once a tree block has been COWed in one reloc tree,
7290 * we can use the resulting new block directly when the same block is
7291 * required to COW again through other reloc trees. By this way, relocated
7292 * tree blocks are shared between reloc trees, so they are also shared
7295 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
7296 struct btrfs_root
*root
,
7297 struct btrfs_path
*path
,
7298 struct btrfs_key
*first_key
,
7299 struct btrfs_ref_path
*ref_path
,
7300 struct btrfs_block_group_cache
*group
,
7301 struct inode
*reloc_inode
)
7303 struct btrfs_root
*reloc_root
;
7304 struct extent_buffer
*eb
= NULL
;
7305 struct btrfs_key
*keys
;
7309 int lowest_level
= 0;
7312 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
7313 lowest_level
= ref_path
->owner_objectid
;
7315 if (!root
->ref_cows
) {
7316 path
->lowest_level
= lowest_level
;
7317 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
7319 path
->lowest_level
= 0;
7320 btrfs_release_path(root
, path
);
7324 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7325 ret
= init_reloc_tree(trans
, root
);
7327 reloc_root
= root
->reloc_root
;
7329 shared_level
= ref_path
->shared_level
;
7330 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
7332 keys
= ref_path
->node_keys
;
7333 nodes
= ref_path
->new_nodes
;
7334 memset(&keys
[shared_level
+ 1], 0,
7335 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7336 memset(&nodes
[shared_level
+ 1], 0,
7337 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7339 if (nodes
[lowest_level
] == 0) {
7340 path
->lowest_level
= lowest_level
;
7341 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7344 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
7345 eb
= path
->nodes
[level
];
7346 if (!eb
|| eb
== reloc_root
->node
)
7348 nodes
[level
] = eb
->start
;
7350 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
7352 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
7355 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7356 eb
= path
->nodes
[0];
7357 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
7358 group
, reloc_inode
);
7361 btrfs_release_path(reloc_root
, path
);
7363 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
7369 * replace tree blocks in the fs tree with tree blocks in
7372 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
7375 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7376 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7379 extent_buffer_get(path
->nodes
[0]);
7380 eb
= path
->nodes
[0];
7381 btrfs_release_path(reloc_root
, path
);
7382 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
7384 free_extent_buffer(eb
);
7387 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7388 path
->lowest_level
= 0;
7392 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
7393 struct btrfs_root
*root
,
7394 struct btrfs_path
*path
,
7395 struct btrfs_key
*first_key
,
7396 struct btrfs_ref_path
*ref_path
)
7400 ret
= relocate_one_path(trans
, root
, path
, first_key
,
7401 ref_path
, NULL
, NULL
);
7407 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
7408 struct btrfs_root
*extent_root
,
7409 struct btrfs_path
*path
,
7410 struct btrfs_key
*extent_key
)
7414 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
7417 ret
= btrfs_del_item(trans
, extent_root
, path
);
7419 btrfs_release_path(extent_root
, path
);
7423 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
7424 struct btrfs_ref_path
*ref_path
)
7426 struct btrfs_key root_key
;
7428 root_key
.objectid
= ref_path
->root_objectid
;
7429 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7430 if (is_cowonly_root(ref_path
->root_objectid
))
7431 root_key
.offset
= 0;
7433 root_key
.offset
= (u64
)-1;
7435 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
7438 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
7439 struct btrfs_path
*path
,
7440 struct btrfs_key
*extent_key
,
7441 struct btrfs_block_group_cache
*group
,
7442 struct inode
*reloc_inode
, int pass
)
7444 struct btrfs_trans_handle
*trans
;
7445 struct btrfs_root
*found_root
;
7446 struct btrfs_ref_path
*ref_path
= NULL
;
7447 struct disk_extent
*new_extents
= NULL
;
7452 struct btrfs_key first_key
;
7456 trans
= btrfs_start_transaction(extent_root
, 1);
7459 if (extent_key
->objectid
== 0) {
7460 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
7464 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
7470 for (loops
= 0; ; loops
++) {
7472 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
7473 extent_key
->objectid
);
7475 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
7482 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
7483 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
7486 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
7487 BUG_ON(!found_root
);
7489 * for reference counted tree, only process reference paths
7490 * rooted at the latest committed root.
7492 if (found_root
->ref_cows
&&
7493 ref_path
->root_generation
!= found_root
->root_key
.offset
)
7496 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7499 * copy data extents to new locations
7501 u64 group_start
= group
->key
.objectid
;
7502 ret
= relocate_data_extent(reloc_inode
,
7511 level
= ref_path
->owner_objectid
;
7514 if (prev_block
!= ref_path
->nodes
[level
]) {
7515 struct extent_buffer
*eb
;
7516 u64 block_start
= ref_path
->nodes
[level
];
7517 u64 block_size
= btrfs_level_size(found_root
, level
);
7519 eb
= read_tree_block(found_root
, block_start
,
7521 btrfs_tree_lock(eb
);
7522 BUG_ON(level
!= btrfs_header_level(eb
));
7525 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
7527 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
7529 btrfs_tree_unlock(eb
);
7530 free_extent_buffer(eb
);
7531 prev_block
= block_start
;
7534 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
7535 btrfs_record_root_in_trans(found_root
);
7536 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
7537 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7539 * try to update data extent references while
7540 * keeping metadata shared between snapshots.
7543 ret
= relocate_one_path(trans
, found_root
,
7544 path
, &first_key
, ref_path
,
7545 group
, reloc_inode
);
7551 * use fallback method to process the remaining
7555 u64 group_start
= group
->key
.objectid
;
7556 new_extents
= kmalloc(sizeof(*new_extents
),
7559 ret
= get_new_locations(reloc_inode
,
7567 ret
= replace_one_extent(trans
, found_root
,
7569 &first_key
, ref_path
,
7570 new_extents
, nr_extents
);
7572 ret
= relocate_tree_block(trans
, found_root
, path
,
7573 &first_key
, ref_path
);
7580 btrfs_end_transaction(trans
, extent_root
);
7587 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
7590 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
7591 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
7593 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
7594 if (num_devices
== 1) {
7595 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7596 stripped
= flags
& ~stripped
;
7598 /* turn raid0 into single device chunks */
7599 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7602 /* turn mirroring into duplication */
7603 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7604 BTRFS_BLOCK_GROUP_RAID10
))
7605 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7608 /* they already had raid on here, just return */
7609 if (flags
& stripped
)
7612 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7613 stripped
= flags
& ~stripped
;
7615 /* switch duplicated blocks with raid1 */
7616 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7617 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7619 /* turn single device chunks into raid0 */
7620 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
7625 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
)
7627 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7634 spin_lock(&sinfo
->lock
);
7635 spin_lock(&cache
->lock
);
7636 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7637 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7639 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
7640 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
7641 cache
->reserved_pinned
+ num_bytes
< sinfo
->total_bytes
) {
7642 sinfo
->bytes_readonly
+= num_bytes
;
7643 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
7644 cache
->reserved_pinned
= 0;
7648 spin_unlock(&cache
->lock
);
7649 spin_unlock(&sinfo
->lock
);
7653 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
7654 struct btrfs_block_group_cache
*cache
)
7657 struct btrfs_trans_handle
*trans
;
7663 trans
= btrfs_join_transaction(root
, 1);
7664 BUG_ON(IS_ERR(trans
));
7666 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
7667 if (alloc_flags
!= cache
->flags
)
7668 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
7670 ret
= set_block_group_ro(cache
);
7673 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
7674 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
7677 ret
= set_block_group_ro(cache
);
7679 btrfs_end_transaction(trans
, root
);
7683 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
7684 struct btrfs_block_group_cache
*cache
)
7686 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7691 spin_lock(&sinfo
->lock
);
7692 spin_lock(&cache
->lock
);
7693 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7694 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7695 sinfo
->bytes_readonly
-= num_bytes
;
7697 spin_unlock(&cache
->lock
);
7698 spin_unlock(&sinfo
->lock
);
7703 * checks to see if its even possible to relocate this block group.
7705 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7706 * ok to go ahead and try.
7708 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
7710 struct btrfs_block_group_cache
*block_group
;
7711 struct btrfs_space_info
*space_info
;
7712 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7713 struct btrfs_device
*device
;
7717 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
7719 /* odd, couldn't find the block group, leave it alone */
7723 /* no bytes used, we're good */
7724 if (!btrfs_block_group_used(&block_group
->item
))
7727 space_info
= block_group
->space_info
;
7728 spin_lock(&space_info
->lock
);
7730 full
= space_info
->full
;
7733 * if this is the last block group we have in this space, we can't
7734 * relocate it unless we're able to allocate a new chunk below.
7736 * Otherwise, we need to make sure we have room in the space to handle
7737 * all of the extents from this block group. If we can, we're good
7739 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
7740 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
7741 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
7742 btrfs_block_group_used(&block_group
->item
) <
7743 space_info
->total_bytes
)) {
7744 spin_unlock(&space_info
->lock
);
7747 spin_unlock(&space_info
->lock
);
7750 * ok we don't have enough space, but maybe we have free space on our
7751 * devices to allocate new chunks for relocation, so loop through our
7752 * alloc devices and guess if we have enough space. However, if we
7753 * were marked as full, then we know there aren't enough chunks, and we
7760 mutex_lock(&root
->fs_info
->chunk_mutex
);
7761 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
7762 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
7763 u64 dev_offset
, max_avail
;
7766 * check to make sure we can actually find a chunk with enough
7767 * space to fit our block group in.
7769 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
7770 ret
= find_free_dev_extent(NULL
, device
, min_free
,
7771 &dev_offset
, &max_avail
);
7777 mutex_unlock(&root
->fs_info
->chunk_mutex
);
7779 btrfs_put_block_group(block_group
);
7783 static int find_first_block_group(struct btrfs_root
*root
,
7784 struct btrfs_path
*path
, struct btrfs_key
*key
)
7787 struct btrfs_key found_key
;
7788 struct extent_buffer
*leaf
;
7791 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7796 slot
= path
->slots
[0];
7797 leaf
= path
->nodes
[0];
7798 if (slot
>= btrfs_header_nritems(leaf
)) {
7799 ret
= btrfs_next_leaf(root
, path
);
7806 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7808 if (found_key
.objectid
>= key
->objectid
&&
7809 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7819 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7821 struct btrfs_block_group_cache
*block_group
;
7822 struct btrfs_space_info
*space_info
;
7823 struct btrfs_caching_control
*caching_ctl
;
7826 down_write(&info
->extent_commit_sem
);
7827 while (!list_empty(&info
->caching_block_groups
)) {
7828 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
7829 struct btrfs_caching_control
, list
);
7830 list_del(&caching_ctl
->list
);
7831 put_caching_control(caching_ctl
);
7833 up_write(&info
->extent_commit_sem
);
7835 spin_lock(&info
->block_group_cache_lock
);
7836 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7837 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7839 rb_erase(&block_group
->cache_node
,
7840 &info
->block_group_cache_tree
);
7841 spin_unlock(&info
->block_group_cache_lock
);
7843 down_write(&block_group
->space_info
->groups_sem
);
7844 list_del(&block_group
->list
);
7845 up_write(&block_group
->space_info
->groups_sem
);
7847 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7848 wait_block_group_cache_done(block_group
);
7850 btrfs_remove_free_space_cache(block_group
);
7851 btrfs_put_block_group(block_group
);
7853 spin_lock(&info
->block_group_cache_lock
);
7855 spin_unlock(&info
->block_group_cache_lock
);
7857 /* now that all the block groups are freed, go through and
7858 * free all the space_info structs. This is only called during
7859 * the final stages of unmount, and so we know nobody is
7860 * using them. We call synchronize_rcu() once before we start,
7861 * just to be on the safe side.
7865 release_global_block_rsv(info
);
7867 while(!list_empty(&info
->space_info
)) {
7868 space_info
= list_entry(info
->space_info
.next
,
7869 struct btrfs_space_info
,
7871 if (space_info
->bytes_pinned
> 0 ||
7872 space_info
->bytes_reserved
> 0) {
7874 dump_space_info(space_info
, 0, 0);
7876 list_del(&space_info
->list
);
7882 static void __link_block_group(struct btrfs_space_info
*space_info
,
7883 struct btrfs_block_group_cache
*cache
)
7885 int index
= get_block_group_index(cache
);
7887 down_write(&space_info
->groups_sem
);
7888 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
7889 up_write(&space_info
->groups_sem
);
7892 int btrfs_read_block_groups(struct btrfs_root
*root
)
7894 struct btrfs_path
*path
;
7896 struct btrfs_block_group_cache
*cache
;
7897 struct btrfs_fs_info
*info
= root
->fs_info
;
7898 struct btrfs_space_info
*space_info
;
7899 struct btrfs_key key
;
7900 struct btrfs_key found_key
;
7901 struct extent_buffer
*leaf
;
7903 root
= info
->extent_root
;
7906 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7907 path
= btrfs_alloc_path();
7912 ret
= find_first_block_group(root
, path
, &key
);
7918 leaf
= path
->nodes
[0];
7919 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7920 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7926 atomic_set(&cache
->count
, 1);
7927 spin_lock_init(&cache
->lock
);
7928 spin_lock_init(&cache
->tree_lock
);
7929 cache
->fs_info
= info
;
7930 INIT_LIST_HEAD(&cache
->list
);
7931 INIT_LIST_HEAD(&cache
->cluster_list
);
7934 * we only want to have 32k of ram per block group for keeping
7935 * track of free space, and if we pass 1/2 of that we want to
7936 * start converting things over to using bitmaps
7938 cache
->extents_thresh
= ((1024 * 32) / 2) /
7939 sizeof(struct btrfs_free_space
);
7941 read_extent_buffer(leaf
, &cache
->item
,
7942 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7943 sizeof(cache
->item
));
7944 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7946 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7947 btrfs_release_path(root
, path
);
7948 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7949 cache
->sectorsize
= root
->sectorsize
;
7952 * check for two cases, either we are full, and therefore
7953 * don't need to bother with the caching work since we won't
7954 * find any space, or we are empty, and we can just add all
7955 * the space in and be done with it. This saves us _alot_ of
7956 * time, particularly in the full case.
7958 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7959 exclude_super_stripes(root
, cache
);
7960 cache
->last_byte_to_unpin
= (u64
)-1;
7961 cache
->cached
= BTRFS_CACHE_FINISHED
;
7962 free_excluded_extents(root
, cache
);
7963 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7964 exclude_super_stripes(root
, cache
);
7965 cache
->last_byte_to_unpin
= (u64
)-1;
7966 cache
->cached
= BTRFS_CACHE_FINISHED
;
7967 add_new_free_space(cache
, root
->fs_info
,
7969 found_key
.objectid
+
7971 free_excluded_extents(root
, cache
);
7974 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7975 btrfs_block_group_used(&cache
->item
),
7978 cache
->space_info
= space_info
;
7979 spin_lock(&cache
->space_info
->lock
);
7980 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7981 spin_unlock(&cache
->space_info
->lock
);
7983 __link_block_group(space_info
, cache
);
7985 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7988 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7989 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7990 set_block_group_ro(cache
);
7993 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7994 if (!(get_alloc_profile(root
, space_info
->flags
) &
7995 (BTRFS_BLOCK_GROUP_RAID10
|
7996 BTRFS_BLOCK_GROUP_RAID1
|
7997 BTRFS_BLOCK_GROUP_DUP
)))
8000 * avoid allocating from un-mirrored block group if there are
8001 * mirrored block groups.
8003 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
8004 set_block_group_ro(cache
);
8005 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
8006 set_block_group_ro(cache
);
8009 init_global_block_rsv(info
);
8012 btrfs_free_path(path
);
8016 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8017 struct btrfs_root
*root
, u64 bytes_used
,
8018 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8022 struct btrfs_root
*extent_root
;
8023 struct btrfs_block_group_cache
*cache
;
8025 extent_root
= root
->fs_info
->extent_root
;
8027 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8029 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8033 cache
->key
.objectid
= chunk_offset
;
8034 cache
->key
.offset
= size
;
8035 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8036 cache
->sectorsize
= root
->sectorsize
;
8039 * we only want to have 32k of ram per block group for keeping track
8040 * of free space, and if we pass 1/2 of that we want to start
8041 * converting things over to using bitmaps
8043 cache
->extents_thresh
= ((1024 * 32) / 2) /
8044 sizeof(struct btrfs_free_space
);
8045 atomic_set(&cache
->count
, 1);
8046 spin_lock_init(&cache
->lock
);
8047 spin_lock_init(&cache
->tree_lock
);
8048 INIT_LIST_HEAD(&cache
->list
);
8049 INIT_LIST_HEAD(&cache
->cluster_list
);
8051 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8052 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8053 cache
->flags
= type
;
8054 btrfs_set_block_group_flags(&cache
->item
, type
);
8056 cache
->last_byte_to_unpin
= (u64
)-1;
8057 cache
->cached
= BTRFS_CACHE_FINISHED
;
8058 exclude_super_stripes(root
, cache
);
8060 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8061 chunk_offset
+ size
);
8063 free_excluded_extents(root
, cache
);
8065 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8066 &cache
->space_info
);
8069 spin_lock(&cache
->space_info
->lock
);
8070 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8071 spin_unlock(&cache
->space_info
->lock
);
8073 __link_block_group(cache
->space_info
, cache
);
8075 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8078 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
8079 sizeof(cache
->item
));
8082 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8087 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8088 struct btrfs_root
*root
, u64 group_start
)
8090 struct btrfs_path
*path
;
8091 struct btrfs_block_group_cache
*block_group
;
8092 struct btrfs_free_cluster
*cluster
;
8093 struct btrfs_key key
;
8096 root
= root
->fs_info
->extent_root
;
8098 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8099 BUG_ON(!block_group
);
8100 BUG_ON(!block_group
->ro
);
8102 memcpy(&key
, &block_group
->key
, sizeof(key
));
8104 /* make sure this block group isn't part of an allocation cluster */
8105 cluster
= &root
->fs_info
->data_alloc_cluster
;
8106 spin_lock(&cluster
->refill_lock
);
8107 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8108 spin_unlock(&cluster
->refill_lock
);
8111 * make sure this block group isn't part of a metadata
8112 * allocation cluster
8114 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8115 spin_lock(&cluster
->refill_lock
);
8116 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8117 spin_unlock(&cluster
->refill_lock
);
8119 path
= btrfs_alloc_path();
8122 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8123 rb_erase(&block_group
->cache_node
,
8124 &root
->fs_info
->block_group_cache_tree
);
8125 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8127 down_write(&block_group
->space_info
->groups_sem
);
8129 * we must use list_del_init so people can check to see if they
8130 * are still on the list after taking the semaphore
8132 list_del_init(&block_group
->list
);
8133 up_write(&block_group
->space_info
->groups_sem
);
8135 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8136 wait_block_group_cache_done(block_group
);
8138 btrfs_remove_free_space_cache(block_group
);
8140 spin_lock(&block_group
->space_info
->lock
);
8141 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8142 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8143 spin_unlock(&block_group
->space_info
->lock
);
8145 btrfs_clear_space_info_full(root
->fs_info
);
8147 btrfs_put_block_group(block_group
);
8148 btrfs_put_block_group(block_group
);
8150 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8156 ret
= btrfs_del_item(trans
, root
, path
);
8158 btrfs_free_path(path
);