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>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "free-space-cache.h"
35 static int update_block_group(struct btrfs_trans_handle
*trans
,
36 struct btrfs_root
*root
,
37 u64 bytenr
, u64 num_bytes
, int alloc
,
39 static int update_reserved_extents(struct btrfs_block_group_cache
*cache
,
40 u64 num_bytes
, int reserve
);
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 pin_down_bytes(struct btrfs_trans_handle
*trans
,
64 struct btrfs_root
*root
,
65 struct btrfs_path
*path
,
66 u64 bytenr
, u64 num_bytes
,
67 int is_data
, int reserved
,
68 struct extent_buffer
**must_clean
);
69 static int find_next_key(struct btrfs_path
*path
, int level
,
70 struct btrfs_key
*key
);
71 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
72 int dump_block_groups
);
75 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
78 return cache
->cached
== BTRFS_CACHE_FINISHED
;
81 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
83 return (cache
->flags
& bits
) == bits
;
87 * this adds the block group to the fs_info rb tree for the block group
90 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
91 struct btrfs_block_group_cache
*block_group
)
94 struct rb_node
*parent
= NULL
;
95 struct btrfs_block_group_cache
*cache
;
97 spin_lock(&info
->block_group_cache_lock
);
98 p
= &info
->block_group_cache_tree
.rb_node
;
102 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
104 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
106 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
109 spin_unlock(&info
->block_group_cache_lock
);
114 rb_link_node(&block_group
->cache_node
, parent
, p
);
115 rb_insert_color(&block_group
->cache_node
,
116 &info
->block_group_cache_tree
);
117 spin_unlock(&info
->block_group_cache_lock
);
123 * This will return the block group at or after bytenr if contains is 0, else
124 * it will return the block group that contains the bytenr
126 static struct btrfs_block_group_cache
*
127 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
130 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
134 spin_lock(&info
->block_group_cache_lock
);
135 n
= info
->block_group_cache_tree
.rb_node
;
138 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
140 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
141 start
= cache
->key
.objectid
;
143 if (bytenr
< start
) {
144 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
147 } else if (bytenr
> start
) {
148 if (contains
&& bytenr
<= end
) {
159 atomic_inc(&ret
->count
);
160 spin_unlock(&info
->block_group_cache_lock
);
165 static int add_excluded_extent(struct btrfs_root
*root
,
166 u64 start
, u64 num_bytes
)
168 u64 end
= start
+ num_bytes
- 1;
169 set_extent_bits(&root
->fs_info
->freed_extents
[0],
170 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
171 set_extent_bits(&root
->fs_info
->freed_extents
[1],
172 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
176 static void free_excluded_extents(struct btrfs_root
*root
,
177 struct btrfs_block_group_cache
*cache
)
181 start
= cache
->key
.objectid
;
182 end
= start
+ cache
->key
.offset
- 1;
184 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
185 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
186 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
187 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
190 static int exclude_super_stripes(struct btrfs_root
*root
,
191 struct btrfs_block_group_cache
*cache
)
198 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
199 bytenr
= btrfs_sb_offset(i
);
200 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
201 cache
->key
.objectid
, bytenr
,
202 0, &logical
, &nr
, &stripe_len
);
206 cache
->bytes_super
+= stripe_len
;
207 ret
= add_excluded_extent(root
, logical
[nr
],
217 static struct btrfs_caching_control
*
218 get_caching_control(struct btrfs_block_group_cache
*cache
)
220 struct btrfs_caching_control
*ctl
;
222 spin_lock(&cache
->lock
);
223 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
224 spin_unlock(&cache
->lock
);
228 ctl
= cache
->caching_ctl
;
229 atomic_inc(&ctl
->count
);
230 spin_unlock(&cache
->lock
);
234 static void put_caching_control(struct btrfs_caching_control
*ctl
)
236 if (atomic_dec_and_test(&ctl
->count
))
241 * this is only called by cache_block_group, since we could have freed extents
242 * we need to check the pinned_extents for any extents that can't be used yet
243 * since their free space will be released as soon as the transaction commits.
245 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
246 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
248 u64 extent_start
, extent_end
, size
, total_added
= 0;
251 while (start
< end
) {
252 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
253 &extent_start
, &extent_end
,
254 EXTENT_DIRTY
| EXTENT_UPTODATE
);
258 if (extent_start
== start
) {
259 start
= extent_end
+ 1;
260 } else if (extent_start
> start
&& extent_start
< end
) {
261 size
= extent_start
- start
;
263 ret
= btrfs_add_free_space(block_group
, start
,
266 start
= extent_end
+ 1;
275 ret
= btrfs_add_free_space(block_group
, start
, size
);
282 static int caching_kthread(void *data
)
284 struct btrfs_block_group_cache
*block_group
= data
;
285 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
286 struct btrfs_caching_control
*caching_ctl
= block_group
->caching_ctl
;
287 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
288 struct btrfs_path
*path
;
289 struct extent_buffer
*leaf
;
290 struct btrfs_key key
;
296 path
= btrfs_alloc_path();
300 exclude_super_stripes(extent_root
, block_group
);
301 spin_lock(&block_group
->space_info
->lock
);
302 block_group
->space_info
->bytes_super
+= block_group
->bytes_super
;
303 spin_unlock(&block_group
->space_info
->lock
);
305 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
308 * We don't want to deadlock with somebody trying to allocate a new
309 * extent for the extent root while also trying to search the extent
310 * root to add free space. So we skip locking and search the commit
311 * root, since its read-only
313 path
->skip_locking
= 1;
314 path
->search_commit_root
= 1;
319 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
321 mutex_lock(&caching_ctl
->mutex
);
322 /* need to make sure the commit_root doesn't disappear */
323 down_read(&fs_info
->extent_commit_sem
);
325 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
329 leaf
= path
->nodes
[0];
330 nritems
= btrfs_header_nritems(leaf
);
334 if (fs_info
->closing
> 1) {
339 if (path
->slots
[0] < nritems
) {
340 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
342 ret
= find_next_key(path
, 0, &key
);
346 caching_ctl
->progress
= last
;
347 btrfs_release_path(extent_root
, path
);
348 up_read(&fs_info
->extent_commit_sem
);
349 mutex_unlock(&caching_ctl
->mutex
);
350 if (btrfs_transaction_in_commit(fs_info
))
357 if (key
.objectid
< block_group
->key
.objectid
) {
362 if (key
.objectid
>= block_group
->key
.objectid
+
363 block_group
->key
.offset
)
366 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
367 total_found
+= add_new_free_space(block_group
,
370 last
= key
.objectid
+ key
.offset
;
372 if (total_found
> (1024 * 1024 * 2)) {
374 wake_up(&caching_ctl
->wait
);
381 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
382 block_group
->key
.objectid
+
383 block_group
->key
.offset
);
384 caching_ctl
->progress
= (u64
)-1;
386 spin_lock(&block_group
->lock
);
387 block_group
->caching_ctl
= NULL
;
388 block_group
->cached
= BTRFS_CACHE_FINISHED
;
389 spin_unlock(&block_group
->lock
);
392 btrfs_free_path(path
);
393 up_read(&fs_info
->extent_commit_sem
);
395 free_excluded_extents(extent_root
, block_group
);
397 mutex_unlock(&caching_ctl
->mutex
);
398 wake_up(&caching_ctl
->wait
);
400 put_caching_control(caching_ctl
);
401 atomic_dec(&block_group
->space_info
->caching_threads
);
405 static int cache_block_group(struct btrfs_block_group_cache
*cache
)
407 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
408 struct btrfs_caching_control
*caching_ctl
;
409 struct task_struct
*tsk
;
413 if (cache
->cached
!= BTRFS_CACHE_NO
)
416 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_KERNEL
);
417 BUG_ON(!caching_ctl
);
419 INIT_LIST_HEAD(&caching_ctl
->list
);
420 mutex_init(&caching_ctl
->mutex
);
421 init_waitqueue_head(&caching_ctl
->wait
);
422 caching_ctl
->block_group
= cache
;
423 caching_ctl
->progress
= cache
->key
.objectid
;
424 /* one for caching kthread, one for caching block group list */
425 atomic_set(&caching_ctl
->count
, 2);
427 spin_lock(&cache
->lock
);
428 if (cache
->cached
!= BTRFS_CACHE_NO
) {
429 spin_unlock(&cache
->lock
);
433 cache
->caching_ctl
= caching_ctl
;
434 cache
->cached
= BTRFS_CACHE_STARTED
;
435 spin_unlock(&cache
->lock
);
437 down_write(&fs_info
->extent_commit_sem
);
438 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
439 up_write(&fs_info
->extent_commit_sem
);
441 atomic_inc(&cache
->space_info
->caching_threads
);
443 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
444 cache
->key
.objectid
);
447 printk(KERN_ERR
"error running thread %d\n", ret
);
455 * return the block group that starts at or after bytenr
457 static struct btrfs_block_group_cache
*
458 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
460 struct btrfs_block_group_cache
*cache
;
462 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
468 * return the block group that contains the given bytenr
470 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
471 struct btrfs_fs_info
*info
,
474 struct btrfs_block_group_cache
*cache
;
476 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
481 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
483 if (atomic_dec_and_test(&cache
->count
))
487 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
490 struct list_head
*head
= &info
->space_info
;
491 struct btrfs_space_info
*found
;
494 list_for_each_entry_rcu(found
, head
, list
) {
495 if (found
->flags
== flags
) {
505 * after adding space to the filesystem, we need to clear the full flags
506 * on all the space infos.
508 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
510 struct list_head
*head
= &info
->space_info
;
511 struct btrfs_space_info
*found
;
514 list_for_each_entry_rcu(found
, head
, list
)
519 static u64
div_factor(u64 num
, int factor
)
528 u64
btrfs_find_block_group(struct btrfs_root
*root
,
529 u64 search_start
, u64 search_hint
, int owner
)
531 struct btrfs_block_group_cache
*cache
;
533 u64 last
= max(search_hint
, search_start
);
540 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
544 spin_lock(&cache
->lock
);
545 last
= cache
->key
.objectid
+ cache
->key
.offset
;
546 used
= btrfs_block_group_used(&cache
->item
);
548 if ((full_search
|| !cache
->ro
) &&
549 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
550 if (used
+ cache
->pinned
+ cache
->reserved
<
551 div_factor(cache
->key
.offset
, factor
)) {
552 group_start
= cache
->key
.objectid
;
553 spin_unlock(&cache
->lock
);
554 btrfs_put_block_group(cache
);
558 spin_unlock(&cache
->lock
);
559 btrfs_put_block_group(cache
);
567 if (!full_search
&& factor
< 10) {
577 /* simple helper to search for an existing extent at a given offset */
578 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
581 struct btrfs_key key
;
582 struct btrfs_path
*path
;
584 path
= btrfs_alloc_path();
586 key
.objectid
= start
;
588 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
589 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
591 btrfs_free_path(path
);
596 * Back reference rules. Back refs have three main goals:
598 * 1) differentiate between all holders of references to an extent so that
599 * when a reference is dropped we can make sure it was a valid reference
600 * before freeing the extent.
602 * 2) Provide enough information to quickly find the holders of an extent
603 * if we notice a given block is corrupted or bad.
605 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
606 * maintenance. This is actually the same as #2, but with a slightly
607 * different use case.
609 * There are two kinds of back refs. The implicit back refs is optimized
610 * for pointers in non-shared tree blocks. For a given pointer in a block,
611 * back refs of this kind provide information about the block's owner tree
612 * and the pointer's key. These information allow us to find the block by
613 * b-tree searching. The full back refs is for pointers in tree blocks not
614 * referenced by their owner trees. The location of tree block is recorded
615 * in the back refs. Actually the full back refs is generic, and can be
616 * used in all cases the implicit back refs is used. The major shortcoming
617 * of the full back refs is its overhead. Every time a tree block gets
618 * COWed, we have to update back refs entry for all pointers in it.
620 * For a newly allocated tree block, we use implicit back refs for
621 * pointers in it. This means most tree related operations only involve
622 * implicit back refs. For a tree block created in old transaction, the
623 * only way to drop a reference to it is COW it. So we can detect the
624 * event that tree block loses its owner tree's reference and do the
625 * back refs conversion.
627 * When a tree block is COW'd through a tree, there are four cases:
629 * The reference count of the block is one and the tree is the block's
630 * owner tree. Nothing to do in this case.
632 * The reference count of the block is one and the tree is not the
633 * block's owner tree. In this case, full back refs is used for pointers
634 * in the block. Remove these full back refs, add implicit back refs for
635 * every pointers in the new block.
637 * The reference count of the block is greater than one and the tree is
638 * the block's owner tree. In this case, implicit back refs is used for
639 * pointers in the block. Add full back refs for every pointers in the
640 * block, increase lower level extents' reference counts. The original
641 * implicit back refs are entailed to the new block.
643 * The reference count of the block is greater than one and the tree is
644 * not the block's owner tree. Add implicit back refs for every pointer in
645 * the new block, increase lower level extents' reference count.
647 * Back Reference Key composing:
649 * The key objectid corresponds to the first byte in the extent,
650 * The key type is used to differentiate between types of back refs.
651 * There are different meanings of the key offset for different types
654 * File extents can be referenced by:
656 * - multiple snapshots, subvolumes, or different generations in one subvol
657 * - different files inside a single subvolume
658 * - different offsets inside a file (bookend extents in file.c)
660 * The extent ref structure for the implicit back refs has fields for:
662 * - Objectid of the subvolume root
663 * - objectid of the file holding the reference
664 * - original offset in the file
665 * - how many bookend extents
667 * The key offset for the implicit back refs is hash of the first
670 * The extent ref structure for the full back refs has field for:
672 * - number of pointers in the tree leaf
674 * The key offset for the implicit back refs is the first byte of
677 * When a file extent is allocated, The implicit back refs is used.
678 * the fields are filled in:
680 * (root_key.objectid, inode objectid, offset in file, 1)
682 * When a file extent is removed file truncation, we find the
683 * corresponding implicit back refs and check the following fields:
685 * (btrfs_header_owner(leaf), inode objectid, offset in file)
687 * Btree extents can be referenced by:
689 * - Different subvolumes
691 * Both the implicit back refs and the full back refs for tree blocks
692 * only consist of key. The key offset for the implicit back refs is
693 * objectid of block's owner tree. The key offset for the full back refs
694 * is the first byte of parent block.
696 * When implicit back refs is used, information about the lowest key and
697 * level of the tree block are required. These information are stored in
698 * tree block info structure.
701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
702 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
703 struct btrfs_root
*root
,
704 struct btrfs_path
*path
,
705 u64 owner
, u32 extra_size
)
707 struct btrfs_extent_item
*item
;
708 struct btrfs_extent_item_v0
*ei0
;
709 struct btrfs_extent_ref_v0
*ref0
;
710 struct btrfs_tree_block_info
*bi
;
711 struct extent_buffer
*leaf
;
712 struct btrfs_key key
;
713 struct btrfs_key found_key
;
714 u32 new_size
= sizeof(*item
);
718 leaf
= path
->nodes
[0];
719 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
721 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
722 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
723 struct btrfs_extent_item_v0
);
724 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
726 if (owner
== (u64
)-1) {
728 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
729 ret
= btrfs_next_leaf(root
, path
);
733 leaf
= path
->nodes
[0];
735 btrfs_item_key_to_cpu(leaf
, &found_key
,
737 BUG_ON(key
.objectid
!= found_key
.objectid
);
738 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
742 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
743 struct btrfs_extent_ref_v0
);
744 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
748 btrfs_release_path(root
, path
);
750 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
751 new_size
+= sizeof(*bi
);
753 new_size
-= sizeof(*ei0
);
754 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
755 new_size
+ extra_size
, 1);
760 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
763 leaf
= path
->nodes
[0];
764 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
765 btrfs_set_extent_refs(leaf
, item
, refs
);
766 /* FIXME: get real generation */
767 btrfs_set_extent_generation(leaf
, item
, 0);
768 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
769 btrfs_set_extent_flags(leaf
, item
,
770 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
771 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
772 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
773 /* FIXME: get first key of the block */
774 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
775 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
777 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
779 btrfs_mark_buffer_dirty(leaf
);
784 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
786 u32 high_crc
= ~(u32
)0;
787 u32 low_crc
= ~(u32
)0;
790 lenum
= cpu_to_le64(root_objectid
);
791 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
792 lenum
= cpu_to_le64(owner
);
793 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
794 lenum
= cpu_to_le64(offset
);
795 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
797 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
800 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
801 struct btrfs_extent_data_ref
*ref
)
803 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
804 btrfs_extent_data_ref_objectid(leaf
, ref
),
805 btrfs_extent_data_ref_offset(leaf
, ref
));
808 static int match_extent_data_ref(struct extent_buffer
*leaf
,
809 struct btrfs_extent_data_ref
*ref
,
810 u64 root_objectid
, u64 owner
, u64 offset
)
812 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
813 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
814 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
819 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
820 struct btrfs_root
*root
,
821 struct btrfs_path
*path
,
822 u64 bytenr
, u64 parent
,
824 u64 owner
, u64 offset
)
826 struct btrfs_key key
;
827 struct btrfs_extent_data_ref
*ref
;
828 struct extent_buffer
*leaf
;
834 key
.objectid
= bytenr
;
836 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
839 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
840 key
.offset
= hash_extent_data_ref(root_objectid
,
845 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
854 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
855 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
856 btrfs_release_path(root
, path
);
857 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
868 leaf
= path
->nodes
[0];
869 nritems
= btrfs_header_nritems(leaf
);
871 if (path
->slots
[0] >= nritems
) {
872 ret
= btrfs_next_leaf(root
, path
);
878 leaf
= path
->nodes
[0];
879 nritems
= btrfs_header_nritems(leaf
);
883 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
884 if (key
.objectid
!= bytenr
||
885 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
888 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
889 struct btrfs_extent_data_ref
);
891 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
894 btrfs_release_path(root
, path
);
906 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
907 struct btrfs_root
*root
,
908 struct btrfs_path
*path
,
909 u64 bytenr
, u64 parent
,
910 u64 root_objectid
, u64 owner
,
911 u64 offset
, int refs_to_add
)
913 struct btrfs_key key
;
914 struct extent_buffer
*leaf
;
919 key
.objectid
= bytenr
;
921 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
923 size
= sizeof(struct btrfs_shared_data_ref
);
925 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
926 key
.offset
= hash_extent_data_ref(root_objectid
,
928 size
= sizeof(struct btrfs_extent_data_ref
);
931 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
932 if (ret
&& ret
!= -EEXIST
)
935 leaf
= path
->nodes
[0];
937 struct btrfs_shared_data_ref
*ref
;
938 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
939 struct btrfs_shared_data_ref
);
941 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
943 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
944 num_refs
+= refs_to_add
;
945 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
948 struct btrfs_extent_data_ref
*ref
;
949 while (ret
== -EEXIST
) {
950 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
951 struct btrfs_extent_data_ref
);
952 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
955 btrfs_release_path(root
, path
);
957 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
959 if (ret
&& ret
!= -EEXIST
)
962 leaf
= path
->nodes
[0];
964 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
965 struct btrfs_extent_data_ref
);
967 btrfs_set_extent_data_ref_root(leaf
, ref
,
969 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
970 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
971 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
973 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
974 num_refs
+= refs_to_add
;
975 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
978 btrfs_mark_buffer_dirty(leaf
);
981 btrfs_release_path(root
, path
);
985 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
986 struct btrfs_root
*root
,
987 struct btrfs_path
*path
,
990 struct btrfs_key key
;
991 struct btrfs_extent_data_ref
*ref1
= NULL
;
992 struct btrfs_shared_data_ref
*ref2
= NULL
;
993 struct extent_buffer
*leaf
;
997 leaf
= path
->nodes
[0];
998 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1000 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1001 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1002 struct btrfs_extent_data_ref
);
1003 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1004 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1005 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1006 struct btrfs_shared_data_ref
);
1007 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1008 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1009 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1010 struct btrfs_extent_ref_v0
*ref0
;
1011 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1012 struct btrfs_extent_ref_v0
);
1013 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1019 BUG_ON(num_refs
< refs_to_drop
);
1020 num_refs
-= refs_to_drop
;
1022 if (num_refs
== 0) {
1023 ret
= btrfs_del_item(trans
, root
, path
);
1025 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1026 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1027 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1028 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1031 struct btrfs_extent_ref_v0
*ref0
;
1032 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1033 struct btrfs_extent_ref_v0
);
1034 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1037 btrfs_mark_buffer_dirty(leaf
);
1042 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1043 struct btrfs_path
*path
,
1044 struct btrfs_extent_inline_ref
*iref
)
1046 struct btrfs_key key
;
1047 struct extent_buffer
*leaf
;
1048 struct btrfs_extent_data_ref
*ref1
;
1049 struct btrfs_shared_data_ref
*ref2
;
1052 leaf
= path
->nodes
[0];
1053 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1055 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1056 BTRFS_EXTENT_DATA_REF_KEY
) {
1057 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1058 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1060 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1061 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1063 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1064 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1065 struct btrfs_extent_data_ref
);
1066 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1067 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1068 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1069 struct btrfs_shared_data_ref
);
1070 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1071 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1072 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1073 struct btrfs_extent_ref_v0
*ref0
;
1074 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1075 struct btrfs_extent_ref_v0
);
1076 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1084 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1085 struct btrfs_root
*root
,
1086 struct btrfs_path
*path
,
1087 u64 bytenr
, u64 parent
,
1090 struct btrfs_key key
;
1093 key
.objectid
= bytenr
;
1095 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1096 key
.offset
= parent
;
1098 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1099 key
.offset
= root_objectid
;
1102 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 if (ret
== -ENOENT
&& parent
) {
1107 btrfs_release_path(root
, path
);
1108 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1109 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1117 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1118 struct btrfs_root
*root
,
1119 struct btrfs_path
*path
,
1120 u64 bytenr
, u64 parent
,
1123 struct btrfs_key key
;
1126 key
.objectid
= bytenr
;
1128 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1129 key
.offset
= parent
;
1131 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1132 key
.offset
= root_objectid
;
1135 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1136 btrfs_release_path(root
, path
);
1140 static inline int extent_ref_type(u64 parent
, u64 owner
)
1143 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1145 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1147 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1150 type
= BTRFS_SHARED_DATA_REF_KEY
;
1152 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1157 static int find_next_key(struct btrfs_path
*path
, int level
,
1158 struct btrfs_key
*key
)
1161 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1162 if (!path
->nodes
[level
])
1164 if (path
->slots
[level
] + 1 >=
1165 btrfs_header_nritems(path
->nodes
[level
]))
1168 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1169 path
->slots
[level
] + 1);
1171 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1172 path
->slots
[level
] + 1);
1179 * look for inline back ref. if back ref is found, *ref_ret is set
1180 * to the address of inline back ref, and 0 is returned.
1182 * if back ref isn't found, *ref_ret is set to the address where it
1183 * should be inserted, and -ENOENT is returned.
1185 * if insert is true and there are too many inline back refs, the path
1186 * points to the extent item, and -EAGAIN is returned.
1188 * NOTE: inline back refs are ordered in the same way that back ref
1189 * items in the tree are ordered.
1191 static noinline_for_stack
1192 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1193 struct btrfs_root
*root
,
1194 struct btrfs_path
*path
,
1195 struct btrfs_extent_inline_ref
**ref_ret
,
1196 u64 bytenr
, u64 num_bytes
,
1197 u64 parent
, u64 root_objectid
,
1198 u64 owner
, u64 offset
, int insert
)
1200 struct btrfs_key key
;
1201 struct extent_buffer
*leaf
;
1202 struct btrfs_extent_item
*ei
;
1203 struct btrfs_extent_inline_ref
*iref
;
1214 key
.objectid
= bytenr
;
1215 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1216 key
.offset
= num_bytes
;
1218 want
= extent_ref_type(parent
, owner
);
1220 extra_size
= btrfs_extent_inline_ref_size(want
);
1221 path
->keep_locks
= 1;
1224 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1231 leaf
= path
->nodes
[0];
1232 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234 if (item_size
< sizeof(*ei
)) {
1239 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1245 leaf
= path
->nodes
[0];
1246 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1249 BUG_ON(item_size
< sizeof(*ei
));
1251 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1252 flags
= btrfs_extent_flags(leaf
, ei
);
1254 ptr
= (unsigned long)(ei
+ 1);
1255 end
= (unsigned long)ei
+ item_size
;
1257 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1258 ptr
+= sizeof(struct btrfs_tree_block_info
);
1261 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1270 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1271 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1275 ptr
+= btrfs_extent_inline_ref_size(type
);
1279 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1280 struct btrfs_extent_data_ref
*dref
;
1281 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1282 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1287 if (hash_extent_data_ref_item(leaf
, dref
) <
1288 hash_extent_data_ref(root_objectid
, owner
, offset
))
1292 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1294 if (parent
== ref_offset
) {
1298 if (ref_offset
< parent
)
1301 if (root_objectid
== ref_offset
) {
1305 if (ref_offset
< root_objectid
)
1309 ptr
+= btrfs_extent_inline_ref_size(type
);
1311 if (err
== -ENOENT
&& insert
) {
1312 if (item_size
+ extra_size
>=
1313 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1318 * To add new inline back ref, we have to make sure
1319 * there is no corresponding back ref item.
1320 * For simplicity, we just do not add new inline back
1321 * ref if there is any kind of item for this block
1323 if (find_next_key(path
, 0, &key
) == 0 &&
1324 key
.objectid
== bytenr
&&
1325 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1330 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1333 path
->keep_locks
= 0;
1334 btrfs_unlock_up_safe(path
, 1);
1340 * helper to add new inline back ref
1342 static noinline_for_stack
1343 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1344 struct btrfs_root
*root
,
1345 struct btrfs_path
*path
,
1346 struct btrfs_extent_inline_ref
*iref
,
1347 u64 parent
, u64 root_objectid
,
1348 u64 owner
, u64 offset
, int refs_to_add
,
1349 struct btrfs_delayed_extent_op
*extent_op
)
1351 struct extent_buffer
*leaf
;
1352 struct btrfs_extent_item
*ei
;
1355 unsigned long item_offset
;
1361 leaf
= path
->nodes
[0];
1362 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1363 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1365 type
= extent_ref_type(parent
, owner
);
1366 size
= btrfs_extent_inline_ref_size(type
);
1368 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1371 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1372 refs
= btrfs_extent_refs(leaf
, ei
);
1373 refs
+= refs_to_add
;
1374 btrfs_set_extent_refs(leaf
, ei
, refs
);
1376 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1378 ptr
= (unsigned long)ei
+ item_offset
;
1379 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1380 if (ptr
< end
- size
)
1381 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1384 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1385 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1386 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1387 struct btrfs_extent_data_ref
*dref
;
1388 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1389 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1390 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1391 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1392 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1393 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1394 struct btrfs_shared_data_ref
*sref
;
1395 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1396 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1397 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1398 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1399 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1401 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1403 btrfs_mark_buffer_dirty(leaf
);
1407 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1408 struct btrfs_root
*root
,
1409 struct btrfs_path
*path
,
1410 struct btrfs_extent_inline_ref
**ref_ret
,
1411 u64 bytenr
, u64 num_bytes
, u64 parent
,
1412 u64 root_objectid
, u64 owner
, u64 offset
)
1416 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1417 bytenr
, num_bytes
, parent
,
1418 root_objectid
, owner
, offset
, 0);
1422 btrfs_release_path(root
, path
);
1425 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1426 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1429 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1430 root_objectid
, owner
, offset
);
1436 * helper to update/remove inline back ref
1438 static noinline_for_stack
1439 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1440 struct btrfs_root
*root
,
1441 struct btrfs_path
*path
,
1442 struct btrfs_extent_inline_ref
*iref
,
1444 struct btrfs_delayed_extent_op
*extent_op
)
1446 struct extent_buffer
*leaf
;
1447 struct btrfs_extent_item
*ei
;
1448 struct btrfs_extent_data_ref
*dref
= NULL
;
1449 struct btrfs_shared_data_ref
*sref
= NULL
;
1458 leaf
= path
->nodes
[0];
1459 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1460 refs
= btrfs_extent_refs(leaf
, ei
);
1461 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1462 refs
+= refs_to_mod
;
1463 btrfs_set_extent_refs(leaf
, ei
, refs
);
1465 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1467 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1469 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1470 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1471 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1472 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1473 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1474 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1477 BUG_ON(refs_to_mod
!= -1);
1480 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1481 refs
+= refs_to_mod
;
1484 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1485 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1487 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1489 size
= btrfs_extent_inline_ref_size(type
);
1490 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1491 ptr
= (unsigned long)iref
;
1492 end
= (unsigned long)ei
+ item_size
;
1493 if (ptr
+ size
< end
)
1494 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1497 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1500 btrfs_mark_buffer_dirty(leaf
);
1504 static noinline_for_stack
1505 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1506 struct btrfs_root
*root
,
1507 struct btrfs_path
*path
,
1508 u64 bytenr
, u64 num_bytes
, u64 parent
,
1509 u64 root_objectid
, u64 owner
,
1510 u64 offset
, int refs_to_add
,
1511 struct btrfs_delayed_extent_op
*extent_op
)
1513 struct btrfs_extent_inline_ref
*iref
;
1516 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1517 bytenr
, num_bytes
, parent
,
1518 root_objectid
, owner
, offset
, 1);
1520 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1521 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1522 refs_to_add
, extent_op
);
1523 } else if (ret
== -ENOENT
) {
1524 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1525 parent
, root_objectid
,
1526 owner
, offset
, refs_to_add
,
1532 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1533 struct btrfs_root
*root
,
1534 struct btrfs_path
*path
,
1535 u64 bytenr
, u64 parent
, u64 root_objectid
,
1536 u64 owner
, u64 offset
, int refs_to_add
)
1539 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1540 BUG_ON(refs_to_add
!= 1);
1541 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1542 parent
, root_objectid
);
1544 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1545 parent
, root_objectid
,
1546 owner
, offset
, refs_to_add
);
1551 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1552 struct btrfs_root
*root
,
1553 struct btrfs_path
*path
,
1554 struct btrfs_extent_inline_ref
*iref
,
1555 int refs_to_drop
, int is_data
)
1559 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1561 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1562 -refs_to_drop
, NULL
);
1563 } else if (is_data
) {
1564 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1566 ret
= btrfs_del_item(trans
, root
, path
);
1571 #ifdef BIO_RW_DISCARD
1572 static void btrfs_issue_discard(struct block_device
*bdev
,
1575 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
);
1579 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1582 #ifdef BIO_RW_DISCARD
1584 u64 map_length
= num_bytes
;
1585 struct btrfs_multi_bio
*multi
= NULL
;
1587 /* Tell the block device(s) that the sectors can be discarded */
1588 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1589 bytenr
, &map_length
, &multi
, 0);
1591 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1594 if (map_length
> num_bytes
)
1595 map_length
= num_bytes
;
1597 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1598 btrfs_issue_discard(stripe
->dev
->bdev
,
1611 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1612 struct btrfs_root
*root
,
1613 u64 bytenr
, u64 num_bytes
, u64 parent
,
1614 u64 root_objectid
, u64 owner
, u64 offset
)
1617 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1618 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1620 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1621 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1622 parent
, root_objectid
, (int)owner
,
1623 BTRFS_ADD_DELAYED_REF
, NULL
);
1625 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1626 parent
, root_objectid
, owner
, offset
,
1627 BTRFS_ADD_DELAYED_REF
, NULL
);
1632 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1633 struct btrfs_root
*root
,
1634 u64 bytenr
, u64 num_bytes
,
1635 u64 parent
, u64 root_objectid
,
1636 u64 owner
, u64 offset
, int refs_to_add
,
1637 struct btrfs_delayed_extent_op
*extent_op
)
1639 struct btrfs_path
*path
;
1640 struct extent_buffer
*leaf
;
1641 struct btrfs_extent_item
*item
;
1646 path
= btrfs_alloc_path();
1651 path
->leave_spinning
= 1;
1652 /* this will setup the path even if it fails to insert the back ref */
1653 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1654 path
, bytenr
, num_bytes
, parent
,
1655 root_objectid
, owner
, offset
,
1656 refs_to_add
, extent_op
);
1660 if (ret
!= -EAGAIN
) {
1665 leaf
= path
->nodes
[0];
1666 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1667 refs
= btrfs_extent_refs(leaf
, item
);
1668 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1670 __run_delayed_extent_op(extent_op
, leaf
, item
);
1672 btrfs_mark_buffer_dirty(leaf
);
1673 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1676 path
->leave_spinning
= 1;
1678 /* now insert the actual backref */
1679 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1680 path
, bytenr
, parent
, root_objectid
,
1681 owner
, offset
, refs_to_add
);
1684 btrfs_free_path(path
);
1688 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1689 struct btrfs_root
*root
,
1690 struct btrfs_delayed_ref_node
*node
,
1691 struct btrfs_delayed_extent_op
*extent_op
,
1692 int insert_reserved
)
1695 struct btrfs_delayed_data_ref
*ref
;
1696 struct btrfs_key ins
;
1701 ins
.objectid
= node
->bytenr
;
1702 ins
.offset
= node
->num_bytes
;
1703 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1705 ref
= btrfs_delayed_node_to_data_ref(node
);
1706 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1707 parent
= ref
->parent
;
1709 ref_root
= ref
->root
;
1711 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1713 BUG_ON(extent_op
->update_key
);
1714 flags
|= extent_op
->flags_to_set
;
1716 ret
= alloc_reserved_file_extent(trans
, root
,
1717 parent
, ref_root
, flags
,
1718 ref
->objectid
, ref
->offset
,
1719 &ins
, node
->ref_mod
);
1720 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1721 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1722 node
->num_bytes
, parent
,
1723 ref_root
, ref
->objectid
,
1724 ref
->offset
, node
->ref_mod
,
1726 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1727 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1728 node
->num_bytes
, parent
,
1729 ref_root
, ref
->objectid
,
1730 ref
->offset
, node
->ref_mod
,
1738 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1739 struct extent_buffer
*leaf
,
1740 struct btrfs_extent_item
*ei
)
1742 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1743 if (extent_op
->update_flags
) {
1744 flags
|= extent_op
->flags_to_set
;
1745 btrfs_set_extent_flags(leaf
, ei
, flags
);
1748 if (extent_op
->update_key
) {
1749 struct btrfs_tree_block_info
*bi
;
1750 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1751 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1752 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1756 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1757 struct btrfs_root
*root
,
1758 struct btrfs_delayed_ref_node
*node
,
1759 struct btrfs_delayed_extent_op
*extent_op
)
1761 struct btrfs_key key
;
1762 struct btrfs_path
*path
;
1763 struct btrfs_extent_item
*ei
;
1764 struct extent_buffer
*leaf
;
1769 path
= btrfs_alloc_path();
1773 key
.objectid
= node
->bytenr
;
1774 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1775 key
.offset
= node
->num_bytes
;
1778 path
->leave_spinning
= 1;
1779 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1790 leaf
= path
->nodes
[0];
1791 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1792 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1793 if (item_size
< sizeof(*ei
)) {
1794 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1800 leaf
= path
->nodes
[0];
1801 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1804 BUG_ON(item_size
< sizeof(*ei
));
1805 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1806 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1808 btrfs_mark_buffer_dirty(leaf
);
1810 btrfs_free_path(path
);
1814 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1815 struct btrfs_root
*root
,
1816 struct btrfs_delayed_ref_node
*node
,
1817 struct btrfs_delayed_extent_op
*extent_op
,
1818 int insert_reserved
)
1821 struct btrfs_delayed_tree_ref
*ref
;
1822 struct btrfs_key ins
;
1826 ins
.objectid
= node
->bytenr
;
1827 ins
.offset
= node
->num_bytes
;
1828 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1830 ref
= btrfs_delayed_node_to_tree_ref(node
);
1831 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1832 parent
= ref
->parent
;
1834 ref_root
= ref
->root
;
1836 BUG_ON(node
->ref_mod
!= 1);
1837 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1838 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
1839 !extent_op
->update_key
);
1840 ret
= alloc_reserved_tree_block(trans
, root
,
1842 extent_op
->flags_to_set
,
1845 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1846 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1847 node
->num_bytes
, parent
, ref_root
,
1848 ref
->level
, 0, 1, extent_op
);
1849 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1850 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1851 node
->num_bytes
, parent
, ref_root
,
1852 ref
->level
, 0, 1, extent_op
);
1860 /* helper function to actually process a single delayed ref entry */
1861 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
1862 struct btrfs_root
*root
,
1863 struct btrfs_delayed_ref_node
*node
,
1864 struct btrfs_delayed_extent_op
*extent_op
,
1865 int insert_reserved
)
1868 if (btrfs_delayed_ref_is_head(node
)) {
1869 struct btrfs_delayed_ref_head
*head
;
1871 * we've hit the end of the chain and we were supposed
1872 * to insert this extent into the tree. But, it got
1873 * deleted before we ever needed to insert it, so all
1874 * we have to do is clean up the accounting
1877 head
= btrfs_delayed_node_to_head(node
);
1878 if (insert_reserved
) {
1880 struct extent_buffer
*must_clean
= NULL
;
1882 ret
= pin_down_bytes(trans
, root
, NULL
,
1883 node
->bytenr
, node
->num_bytes
,
1884 head
->is_data
, 1, &must_clean
);
1889 clean_tree_block(NULL
, root
, must_clean
);
1890 btrfs_tree_unlock(must_clean
);
1891 free_extent_buffer(must_clean
);
1893 if (head
->is_data
) {
1894 ret
= btrfs_del_csums(trans
, root
,
1900 ret
= btrfs_free_reserved_extent(root
,
1906 mutex_unlock(&head
->mutex
);
1910 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
1911 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1912 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
1914 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
1915 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1916 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
1923 static noinline
struct btrfs_delayed_ref_node
*
1924 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
1926 struct rb_node
*node
;
1927 struct btrfs_delayed_ref_node
*ref
;
1928 int action
= BTRFS_ADD_DELAYED_REF
;
1931 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1932 * this prevents ref count from going down to zero when
1933 * there still are pending delayed ref.
1935 node
= rb_prev(&head
->node
.rb_node
);
1939 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
1941 if (ref
->bytenr
!= head
->node
.bytenr
)
1943 if (ref
->action
== action
)
1945 node
= rb_prev(node
);
1947 if (action
== BTRFS_ADD_DELAYED_REF
) {
1948 action
= BTRFS_DROP_DELAYED_REF
;
1954 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
1955 struct btrfs_root
*root
,
1956 struct list_head
*cluster
)
1958 struct btrfs_delayed_ref_root
*delayed_refs
;
1959 struct btrfs_delayed_ref_node
*ref
;
1960 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
1961 struct btrfs_delayed_extent_op
*extent_op
;
1964 int must_insert_reserved
= 0;
1966 delayed_refs
= &trans
->transaction
->delayed_refs
;
1969 /* pick a new head ref from the cluster list */
1970 if (list_empty(cluster
))
1973 locked_ref
= list_entry(cluster
->next
,
1974 struct btrfs_delayed_ref_head
, cluster
);
1976 /* grab the lock that says we are going to process
1977 * all the refs for this head */
1978 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
1981 * we may have dropped the spin lock to get the head
1982 * mutex lock, and that might have given someone else
1983 * time to free the head. If that's true, it has been
1984 * removed from our list and we can move on.
1986 if (ret
== -EAGAIN
) {
1994 * record the must insert reserved flag before we
1995 * drop the spin lock.
1997 must_insert_reserved
= locked_ref
->must_insert_reserved
;
1998 locked_ref
->must_insert_reserved
= 0;
2000 extent_op
= locked_ref
->extent_op
;
2001 locked_ref
->extent_op
= NULL
;
2004 * locked_ref is the head node, so we have to go one
2005 * node back for any delayed ref updates
2007 ref
= select_delayed_ref(locked_ref
);
2009 /* All delayed refs have been processed, Go ahead
2010 * and send the head node to run_one_delayed_ref,
2011 * so that any accounting fixes can happen
2013 ref
= &locked_ref
->node
;
2015 if (extent_op
&& must_insert_reserved
) {
2021 spin_unlock(&delayed_refs
->lock
);
2023 ret
= run_delayed_extent_op(trans
, root
,
2029 spin_lock(&delayed_refs
->lock
);
2033 list_del_init(&locked_ref
->cluster
);
2038 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2039 delayed_refs
->num_entries
--;
2041 spin_unlock(&delayed_refs
->lock
);
2043 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2044 must_insert_reserved
);
2047 btrfs_put_delayed_ref(ref
);
2052 spin_lock(&delayed_refs
->lock
);
2058 * this starts processing the delayed reference count updates and
2059 * extent insertions we have queued up so far. count can be
2060 * 0, which means to process everything in the tree at the start
2061 * of the run (but not newly added entries), or it can be some target
2062 * number you'd like to process.
2064 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2065 struct btrfs_root
*root
, unsigned long count
)
2067 struct rb_node
*node
;
2068 struct btrfs_delayed_ref_root
*delayed_refs
;
2069 struct btrfs_delayed_ref_node
*ref
;
2070 struct list_head cluster
;
2072 int run_all
= count
== (unsigned long)-1;
2075 if (root
== root
->fs_info
->extent_root
)
2076 root
= root
->fs_info
->tree_root
;
2078 delayed_refs
= &trans
->transaction
->delayed_refs
;
2079 INIT_LIST_HEAD(&cluster
);
2081 spin_lock(&delayed_refs
->lock
);
2083 count
= delayed_refs
->num_entries
* 2;
2087 if (!(run_all
|| run_most
) &&
2088 delayed_refs
->num_heads_ready
< 64)
2092 * go find something we can process in the rbtree. We start at
2093 * the beginning of the tree, and then build a cluster
2094 * of refs to process starting at the first one we are able to
2097 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2098 delayed_refs
->run_delayed_start
);
2102 ret
= run_clustered_refs(trans
, root
, &cluster
);
2105 count
-= min_t(unsigned long, ret
, count
);
2112 node
= rb_first(&delayed_refs
->root
);
2115 count
= (unsigned long)-1;
2118 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2120 if (btrfs_delayed_ref_is_head(ref
)) {
2121 struct btrfs_delayed_ref_head
*head
;
2123 head
= btrfs_delayed_node_to_head(ref
);
2124 atomic_inc(&ref
->refs
);
2126 spin_unlock(&delayed_refs
->lock
);
2127 mutex_lock(&head
->mutex
);
2128 mutex_unlock(&head
->mutex
);
2130 btrfs_put_delayed_ref(ref
);
2134 node
= rb_next(node
);
2136 spin_unlock(&delayed_refs
->lock
);
2137 schedule_timeout(1);
2141 spin_unlock(&delayed_refs
->lock
);
2145 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2146 struct btrfs_root
*root
,
2147 u64 bytenr
, u64 num_bytes
, u64 flags
,
2150 struct btrfs_delayed_extent_op
*extent_op
;
2153 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2157 extent_op
->flags_to_set
= flags
;
2158 extent_op
->update_flags
= 1;
2159 extent_op
->update_key
= 0;
2160 extent_op
->is_data
= is_data
? 1 : 0;
2162 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2168 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2169 struct btrfs_root
*root
,
2170 struct btrfs_path
*path
,
2171 u64 objectid
, u64 offset
, u64 bytenr
)
2173 struct btrfs_delayed_ref_head
*head
;
2174 struct btrfs_delayed_ref_node
*ref
;
2175 struct btrfs_delayed_data_ref
*data_ref
;
2176 struct btrfs_delayed_ref_root
*delayed_refs
;
2177 struct rb_node
*node
;
2181 delayed_refs
= &trans
->transaction
->delayed_refs
;
2182 spin_lock(&delayed_refs
->lock
);
2183 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2187 if (!mutex_trylock(&head
->mutex
)) {
2188 atomic_inc(&head
->node
.refs
);
2189 spin_unlock(&delayed_refs
->lock
);
2191 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2193 mutex_lock(&head
->mutex
);
2194 mutex_unlock(&head
->mutex
);
2195 btrfs_put_delayed_ref(&head
->node
);
2199 node
= rb_prev(&head
->node
.rb_node
);
2203 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2205 if (ref
->bytenr
!= bytenr
)
2209 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2212 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2214 node
= rb_prev(node
);
2216 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2217 if (ref
->bytenr
== bytenr
)
2221 if (data_ref
->root
!= root
->root_key
.objectid
||
2222 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2227 mutex_unlock(&head
->mutex
);
2229 spin_unlock(&delayed_refs
->lock
);
2233 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2234 struct btrfs_root
*root
,
2235 struct btrfs_path
*path
,
2236 u64 objectid
, u64 offset
, u64 bytenr
)
2238 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2239 struct extent_buffer
*leaf
;
2240 struct btrfs_extent_data_ref
*ref
;
2241 struct btrfs_extent_inline_ref
*iref
;
2242 struct btrfs_extent_item
*ei
;
2243 struct btrfs_key key
;
2247 key
.objectid
= bytenr
;
2248 key
.offset
= (u64
)-1;
2249 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2251 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2257 if (path
->slots
[0] == 0)
2261 leaf
= path
->nodes
[0];
2262 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2264 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2268 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2269 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2270 if (item_size
< sizeof(*ei
)) {
2271 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2275 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2277 if (item_size
!= sizeof(*ei
) +
2278 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2281 if (btrfs_extent_generation(leaf
, ei
) <=
2282 btrfs_root_last_snapshot(&root
->root_item
))
2285 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2286 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2287 BTRFS_EXTENT_DATA_REF_KEY
)
2290 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2291 if (btrfs_extent_refs(leaf
, ei
) !=
2292 btrfs_extent_data_ref_count(leaf
, ref
) ||
2293 btrfs_extent_data_ref_root(leaf
, ref
) !=
2294 root
->root_key
.objectid
||
2295 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2296 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2304 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2305 struct btrfs_root
*root
,
2306 u64 objectid
, u64 offset
, u64 bytenr
)
2308 struct btrfs_path
*path
;
2312 path
= btrfs_alloc_path();
2317 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2319 if (ret
&& ret
!= -ENOENT
)
2322 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2324 } while (ret2
== -EAGAIN
);
2326 if (ret2
&& ret2
!= -ENOENT
) {
2331 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2334 btrfs_free_path(path
);
2339 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2340 struct extent_buffer
*buf
, u32 nr_extents
)
2342 struct btrfs_key key
;
2343 struct btrfs_file_extent_item
*fi
;
2351 if (!root
->ref_cows
)
2354 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2356 root_gen
= root
->root_key
.offset
;
2359 root_gen
= trans
->transid
- 1;
2362 level
= btrfs_header_level(buf
);
2363 nritems
= btrfs_header_nritems(buf
);
2366 struct btrfs_leaf_ref
*ref
;
2367 struct btrfs_extent_info
*info
;
2369 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2375 ref
->root_gen
= root_gen
;
2376 ref
->bytenr
= buf
->start
;
2377 ref
->owner
= btrfs_header_owner(buf
);
2378 ref
->generation
= btrfs_header_generation(buf
);
2379 ref
->nritems
= nr_extents
;
2380 info
= ref
->extents
;
2382 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2384 btrfs_item_key_to_cpu(buf
, &key
, i
);
2385 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2387 fi
= btrfs_item_ptr(buf
, i
,
2388 struct btrfs_file_extent_item
);
2389 if (btrfs_file_extent_type(buf
, fi
) ==
2390 BTRFS_FILE_EXTENT_INLINE
)
2392 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2393 if (disk_bytenr
== 0)
2396 info
->bytenr
= disk_bytenr
;
2398 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2399 info
->objectid
= key
.objectid
;
2400 info
->offset
= key
.offset
;
2404 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2405 if (ret
== -EEXIST
&& shared
) {
2406 struct btrfs_leaf_ref
*old
;
2407 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2409 btrfs_remove_leaf_ref(root
, old
);
2410 btrfs_free_leaf_ref(root
, old
);
2411 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2414 btrfs_free_leaf_ref(root
, ref
);
2420 /* when a block goes through cow, we update the reference counts of
2421 * everything that block points to. The internal pointers of the block
2422 * can be in just about any order, and it is likely to have clusters of
2423 * things that are close together and clusters of things that are not.
2425 * To help reduce the seeks that come with updating all of these reference
2426 * counts, sort them by byte number before actual updates are done.
2428 * struct refsort is used to match byte number to slot in the btree block.
2429 * we sort based on the byte number and then use the slot to actually
2432 * struct refsort is smaller than strcut btrfs_item and smaller than
2433 * struct btrfs_key_ptr. Since we're currently limited to the page size
2434 * for a btree block, there's no way for a kmalloc of refsorts for a
2435 * single node to be bigger than a page.
2443 * for passing into sort()
2445 static int refsort_cmp(const void *a_void
, const void *b_void
)
2447 const struct refsort
*a
= a_void
;
2448 const struct refsort
*b
= b_void
;
2450 if (a
->bytenr
< b
->bytenr
)
2452 if (a
->bytenr
> b
->bytenr
)
2458 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2459 struct btrfs_root
*root
,
2460 struct extent_buffer
*buf
,
2461 int full_backref
, int inc
)
2468 struct btrfs_key key
;
2469 struct btrfs_file_extent_item
*fi
;
2473 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2474 u64
, u64
, u64
, u64
, u64
, u64
);
2476 ref_root
= btrfs_header_owner(buf
);
2477 nritems
= btrfs_header_nritems(buf
);
2478 level
= btrfs_header_level(buf
);
2480 if (!root
->ref_cows
&& level
== 0)
2484 process_func
= btrfs_inc_extent_ref
;
2486 process_func
= btrfs_free_extent
;
2489 parent
= buf
->start
;
2493 for (i
= 0; i
< nritems
; i
++) {
2495 btrfs_item_key_to_cpu(buf
, &key
, i
);
2496 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2498 fi
= btrfs_item_ptr(buf
, i
,
2499 struct btrfs_file_extent_item
);
2500 if (btrfs_file_extent_type(buf
, fi
) ==
2501 BTRFS_FILE_EXTENT_INLINE
)
2503 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2507 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2508 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2509 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2510 parent
, ref_root
, key
.objectid
,
2515 bytenr
= btrfs_node_blockptr(buf
, i
);
2516 num_bytes
= btrfs_level_size(root
, level
- 1);
2517 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2518 parent
, ref_root
, level
- 1, 0);
2529 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2530 struct extent_buffer
*buf
, int full_backref
)
2532 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2535 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2536 struct extent_buffer
*buf
, int full_backref
)
2538 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2541 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2542 struct btrfs_root
*root
,
2543 struct btrfs_path
*path
,
2544 struct btrfs_block_group_cache
*cache
)
2547 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2549 struct extent_buffer
*leaf
;
2551 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2556 leaf
= path
->nodes
[0];
2557 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2558 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2559 btrfs_mark_buffer_dirty(leaf
);
2560 btrfs_release_path(extent_root
, path
);
2568 static struct btrfs_block_group_cache
*
2569 next_block_group(struct btrfs_root
*root
,
2570 struct btrfs_block_group_cache
*cache
)
2572 struct rb_node
*node
;
2573 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2574 node
= rb_next(&cache
->cache_node
);
2575 btrfs_put_block_group(cache
);
2577 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2579 atomic_inc(&cache
->count
);
2582 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2586 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2587 struct btrfs_root
*root
)
2589 struct btrfs_block_group_cache
*cache
;
2591 struct btrfs_path
*path
;
2594 path
= btrfs_alloc_path();
2600 err
= btrfs_run_delayed_refs(trans
, root
,
2605 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2609 cache
= next_block_group(root
, cache
);
2619 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2621 err
= write_one_cache_group(trans
, root
, path
, cache
);
2623 btrfs_put_block_group(cache
);
2626 btrfs_free_path(path
);
2630 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2632 struct btrfs_block_group_cache
*block_group
;
2635 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2636 if (!block_group
|| block_group
->ro
)
2639 btrfs_put_block_group(block_group
);
2643 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2644 u64 total_bytes
, u64 bytes_used
,
2645 struct btrfs_space_info
**space_info
)
2647 struct btrfs_space_info
*found
;
2649 found
= __find_space_info(info
, flags
);
2651 spin_lock(&found
->lock
);
2652 found
->total_bytes
+= total_bytes
;
2653 found
->bytes_used
+= bytes_used
;
2655 spin_unlock(&found
->lock
);
2656 *space_info
= found
;
2659 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2663 INIT_LIST_HEAD(&found
->block_groups
);
2664 init_rwsem(&found
->groups_sem
);
2665 spin_lock_init(&found
->lock
);
2666 found
->flags
= flags
;
2667 found
->total_bytes
= total_bytes
;
2668 found
->bytes_used
= bytes_used
;
2669 found
->bytes_pinned
= 0;
2670 found
->bytes_reserved
= 0;
2671 found
->bytes_readonly
= 0;
2672 found
->bytes_delalloc
= 0;
2674 found
->force_alloc
= 0;
2675 *space_info
= found
;
2676 list_add_rcu(&found
->list
, &info
->space_info
);
2677 atomic_set(&found
->caching_threads
, 0);
2681 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2683 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2684 BTRFS_BLOCK_GROUP_RAID1
|
2685 BTRFS_BLOCK_GROUP_RAID10
|
2686 BTRFS_BLOCK_GROUP_DUP
);
2688 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2689 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2690 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2691 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2692 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2693 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2697 static void set_block_group_readonly(struct btrfs_block_group_cache
*cache
)
2699 spin_lock(&cache
->space_info
->lock
);
2700 spin_lock(&cache
->lock
);
2702 cache
->space_info
->bytes_readonly
+= cache
->key
.offset
-
2703 btrfs_block_group_used(&cache
->item
);
2706 spin_unlock(&cache
->lock
);
2707 spin_unlock(&cache
->space_info
->lock
);
2710 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2712 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
2714 if (num_devices
== 1)
2715 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2716 if (num_devices
< 4)
2717 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2719 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2720 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2721 BTRFS_BLOCK_GROUP_RAID10
))) {
2722 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
2725 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
2726 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
2727 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
2730 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
2731 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
2732 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
2733 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
2734 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
2738 static u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, u64 data
)
2740 struct btrfs_fs_info
*info
= root
->fs_info
;
2744 alloc_profile
= info
->avail_data_alloc_bits
&
2745 info
->data_alloc_profile
;
2746 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
2747 } else if (root
== root
->fs_info
->chunk_root
) {
2748 alloc_profile
= info
->avail_system_alloc_bits
&
2749 info
->system_alloc_profile
;
2750 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
2752 alloc_profile
= info
->avail_metadata_alloc_bits
&
2753 info
->metadata_alloc_profile
;
2754 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
2757 return btrfs_reduce_alloc_profile(root
, data
);
2760 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
2764 alloc_target
= btrfs_get_alloc_profile(root
, 1);
2765 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
2769 static u64
calculate_bytes_needed(struct btrfs_root
*root
, int num_items
)
2774 level
= BTRFS_MAX_LEVEL
- 2;
2776 * NOTE: these calculations are absolutely the worst possible case.
2777 * This assumes that _every_ item we insert will require a new leaf, and
2778 * that the tree has grown to its maximum level size.
2782 * for every item we insert we could insert both an extent item and a
2783 * extent ref item. Then for ever item we insert, we will need to cow
2784 * both the original leaf, plus the leaf to the left and right of it.
2786 * Unless we are talking about the extent root, then we just want the
2787 * number of items * 2, since we just need the extent item plus its ref.
2789 if (root
== root
->fs_info
->extent_root
)
2790 num_bytes
= num_items
* 2;
2792 num_bytes
= (num_items
+ (2 * num_items
)) * 3;
2795 * num_bytes is total number of leaves we could need times the leaf
2796 * size, and then for every leaf we could end up cow'ing 2 nodes per
2797 * level, down to the leaf level.
2799 num_bytes
= (num_bytes
* root
->leafsize
) +
2800 (num_bytes
* (level
* 2)) * root
->nodesize
;
2806 * Unreserve metadata space for delalloc. If we have less reserved credits than
2807 * we have extents, this function does nothing.
2809 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root
*root
,
2810 struct inode
*inode
, int num_items
)
2812 struct btrfs_fs_info
*info
= root
->fs_info
;
2813 struct btrfs_space_info
*meta_sinfo
;
2818 /* get the space info for where the metadata will live */
2819 alloc_target
= btrfs_get_alloc_profile(root
, 0);
2820 meta_sinfo
= __find_space_info(info
, alloc_target
);
2822 num_bytes
= calculate_bytes_needed(root
->fs_info
->extent_root
,
2825 spin_lock(&meta_sinfo
->lock
);
2826 if (BTRFS_I(inode
)->delalloc_reserved_extents
<=
2827 BTRFS_I(inode
)->delalloc_extents
) {
2828 spin_unlock(&meta_sinfo
->lock
);
2832 BTRFS_I(inode
)->delalloc_reserved_extents
--;
2833 BUG_ON(BTRFS_I(inode
)->delalloc_reserved_extents
< 0);
2835 if (meta_sinfo
->bytes_delalloc
< num_bytes
) {
2837 meta_sinfo
->bytes_delalloc
= 0;
2839 meta_sinfo
->bytes_delalloc
-= num_bytes
;
2841 spin_unlock(&meta_sinfo
->lock
);
2848 static void check_force_delalloc(struct btrfs_space_info
*meta_sinfo
)
2852 thresh
= meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
2853 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
+
2854 meta_sinfo
->bytes_super
+ meta_sinfo
->bytes_root
+
2855 meta_sinfo
->bytes_may_use
;
2857 thresh
= meta_sinfo
->total_bytes
- thresh
;
2859 do_div(thresh
, 100);
2860 if (thresh
<= meta_sinfo
->bytes_delalloc
)
2861 meta_sinfo
->force_delalloc
= 1;
2863 meta_sinfo
->force_delalloc
= 0;
2866 static int maybe_allocate_chunk(struct btrfs_root
*root
,
2867 struct btrfs_space_info
*info
)
2869 struct btrfs_super_block
*disk_super
= &root
->fs_info
->super_copy
;
2870 struct btrfs_trans_handle
*trans
;
2876 free_space
= btrfs_super_total_bytes(disk_super
);
2878 * we allow the metadata to grow to a max of either 5gb or 5% of the
2879 * space in the volume.
2881 min_metadata
= min((u64
)5 * 1024 * 1024 * 1024,
2882 div64_u64(free_space
* 5, 100));
2883 if (info
->total_bytes
>= min_metadata
) {
2884 spin_unlock(&info
->lock
);
2889 spin_unlock(&info
->lock
);
2893 if (!info
->allocating_chunk
) {
2894 info
->force_alloc
= 1;
2895 info
->allocating_chunk
= 1;
2896 init_waitqueue_head(&info
->wait
);
2901 spin_unlock(&info
->lock
);
2904 wait_event(info
->wait
,
2905 !info
->allocating_chunk
);
2909 trans
= btrfs_start_transaction(root
, 1);
2915 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2916 4096 + 2 * 1024 * 1024,
2918 btrfs_end_transaction(trans
, root
);
2922 spin_lock(&info
->lock
);
2923 info
->allocating_chunk
= 0;
2924 spin_unlock(&info
->lock
);
2925 wake_up(&info
->wait
);
2933 * Reserve metadata space for delalloc.
2935 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root
*root
,
2936 struct inode
*inode
, int num_items
)
2938 struct btrfs_fs_info
*info
= root
->fs_info
;
2939 struct btrfs_space_info
*meta_sinfo
;
2946 /* get the space info for where the metadata will live */
2947 alloc_target
= btrfs_get_alloc_profile(root
, 0);
2948 meta_sinfo
= __find_space_info(info
, alloc_target
);
2950 num_bytes
= calculate_bytes_needed(root
->fs_info
->extent_root
,
2953 spin_lock(&meta_sinfo
->lock
);
2955 force_delalloc
= meta_sinfo
->force_delalloc
;
2957 if (unlikely(!meta_sinfo
->bytes_root
))
2958 meta_sinfo
->bytes_root
= calculate_bytes_needed(root
, 6);
2961 meta_sinfo
->bytes_delalloc
+= num_bytes
;
2963 used
= meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
2964 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
+
2965 meta_sinfo
->bytes_super
+ meta_sinfo
->bytes_root
+
2966 meta_sinfo
->bytes_may_use
+ meta_sinfo
->bytes_delalloc
;
2968 if (used
> meta_sinfo
->total_bytes
) {
2972 if (maybe_allocate_chunk(root
, meta_sinfo
))
2976 spin_unlock(&meta_sinfo
->lock
);
2980 filemap_flush(inode
->i_mapping
);
2982 } else if (flushed
== 3) {
2983 btrfs_start_delalloc_inodes(root
);
2984 btrfs_wait_ordered_extents(root
, 0);
2987 spin_lock(&meta_sinfo
->lock
);
2988 meta_sinfo
->bytes_delalloc
-= num_bytes
;
2989 spin_unlock(&meta_sinfo
->lock
);
2990 printk(KERN_ERR
"enospc, has %d, reserved %d\n",
2991 BTRFS_I(inode
)->delalloc_extents
,
2992 BTRFS_I(inode
)->delalloc_reserved_extents
);
2993 dump_space_info(meta_sinfo
, 0, 0);
2997 BTRFS_I(inode
)->delalloc_reserved_extents
++;
2998 check_force_delalloc(meta_sinfo
);
2999 spin_unlock(&meta_sinfo
->lock
);
3001 if (!flushed
&& force_delalloc
)
3002 filemap_flush(inode
->i_mapping
);
3008 * unreserve num_items number of items worth of metadata space. This needs to
3009 * be paired with btrfs_reserve_metadata_space.
3011 * NOTE: if you have the option, run this _AFTER_ you do a
3012 * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3013 * oprations which will result in more used metadata, so we want to make sure we
3014 * can do that without issue.
3016 int btrfs_unreserve_metadata_space(struct btrfs_root
*root
, int num_items
)
3018 struct btrfs_fs_info
*info
= root
->fs_info
;
3019 struct btrfs_space_info
*meta_sinfo
;
3024 /* get the space info for where the metadata will live */
3025 alloc_target
= btrfs_get_alloc_profile(root
, 0);
3026 meta_sinfo
= __find_space_info(info
, alloc_target
);
3028 num_bytes
= calculate_bytes_needed(root
, num_items
);
3030 spin_lock(&meta_sinfo
->lock
);
3031 if (meta_sinfo
->bytes_may_use
< num_bytes
) {
3033 meta_sinfo
->bytes_may_use
= 0;
3035 meta_sinfo
->bytes_may_use
-= num_bytes
;
3037 spin_unlock(&meta_sinfo
->lock
);
3045 * Reserve some metadata space for use. We'll calculate the worste case number
3046 * of bytes that would be needed to modify num_items number of items. If we
3047 * have space, fantastic, if not, you get -ENOSPC. Please call
3048 * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3049 * items you reserved, since whatever metadata you needed should have already
3052 * This will commit the transaction to make more space if we don't have enough
3053 * metadata space. THe only time we don't do this is if we're reserving space
3054 * inside of a transaction, then we will just return -ENOSPC and it is the
3055 * callers responsibility to handle it properly.
3057 int btrfs_reserve_metadata_space(struct btrfs_root
*root
, int num_items
)
3059 struct btrfs_fs_info
*info
= root
->fs_info
;
3060 struct btrfs_space_info
*meta_sinfo
;
3066 /* get the space info for where the metadata will live */
3067 alloc_target
= btrfs_get_alloc_profile(root
, 0);
3068 meta_sinfo
= __find_space_info(info
, alloc_target
);
3070 num_bytes
= calculate_bytes_needed(root
, num_items
);
3072 spin_lock(&meta_sinfo
->lock
);
3074 if (unlikely(!meta_sinfo
->bytes_root
))
3075 meta_sinfo
->bytes_root
= calculate_bytes_needed(root
, 6);
3078 meta_sinfo
->bytes_may_use
+= num_bytes
;
3080 used
= meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
3081 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
+
3082 meta_sinfo
->bytes_super
+ meta_sinfo
->bytes_root
+
3083 meta_sinfo
->bytes_may_use
+ meta_sinfo
->bytes_delalloc
;
3085 if (used
> meta_sinfo
->total_bytes
) {
3088 if (maybe_allocate_chunk(root
, meta_sinfo
))
3092 spin_unlock(&meta_sinfo
->lock
);
3096 btrfs_start_delalloc_inodes(root
);
3097 btrfs_wait_ordered_extents(root
, 0);
3100 spin_lock(&meta_sinfo
->lock
);
3101 meta_sinfo
->bytes_may_use
-= num_bytes
;
3102 spin_unlock(&meta_sinfo
->lock
);
3104 dump_space_info(meta_sinfo
, 0, 0);
3108 check_force_delalloc(meta_sinfo
);
3109 spin_unlock(&meta_sinfo
->lock
);
3115 * This will check the space that the inode allocates from to make sure we have
3116 * enough space for bytes.
3118 int btrfs_check_data_free_space(struct btrfs_root
*root
, struct inode
*inode
,
3121 struct btrfs_space_info
*data_sinfo
;
3122 int ret
= 0, committed
= 0;
3124 /* make sure bytes are sectorsize aligned */
3125 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3127 data_sinfo
= BTRFS_I(inode
)->space_info
;
3132 /* make sure we have enough space to handle the data first */
3133 spin_lock(&data_sinfo
->lock
);
3134 if (data_sinfo
->total_bytes
- data_sinfo
->bytes_used
-
3135 data_sinfo
->bytes_delalloc
- data_sinfo
->bytes_reserved
-
3136 data_sinfo
->bytes_pinned
- data_sinfo
->bytes_readonly
-
3137 data_sinfo
->bytes_may_use
- data_sinfo
->bytes_super
< bytes
) {
3138 struct btrfs_trans_handle
*trans
;
3141 * if we don't have enough free bytes in this space then we need
3142 * to alloc a new chunk.
3144 if (!data_sinfo
->full
) {
3147 data_sinfo
->force_alloc
= 1;
3148 spin_unlock(&data_sinfo
->lock
);
3150 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3151 trans
= btrfs_start_transaction(root
, 1);
3155 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3156 bytes
+ 2 * 1024 * 1024,
3158 btrfs_end_transaction(trans
, root
);
3163 btrfs_set_inode_space_info(root
, inode
);
3164 data_sinfo
= BTRFS_I(inode
)->space_info
;
3168 spin_unlock(&data_sinfo
->lock
);
3170 /* commit the current transaction and try again */
3173 trans
= btrfs_join_transaction(root
, 1);
3176 ret
= btrfs_commit_transaction(trans
, root
);
3182 printk(KERN_ERR
"no space left, need %llu, %llu delalloc bytes"
3183 ", %llu bytes_used, %llu bytes_reserved, "
3184 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3185 "%llu total\n", (unsigned long long)bytes
,
3186 (unsigned long long)data_sinfo
->bytes_delalloc
,
3187 (unsigned long long)data_sinfo
->bytes_used
,
3188 (unsigned long long)data_sinfo
->bytes_reserved
,
3189 (unsigned long long)data_sinfo
->bytes_pinned
,
3190 (unsigned long long)data_sinfo
->bytes_readonly
,
3191 (unsigned long long)data_sinfo
->bytes_may_use
,
3192 (unsigned long long)data_sinfo
->total_bytes
);
3195 data_sinfo
->bytes_may_use
+= bytes
;
3196 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3197 spin_unlock(&data_sinfo
->lock
);
3203 * if there was an error for whatever reason after calling
3204 * btrfs_check_data_free_space, call this so we can cleanup the counters.
3206 void btrfs_free_reserved_data_space(struct btrfs_root
*root
,
3207 struct inode
*inode
, u64 bytes
)
3209 struct btrfs_space_info
*data_sinfo
;
3211 /* make sure bytes are sectorsize aligned */
3212 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3214 data_sinfo
= BTRFS_I(inode
)->space_info
;
3215 spin_lock(&data_sinfo
->lock
);
3216 data_sinfo
->bytes_may_use
-= bytes
;
3217 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3218 spin_unlock(&data_sinfo
->lock
);
3221 /* called when we are adding a delalloc extent to the inode's io_tree */
3222 void btrfs_delalloc_reserve_space(struct btrfs_root
*root
, struct inode
*inode
,
3225 struct btrfs_space_info
*data_sinfo
;
3227 /* get the space info for where this inode will be storing its data */
3228 data_sinfo
= BTRFS_I(inode
)->space_info
;
3230 /* make sure we have enough space to handle the data first */
3231 spin_lock(&data_sinfo
->lock
);
3232 data_sinfo
->bytes_delalloc
+= bytes
;
3235 * we are adding a delalloc extent without calling
3236 * btrfs_check_data_free_space first. This happens on a weird
3237 * writepage condition, but shouldn't hurt our accounting
3239 if (unlikely(bytes
> BTRFS_I(inode
)->reserved_bytes
)) {
3240 data_sinfo
->bytes_may_use
-= BTRFS_I(inode
)->reserved_bytes
;
3241 BTRFS_I(inode
)->reserved_bytes
= 0;
3243 data_sinfo
->bytes_may_use
-= bytes
;
3244 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3247 spin_unlock(&data_sinfo
->lock
);
3250 /* called when we are clearing an delalloc extent from the inode's io_tree */
3251 void btrfs_delalloc_free_space(struct btrfs_root
*root
, struct inode
*inode
,
3254 struct btrfs_space_info
*info
;
3256 info
= BTRFS_I(inode
)->space_info
;
3258 spin_lock(&info
->lock
);
3259 info
->bytes_delalloc
-= bytes
;
3260 spin_unlock(&info
->lock
);
3263 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3265 struct list_head
*head
= &info
->space_info
;
3266 struct btrfs_space_info
*found
;
3269 list_for_each_entry_rcu(found
, head
, list
) {
3270 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3271 found
->force_alloc
= 1;
3276 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3277 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3278 u64 flags
, int force
)
3280 struct btrfs_space_info
*space_info
;
3281 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3285 mutex_lock(&fs_info
->chunk_mutex
);
3287 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3289 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3291 ret
= update_space_info(extent_root
->fs_info
, flags
,
3295 BUG_ON(!space_info
);
3297 spin_lock(&space_info
->lock
);
3298 if (space_info
->force_alloc
)
3300 if (space_info
->full
) {
3301 spin_unlock(&space_info
->lock
);
3305 thresh
= space_info
->total_bytes
- space_info
->bytes_readonly
;
3306 thresh
= div_factor(thresh
, 8);
3308 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
3309 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
) {
3310 spin_unlock(&space_info
->lock
);
3313 spin_unlock(&space_info
->lock
);
3316 * if we're doing a data chunk, go ahead and make sure that
3317 * we keep a reasonable number of metadata chunks allocated in the
3320 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3321 fs_info
->data_chunk_allocations
++;
3322 if (!(fs_info
->data_chunk_allocations
%
3323 fs_info
->metadata_ratio
))
3324 force_metadata_allocation(fs_info
);
3327 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3328 spin_lock(&space_info
->lock
);
3330 space_info
->full
= 1;
3331 space_info
->force_alloc
= 0;
3332 spin_unlock(&space_info
->lock
);
3334 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3338 static int update_block_group(struct btrfs_trans_handle
*trans
,
3339 struct btrfs_root
*root
,
3340 u64 bytenr
, u64 num_bytes
, int alloc
,
3343 struct btrfs_block_group_cache
*cache
;
3344 struct btrfs_fs_info
*info
= root
->fs_info
;
3345 u64 total
= num_bytes
;
3349 /* block accounting for super block */
3350 spin_lock(&info
->delalloc_lock
);
3351 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
3353 old_val
+= num_bytes
;
3355 old_val
-= num_bytes
;
3356 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
3358 /* block accounting for root item */
3359 old_val
= btrfs_root_used(&root
->root_item
);
3361 old_val
+= num_bytes
;
3363 old_val
-= num_bytes
;
3364 btrfs_set_root_used(&root
->root_item
, old_val
);
3365 spin_unlock(&info
->delalloc_lock
);
3368 cache
= btrfs_lookup_block_group(info
, bytenr
);
3371 byte_in_group
= bytenr
- cache
->key
.objectid
;
3372 WARN_ON(byte_in_group
> cache
->key
.offset
);
3374 spin_lock(&cache
->space_info
->lock
);
3375 spin_lock(&cache
->lock
);
3377 old_val
= btrfs_block_group_used(&cache
->item
);
3378 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
3380 old_val
+= num_bytes
;
3381 btrfs_set_block_group_used(&cache
->item
, old_val
);
3382 cache
->reserved
-= num_bytes
;
3383 cache
->space_info
->bytes_used
+= num_bytes
;
3384 cache
->space_info
->bytes_reserved
-= num_bytes
;
3386 cache
->space_info
->bytes_readonly
-= num_bytes
;
3387 spin_unlock(&cache
->lock
);
3388 spin_unlock(&cache
->space_info
->lock
);
3390 old_val
-= num_bytes
;
3391 cache
->space_info
->bytes_used
-= num_bytes
;
3393 cache
->space_info
->bytes_readonly
+= num_bytes
;
3394 btrfs_set_block_group_used(&cache
->item
, old_val
);
3395 spin_unlock(&cache
->lock
);
3396 spin_unlock(&cache
->space_info
->lock
);
3400 ret
= btrfs_discard_extent(root
, bytenr
,
3404 ret
= btrfs_add_free_space(cache
, bytenr
,
3409 btrfs_put_block_group(cache
);
3411 bytenr
+= num_bytes
;
3416 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
3418 struct btrfs_block_group_cache
*cache
;
3421 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
3425 bytenr
= cache
->key
.objectid
;
3426 btrfs_put_block_group(cache
);
3432 * this function must be called within transaction
3434 int btrfs_pin_extent(struct btrfs_root
*root
,
3435 u64 bytenr
, u64 num_bytes
, int reserved
)
3437 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3438 struct btrfs_block_group_cache
*cache
;
3440 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
3443 spin_lock(&cache
->space_info
->lock
);
3444 spin_lock(&cache
->lock
);
3445 cache
->pinned
+= num_bytes
;
3446 cache
->space_info
->bytes_pinned
+= num_bytes
;
3448 cache
->reserved
-= num_bytes
;
3449 cache
->space_info
->bytes_reserved
-= num_bytes
;
3451 spin_unlock(&cache
->lock
);
3452 spin_unlock(&cache
->space_info
->lock
);
3454 btrfs_put_block_group(cache
);
3456 set_extent_dirty(fs_info
->pinned_extents
,
3457 bytenr
, bytenr
+ num_bytes
- 1, GFP_NOFS
);
3461 static int update_reserved_extents(struct btrfs_block_group_cache
*cache
,
3462 u64 num_bytes
, int reserve
)
3464 spin_lock(&cache
->space_info
->lock
);
3465 spin_lock(&cache
->lock
);
3467 cache
->reserved
+= num_bytes
;
3468 cache
->space_info
->bytes_reserved
+= num_bytes
;
3470 cache
->reserved
-= num_bytes
;
3471 cache
->space_info
->bytes_reserved
-= num_bytes
;
3473 spin_unlock(&cache
->lock
);
3474 spin_unlock(&cache
->space_info
->lock
);
3478 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
3479 struct btrfs_root
*root
)
3481 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3482 struct btrfs_caching_control
*next
;
3483 struct btrfs_caching_control
*caching_ctl
;
3484 struct btrfs_block_group_cache
*cache
;
3486 down_write(&fs_info
->extent_commit_sem
);
3488 list_for_each_entry_safe(caching_ctl
, next
,
3489 &fs_info
->caching_block_groups
, list
) {
3490 cache
= caching_ctl
->block_group
;
3491 if (block_group_cache_done(cache
)) {
3492 cache
->last_byte_to_unpin
= (u64
)-1;
3493 list_del_init(&caching_ctl
->list
);
3494 put_caching_control(caching_ctl
);
3496 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
3500 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
3501 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
3503 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
3505 up_write(&fs_info
->extent_commit_sem
);
3509 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
3511 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3512 struct btrfs_block_group_cache
*cache
= NULL
;
3515 while (start
<= end
) {
3517 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
3519 btrfs_put_block_group(cache
);
3520 cache
= btrfs_lookup_block_group(fs_info
, start
);
3524 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
3525 len
= min(len
, end
+ 1 - start
);
3527 if (start
< cache
->last_byte_to_unpin
) {
3528 len
= min(len
, cache
->last_byte_to_unpin
- start
);
3529 btrfs_add_free_space(cache
, start
, len
);
3532 spin_lock(&cache
->space_info
->lock
);
3533 spin_lock(&cache
->lock
);
3534 cache
->pinned
-= len
;
3535 cache
->space_info
->bytes_pinned
-= len
;
3536 spin_unlock(&cache
->lock
);
3537 spin_unlock(&cache
->space_info
->lock
);
3543 btrfs_put_block_group(cache
);
3547 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
3548 struct btrfs_root
*root
)
3550 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3551 struct extent_io_tree
*unpin
;
3556 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
3557 unpin
= &fs_info
->freed_extents
[1];
3559 unpin
= &fs_info
->freed_extents
[0];
3562 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3567 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
3569 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3570 unpin_extent_range(root
, start
, end
);
3577 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
3578 struct btrfs_root
*root
,
3579 struct btrfs_path
*path
,
3580 u64 bytenr
, u64 num_bytes
,
3581 int is_data
, int reserved
,
3582 struct extent_buffer
**must_clean
)
3585 struct extent_buffer
*buf
;
3590 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
3594 /* we can reuse a block if it hasn't been written
3595 * and it is from this transaction. We can't
3596 * reuse anything from the tree log root because
3597 * it has tiny sub-transactions.
3599 if (btrfs_buffer_uptodate(buf
, 0) &&
3600 btrfs_try_tree_lock(buf
)) {
3601 u64 header_owner
= btrfs_header_owner(buf
);
3602 u64 header_transid
= btrfs_header_generation(buf
);
3603 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
3604 header_transid
== trans
->transid
&&
3605 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
3609 btrfs_tree_unlock(buf
);
3611 free_extent_buffer(buf
);
3614 btrfs_set_path_blocking(path
);
3615 /* unlocks the pinned mutex */
3616 btrfs_pin_extent(root
, bytenr
, num_bytes
, reserved
);
3622 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
3623 struct btrfs_root
*root
,
3624 u64 bytenr
, u64 num_bytes
, u64 parent
,
3625 u64 root_objectid
, u64 owner_objectid
,
3626 u64 owner_offset
, int refs_to_drop
,
3627 struct btrfs_delayed_extent_op
*extent_op
)
3629 struct btrfs_key key
;
3630 struct btrfs_path
*path
;
3631 struct btrfs_fs_info
*info
= root
->fs_info
;
3632 struct btrfs_root
*extent_root
= info
->extent_root
;
3633 struct extent_buffer
*leaf
;
3634 struct btrfs_extent_item
*ei
;
3635 struct btrfs_extent_inline_ref
*iref
;
3638 int extent_slot
= 0;
3639 int found_extent
= 0;
3644 path
= btrfs_alloc_path();
3649 path
->leave_spinning
= 1;
3651 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
3652 BUG_ON(!is_data
&& refs_to_drop
!= 1);
3654 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
3655 bytenr
, num_bytes
, parent
,
3656 root_objectid
, owner_objectid
,
3659 extent_slot
= path
->slots
[0];
3660 while (extent_slot
>= 0) {
3661 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3663 if (key
.objectid
!= bytenr
)
3665 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3666 key
.offset
== num_bytes
) {
3670 if (path
->slots
[0] - extent_slot
> 5)
3674 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3675 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
3676 if (found_extent
&& item_size
< sizeof(*ei
))
3679 if (!found_extent
) {
3681 ret
= remove_extent_backref(trans
, extent_root
, path
,
3685 btrfs_release_path(extent_root
, path
);
3686 path
->leave_spinning
= 1;
3688 key
.objectid
= bytenr
;
3689 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3690 key
.offset
= num_bytes
;
3692 ret
= btrfs_search_slot(trans
, extent_root
,
3695 printk(KERN_ERR
"umm, got %d back from search"
3696 ", was looking for %llu\n", ret
,
3697 (unsigned long long)bytenr
);
3698 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3701 extent_slot
= path
->slots
[0];
3704 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3706 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
3707 "parent %llu root %llu owner %llu offset %llu\n",
3708 (unsigned long long)bytenr
,
3709 (unsigned long long)parent
,
3710 (unsigned long long)root_objectid
,
3711 (unsigned long long)owner_objectid
,
3712 (unsigned long long)owner_offset
);
3715 leaf
= path
->nodes
[0];
3716 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3717 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3718 if (item_size
< sizeof(*ei
)) {
3719 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
3720 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
3724 btrfs_release_path(extent_root
, path
);
3725 path
->leave_spinning
= 1;
3727 key
.objectid
= bytenr
;
3728 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3729 key
.offset
= num_bytes
;
3731 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
3734 printk(KERN_ERR
"umm, got %d back from search"
3735 ", was looking for %llu\n", ret
,
3736 (unsigned long long)bytenr
);
3737 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3740 extent_slot
= path
->slots
[0];
3741 leaf
= path
->nodes
[0];
3742 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3745 BUG_ON(item_size
< sizeof(*ei
));
3746 ei
= btrfs_item_ptr(leaf
, extent_slot
,
3747 struct btrfs_extent_item
);
3748 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
3749 struct btrfs_tree_block_info
*bi
;
3750 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
3751 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
3752 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
3755 refs
= btrfs_extent_refs(leaf
, ei
);
3756 BUG_ON(refs
< refs_to_drop
);
3757 refs
-= refs_to_drop
;
3761 __run_delayed_extent_op(extent_op
, leaf
, ei
);
3763 * In the case of inline back ref, reference count will
3764 * be updated by remove_extent_backref
3767 BUG_ON(!found_extent
);
3769 btrfs_set_extent_refs(leaf
, ei
, refs
);
3770 btrfs_mark_buffer_dirty(leaf
);
3773 ret
= remove_extent_backref(trans
, extent_root
, path
,
3780 struct extent_buffer
*must_clean
= NULL
;
3783 BUG_ON(is_data
&& refs_to_drop
!=
3784 extent_data_ref_count(root
, path
, iref
));
3786 BUG_ON(path
->slots
[0] != extent_slot
);
3788 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
3789 path
->slots
[0] = extent_slot
;
3794 ret
= pin_down_bytes(trans
, root
, path
, bytenr
,
3795 num_bytes
, is_data
, 0, &must_clean
);
3800 * it is going to be very rare for someone to be waiting
3801 * on the block we're freeing. del_items might need to
3802 * schedule, so rather than get fancy, just force it
3806 btrfs_set_lock_blocking(must_clean
);
3808 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
3811 btrfs_release_path(extent_root
, path
);
3814 clean_tree_block(NULL
, root
, must_clean
);
3815 btrfs_tree_unlock(must_clean
);
3816 free_extent_buffer(must_clean
);
3820 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
3823 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
3824 bytenr
>> PAGE_CACHE_SHIFT
,
3825 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
3828 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
3832 btrfs_free_path(path
);
3837 * when we free an extent, it is possible (and likely) that we free the last
3838 * delayed ref for that extent as well. This searches the delayed ref tree for
3839 * a given extent, and if there are no other delayed refs to be processed, it
3840 * removes it from the tree.
3842 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
3843 struct btrfs_root
*root
, u64 bytenr
)
3845 struct btrfs_delayed_ref_head
*head
;
3846 struct btrfs_delayed_ref_root
*delayed_refs
;
3847 struct btrfs_delayed_ref_node
*ref
;
3848 struct rb_node
*node
;
3851 delayed_refs
= &trans
->transaction
->delayed_refs
;
3852 spin_lock(&delayed_refs
->lock
);
3853 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3857 node
= rb_prev(&head
->node
.rb_node
);
3861 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3863 /* there are still entries for this ref, we can't drop it */
3864 if (ref
->bytenr
== bytenr
)
3867 if (head
->extent_op
) {
3868 if (!head
->must_insert_reserved
)
3870 kfree(head
->extent_op
);
3871 head
->extent_op
= NULL
;
3875 * waiting for the lock here would deadlock. If someone else has it
3876 * locked they are already in the process of dropping it anyway
3878 if (!mutex_trylock(&head
->mutex
))
3882 * at this point we have a head with no other entries. Go
3883 * ahead and process it.
3885 head
->node
.in_tree
= 0;
3886 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
3888 delayed_refs
->num_entries
--;
3891 * we don't take a ref on the node because we're removing it from the
3892 * tree, so we just steal the ref the tree was holding.
3894 delayed_refs
->num_heads
--;
3895 if (list_empty(&head
->cluster
))
3896 delayed_refs
->num_heads_ready
--;
3898 list_del_init(&head
->cluster
);
3899 spin_unlock(&delayed_refs
->lock
);
3901 ret
= run_one_delayed_ref(trans
, root
->fs_info
->tree_root
,
3902 &head
->node
, head
->extent_op
,
3903 head
->must_insert_reserved
);
3905 btrfs_put_delayed_ref(&head
->node
);
3908 spin_unlock(&delayed_refs
->lock
);
3912 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
3913 struct btrfs_root
*root
,
3914 u64 bytenr
, u64 num_bytes
, u64 parent
,
3915 u64 root_objectid
, u64 owner
, u64 offset
)
3920 * tree log blocks never actually go into the extent allocation
3921 * tree, just update pinning info and exit early.
3923 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
3924 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
3925 /* unlocks the pinned mutex */
3926 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
3928 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
3929 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
3930 parent
, root_objectid
, (int)owner
,
3931 BTRFS_DROP_DELAYED_REF
, NULL
);
3933 ret
= check_ref_cleanup(trans
, root
, bytenr
);
3936 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
3937 parent
, root_objectid
, owner
,
3938 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
3944 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
3946 u64 mask
= ((u64
)root
->stripesize
- 1);
3947 u64 ret
= (val
+ mask
) & ~mask
;
3952 * when we wait for progress in the block group caching, its because
3953 * our allocation attempt failed at least once. So, we must sleep
3954 * and let some progress happen before we try again.
3956 * This function will sleep at least once waiting for new free space to
3957 * show up, and then it will check the block group free space numbers
3958 * for our min num_bytes. Another option is to have it go ahead
3959 * and look in the rbtree for a free extent of a given size, but this
3963 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
3966 struct btrfs_caching_control
*caching_ctl
;
3969 caching_ctl
= get_caching_control(cache
);
3973 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
3974 (cache
->free_space
>= num_bytes
));
3976 put_caching_control(caching_ctl
);
3981 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
3983 struct btrfs_caching_control
*caching_ctl
;
3986 caching_ctl
= get_caching_control(cache
);
3990 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
3992 put_caching_control(caching_ctl
);
3996 enum btrfs_loop_type
{
3997 LOOP_CACHED_ONLY
= 0,
3998 LOOP_CACHING_NOWAIT
= 1,
3999 LOOP_CACHING_WAIT
= 2,
4000 LOOP_ALLOC_CHUNK
= 3,
4001 LOOP_NO_EMPTY_SIZE
= 4,
4005 * walks the btree of allocated extents and find a hole of a given size.
4006 * The key ins is changed to record the hole:
4007 * ins->objectid == block start
4008 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4009 * ins->offset == number of blocks
4010 * Any available blocks before search_start are skipped.
4012 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4013 struct btrfs_root
*orig_root
,
4014 u64 num_bytes
, u64 empty_size
,
4015 u64 search_start
, u64 search_end
,
4016 u64 hint_byte
, struct btrfs_key
*ins
,
4017 u64 exclude_start
, u64 exclude_nr
,
4021 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4022 struct btrfs_free_cluster
*last_ptr
= NULL
;
4023 struct btrfs_block_group_cache
*block_group
= NULL
;
4024 int empty_cluster
= 2 * 1024 * 1024;
4025 int allowed_chunk_alloc
= 0;
4026 struct btrfs_space_info
*space_info
;
4027 int last_ptr_loop
= 0;
4029 bool found_uncached_bg
= false;
4030 bool failed_cluster_refill
= false;
4032 WARN_ON(num_bytes
< root
->sectorsize
);
4033 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4037 space_info
= __find_space_info(root
->fs_info
, data
);
4039 if (orig_root
->ref_cows
|| empty_size
)
4040 allowed_chunk_alloc
= 1;
4042 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
4043 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4044 if (!btrfs_test_opt(root
, SSD
))
4045 empty_cluster
= 64 * 1024;
4048 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
4049 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4053 spin_lock(&last_ptr
->lock
);
4054 if (last_ptr
->block_group
)
4055 hint_byte
= last_ptr
->window_start
;
4056 spin_unlock(&last_ptr
->lock
);
4059 search_start
= max(search_start
, first_logical_byte(root
, 0));
4060 search_start
= max(search_start
, hint_byte
);
4065 if (search_start
== hint_byte
) {
4066 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4069 * we don't want to use the block group if it doesn't match our
4070 * allocation bits, or if its not cached.
4072 if (block_group
&& block_group_bits(block_group
, data
) &&
4073 block_group_cache_done(block_group
)) {
4074 down_read(&space_info
->groups_sem
);
4075 if (list_empty(&block_group
->list
) ||
4078 * someone is removing this block group,
4079 * we can't jump into the have_block_group
4080 * target because our list pointers are not
4083 btrfs_put_block_group(block_group
);
4084 up_read(&space_info
->groups_sem
);
4086 goto have_block_group
;
4087 } else if (block_group
) {
4088 btrfs_put_block_group(block_group
);
4093 down_read(&space_info
->groups_sem
);
4094 list_for_each_entry(block_group
, &space_info
->block_groups
, list
) {
4098 atomic_inc(&block_group
->count
);
4099 search_start
= block_group
->key
.objectid
;
4102 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
4104 * we want to start caching kthreads, but not too many
4105 * right off the bat so we don't overwhelm the system,
4106 * so only start them if there are less than 2 and we're
4107 * in the initial allocation phase.
4109 if (loop
> LOOP_CACHING_NOWAIT
||
4110 atomic_read(&space_info
->caching_threads
) < 2) {
4111 ret
= cache_block_group(block_group
);
4116 cached
= block_group_cache_done(block_group
);
4117 if (unlikely(!cached
)) {
4118 found_uncached_bg
= true;
4120 /* if we only want cached bgs, loop */
4121 if (loop
== LOOP_CACHED_ONLY
)
4125 if (unlikely(block_group
->ro
))
4129 * Ok we want to try and use the cluster allocator, so lets look
4130 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4131 * have tried the cluster allocator plenty of times at this
4132 * point and not have found anything, so we are likely way too
4133 * fragmented for the clustering stuff to find anything, so lets
4134 * just skip it and let the allocator find whatever block it can
4137 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
4139 * the refill lock keeps out other
4140 * people trying to start a new cluster
4142 spin_lock(&last_ptr
->refill_lock
);
4143 if (last_ptr
->block_group
&&
4144 (last_ptr
->block_group
->ro
||
4145 !block_group_bits(last_ptr
->block_group
, data
))) {
4147 goto refill_cluster
;
4150 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
4151 num_bytes
, search_start
);
4153 /* we have a block, we're done */
4154 spin_unlock(&last_ptr
->refill_lock
);
4158 spin_lock(&last_ptr
->lock
);
4160 * whoops, this cluster doesn't actually point to
4161 * this block group. Get a ref on the block
4162 * group is does point to and try again
4164 if (!last_ptr_loop
&& last_ptr
->block_group
&&
4165 last_ptr
->block_group
!= block_group
) {
4167 btrfs_put_block_group(block_group
);
4168 block_group
= last_ptr
->block_group
;
4169 atomic_inc(&block_group
->count
);
4170 spin_unlock(&last_ptr
->lock
);
4171 spin_unlock(&last_ptr
->refill_lock
);
4174 search_start
= block_group
->key
.objectid
;
4176 * we know this block group is properly
4177 * in the list because
4178 * btrfs_remove_block_group, drops the
4179 * cluster before it removes the block
4180 * group from the list
4182 goto have_block_group
;
4184 spin_unlock(&last_ptr
->lock
);
4187 * this cluster didn't work out, free it and
4190 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
4194 /* allocate a cluster in this block group */
4195 ret
= btrfs_find_space_cluster(trans
, root
,
4196 block_group
, last_ptr
,
4198 empty_cluster
+ empty_size
);
4201 * now pull our allocation out of this
4204 offset
= btrfs_alloc_from_cluster(block_group
,
4205 last_ptr
, num_bytes
,
4208 /* we found one, proceed */
4209 spin_unlock(&last_ptr
->refill_lock
);
4212 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
4213 && !failed_cluster_refill
) {
4214 spin_unlock(&last_ptr
->refill_lock
);
4216 failed_cluster_refill
= true;
4217 wait_block_group_cache_progress(block_group
,
4218 num_bytes
+ empty_cluster
+ empty_size
);
4219 goto have_block_group
;
4223 * at this point we either didn't find a cluster
4224 * or we weren't able to allocate a block from our
4225 * cluster. Free the cluster we've been trying
4226 * to use, and go to the next block group
4228 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
4229 spin_unlock(&last_ptr
->refill_lock
);
4233 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
4234 num_bytes
, empty_size
);
4235 if (!offset
&& (cached
|| (!cached
&&
4236 loop
== LOOP_CACHING_NOWAIT
))) {
4238 } else if (!offset
&& (!cached
&&
4239 loop
> LOOP_CACHING_NOWAIT
)) {
4240 wait_block_group_cache_progress(block_group
,
4241 num_bytes
+ empty_size
);
4242 goto have_block_group
;
4245 search_start
= stripe_align(root
, offset
);
4246 /* move on to the next group */
4247 if (search_start
+ num_bytes
>= search_end
) {
4248 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4252 /* move on to the next group */
4253 if (search_start
+ num_bytes
>
4254 block_group
->key
.objectid
+ block_group
->key
.offset
) {
4255 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4259 if (exclude_nr
> 0 &&
4260 (search_start
+ num_bytes
> exclude_start
&&
4261 search_start
< exclude_start
+ exclude_nr
)) {
4262 search_start
= exclude_start
+ exclude_nr
;
4264 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4266 * if search_start is still in this block group
4267 * then we just re-search this block group
4269 if (search_start
>= block_group
->key
.objectid
&&
4270 search_start
< (block_group
->key
.objectid
+
4271 block_group
->key
.offset
))
4272 goto have_block_group
;
4276 ins
->objectid
= search_start
;
4277 ins
->offset
= num_bytes
;
4279 if (offset
< search_start
)
4280 btrfs_add_free_space(block_group
, offset
,
4281 search_start
- offset
);
4282 BUG_ON(offset
> search_start
);
4284 update_reserved_extents(block_group
, num_bytes
, 1);
4286 /* we are all good, lets return */
4289 failed_cluster_refill
= false;
4290 btrfs_put_block_group(block_group
);
4292 up_read(&space_info
->groups_sem
);
4294 /* LOOP_CACHED_ONLY, only search fully cached block groups
4295 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4296 * dont wait foR them to finish caching
4297 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4298 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4299 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4302 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
4303 (found_uncached_bg
|| empty_size
|| empty_cluster
||
4304 allowed_chunk_alloc
)) {
4305 if (found_uncached_bg
) {
4306 found_uncached_bg
= false;
4307 if (loop
< LOOP_CACHING_WAIT
) {
4313 if (loop
== LOOP_ALLOC_CHUNK
) {
4318 if (allowed_chunk_alloc
) {
4319 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
4320 2 * 1024 * 1024, data
, 1);
4321 allowed_chunk_alloc
= 0;
4323 space_info
->force_alloc
= 1;
4326 if (loop
< LOOP_NO_EMPTY_SIZE
) {
4331 } else if (!ins
->objectid
) {
4335 /* we found what we needed */
4336 if (ins
->objectid
) {
4337 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
4338 trans
->block_group
= block_group
->key
.objectid
;
4340 btrfs_put_block_group(block_group
);
4347 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
4348 int dump_block_groups
)
4350 struct btrfs_block_group_cache
*cache
;
4352 spin_lock(&info
->lock
);
4353 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
4354 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
4355 info
->bytes_pinned
- info
->bytes_reserved
-
4357 (info
->full
) ? "" : "not ");
4358 printk(KERN_INFO
"space_info total=%llu, pinned=%llu, delalloc=%llu,"
4359 " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4361 (unsigned long long)info
->total_bytes
,
4362 (unsigned long long)info
->bytes_pinned
,
4363 (unsigned long long)info
->bytes_delalloc
,
4364 (unsigned long long)info
->bytes_may_use
,
4365 (unsigned long long)info
->bytes_used
,
4366 (unsigned long long)info
->bytes_root
,
4367 (unsigned long long)info
->bytes_super
,
4368 (unsigned long long)info
->bytes_reserved
);
4369 spin_unlock(&info
->lock
);
4371 if (!dump_block_groups
)
4374 down_read(&info
->groups_sem
);
4375 list_for_each_entry(cache
, &info
->block_groups
, list
) {
4376 spin_lock(&cache
->lock
);
4377 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
4378 "%llu pinned %llu reserved\n",
4379 (unsigned long long)cache
->key
.objectid
,
4380 (unsigned long long)cache
->key
.offset
,
4381 (unsigned long long)btrfs_block_group_used(&cache
->item
),
4382 (unsigned long long)cache
->pinned
,
4383 (unsigned long long)cache
->reserved
);
4384 btrfs_dump_free_space(cache
, bytes
);
4385 spin_unlock(&cache
->lock
);
4387 up_read(&info
->groups_sem
);
4390 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
4391 struct btrfs_root
*root
,
4392 u64 num_bytes
, u64 min_alloc_size
,
4393 u64 empty_size
, u64 hint_byte
,
4394 u64 search_end
, struct btrfs_key
*ins
,
4398 u64 search_start
= 0;
4399 struct btrfs_fs_info
*info
= root
->fs_info
;
4401 data
= btrfs_get_alloc_profile(root
, data
);
4404 * the only place that sets empty_size is btrfs_realloc_node, which
4405 * is not called recursively on allocations
4407 if (empty_size
|| root
->ref_cows
) {
4408 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
4409 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4411 BTRFS_BLOCK_GROUP_METADATA
|
4412 (info
->metadata_alloc_profile
&
4413 info
->avail_metadata_alloc_bits
), 0);
4415 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4416 num_bytes
+ 2 * 1024 * 1024, data
, 0);
4419 WARN_ON(num_bytes
< root
->sectorsize
);
4420 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
4421 search_start
, search_end
, hint_byte
, ins
,
4422 trans
->alloc_exclude_start
,
4423 trans
->alloc_exclude_nr
, data
);
4425 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
4426 num_bytes
= num_bytes
>> 1;
4427 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
4428 num_bytes
= max(num_bytes
, min_alloc_size
);
4429 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4430 num_bytes
, data
, 1);
4433 if (ret
== -ENOSPC
) {
4434 struct btrfs_space_info
*sinfo
;
4436 sinfo
= __find_space_info(root
->fs_info
, data
);
4437 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
4438 "wanted %llu\n", (unsigned long long)data
,
4439 (unsigned long long)num_bytes
);
4440 dump_space_info(sinfo
, num_bytes
, 1);
4446 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
4448 struct btrfs_block_group_cache
*cache
;
4451 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
4453 printk(KERN_ERR
"Unable to find block group for %llu\n",
4454 (unsigned long long)start
);
4458 ret
= btrfs_discard_extent(root
, start
, len
);
4460 btrfs_add_free_space(cache
, start
, len
);
4461 update_reserved_extents(cache
, len
, 0);
4462 btrfs_put_block_group(cache
);
4467 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4468 struct btrfs_root
*root
,
4469 u64 parent
, u64 root_objectid
,
4470 u64 flags
, u64 owner
, u64 offset
,
4471 struct btrfs_key
*ins
, int ref_mod
)
4474 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4475 struct btrfs_extent_item
*extent_item
;
4476 struct btrfs_extent_inline_ref
*iref
;
4477 struct btrfs_path
*path
;
4478 struct extent_buffer
*leaf
;
4483 type
= BTRFS_SHARED_DATA_REF_KEY
;
4485 type
= BTRFS_EXTENT_DATA_REF_KEY
;
4487 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
4489 path
= btrfs_alloc_path();
4492 path
->leave_spinning
= 1;
4493 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
4497 leaf
= path
->nodes
[0];
4498 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4499 struct btrfs_extent_item
);
4500 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
4501 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4502 btrfs_set_extent_flags(leaf
, extent_item
,
4503 flags
| BTRFS_EXTENT_FLAG_DATA
);
4505 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
4506 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
4508 struct btrfs_shared_data_ref
*ref
;
4509 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
4510 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4511 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
4513 struct btrfs_extent_data_ref
*ref
;
4514 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
4515 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
4516 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
4517 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
4518 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
4521 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4522 btrfs_free_path(path
);
4524 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
4527 printk(KERN_ERR
"btrfs update block group failed for %llu "
4528 "%llu\n", (unsigned long long)ins
->objectid
,
4529 (unsigned long long)ins
->offset
);
4535 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
4536 struct btrfs_root
*root
,
4537 u64 parent
, u64 root_objectid
,
4538 u64 flags
, struct btrfs_disk_key
*key
,
4539 int level
, struct btrfs_key
*ins
)
4542 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4543 struct btrfs_extent_item
*extent_item
;
4544 struct btrfs_tree_block_info
*block_info
;
4545 struct btrfs_extent_inline_ref
*iref
;
4546 struct btrfs_path
*path
;
4547 struct extent_buffer
*leaf
;
4548 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
4550 path
= btrfs_alloc_path();
4553 path
->leave_spinning
= 1;
4554 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
4558 leaf
= path
->nodes
[0];
4559 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4560 struct btrfs_extent_item
);
4561 btrfs_set_extent_refs(leaf
, extent_item
, 1);
4562 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4563 btrfs_set_extent_flags(leaf
, extent_item
,
4564 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
4565 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
4567 btrfs_set_tree_block_key(leaf
, block_info
, key
);
4568 btrfs_set_tree_block_level(leaf
, block_info
, level
);
4570 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
4572 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
4573 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4574 BTRFS_SHARED_BLOCK_REF_KEY
);
4575 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4577 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4578 BTRFS_TREE_BLOCK_REF_KEY
);
4579 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
4582 btrfs_mark_buffer_dirty(leaf
);
4583 btrfs_free_path(path
);
4585 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
4588 printk(KERN_ERR
"btrfs update block group failed for %llu "
4589 "%llu\n", (unsigned long long)ins
->objectid
,
4590 (unsigned long long)ins
->offset
);
4596 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4597 struct btrfs_root
*root
,
4598 u64 root_objectid
, u64 owner
,
4599 u64 offset
, struct btrfs_key
*ins
)
4603 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
4605 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
4606 0, root_objectid
, owner
, offset
,
4607 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
4612 * this is used by the tree logging recovery code. It records that
4613 * an extent has been allocated and makes sure to clear the free
4614 * space cache bits as well
4616 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
4617 struct btrfs_root
*root
,
4618 u64 root_objectid
, u64 owner
, u64 offset
,
4619 struct btrfs_key
*ins
)
4622 struct btrfs_block_group_cache
*block_group
;
4623 struct btrfs_caching_control
*caching_ctl
;
4624 u64 start
= ins
->objectid
;
4625 u64 num_bytes
= ins
->offset
;
4627 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
4628 cache_block_group(block_group
);
4629 caching_ctl
= get_caching_control(block_group
);
4632 BUG_ON(!block_group_cache_done(block_group
));
4633 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
4636 mutex_lock(&caching_ctl
->mutex
);
4638 if (start
>= caching_ctl
->progress
) {
4639 ret
= add_excluded_extent(root
, start
, num_bytes
);
4641 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
4642 ret
= btrfs_remove_free_space(block_group
,
4646 num_bytes
= caching_ctl
->progress
- start
;
4647 ret
= btrfs_remove_free_space(block_group
,
4651 start
= caching_ctl
->progress
;
4652 num_bytes
= ins
->objectid
+ ins
->offset
-
4653 caching_ctl
->progress
;
4654 ret
= add_excluded_extent(root
, start
, num_bytes
);
4658 mutex_unlock(&caching_ctl
->mutex
);
4659 put_caching_control(caching_ctl
);
4662 update_reserved_extents(block_group
, ins
->offset
, 1);
4663 btrfs_put_block_group(block_group
);
4664 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
4665 0, owner
, offset
, ins
, 1);
4670 * finds a free extent and does all the dirty work required for allocation
4671 * returns the key for the extent through ins, and a tree buffer for
4672 * the first block of the extent through buf.
4674 * returns 0 if everything worked, non-zero otherwise.
4676 static int alloc_tree_block(struct btrfs_trans_handle
*trans
,
4677 struct btrfs_root
*root
,
4678 u64 num_bytes
, u64 parent
, u64 root_objectid
,
4679 struct btrfs_disk_key
*key
, int level
,
4680 u64 empty_size
, u64 hint_byte
, u64 search_end
,
4681 struct btrfs_key
*ins
)
4686 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, num_bytes
,
4687 empty_size
, hint_byte
, search_end
,
4692 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
4694 parent
= ins
->objectid
;
4695 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4699 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4700 struct btrfs_delayed_extent_op
*extent_op
;
4701 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
4704 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
4706 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
4707 extent_op
->flags_to_set
= flags
;
4708 extent_op
->update_key
= 1;
4709 extent_op
->update_flags
= 1;
4710 extent_op
->is_data
= 0;
4712 ret
= btrfs_add_delayed_tree_ref(trans
, ins
->objectid
,
4713 ins
->offset
, parent
, root_objectid
,
4714 level
, BTRFS_ADD_DELAYED_EXTENT
,
4721 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
4722 struct btrfs_root
*root
,
4723 u64 bytenr
, u32 blocksize
,
4726 struct extent_buffer
*buf
;
4728 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
4730 return ERR_PTR(-ENOMEM
);
4731 btrfs_set_header_generation(buf
, trans
->transid
);
4732 btrfs_set_buffer_lockdep_class(buf
, level
);
4733 btrfs_tree_lock(buf
);
4734 clean_tree_block(trans
, root
, buf
);
4736 btrfs_set_lock_blocking(buf
);
4737 btrfs_set_buffer_uptodate(buf
);
4739 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4740 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
4741 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4743 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
4744 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4746 trans
->blocks_used
++;
4747 /* this returns a buffer locked for blocking */
4752 * helper function to allocate a block for a given tree
4753 * returns the tree buffer or NULL.
4755 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
4756 struct btrfs_root
*root
, u32 blocksize
,
4757 u64 parent
, u64 root_objectid
,
4758 struct btrfs_disk_key
*key
, int level
,
4759 u64 hint
, u64 empty_size
)
4761 struct btrfs_key ins
;
4763 struct extent_buffer
*buf
;
4765 ret
= alloc_tree_block(trans
, root
, blocksize
, parent
, root_objectid
,
4766 key
, level
, empty_size
, hint
, (u64
)-1, &ins
);
4769 return ERR_PTR(ret
);
4772 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
4777 struct walk_control
{
4778 u64 refs
[BTRFS_MAX_LEVEL
];
4779 u64 flags
[BTRFS_MAX_LEVEL
];
4780 struct btrfs_key update_progress
;
4790 #define DROP_REFERENCE 1
4791 #define UPDATE_BACKREF 2
4793 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
4794 struct btrfs_root
*root
,
4795 struct walk_control
*wc
,
4796 struct btrfs_path
*path
)
4804 struct btrfs_key key
;
4805 struct extent_buffer
*eb
;
4810 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
4811 wc
->reada_count
= wc
->reada_count
* 2 / 3;
4812 wc
->reada_count
= max(wc
->reada_count
, 2);
4814 wc
->reada_count
= wc
->reada_count
* 3 / 2;
4815 wc
->reada_count
= min_t(int, wc
->reada_count
,
4816 BTRFS_NODEPTRS_PER_BLOCK(root
));
4819 eb
= path
->nodes
[wc
->level
];
4820 nritems
= btrfs_header_nritems(eb
);
4821 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
4823 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
4824 if (nread
>= wc
->reada_count
)
4828 bytenr
= btrfs_node_blockptr(eb
, slot
);
4829 generation
= btrfs_node_ptr_generation(eb
, slot
);
4831 if (slot
== path
->slots
[wc
->level
])
4834 if (wc
->stage
== UPDATE_BACKREF
&&
4835 generation
<= root
->root_key
.offset
)
4838 if (wc
->stage
== DROP_REFERENCE
) {
4839 ret
= btrfs_lookup_extent_info(trans
, root
,
4847 if (!wc
->update_ref
||
4848 generation
<= root
->root_key
.offset
)
4850 btrfs_node_key_to_cpu(eb
, &key
, slot
);
4851 ret
= btrfs_comp_cpu_keys(&key
,
4852 &wc
->update_progress
);
4857 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
4861 last
= bytenr
+ blocksize
;
4864 wc
->reada_slot
= slot
;
4868 * hepler to process tree block while walking down the tree.
4870 * when wc->stage == UPDATE_BACKREF, this function updates
4871 * back refs for pointers in the block.
4873 * NOTE: return value 1 means we should stop walking down.
4875 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
4876 struct btrfs_root
*root
,
4877 struct btrfs_path
*path
,
4878 struct walk_control
*wc
)
4880 int level
= wc
->level
;
4881 struct extent_buffer
*eb
= path
->nodes
[level
];
4882 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4885 if (wc
->stage
== UPDATE_BACKREF
&&
4886 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
4890 * when reference count of tree block is 1, it won't increase
4891 * again. once full backref flag is set, we never clear it.
4893 if ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
4894 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
))) {
4895 BUG_ON(!path
->locks
[level
]);
4896 ret
= btrfs_lookup_extent_info(trans
, root
,
4901 BUG_ON(wc
->refs
[level
] == 0);
4904 if (wc
->stage
== DROP_REFERENCE
) {
4905 if (wc
->refs
[level
] > 1)
4908 if (path
->locks
[level
] && !wc
->keep_locks
) {
4909 btrfs_tree_unlock(eb
);
4910 path
->locks
[level
] = 0;
4915 /* wc->stage == UPDATE_BACKREF */
4916 if (!(wc
->flags
[level
] & flag
)) {
4917 BUG_ON(!path
->locks
[level
]);
4918 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
4920 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
4922 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
4925 wc
->flags
[level
] |= flag
;
4929 * the block is shared by multiple trees, so it's not good to
4930 * keep the tree lock
4932 if (path
->locks
[level
] && level
> 0) {
4933 btrfs_tree_unlock(eb
);
4934 path
->locks
[level
] = 0;
4940 * hepler to process tree block pointer.
4942 * when wc->stage == DROP_REFERENCE, this function checks
4943 * reference count of the block pointed to. if the block
4944 * is shared and we need update back refs for the subtree
4945 * rooted at the block, this function changes wc->stage to
4946 * UPDATE_BACKREF. if the block is shared and there is no
4947 * need to update back, this function drops the reference
4950 * NOTE: return value 1 means we should stop walking down.
4952 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
4953 struct btrfs_root
*root
,
4954 struct btrfs_path
*path
,
4955 struct walk_control
*wc
)
4961 struct btrfs_key key
;
4962 struct extent_buffer
*next
;
4963 int level
= wc
->level
;
4967 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
4968 path
->slots
[level
]);
4970 * if the lower level block was created before the snapshot
4971 * was created, we know there is no need to update back refs
4974 if (wc
->stage
== UPDATE_BACKREF
&&
4975 generation
<= root
->root_key
.offset
)
4978 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
4979 blocksize
= btrfs_level_size(root
, level
- 1);
4981 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
4983 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
4986 btrfs_tree_lock(next
);
4987 btrfs_set_lock_blocking(next
);
4989 if (wc
->stage
== DROP_REFERENCE
) {
4990 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
4991 &wc
->refs
[level
- 1],
4992 &wc
->flags
[level
- 1]);
4994 BUG_ON(wc
->refs
[level
- 1] == 0);
4996 if (wc
->refs
[level
- 1] > 1) {
4997 if (!wc
->update_ref
||
4998 generation
<= root
->root_key
.offset
)
5001 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
5002 path
->slots
[level
]);
5003 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
5007 wc
->stage
= UPDATE_BACKREF
;
5008 wc
->shared_level
= level
- 1;
5012 if (!btrfs_buffer_uptodate(next
, generation
)) {
5013 btrfs_tree_unlock(next
);
5014 free_extent_buffer(next
);
5019 if (reada
&& level
== 1)
5020 reada_walk_down(trans
, root
, wc
, path
);
5021 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
5022 btrfs_tree_lock(next
);
5023 btrfs_set_lock_blocking(next
);
5027 BUG_ON(level
!= btrfs_header_level(next
));
5028 path
->nodes
[level
] = next
;
5029 path
->slots
[level
] = 0;
5030 path
->locks
[level
] = 1;
5036 wc
->refs
[level
- 1] = 0;
5037 wc
->flags
[level
- 1] = 0;
5039 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
5040 parent
= path
->nodes
[level
]->start
;
5042 BUG_ON(root
->root_key
.objectid
!=
5043 btrfs_header_owner(path
->nodes
[level
]));
5047 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
5048 root
->root_key
.objectid
, level
- 1, 0);
5051 btrfs_tree_unlock(next
);
5052 free_extent_buffer(next
);
5057 * hepler to process tree block while walking up the tree.
5059 * when wc->stage == DROP_REFERENCE, this function drops
5060 * reference count on the block.
5062 * when wc->stage == UPDATE_BACKREF, this function changes
5063 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5064 * to UPDATE_BACKREF previously while processing the block.
5066 * NOTE: return value 1 means we should stop walking up.
5068 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
5069 struct btrfs_root
*root
,
5070 struct btrfs_path
*path
,
5071 struct walk_control
*wc
)
5074 int level
= wc
->level
;
5075 struct extent_buffer
*eb
= path
->nodes
[level
];
5078 if (wc
->stage
== UPDATE_BACKREF
) {
5079 BUG_ON(wc
->shared_level
< level
);
5080 if (level
< wc
->shared_level
)
5083 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
5087 wc
->stage
= DROP_REFERENCE
;
5088 wc
->shared_level
= -1;
5089 path
->slots
[level
] = 0;
5092 * check reference count again if the block isn't locked.
5093 * we should start walking down the tree again if reference
5096 if (!path
->locks
[level
]) {
5098 btrfs_tree_lock(eb
);
5099 btrfs_set_lock_blocking(eb
);
5100 path
->locks
[level
] = 1;
5102 ret
= btrfs_lookup_extent_info(trans
, root
,
5107 BUG_ON(wc
->refs
[level
] == 0);
5108 if (wc
->refs
[level
] == 1) {
5109 btrfs_tree_unlock(eb
);
5110 path
->locks
[level
] = 0;
5116 /* wc->stage == DROP_REFERENCE */
5117 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
5119 if (wc
->refs
[level
] == 1) {
5121 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5122 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
5124 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5127 /* make block locked assertion in clean_tree_block happy */
5128 if (!path
->locks
[level
] &&
5129 btrfs_header_generation(eb
) == trans
->transid
) {
5130 btrfs_tree_lock(eb
);
5131 btrfs_set_lock_blocking(eb
);
5132 path
->locks
[level
] = 1;
5134 clean_tree_block(trans
, root
, eb
);
5137 if (eb
== root
->node
) {
5138 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5141 BUG_ON(root
->root_key
.objectid
!=
5142 btrfs_header_owner(eb
));
5144 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5145 parent
= path
->nodes
[level
+ 1]->start
;
5147 BUG_ON(root
->root_key
.objectid
!=
5148 btrfs_header_owner(path
->nodes
[level
+ 1]));
5151 ret
= btrfs_free_extent(trans
, root
, eb
->start
, eb
->len
, parent
,
5152 root
->root_key
.objectid
, level
, 0);
5155 wc
->refs
[level
] = 0;
5156 wc
->flags
[level
] = 0;
5160 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
5161 struct btrfs_root
*root
,
5162 struct btrfs_path
*path
,
5163 struct walk_control
*wc
)
5165 int level
= wc
->level
;
5168 while (level
>= 0) {
5169 if (path
->slots
[level
] >=
5170 btrfs_header_nritems(path
->nodes
[level
]))
5173 ret
= walk_down_proc(trans
, root
, path
, wc
);
5180 ret
= do_walk_down(trans
, root
, path
, wc
);
5182 path
->slots
[level
]++;
5190 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
5191 struct btrfs_root
*root
,
5192 struct btrfs_path
*path
,
5193 struct walk_control
*wc
, int max_level
)
5195 int level
= wc
->level
;
5198 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
5199 while (level
< max_level
&& path
->nodes
[level
]) {
5201 if (path
->slots
[level
] + 1 <
5202 btrfs_header_nritems(path
->nodes
[level
])) {
5203 path
->slots
[level
]++;
5206 ret
= walk_up_proc(trans
, root
, path
, wc
);
5210 if (path
->locks
[level
]) {
5211 btrfs_tree_unlock(path
->nodes
[level
]);
5212 path
->locks
[level
] = 0;
5214 free_extent_buffer(path
->nodes
[level
]);
5215 path
->nodes
[level
] = NULL
;
5223 * drop a subvolume tree.
5225 * this function traverses the tree freeing any blocks that only
5226 * referenced by the tree.
5228 * when a shared tree block is found. this function decreases its
5229 * reference count by one. if update_ref is true, this function
5230 * also make sure backrefs for the shared block and all lower level
5231 * blocks are properly updated.
5233 int btrfs_drop_snapshot(struct btrfs_root
*root
, int update_ref
)
5235 struct btrfs_path
*path
;
5236 struct btrfs_trans_handle
*trans
;
5237 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
5238 struct btrfs_root_item
*root_item
= &root
->root_item
;
5239 struct walk_control
*wc
;
5240 struct btrfs_key key
;
5245 path
= btrfs_alloc_path();
5248 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5251 trans
= btrfs_start_transaction(tree_root
, 1);
5253 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
5254 level
= btrfs_header_level(root
->node
);
5255 path
->nodes
[level
] = btrfs_lock_root_node(root
);
5256 btrfs_set_lock_blocking(path
->nodes
[level
]);
5257 path
->slots
[level
] = 0;
5258 path
->locks
[level
] = 1;
5259 memset(&wc
->update_progress
, 0,
5260 sizeof(wc
->update_progress
));
5262 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
5263 memcpy(&wc
->update_progress
, &key
,
5264 sizeof(wc
->update_progress
));
5266 level
= root_item
->drop_level
;
5268 path
->lowest_level
= level
;
5269 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5270 path
->lowest_level
= 0;
5278 * unlock our path, this is safe because only this
5279 * function is allowed to delete this snapshot
5281 btrfs_unlock_up_safe(path
, 0);
5283 level
= btrfs_header_level(root
->node
);
5285 btrfs_tree_lock(path
->nodes
[level
]);
5286 btrfs_set_lock_blocking(path
->nodes
[level
]);
5288 ret
= btrfs_lookup_extent_info(trans
, root
,
5289 path
->nodes
[level
]->start
,
5290 path
->nodes
[level
]->len
,
5294 BUG_ON(wc
->refs
[level
] == 0);
5296 if (level
== root_item
->drop_level
)
5299 btrfs_tree_unlock(path
->nodes
[level
]);
5300 WARN_ON(wc
->refs
[level
] != 1);
5306 wc
->shared_level
= -1;
5307 wc
->stage
= DROP_REFERENCE
;
5308 wc
->update_ref
= update_ref
;
5310 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
5313 ret
= walk_down_tree(trans
, root
, path
, wc
);
5319 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
5326 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
5330 if (wc
->stage
== DROP_REFERENCE
) {
5332 btrfs_node_key(path
->nodes
[level
],
5333 &root_item
->drop_progress
,
5334 path
->slots
[level
]);
5335 root_item
->drop_level
= level
;
5338 BUG_ON(wc
->level
== 0);
5339 if (trans
->transaction
->in_commit
||
5340 trans
->transaction
->delayed_refs
.flushing
) {
5341 ret
= btrfs_update_root(trans
, tree_root
,
5346 btrfs_end_transaction(trans
, tree_root
);
5347 trans
= btrfs_start_transaction(tree_root
, 1);
5349 unsigned long update
;
5350 update
= trans
->delayed_ref_updates
;
5351 trans
->delayed_ref_updates
= 0;
5353 btrfs_run_delayed_refs(trans
, tree_root
,
5357 btrfs_release_path(root
, path
);
5360 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
5363 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
5364 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
5368 ret
= btrfs_del_orphan_item(trans
, tree_root
,
5369 root
->root_key
.objectid
);
5374 if (root
->in_radix
) {
5375 btrfs_free_fs_root(tree_root
->fs_info
, root
);
5377 free_extent_buffer(root
->node
);
5378 free_extent_buffer(root
->commit_root
);
5382 btrfs_end_transaction(trans
, tree_root
);
5384 btrfs_free_path(path
);
5389 * drop subtree rooted at tree block 'node'.
5391 * NOTE: this function will unlock and release tree block 'node'
5393 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
5394 struct btrfs_root
*root
,
5395 struct extent_buffer
*node
,
5396 struct extent_buffer
*parent
)
5398 struct btrfs_path
*path
;
5399 struct walk_control
*wc
;
5405 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
5407 path
= btrfs_alloc_path();
5410 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5413 btrfs_assert_tree_locked(parent
);
5414 parent_level
= btrfs_header_level(parent
);
5415 extent_buffer_get(parent
);
5416 path
->nodes
[parent_level
] = parent
;
5417 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
5419 btrfs_assert_tree_locked(node
);
5420 level
= btrfs_header_level(node
);
5421 path
->nodes
[level
] = node
;
5422 path
->slots
[level
] = 0;
5423 path
->locks
[level
] = 1;
5425 wc
->refs
[parent_level
] = 1;
5426 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5428 wc
->shared_level
= -1;
5429 wc
->stage
= DROP_REFERENCE
;
5432 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
5435 wret
= walk_down_tree(trans
, root
, path
, wc
);
5441 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
5449 btrfs_free_path(path
);
5454 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
5457 return min(last
, start
+ nr
- 1);
5460 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
5465 unsigned long first_index
;
5466 unsigned long last_index
;
5469 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5470 struct file_ra_state
*ra
;
5471 struct btrfs_ordered_extent
*ordered
;
5472 unsigned int total_read
= 0;
5473 unsigned int total_dirty
= 0;
5476 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
5478 mutex_lock(&inode
->i_mutex
);
5479 first_index
= start
>> PAGE_CACHE_SHIFT
;
5480 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
5482 /* make sure the dirty trick played by the caller work */
5483 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
5484 first_index
, last_index
);
5488 file_ra_state_init(ra
, inode
->i_mapping
);
5490 for (i
= first_index
; i
<= last_index
; i
++) {
5491 if (total_read
% ra
->ra_pages
== 0) {
5492 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
5493 calc_ra(i
, last_index
, ra
->ra_pages
));
5497 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
5499 page
= grab_cache_page(inode
->i_mapping
, i
);
5504 if (!PageUptodate(page
)) {
5505 btrfs_readpage(NULL
, page
);
5507 if (!PageUptodate(page
)) {
5509 page_cache_release(page
);
5514 wait_on_page_writeback(page
);
5516 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
5517 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
5518 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5520 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
5522 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5524 page_cache_release(page
);
5525 btrfs_start_ordered_extent(inode
, ordered
, 1);
5526 btrfs_put_ordered_extent(ordered
);
5529 set_page_extent_mapped(page
);
5531 if (i
== first_index
)
5532 set_extent_bits(io_tree
, page_start
, page_end
,
5533 EXTENT_BOUNDARY
, GFP_NOFS
);
5534 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
5536 set_page_dirty(page
);
5539 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5541 page_cache_release(page
);
5546 mutex_unlock(&inode
->i_mutex
);
5547 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
5551 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
5552 struct btrfs_key
*extent_key
,
5555 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5556 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
5557 struct extent_map
*em
;
5558 u64 start
= extent_key
->objectid
- offset
;
5559 u64 end
= start
+ extent_key
->offset
- 1;
5561 em
= alloc_extent_map(GFP_NOFS
);
5562 BUG_ON(!em
|| IS_ERR(em
));
5565 em
->len
= extent_key
->offset
;
5566 em
->block_len
= extent_key
->offset
;
5567 em
->block_start
= extent_key
->objectid
;
5568 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5569 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5571 /* setup extent map to cheat btrfs_readpage */
5572 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5575 write_lock(&em_tree
->lock
);
5576 ret
= add_extent_mapping(em_tree
, em
);
5577 write_unlock(&em_tree
->lock
);
5578 if (ret
!= -EEXIST
) {
5579 free_extent_map(em
);
5582 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
5584 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5586 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
5589 struct btrfs_ref_path
{
5591 u64 nodes
[BTRFS_MAX_LEVEL
];
5593 u64 root_generation
;
5600 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
5601 u64 new_nodes
[BTRFS_MAX_LEVEL
];
5604 struct disk_extent
{
5615 static int is_cowonly_root(u64 root_objectid
)
5617 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
5618 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
5619 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
5620 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
5621 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
5622 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
5627 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
5628 struct btrfs_root
*extent_root
,
5629 struct btrfs_ref_path
*ref_path
,
5632 struct extent_buffer
*leaf
;
5633 struct btrfs_path
*path
;
5634 struct btrfs_extent_ref
*ref
;
5635 struct btrfs_key key
;
5636 struct btrfs_key found_key
;
5642 path
= btrfs_alloc_path();
5647 ref_path
->lowest_level
= -1;
5648 ref_path
->current_level
= -1;
5649 ref_path
->shared_level
= -1;
5653 level
= ref_path
->current_level
- 1;
5654 while (level
>= -1) {
5656 if (level
< ref_path
->lowest_level
)
5660 bytenr
= ref_path
->nodes
[level
];
5662 bytenr
= ref_path
->extent_start
;
5663 BUG_ON(bytenr
== 0);
5665 parent
= ref_path
->nodes
[level
+ 1];
5666 ref_path
->nodes
[level
+ 1] = 0;
5667 ref_path
->current_level
= level
;
5668 BUG_ON(parent
== 0);
5670 key
.objectid
= bytenr
;
5671 key
.offset
= parent
+ 1;
5672 key
.type
= BTRFS_EXTENT_REF_KEY
;
5674 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5679 leaf
= path
->nodes
[0];
5680 nritems
= btrfs_header_nritems(leaf
);
5681 if (path
->slots
[0] >= nritems
) {
5682 ret
= btrfs_next_leaf(extent_root
, path
);
5687 leaf
= path
->nodes
[0];
5690 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5691 if (found_key
.objectid
== bytenr
&&
5692 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
5693 if (level
< ref_path
->shared_level
)
5694 ref_path
->shared_level
= level
;
5699 btrfs_release_path(extent_root
, path
);
5702 /* reached lowest level */
5706 level
= ref_path
->current_level
;
5707 while (level
< BTRFS_MAX_LEVEL
- 1) {
5711 bytenr
= ref_path
->nodes
[level
];
5713 bytenr
= ref_path
->extent_start
;
5715 BUG_ON(bytenr
== 0);
5717 key
.objectid
= bytenr
;
5719 key
.type
= BTRFS_EXTENT_REF_KEY
;
5721 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5725 leaf
= path
->nodes
[0];
5726 nritems
= btrfs_header_nritems(leaf
);
5727 if (path
->slots
[0] >= nritems
) {
5728 ret
= btrfs_next_leaf(extent_root
, path
);
5732 /* the extent was freed by someone */
5733 if (ref_path
->lowest_level
== level
)
5735 btrfs_release_path(extent_root
, path
);
5738 leaf
= path
->nodes
[0];
5741 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5742 if (found_key
.objectid
!= bytenr
||
5743 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
5744 /* the extent was freed by someone */
5745 if (ref_path
->lowest_level
== level
) {
5749 btrfs_release_path(extent_root
, path
);
5753 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
5754 struct btrfs_extent_ref
);
5755 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
5756 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
5758 level
= (int)ref_objectid
;
5759 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
5760 ref_path
->lowest_level
= level
;
5761 ref_path
->current_level
= level
;
5762 ref_path
->nodes
[level
] = bytenr
;
5764 WARN_ON(ref_objectid
!= level
);
5767 WARN_ON(level
!= -1);
5771 if (ref_path
->lowest_level
== level
) {
5772 ref_path
->owner_objectid
= ref_objectid
;
5773 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
5777 * the block is tree root or the block isn't in reference
5780 if (found_key
.objectid
== found_key
.offset
||
5781 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
5782 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5783 ref_path
->root_generation
=
5784 btrfs_ref_generation(leaf
, ref
);
5786 /* special reference from the tree log */
5787 ref_path
->nodes
[0] = found_key
.offset
;
5788 ref_path
->current_level
= 0;
5795 BUG_ON(ref_path
->nodes
[level
] != 0);
5796 ref_path
->nodes
[level
] = found_key
.offset
;
5797 ref_path
->current_level
= level
;
5800 * the reference was created in the running transaction,
5801 * no need to continue walking up.
5803 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
5804 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5805 ref_path
->root_generation
=
5806 btrfs_ref_generation(leaf
, ref
);
5811 btrfs_release_path(extent_root
, path
);
5814 /* reached max tree level, but no tree root found. */
5817 btrfs_free_path(path
);
5821 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
5822 struct btrfs_root
*extent_root
,
5823 struct btrfs_ref_path
*ref_path
,
5826 memset(ref_path
, 0, sizeof(*ref_path
));
5827 ref_path
->extent_start
= extent_start
;
5829 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
5832 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
5833 struct btrfs_root
*extent_root
,
5834 struct btrfs_ref_path
*ref_path
)
5836 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
5839 static noinline
int get_new_locations(struct inode
*reloc_inode
,
5840 struct btrfs_key
*extent_key
,
5841 u64 offset
, int no_fragment
,
5842 struct disk_extent
**extents
,
5845 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5846 struct btrfs_path
*path
;
5847 struct btrfs_file_extent_item
*fi
;
5848 struct extent_buffer
*leaf
;
5849 struct disk_extent
*exts
= *extents
;
5850 struct btrfs_key found_key
;
5855 int max
= *nr_extents
;
5858 WARN_ON(!no_fragment
&& *extents
);
5861 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
5866 path
= btrfs_alloc_path();
5869 cur_pos
= extent_key
->objectid
- offset
;
5870 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
5871 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
5881 leaf
= path
->nodes
[0];
5882 nritems
= btrfs_header_nritems(leaf
);
5883 if (path
->slots
[0] >= nritems
) {
5884 ret
= btrfs_next_leaf(root
, path
);
5889 leaf
= path
->nodes
[0];
5892 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5893 if (found_key
.offset
!= cur_pos
||
5894 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
5895 found_key
.objectid
!= reloc_inode
->i_ino
)
5898 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
5899 struct btrfs_file_extent_item
);
5900 if (btrfs_file_extent_type(leaf
, fi
) !=
5901 BTRFS_FILE_EXTENT_REG
||
5902 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
5906 struct disk_extent
*old
= exts
;
5908 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
5909 memcpy(exts
, old
, sizeof(*exts
) * nr
);
5910 if (old
!= *extents
)
5914 exts
[nr
].disk_bytenr
=
5915 btrfs_file_extent_disk_bytenr(leaf
, fi
);
5916 exts
[nr
].disk_num_bytes
=
5917 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
5918 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
5919 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
5920 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
5921 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
5922 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
5923 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
5925 BUG_ON(exts
[nr
].offset
> 0);
5926 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
5927 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
5929 cur_pos
+= exts
[nr
].num_bytes
;
5932 if (cur_pos
+ offset
>= last_byte
)
5942 BUG_ON(cur_pos
+ offset
> last_byte
);
5943 if (cur_pos
+ offset
< last_byte
) {
5949 btrfs_free_path(path
);
5951 if (exts
!= *extents
)
5960 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
5961 struct btrfs_root
*root
,
5962 struct btrfs_path
*path
,
5963 struct btrfs_key
*extent_key
,
5964 struct btrfs_key
*leaf_key
,
5965 struct btrfs_ref_path
*ref_path
,
5966 struct disk_extent
*new_extents
,
5969 struct extent_buffer
*leaf
;
5970 struct btrfs_file_extent_item
*fi
;
5971 struct inode
*inode
= NULL
;
5972 struct btrfs_key key
;
5977 u64 search_end
= (u64
)-1;
5980 int extent_locked
= 0;
5984 memcpy(&key
, leaf_key
, sizeof(key
));
5985 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
5986 if (key
.objectid
< ref_path
->owner_objectid
||
5987 (key
.objectid
== ref_path
->owner_objectid
&&
5988 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
5989 key
.objectid
= ref_path
->owner_objectid
;
5990 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5996 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
6000 leaf
= path
->nodes
[0];
6001 nritems
= btrfs_header_nritems(leaf
);
6003 if (extent_locked
&& ret
> 0) {
6005 * the file extent item was modified by someone
6006 * before the extent got locked.
6008 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6009 lock_end
, GFP_NOFS
);
6013 if (path
->slots
[0] >= nritems
) {
6014 if (++nr_scaned
> 2)
6017 BUG_ON(extent_locked
);
6018 ret
= btrfs_next_leaf(root
, path
);
6023 leaf
= path
->nodes
[0];
6024 nritems
= btrfs_header_nritems(leaf
);
6027 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
6029 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
6030 if ((key
.objectid
> ref_path
->owner_objectid
) ||
6031 (key
.objectid
== ref_path
->owner_objectid
&&
6032 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
6033 key
.offset
>= search_end
)
6037 if (inode
&& key
.objectid
!= inode
->i_ino
) {
6038 BUG_ON(extent_locked
);
6039 btrfs_release_path(root
, path
);
6040 mutex_unlock(&inode
->i_mutex
);
6046 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6051 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6052 struct btrfs_file_extent_item
);
6053 extent_type
= btrfs_file_extent_type(leaf
, fi
);
6054 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
6055 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
6056 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
6057 extent_key
->objectid
)) {
6063 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6064 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
6066 if (search_end
== (u64
)-1) {
6067 search_end
= key
.offset
- ext_offset
+
6068 btrfs_file_extent_ram_bytes(leaf
, fi
);
6071 if (!extent_locked
) {
6072 lock_start
= key
.offset
;
6073 lock_end
= lock_start
+ num_bytes
- 1;
6075 if (lock_start
> key
.offset
||
6076 lock_end
+ 1 < key
.offset
+ num_bytes
) {
6077 unlock_extent(&BTRFS_I(inode
)->io_tree
,
6078 lock_start
, lock_end
, GFP_NOFS
);
6084 btrfs_release_path(root
, path
);
6086 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
6087 key
.objectid
, root
);
6088 if (inode
->i_state
& I_NEW
) {
6089 BTRFS_I(inode
)->root
= root
;
6090 BTRFS_I(inode
)->location
.objectid
=
6092 BTRFS_I(inode
)->location
.type
=
6093 BTRFS_INODE_ITEM_KEY
;
6094 BTRFS_I(inode
)->location
.offset
= 0;
6095 btrfs_read_locked_inode(inode
);
6096 unlock_new_inode(inode
);
6099 * some code call btrfs_commit_transaction while
6100 * holding the i_mutex, so we can't use mutex_lock
6103 if (is_bad_inode(inode
) ||
6104 !mutex_trylock(&inode
->i_mutex
)) {
6107 key
.offset
= (u64
)-1;
6112 if (!extent_locked
) {
6113 struct btrfs_ordered_extent
*ordered
;
6115 btrfs_release_path(root
, path
);
6117 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6118 lock_end
, GFP_NOFS
);
6119 ordered
= btrfs_lookup_first_ordered_extent(inode
,
6122 ordered
->file_offset
<= lock_end
&&
6123 ordered
->file_offset
+ ordered
->len
> lock_start
) {
6124 unlock_extent(&BTRFS_I(inode
)->io_tree
,
6125 lock_start
, lock_end
, GFP_NOFS
);
6126 btrfs_start_ordered_extent(inode
, ordered
, 1);
6127 btrfs_put_ordered_extent(ordered
);
6128 key
.offset
+= num_bytes
;
6132 btrfs_put_ordered_extent(ordered
);
6138 if (nr_extents
== 1) {
6139 /* update extent pointer in place */
6140 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6141 new_extents
[0].disk_bytenr
);
6142 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6143 new_extents
[0].disk_num_bytes
);
6144 btrfs_mark_buffer_dirty(leaf
);
6146 btrfs_drop_extent_cache(inode
, key
.offset
,
6147 key
.offset
+ num_bytes
- 1, 0);
6149 ret
= btrfs_inc_extent_ref(trans
, root
,
6150 new_extents
[0].disk_bytenr
,
6151 new_extents
[0].disk_num_bytes
,
6153 root
->root_key
.objectid
,
6158 ret
= btrfs_free_extent(trans
, root
,
6159 extent_key
->objectid
,
6162 btrfs_header_owner(leaf
),
6163 btrfs_header_generation(leaf
),
6167 btrfs_release_path(root
, path
);
6168 key
.offset
+= num_bytes
;
6176 * drop old extent pointer at first, then insert the
6177 * new pointers one bye one
6179 btrfs_release_path(root
, path
);
6180 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
6181 key
.offset
+ num_bytes
,
6182 key
.offset
, &alloc_hint
);
6185 for (i
= 0; i
< nr_extents
; i
++) {
6186 if (ext_offset
>= new_extents
[i
].num_bytes
) {
6187 ext_offset
-= new_extents
[i
].num_bytes
;
6190 extent_len
= min(new_extents
[i
].num_bytes
-
6191 ext_offset
, num_bytes
);
6193 ret
= btrfs_insert_empty_item(trans
, root
,
6198 leaf
= path
->nodes
[0];
6199 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6200 struct btrfs_file_extent_item
);
6201 btrfs_set_file_extent_generation(leaf
, fi
,
6203 btrfs_set_file_extent_type(leaf
, fi
,
6204 BTRFS_FILE_EXTENT_REG
);
6205 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6206 new_extents
[i
].disk_bytenr
);
6207 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6208 new_extents
[i
].disk_num_bytes
);
6209 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
6210 new_extents
[i
].ram_bytes
);
6212 btrfs_set_file_extent_compression(leaf
, fi
,
6213 new_extents
[i
].compression
);
6214 btrfs_set_file_extent_encryption(leaf
, fi
,
6215 new_extents
[i
].encryption
);
6216 btrfs_set_file_extent_other_encoding(leaf
, fi
,
6217 new_extents
[i
].other_encoding
);
6219 btrfs_set_file_extent_num_bytes(leaf
, fi
,
6221 ext_offset
+= new_extents
[i
].offset
;
6222 btrfs_set_file_extent_offset(leaf
, fi
,
6224 btrfs_mark_buffer_dirty(leaf
);
6226 btrfs_drop_extent_cache(inode
, key
.offset
,
6227 key
.offset
+ extent_len
- 1, 0);
6229 ret
= btrfs_inc_extent_ref(trans
, root
,
6230 new_extents
[i
].disk_bytenr
,
6231 new_extents
[i
].disk_num_bytes
,
6233 root
->root_key
.objectid
,
6234 trans
->transid
, key
.objectid
);
6236 btrfs_release_path(root
, path
);
6238 inode_add_bytes(inode
, extent_len
);
6241 num_bytes
-= extent_len
;
6242 key
.offset
+= extent_len
;
6247 BUG_ON(i
>= nr_extents
);
6251 if (extent_locked
) {
6252 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6253 lock_end
, GFP_NOFS
);
6257 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
6258 key
.offset
>= search_end
)
6265 btrfs_release_path(root
, path
);
6267 mutex_unlock(&inode
->i_mutex
);
6268 if (extent_locked
) {
6269 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6270 lock_end
, GFP_NOFS
);
6277 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
6278 struct btrfs_root
*root
,
6279 struct extent_buffer
*buf
, u64 orig_start
)
6284 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
6285 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6287 level
= btrfs_header_level(buf
);
6289 struct btrfs_leaf_ref
*ref
;
6290 struct btrfs_leaf_ref
*orig_ref
;
6292 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
6296 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
6298 btrfs_free_leaf_ref(root
, orig_ref
);
6302 ref
->nritems
= orig_ref
->nritems
;
6303 memcpy(ref
->extents
, orig_ref
->extents
,
6304 sizeof(ref
->extents
[0]) * ref
->nritems
);
6306 btrfs_free_leaf_ref(root
, orig_ref
);
6308 ref
->root_gen
= trans
->transid
;
6309 ref
->bytenr
= buf
->start
;
6310 ref
->owner
= btrfs_header_owner(buf
);
6311 ref
->generation
= btrfs_header_generation(buf
);
6313 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
6315 btrfs_free_leaf_ref(root
, ref
);
6320 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
6321 struct extent_buffer
*leaf
,
6322 struct btrfs_block_group_cache
*group
,
6323 struct btrfs_root
*target_root
)
6325 struct btrfs_key key
;
6326 struct inode
*inode
= NULL
;
6327 struct btrfs_file_extent_item
*fi
;
6329 u64 skip_objectid
= 0;
6333 nritems
= btrfs_header_nritems(leaf
);
6334 for (i
= 0; i
< nritems
; i
++) {
6335 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6336 if (key
.objectid
== skip_objectid
||
6337 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6339 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6340 if (btrfs_file_extent_type(leaf
, fi
) ==
6341 BTRFS_FILE_EXTENT_INLINE
)
6343 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6345 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
6347 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
6348 key
.objectid
, target_root
, 1);
6351 skip_objectid
= key
.objectid
;
6354 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6356 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
6357 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
6358 btrfs_drop_extent_cache(inode
, key
.offset
,
6359 key
.offset
+ num_bytes
- 1, 1);
6360 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
6361 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
6368 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
6369 struct btrfs_root
*root
,
6370 struct extent_buffer
*leaf
,
6371 struct btrfs_block_group_cache
*group
,
6372 struct inode
*reloc_inode
)
6374 struct btrfs_key key
;
6375 struct btrfs_key extent_key
;
6376 struct btrfs_file_extent_item
*fi
;
6377 struct btrfs_leaf_ref
*ref
;
6378 struct disk_extent
*new_extent
;
6387 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
6388 BUG_ON(!new_extent
);
6390 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
6394 nritems
= btrfs_header_nritems(leaf
);
6395 for (i
= 0; i
< nritems
; i
++) {
6396 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6397 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
6399 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6400 if (btrfs_file_extent_type(leaf
, fi
) ==
6401 BTRFS_FILE_EXTENT_INLINE
)
6403 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
6404 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6409 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
6410 bytenr
+ num_bytes
<= group
->key
.objectid
)
6413 extent_key
.objectid
= bytenr
;
6414 extent_key
.offset
= num_bytes
;
6415 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6417 ret
= get_new_locations(reloc_inode
, &extent_key
,
6418 group
->key
.objectid
, 1,
6419 &new_extent
, &nr_extent
);
6424 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
6425 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
6426 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
6427 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
6429 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6430 new_extent
->disk_bytenr
);
6431 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6432 new_extent
->disk_num_bytes
);
6433 btrfs_mark_buffer_dirty(leaf
);
6435 ret
= btrfs_inc_extent_ref(trans
, root
,
6436 new_extent
->disk_bytenr
,
6437 new_extent
->disk_num_bytes
,
6439 root
->root_key
.objectid
,
6440 trans
->transid
, key
.objectid
);
6443 ret
= btrfs_free_extent(trans
, root
,
6444 bytenr
, num_bytes
, leaf
->start
,
6445 btrfs_header_owner(leaf
),
6446 btrfs_header_generation(leaf
),
6452 BUG_ON(ext_index
+ 1 != ref
->nritems
);
6453 btrfs_free_leaf_ref(root
, ref
);
6457 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
6458 struct btrfs_root
*root
)
6460 struct btrfs_root
*reloc_root
;
6463 if (root
->reloc_root
) {
6464 reloc_root
= root
->reloc_root
;
6465 root
->reloc_root
= NULL
;
6466 list_add(&reloc_root
->dead_list
,
6467 &root
->fs_info
->dead_reloc_roots
);
6469 btrfs_set_root_bytenr(&reloc_root
->root_item
,
6470 reloc_root
->node
->start
);
6471 btrfs_set_root_level(&root
->root_item
,
6472 btrfs_header_level(reloc_root
->node
));
6473 memset(&reloc_root
->root_item
.drop_progress
, 0,
6474 sizeof(struct btrfs_disk_key
));
6475 reloc_root
->root_item
.drop_level
= 0;
6477 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
6478 &reloc_root
->root_key
,
6479 &reloc_root
->root_item
);
6485 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
6487 struct btrfs_trans_handle
*trans
;
6488 struct btrfs_root
*reloc_root
;
6489 struct btrfs_root
*prev_root
= NULL
;
6490 struct list_head dead_roots
;
6494 INIT_LIST_HEAD(&dead_roots
);
6495 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
6497 while (!list_empty(&dead_roots
)) {
6498 reloc_root
= list_entry(dead_roots
.prev
,
6499 struct btrfs_root
, dead_list
);
6500 list_del_init(&reloc_root
->dead_list
);
6502 BUG_ON(reloc_root
->commit_root
!= NULL
);
6504 trans
= btrfs_join_transaction(root
, 1);
6507 mutex_lock(&root
->fs_info
->drop_mutex
);
6508 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
6511 mutex_unlock(&root
->fs_info
->drop_mutex
);
6513 nr
= trans
->blocks_used
;
6514 ret
= btrfs_end_transaction(trans
, root
);
6516 btrfs_btree_balance_dirty(root
, nr
);
6519 free_extent_buffer(reloc_root
->node
);
6521 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
6522 &reloc_root
->root_key
);
6524 mutex_unlock(&root
->fs_info
->drop_mutex
);
6526 nr
= trans
->blocks_used
;
6527 ret
= btrfs_end_transaction(trans
, root
);
6529 btrfs_btree_balance_dirty(root
, nr
);
6532 prev_root
= reloc_root
;
6535 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
6541 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
6543 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
6547 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
6549 struct btrfs_root
*reloc_root
;
6550 struct btrfs_trans_handle
*trans
;
6551 struct btrfs_key location
;
6555 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6556 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
6558 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
6559 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6562 trans
= btrfs_start_transaction(root
, 1);
6564 ret
= btrfs_commit_transaction(trans
, root
);
6568 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
6569 location
.offset
= (u64
)-1;
6570 location
.type
= BTRFS_ROOT_ITEM_KEY
;
6572 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
6573 BUG_ON(!reloc_root
);
6574 btrfs_orphan_cleanup(reloc_root
);
6578 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
6579 struct btrfs_root
*root
)
6581 struct btrfs_root
*reloc_root
;
6582 struct extent_buffer
*eb
;
6583 struct btrfs_root_item
*root_item
;
6584 struct btrfs_key root_key
;
6587 BUG_ON(!root
->ref_cows
);
6588 if (root
->reloc_root
)
6591 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
6594 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
6595 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
6598 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
6599 root_key
.offset
= root
->root_key
.objectid
;
6600 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6602 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
6603 btrfs_set_root_refs(root_item
, 0);
6604 btrfs_set_root_bytenr(root_item
, eb
->start
);
6605 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
6606 btrfs_set_root_generation(root_item
, trans
->transid
);
6608 btrfs_tree_unlock(eb
);
6609 free_extent_buffer(eb
);
6611 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
6612 &root_key
, root_item
);
6616 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
6618 BUG_ON(!reloc_root
);
6619 reloc_root
->last_trans
= trans
->transid
;
6620 reloc_root
->commit_root
= NULL
;
6621 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
6623 root
->reloc_root
= reloc_root
;
6628 * Core function of space balance.
6630 * The idea is using reloc trees to relocate tree blocks in reference
6631 * counted roots. There is one reloc tree for each subvol, and all
6632 * reloc trees share same root key objectid. Reloc trees are snapshots
6633 * of the latest committed roots of subvols (root->commit_root).
6635 * To relocate a tree block referenced by a subvol, there are two steps.
6636 * COW the block through subvol's reloc tree, then update block pointer
6637 * in the subvol to point to the new block. Since all reloc trees share
6638 * same root key objectid, doing special handing for tree blocks owned
6639 * by them is easy. Once a tree block has been COWed in one reloc tree,
6640 * we can use the resulting new block directly when the same block is
6641 * required to COW again through other reloc trees. By this way, relocated
6642 * tree blocks are shared between reloc trees, so they are also shared
6645 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
6646 struct btrfs_root
*root
,
6647 struct btrfs_path
*path
,
6648 struct btrfs_key
*first_key
,
6649 struct btrfs_ref_path
*ref_path
,
6650 struct btrfs_block_group_cache
*group
,
6651 struct inode
*reloc_inode
)
6653 struct btrfs_root
*reloc_root
;
6654 struct extent_buffer
*eb
= NULL
;
6655 struct btrfs_key
*keys
;
6659 int lowest_level
= 0;
6662 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
6663 lowest_level
= ref_path
->owner_objectid
;
6665 if (!root
->ref_cows
) {
6666 path
->lowest_level
= lowest_level
;
6667 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
6669 path
->lowest_level
= 0;
6670 btrfs_release_path(root
, path
);
6674 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6675 ret
= init_reloc_tree(trans
, root
);
6677 reloc_root
= root
->reloc_root
;
6679 shared_level
= ref_path
->shared_level
;
6680 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
6682 keys
= ref_path
->node_keys
;
6683 nodes
= ref_path
->new_nodes
;
6684 memset(&keys
[shared_level
+ 1], 0,
6685 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6686 memset(&nodes
[shared_level
+ 1], 0,
6687 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6689 if (nodes
[lowest_level
] == 0) {
6690 path
->lowest_level
= lowest_level
;
6691 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6694 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
6695 eb
= path
->nodes
[level
];
6696 if (!eb
|| eb
== reloc_root
->node
)
6698 nodes
[level
] = eb
->start
;
6700 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
6702 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
6705 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6706 eb
= path
->nodes
[0];
6707 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
6708 group
, reloc_inode
);
6711 btrfs_release_path(reloc_root
, path
);
6713 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
6719 * replace tree blocks in the fs tree with tree blocks in
6722 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
6725 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6726 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6729 extent_buffer_get(path
->nodes
[0]);
6730 eb
= path
->nodes
[0];
6731 btrfs_release_path(reloc_root
, path
);
6732 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
6734 free_extent_buffer(eb
);
6737 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6738 path
->lowest_level
= 0;
6742 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
6743 struct btrfs_root
*root
,
6744 struct btrfs_path
*path
,
6745 struct btrfs_key
*first_key
,
6746 struct btrfs_ref_path
*ref_path
)
6750 ret
= relocate_one_path(trans
, root
, path
, first_key
,
6751 ref_path
, NULL
, NULL
);
6757 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
6758 struct btrfs_root
*extent_root
,
6759 struct btrfs_path
*path
,
6760 struct btrfs_key
*extent_key
)
6764 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
6767 ret
= btrfs_del_item(trans
, extent_root
, path
);
6769 btrfs_release_path(extent_root
, path
);
6773 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
6774 struct btrfs_ref_path
*ref_path
)
6776 struct btrfs_key root_key
;
6778 root_key
.objectid
= ref_path
->root_objectid
;
6779 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6780 if (is_cowonly_root(ref_path
->root_objectid
))
6781 root_key
.offset
= 0;
6783 root_key
.offset
= (u64
)-1;
6785 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
6788 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
6789 struct btrfs_path
*path
,
6790 struct btrfs_key
*extent_key
,
6791 struct btrfs_block_group_cache
*group
,
6792 struct inode
*reloc_inode
, int pass
)
6794 struct btrfs_trans_handle
*trans
;
6795 struct btrfs_root
*found_root
;
6796 struct btrfs_ref_path
*ref_path
= NULL
;
6797 struct disk_extent
*new_extents
= NULL
;
6802 struct btrfs_key first_key
;
6806 trans
= btrfs_start_transaction(extent_root
, 1);
6809 if (extent_key
->objectid
== 0) {
6810 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
6814 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
6820 for (loops
= 0; ; loops
++) {
6822 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
6823 extent_key
->objectid
);
6825 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
6832 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6833 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
6836 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
6837 BUG_ON(!found_root
);
6839 * for reference counted tree, only process reference paths
6840 * rooted at the latest committed root.
6842 if (found_root
->ref_cows
&&
6843 ref_path
->root_generation
!= found_root
->root_key
.offset
)
6846 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6849 * copy data extents to new locations
6851 u64 group_start
= group
->key
.objectid
;
6852 ret
= relocate_data_extent(reloc_inode
,
6861 level
= ref_path
->owner_objectid
;
6864 if (prev_block
!= ref_path
->nodes
[level
]) {
6865 struct extent_buffer
*eb
;
6866 u64 block_start
= ref_path
->nodes
[level
];
6867 u64 block_size
= btrfs_level_size(found_root
, level
);
6869 eb
= read_tree_block(found_root
, block_start
,
6871 btrfs_tree_lock(eb
);
6872 BUG_ON(level
!= btrfs_header_level(eb
));
6875 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
6877 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
6879 btrfs_tree_unlock(eb
);
6880 free_extent_buffer(eb
);
6881 prev_block
= block_start
;
6884 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
6885 btrfs_record_root_in_trans(found_root
);
6886 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
6887 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6889 * try to update data extent references while
6890 * keeping metadata shared between snapshots.
6893 ret
= relocate_one_path(trans
, found_root
,
6894 path
, &first_key
, ref_path
,
6895 group
, reloc_inode
);
6901 * use fallback method to process the remaining
6905 u64 group_start
= group
->key
.objectid
;
6906 new_extents
= kmalloc(sizeof(*new_extents
),
6909 ret
= get_new_locations(reloc_inode
,
6917 ret
= replace_one_extent(trans
, found_root
,
6919 &first_key
, ref_path
,
6920 new_extents
, nr_extents
);
6922 ret
= relocate_tree_block(trans
, found_root
, path
,
6923 &first_key
, ref_path
);
6930 btrfs_end_transaction(trans
, extent_root
);
6937 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6940 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6941 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6943 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
6944 if (num_devices
== 1) {
6945 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6946 stripped
= flags
& ~stripped
;
6948 /* turn raid0 into single device chunks */
6949 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6952 /* turn mirroring into duplication */
6953 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6954 BTRFS_BLOCK_GROUP_RAID10
))
6955 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6958 /* they already had raid on here, just return */
6959 if (flags
& stripped
)
6962 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6963 stripped
= flags
& ~stripped
;
6965 /* switch duplicated blocks with raid1 */
6966 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6967 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6969 /* turn single device chunks into raid0 */
6970 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6975 static int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
6976 struct btrfs_block_group_cache
*shrink_block_group
,
6979 struct btrfs_trans_handle
*trans
;
6980 u64 new_alloc_flags
;
6983 spin_lock(&shrink_block_group
->lock
);
6984 if (btrfs_block_group_used(&shrink_block_group
->item
) +
6985 shrink_block_group
->reserved
> 0) {
6986 spin_unlock(&shrink_block_group
->lock
);
6988 trans
= btrfs_start_transaction(root
, 1);
6989 spin_lock(&shrink_block_group
->lock
);
6991 new_alloc_flags
= update_block_group_flags(root
,
6992 shrink_block_group
->flags
);
6993 if (new_alloc_flags
!= shrink_block_group
->flags
) {
6995 btrfs_block_group_used(&shrink_block_group
->item
);
6997 calc
= shrink_block_group
->key
.offset
;
6999 spin_unlock(&shrink_block_group
->lock
);
7001 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
7002 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
7004 btrfs_end_transaction(trans
, root
);
7006 spin_unlock(&shrink_block_group
->lock
);
7011 int btrfs_prepare_block_group_relocation(struct btrfs_root
*root
,
7012 struct btrfs_block_group_cache
*group
)
7015 __alloc_chunk_for_shrink(root
, group
, 1);
7016 set_block_group_readonly(group
);
7021 * checks to see if its even possible to relocate this block group.
7023 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7024 * ok to go ahead and try.
7026 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
7028 struct btrfs_block_group_cache
*block_group
;
7029 struct btrfs_space_info
*space_info
;
7030 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7031 struct btrfs_device
*device
;
7035 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
7037 /* odd, couldn't find the block group, leave it alone */
7041 /* no bytes used, we're good */
7042 if (!btrfs_block_group_used(&block_group
->item
))
7045 space_info
= block_group
->space_info
;
7046 spin_lock(&space_info
->lock
);
7048 full
= space_info
->full
;
7051 * if this is the last block group we have in this space, we can't
7052 * relocate it unless we're able to allocate a new chunk below.
7054 * Otherwise, we need to make sure we have room in the space to handle
7055 * all of the extents from this block group. If we can, we're good
7057 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
7058 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
7059 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
7060 btrfs_block_group_used(&block_group
->item
) <
7061 space_info
->total_bytes
)) {
7062 spin_unlock(&space_info
->lock
);
7065 spin_unlock(&space_info
->lock
);
7068 * ok we don't have enough space, but maybe we have free space on our
7069 * devices to allocate new chunks for relocation, so loop through our
7070 * alloc devices and guess if we have enough space. However, if we
7071 * were marked as full, then we know there aren't enough chunks, and we
7078 mutex_lock(&root
->fs_info
->chunk_mutex
);
7079 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
7080 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
7081 u64 dev_offset
, max_avail
;
7084 * check to make sure we can actually find a chunk with enough
7085 * space to fit our block group in.
7087 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
7088 ret
= find_free_dev_extent(NULL
, device
, min_free
,
7089 &dev_offset
, &max_avail
);
7095 mutex_unlock(&root
->fs_info
->chunk_mutex
);
7097 btrfs_put_block_group(block_group
);
7101 static int find_first_block_group(struct btrfs_root
*root
,
7102 struct btrfs_path
*path
, struct btrfs_key
*key
)
7105 struct btrfs_key found_key
;
7106 struct extent_buffer
*leaf
;
7109 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7114 slot
= path
->slots
[0];
7115 leaf
= path
->nodes
[0];
7116 if (slot
>= btrfs_header_nritems(leaf
)) {
7117 ret
= btrfs_next_leaf(root
, path
);
7124 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7126 if (found_key
.objectid
>= key
->objectid
&&
7127 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7138 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7140 struct btrfs_block_group_cache
*block_group
;
7141 struct btrfs_space_info
*space_info
;
7142 struct btrfs_caching_control
*caching_ctl
;
7145 down_write(&info
->extent_commit_sem
);
7146 while (!list_empty(&info
->caching_block_groups
)) {
7147 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
7148 struct btrfs_caching_control
, list
);
7149 list_del(&caching_ctl
->list
);
7150 put_caching_control(caching_ctl
);
7152 up_write(&info
->extent_commit_sem
);
7154 spin_lock(&info
->block_group_cache_lock
);
7155 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7156 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7158 rb_erase(&block_group
->cache_node
,
7159 &info
->block_group_cache_tree
);
7160 spin_unlock(&info
->block_group_cache_lock
);
7162 down_write(&block_group
->space_info
->groups_sem
);
7163 list_del(&block_group
->list
);
7164 up_write(&block_group
->space_info
->groups_sem
);
7166 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7167 wait_block_group_cache_done(block_group
);
7169 btrfs_remove_free_space_cache(block_group
);
7171 WARN_ON(atomic_read(&block_group
->count
) != 1);
7174 spin_lock(&info
->block_group_cache_lock
);
7176 spin_unlock(&info
->block_group_cache_lock
);
7178 /* now that all the block groups are freed, go through and
7179 * free all the space_info structs. This is only called during
7180 * the final stages of unmount, and so we know nobody is
7181 * using them. We call synchronize_rcu() once before we start,
7182 * just to be on the safe side.
7186 while(!list_empty(&info
->space_info
)) {
7187 space_info
= list_entry(info
->space_info
.next
,
7188 struct btrfs_space_info
,
7191 list_del(&space_info
->list
);
7197 int btrfs_read_block_groups(struct btrfs_root
*root
)
7199 struct btrfs_path
*path
;
7201 struct btrfs_block_group_cache
*cache
;
7202 struct btrfs_fs_info
*info
= root
->fs_info
;
7203 struct btrfs_space_info
*space_info
;
7204 struct btrfs_key key
;
7205 struct btrfs_key found_key
;
7206 struct extent_buffer
*leaf
;
7208 root
= info
->extent_root
;
7211 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7212 path
= btrfs_alloc_path();
7217 ret
= find_first_block_group(root
, path
, &key
);
7225 leaf
= path
->nodes
[0];
7226 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7227 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7233 atomic_set(&cache
->count
, 1);
7234 spin_lock_init(&cache
->lock
);
7235 spin_lock_init(&cache
->tree_lock
);
7236 cache
->fs_info
= info
;
7237 INIT_LIST_HEAD(&cache
->list
);
7238 INIT_LIST_HEAD(&cache
->cluster_list
);
7241 * we only want to have 32k of ram per block group for keeping
7242 * track of free space, and if we pass 1/2 of that we want to
7243 * start converting things over to using bitmaps
7245 cache
->extents_thresh
= ((1024 * 32) / 2) /
7246 sizeof(struct btrfs_free_space
);
7248 read_extent_buffer(leaf
, &cache
->item
,
7249 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7250 sizeof(cache
->item
));
7251 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7253 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7254 btrfs_release_path(root
, path
);
7255 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7256 cache
->sectorsize
= root
->sectorsize
;
7259 * check for two cases, either we are full, and therefore
7260 * don't need to bother with the caching work since we won't
7261 * find any space, or we are empty, and we can just add all
7262 * the space in and be done with it. This saves us _alot_ of
7263 * time, particularly in the full case.
7265 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7266 exclude_super_stripes(root
, cache
);
7267 cache
->last_byte_to_unpin
= (u64
)-1;
7268 cache
->cached
= BTRFS_CACHE_FINISHED
;
7269 free_excluded_extents(root
, cache
);
7270 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7271 exclude_super_stripes(root
, cache
);
7272 cache
->last_byte_to_unpin
= (u64
)-1;
7273 cache
->cached
= BTRFS_CACHE_FINISHED
;
7274 add_new_free_space(cache
, root
->fs_info
,
7276 found_key
.objectid
+
7278 free_excluded_extents(root
, cache
);
7281 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7282 btrfs_block_group_used(&cache
->item
),
7285 cache
->space_info
= space_info
;
7286 spin_lock(&cache
->space_info
->lock
);
7287 cache
->space_info
->bytes_super
+= cache
->bytes_super
;
7288 spin_unlock(&cache
->space_info
->lock
);
7290 down_write(&space_info
->groups_sem
);
7291 list_add_tail(&cache
->list
, &space_info
->block_groups
);
7292 up_write(&space_info
->groups_sem
);
7294 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7297 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7298 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7299 set_block_group_readonly(cache
);
7303 btrfs_free_path(path
);
7307 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7308 struct btrfs_root
*root
, u64 bytes_used
,
7309 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7313 struct btrfs_root
*extent_root
;
7314 struct btrfs_block_group_cache
*cache
;
7316 extent_root
= root
->fs_info
->extent_root
;
7318 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7320 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7324 cache
->key
.objectid
= chunk_offset
;
7325 cache
->key
.offset
= size
;
7326 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7327 cache
->sectorsize
= root
->sectorsize
;
7330 * we only want to have 32k of ram per block group for keeping track
7331 * of free space, and if we pass 1/2 of that we want to start
7332 * converting things over to using bitmaps
7334 cache
->extents_thresh
= ((1024 * 32) / 2) /
7335 sizeof(struct btrfs_free_space
);
7336 atomic_set(&cache
->count
, 1);
7337 spin_lock_init(&cache
->lock
);
7338 spin_lock_init(&cache
->tree_lock
);
7339 INIT_LIST_HEAD(&cache
->list
);
7340 INIT_LIST_HEAD(&cache
->cluster_list
);
7342 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7343 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7344 cache
->flags
= type
;
7345 btrfs_set_block_group_flags(&cache
->item
, type
);
7347 cache
->last_byte_to_unpin
= (u64
)-1;
7348 cache
->cached
= BTRFS_CACHE_FINISHED
;
7349 exclude_super_stripes(root
, cache
);
7351 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7352 chunk_offset
+ size
);
7354 free_excluded_extents(root
, cache
);
7356 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7357 &cache
->space_info
);
7360 spin_lock(&cache
->space_info
->lock
);
7361 cache
->space_info
->bytes_super
+= cache
->bytes_super
;
7362 spin_unlock(&cache
->space_info
->lock
);
7364 down_write(&cache
->space_info
->groups_sem
);
7365 list_add_tail(&cache
->list
, &cache
->space_info
->block_groups
);
7366 up_write(&cache
->space_info
->groups_sem
);
7368 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7371 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7372 sizeof(cache
->item
));
7375 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7380 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7381 struct btrfs_root
*root
, u64 group_start
)
7383 struct btrfs_path
*path
;
7384 struct btrfs_block_group_cache
*block_group
;
7385 struct btrfs_free_cluster
*cluster
;
7386 struct btrfs_key key
;
7389 root
= root
->fs_info
->extent_root
;
7391 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7392 BUG_ON(!block_group
);
7393 BUG_ON(!block_group
->ro
);
7395 memcpy(&key
, &block_group
->key
, sizeof(key
));
7397 /* make sure this block group isn't part of an allocation cluster */
7398 cluster
= &root
->fs_info
->data_alloc_cluster
;
7399 spin_lock(&cluster
->refill_lock
);
7400 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7401 spin_unlock(&cluster
->refill_lock
);
7404 * make sure this block group isn't part of a metadata
7405 * allocation cluster
7407 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7408 spin_lock(&cluster
->refill_lock
);
7409 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7410 spin_unlock(&cluster
->refill_lock
);
7412 path
= btrfs_alloc_path();
7415 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7416 rb_erase(&block_group
->cache_node
,
7417 &root
->fs_info
->block_group_cache_tree
);
7418 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7420 down_write(&block_group
->space_info
->groups_sem
);
7422 * we must use list_del_init so people can check to see if they
7423 * are still on the list after taking the semaphore
7425 list_del_init(&block_group
->list
);
7426 up_write(&block_group
->space_info
->groups_sem
);
7428 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7429 wait_block_group_cache_done(block_group
);
7431 btrfs_remove_free_space_cache(block_group
);
7433 spin_lock(&block_group
->space_info
->lock
);
7434 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7435 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7436 spin_unlock(&block_group
->space_info
->lock
);
7438 btrfs_clear_space_info_full(root
->fs_info
);
7440 btrfs_put_block_group(block_group
);
7441 btrfs_put_block_group(block_group
);
7443 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7449 ret
= btrfs_del_item(trans
, root
, path
);
7451 btrfs_free_path(path
);