2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
39 #undef SCRAMBLE_DELAYED_REFS
42 * control flags for do_chunk_alloc's force field
43 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
46 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
52 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 CHUNK_ALLOC_NO_FORCE
= 0,
57 CHUNK_ALLOC_LIMITED
= 1,
58 CHUNK_ALLOC_FORCE
= 2,
62 * Control how reservations are dealt with.
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
73 RESERVE_ALLOC_NO_ACCOUNT
= 2,
76 static int update_block_group(struct btrfs_root
*root
,
77 u64 bytenr
, u64 num_bytes
, int alloc
);
78 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
79 struct btrfs_root
*root
,
80 u64 bytenr
, u64 num_bytes
, u64 parent
,
81 u64 root_objectid
, u64 owner_objectid
,
82 u64 owner_offset
, int refs_to_drop
,
83 struct btrfs_delayed_extent_op
*extra_op
);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
85 struct extent_buffer
*leaf
,
86 struct btrfs_extent_item
*ei
);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
88 struct btrfs_root
*root
,
89 u64 parent
, u64 root_objectid
,
90 u64 flags
, u64 owner
, u64 offset
,
91 struct btrfs_key
*ins
, int ref_mod
);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
93 struct btrfs_root
*root
,
94 u64 parent
, u64 root_objectid
,
95 u64 flags
, struct btrfs_disk_key
*key
,
96 int level
, struct btrfs_key
*ins
);
97 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
98 struct btrfs_root
*extent_root
, u64 flags
,
100 static int find_next_key(struct btrfs_path
*path
, int level
,
101 struct btrfs_key
*key
);
102 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
103 int dump_block_groups
);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
105 u64 num_bytes
, int reserve
);
106 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
108 int btrfs_pin_extent(struct btrfs_root
*root
,
109 u64 bytenr
, u64 num_bytes
, int reserved
);
112 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
115 return cache
->cached
== BTRFS_CACHE_FINISHED
||
116 cache
->cached
== BTRFS_CACHE_ERROR
;
119 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
121 return (cache
->flags
& bits
) == bits
;
124 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
126 atomic_inc(&cache
->count
);
129 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
131 if (atomic_dec_and_test(&cache
->count
)) {
132 WARN_ON(cache
->pinned
> 0);
133 WARN_ON(cache
->reserved
> 0);
134 kfree(cache
->free_space_ctl
);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
144 struct btrfs_block_group_cache
*block_group
)
147 struct rb_node
*parent
= NULL
;
148 struct btrfs_block_group_cache
*cache
;
150 spin_lock(&info
->block_group_cache_lock
);
151 p
= &info
->block_group_cache_tree
.rb_node
;
155 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
157 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
159 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
162 spin_unlock(&info
->block_group_cache_lock
);
167 rb_link_node(&block_group
->cache_node
, parent
, p
);
168 rb_insert_color(&block_group
->cache_node
,
169 &info
->block_group_cache_tree
);
171 if (info
->first_logical_byte
> block_group
->key
.objectid
)
172 info
->first_logical_byte
= block_group
->key
.objectid
;
174 spin_unlock(&info
->block_group_cache_lock
);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache
*
184 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
187 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
191 spin_lock(&info
->block_group_cache_lock
);
192 n
= info
->block_group_cache_tree
.rb_node
;
195 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
197 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
198 start
= cache
->key
.objectid
;
200 if (bytenr
< start
) {
201 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
204 } else if (bytenr
> start
) {
205 if (contains
&& bytenr
<= end
) {
216 btrfs_get_block_group(ret
);
217 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
218 info
->first_logical_byte
= ret
->key
.objectid
;
220 spin_unlock(&info
->block_group_cache_lock
);
225 static int add_excluded_extent(struct btrfs_root
*root
,
226 u64 start
, u64 num_bytes
)
228 u64 end
= start
+ num_bytes
- 1;
229 set_extent_bits(&root
->fs_info
->freed_extents
[0],
230 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
231 set_extent_bits(&root
->fs_info
->freed_extents
[1],
232 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
236 static void free_excluded_extents(struct btrfs_root
*root
,
237 struct btrfs_block_group_cache
*cache
)
241 start
= cache
->key
.objectid
;
242 end
= start
+ cache
->key
.offset
- 1;
244 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
245 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
246 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
247 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
250 static int exclude_super_stripes(struct btrfs_root
*root
,
251 struct btrfs_block_group_cache
*cache
)
258 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
259 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
260 cache
->bytes_super
+= stripe_len
;
261 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
267 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
268 bytenr
= btrfs_sb_offset(i
);
269 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
270 cache
->key
.objectid
, bytenr
,
271 0, &logical
, &nr
, &stripe_len
);
278 if (logical
[nr
] > cache
->key
.objectid
+
282 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
286 if (start
< cache
->key
.objectid
) {
287 start
= cache
->key
.objectid
;
288 len
= (logical
[nr
] + stripe_len
) - start
;
290 len
= min_t(u64
, stripe_len
,
291 cache
->key
.objectid
+
292 cache
->key
.offset
- start
);
295 cache
->bytes_super
+= len
;
296 ret
= add_excluded_extent(root
, start
, len
);
308 static struct btrfs_caching_control
*
309 get_caching_control(struct btrfs_block_group_cache
*cache
)
311 struct btrfs_caching_control
*ctl
;
313 spin_lock(&cache
->lock
);
314 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
315 spin_unlock(&cache
->lock
);
319 /* We're loading it the fast way, so we don't have a caching_ctl. */
320 if (!cache
->caching_ctl
) {
321 spin_unlock(&cache
->lock
);
325 ctl
= cache
->caching_ctl
;
326 atomic_inc(&ctl
->count
);
327 spin_unlock(&cache
->lock
);
331 static void put_caching_control(struct btrfs_caching_control
*ctl
)
333 if (atomic_dec_and_test(&ctl
->count
))
338 * this is only called by cache_block_group, since we could have freed extents
339 * we need to check the pinned_extents for any extents that can't be used yet
340 * since their free space will be released as soon as the transaction commits.
342 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
343 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
345 u64 extent_start
, extent_end
, size
, total_added
= 0;
348 while (start
< end
) {
349 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
350 &extent_start
, &extent_end
,
351 EXTENT_DIRTY
| EXTENT_UPTODATE
,
356 if (extent_start
<= start
) {
357 start
= extent_end
+ 1;
358 } else if (extent_start
> start
&& extent_start
< end
) {
359 size
= extent_start
- start
;
361 ret
= btrfs_add_free_space(block_group
, start
,
363 BUG_ON(ret
); /* -ENOMEM or logic error */
364 start
= extent_end
+ 1;
373 ret
= btrfs_add_free_space(block_group
, start
, size
);
374 BUG_ON(ret
); /* -ENOMEM or logic error */
380 static noinline
void caching_thread(struct btrfs_work
*work
)
382 struct btrfs_block_group_cache
*block_group
;
383 struct btrfs_fs_info
*fs_info
;
384 struct btrfs_caching_control
*caching_ctl
;
385 struct btrfs_root
*extent_root
;
386 struct btrfs_path
*path
;
387 struct extent_buffer
*leaf
;
388 struct btrfs_key key
;
394 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
395 block_group
= caching_ctl
->block_group
;
396 fs_info
= block_group
->fs_info
;
397 extent_root
= fs_info
->extent_root
;
399 path
= btrfs_alloc_path();
403 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
406 * We don't want to deadlock with somebody trying to allocate a new
407 * extent for the extent root while also trying to search the extent
408 * root to add free space. So we skip locking and search the commit
409 * root, since its read-only
411 path
->skip_locking
= 1;
412 path
->search_commit_root
= 1;
417 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
419 mutex_lock(&caching_ctl
->mutex
);
420 /* need to make sure the commit_root doesn't disappear */
421 down_read(&fs_info
->extent_commit_sem
);
424 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
428 leaf
= path
->nodes
[0];
429 nritems
= btrfs_header_nritems(leaf
);
432 if (btrfs_fs_closing(fs_info
) > 1) {
437 if (path
->slots
[0] < nritems
) {
438 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
440 ret
= find_next_key(path
, 0, &key
);
444 if (need_resched()) {
445 caching_ctl
->progress
= last
;
446 btrfs_release_path(path
);
447 up_read(&fs_info
->extent_commit_sem
);
448 mutex_unlock(&caching_ctl
->mutex
);
453 ret
= btrfs_next_leaf(extent_root
, path
);
458 leaf
= path
->nodes
[0];
459 nritems
= btrfs_header_nritems(leaf
);
463 if (key
.objectid
< last
) {
466 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
468 caching_ctl
->progress
= last
;
469 btrfs_release_path(path
);
473 if (key
.objectid
< block_group
->key
.objectid
) {
478 if (key
.objectid
>= block_group
->key
.objectid
+
479 block_group
->key
.offset
)
482 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
483 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
484 total_found
+= add_new_free_space(block_group
,
487 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
488 last
= key
.objectid
+
489 fs_info
->tree_root
->leafsize
;
491 last
= key
.objectid
+ key
.offset
;
493 if (total_found
> (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl
->wait
);
502 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
503 block_group
->key
.objectid
+
504 block_group
->key
.offset
);
505 caching_ctl
->progress
= (u64
)-1;
507 spin_lock(&block_group
->lock
);
508 block_group
->caching_ctl
= NULL
;
509 block_group
->cached
= BTRFS_CACHE_FINISHED
;
510 spin_unlock(&block_group
->lock
);
513 btrfs_free_path(path
);
514 up_read(&fs_info
->extent_commit_sem
);
516 free_excluded_extents(extent_root
, block_group
);
518 mutex_unlock(&caching_ctl
->mutex
);
521 spin_lock(&block_group
->lock
);
522 block_group
->caching_ctl
= NULL
;
523 block_group
->cached
= BTRFS_CACHE_ERROR
;
524 spin_unlock(&block_group
->lock
);
526 wake_up(&caching_ctl
->wait
);
528 put_caching_control(caching_ctl
);
529 btrfs_put_block_group(block_group
);
532 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
536 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
537 struct btrfs_caching_control
*caching_ctl
;
540 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
544 INIT_LIST_HEAD(&caching_ctl
->list
);
545 mutex_init(&caching_ctl
->mutex
);
546 init_waitqueue_head(&caching_ctl
->wait
);
547 caching_ctl
->block_group
= cache
;
548 caching_ctl
->progress
= cache
->key
.objectid
;
549 atomic_set(&caching_ctl
->count
, 1);
550 caching_ctl
->work
.func
= caching_thread
;
552 spin_lock(&cache
->lock
);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache
->cached
== BTRFS_CACHE_FAST
) {
566 struct btrfs_caching_control
*ctl
;
568 ctl
= cache
->caching_ctl
;
569 atomic_inc(&ctl
->count
);
570 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
571 spin_unlock(&cache
->lock
);
575 finish_wait(&ctl
->wait
, &wait
);
576 put_caching_control(ctl
);
577 spin_lock(&cache
->lock
);
580 if (cache
->cached
!= BTRFS_CACHE_NO
) {
581 spin_unlock(&cache
->lock
);
585 WARN_ON(cache
->caching_ctl
);
586 cache
->caching_ctl
= caching_ctl
;
587 cache
->cached
= BTRFS_CACHE_FAST
;
588 spin_unlock(&cache
->lock
);
590 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
591 ret
= load_free_space_cache(fs_info
, cache
);
593 spin_lock(&cache
->lock
);
595 cache
->caching_ctl
= NULL
;
596 cache
->cached
= BTRFS_CACHE_FINISHED
;
597 cache
->last_byte_to_unpin
= (u64
)-1;
599 if (load_cache_only
) {
600 cache
->caching_ctl
= NULL
;
601 cache
->cached
= BTRFS_CACHE_NO
;
603 cache
->cached
= BTRFS_CACHE_STARTED
;
606 spin_unlock(&cache
->lock
);
607 wake_up(&caching_ctl
->wait
);
609 put_caching_control(caching_ctl
);
610 free_excluded_extents(fs_info
->extent_root
, cache
);
615 * We are not going to do the fast caching, set cached to the
616 * appropriate value and wakeup any waiters.
618 spin_lock(&cache
->lock
);
619 if (load_cache_only
) {
620 cache
->caching_ctl
= NULL
;
621 cache
->cached
= BTRFS_CACHE_NO
;
623 cache
->cached
= BTRFS_CACHE_STARTED
;
625 spin_unlock(&cache
->lock
);
626 wake_up(&caching_ctl
->wait
);
629 if (load_cache_only
) {
630 put_caching_control(caching_ctl
);
634 down_write(&fs_info
->extent_commit_sem
);
635 atomic_inc(&caching_ctl
->count
);
636 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
637 up_write(&fs_info
->extent_commit_sem
);
639 btrfs_get_block_group(cache
);
641 btrfs_queue_worker(&fs_info
->caching_workers
, &caching_ctl
->work
);
647 * return the block group that starts at or after bytenr
649 static struct btrfs_block_group_cache
*
650 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
652 struct btrfs_block_group_cache
*cache
;
654 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
660 * return the block group that contains the given bytenr
662 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
663 struct btrfs_fs_info
*info
,
666 struct btrfs_block_group_cache
*cache
;
668 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
673 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
676 struct list_head
*head
= &info
->space_info
;
677 struct btrfs_space_info
*found
;
679 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
682 list_for_each_entry_rcu(found
, head
, list
) {
683 if (found
->flags
& flags
) {
693 * after adding space to the filesystem, we need to clear the full flags
694 * on all the space infos.
696 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
698 struct list_head
*head
= &info
->space_info
;
699 struct btrfs_space_info
*found
;
702 list_for_each_entry_rcu(found
, head
, list
)
707 /* simple helper to search for an existing extent at a given offset */
708 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
711 struct btrfs_key key
;
712 struct btrfs_path
*path
;
714 path
= btrfs_alloc_path();
718 key
.objectid
= start
;
720 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
721 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
724 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
725 if (key
.objectid
== start
&&
726 key
.type
== BTRFS_METADATA_ITEM_KEY
)
729 btrfs_free_path(path
);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
743 struct btrfs_root
*root
, u64 bytenr
,
744 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
746 struct btrfs_delayed_ref_head
*head
;
747 struct btrfs_delayed_ref_root
*delayed_refs
;
748 struct btrfs_path
*path
;
749 struct btrfs_extent_item
*ei
;
750 struct extent_buffer
*leaf
;
751 struct btrfs_key key
;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
762 offset
= root
->leafsize
;
766 path
= btrfs_alloc_path();
771 path
->skip_locking
= 1;
772 path
->search_commit_root
= 1;
776 key
.objectid
= bytenr
;
779 key
.type
= BTRFS_METADATA_ITEM_KEY
;
781 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
784 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
789 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
790 if (path
->slots
[0]) {
792 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
794 if (key
.objectid
== bytenr
&&
795 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
796 key
.offset
== root
->leafsize
)
800 key
.objectid
= bytenr
;
801 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
802 key
.offset
= root
->leafsize
;
803 btrfs_release_path(path
);
809 leaf
= path
->nodes
[0];
810 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
811 if (item_size
>= sizeof(*ei
)) {
812 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
813 struct btrfs_extent_item
);
814 num_refs
= btrfs_extent_refs(leaf
, ei
);
815 extent_flags
= btrfs_extent_flags(leaf
, ei
);
817 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
818 struct btrfs_extent_item_v0
*ei0
;
819 BUG_ON(item_size
!= sizeof(*ei0
));
820 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
821 struct btrfs_extent_item_v0
);
822 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
823 /* FIXME: this isn't correct for data */
824 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
829 BUG_ON(num_refs
== 0);
839 delayed_refs
= &trans
->transaction
->delayed_refs
;
840 spin_lock(&delayed_refs
->lock
);
841 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
843 if (!mutex_trylock(&head
->mutex
)) {
844 atomic_inc(&head
->node
.refs
);
845 spin_unlock(&delayed_refs
->lock
);
847 btrfs_release_path(path
);
850 * Mutex was contended, block until it's released and try
853 mutex_lock(&head
->mutex
);
854 mutex_unlock(&head
->mutex
);
855 btrfs_put_delayed_ref(&head
->node
);
858 if (head
->extent_op
&& head
->extent_op
->update_flags
)
859 extent_flags
|= head
->extent_op
->flags_to_set
;
861 BUG_ON(num_refs
== 0);
863 num_refs
+= head
->node
.ref_mod
;
864 mutex_unlock(&head
->mutex
);
866 spin_unlock(&delayed_refs
->lock
);
868 WARN_ON(num_refs
== 0);
872 *flags
= extent_flags
;
874 btrfs_free_path(path
);
879 * Back reference rules. Back refs have three main goals:
881 * 1) differentiate between all holders of references to an extent so that
882 * when a reference is dropped we can make sure it was a valid reference
883 * before freeing the extent.
885 * 2) Provide enough information to quickly find the holders of an extent
886 * if we notice a given block is corrupted or bad.
888 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
889 * maintenance. This is actually the same as #2, but with a slightly
890 * different use case.
892 * There are two kinds of back refs. The implicit back refs is optimized
893 * for pointers in non-shared tree blocks. For a given pointer in a block,
894 * back refs of this kind provide information about the block's owner tree
895 * and the pointer's key. These information allow us to find the block by
896 * b-tree searching. The full back refs is for pointers in tree blocks not
897 * referenced by their owner trees. The location of tree block is recorded
898 * in the back refs. Actually the full back refs is generic, and can be
899 * used in all cases the implicit back refs is used. The major shortcoming
900 * of the full back refs is its overhead. Every time a tree block gets
901 * COWed, we have to update back refs entry for all pointers in it.
903 * For a newly allocated tree block, we use implicit back refs for
904 * pointers in it. This means most tree related operations only involve
905 * implicit back refs. For a tree block created in old transaction, the
906 * only way to drop a reference to it is COW it. So we can detect the
907 * event that tree block loses its owner tree's reference and do the
908 * back refs conversion.
910 * When a tree block is COW'd through a tree, there are four cases:
912 * The reference count of the block is one and the tree is the block's
913 * owner tree. Nothing to do in this case.
915 * The reference count of the block is one and the tree is not the
916 * block's owner tree. In this case, full back refs is used for pointers
917 * in the block. Remove these full back refs, add implicit back refs for
918 * every pointers in the new block.
920 * The reference count of the block is greater than one and the tree is
921 * the block's owner tree. In this case, implicit back refs is used for
922 * pointers in the block. Add full back refs for every pointers in the
923 * block, increase lower level extents' reference counts. The original
924 * implicit back refs are entailed to the new block.
926 * The reference count of the block is greater than one and the tree is
927 * not the block's owner tree. Add implicit back refs for every pointer in
928 * the new block, increase lower level extents' reference count.
930 * Back Reference Key composing:
932 * The key objectid corresponds to the first byte in the extent,
933 * The key type is used to differentiate between types of back refs.
934 * There are different meanings of the key offset for different types
937 * File extents can be referenced by:
939 * - multiple snapshots, subvolumes, or different generations in one subvol
940 * - different files inside a single subvolume
941 * - different offsets inside a file (bookend extents in file.c)
943 * The extent ref structure for the implicit back refs has fields for:
945 * - Objectid of the subvolume root
946 * - objectid of the file holding the reference
947 * - original offset in the file
948 * - how many bookend extents
950 * The key offset for the implicit back refs is hash of the first
953 * The extent ref structure for the full back refs has field for:
955 * - number of pointers in the tree leaf
957 * The key offset for the implicit back refs is the first byte of
960 * When a file extent is allocated, The implicit back refs is used.
961 * the fields are filled in:
963 * (root_key.objectid, inode objectid, offset in file, 1)
965 * When a file extent is removed file truncation, we find the
966 * corresponding implicit back refs and check the following fields:
968 * (btrfs_header_owner(leaf), inode objectid, offset in file)
970 * Btree extents can be referenced by:
972 * - Different subvolumes
974 * Both the implicit back refs and the full back refs for tree blocks
975 * only consist of key. The key offset for the implicit back refs is
976 * objectid of block's owner tree. The key offset for the full back refs
977 * is the first byte of parent block.
979 * When implicit back refs is used, information about the lowest key and
980 * level of the tree block are required. These information are stored in
981 * tree block info structure.
984 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
985 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
986 struct btrfs_root
*root
,
987 struct btrfs_path
*path
,
988 u64 owner
, u32 extra_size
)
990 struct btrfs_extent_item
*item
;
991 struct btrfs_extent_item_v0
*ei0
;
992 struct btrfs_extent_ref_v0
*ref0
;
993 struct btrfs_tree_block_info
*bi
;
994 struct extent_buffer
*leaf
;
995 struct btrfs_key key
;
996 struct btrfs_key found_key
;
997 u32 new_size
= sizeof(*item
);
1001 leaf
= path
->nodes
[0];
1002 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1004 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1005 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1006 struct btrfs_extent_item_v0
);
1007 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1009 if (owner
== (u64
)-1) {
1011 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1012 ret
= btrfs_next_leaf(root
, path
);
1015 BUG_ON(ret
> 0); /* Corruption */
1016 leaf
= path
->nodes
[0];
1018 btrfs_item_key_to_cpu(leaf
, &found_key
,
1020 BUG_ON(key
.objectid
!= found_key
.objectid
);
1021 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1025 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1026 struct btrfs_extent_ref_v0
);
1027 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1031 btrfs_release_path(path
);
1033 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1034 new_size
+= sizeof(*bi
);
1036 new_size
-= sizeof(*ei0
);
1037 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1038 new_size
+ extra_size
, 1);
1041 BUG_ON(ret
); /* Corruption */
1043 btrfs_extend_item(root
, path
, new_size
);
1045 leaf
= path
->nodes
[0];
1046 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1047 btrfs_set_extent_refs(leaf
, item
, refs
);
1048 /* FIXME: get real generation */
1049 btrfs_set_extent_generation(leaf
, item
, 0);
1050 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1051 btrfs_set_extent_flags(leaf
, item
,
1052 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1053 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1054 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1055 /* FIXME: get first key of the block */
1056 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1057 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1059 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1061 btrfs_mark_buffer_dirty(leaf
);
1066 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1068 u32 high_crc
= ~(u32
)0;
1069 u32 low_crc
= ~(u32
)0;
1072 lenum
= cpu_to_le64(root_objectid
);
1073 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1074 lenum
= cpu_to_le64(owner
);
1075 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1076 lenum
= cpu_to_le64(offset
);
1077 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1079 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1082 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1083 struct btrfs_extent_data_ref
*ref
)
1085 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1086 btrfs_extent_data_ref_objectid(leaf
, ref
),
1087 btrfs_extent_data_ref_offset(leaf
, ref
));
1090 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1091 struct btrfs_extent_data_ref
*ref
,
1092 u64 root_objectid
, u64 owner
, u64 offset
)
1094 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1095 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1096 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1101 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1102 struct btrfs_root
*root
,
1103 struct btrfs_path
*path
,
1104 u64 bytenr
, u64 parent
,
1106 u64 owner
, u64 offset
)
1108 struct btrfs_key key
;
1109 struct btrfs_extent_data_ref
*ref
;
1110 struct extent_buffer
*leaf
;
1116 key
.objectid
= bytenr
;
1118 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1119 key
.offset
= parent
;
1121 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1122 key
.offset
= hash_extent_data_ref(root_objectid
,
1127 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1136 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1137 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1138 btrfs_release_path(path
);
1139 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1150 leaf
= path
->nodes
[0];
1151 nritems
= btrfs_header_nritems(leaf
);
1153 if (path
->slots
[0] >= nritems
) {
1154 ret
= btrfs_next_leaf(root
, path
);
1160 leaf
= path
->nodes
[0];
1161 nritems
= btrfs_header_nritems(leaf
);
1165 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1166 if (key
.objectid
!= bytenr
||
1167 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1170 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1171 struct btrfs_extent_data_ref
);
1173 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1176 btrfs_release_path(path
);
1188 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1189 struct btrfs_root
*root
,
1190 struct btrfs_path
*path
,
1191 u64 bytenr
, u64 parent
,
1192 u64 root_objectid
, u64 owner
,
1193 u64 offset
, int refs_to_add
)
1195 struct btrfs_key key
;
1196 struct extent_buffer
*leaf
;
1201 key
.objectid
= bytenr
;
1203 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1204 key
.offset
= parent
;
1205 size
= sizeof(struct btrfs_shared_data_ref
);
1207 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1208 key
.offset
= hash_extent_data_ref(root_objectid
,
1210 size
= sizeof(struct btrfs_extent_data_ref
);
1213 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1214 if (ret
&& ret
!= -EEXIST
)
1217 leaf
= path
->nodes
[0];
1219 struct btrfs_shared_data_ref
*ref
;
1220 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1221 struct btrfs_shared_data_ref
);
1223 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1225 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1226 num_refs
+= refs_to_add
;
1227 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1230 struct btrfs_extent_data_ref
*ref
;
1231 while (ret
== -EEXIST
) {
1232 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1233 struct btrfs_extent_data_ref
);
1234 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1237 btrfs_release_path(path
);
1239 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1241 if (ret
&& ret
!= -EEXIST
)
1244 leaf
= path
->nodes
[0];
1246 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1247 struct btrfs_extent_data_ref
);
1249 btrfs_set_extent_data_ref_root(leaf
, ref
,
1251 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1252 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1253 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1255 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1256 num_refs
+= refs_to_add
;
1257 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1260 btrfs_mark_buffer_dirty(leaf
);
1263 btrfs_release_path(path
);
1267 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1268 struct btrfs_root
*root
,
1269 struct btrfs_path
*path
,
1272 struct btrfs_key key
;
1273 struct btrfs_extent_data_ref
*ref1
= NULL
;
1274 struct btrfs_shared_data_ref
*ref2
= NULL
;
1275 struct extent_buffer
*leaf
;
1279 leaf
= path
->nodes
[0];
1280 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1282 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1283 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1284 struct btrfs_extent_data_ref
);
1285 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1286 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1287 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1288 struct btrfs_shared_data_ref
);
1289 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1290 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1291 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1292 struct btrfs_extent_ref_v0
*ref0
;
1293 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1294 struct btrfs_extent_ref_v0
);
1295 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1301 BUG_ON(num_refs
< refs_to_drop
);
1302 num_refs
-= refs_to_drop
;
1304 if (num_refs
== 0) {
1305 ret
= btrfs_del_item(trans
, root
, path
);
1307 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1308 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1309 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1310 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1311 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1313 struct btrfs_extent_ref_v0
*ref0
;
1314 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1315 struct btrfs_extent_ref_v0
);
1316 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1319 btrfs_mark_buffer_dirty(leaf
);
1324 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1325 struct btrfs_path
*path
,
1326 struct btrfs_extent_inline_ref
*iref
)
1328 struct btrfs_key key
;
1329 struct extent_buffer
*leaf
;
1330 struct btrfs_extent_data_ref
*ref1
;
1331 struct btrfs_shared_data_ref
*ref2
;
1334 leaf
= path
->nodes
[0];
1335 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1337 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1338 BTRFS_EXTENT_DATA_REF_KEY
) {
1339 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1340 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1342 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1343 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1345 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1346 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1347 struct btrfs_extent_data_ref
);
1348 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1349 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1350 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1351 struct btrfs_shared_data_ref
);
1352 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1353 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1354 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1355 struct btrfs_extent_ref_v0
*ref0
;
1356 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1357 struct btrfs_extent_ref_v0
);
1358 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1366 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1367 struct btrfs_root
*root
,
1368 struct btrfs_path
*path
,
1369 u64 bytenr
, u64 parent
,
1372 struct btrfs_key key
;
1375 key
.objectid
= bytenr
;
1377 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1378 key
.offset
= parent
;
1380 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1381 key
.offset
= root_objectid
;
1384 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1387 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1388 if (ret
== -ENOENT
&& parent
) {
1389 btrfs_release_path(path
);
1390 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1391 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1399 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1400 struct btrfs_root
*root
,
1401 struct btrfs_path
*path
,
1402 u64 bytenr
, u64 parent
,
1405 struct btrfs_key key
;
1408 key
.objectid
= bytenr
;
1410 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1411 key
.offset
= parent
;
1413 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1414 key
.offset
= root_objectid
;
1417 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1418 btrfs_release_path(path
);
1422 static inline int extent_ref_type(u64 parent
, u64 owner
)
1425 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1427 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1429 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1432 type
= BTRFS_SHARED_DATA_REF_KEY
;
1434 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1439 static int find_next_key(struct btrfs_path
*path
, int level
,
1440 struct btrfs_key
*key
)
1443 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1444 if (!path
->nodes
[level
])
1446 if (path
->slots
[level
] + 1 >=
1447 btrfs_header_nritems(path
->nodes
[level
]))
1450 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1451 path
->slots
[level
] + 1);
1453 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1454 path
->slots
[level
] + 1);
1461 * look for inline back ref. if back ref is found, *ref_ret is set
1462 * to the address of inline back ref, and 0 is returned.
1464 * if back ref isn't found, *ref_ret is set to the address where it
1465 * should be inserted, and -ENOENT is returned.
1467 * if insert is true and there are too many inline back refs, the path
1468 * points to the extent item, and -EAGAIN is returned.
1470 * NOTE: inline back refs are ordered in the same way that back ref
1471 * items in the tree are ordered.
1473 static noinline_for_stack
1474 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1475 struct btrfs_root
*root
,
1476 struct btrfs_path
*path
,
1477 struct btrfs_extent_inline_ref
**ref_ret
,
1478 u64 bytenr
, u64 num_bytes
,
1479 u64 parent
, u64 root_objectid
,
1480 u64 owner
, u64 offset
, int insert
)
1482 struct btrfs_key key
;
1483 struct extent_buffer
*leaf
;
1484 struct btrfs_extent_item
*ei
;
1485 struct btrfs_extent_inline_ref
*iref
;
1495 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1498 key
.objectid
= bytenr
;
1499 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1500 key
.offset
= num_bytes
;
1502 want
= extent_ref_type(parent
, owner
);
1504 extra_size
= btrfs_extent_inline_ref_size(want
);
1505 path
->keep_locks
= 1;
1510 * Owner is our parent level, so we can just add one to get the level
1511 * for the block we are interested in.
1513 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1514 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1519 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1526 * We may be a newly converted file system which still has the old fat
1527 * extent entries for metadata, so try and see if we have one of those.
1529 if (ret
> 0 && skinny_metadata
) {
1530 skinny_metadata
= false;
1531 if (path
->slots
[0]) {
1533 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1535 if (key
.objectid
== bytenr
&&
1536 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1537 key
.offset
== num_bytes
)
1541 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1542 key
.offset
= num_bytes
;
1543 btrfs_release_path(path
);
1548 if (ret
&& !insert
) {
1551 } else if (WARN_ON(ret
)) {
1556 leaf
= path
->nodes
[0];
1557 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1558 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1559 if (item_size
< sizeof(*ei
)) {
1564 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1570 leaf
= path
->nodes
[0];
1571 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1574 BUG_ON(item_size
< sizeof(*ei
));
1576 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1577 flags
= btrfs_extent_flags(leaf
, ei
);
1579 ptr
= (unsigned long)(ei
+ 1);
1580 end
= (unsigned long)ei
+ item_size
;
1582 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1583 ptr
+= sizeof(struct btrfs_tree_block_info
);
1593 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1594 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1598 ptr
+= btrfs_extent_inline_ref_size(type
);
1602 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1603 struct btrfs_extent_data_ref
*dref
;
1604 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1605 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1610 if (hash_extent_data_ref_item(leaf
, dref
) <
1611 hash_extent_data_ref(root_objectid
, owner
, offset
))
1615 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1617 if (parent
== ref_offset
) {
1621 if (ref_offset
< parent
)
1624 if (root_objectid
== ref_offset
) {
1628 if (ref_offset
< root_objectid
)
1632 ptr
+= btrfs_extent_inline_ref_size(type
);
1634 if (err
== -ENOENT
&& insert
) {
1635 if (item_size
+ extra_size
>=
1636 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1641 * To add new inline back ref, we have to make sure
1642 * there is no corresponding back ref item.
1643 * For simplicity, we just do not add new inline back
1644 * ref if there is any kind of item for this block
1646 if (find_next_key(path
, 0, &key
) == 0 &&
1647 key
.objectid
== bytenr
&&
1648 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1653 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1656 path
->keep_locks
= 0;
1657 btrfs_unlock_up_safe(path
, 1);
1663 * helper to add new inline back ref
1665 static noinline_for_stack
1666 void setup_inline_extent_backref(struct btrfs_root
*root
,
1667 struct btrfs_path
*path
,
1668 struct btrfs_extent_inline_ref
*iref
,
1669 u64 parent
, u64 root_objectid
,
1670 u64 owner
, u64 offset
, int refs_to_add
,
1671 struct btrfs_delayed_extent_op
*extent_op
)
1673 struct extent_buffer
*leaf
;
1674 struct btrfs_extent_item
*ei
;
1677 unsigned long item_offset
;
1682 leaf
= path
->nodes
[0];
1683 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1684 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1686 type
= extent_ref_type(parent
, owner
);
1687 size
= btrfs_extent_inline_ref_size(type
);
1689 btrfs_extend_item(root
, path
, size
);
1691 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1692 refs
= btrfs_extent_refs(leaf
, ei
);
1693 refs
+= refs_to_add
;
1694 btrfs_set_extent_refs(leaf
, ei
, refs
);
1696 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1698 ptr
= (unsigned long)ei
+ item_offset
;
1699 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1700 if (ptr
< end
- size
)
1701 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1704 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1705 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1706 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1707 struct btrfs_extent_data_ref
*dref
;
1708 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1709 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1710 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1711 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1712 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1713 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1714 struct btrfs_shared_data_ref
*sref
;
1715 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1716 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1717 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1718 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1719 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1721 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1723 btrfs_mark_buffer_dirty(leaf
);
1726 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1727 struct btrfs_root
*root
,
1728 struct btrfs_path
*path
,
1729 struct btrfs_extent_inline_ref
**ref_ret
,
1730 u64 bytenr
, u64 num_bytes
, u64 parent
,
1731 u64 root_objectid
, u64 owner
, u64 offset
)
1735 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1736 bytenr
, num_bytes
, parent
,
1737 root_objectid
, owner
, offset
, 0);
1741 btrfs_release_path(path
);
1744 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1745 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1748 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1749 root_objectid
, owner
, offset
);
1755 * helper to update/remove inline back ref
1757 static noinline_for_stack
1758 void update_inline_extent_backref(struct btrfs_root
*root
,
1759 struct btrfs_path
*path
,
1760 struct btrfs_extent_inline_ref
*iref
,
1762 struct btrfs_delayed_extent_op
*extent_op
)
1764 struct extent_buffer
*leaf
;
1765 struct btrfs_extent_item
*ei
;
1766 struct btrfs_extent_data_ref
*dref
= NULL
;
1767 struct btrfs_shared_data_ref
*sref
= NULL
;
1775 leaf
= path
->nodes
[0];
1776 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1777 refs
= btrfs_extent_refs(leaf
, ei
);
1778 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1779 refs
+= refs_to_mod
;
1780 btrfs_set_extent_refs(leaf
, ei
, refs
);
1782 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1784 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1786 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1787 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1788 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1789 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1790 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1791 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1794 BUG_ON(refs_to_mod
!= -1);
1797 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1798 refs
+= refs_to_mod
;
1801 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1802 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1804 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1806 size
= btrfs_extent_inline_ref_size(type
);
1807 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1808 ptr
= (unsigned long)iref
;
1809 end
= (unsigned long)ei
+ item_size
;
1810 if (ptr
+ size
< end
)
1811 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1814 btrfs_truncate_item(root
, path
, item_size
, 1);
1816 btrfs_mark_buffer_dirty(leaf
);
1819 static noinline_for_stack
1820 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1821 struct btrfs_root
*root
,
1822 struct btrfs_path
*path
,
1823 u64 bytenr
, u64 num_bytes
, u64 parent
,
1824 u64 root_objectid
, u64 owner
,
1825 u64 offset
, int refs_to_add
,
1826 struct btrfs_delayed_extent_op
*extent_op
)
1828 struct btrfs_extent_inline_ref
*iref
;
1831 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1832 bytenr
, num_bytes
, parent
,
1833 root_objectid
, owner
, offset
, 1);
1835 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1836 update_inline_extent_backref(root
, path
, iref
,
1837 refs_to_add
, extent_op
);
1838 } else if (ret
== -ENOENT
) {
1839 setup_inline_extent_backref(root
, path
, iref
, parent
,
1840 root_objectid
, owner
, offset
,
1841 refs_to_add
, extent_op
);
1847 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1848 struct btrfs_root
*root
,
1849 struct btrfs_path
*path
,
1850 u64 bytenr
, u64 parent
, u64 root_objectid
,
1851 u64 owner
, u64 offset
, int refs_to_add
)
1854 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1855 BUG_ON(refs_to_add
!= 1);
1856 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1857 parent
, root_objectid
);
1859 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1860 parent
, root_objectid
,
1861 owner
, offset
, refs_to_add
);
1866 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1867 struct btrfs_root
*root
,
1868 struct btrfs_path
*path
,
1869 struct btrfs_extent_inline_ref
*iref
,
1870 int refs_to_drop
, int is_data
)
1874 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1876 update_inline_extent_backref(root
, path
, iref
,
1877 -refs_to_drop
, NULL
);
1878 } else if (is_data
) {
1879 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1881 ret
= btrfs_del_item(trans
, root
, path
);
1886 static int btrfs_issue_discard(struct block_device
*bdev
,
1889 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1892 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1893 u64 num_bytes
, u64
*actual_bytes
)
1896 u64 discarded_bytes
= 0;
1897 struct btrfs_bio
*bbio
= NULL
;
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1902 bytenr
, &num_bytes
, &bbio
, 0);
1903 /* Error condition is -ENOMEM */
1905 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1909 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1910 if (!stripe
->dev
->can_discard
)
1913 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1917 discarded_bytes
+= stripe
->length
;
1918 else if (ret
!= -EOPNOTSUPP
)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1932 *actual_bytes
= discarded_bytes
;
1935 if (ret
== -EOPNOTSUPP
)
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1942 struct btrfs_root
*root
,
1943 u64 bytenr
, u64 num_bytes
, u64 parent
,
1944 u64 root_objectid
, u64 owner
, u64 offset
, int for_cow
)
1947 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1949 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1950 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1952 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1953 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1955 parent
, root_objectid
, (int)owner
,
1956 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1958 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1960 parent
, root_objectid
, owner
, offset
,
1961 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1966 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1967 struct btrfs_root
*root
,
1968 u64 bytenr
, u64 num_bytes
,
1969 u64 parent
, u64 root_objectid
,
1970 u64 owner
, u64 offset
, int refs_to_add
,
1971 struct btrfs_delayed_extent_op
*extent_op
)
1973 struct btrfs_path
*path
;
1974 struct extent_buffer
*leaf
;
1975 struct btrfs_extent_item
*item
;
1979 path
= btrfs_alloc_path();
1984 path
->leave_spinning
= 1;
1985 /* this will setup the path even if it fails to insert the back ref */
1986 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1987 path
, bytenr
, num_bytes
, parent
,
1988 root_objectid
, owner
, offset
,
1989 refs_to_add
, extent_op
);
1993 leaf
= path
->nodes
[0];
1994 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1995 refs
= btrfs_extent_refs(leaf
, item
);
1996 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1998 __run_delayed_extent_op(extent_op
, leaf
, item
);
2000 btrfs_mark_buffer_dirty(leaf
);
2001 btrfs_release_path(path
);
2004 path
->leave_spinning
= 1;
2006 /* now insert the actual backref */
2007 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2008 path
, bytenr
, parent
, root_objectid
,
2009 owner
, offset
, refs_to_add
);
2011 btrfs_abort_transaction(trans
, root
, ret
);
2013 btrfs_free_path(path
);
2017 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2018 struct btrfs_root
*root
,
2019 struct btrfs_delayed_ref_node
*node
,
2020 struct btrfs_delayed_extent_op
*extent_op
,
2021 int insert_reserved
)
2024 struct btrfs_delayed_data_ref
*ref
;
2025 struct btrfs_key ins
;
2030 ins
.objectid
= node
->bytenr
;
2031 ins
.offset
= node
->num_bytes
;
2032 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2034 ref
= btrfs_delayed_node_to_data_ref(node
);
2035 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2037 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2038 parent
= ref
->parent
;
2040 ref_root
= ref
->root
;
2042 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2044 flags
|= extent_op
->flags_to_set
;
2045 ret
= alloc_reserved_file_extent(trans
, root
,
2046 parent
, ref_root
, flags
,
2047 ref
->objectid
, ref
->offset
,
2048 &ins
, node
->ref_mod
);
2049 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2050 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2051 node
->num_bytes
, parent
,
2052 ref_root
, ref
->objectid
,
2053 ref
->offset
, node
->ref_mod
,
2055 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2056 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2057 node
->num_bytes
, parent
,
2058 ref_root
, ref
->objectid
,
2059 ref
->offset
, node
->ref_mod
,
2067 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2068 struct extent_buffer
*leaf
,
2069 struct btrfs_extent_item
*ei
)
2071 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2072 if (extent_op
->update_flags
) {
2073 flags
|= extent_op
->flags_to_set
;
2074 btrfs_set_extent_flags(leaf
, ei
, flags
);
2077 if (extent_op
->update_key
) {
2078 struct btrfs_tree_block_info
*bi
;
2079 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2080 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2081 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2085 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2086 struct btrfs_root
*root
,
2087 struct btrfs_delayed_ref_node
*node
,
2088 struct btrfs_delayed_extent_op
*extent_op
)
2090 struct btrfs_key key
;
2091 struct btrfs_path
*path
;
2092 struct btrfs_extent_item
*ei
;
2093 struct extent_buffer
*leaf
;
2097 int metadata
= !extent_op
->is_data
;
2102 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2105 path
= btrfs_alloc_path();
2109 key
.objectid
= node
->bytenr
;
2112 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2113 key
.offset
= extent_op
->level
;
2115 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2116 key
.offset
= node
->num_bytes
;
2121 path
->leave_spinning
= 1;
2122 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2130 if (path
->slots
[0] > 0) {
2132 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2134 if (key
.objectid
== node
->bytenr
&&
2135 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2136 key
.offset
== node
->num_bytes
)
2140 btrfs_release_path(path
);
2143 key
.objectid
= node
->bytenr
;
2144 key
.offset
= node
->num_bytes
;
2145 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2154 leaf
= path
->nodes
[0];
2155 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2156 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2157 if (item_size
< sizeof(*ei
)) {
2158 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2164 leaf
= path
->nodes
[0];
2165 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2168 BUG_ON(item_size
< sizeof(*ei
));
2169 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2170 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2172 btrfs_mark_buffer_dirty(leaf
);
2174 btrfs_free_path(path
);
2178 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2179 struct btrfs_root
*root
,
2180 struct btrfs_delayed_ref_node
*node
,
2181 struct btrfs_delayed_extent_op
*extent_op
,
2182 int insert_reserved
)
2185 struct btrfs_delayed_tree_ref
*ref
;
2186 struct btrfs_key ins
;
2189 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2192 ref
= btrfs_delayed_node_to_tree_ref(node
);
2193 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2195 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2196 parent
= ref
->parent
;
2198 ref_root
= ref
->root
;
2200 ins
.objectid
= node
->bytenr
;
2201 if (skinny_metadata
) {
2202 ins
.offset
= ref
->level
;
2203 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2205 ins
.offset
= node
->num_bytes
;
2206 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2209 BUG_ON(node
->ref_mod
!= 1);
2210 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2211 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2212 ret
= alloc_reserved_tree_block(trans
, root
,
2214 extent_op
->flags_to_set
,
2217 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2218 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2219 node
->num_bytes
, parent
, ref_root
,
2220 ref
->level
, 0, 1, extent_op
);
2221 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2222 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2223 node
->num_bytes
, parent
, ref_root
,
2224 ref
->level
, 0, 1, extent_op
);
2231 /* helper function to actually process a single delayed ref entry */
2232 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2233 struct btrfs_root
*root
,
2234 struct btrfs_delayed_ref_node
*node
,
2235 struct btrfs_delayed_extent_op
*extent_op
,
2236 int insert_reserved
)
2240 if (trans
->aborted
) {
2241 if (insert_reserved
)
2242 btrfs_pin_extent(root
, node
->bytenr
,
2243 node
->num_bytes
, 1);
2247 if (btrfs_delayed_ref_is_head(node
)) {
2248 struct btrfs_delayed_ref_head
*head
;
2250 * we've hit the end of the chain and we were supposed
2251 * to insert this extent into the tree. But, it got
2252 * deleted before we ever needed to insert it, so all
2253 * we have to do is clean up the accounting
2256 head
= btrfs_delayed_node_to_head(node
);
2257 trace_run_delayed_ref_head(node
, head
, node
->action
);
2259 if (insert_reserved
) {
2260 btrfs_pin_extent(root
, node
->bytenr
,
2261 node
->num_bytes
, 1);
2262 if (head
->is_data
) {
2263 ret
= btrfs_del_csums(trans
, root
,
2271 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2272 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2273 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2275 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2276 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2277 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2284 static noinline
struct btrfs_delayed_ref_node
*
2285 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2287 struct rb_node
*node
;
2288 struct btrfs_delayed_ref_node
*ref
;
2289 int action
= BTRFS_ADD_DELAYED_REF
;
2292 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2293 * this prevents ref count from going down to zero when
2294 * there still are pending delayed ref.
2296 node
= rb_prev(&head
->node
.rb_node
);
2300 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2302 if (ref
->bytenr
!= head
->node
.bytenr
)
2304 if (ref
->action
== action
)
2306 node
= rb_prev(node
);
2308 if (action
== BTRFS_ADD_DELAYED_REF
) {
2309 action
= BTRFS_DROP_DELAYED_REF
;
2316 * Returns 0 on success or if called with an already aborted transaction.
2317 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2319 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2320 struct btrfs_root
*root
,
2321 struct list_head
*cluster
)
2323 struct btrfs_delayed_ref_root
*delayed_refs
;
2324 struct btrfs_delayed_ref_node
*ref
;
2325 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2326 struct btrfs_delayed_extent_op
*extent_op
;
2327 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2330 int must_insert_reserved
= 0;
2332 delayed_refs
= &trans
->transaction
->delayed_refs
;
2335 /* pick a new head ref from the cluster list */
2336 if (list_empty(cluster
))
2339 locked_ref
= list_entry(cluster
->next
,
2340 struct btrfs_delayed_ref_head
, cluster
);
2342 /* grab the lock that says we are going to process
2343 * all the refs for this head */
2344 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2347 * we may have dropped the spin lock to get the head
2348 * mutex lock, and that might have given someone else
2349 * time to free the head. If that's true, it has been
2350 * removed from our list and we can move on.
2352 if (ret
== -EAGAIN
) {
2360 * We need to try and merge add/drops of the same ref since we
2361 * can run into issues with relocate dropping the implicit ref
2362 * and then it being added back again before the drop can
2363 * finish. If we merged anything we need to re-loop so we can
2366 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2370 * locked_ref is the head node, so we have to go one
2371 * node back for any delayed ref updates
2373 ref
= select_delayed_ref(locked_ref
);
2375 if (ref
&& ref
->seq
&&
2376 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2378 * there are still refs with lower seq numbers in the
2379 * process of being added. Don't run this ref yet.
2381 list_del_init(&locked_ref
->cluster
);
2382 btrfs_delayed_ref_unlock(locked_ref
);
2384 delayed_refs
->num_heads_ready
++;
2385 spin_unlock(&delayed_refs
->lock
);
2387 spin_lock(&delayed_refs
->lock
);
2392 * record the must insert reserved flag before we
2393 * drop the spin lock.
2395 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2396 locked_ref
->must_insert_reserved
= 0;
2398 extent_op
= locked_ref
->extent_op
;
2399 locked_ref
->extent_op
= NULL
;
2402 /* All delayed refs have been processed, Go ahead
2403 * and send the head node to run_one_delayed_ref,
2404 * so that any accounting fixes can happen
2406 ref
= &locked_ref
->node
;
2408 if (extent_op
&& must_insert_reserved
) {
2409 btrfs_free_delayed_extent_op(extent_op
);
2414 spin_unlock(&delayed_refs
->lock
);
2416 ret
= run_delayed_extent_op(trans
, root
,
2418 btrfs_free_delayed_extent_op(extent_op
);
2422 * Need to reset must_insert_reserved if
2423 * there was an error so the abort stuff
2424 * can cleanup the reserved space
2427 if (must_insert_reserved
)
2428 locked_ref
->must_insert_reserved
= 1;
2429 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2430 spin_lock(&delayed_refs
->lock
);
2431 btrfs_delayed_ref_unlock(locked_ref
);
2440 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2441 if (btrfs_delayed_ref_is_head(ref
)) {
2442 rb_erase(&locked_ref
->href_node
,
2443 &delayed_refs
->href_root
);
2445 delayed_refs
->num_entries
--;
2446 if (!btrfs_delayed_ref_is_head(ref
)) {
2448 * when we play the delayed ref, also correct the
2451 switch (ref
->action
) {
2452 case BTRFS_ADD_DELAYED_REF
:
2453 case BTRFS_ADD_DELAYED_EXTENT
:
2454 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2456 case BTRFS_DROP_DELAYED_REF
:
2457 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2463 list_del_init(&locked_ref
->cluster
);
2465 spin_unlock(&delayed_refs
->lock
);
2467 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2468 must_insert_reserved
);
2470 btrfs_free_delayed_extent_op(extent_op
);
2472 btrfs_delayed_ref_unlock(locked_ref
);
2473 btrfs_put_delayed_ref(ref
);
2474 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2475 spin_lock(&delayed_refs
->lock
);
2480 * If this node is a head, that means all the refs in this head
2481 * have been dealt with, and we will pick the next head to deal
2482 * with, so we must unlock the head and drop it from the cluster
2483 * list before we release it.
2485 if (btrfs_delayed_ref_is_head(ref
)) {
2486 btrfs_delayed_ref_unlock(locked_ref
);
2489 btrfs_put_delayed_ref(ref
);
2493 spin_lock(&delayed_refs
->lock
);
2498 #ifdef SCRAMBLE_DELAYED_REFS
2500 * Normally delayed refs get processed in ascending bytenr order. This
2501 * correlates in most cases to the order added. To expose dependencies on this
2502 * order, we start to process the tree in the middle instead of the beginning
2504 static u64
find_middle(struct rb_root
*root
)
2506 struct rb_node
*n
= root
->rb_node
;
2507 struct btrfs_delayed_ref_node
*entry
;
2510 u64 first
= 0, last
= 0;
2514 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2515 first
= entry
->bytenr
;
2519 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2520 last
= entry
->bytenr
;
2525 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2526 WARN_ON(!entry
->in_tree
);
2528 middle
= entry
->bytenr
;
2541 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle
*trans
,
2542 struct btrfs_fs_info
*fs_info
)
2544 struct qgroup_update
*qgroup_update
;
2547 if (list_empty(&trans
->qgroup_ref_list
) !=
2548 !trans
->delayed_ref_elem
.seq
) {
2549 /* list without seq or seq without list */
2551 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2552 list_empty(&trans
->qgroup_ref_list
) ? "" : " not",
2553 (u32
)(trans
->delayed_ref_elem
.seq
>> 32),
2554 (u32
)trans
->delayed_ref_elem
.seq
);
2558 if (!trans
->delayed_ref_elem
.seq
)
2561 while (!list_empty(&trans
->qgroup_ref_list
)) {
2562 qgroup_update
= list_first_entry(&trans
->qgroup_ref_list
,
2563 struct qgroup_update
, list
);
2564 list_del(&qgroup_update
->list
);
2566 ret
= btrfs_qgroup_account_ref(
2567 trans
, fs_info
, qgroup_update
->node
,
2568 qgroup_update
->extent_op
);
2569 kfree(qgroup_update
);
2572 btrfs_put_tree_mod_seq(fs_info
, &trans
->delayed_ref_elem
);
2577 static int refs_newer(struct btrfs_delayed_ref_root
*delayed_refs
, int seq
,
2580 int val
= atomic_read(&delayed_refs
->ref_seq
);
2582 if (val
< seq
|| val
>= seq
+ count
)
2587 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2591 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2592 sizeof(struct btrfs_extent_inline_ref
));
2593 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2594 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2597 * We don't ever fill up leaves all the way so multiply by 2 just to be
2598 * closer to what we're really going to want to ouse.
2600 return div64_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2603 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2604 struct btrfs_root
*root
)
2606 struct btrfs_block_rsv
*global_rsv
;
2607 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2611 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2612 num_heads
= heads_to_leaves(root
, num_heads
);
2614 num_bytes
+= (num_heads
- 1) * root
->leafsize
;
2616 global_rsv
= &root
->fs_info
->global_block_rsv
;
2619 * If we can't allocate any more chunks lets make sure we have _lots_ of
2620 * wiggle room since running delayed refs can create more delayed refs.
2622 if (global_rsv
->space_info
->full
)
2625 spin_lock(&global_rsv
->lock
);
2626 if (global_rsv
->reserved
<= num_bytes
)
2628 spin_unlock(&global_rsv
->lock
);
2633 * this starts processing the delayed reference count updates and
2634 * extent insertions we have queued up so far. count can be
2635 * 0, which means to process everything in the tree at the start
2636 * of the run (but not newly added entries), or it can be some target
2637 * number you'd like to process.
2639 * Returns 0 on success or if called with an aborted transaction
2640 * Returns <0 on error and aborts the transaction
2642 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2643 struct btrfs_root
*root
, unsigned long count
)
2645 struct rb_node
*node
;
2646 struct btrfs_delayed_ref_root
*delayed_refs
;
2647 struct btrfs_delayed_ref_head
*head
;
2648 struct list_head cluster
;
2651 int run_all
= count
== (unsigned long)-1;
2655 /* We'll clean this up in btrfs_cleanup_transaction */
2659 if (root
== root
->fs_info
->extent_root
)
2660 root
= root
->fs_info
->tree_root
;
2662 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
2664 delayed_refs
= &trans
->transaction
->delayed_refs
;
2665 INIT_LIST_HEAD(&cluster
);
2667 count
= delayed_refs
->num_entries
* 2;
2671 if (!run_all
&& !run_most
) {
2673 int seq
= atomic_read(&delayed_refs
->ref_seq
);
2676 old
= atomic_cmpxchg(&delayed_refs
->procs_running_refs
, 0, 1);
2678 DEFINE_WAIT(__wait
);
2679 if (delayed_refs
->flushing
||
2680 !btrfs_should_throttle_delayed_refs(trans
, root
))
2683 prepare_to_wait(&delayed_refs
->wait
, &__wait
,
2684 TASK_UNINTERRUPTIBLE
);
2686 old
= atomic_cmpxchg(&delayed_refs
->procs_running_refs
, 0, 1);
2689 finish_wait(&delayed_refs
->wait
, &__wait
);
2691 if (!refs_newer(delayed_refs
, seq
, 256))
2696 finish_wait(&delayed_refs
->wait
, &__wait
);
2702 atomic_inc(&delayed_refs
->procs_running_refs
);
2707 spin_lock(&delayed_refs
->lock
);
2709 #ifdef SCRAMBLE_DELAYED_REFS
2710 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2714 if (!(run_all
|| run_most
) &&
2715 !btrfs_should_throttle_delayed_refs(trans
, root
))
2719 * go find something we can process in the rbtree. We start at
2720 * the beginning of the tree, and then build a cluster
2721 * of refs to process starting at the first one we are able to
2724 delayed_start
= delayed_refs
->run_delayed_start
;
2725 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2726 delayed_refs
->run_delayed_start
);
2730 ret
= run_clustered_refs(trans
, root
, &cluster
);
2732 btrfs_release_ref_cluster(&cluster
);
2733 spin_unlock(&delayed_refs
->lock
);
2734 btrfs_abort_transaction(trans
, root
, ret
);
2735 atomic_dec(&delayed_refs
->procs_running_refs
);
2736 wake_up(&delayed_refs
->wait
);
2740 atomic_add(ret
, &delayed_refs
->ref_seq
);
2742 count
-= min_t(unsigned long, ret
, count
);
2747 if (delayed_start
>= delayed_refs
->run_delayed_start
) {
2750 * btrfs_find_ref_cluster looped. let's do one
2751 * more cycle. if we don't run any delayed ref
2752 * during that cycle (because we can't because
2753 * all of them are blocked), bail out.
2758 * no runnable refs left, stop trying
2765 /* refs were run, let's reset staleness detection */
2771 if (!list_empty(&trans
->new_bgs
)) {
2772 spin_unlock(&delayed_refs
->lock
);
2773 btrfs_create_pending_block_groups(trans
, root
);
2774 spin_lock(&delayed_refs
->lock
);
2777 node
= rb_first(&delayed_refs
->href_root
);
2780 count
= (unsigned long)-1;
2783 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2785 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2786 struct btrfs_delayed_ref_node
*ref
;
2789 atomic_inc(&ref
->refs
);
2791 spin_unlock(&delayed_refs
->lock
);
2793 * Mutex was contended, block until it's
2794 * released and try again
2796 mutex_lock(&head
->mutex
);
2797 mutex_unlock(&head
->mutex
);
2799 btrfs_put_delayed_ref(ref
);
2805 node
= rb_next(node
);
2807 spin_unlock(&delayed_refs
->lock
);
2808 schedule_timeout(1);
2812 atomic_dec(&delayed_refs
->procs_running_refs
);
2814 if (waitqueue_active(&delayed_refs
->wait
))
2815 wake_up(&delayed_refs
->wait
);
2817 spin_unlock(&delayed_refs
->lock
);
2818 assert_qgroups_uptodate(trans
);
2822 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2823 struct btrfs_root
*root
,
2824 u64 bytenr
, u64 num_bytes
, u64 flags
,
2825 int level
, int is_data
)
2827 struct btrfs_delayed_extent_op
*extent_op
;
2830 extent_op
= btrfs_alloc_delayed_extent_op();
2834 extent_op
->flags_to_set
= flags
;
2835 extent_op
->update_flags
= 1;
2836 extent_op
->update_key
= 0;
2837 extent_op
->is_data
= is_data
? 1 : 0;
2838 extent_op
->level
= level
;
2840 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2841 num_bytes
, extent_op
);
2843 btrfs_free_delayed_extent_op(extent_op
);
2847 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2848 struct btrfs_root
*root
,
2849 struct btrfs_path
*path
,
2850 u64 objectid
, u64 offset
, u64 bytenr
)
2852 struct btrfs_delayed_ref_head
*head
;
2853 struct btrfs_delayed_ref_node
*ref
;
2854 struct btrfs_delayed_data_ref
*data_ref
;
2855 struct btrfs_delayed_ref_root
*delayed_refs
;
2856 struct rb_node
*node
;
2860 delayed_refs
= &trans
->transaction
->delayed_refs
;
2861 spin_lock(&delayed_refs
->lock
);
2862 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2866 if (!mutex_trylock(&head
->mutex
)) {
2867 atomic_inc(&head
->node
.refs
);
2868 spin_unlock(&delayed_refs
->lock
);
2870 btrfs_release_path(path
);
2873 * Mutex was contended, block until it's released and let
2876 mutex_lock(&head
->mutex
);
2877 mutex_unlock(&head
->mutex
);
2878 btrfs_put_delayed_ref(&head
->node
);
2882 node
= rb_prev(&head
->node
.rb_node
);
2886 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2888 if (ref
->bytenr
!= bytenr
)
2892 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2895 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2897 node
= rb_prev(node
);
2901 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2902 if (ref
->bytenr
== bytenr
&& ref
->seq
== seq
)
2906 if (data_ref
->root
!= root
->root_key
.objectid
||
2907 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2912 mutex_unlock(&head
->mutex
);
2914 spin_unlock(&delayed_refs
->lock
);
2918 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2919 struct btrfs_root
*root
,
2920 struct btrfs_path
*path
,
2921 u64 objectid
, u64 offset
, u64 bytenr
)
2923 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2924 struct extent_buffer
*leaf
;
2925 struct btrfs_extent_data_ref
*ref
;
2926 struct btrfs_extent_inline_ref
*iref
;
2927 struct btrfs_extent_item
*ei
;
2928 struct btrfs_key key
;
2932 key
.objectid
= bytenr
;
2933 key
.offset
= (u64
)-1;
2934 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2936 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2939 BUG_ON(ret
== 0); /* Corruption */
2942 if (path
->slots
[0] == 0)
2946 leaf
= path
->nodes
[0];
2947 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2949 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2953 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2955 if (item_size
< sizeof(*ei
)) {
2956 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2960 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2962 if (item_size
!= sizeof(*ei
) +
2963 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2966 if (btrfs_extent_generation(leaf
, ei
) <=
2967 btrfs_root_last_snapshot(&root
->root_item
))
2970 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2971 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2972 BTRFS_EXTENT_DATA_REF_KEY
)
2975 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2976 if (btrfs_extent_refs(leaf
, ei
) !=
2977 btrfs_extent_data_ref_count(leaf
, ref
) ||
2978 btrfs_extent_data_ref_root(leaf
, ref
) !=
2979 root
->root_key
.objectid
||
2980 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2981 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2989 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2990 struct btrfs_root
*root
,
2991 u64 objectid
, u64 offset
, u64 bytenr
)
2993 struct btrfs_path
*path
;
2997 path
= btrfs_alloc_path();
3002 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3004 if (ret
&& ret
!= -ENOENT
)
3007 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3009 } while (ret2
== -EAGAIN
);
3011 if (ret2
&& ret2
!= -ENOENT
) {
3016 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3019 btrfs_free_path(path
);
3020 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3025 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3026 struct btrfs_root
*root
,
3027 struct extent_buffer
*buf
,
3028 int full_backref
, int inc
, int for_cow
)
3035 struct btrfs_key key
;
3036 struct btrfs_file_extent_item
*fi
;
3040 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3041 u64
, u64
, u64
, u64
, u64
, u64
, int);
3043 ref_root
= btrfs_header_owner(buf
);
3044 nritems
= btrfs_header_nritems(buf
);
3045 level
= btrfs_header_level(buf
);
3047 if (!root
->ref_cows
&& level
== 0)
3051 process_func
= btrfs_inc_extent_ref
;
3053 process_func
= btrfs_free_extent
;
3056 parent
= buf
->start
;
3060 for (i
= 0; i
< nritems
; i
++) {
3062 btrfs_item_key_to_cpu(buf
, &key
, i
);
3063 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
3065 fi
= btrfs_item_ptr(buf
, i
,
3066 struct btrfs_file_extent_item
);
3067 if (btrfs_file_extent_type(buf
, fi
) ==
3068 BTRFS_FILE_EXTENT_INLINE
)
3070 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3074 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3075 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3076 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3077 parent
, ref_root
, key
.objectid
,
3078 key
.offset
, for_cow
);
3082 bytenr
= btrfs_node_blockptr(buf
, i
);
3083 num_bytes
= btrfs_level_size(root
, level
- 1);
3084 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3085 parent
, ref_root
, level
- 1, 0,
3096 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3097 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
3099 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1, for_cow
);
3102 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3103 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
3105 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0, for_cow
);
3108 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3109 struct btrfs_root
*root
,
3110 struct btrfs_path
*path
,
3111 struct btrfs_block_group_cache
*cache
)
3114 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3116 struct extent_buffer
*leaf
;
3118 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3121 BUG_ON(ret
); /* Corruption */
3123 leaf
= path
->nodes
[0];
3124 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3125 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3126 btrfs_mark_buffer_dirty(leaf
);
3127 btrfs_release_path(path
);
3130 btrfs_abort_transaction(trans
, root
, ret
);
3137 static struct btrfs_block_group_cache
*
3138 next_block_group(struct btrfs_root
*root
,
3139 struct btrfs_block_group_cache
*cache
)
3141 struct rb_node
*node
;
3142 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3143 node
= rb_next(&cache
->cache_node
);
3144 btrfs_put_block_group(cache
);
3146 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3148 btrfs_get_block_group(cache
);
3151 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3155 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3156 struct btrfs_trans_handle
*trans
,
3157 struct btrfs_path
*path
)
3159 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3160 struct inode
*inode
= NULL
;
3162 int dcs
= BTRFS_DC_ERROR
;
3168 * If this block group is smaller than 100 megs don't bother caching the
3171 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3172 spin_lock(&block_group
->lock
);
3173 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3174 spin_unlock(&block_group
->lock
);
3179 inode
= lookup_free_space_inode(root
, block_group
, path
);
3180 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3181 ret
= PTR_ERR(inode
);
3182 btrfs_release_path(path
);
3186 if (IS_ERR(inode
)) {
3190 if (block_group
->ro
)
3193 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3199 /* We've already setup this transaction, go ahead and exit */
3200 if (block_group
->cache_generation
== trans
->transid
&&
3201 i_size_read(inode
)) {
3202 dcs
= BTRFS_DC_SETUP
;
3207 * We want to set the generation to 0, that way if anything goes wrong
3208 * from here on out we know not to trust this cache when we load up next
3211 BTRFS_I(inode
)->generation
= 0;
3212 ret
= btrfs_update_inode(trans
, root
, inode
);
3215 if (i_size_read(inode
) > 0) {
3216 ret
= btrfs_check_trunc_cache_free_space(root
,
3217 &root
->fs_info
->global_block_rsv
);
3221 ret
= btrfs_truncate_free_space_cache(root
, trans
, inode
);
3226 spin_lock(&block_group
->lock
);
3227 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3228 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3230 * don't bother trying to write stuff out _if_
3231 * a) we're not cached,
3232 * b) we're with nospace_cache mount option.
3234 dcs
= BTRFS_DC_WRITTEN
;
3235 spin_unlock(&block_group
->lock
);
3238 spin_unlock(&block_group
->lock
);
3241 * Try to preallocate enough space based on how big the block group is.
3242 * Keep in mind this has to include any pinned space which could end up
3243 * taking up quite a bit since it's not folded into the other space
3246 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3251 num_pages
*= PAGE_CACHE_SIZE
;
3253 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3257 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3258 num_pages
, num_pages
,
3261 dcs
= BTRFS_DC_SETUP
;
3262 btrfs_free_reserved_data_space(inode
, num_pages
);
3267 btrfs_release_path(path
);
3269 spin_lock(&block_group
->lock
);
3270 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3271 block_group
->cache_generation
= trans
->transid
;
3272 block_group
->disk_cache_state
= dcs
;
3273 spin_unlock(&block_group
->lock
);
3278 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3279 struct btrfs_root
*root
)
3281 struct btrfs_block_group_cache
*cache
;
3283 struct btrfs_path
*path
;
3286 path
= btrfs_alloc_path();
3292 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3294 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3296 cache
= next_block_group(root
, cache
);
3304 err
= cache_save_setup(cache
, trans
, path
);
3305 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3306 btrfs_put_block_group(cache
);
3311 err
= btrfs_run_delayed_refs(trans
, root
,
3313 if (err
) /* File system offline */
3317 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3319 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
3320 btrfs_put_block_group(cache
);
3326 cache
= next_block_group(root
, cache
);
3335 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3336 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
3338 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3340 err
= write_one_cache_group(trans
, root
, path
, cache
);
3341 btrfs_put_block_group(cache
);
3342 if (err
) /* File system offline */
3348 * I don't think this is needed since we're just marking our
3349 * preallocated extent as written, but just in case it can't
3353 err
= btrfs_run_delayed_refs(trans
, root
,
3355 if (err
) /* File system offline */
3359 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3362 * Really this shouldn't happen, but it could if we
3363 * couldn't write the entire preallocated extent and
3364 * splitting the extent resulted in a new block.
3367 btrfs_put_block_group(cache
);
3370 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3372 cache
= next_block_group(root
, cache
);
3381 err
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3384 * If we didn't have an error then the cache state is still
3385 * NEED_WRITE, so we can set it to WRITTEN.
3387 if (!err
&& cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3388 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3389 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3390 btrfs_put_block_group(cache
);
3394 btrfs_free_path(path
);
3398 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3400 struct btrfs_block_group_cache
*block_group
;
3403 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3404 if (!block_group
|| block_group
->ro
)
3407 btrfs_put_block_group(block_group
);
3411 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3412 u64 total_bytes
, u64 bytes_used
,
3413 struct btrfs_space_info
**space_info
)
3415 struct btrfs_space_info
*found
;
3420 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3421 BTRFS_BLOCK_GROUP_RAID10
))
3426 found
= __find_space_info(info
, flags
);
3428 spin_lock(&found
->lock
);
3429 found
->total_bytes
+= total_bytes
;
3430 found
->disk_total
+= total_bytes
* factor
;
3431 found
->bytes_used
+= bytes_used
;
3432 found
->disk_used
+= bytes_used
* factor
;
3434 spin_unlock(&found
->lock
);
3435 *space_info
= found
;
3438 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3442 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0);
3448 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3449 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3450 init_rwsem(&found
->groups_sem
);
3451 spin_lock_init(&found
->lock
);
3452 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3453 found
->total_bytes
= total_bytes
;
3454 found
->disk_total
= total_bytes
* factor
;
3455 found
->bytes_used
= bytes_used
;
3456 found
->disk_used
= bytes_used
* factor
;
3457 found
->bytes_pinned
= 0;
3458 found
->bytes_reserved
= 0;
3459 found
->bytes_readonly
= 0;
3460 found
->bytes_may_use
= 0;
3462 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3463 found
->chunk_alloc
= 0;
3465 init_waitqueue_head(&found
->wait
);
3466 *space_info
= found
;
3467 list_add_rcu(&found
->list
, &info
->space_info
);
3468 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3469 info
->data_sinfo
= found
;
3473 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3475 u64 extra_flags
= chunk_to_extended(flags
) &
3476 BTRFS_EXTENDED_PROFILE_MASK
;
3478 write_seqlock(&fs_info
->profiles_lock
);
3479 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3480 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3481 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3482 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3483 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3484 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3485 write_sequnlock(&fs_info
->profiles_lock
);
3489 * returns target flags in extended format or 0 if restripe for this
3490 * chunk_type is not in progress
3492 * should be called with either volume_mutex or balance_lock held
3494 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3496 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3502 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3503 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3504 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3505 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3506 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3507 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3508 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3509 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3510 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3517 * @flags: available profiles in extended format (see ctree.h)
3519 * Returns reduced profile in chunk format. If profile changing is in
3520 * progress (either running or paused) picks the target profile (if it's
3521 * already available), otherwise falls back to plain reducing.
3523 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3526 * we add in the count of missing devices because we want
3527 * to make sure that any RAID levels on a degraded FS
3528 * continue to be honored.
3530 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3531 root
->fs_info
->fs_devices
->missing_devices
;
3536 * see if restripe for this chunk_type is in progress, if so
3537 * try to reduce to the target profile
3539 spin_lock(&root
->fs_info
->balance_lock
);
3540 target
= get_restripe_target(root
->fs_info
, flags
);
3542 /* pick target profile only if it's already available */
3543 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3544 spin_unlock(&root
->fs_info
->balance_lock
);
3545 return extended_to_chunk(target
);
3548 spin_unlock(&root
->fs_info
->balance_lock
);
3550 /* First, mask out the RAID levels which aren't possible */
3551 if (num_devices
== 1)
3552 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3553 BTRFS_BLOCK_GROUP_RAID5
);
3554 if (num_devices
< 3)
3555 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3556 if (num_devices
< 4)
3557 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3559 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3560 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3561 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3564 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3565 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3566 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3567 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3568 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3569 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3570 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3571 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3572 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3573 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3575 return extended_to_chunk(flags
| tmp
);
3578 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3583 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3585 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3586 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3587 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3588 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3589 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3590 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3591 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3593 return btrfs_reduce_alloc_profile(root
, flags
);
3596 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3602 flags
= BTRFS_BLOCK_GROUP_DATA
;
3603 else if (root
== root
->fs_info
->chunk_root
)
3604 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3606 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3608 ret
= get_alloc_profile(root
, flags
);
3613 * This will check the space that the inode allocates from to make sure we have
3614 * enough space for bytes.
3616 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3618 struct btrfs_space_info
*data_sinfo
;
3619 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3620 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3622 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3624 /* make sure bytes are sectorsize aligned */
3625 bytes
= ALIGN(bytes
, root
->sectorsize
);
3627 if (btrfs_is_free_space_inode(inode
)) {
3629 ASSERT(current
->journal_info
);
3632 data_sinfo
= fs_info
->data_sinfo
;
3637 /* make sure we have enough space to handle the data first */
3638 spin_lock(&data_sinfo
->lock
);
3639 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3640 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3641 data_sinfo
->bytes_may_use
;
3643 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3644 struct btrfs_trans_handle
*trans
;
3647 * if we don't have enough free bytes in this space then we need
3648 * to alloc a new chunk.
3650 if (!data_sinfo
->full
&& alloc_chunk
) {
3653 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3654 spin_unlock(&data_sinfo
->lock
);
3656 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3658 * It is ugly that we don't call nolock join
3659 * transaction for the free space inode case here.
3660 * But it is safe because we only do the data space
3661 * reservation for the free space cache in the
3662 * transaction context, the common join transaction
3663 * just increase the counter of the current transaction
3664 * handler, doesn't try to acquire the trans_lock of
3667 trans
= btrfs_join_transaction(root
);
3669 return PTR_ERR(trans
);
3671 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3673 CHUNK_ALLOC_NO_FORCE
);
3674 btrfs_end_transaction(trans
, root
);
3683 data_sinfo
= fs_info
->data_sinfo
;
3689 * If we don't have enough pinned space to deal with this
3690 * allocation don't bother committing the transaction.
3692 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3695 spin_unlock(&data_sinfo
->lock
);
3697 /* commit the current transaction and try again */
3700 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3703 trans
= btrfs_join_transaction(root
);
3705 return PTR_ERR(trans
);
3706 ret
= btrfs_commit_transaction(trans
, root
);
3712 trace_btrfs_space_reservation(root
->fs_info
,
3713 "space_info:enospc",
3714 data_sinfo
->flags
, bytes
, 1);
3717 data_sinfo
->bytes_may_use
+= bytes
;
3718 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3719 data_sinfo
->flags
, bytes
, 1);
3720 spin_unlock(&data_sinfo
->lock
);
3726 * Called if we need to clear a data reservation for this inode.
3728 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3730 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3731 struct btrfs_space_info
*data_sinfo
;
3733 /* make sure bytes are sectorsize aligned */
3734 bytes
= ALIGN(bytes
, root
->sectorsize
);
3736 data_sinfo
= root
->fs_info
->data_sinfo
;
3737 spin_lock(&data_sinfo
->lock
);
3738 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3739 data_sinfo
->bytes_may_use
-= bytes
;
3740 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3741 data_sinfo
->flags
, bytes
, 0);
3742 spin_unlock(&data_sinfo
->lock
);
3745 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3747 struct list_head
*head
= &info
->space_info
;
3748 struct btrfs_space_info
*found
;
3751 list_for_each_entry_rcu(found
, head
, list
) {
3752 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3753 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3758 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3760 return (global
->size
<< 1);
3763 static int should_alloc_chunk(struct btrfs_root
*root
,
3764 struct btrfs_space_info
*sinfo
, int force
)
3766 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3767 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3768 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3771 if (force
== CHUNK_ALLOC_FORCE
)
3775 * We need to take into account the global rsv because for all intents
3776 * and purposes it's used space. Don't worry about locking the
3777 * global_rsv, it doesn't change except when the transaction commits.
3779 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3780 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
3783 * in limited mode, we want to have some free space up to
3784 * about 1% of the FS size.
3786 if (force
== CHUNK_ALLOC_LIMITED
) {
3787 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3788 thresh
= max_t(u64
, 64 * 1024 * 1024,
3789 div_factor_fine(thresh
, 1));
3791 if (num_bytes
- num_allocated
< thresh
)
3795 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3800 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3804 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
3805 BTRFS_BLOCK_GROUP_RAID0
|
3806 BTRFS_BLOCK_GROUP_RAID5
|
3807 BTRFS_BLOCK_GROUP_RAID6
))
3808 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3809 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3812 num_dev
= 1; /* DUP or single */
3814 /* metadata for updaing devices and chunk tree */
3815 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3818 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3819 struct btrfs_root
*root
, u64 type
)
3821 struct btrfs_space_info
*info
;
3825 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3826 spin_lock(&info
->lock
);
3827 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3828 info
->bytes_reserved
- info
->bytes_readonly
;
3829 spin_unlock(&info
->lock
);
3831 thresh
= get_system_chunk_thresh(root
, type
);
3832 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3833 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
3834 left
, thresh
, type
);
3835 dump_space_info(info
, 0, 0);
3838 if (left
< thresh
) {
3841 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3842 btrfs_alloc_chunk(trans
, root
, flags
);
3846 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3847 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3849 struct btrfs_space_info
*space_info
;
3850 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3851 int wait_for_alloc
= 0;
3854 /* Don't re-enter if we're already allocating a chunk */
3855 if (trans
->allocating_chunk
)
3858 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3860 ret
= update_space_info(extent_root
->fs_info
, flags
,
3862 BUG_ON(ret
); /* -ENOMEM */
3864 BUG_ON(!space_info
); /* Logic error */
3867 spin_lock(&space_info
->lock
);
3868 if (force
< space_info
->force_alloc
)
3869 force
= space_info
->force_alloc
;
3870 if (space_info
->full
) {
3871 if (should_alloc_chunk(extent_root
, space_info
, force
))
3875 spin_unlock(&space_info
->lock
);
3879 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3880 spin_unlock(&space_info
->lock
);
3882 } else if (space_info
->chunk_alloc
) {
3885 space_info
->chunk_alloc
= 1;
3888 spin_unlock(&space_info
->lock
);
3890 mutex_lock(&fs_info
->chunk_mutex
);
3893 * The chunk_mutex is held throughout the entirety of a chunk
3894 * allocation, so once we've acquired the chunk_mutex we know that the
3895 * other guy is done and we need to recheck and see if we should
3898 if (wait_for_alloc
) {
3899 mutex_unlock(&fs_info
->chunk_mutex
);
3904 trans
->allocating_chunk
= true;
3907 * If we have mixed data/metadata chunks we want to make sure we keep
3908 * allocating mixed chunks instead of individual chunks.
3910 if (btrfs_mixed_space_info(space_info
))
3911 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3914 * if we're doing a data chunk, go ahead and make sure that
3915 * we keep a reasonable number of metadata chunks allocated in the
3918 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3919 fs_info
->data_chunk_allocations
++;
3920 if (!(fs_info
->data_chunk_allocations
%
3921 fs_info
->metadata_ratio
))
3922 force_metadata_allocation(fs_info
);
3926 * Check if we have enough space in SYSTEM chunk because we may need
3927 * to update devices.
3929 check_system_chunk(trans
, extent_root
, flags
);
3931 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3932 trans
->allocating_chunk
= false;
3934 spin_lock(&space_info
->lock
);
3935 if (ret
< 0 && ret
!= -ENOSPC
)
3938 space_info
->full
= 1;
3942 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3944 space_info
->chunk_alloc
= 0;
3945 spin_unlock(&space_info
->lock
);
3946 mutex_unlock(&fs_info
->chunk_mutex
);
3950 static int can_overcommit(struct btrfs_root
*root
,
3951 struct btrfs_space_info
*space_info
, u64 bytes
,
3952 enum btrfs_reserve_flush_enum flush
)
3954 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3955 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3960 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3961 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3964 * We only want to allow over committing if we have lots of actual space
3965 * free, but if we don't have enough space to handle the global reserve
3966 * space then we could end up having a real enospc problem when trying
3967 * to allocate a chunk or some other such important allocation.
3969 spin_lock(&global_rsv
->lock
);
3970 space_size
= calc_global_rsv_need_space(global_rsv
);
3971 spin_unlock(&global_rsv
->lock
);
3972 if (used
+ space_size
>= space_info
->total_bytes
)
3975 used
+= space_info
->bytes_may_use
;
3977 spin_lock(&root
->fs_info
->free_chunk_lock
);
3978 avail
= root
->fs_info
->free_chunk_space
;
3979 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3982 * If we have dup, raid1 or raid10 then only half of the free
3983 * space is actually useable. For raid56, the space info used
3984 * doesn't include the parity drive, so we don't have to
3987 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3988 BTRFS_BLOCK_GROUP_RAID1
|
3989 BTRFS_BLOCK_GROUP_RAID10
))
3993 * If we aren't flushing all things, let us overcommit up to
3994 * 1/2th of the space. If we can flush, don't let us overcommit
3995 * too much, let it overcommit up to 1/8 of the space.
3997 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4002 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4007 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4008 unsigned long nr_pages
)
4010 struct super_block
*sb
= root
->fs_info
->sb
;
4012 if (down_read_trylock(&sb
->s_umount
)) {
4013 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4014 up_read(&sb
->s_umount
);
4017 * We needn't worry the filesystem going from r/w to r/o though
4018 * we don't acquire ->s_umount mutex, because the filesystem
4019 * should guarantee the delalloc inodes list be empty after
4020 * the filesystem is readonly(all dirty pages are written to
4023 btrfs_start_delalloc_roots(root
->fs_info
, 0);
4024 if (!current
->journal_info
)
4025 btrfs_wait_ordered_roots(root
->fs_info
, -1);
4029 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4034 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4035 nr
= (int)div64_u64(to_reclaim
, bytes
);
4041 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4044 * shrink metadata reservation for delalloc
4046 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4049 struct btrfs_block_rsv
*block_rsv
;
4050 struct btrfs_space_info
*space_info
;
4051 struct btrfs_trans_handle
*trans
;
4055 unsigned long nr_pages
;
4058 enum btrfs_reserve_flush_enum flush
;
4060 /* Calc the number of the pages we need flush for space reservation */
4061 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4062 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4064 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4065 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4066 space_info
= block_rsv
->space_info
;
4068 delalloc_bytes
= percpu_counter_sum_positive(
4069 &root
->fs_info
->delalloc_bytes
);
4070 if (delalloc_bytes
== 0) {
4074 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4079 while (delalloc_bytes
&& loops
< 3) {
4080 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4081 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4082 btrfs_writeback_inodes_sb_nr(root
, nr_pages
);
4084 * We need to wait for the async pages to actually start before
4087 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4091 if (max_reclaim
<= nr_pages
)
4094 max_reclaim
-= nr_pages
;
4096 wait_event(root
->fs_info
->async_submit_wait
,
4097 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4101 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4103 flush
= BTRFS_RESERVE_NO_FLUSH
;
4104 spin_lock(&space_info
->lock
);
4105 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4106 spin_unlock(&space_info
->lock
);
4109 spin_unlock(&space_info
->lock
);
4112 if (wait_ordered
&& !trans
) {
4113 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4115 time_left
= schedule_timeout_killable(1);
4119 delalloc_bytes
= percpu_counter_sum_positive(
4120 &root
->fs_info
->delalloc_bytes
);
4125 * maybe_commit_transaction - possibly commit the transaction if its ok to
4126 * @root - the root we're allocating for
4127 * @bytes - the number of bytes we want to reserve
4128 * @force - force the commit
4130 * This will check to make sure that committing the transaction will actually
4131 * get us somewhere and then commit the transaction if it does. Otherwise it
4132 * will return -ENOSPC.
4134 static int may_commit_transaction(struct btrfs_root
*root
,
4135 struct btrfs_space_info
*space_info
,
4136 u64 bytes
, int force
)
4138 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4139 struct btrfs_trans_handle
*trans
;
4141 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4148 /* See if there is enough pinned space to make this reservation */
4149 spin_lock(&space_info
->lock
);
4150 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4152 spin_unlock(&space_info
->lock
);
4155 spin_unlock(&space_info
->lock
);
4158 * See if there is some space in the delayed insertion reservation for
4161 if (space_info
!= delayed_rsv
->space_info
)
4164 spin_lock(&space_info
->lock
);
4165 spin_lock(&delayed_rsv
->lock
);
4166 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4167 bytes
- delayed_rsv
->size
) >= 0) {
4168 spin_unlock(&delayed_rsv
->lock
);
4169 spin_unlock(&space_info
->lock
);
4172 spin_unlock(&delayed_rsv
->lock
);
4173 spin_unlock(&space_info
->lock
);
4176 trans
= btrfs_join_transaction(root
);
4180 return btrfs_commit_transaction(trans
, root
);
4184 FLUSH_DELAYED_ITEMS_NR
= 1,
4185 FLUSH_DELAYED_ITEMS
= 2,
4187 FLUSH_DELALLOC_WAIT
= 4,
4192 static int flush_space(struct btrfs_root
*root
,
4193 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4194 u64 orig_bytes
, int state
)
4196 struct btrfs_trans_handle
*trans
;
4201 case FLUSH_DELAYED_ITEMS_NR
:
4202 case FLUSH_DELAYED_ITEMS
:
4203 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4204 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4208 trans
= btrfs_join_transaction(root
);
4209 if (IS_ERR(trans
)) {
4210 ret
= PTR_ERR(trans
);
4213 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4214 btrfs_end_transaction(trans
, root
);
4216 case FLUSH_DELALLOC
:
4217 case FLUSH_DELALLOC_WAIT
:
4218 shrink_delalloc(root
, num_bytes
, orig_bytes
,
4219 state
== FLUSH_DELALLOC_WAIT
);
4222 trans
= btrfs_join_transaction(root
);
4223 if (IS_ERR(trans
)) {
4224 ret
= PTR_ERR(trans
);
4227 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4228 btrfs_get_alloc_profile(root
, 0),
4229 CHUNK_ALLOC_NO_FORCE
);
4230 btrfs_end_transaction(trans
, root
);
4235 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4245 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4246 * @root - the root we're allocating for
4247 * @block_rsv - the block_rsv we're allocating for
4248 * @orig_bytes - the number of bytes we want
4249 * @flush - whether or not we can flush to make our reservation
4251 * This will reserve orgi_bytes number of bytes from the space info associated
4252 * with the block_rsv. If there is not enough space it will make an attempt to
4253 * flush out space to make room. It will do this by flushing delalloc if
4254 * possible or committing the transaction. If flush is 0 then no attempts to
4255 * regain reservations will be made and this will fail if there is not enough
4258 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4259 struct btrfs_block_rsv
*block_rsv
,
4261 enum btrfs_reserve_flush_enum flush
)
4263 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4265 u64 num_bytes
= orig_bytes
;
4266 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4268 bool flushing
= false;
4272 spin_lock(&space_info
->lock
);
4274 * We only want to wait if somebody other than us is flushing and we
4275 * are actually allowed to flush all things.
4277 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4278 space_info
->flush
) {
4279 spin_unlock(&space_info
->lock
);
4281 * If we have a trans handle we can't wait because the flusher
4282 * may have to commit the transaction, which would mean we would
4283 * deadlock since we are waiting for the flusher to finish, but
4284 * hold the current transaction open.
4286 if (current
->journal_info
)
4288 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4289 /* Must have been killed, return */
4293 spin_lock(&space_info
->lock
);
4297 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4298 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4299 space_info
->bytes_may_use
;
4302 * The idea here is that we've not already over-reserved the block group
4303 * then we can go ahead and save our reservation first and then start
4304 * flushing if we need to. Otherwise if we've already overcommitted
4305 * lets start flushing stuff first and then come back and try to make
4308 if (used
<= space_info
->total_bytes
) {
4309 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4310 space_info
->bytes_may_use
+= orig_bytes
;
4311 trace_btrfs_space_reservation(root
->fs_info
,
4312 "space_info", space_info
->flags
, orig_bytes
, 1);
4316 * Ok set num_bytes to orig_bytes since we aren't
4317 * overocmmitted, this way we only try and reclaim what
4320 num_bytes
= orig_bytes
;
4324 * Ok we're over committed, set num_bytes to the overcommitted
4325 * amount plus the amount of bytes that we need for this
4328 num_bytes
= used
- space_info
->total_bytes
+
4332 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4333 space_info
->bytes_may_use
+= orig_bytes
;
4334 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4335 space_info
->flags
, orig_bytes
,
4341 * Couldn't make our reservation, save our place so while we're trying
4342 * to reclaim space we can actually use it instead of somebody else
4343 * stealing it from us.
4345 * We make the other tasks wait for the flush only when we can flush
4348 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4350 space_info
->flush
= 1;
4353 spin_unlock(&space_info
->lock
);
4355 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4358 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4363 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4364 * would happen. So skip delalloc flush.
4366 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4367 (flush_state
== FLUSH_DELALLOC
||
4368 flush_state
== FLUSH_DELALLOC_WAIT
))
4369 flush_state
= ALLOC_CHUNK
;
4373 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4374 flush_state
< COMMIT_TRANS
)
4376 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4377 flush_state
<= COMMIT_TRANS
)
4381 if (ret
== -ENOSPC
&&
4382 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4383 struct btrfs_block_rsv
*global_rsv
=
4384 &root
->fs_info
->global_block_rsv
;
4386 if (block_rsv
!= global_rsv
&&
4387 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4391 trace_btrfs_space_reservation(root
->fs_info
,
4392 "space_info:enospc",
4393 space_info
->flags
, orig_bytes
, 1);
4395 spin_lock(&space_info
->lock
);
4396 space_info
->flush
= 0;
4397 wake_up_all(&space_info
->wait
);
4398 spin_unlock(&space_info
->lock
);
4403 static struct btrfs_block_rsv
*get_block_rsv(
4404 const struct btrfs_trans_handle
*trans
,
4405 const struct btrfs_root
*root
)
4407 struct btrfs_block_rsv
*block_rsv
= NULL
;
4410 block_rsv
= trans
->block_rsv
;
4412 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4413 block_rsv
= trans
->block_rsv
;
4415 if (root
== root
->fs_info
->uuid_root
)
4416 block_rsv
= trans
->block_rsv
;
4419 block_rsv
= root
->block_rsv
;
4422 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4427 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4431 spin_lock(&block_rsv
->lock
);
4432 if (block_rsv
->reserved
>= num_bytes
) {
4433 block_rsv
->reserved
-= num_bytes
;
4434 if (block_rsv
->reserved
< block_rsv
->size
)
4435 block_rsv
->full
= 0;
4438 spin_unlock(&block_rsv
->lock
);
4442 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4443 u64 num_bytes
, int update_size
)
4445 spin_lock(&block_rsv
->lock
);
4446 block_rsv
->reserved
+= num_bytes
;
4448 block_rsv
->size
+= num_bytes
;
4449 else if (block_rsv
->reserved
>= block_rsv
->size
)
4450 block_rsv
->full
= 1;
4451 spin_unlock(&block_rsv
->lock
);
4454 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4455 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4458 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4461 if (global_rsv
->space_info
!= dest
->space_info
)
4464 spin_lock(&global_rsv
->lock
);
4465 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4466 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4467 spin_unlock(&global_rsv
->lock
);
4470 global_rsv
->reserved
-= num_bytes
;
4471 if (global_rsv
->reserved
< global_rsv
->size
)
4472 global_rsv
->full
= 0;
4473 spin_unlock(&global_rsv
->lock
);
4475 block_rsv_add_bytes(dest
, num_bytes
, 1);
4479 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4480 struct btrfs_block_rsv
*block_rsv
,
4481 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4483 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4485 spin_lock(&block_rsv
->lock
);
4486 if (num_bytes
== (u64
)-1)
4487 num_bytes
= block_rsv
->size
;
4488 block_rsv
->size
-= num_bytes
;
4489 if (block_rsv
->reserved
>= block_rsv
->size
) {
4490 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4491 block_rsv
->reserved
= block_rsv
->size
;
4492 block_rsv
->full
= 1;
4496 spin_unlock(&block_rsv
->lock
);
4498 if (num_bytes
> 0) {
4500 spin_lock(&dest
->lock
);
4504 bytes_to_add
= dest
->size
- dest
->reserved
;
4505 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4506 dest
->reserved
+= bytes_to_add
;
4507 if (dest
->reserved
>= dest
->size
)
4509 num_bytes
-= bytes_to_add
;
4511 spin_unlock(&dest
->lock
);
4514 spin_lock(&space_info
->lock
);
4515 space_info
->bytes_may_use
-= num_bytes
;
4516 trace_btrfs_space_reservation(fs_info
, "space_info",
4517 space_info
->flags
, num_bytes
, 0);
4518 spin_unlock(&space_info
->lock
);
4523 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4524 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4528 ret
= block_rsv_use_bytes(src
, num_bytes
);
4532 block_rsv_add_bytes(dst
, num_bytes
, 1);
4536 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4538 memset(rsv
, 0, sizeof(*rsv
));
4539 spin_lock_init(&rsv
->lock
);
4543 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4544 unsigned short type
)
4546 struct btrfs_block_rsv
*block_rsv
;
4547 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4549 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4553 btrfs_init_block_rsv(block_rsv
, type
);
4554 block_rsv
->space_info
= __find_space_info(fs_info
,
4555 BTRFS_BLOCK_GROUP_METADATA
);
4559 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4560 struct btrfs_block_rsv
*rsv
)
4564 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4568 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4569 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4570 enum btrfs_reserve_flush_enum flush
)
4577 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4579 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4586 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4587 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4595 spin_lock(&block_rsv
->lock
);
4596 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4597 if (block_rsv
->reserved
>= num_bytes
)
4599 spin_unlock(&block_rsv
->lock
);
4604 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4605 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4606 enum btrfs_reserve_flush_enum flush
)
4614 spin_lock(&block_rsv
->lock
);
4615 num_bytes
= min_reserved
;
4616 if (block_rsv
->reserved
>= num_bytes
)
4619 num_bytes
-= block_rsv
->reserved
;
4620 spin_unlock(&block_rsv
->lock
);
4625 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4627 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4634 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4635 struct btrfs_block_rsv
*dst_rsv
,
4638 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4641 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4642 struct btrfs_block_rsv
*block_rsv
,
4645 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4646 if (global_rsv
->full
|| global_rsv
== block_rsv
||
4647 block_rsv
->space_info
!= global_rsv
->space_info
)
4649 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4654 * helper to calculate size of global block reservation.
4655 * the desired value is sum of space used by extent tree,
4656 * checksum tree and root tree
4658 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4660 struct btrfs_space_info
*sinfo
;
4664 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4666 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4667 spin_lock(&sinfo
->lock
);
4668 data_used
= sinfo
->bytes_used
;
4669 spin_unlock(&sinfo
->lock
);
4671 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4672 spin_lock(&sinfo
->lock
);
4673 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4675 meta_used
= sinfo
->bytes_used
;
4676 spin_unlock(&sinfo
->lock
);
4678 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4680 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4682 if (num_bytes
* 3 > meta_used
)
4683 num_bytes
= div64_u64(meta_used
, 3);
4685 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
4688 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4690 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4691 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4694 num_bytes
= calc_global_metadata_size(fs_info
);
4696 spin_lock(&sinfo
->lock
);
4697 spin_lock(&block_rsv
->lock
);
4699 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
4701 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4702 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4703 sinfo
->bytes_may_use
;
4705 if (sinfo
->total_bytes
> num_bytes
) {
4706 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4707 block_rsv
->reserved
+= num_bytes
;
4708 sinfo
->bytes_may_use
+= num_bytes
;
4709 trace_btrfs_space_reservation(fs_info
, "space_info",
4710 sinfo
->flags
, num_bytes
, 1);
4713 if (block_rsv
->reserved
>= block_rsv
->size
) {
4714 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4715 sinfo
->bytes_may_use
-= num_bytes
;
4716 trace_btrfs_space_reservation(fs_info
, "space_info",
4717 sinfo
->flags
, num_bytes
, 0);
4718 block_rsv
->reserved
= block_rsv
->size
;
4719 block_rsv
->full
= 1;
4722 spin_unlock(&block_rsv
->lock
);
4723 spin_unlock(&sinfo
->lock
);
4726 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4728 struct btrfs_space_info
*space_info
;
4730 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4731 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4733 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4734 fs_info
->global_block_rsv
.space_info
= space_info
;
4735 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4736 fs_info
->trans_block_rsv
.space_info
= space_info
;
4737 fs_info
->empty_block_rsv
.space_info
= space_info
;
4738 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4740 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4741 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4742 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4743 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4744 if (fs_info
->quota_root
)
4745 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
4746 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4748 update_global_block_rsv(fs_info
);
4751 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4753 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4755 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4756 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4757 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4758 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4759 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4760 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4761 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4762 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4765 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4766 struct btrfs_root
*root
)
4768 if (!trans
->block_rsv
)
4771 if (!trans
->bytes_reserved
)
4774 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4775 trans
->transid
, trans
->bytes_reserved
, 0);
4776 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4777 trans
->bytes_reserved
= 0;
4780 /* Can only return 0 or -ENOSPC */
4781 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4782 struct inode
*inode
)
4784 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4785 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4786 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4789 * We need to hold space in order to delete our orphan item once we've
4790 * added it, so this takes the reservation so we can release it later
4791 * when we are truly done with the orphan item.
4793 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4794 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4795 btrfs_ino(inode
), num_bytes
, 1);
4796 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4799 void btrfs_orphan_release_metadata(struct inode
*inode
)
4801 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4802 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4803 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4804 btrfs_ino(inode
), num_bytes
, 0);
4805 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4809 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4810 * root: the root of the parent directory
4811 * rsv: block reservation
4812 * items: the number of items that we need do reservation
4813 * qgroup_reserved: used to return the reserved size in qgroup
4815 * This function is used to reserve the space for snapshot/subvolume
4816 * creation and deletion. Those operations are different with the
4817 * common file/directory operations, they change two fs/file trees
4818 * and root tree, the number of items that the qgroup reserves is
4819 * different with the free space reservation. So we can not use
4820 * the space reseravtion mechanism in start_transaction().
4822 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
4823 struct btrfs_block_rsv
*rsv
,
4825 u64
*qgroup_reserved
,
4826 bool use_global_rsv
)
4830 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4832 if (root
->fs_info
->quota_enabled
) {
4833 /* One for parent inode, two for dir entries */
4834 num_bytes
= 3 * root
->leafsize
;
4835 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
4842 *qgroup_reserved
= num_bytes
;
4844 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
4845 rsv
->space_info
= __find_space_info(root
->fs_info
,
4846 BTRFS_BLOCK_GROUP_METADATA
);
4847 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
4848 BTRFS_RESERVE_FLUSH_ALL
);
4850 if (ret
== -ENOSPC
&& use_global_rsv
)
4851 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
4854 if (*qgroup_reserved
)
4855 btrfs_qgroup_free(root
, *qgroup_reserved
);
4861 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
4862 struct btrfs_block_rsv
*rsv
,
4863 u64 qgroup_reserved
)
4865 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4866 if (qgroup_reserved
)
4867 btrfs_qgroup_free(root
, qgroup_reserved
);
4871 * drop_outstanding_extent - drop an outstanding extent
4872 * @inode: the inode we're dropping the extent for
4874 * This is called when we are freeing up an outstanding extent, either called
4875 * after an error or after an extent is written. This will return the number of
4876 * reserved extents that need to be freed. This must be called with
4877 * BTRFS_I(inode)->lock held.
4879 static unsigned drop_outstanding_extent(struct inode
*inode
)
4881 unsigned drop_inode_space
= 0;
4882 unsigned dropped_extents
= 0;
4884 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4885 BTRFS_I(inode
)->outstanding_extents
--;
4887 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4888 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4889 &BTRFS_I(inode
)->runtime_flags
))
4890 drop_inode_space
= 1;
4893 * If we have more or the same amount of outsanding extents than we have
4894 * reserved then we need to leave the reserved extents count alone.
4896 if (BTRFS_I(inode
)->outstanding_extents
>=
4897 BTRFS_I(inode
)->reserved_extents
)
4898 return drop_inode_space
;
4900 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4901 BTRFS_I(inode
)->outstanding_extents
;
4902 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4903 return dropped_extents
+ drop_inode_space
;
4907 * calc_csum_metadata_size - return the amount of metada space that must be
4908 * reserved/free'd for the given bytes.
4909 * @inode: the inode we're manipulating
4910 * @num_bytes: the number of bytes in question
4911 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4913 * This adjusts the number of csum_bytes in the inode and then returns the
4914 * correct amount of metadata that must either be reserved or freed. We
4915 * calculate how many checksums we can fit into one leaf and then divide the
4916 * number of bytes that will need to be checksumed by this value to figure out
4917 * how many checksums will be required. If we are adding bytes then the number
4918 * may go up and we will return the number of additional bytes that must be
4919 * reserved. If it is going down we will return the number of bytes that must
4922 * This must be called with BTRFS_I(inode)->lock held.
4924 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4927 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4929 int num_csums_per_leaf
;
4933 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4934 BTRFS_I(inode
)->csum_bytes
== 0)
4937 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4939 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4941 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4942 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4943 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4944 sizeof(struct btrfs_csum_item
) +
4945 sizeof(struct btrfs_disk_key
));
4946 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4947 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4948 num_csums
= num_csums
/ num_csums_per_leaf
;
4950 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4951 old_csums
= old_csums
/ num_csums_per_leaf
;
4953 /* No change, no need to reserve more */
4954 if (old_csums
== num_csums
)
4958 return btrfs_calc_trans_metadata_size(root
,
4959 num_csums
- old_csums
);
4961 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4964 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4966 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4967 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4970 unsigned nr_extents
= 0;
4971 int extra_reserve
= 0;
4972 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
4974 bool delalloc_lock
= true;
4978 /* If we are a free space inode we need to not flush since we will be in
4979 * the middle of a transaction commit. We also don't need the delalloc
4980 * mutex since we won't race with anybody. We need this mostly to make
4981 * lockdep shut its filthy mouth.
4983 if (btrfs_is_free_space_inode(inode
)) {
4984 flush
= BTRFS_RESERVE_NO_FLUSH
;
4985 delalloc_lock
= false;
4988 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
4989 btrfs_transaction_in_commit(root
->fs_info
))
4990 schedule_timeout(1);
4993 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
4995 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4997 spin_lock(&BTRFS_I(inode
)->lock
);
4998 BTRFS_I(inode
)->outstanding_extents
++;
5000 if (BTRFS_I(inode
)->outstanding_extents
>
5001 BTRFS_I(inode
)->reserved_extents
)
5002 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5003 BTRFS_I(inode
)->reserved_extents
;
5006 * Add an item to reserve for updating the inode when we complete the
5009 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5010 &BTRFS_I(inode
)->runtime_flags
)) {
5015 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5016 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5017 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5018 spin_unlock(&BTRFS_I(inode
)->lock
);
5020 if (root
->fs_info
->quota_enabled
) {
5021 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
5022 nr_extents
* root
->leafsize
);
5027 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5028 if (unlikely(ret
)) {
5029 if (root
->fs_info
->quota_enabled
)
5030 btrfs_qgroup_free(root
, num_bytes
+
5031 nr_extents
* root
->leafsize
);
5035 spin_lock(&BTRFS_I(inode
)->lock
);
5036 if (extra_reserve
) {
5037 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5038 &BTRFS_I(inode
)->runtime_flags
);
5041 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5042 spin_unlock(&BTRFS_I(inode
)->lock
);
5045 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5048 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5049 btrfs_ino(inode
), to_reserve
, 1);
5050 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5055 spin_lock(&BTRFS_I(inode
)->lock
);
5056 dropped
= drop_outstanding_extent(inode
);
5058 * If the inodes csum_bytes is the same as the original
5059 * csum_bytes then we know we haven't raced with any free()ers
5060 * so we can just reduce our inodes csum bytes and carry on.
5062 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5063 calc_csum_metadata_size(inode
, num_bytes
, 0);
5065 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5069 * This is tricky, but first we need to figure out how much we
5070 * free'd from any free-ers that occured during this
5071 * reservation, so we reset ->csum_bytes to the csum_bytes
5072 * before we dropped our lock, and then call the free for the
5073 * number of bytes that were freed while we were trying our
5076 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5077 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5078 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5082 * Now we need to see how much we would have freed had we not
5083 * been making this reservation and our ->csum_bytes were not
5084 * artificially inflated.
5086 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5087 bytes
= csum_bytes
- orig_csum_bytes
;
5088 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5091 * Now reset ->csum_bytes to what it should be. If bytes is
5092 * more than to_free then we would have free'd more space had we
5093 * not had an artificially high ->csum_bytes, so we need to free
5094 * the remainder. If bytes is the same or less then we don't
5095 * need to do anything, the other free-ers did the correct
5098 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5099 if (bytes
> to_free
)
5100 to_free
= bytes
- to_free
;
5104 spin_unlock(&BTRFS_I(inode
)->lock
);
5106 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5109 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5110 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5111 btrfs_ino(inode
), to_free
, 0);
5114 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5119 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5120 * @inode: the inode to release the reservation for
5121 * @num_bytes: the number of bytes we're releasing
5123 * This will release the metadata reservation for an inode. This can be called
5124 * once we complete IO for a given set of bytes to release their metadata
5127 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5129 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5133 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5134 spin_lock(&BTRFS_I(inode
)->lock
);
5135 dropped
= drop_outstanding_extent(inode
);
5138 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5139 spin_unlock(&BTRFS_I(inode
)->lock
);
5141 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5143 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5144 btrfs_ino(inode
), to_free
, 0);
5145 if (root
->fs_info
->quota_enabled
) {
5146 btrfs_qgroup_free(root
, num_bytes
+
5147 dropped
* root
->leafsize
);
5150 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5155 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5156 * @inode: inode we're writing to
5157 * @num_bytes: the number of bytes we want to allocate
5159 * This will do the following things
5161 * o reserve space in the data space info for num_bytes
5162 * o reserve space in the metadata space info based on number of outstanding
5163 * extents and how much csums will be needed
5164 * o add to the inodes ->delalloc_bytes
5165 * o add it to the fs_info's delalloc inodes list.
5167 * This will return 0 for success and -ENOSPC if there is no space left.
5169 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5173 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5177 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5179 btrfs_free_reserved_data_space(inode
, num_bytes
);
5187 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5188 * @inode: inode we're releasing space for
5189 * @num_bytes: the number of bytes we want to free up
5191 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5192 * called in the case that we don't need the metadata AND data reservations
5193 * anymore. So if there is an error or we insert an inline extent.
5195 * This function will release the metadata space that was not used and will
5196 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5197 * list if there are no delalloc bytes left.
5199 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5201 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5202 btrfs_free_reserved_data_space(inode
, num_bytes
);
5205 static int update_block_group(struct btrfs_root
*root
,
5206 u64 bytenr
, u64 num_bytes
, int alloc
)
5208 struct btrfs_block_group_cache
*cache
= NULL
;
5209 struct btrfs_fs_info
*info
= root
->fs_info
;
5210 u64 total
= num_bytes
;
5215 /* block accounting for super block */
5216 spin_lock(&info
->delalloc_root_lock
);
5217 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5219 old_val
+= num_bytes
;
5221 old_val
-= num_bytes
;
5222 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5223 spin_unlock(&info
->delalloc_root_lock
);
5226 cache
= btrfs_lookup_block_group(info
, bytenr
);
5229 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5230 BTRFS_BLOCK_GROUP_RAID1
|
5231 BTRFS_BLOCK_GROUP_RAID10
))
5236 * If this block group has free space cache written out, we
5237 * need to make sure to load it if we are removing space. This
5238 * is because we need the unpinning stage to actually add the
5239 * space back to the block group, otherwise we will leak space.
5241 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5242 cache_block_group(cache
, 1);
5244 byte_in_group
= bytenr
- cache
->key
.objectid
;
5245 WARN_ON(byte_in_group
> cache
->key
.offset
);
5247 spin_lock(&cache
->space_info
->lock
);
5248 spin_lock(&cache
->lock
);
5250 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5251 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5252 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5255 old_val
= btrfs_block_group_used(&cache
->item
);
5256 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5258 old_val
+= num_bytes
;
5259 btrfs_set_block_group_used(&cache
->item
, old_val
);
5260 cache
->reserved
-= num_bytes
;
5261 cache
->space_info
->bytes_reserved
-= num_bytes
;
5262 cache
->space_info
->bytes_used
+= num_bytes
;
5263 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5264 spin_unlock(&cache
->lock
);
5265 spin_unlock(&cache
->space_info
->lock
);
5267 old_val
-= num_bytes
;
5268 btrfs_set_block_group_used(&cache
->item
, old_val
);
5269 cache
->pinned
+= num_bytes
;
5270 cache
->space_info
->bytes_pinned
+= num_bytes
;
5271 cache
->space_info
->bytes_used
-= num_bytes
;
5272 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5273 spin_unlock(&cache
->lock
);
5274 spin_unlock(&cache
->space_info
->lock
);
5276 set_extent_dirty(info
->pinned_extents
,
5277 bytenr
, bytenr
+ num_bytes
- 1,
5278 GFP_NOFS
| __GFP_NOFAIL
);
5280 btrfs_put_block_group(cache
);
5282 bytenr
+= num_bytes
;
5287 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5289 struct btrfs_block_group_cache
*cache
;
5292 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5293 bytenr
= root
->fs_info
->first_logical_byte
;
5294 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5296 if (bytenr
< (u64
)-1)
5299 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5303 bytenr
= cache
->key
.objectid
;
5304 btrfs_put_block_group(cache
);
5309 static int pin_down_extent(struct btrfs_root
*root
,
5310 struct btrfs_block_group_cache
*cache
,
5311 u64 bytenr
, u64 num_bytes
, int reserved
)
5313 spin_lock(&cache
->space_info
->lock
);
5314 spin_lock(&cache
->lock
);
5315 cache
->pinned
+= num_bytes
;
5316 cache
->space_info
->bytes_pinned
+= num_bytes
;
5318 cache
->reserved
-= num_bytes
;
5319 cache
->space_info
->bytes_reserved
-= num_bytes
;
5321 spin_unlock(&cache
->lock
);
5322 spin_unlock(&cache
->space_info
->lock
);
5324 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5325 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5327 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5332 * this function must be called within transaction
5334 int btrfs_pin_extent(struct btrfs_root
*root
,
5335 u64 bytenr
, u64 num_bytes
, int reserved
)
5337 struct btrfs_block_group_cache
*cache
;
5339 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5340 BUG_ON(!cache
); /* Logic error */
5342 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5344 btrfs_put_block_group(cache
);
5349 * this function must be called within transaction
5351 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5352 u64 bytenr
, u64 num_bytes
)
5354 struct btrfs_block_group_cache
*cache
;
5357 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5362 * pull in the free space cache (if any) so that our pin
5363 * removes the free space from the cache. We have load_only set
5364 * to one because the slow code to read in the free extents does check
5365 * the pinned extents.
5367 cache_block_group(cache
, 1);
5369 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5371 /* remove us from the free space cache (if we're there at all) */
5372 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5373 btrfs_put_block_group(cache
);
5377 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5380 struct btrfs_block_group_cache
*block_group
;
5381 struct btrfs_caching_control
*caching_ctl
;
5383 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5387 cache_block_group(block_group
, 0);
5388 caching_ctl
= get_caching_control(block_group
);
5392 BUG_ON(!block_group_cache_done(block_group
));
5393 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5395 mutex_lock(&caching_ctl
->mutex
);
5397 if (start
>= caching_ctl
->progress
) {
5398 ret
= add_excluded_extent(root
, start
, num_bytes
);
5399 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5400 ret
= btrfs_remove_free_space(block_group
,
5403 num_bytes
= caching_ctl
->progress
- start
;
5404 ret
= btrfs_remove_free_space(block_group
,
5409 num_bytes
= (start
+ num_bytes
) -
5410 caching_ctl
->progress
;
5411 start
= caching_ctl
->progress
;
5412 ret
= add_excluded_extent(root
, start
, num_bytes
);
5415 mutex_unlock(&caching_ctl
->mutex
);
5416 put_caching_control(caching_ctl
);
5418 btrfs_put_block_group(block_group
);
5422 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5423 struct extent_buffer
*eb
)
5425 struct btrfs_file_extent_item
*item
;
5426 struct btrfs_key key
;
5430 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5433 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5434 btrfs_item_key_to_cpu(eb
, &key
, i
);
5435 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5437 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5438 found_type
= btrfs_file_extent_type(eb
, item
);
5439 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5441 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5443 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5444 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5445 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5452 * btrfs_update_reserved_bytes - update the block_group and space info counters
5453 * @cache: The cache we are manipulating
5454 * @num_bytes: The number of bytes in question
5455 * @reserve: One of the reservation enums
5457 * This is called by the allocator when it reserves space, or by somebody who is
5458 * freeing space that was never actually used on disk. For example if you
5459 * reserve some space for a new leaf in transaction A and before transaction A
5460 * commits you free that leaf, you call this with reserve set to 0 in order to
5461 * clear the reservation.
5463 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5464 * ENOSPC accounting. For data we handle the reservation through clearing the
5465 * delalloc bits in the io_tree. We have to do this since we could end up
5466 * allocating less disk space for the amount of data we have reserved in the
5467 * case of compression.
5469 * If this is a reservation and the block group has become read only we cannot
5470 * make the reservation and return -EAGAIN, otherwise this function always
5473 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5474 u64 num_bytes
, int reserve
)
5476 struct btrfs_space_info
*space_info
= cache
->space_info
;
5479 spin_lock(&space_info
->lock
);
5480 spin_lock(&cache
->lock
);
5481 if (reserve
!= RESERVE_FREE
) {
5485 cache
->reserved
+= num_bytes
;
5486 space_info
->bytes_reserved
+= num_bytes
;
5487 if (reserve
== RESERVE_ALLOC
) {
5488 trace_btrfs_space_reservation(cache
->fs_info
,
5489 "space_info", space_info
->flags
,
5491 space_info
->bytes_may_use
-= num_bytes
;
5496 space_info
->bytes_readonly
+= num_bytes
;
5497 cache
->reserved
-= num_bytes
;
5498 space_info
->bytes_reserved
-= num_bytes
;
5500 spin_unlock(&cache
->lock
);
5501 spin_unlock(&space_info
->lock
);
5505 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5506 struct btrfs_root
*root
)
5508 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5509 struct btrfs_caching_control
*next
;
5510 struct btrfs_caching_control
*caching_ctl
;
5511 struct btrfs_block_group_cache
*cache
;
5512 struct btrfs_space_info
*space_info
;
5514 down_write(&fs_info
->extent_commit_sem
);
5516 list_for_each_entry_safe(caching_ctl
, next
,
5517 &fs_info
->caching_block_groups
, list
) {
5518 cache
= caching_ctl
->block_group
;
5519 if (block_group_cache_done(cache
)) {
5520 cache
->last_byte_to_unpin
= (u64
)-1;
5521 list_del_init(&caching_ctl
->list
);
5522 put_caching_control(caching_ctl
);
5524 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5528 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5529 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5531 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5533 up_write(&fs_info
->extent_commit_sem
);
5535 list_for_each_entry_rcu(space_info
, &fs_info
->space_info
, list
)
5536 percpu_counter_set(&space_info
->total_bytes_pinned
, 0);
5538 update_global_block_rsv(fs_info
);
5541 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
5543 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5544 struct btrfs_block_group_cache
*cache
= NULL
;
5545 struct btrfs_space_info
*space_info
;
5546 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5550 while (start
<= end
) {
5553 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5555 btrfs_put_block_group(cache
);
5556 cache
= btrfs_lookup_block_group(fs_info
, start
);
5557 BUG_ON(!cache
); /* Logic error */
5560 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5561 len
= min(len
, end
+ 1 - start
);
5563 if (start
< cache
->last_byte_to_unpin
) {
5564 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5565 btrfs_add_free_space(cache
, start
, len
);
5569 space_info
= cache
->space_info
;
5571 spin_lock(&space_info
->lock
);
5572 spin_lock(&cache
->lock
);
5573 cache
->pinned
-= len
;
5574 space_info
->bytes_pinned
-= len
;
5576 space_info
->bytes_readonly
+= len
;
5579 spin_unlock(&cache
->lock
);
5580 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5581 spin_lock(&global_rsv
->lock
);
5582 if (!global_rsv
->full
) {
5583 len
= min(len
, global_rsv
->size
-
5584 global_rsv
->reserved
);
5585 global_rsv
->reserved
+= len
;
5586 space_info
->bytes_may_use
+= len
;
5587 if (global_rsv
->reserved
>= global_rsv
->size
)
5588 global_rsv
->full
= 1;
5590 spin_unlock(&global_rsv
->lock
);
5592 spin_unlock(&space_info
->lock
);
5596 btrfs_put_block_group(cache
);
5600 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5601 struct btrfs_root
*root
)
5603 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5604 struct extent_io_tree
*unpin
;
5612 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5613 unpin
= &fs_info
->freed_extents
[1];
5615 unpin
= &fs_info
->freed_extents
[0];
5618 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5619 EXTENT_DIRTY
, NULL
);
5623 if (btrfs_test_opt(root
, DISCARD
))
5624 ret
= btrfs_discard_extent(root
, start
,
5625 end
+ 1 - start
, NULL
);
5627 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
5628 unpin_extent_range(root
, start
, end
);
5635 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
5636 u64 owner
, u64 root_objectid
)
5638 struct btrfs_space_info
*space_info
;
5641 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5642 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
5643 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
5645 flags
= BTRFS_BLOCK_GROUP_METADATA
;
5647 flags
= BTRFS_BLOCK_GROUP_DATA
;
5650 space_info
= __find_space_info(fs_info
, flags
);
5651 BUG_ON(!space_info
); /* Logic bug */
5652 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
5656 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5657 struct btrfs_root
*root
,
5658 u64 bytenr
, u64 num_bytes
, u64 parent
,
5659 u64 root_objectid
, u64 owner_objectid
,
5660 u64 owner_offset
, int refs_to_drop
,
5661 struct btrfs_delayed_extent_op
*extent_op
)
5663 struct btrfs_key key
;
5664 struct btrfs_path
*path
;
5665 struct btrfs_fs_info
*info
= root
->fs_info
;
5666 struct btrfs_root
*extent_root
= info
->extent_root
;
5667 struct extent_buffer
*leaf
;
5668 struct btrfs_extent_item
*ei
;
5669 struct btrfs_extent_inline_ref
*iref
;
5672 int extent_slot
= 0;
5673 int found_extent
= 0;
5677 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
5680 path
= btrfs_alloc_path();
5685 path
->leave_spinning
= 1;
5687 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5688 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5691 skinny_metadata
= 0;
5693 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5694 bytenr
, num_bytes
, parent
,
5695 root_objectid
, owner_objectid
,
5698 extent_slot
= path
->slots
[0];
5699 while (extent_slot
>= 0) {
5700 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5702 if (key
.objectid
!= bytenr
)
5704 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5705 key
.offset
== num_bytes
) {
5709 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
5710 key
.offset
== owner_objectid
) {
5714 if (path
->slots
[0] - extent_slot
> 5)
5718 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5719 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5720 if (found_extent
&& item_size
< sizeof(*ei
))
5723 if (!found_extent
) {
5725 ret
= remove_extent_backref(trans
, extent_root
, path
,
5729 btrfs_abort_transaction(trans
, extent_root
, ret
);
5732 btrfs_release_path(path
);
5733 path
->leave_spinning
= 1;
5735 key
.objectid
= bytenr
;
5736 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5737 key
.offset
= num_bytes
;
5739 if (!is_data
&& skinny_metadata
) {
5740 key
.type
= BTRFS_METADATA_ITEM_KEY
;
5741 key
.offset
= owner_objectid
;
5744 ret
= btrfs_search_slot(trans
, extent_root
,
5746 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
5748 * Couldn't find our skinny metadata item,
5749 * see if we have ye olde extent item.
5752 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5754 if (key
.objectid
== bytenr
&&
5755 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5756 key
.offset
== num_bytes
)
5760 if (ret
> 0 && skinny_metadata
) {
5761 skinny_metadata
= false;
5762 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5763 key
.offset
= num_bytes
;
5764 btrfs_release_path(path
);
5765 ret
= btrfs_search_slot(trans
, extent_root
,
5770 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5773 btrfs_print_leaf(extent_root
,
5777 btrfs_abort_transaction(trans
, extent_root
, ret
);
5780 extent_slot
= path
->slots
[0];
5782 } else if (WARN_ON(ret
== -ENOENT
)) {
5783 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5785 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5786 bytenr
, parent
, root_objectid
, owner_objectid
,
5789 btrfs_abort_transaction(trans
, extent_root
, ret
);
5793 leaf
= path
->nodes
[0];
5794 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5795 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5796 if (item_size
< sizeof(*ei
)) {
5797 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5798 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5801 btrfs_abort_transaction(trans
, extent_root
, ret
);
5805 btrfs_release_path(path
);
5806 path
->leave_spinning
= 1;
5808 key
.objectid
= bytenr
;
5809 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5810 key
.offset
= num_bytes
;
5812 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5815 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5817 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5820 btrfs_abort_transaction(trans
, extent_root
, ret
);
5824 extent_slot
= path
->slots
[0];
5825 leaf
= path
->nodes
[0];
5826 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5829 BUG_ON(item_size
< sizeof(*ei
));
5830 ei
= btrfs_item_ptr(leaf
, extent_slot
,
5831 struct btrfs_extent_item
);
5832 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
5833 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
5834 struct btrfs_tree_block_info
*bi
;
5835 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
5836 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
5837 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
5840 refs
= btrfs_extent_refs(leaf
, ei
);
5841 if (refs
< refs_to_drop
) {
5842 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
5843 "for bytenr %Lu\n", refs_to_drop
, refs
, bytenr
);
5845 btrfs_abort_transaction(trans
, extent_root
, ret
);
5848 refs
-= refs_to_drop
;
5852 __run_delayed_extent_op(extent_op
, leaf
, ei
);
5854 * In the case of inline back ref, reference count will
5855 * be updated by remove_extent_backref
5858 BUG_ON(!found_extent
);
5860 btrfs_set_extent_refs(leaf
, ei
, refs
);
5861 btrfs_mark_buffer_dirty(leaf
);
5864 ret
= remove_extent_backref(trans
, extent_root
, path
,
5868 btrfs_abort_transaction(trans
, extent_root
, ret
);
5872 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
5876 BUG_ON(is_data
&& refs_to_drop
!=
5877 extent_data_ref_count(root
, path
, iref
));
5879 BUG_ON(path
->slots
[0] != extent_slot
);
5881 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
5882 path
->slots
[0] = extent_slot
;
5887 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
5890 btrfs_abort_transaction(trans
, extent_root
, ret
);
5893 btrfs_release_path(path
);
5896 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
5898 btrfs_abort_transaction(trans
, extent_root
, ret
);
5903 ret
= update_block_group(root
, bytenr
, num_bytes
, 0);
5905 btrfs_abort_transaction(trans
, extent_root
, ret
);
5910 btrfs_free_path(path
);
5915 * when we free an block, it is possible (and likely) that we free the last
5916 * delayed ref for that extent as well. This searches the delayed ref tree for
5917 * a given extent, and if there are no other delayed refs to be processed, it
5918 * removes it from the tree.
5920 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
5921 struct btrfs_root
*root
, u64 bytenr
)
5923 struct btrfs_delayed_ref_head
*head
;
5924 struct btrfs_delayed_ref_root
*delayed_refs
;
5925 struct btrfs_delayed_ref_node
*ref
;
5926 struct rb_node
*node
;
5929 delayed_refs
= &trans
->transaction
->delayed_refs
;
5930 spin_lock(&delayed_refs
->lock
);
5931 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
5935 node
= rb_prev(&head
->node
.rb_node
);
5939 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
5941 /* there are still entries for this ref, we can't drop it */
5942 if (ref
->bytenr
== bytenr
)
5945 if (head
->extent_op
) {
5946 if (!head
->must_insert_reserved
)
5948 btrfs_free_delayed_extent_op(head
->extent_op
);
5949 head
->extent_op
= NULL
;
5953 * waiting for the lock here would deadlock. If someone else has it
5954 * locked they are already in the process of dropping it anyway
5956 if (!mutex_trylock(&head
->mutex
))
5960 * at this point we have a head with no other entries. Go
5961 * ahead and process it.
5963 head
->node
.in_tree
= 0;
5964 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
5965 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
5967 delayed_refs
->num_entries
--;
5970 * we don't take a ref on the node because we're removing it from the
5971 * tree, so we just steal the ref the tree was holding.
5973 delayed_refs
->num_heads
--;
5974 if (list_empty(&head
->cluster
))
5975 delayed_refs
->num_heads_ready
--;
5977 list_del_init(&head
->cluster
);
5978 spin_unlock(&delayed_refs
->lock
);
5980 BUG_ON(head
->extent_op
);
5981 if (head
->must_insert_reserved
)
5984 mutex_unlock(&head
->mutex
);
5985 btrfs_put_delayed_ref(&head
->node
);
5988 spin_unlock(&delayed_refs
->lock
);
5992 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
5993 struct btrfs_root
*root
,
5994 struct extent_buffer
*buf
,
5995 u64 parent
, int last_ref
)
5997 struct btrfs_block_group_cache
*cache
= NULL
;
6001 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6002 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6003 buf
->start
, buf
->len
,
6004 parent
, root
->root_key
.objectid
,
6005 btrfs_header_level(buf
),
6006 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6007 BUG_ON(ret
); /* -ENOMEM */
6013 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6015 if (btrfs_header_generation(buf
) == trans
->transid
) {
6016 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6017 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6022 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6023 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6027 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6029 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6030 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
6031 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6036 add_pinned_bytes(root
->fs_info
, buf
->len
,
6037 btrfs_header_level(buf
),
6038 root
->root_key
.objectid
);
6041 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6044 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6045 btrfs_put_block_group(cache
);
6048 /* Can return -ENOMEM */
6049 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6050 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6051 u64 owner
, u64 offset
, int for_cow
)
6054 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6056 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6059 * tree log blocks never actually go into the extent allocation
6060 * tree, just update pinning info and exit early.
6062 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6063 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6064 /* unlocks the pinned mutex */
6065 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6067 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6068 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6070 parent
, root_objectid
, (int)owner
,
6071 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
6073 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6075 parent
, root_objectid
, owner
,
6076 offset
, BTRFS_DROP_DELAYED_REF
,
6082 static u64
stripe_align(struct btrfs_root
*root
,
6083 struct btrfs_block_group_cache
*cache
,
6084 u64 val
, u64 num_bytes
)
6086 u64 ret
= ALIGN(val
, root
->stripesize
);
6091 * when we wait for progress in the block group caching, its because
6092 * our allocation attempt failed at least once. So, we must sleep
6093 * and let some progress happen before we try again.
6095 * This function will sleep at least once waiting for new free space to
6096 * show up, and then it will check the block group free space numbers
6097 * for our min num_bytes. Another option is to have it go ahead
6098 * and look in the rbtree for a free extent of a given size, but this
6101 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6102 * any of the information in this block group.
6104 static noinline
void
6105 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6108 struct btrfs_caching_control
*caching_ctl
;
6110 caching_ctl
= get_caching_control(cache
);
6114 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6115 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6117 put_caching_control(caching_ctl
);
6121 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6123 struct btrfs_caching_control
*caching_ctl
;
6126 caching_ctl
= get_caching_control(cache
);
6128 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6130 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6131 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6133 put_caching_control(caching_ctl
);
6137 int __get_raid_index(u64 flags
)
6139 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6140 return BTRFS_RAID_RAID10
;
6141 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6142 return BTRFS_RAID_RAID1
;
6143 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6144 return BTRFS_RAID_DUP
;
6145 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6146 return BTRFS_RAID_RAID0
;
6147 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6148 return BTRFS_RAID_RAID5
;
6149 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6150 return BTRFS_RAID_RAID6
;
6152 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6155 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6157 return __get_raid_index(cache
->flags
);
6160 enum btrfs_loop_type
{
6161 LOOP_CACHING_NOWAIT
= 0,
6162 LOOP_CACHING_WAIT
= 1,
6163 LOOP_ALLOC_CHUNK
= 2,
6164 LOOP_NO_EMPTY_SIZE
= 3,
6168 * walks the btree of allocated extents and find a hole of a given size.
6169 * The key ins is changed to record the hole:
6170 * ins->objectid == start position
6171 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6172 * ins->offset == the size of the hole.
6173 * Any available blocks before search_start are skipped.
6175 * If there is no suitable free space, we will record the max size of
6176 * the free space extent currently.
6178 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6179 u64 num_bytes
, u64 empty_size
,
6180 u64 hint_byte
, struct btrfs_key
*ins
,
6184 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6185 struct btrfs_free_cluster
*last_ptr
= NULL
;
6186 struct btrfs_block_group_cache
*block_group
= NULL
;
6187 struct btrfs_block_group_cache
*used_block_group
;
6188 u64 search_start
= 0;
6189 u64 max_extent_size
= 0;
6190 int empty_cluster
= 2 * 1024 * 1024;
6191 struct btrfs_space_info
*space_info
;
6193 int index
= __get_raid_index(flags
);
6194 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6195 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6196 bool found_uncached_bg
= false;
6197 bool failed_cluster_refill
= false;
6198 bool failed_alloc
= false;
6199 bool use_cluster
= true;
6200 bool have_caching_bg
= false;
6202 WARN_ON(num_bytes
< root
->sectorsize
);
6203 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
6207 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6209 space_info
= __find_space_info(root
->fs_info
, flags
);
6211 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6216 * If the space info is for both data and metadata it means we have a
6217 * small filesystem and we can't use the clustering stuff.
6219 if (btrfs_mixed_space_info(space_info
))
6220 use_cluster
= false;
6222 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6223 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6224 if (!btrfs_test_opt(root
, SSD
))
6225 empty_cluster
= 64 * 1024;
6228 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6229 btrfs_test_opt(root
, SSD
)) {
6230 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6234 spin_lock(&last_ptr
->lock
);
6235 if (last_ptr
->block_group
)
6236 hint_byte
= last_ptr
->window_start
;
6237 spin_unlock(&last_ptr
->lock
);
6240 search_start
= max(search_start
, first_logical_byte(root
, 0));
6241 search_start
= max(search_start
, hint_byte
);
6246 if (search_start
== hint_byte
) {
6247 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6249 used_block_group
= block_group
;
6251 * we don't want to use the block group if it doesn't match our
6252 * allocation bits, or if its not cached.
6254 * However if we are re-searching with an ideal block group
6255 * picked out then we don't care that the block group is cached.
6257 if (block_group
&& block_group_bits(block_group
, flags
) &&
6258 block_group
->cached
!= BTRFS_CACHE_NO
) {
6259 down_read(&space_info
->groups_sem
);
6260 if (list_empty(&block_group
->list
) ||
6263 * someone is removing this block group,
6264 * we can't jump into the have_block_group
6265 * target because our list pointers are not
6268 btrfs_put_block_group(block_group
);
6269 up_read(&space_info
->groups_sem
);
6271 index
= get_block_group_index(block_group
);
6272 goto have_block_group
;
6274 } else if (block_group
) {
6275 btrfs_put_block_group(block_group
);
6279 have_caching_bg
= false;
6280 down_read(&space_info
->groups_sem
);
6281 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6286 used_block_group
= block_group
;
6287 btrfs_get_block_group(block_group
);
6288 search_start
= block_group
->key
.objectid
;
6291 * this can happen if we end up cycling through all the
6292 * raid types, but we want to make sure we only allocate
6293 * for the proper type.
6295 if (!block_group_bits(block_group
, flags
)) {
6296 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6297 BTRFS_BLOCK_GROUP_RAID1
|
6298 BTRFS_BLOCK_GROUP_RAID5
|
6299 BTRFS_BLOCK_GROUP_RAID6
|
6300 BTRFS_BLOCK_GROUP_RAID10
;
6303 * if they asked for extra copies and this block group
6304 * doesn't provide them, bail. This does allow us to
6305 * fill raid0 from raid1.
6307 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6312 cached
= block_group_cache_done(block_group
);
6313 if (unlikely(!cached
)) {
6314 found_uncached_bg
= true;
6315 ret
= cache_block_group(block_group
, 0);
6320 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6322 if (unlikely(block_group
->ro
))
6326 * Ok we want to try and use the cluster allocator, so
6330 unsigned long aligned_cluster
;
6332 * the refill lock keeps out other
6333 * people trying to start a new cluster
6335 spin_lock(&last_ptr
->refill_lock
);
6336 used_block_group
= last_ptr
->block_group
;
6337 if (used_block_group
!= block_group
&&
6338 (!used_block_group
||
6339 used_block_group
->ro
||
6340 !block_group_bits(used_block_group
, flags
))) {
6341 used_block_group
= block_group
;
6342 goto refill_cluster
;
6345 if (used_block_group
!= block_group
)
6346 btrfs_get_block_group(used_block_group
);
6348 offset
= btrfs_alloc_from_cluster(used_block_group
,
6351 used_block_group
->key
.objectid
,
6354 /* we have a block, we're done */
6355 spin_unlock(&last_ptr
->refill_lock
);
6356 trace_btrfs_reserve_extent_cluster(root
,
6357 block_group
, search_start
, num_bytes
);
6361 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6362 if (used_block_group
!= block_group
) {
6363 btrfs_put_block_group(used_block_group
);
6364 used_block_group
= block_group
;
6367 BUG_ON(used_block_group
!= block_group
);
6368 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6369 * set up a new clusters, so lets just skip it
6370 * and let the allocator find whatever block
6371 * it can find. If we reach this point, we
6372 * will have tried the cluster allocator
6373 * plenty of times and not have found
6374 * anything, so we are likely way too
6375 * fragmented for the clustering stuff to find
6378 * However, if the cluster is taken from the
6379 * current block group, release the cluster
6380 * first, so that we stand a better chance of
6381 * succeeding in the unclustered
6383 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6384 last_ptr
->block_group
!= block_group
) {
6385 spin_unlock(&last_ptr
->refill_lock
);
6386 goto unclustered_alloc
;
6390 * this cluster didn't work out, free it and
6393 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6395 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6396 spin_unlock(&last_ptr
->refill_lock
);
6397 goto unclustered_alloc
;
6400 aligned_cluster
= max_t(unsigned long,
6401 empty_cluster
+ empty_size
,
6402 block_group
->full_stripe_len
);
6404 /* allocate a cluster in this block group */
6405 ret
= btrfs_find_space_cluster(root
, block_group
,
6406 last_ptr
, search_start
,
6411 * now pull our allocation out of this
6414 offset
= btrfs_alloc_from_cluster(block_group
,
6420 /* we found one, proceed */
6421 spin_unlock(&last_ptr
->refill_lock
);
6422 trace_btrfs_reserve_extent_cluster(root
,
6423 block_group
, search_start
,
6427 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6428 && !failed_cluster_refill
) {
6429 spin_unlock(&last_ptr
->refill_lock
);
6431 failed_cluster_refill
= true;
6432 wait_block_group_cache_progress(block_group
,
6433 num_bytes
+ empty_cluster
+ empty_size
);
6434 goto have_block_group
;
6438 * at this point we either didn't find a cluster
6439 * or we weren't able to allocate a block from our
6440 * cluster. Free the cluster we've been trying
6441 * to use, and go to the next block group
6443 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6444 spin_unlock(&last_ptr
->refill_lock
);
6449 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6451 block_group
->free_space_ctl
->free_space
<
6452 num_bytes
+ empty_cluster
+ empty_size
) {
6453 if (block_group
->free_space_ctl
->free_space
>
6456 block_group
->free_space_ctl
->free_space
;
6457 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6460 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6462 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6463 num_bytes
, empty_size
,
6466 * If we didn't find a chunk, and we haven't failed on this
6467 * block group before, and this block group is in the middle of
6468 * caching and we are ok with waiting, then go ahead and wait
6469 * for progress to be made, and set failed_alloc to true.
6471 * If failed_alloc is true then we've already waited on this
6472 * block group once and should move on to the next block group.
6474 if (!offset
&& !failed_alloc
&& !cached
&&
6475 loop
> LOOP_CACHING_NOWAIT
) {
6476 wait_block_group_cache_progress(block_group
,
6477 num_bytes
+ empty_size
);
6478 failed_alloc
= true;
6479 goto have_block_group
;
6480 } else if (!offset
) {
6482 have_caching_bg
= true;
6486 search_start
= stripe_align(root
, used_block_group
,
6489 /* move on to the next group */
6490 if (search_start
+ num_bytes
>
6491 used_block_group
->key
.objectid
+ used_block_group
->key
.offset
) {
6492 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
6496 if (offset
< search_start
)
6497 btrfs_add_free_space(used_block_group
, offset
,
6498 search_start
- offset
);
6499 BUG_ON(offset
> search_start
);
6501 ret
= btrfs_update_reserved_bytes(used_block_group
, num_bytes
,
6503 if (ret
== -EAGAIN
) {
6504 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
6508 /* we are all good, lets return */
6509 ins
->objectid
= search_start
;
6510 ins
->offset
= num_bytes
;
6512 trace_btrfs_reserve_extent(orig_root
, block_group
,
6513 search_start
, num_bytes
);
6514 if (used_block_group
!= block_group
)
6515 btrfs_put_block_group(used_block_group
);
6516 btrfs_put_block_group(block_group
);
6519 failed_cluster_refill
= false;
6520 failed_alloc
= false;
6521 BUG_ON(index
!= get_block_group_index(block_group
));
6522 if (used_block_group
!= block_group
)
6523 btrfs_put_block_group(used_block_group
);
6524 btrfs_put_block_group(block_group
);
6526 up_read(&space_info
->groups_sem
);
6528 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
6531 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
6535 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6536 * caching kthreads as we move along
6537 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6538 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6539 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6542 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
6545 if (loop
== LOOP_ALLOC_CHUNK
) {
6546 struct btrfs_trans_handle
*trans
;
6548 trans
= btrfs_join_transaction(root
);
6549 if (IS_ERR(trans
)) {
6550 ret
= PTR_ERR(trans
);
6554 ret
= do_chunk_alloc(trans
, root
, flags
,
6557 * Do not bail out on ENOSPC since we
6558 * can do more things.
6560 if (ret
< 0 && ret
!= -ENOSPC
)
6561 btrfs_abort_transaction(trans
,
6565 btrfs_end_transaction(trans
, root
);
6570 if (loop
== LOOP_NO_EMPTY_SIZE
) {
6576 } else if (!ins
->objectid
) {
6578 } else if (ins
->objectid
) {
6583 ins
->offset
= max_extent_size
;
6587 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
6588 int dump_block_groups
)
6590 struct btrfs_block_group_cache
*cache
;
6593 spin_lock(&info
->lock
);
6594 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
6596 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
6597 info
->bytes_reserved
- info
->bytes_readonly
,
6598 (info
->full
) ? "" : "not ");
6599 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
6600 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6601 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
6602 info
->bytes_reserved
, info
->bytes_may_use
,
6603 info
->bytes_readonly
);
6604 spin_unlock(&info
->lock
);
6606 if (!dump_block_groups
)
6609 down_read(&info
->groups_sem
);
6611 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6612 spin_lock(&cache
->lock
);
6613 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6614 cache
->key
.objectid
, cache
->key
.offset
,
6615 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
6616 cache
->reserved
, cache
->ro
? "[readonly]" : "");
6617 btrfs_dump_free_space(cache
, bytes
);
6618 spin_unlock(&cache
->lock
);
6620 if (++index
< BTRFS_NR_RAID_TYPES
)
6622 up_read(&info
->groups_sem
);
6625 int btrfs_reserve_extent(struct btrfs_root
*root
,
6626 u64 num_bytes
, u64 min_alloc_size
,
6627 u64 empty_size
, u64 hint_byte
,
6628 struct btrfs_key
*ins
, int is_data
)
6630 bool final_tried
= false;
6634 flags
= btrfs_get_alloc_profile(root
, is_data
);
6636 WARN_ON(num_bytes
< root
->sectorsize
);
6637 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
6640 if (ret
== -ENOSPC
) {
6641 if (!final_tried
&& ins
->offset
) {
6642 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
6643 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
6644 num_bytes
= max(num_bytes
, min_alloc_size
);
6645 if (num_bytes
== min_alloc_size
)
6648 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6649 struct btrfs_space_info
*sinfo
;
6651 sinfo
= __find_space_info(root
->fs_info
, flags
);
6652 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
6655 dump_space_info(sinfo
, num_bytes
, 1);
6662 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6663 u64 start
, u64 len
, int pin
)
6665 struct btrfs_block_group_cache
*cache
;
6668 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6670 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
6675 if (btrfs_test_opt(root
, DISCARD
))
6676 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6679 pin_down_extent(root
, cache
, start
, len
, 1);
6681 btrfs_add_free_space(cache
, start
, len
);
6682 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
6684 btrfs_put_block_group(cache
);
6686 trace_btrfs_reserved_extent_free(root
, start
, len
);
6691 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6694 return __btrfs_free_reserved_extent(root
, start
, len
, 0);
6697 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6700 return __btrfs_free_reserved_extent(root
, start
, len
, 1);
6703 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6704 struct btrfs_root
*root
,
6705 u64 parent
, u64 root_objectid
,
6706 u64 flags
, u64 owner
, u64 offset
,
6707 struct btrfs_key
*ins
, int ref_mod
)
6710 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6711 struct btrfs_extent_item
*extent_item
;
6712 struct btrfs_extent_inline_ref
*iref
;
6713 struct btrfs_path
*path
;
6714 struct extent_buffer
*leaf
;
6719 type
= BTRFS_SHARED_DATA_REF_KEY
;
6721 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6723 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6725 path
= btrfs_alloc_path();
6729 path
->leave_spinning
= 1;
6730 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6733 btrfs_free_path(path
);
6737 leaf
= path
->nodes
[0];
6738 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6739 struct btrfs_extent_item
);
6740 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
6741 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6742 btrfs_set_extent_flags(leaf
, extent_item
,
6743 flags
| BTRFS_EXTENT_FLAG_DATA
);
6745 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6746 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
6748 struct btrfs_shared_data_ref
*ref
;
6749 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
6750 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6751 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
6753 struct btrfs_extent_data_ref
*ref
;
6754 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
6755 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
6756 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
6757 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
6758 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
6761 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6762 btrfs_free_path(path
);
6764 ret
= update_block_group(root
, ins
->objectid
, ins
->offset
, 1);
6765 if (ret
) { /* -ENOENT, logic error */
6766 btrfs_err(fs_info
, "update block group failed for %llu %llu",
6767 ins
->objectid
, ins
->offset
);
6770 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
6774 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
6775 struct btrfs_root
*root
,
6776 u64 parent
, u64 root_objectid
,
6777 u64 flags
, struct btrfs_disk_key
*key
,
6778 int level
, struct btrfs_key
*ins
)
6781 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6782 struct btrfs_extent_item
*extent_item
;
6783 struct btrfs_tree_block_info
*block_info
;
6784 struct btrfs_extent_inline_ref
*iref
;
6785 struct btrfs_path
*path
;
6786 struct extent_buffer
*leaf
;
6787 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
6788 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6791 if (!skinny_metadata
)
6792 size
+= sizeof(*block_info
);
6794 path
= btrfs_alloc_path();
6796 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
6801 path
->leave_spinning
= 1;
6802 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6805 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
6807 btrfs_free_path(path
);
6811 leaf
= path
->nodes
[0];
6812 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6813 struct btrfs_extent_item
);
6814 btrfs_set_extent_refs(leaf
, extent_item
, 1);
6815 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6816 btrfs_set_extent_flags(leaf
, extent_item
,
6817 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
6819 if (skinny_metadata
) {
6820 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6822 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
6823 btrfs_set_tree_block_key(leaf
, block_info
, key
);
6824 btrfs_set_tree_block_level(leaf
, block_info
, level
);
6825 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
6829 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
6830 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6831 BTRFS_SHARED_BLOCK_REF_KEY
);
6832 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6834 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6835 BTRFS_TREE_BLOCK_REF_KEY
);
6836 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
6839 btrfs_mark_buffer_dirty(leaf
);
6840 btrfs_free_path(path
);
6842 ret
= update_block_group(root
, ins
->objectid
, root
->leafsize
, 1);
6843 if (ret
) { /* -ENOENT, logic error */
6844 btrfs_err(fs_info
, "update block group failed for %llu %llu",
6845 ins
->objectid
, ins
->offset
);
6849 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->leafsize
);
6853 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6854 struct btrfs_root
*root
,
6855 u64 root_objectid
, u64 owner
,
6856 u64 offset
, struct btrfs_key
*ins
)
6860 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
6862 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
6864 root_objectid
, owner
, offset
,
6865 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
6870 * this is used by the tree logging recovery code. It records that
6871 * an extent has been allocated and makes sure to clear the free
6872 * space cache bits as well
6874 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
6875 struct btrfs_root
*root
,
6876 u64 root_objectid
, u64 owner
, u64 offset
,
6877 struct btrfs_key
*ins
)
6880 struct btrfs_block_group_cache
*block_group
;
6883 * Mixed block groups will exclude before processing the log so we only
6884 * need to do the exlude dance if this fs isn't mixed.
6886 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
6887 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
6892 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
6896 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
6897 RESERVE_ALLOC_NO_ACCOUNT
);
6898 BUG_ON(ret
); /* logic error */
6899 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
6900 0, owner
, offset
, ins
, 1);
6901 btrfs_put_block_group(block_group
);
6905 static struct extent_buffer
*
6906 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6907 u64 bytenr
, u32 blocksize
, int level
)
6909 struct extent_buffer
*buf
;
6911 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6913 return ERR_PTR(-ENOMEM
);
6914 btrfs_set_header_generation(buf
, trans
->transid
);
6915 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
6916 btrfs_tree_lock(buf
);
6917 clean_tree_block(trans
, root
, buf
);
6918 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
6920 btrfs_set_lock_blocking(buf
);
6921 btrfs_set_buffer_uptodate(buf
);
6923 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6925 * we allow two log transactions at a time, use different
6926 * EXENT bit to differentiate dirty pages.
6928 if (root
->log_transid
% 2 == 0)
6929 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
6930 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6932 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
6933 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6935 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
6936 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6938 trans
->blocks_used
++;
6939 /* this returns a buffer locked for blocking */
6943 static struct btrfs_block_rsv
*
6944 use_block_rsv(struct btrfs_trans_handle
*trans
,
6945 struct btrfs_root
*root
, u32 blocksize
)
6947 struct btrfs_block_rsv
*block_rsv
;
6948 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
6950 bool global_updated
= false;
6952 block_rsv
= get_block_rsv(trans
, root
);
6954 if (unlikely(block_rsv
->size
== 0))
6957 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
6961 if (block_rsv
->failfast
)
6962 return ERR_PTR(ret
);
6964 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
6965 global_updated
= true;
6966 update_global_block_rsv(root
->fs_info
);
6970 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6971 static DEFINE_RATELIMIT_STATE(_rs
,
6972 DEFAULT_RATELIMIT_INTERVAL
* 10,
6973 /*DEFAULT_RATELIMIT_BURST*/ 1);
6974 if (__ratelimit(&_rs
))
6976 "btrfs: block rsv returned %d\n", ret
);
6979 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
6980 BTRFS_RESERVE_NO_FLUSH
);
6984 * If we couldn't reserve metadata bytes try and use some from
6985 * the global reserve if its space type is the same as the global
6988 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
6989 block_rsv
->space_info
== global_rsv
->space_info
) {
6990 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6994 return ERR_PTR(ret
);
6997 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
6998 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7000 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7001 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7005 * finds a free extent and does all the dirty work required for allocation
7006 * returns the key for the extent through ins, and a tree buffer for
7007 * the first block of the extent through buf.
7009 * returns the tree buffer or NULL.
7011 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
7012 struct btrfs_root
*root
, u32 blocksize
,
7013 u64 parent
, u64 root_objectid
,
7014 struct btrfs_disk_key
*key
, int level
,
7015 u64 hint
, u64 empty_size
)
7017 struct btrfs_key ins
;
7018 struct btrfs_block_rsv
*block_rsv
;
7019 struct extent_buffer
*buf
;
7022 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7025 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7026 if (IS_ERR(block_rsv
))
7027 return ERR_CAST(block_rsv
);
7029 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7030 empty_size
, hint
, &ins
, 0);
7032 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7033 return ERR_PTR(ret
);
7036 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
7038 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7040 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7042 parent
= ins
.objectid
;
7043 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7047 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7048 struct btrfs_delayed_extent_op
*extent_op
;
7049 extent_op
= btrfs_alloc_delayed_extent_op();
7050 BUG_ON(!extent_op
); /* -ENOMEM */
7052 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7054 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7055 extent_op
->flags_to_set
= flags
;
7056 if (skinny_metadata
)
7057 extent_op
->update_key
= 0;
7059 extent_op
->update_key
= 1;
7060 extent_op
->update_flags
= 1;
7061 extent_op
->is_data
= 0;
7062 extent_op
->level
= level
;
7064 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7066 ins
.offset
, parent
, root_objectid
,
7067 level
, BTRFS_ADD_DELAYED_EXTENT
,
7069 BUG_ON(ret
); /* -ENOMEM */
7074 struct walk_control
{
7075 u64 refs
[BTRFS_MAX_LEVEL
];
7076 u64 flags
[BTRFS_MAX_LEVEL
];
7077 struct btrfs_key update_progress
;
7088 #define DROP_REFERENCE 1
7089 #define UPDATE_BACKREF 2
7091 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7092 struct btrfs_root
*root
,
7093 struct walk_control
*wc
,
7094 struct btrfs_path
*path
)
7102 struct btrfs_key key
;
7103 struct extent_buffer
*eb
;
7108 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7109 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7110 wc
->reada_count
= max(wc
->reada_count
, 2);
7112 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7113 wc
->reada_count
= min_t(int, wc
->reada_count
,
7114 BTRFS_NODEPTRS_PER_BLOCK(root
));
7117 eb
= path
->nodes
[wc
->level
];
7118 nritems
= btrfs_header_nritems(eb
);
7119 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
7121 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7122 if (nread
>= wc
->reada_count
)
7126 bytenr
= btrfs_node_blockptr(eb
, slot
);
7127 generation
= btrfs_node_ptr_generation(eb
, slot
);
7129 if (slot
== path
->slots
[wc
->level
])
7132 if (wc
->stage
== UPDATE_BACKREF
&&
7133 generation
<= root
->root_key
.offset
)
7136 /* We don't lock the tree block, it's OK to be racy here */
7137 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7138 wc
->level
- 1, 1, &refs
,
7140 /* We don't care about errors in readahead. */
7145 if (wc
->stage
== DROP_REFERENCE
) {
7149 if (wc
->level
== 1 &&
7150 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7152 if (!wc
->update_ref
||
7153 generation
<= root
->root_key
.offset
)
7155 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7156 ret
= btrfs_comp_cpu_keys(&key
,
7157 &wc
->update_progress
);
7161 if (wc
->level
== 1 &&
7162 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7166 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
7172 wc
->reada_slot
= slot
;
7176 * helper to process tree block while walking down the tree.
7178 * when wc->stage == UPDATE_BACKREF, this function updates
7179 * back refs for pointers in the block.
7181 * NOTE: return value 1 means we should stop walking down.
7183 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7184 struct btrfs_root
*root
,
7185 struct btrfs_path
*path
,
7186 struct walk_control
*wc
, int lookup_info
)
7188 int level
= wc
->level
;
7189 struct extent_buffer
*eb
= path
->nodes
[level
];
7190 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7193 if (wc
->stage
== UPDATE_BACKREF
&&
7194 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7198 * when reference count of tree block is 1, it won't increase
7199 * again. once full backref flag is set, we never clear it.
7202 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7203 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7204 BUG_ON(!path
->locks
[level
]);
7205 ret
= btrfs_lookup_extent_info(trans
, root
,
7206 eb
->start
, level
, 1,
7209 BUG_ON(ret
== -ENOMEM
);
7212 BUG_ON(wc
->refs
[level
] == 0);
7215 if (wc
->stage
== DROP_REFERENCE
) {
7216 if (wc
->refs
[level
] > 1)
7219 if (path
->locks
[level
] && !wc
->keep_locks
) {
7220 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7221 path
->locks
[level
] = 0;
7226 /* wc->stage == UPDATE_BACKREF */
7227 if (!(wc
->flags
[level
] & flag
)) {
7228 BUG_ON(!path
->locks
[level
]);
7229 ret
= btrfs_inc_ref(trans
, root
, eb
, 1, wc
->for_reloc
);
7230 BUG_ON(ret
); /* -ENOMEM */
7231 ret
= btrfs_dec_ref(trans
, root
, eb
, 0, wc
->for_reloc
);
7232 BUG_ON(ret
); /* -ENOMEM */
7233 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7235 btrfs_header_level(eb
), 0);
7236 BUG_ON(ret
); /* -ENOMEM */
7237 wc
->flags
[level
] |= flag
;
7241 * the block is shared by multiple trees, so it's not good to
7242 * keep the tree lock
7244 if (path
->locks
[level
] && level
> 0) {
7245 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7246 path
->locks
[level
] = 0;
7252 * helper to process tree block pointer.
7254 * when wc->stage == DROP_REFERENCE, this function checks
7255 * reference count of the block pointed to. if the block
7256 * is shared and we need update back refs for the subtree
7257 * rooted at the block, this function changes wc->stage to
7258 * UPDATE_BACKREF. if the block is shared and there is no
7259 * need to update back, this function drops the reference
7262 * NOTE: return value 1 means we should stop walking down.
7264 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
7265 struct btrfs_root
*root
,
7266 struct btrfs_path
*path
,
7267 struct walk_control
*wc
, int *lookup_info
)
7273 struct btrfs_key key
;
7274 struct extent_buffer
*next
;
7275 int level
= wc
->level
;
7279 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
7280 path
->slots
[level
]);
7282 * if the lower level block was created before the snapshot
7283 * was created, we know there is no need to update back refs
7286 if (wc
->stage
== UPDATE_BACKREF
&&
7287 generation
<= root
->root_key
.offset
) {
7292 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
7293 blocksize
= btrfs_level_size(root
, level
- 1);
7295 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
7297 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
7300 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
7304 btrfs_tree_lock(next
);
7305 btrfs_set_lock_blocking(next
);
7307 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
7308 &wc
->refs
[level
- 1],
7309 &wc
->flags
[level
- 1]);
7311 btrfs_tree_unlock(next
);
7315 if (unlikely(wc
->refs
[level
- 1] == 0)) {
7316 btrfs_err(root
->fs_info
, "Missing references.");
7321 if (wc
->stage
== DROP_REFERENCE
) {
7322 if (wc
->refs
[level
- 1] > 1) {
7324 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7327 if (!wc
->update_ref
||
7328 generation
<= root
->root_key
.offset
)
7331 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
7332 path
->slots
[level
]);
7333 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
7337 wc
->stage
= UPDATE_BACKREF
;
7338 wc
->shared_level
= level
- 1;
7342 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7346 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
7347 btrfs_tree_unlock(next
);
7348 free_extent_buffer(next
);
7354 if (reada
&& level
== 1)
7355 reada_walk_down(trans
, root
, wc
, path
);
7356 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
7357 if (!next
|| !extent_buffer_uptodate(next
)) {
7358 free_extent_buffer(next
);
7361 btrfs_tree_lock(next
);
7362 btrfs_set_lock_blocking(next
);
7366 BUG_ON(level
!= btrfs_header_level(next
));
7367 path
->nodes
[level
] = next
;
7368 path
->slots
[level
] = 0;
7369 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7375 wc
->refs
[level
- 1] = 0;
7376 wc
->flags
[level
- 1] = 0;
7377 if (wc
->stage
== DROP_REFERENCE
) {
7378 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
7379 parent
= path
->nodes
[level
]->start
;
7381 BUG_ON(root
->root_key
.objectid
!=
7382 btrfs_header_owner(path
->nodes
[level
]));
7386 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
7387 root
->root_key
.objectid
, level
- 1, 0, 0);
7388 BUG_ON(ret
); /* -ENOMEM */
7390 btrfs_tree_unlock(next
);
7391 free_extent_buffer(next
);
7397 * helper to process tree block while walking up the tree.
7399 * when wc->stage == DROP_REFERENCE, this function drops
7400 * reference count on the block.
7402 * when wc->stage == UPDATE_BACKREF, this function changes
7403 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7404 * to UPDATE_BACKREF previously while processing the block.
7406 * NOTE: return value 1 means we should stop walking up.
7408 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
7409 struct btrfs_root
*root
,
7410 struct btrfs_path
*path
,
7411 struct walk_control
*wc
)
7414 int level
= wc
->level
;
7415 struct extent_buffer
*eb
= path
->nodes
[level
];
7418 if (wc
->stage
== UPDATE_BACKREF
) {
7419 BUG_ON(wc
->shared_level
< level
);
7420 if (level
< wc
->shared_level
)
7423 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
7427 wc
->stage
= DROP_REFERENCE
;
7428 wc
->shared_level
= -1;
7429 path
->slots
[level
] = 0;
7432 * check reference count again if the block isn't locked.
7433 * we should start walking down the tree again if reference
7436 if (!path
->locks
[level
]) {
7438 btrfs_tree_lock(eb
);
7439 btrfs_set_lock_blocking(eb
);
7440 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7442 ret
= btrfs_lookup_extent_info(trans
, root
,
7443 eb
->start
, level
, 1,
7447 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7448 path
->locks
[level
] = 0;
7451 BUG_ON(wc
->refs
[level
] == 0);
7452 if (wc
->refs
[level
] == 1) {
7453 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7454 path
->locks
[level
] = 0;
7460 /* wc->stage == DROP_REFERENCE */
7461 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
7463 if (wc
->refs
[level
] == 1) {
7465 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7466 ret
= btrfs_dec_ref(trans
, root
, eb
, 1,
7469 ret
= btrfs_dec_ref(trans
, root
, eb
, 0,
7471 BUG_ON(ret
); /* -ENOMEM */
7473 /* make block locked assertion in clean_tree_block happy */
7474 if (!path
->locks
[level
] &&
7475 btrfs_header_generation(eb
) == trans
->transid
) {
7476 btrfs_tree_lock(eb
);
7477 btrfs_set_lock_blocking(eb
);
7478 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7480 clean_tree_block(trans
, root
, eb
);
7483 if (eb
== root
->node
) {
7484 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7487 BUG_ON(root
->root_key
.objectid
!=
7488 btrfs_header_owner(eb
));
7490 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7491 parent
= path
->nodes
[level
+ 1]->start
;
7493 BUG_ON(root
->root_key
.objectid
!=
7494 btrfs_header_owner(path
->nodes
[level
+ 1]));
7497 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
7499 wc
->refs
[level
] = 0;
7500 wc
->flags
[level
] = 0;
7504 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
7505 struct btrfs_root
*root
,
7506 struct btrfs_path
*path
,
7507 struct walk_control
*wc
)
7509 int level
= wc
->level
;
7510 int lookup_info
= 1;
7513 while (level
>= 0) {
7514 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
7521 if (path
->slots
[level
] >=
7522 btrfs_header_nritems(path
->nodes
[level
]))
7525 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
7527 path
->slots
[level
]++;
7536 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
7537 struct btrfs_root
*root
,
7538 struct btrfs_path
*path
,
7539 struct walk_control
*wc
, int max_level
)
7541 int level
= wc
->level
;
7544 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
7545 while (level
< max_level
&& path
->nodes
[level
]) {
7547 if (path
->slots
[level
] + 1 <
7548 btrfs_header_nritems(path
->nodes
[level
])) {
7549 path
->slots
[level
]++;
7552 ret
= walk_up_proc(trans
, root
, path
, wc
);
7556 if (path
->locks
[level
]) {
7557 btrfs_tree_unlock_rw(path
->nodes
[level
],
7558 path
->locks
[level
]);
7559 path
->locks
[level
] = 0;
7561 free_extent_buffer(path
->nodes
[level
]);
7562 path
->nodes
[level
] = NULL
;
7570 * drop a subvolume tree.
7572 * this function traverses the tree freeing any blocks that only
7573 * referenced by the tree.
7575 * when a shared tree block is found. this function decreases its
7576 * reference count by one. if update_ref is true, this function
7577 * also make sure backrefs for the shared block and all lower level
7578 * blocks are properly updated.
7580 * If called with for_reloc == 0, may exit early with -EAGAIN
7582 int btrfs_drop_snapshot(struct btrfs_root
*root
,
7583 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
7586 struct btrfs_path
*path
;
7587 struct btrfs_trans_handle
*trans
;
7588 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7589 struct btrfs_root_item
*root_item
= &root
->root_item
;
7590 struct walk_control
*wc
;
7591 struct btrfs_key key
;
7595 bool root_dropped
= false;
7597 path
= btrfs_alloc_path();
7603 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7605 btrfs_free_path(path
);
7610 trans
= btrfs_start_transaction(tree_root
, 0);
7611 if (IS_ERR(trans
)) {
7612 err
= PTR_ERR(trans
);
7617 trans
->block_rsv
= block_rsv
;
7619 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
7620 level
= btrfs_header_level(root
->node
);
7621 path
->nodes
[level
] = btrfs_lock_root_node(root
);
7622 btrfs_set_lock_blocking(path
->nodes
[level
]);
7623 path
->slots
[level
] = 0;
7624 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7625 memset(&wc
->update_progress
, 0,
7626 sizeof(wc
->update_progress
));
7628 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
7629 memcpy(&wc
->update_progress
, &key
,
7630 sizeof(wc
->update_progress
));
7632 level
= root_item
->drop_level
;
7634 path
->lowest_level
= level
;
7635 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7636 path
->lowest_level
= 0;
7644 * unlock our path, this is safe because only this
7645 * function is allowed to delete this snapshot
7647 btrfs_unlock_up_safe(path
, 0);
7649 level
= btrfs_header_level(root
->node
);
7651 btrfs_tree_lock(path
->nodes
[level
]);
7652 btrfs_set_lock_blocking(path
->nodes
[level
]);
7653 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7655 ret
= btrfs_lookup_extent_info(trans
, root
,
7656 path
->nodes
[level
]->start
,
7657 level
, 1, &wc
->refs
[level
],
7663 BUG_ON(wc
->refs
[level
] == 0);
7665 if (level
== root_item
->drop_level
)
7668 btrfs_tree_unlock(path
->nodes
[level
]);
7669 path
->locks
[level
] = 0;
7670 WARN_ON(wc
->refs
[level
] != 1);
7676 wc
->shared_level
= -1;
7677 wc
->stage
= DROP_REFERENCE
;
7678 wc
->update_ref
= update_ref
;
7680 wc
->for_reloc
= for_reloc
;
7681 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7685 ret
= walk_down_tree(trans
, root
, path
, wc
);
7691 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
7698 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
7702 if (wc
->stage
== DROP_REFERENCE
) {
7704 btrfs_node_key(path
->nodes
[level
],
7705 &root_item
->drop_progress
,
7706 path
->slots
[level
]);
7707 root_item
->drop_level
= level
;
7710 BUG_ON(wc
->level
== 0);
7711 if (btrfs_should_end_transaction(trans
, tree_root
) ||
7712 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
7713 ret
= btrfs_update_root(trans
, tree_root
,
7717 btrfs_abort_transaction(trans
, tree_root
, ret
);
7722 btrfs_end_transaction_throttle(trans
, tree_root
);
7723 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
7724 pr_debug("btrfs: drop snapshot early exit\n");
7729 trans
= btrfs_start_transaction(tree_root
, 0);
7730 if (IS_ERR(trans
)) {
7731 err
= PTR_ERR(trans
);
7735 trans
->block_rsv
= block_rsv
;
7738 btrfs_release_path(path
);
7742 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
7744 btrfs_abort_transaction(trans
, tree_root
, ret
);
7748 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
7749 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
7752 btrfs_abort_transaction(trans
, tree_root
, ret
);
7755 } else if (ret
> 0) {
7756 /* if we fail to delete the orphan item this time
7757 * around, it'll get picked up the next time.
7759 * The most common failure here is just -ENOENT.
7761 btrfs_del_orphan_item(trans
, tree_root
,
7762 root
->root_key
.objectid
);
7766 if (root
->in_radix
) {
7767 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
7769 free_extent_buffer(root
->node
);
7770 free_extent_buffer(root
->commit_root
);
7771 btrfs_put_fs_root(root
);
7773 root_dropped
= true;
7775 btrfs_end_transaction_throttle(trans
, tree_root
);
7778 btrfs_free_path(path
);
7781 * So if we need to stop dropping the snapshot for whatever reason we
7782 * need to make sure to add it back to the dead root list so that we
7783 * keep trying to do the work later. This also cleans up roots if we
7784 * don't have it in the radix (like when we recover after a power fail
7785 * or unmount) so we don't leak memory.
7787 if (!for_reloc
&& root_dropped
== false)
7788 btrfs_add_dead_root(root
);
7790 btrfs_std_error(root
->fs_info
, err
);
7795 * drop subtree rooted at tree block 'node'.
7797 * NOTE: this function will unlock and release tree block 'node'
7798 * only used by relocation code
7800 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
7801 struct btrfs_root
*root
,
7802 struct extent_buffer
*node
,
7803 struct extent_buffer
*parent
)
7805 struct btrfs_path
*path
;
7806 struct walk_control
*wc
;
7812 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7814 path
= btrfs_alloc_path();
7818 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7820 btrfs_free_path(path
);
7824 btrfs_assert_tree_locked(parent
);
7825 parent_level
= btrfs_header_level(parent
);
7826 extent_buffer_get(parent
);
7827 path
->nodes
[parent_level
] = parent
;
7828 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
7830 btrfs_assert_tree_locked(node
);
7831 level
= btrfs_header_level(node
);
7832 path
->nodes
[level
] = node
;
7833 path
->slots
[level
] = 0;
7834 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7836 wc
->refs
[parent_level
] = 1;
7837 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7839 wc
->shared_level
= -1;
7840 wc
->stage
= DROP_REFERENCE
;
7844 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7847 wret
= walk_down_tree(trans
, root
, path
, wc
);
7853 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
7861 btrfs_free_path(path
);
7865 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
7871 * if restripe for this chunk_type is on pick target profile and
7872 * return, otherwise do the usual balance
7874 stripped
= get_restripe_target(root
->fs_info
, flags
);
7876 return extended_to_chunk(stripped
);
7879 * we add in the count of missing devices because we want
7880 * to make sure that any RAID levels on a degraded FS
7881 * continue to be honored.
7883 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
7884 root
->fs_info
->fs_devices
->missing_devices
;
7886 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
7887 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
7888 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
7890 if (num_devices
== 1) {
7891 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7892 stripped
= flags
& ~stripped
;
7894 /* turn raid0 into single device chunks */
7895 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7898 /* turn mirroring into duplication */
7899 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7900 BTRFS_BLOCK_GROUP_RAID10
))
7901 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7903 /* they already had raid on here, just return */
7904 if (flags
& stripped
)
7907 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7908 stripped
= flags
& ~stripped
;
7910 /* switch duplicated blocks with raid1 */
7911 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7912 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7914 /* this is drive concat, leave it alone */
7920 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
7922 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7924 u64 min_allocable_bytes
;
7929 * We need some metadata space and system metadata space for
7930 * allocating chunks in some corner cases until we force to set
7931 * it to be readonly.
7934 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
7936 min_allocable_bytes
= 1 * 1024 * 1024;
7938 min_allocable_bytes
= 0;
7940 spin_lock(&sinfo
->lock
);
7941 spin_lock(&cache
->lock
);
7948 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7949 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7951 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
7952 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
7953 min_allocable_bytes
<= sinfo
->total_bytes
) {
7954 sinfo
->bytes_readonly
+= num_bytes
;
7959 spin_unlock(&cache
->lock
);
7960 spin_unlock(&sinfo
->lock
);
7964 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
7965 struct btrfs_block_group_cache
*cache
)
7968 struct btrfs_trans_handle
*trans
;
7974 trans
= btrfs_join_transaction(root
);
7976 return PTR_ERR(trans
);
7978 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
7979 if (alloc_flags
!= cache
->flags
) {
7980 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7986 ret
= set_block_group_ro(cache
, 0);
7989 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
7990 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7994 ret
= set_block_group_ro(cache
, 0);
7996 btrfs_end_transaction(trans
, root
);
8000 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8001 struct btrfs_root
*root
, u64 type
)
8003 u64 alloc_flags
= get_alloc_profile(root
, type
);
8004 return do_chunk_alloc(trans
, root
, alloc_flags
,
8009 * helper to account the unused space of all the readonly block group in the
8010 * list. takes mirrors into account.
8012 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
8014 struct btrfs_block_group_cache
*block_group
;
8018 list_for_each_entry(block_group
, groups_list
, list
) {
8019 spin_lock(&block_group
->lock
);
8021 if (!block_group
->ro
) {
8022 spin_unlock(&block_group
->lock
);
8026 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8027 BTRFS_BLOCK_GROUP_RAID10
|
8028 BTRFS_BLOCK_GROUP_DUP
))
8033 free_bytes
+= (block_group
->key
.offset
-
8034 btrfs_block_group_used(&block_group
->item
)) *
8037 spin_unlock(&block_group
->lock
);
8044 * helper to account the unused space of all the readonly block group in the
8045 * space_info. takes mirrors into account.
8047 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8052 spin_lock(&sinfo
->lock
);
8054 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
8055 if (!list_empty(&sinfo
->block_groups
[i
]))
8056 free_bytes
+= __btrfs_get_ro_block_group_free_space(
8057 &sinfo
->block_groups
[i
]);
8059 spin_unlock(&sinfo
->lock
);
8064 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8065 struct btrfs_block_group_cache
*cache
)
8067 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8072 spin_lock(&sinfo
->lock
);
8073 spin_lock(&cache
->lock
);
8074 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8075 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8076 sinfo
->bytes_readonly
-= num_bytes
;
8078 spin_unlock(&cache
->lock
);
8079 spin_unlock(&sinfo
->lock
);
8083 * checks to see if its even possible to relocate this block group.
8085 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8086 * ok to go ahead and try.
8088 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8090 struct btrfs_block_group_cache
*block_group
;
8091 struct btrfs_space_info
*space_info
;
8092 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8093 struct btrfs_device
*device
;
8094 struct btrfs_trans_handle
*trans
;
8103 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8105 /* odd, couldn't find the block group, leave it alone */
8109 min_free
= btrfs_block_group_used(&block_group
->item
);
8111 /* no bytes used, we're good */
8115 space_info
= block_group
->space_info
;
8116 spin_lock(&space_info
->lock
);
8118 full
= space_info
->full
;
8121 * if this is the last block group we have in this space, we can't
8122 * relocate it unless we're able to allocate a new chunk below.
8124 * Otherwise, we need to make sure we have room in the space to handle
8125 * all of the extents from this block group. If we can, we're good
8127 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8128 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8129 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8130 min_free
< space_info
->total_bytes
)) {
8131 spin_unlock(&space_info
->lock
);
8134 spin_unlock(&space_info
->lock
);
8137 * ok we don't have enough space, but maybe we have free space on our
8138 * devices to allocate new chunks for relocation, so loop through our
8139 * alloc devices and guess if we have enough space. if this block
8140 * group is going to be restriped, run checks against the target
8141 * profile instead of the current one.
8153 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8155 index
= __get_raid_index(extended_to_chunk(target
));
8158 * this is just a balance, so if we were marked as full
8159 * we know there is no space for a new chunk
8164 index
= get_block_group_index(block_group
);
8167 if (index
== BTRFS_RAID_RAID10
) {
8171 } else if (index
== BTRFS_RAID_RAID1
) {
8173 } else if (index
== BTRFS_RAID_DUP
) {
8176 } else if (index
== BTRFS_RAID_RAID0
) {
8177 dev_min
= fs_devices
->rw_devices
;
8178 do_div(min_free
, dev_min
);
8181 /* We need to do this so that we can look at pending chunks */
8182 trans
= btrfs_join_transaction(root
);
8183 if (IS_ERR(trans
)) {
8184 ret
= PTR_ERR(trans
);
8188 mutex_lock(&root
->fs_info
->chunk_mutex
);
8189 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8193 * check to make sure we can actually find a chunk with enough
8194 * space to fit our block group in.
8196 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8197 !device
->is_tgtdev_for_dev_replace
) {
8198 ret
= find_free_dev_extent(trans
, device
, min_free
,
8203 if (dev_nr
>= dev_min
)
8209 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8210 btrfs_end_transaction(trans
, root
);
8212 btrfs_put_block_group(block_group
);
8216 static int find_first_block_group(struct btrfs_root
*root
,
8217 struct btrfs_path
*path
, struct btrfs_key
*key
)
8220 struct btrfs_key found_key
;
8221 struct extent_buffer
*leaf
;
8224 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8229 slot
= path
->slots
[0];
8230 leaf
= path
->nodes
[0];
8231 if (slot
>= btrfs_header_nritems(leaf
)) {
8232 ret
= btrfs_next_leaf(root
, path
);
8239 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8241 if (found_key
.objectid
>= key
->objectid
&&
8242 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8252 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8254 struct btrfs_block_group_cache
*block_group
;
8258 struct inode
*inode
;
8260 block_group
= btrfs_lookup_first_block_group(info
, last
);
8261 while (block_group
) {
8262 spin_lock(&block_group
->lock
);
8263 if (block_group
->iref
)
8265 spin_unlock(&block_group
->lock
);
8266 block_group
= next_block_group(info
->tree_root
,
8276 inode
= block_group
->inode
;
8277 block_group
->iref
= 0;
8278 block_group
->inode
= NULL
;
8279 spin_unlock(&block_group
->lock
);
8281 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8282 btrfs_put_block_group(block_group
);
8286 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8288 struct btrfs_block_group_cache
*block_group
;
8289 struct btrfs_space_info
*space_info
;
8290 struct btrfs_caching_control
*caching_ctl
;
8293 down_write(&info
->extent_commit_sem
);
8294 while (!list_empty(&info
->caching_block_groups
)) {
8295 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8296 struct btrfs_caching_control
, list
);
8297 list_del(&caching_ctl
->list
);
8298 put_caching_control(caching_ctl
);
8300 up_write(&info
->extent_commit_sem
);
8302 spin_lock(&info
->block_group_cache_lock
);
8303 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8304 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8306 rb_erase(&block_group
->cache_node
,
8307 &info
->block_group_cache_tree
);
8308 spin_unlock(&info
->block_group_cache_lock
);
8310 down_write(&block_group
->space_info
->groups_sem
);
8311 list_del(&block_group
->list
);
8312 up_write(&block_group
->space_info
->groups_sem
);
8314 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8315 wait_block_group_cache_done(block_group
);
8318 * We haven't cached this block group, which means we could
8319 * possibly have excluded extents on this block group.
8321 if (block_group
->cached
== BTRFS_CACHE_NO
||
8322 block_group
->cached
== BTRFS_CACHE_ERROR
)
8323 free_excluded_extents(info
->extent_root
, block_group
);
8325 btrfs_remove_free_space_cache(block_group
);
8326 btrfs_put_block_group(block_group
);
8328 spin_lock(&info
->block_group_cache_lock
);
8330 spin_unlock(&info
->block_group_cache_lock
);
8332 /* now that all the block groups are freed, go through and
8333 * free all the space_info structs. This is only called during
8334 * the final stages of unmount, and so we know nobody is
8335 * using them. We call synchronize_rcu() once before we start,
8336 * just to be on the safe side.
8340 release_global_block_rsv(info
);
8342 while (!list_empty(&info
->space_info
)) {
8343 space_info
= list_entry(info
->space_info
.next
,
8344 struct btrfs_space_info
,
8346 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
8347 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
8348 space_info
->bytes_reserved
> 0 ||
8349 space_info
->bytes_may_use
> 0)) {
8350 dump_space_info(space_info
, 0, 0);
8353 percpu_counter_destroy(&space_info
->total_bytes_pinned
);
8354 list_del(&space_info
->list
);
8360 static void __link_block_group(struct btrfs_space_info
*space_info
,
8361 struct btrfs_block_group_cache
*cache
)
8363 int index
= get_block_group_index(cache
);
8365 down_write(&space_info
->groups_sem
);
8366 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8367 up_write(&space_info
->groups_sem
);
8370 int btrfs_read_block_groups(struct btrfs_root
*root
)
8372 struct btrfs_path
*path
;
8374 struct btrfs_block_group_cache
*cache
;
8375 struct btrfs_fs_info
*info
= root
->fs_info
;
8376 struct btrfs_space_info
*space_info
;
8377 struct btrfs_key key
;
8378 struct btrfs_key found_key
;
8379 struct extent_buffer
*leaf
;
8383 root
= info
->extent_root
;
8386 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8387 path
= btrfs_alloc_path();
8392 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
8393 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
8394 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
8396 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8400 ret
= find_first_block_group(root
, path
, &key
);
8405 leaf
= path
->nodes
[0];
8406 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8407 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8412 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8414 if (!cache
->free_space_ctl
) {
8420 atomic_set(&cache
->count
, 1);
8421 spin_lock_init(&cache
->lock
);
8422 cache
->fs_info
= info
;
8423 INIT_LIST_HEAD(&cache
->list
);
8424 INIT_LIST_HEAD(&cache
->cluster_list
);
8428 * When we mount with old space cache, we need to
8429 * set BTRFS_DC_CLEAR and set dirty flag.
8431 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8432 * truncate the old free space cache inode and
8434 * b) Setting 'dirty flag' makes sure that we flush
8435 * the new space cache info onto disk.
8437 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8438 if (btrfs_test_opt(root
, SPACE_CACHE
))
8442 read_extent_buffer(leaf
, &cache
->item
,
8443 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8444 sizeof(cache
->item
));
8445 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
8447 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8448 btrfs_release_path(path
);
8449 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8450 cache
->sectorsize
= root
->sectorsize
;
8451 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8452 &root
->fs_info
->mapping_tree
,
8453 found_key
.objectid
);
8454 btrfs_init_free_space_ctl(cache
);
8457 * We need to exclude the super stripes now so that the space
8458 * info has super bytes accounted for, otherwise we'll think
8459 * we have more space than we actually do.
8461 ret
= exclude_super_stripes(root
, cache
);
8464 * We may have excluded something, so call this just in
8467 free_excluded_extents(root
, cache
);
8468 kfree(cache
->free_space_ctl
);
8474 * check for two cases, either we are full, and therefore
8475 * don't need to bother with the caching work since we won't
8476 * find any space, or we are empty, and we can just add all
8477 * the space in and be done with it. This saves us _alot_ of
8478 * time, particularly in the full case.
8480 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8481 cache
->last_byte_to_unpin
= (u64
)-1;
8482 cache
->cached
= BTRFS_CACHE_FINISHED
;
8483 free_excluded_extents(root
, cache
);
8484 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8485 cache
->last_byte_to_unpin
= (u64
)-1;
8486 cache
->cached
= BTRFS_CACHE_FINISHED
;
8487 add_new_free_space(cache
, root
->fs_info
,
8489 found_key
.objectid
+
8491 free_excluded_extents(root
, cache
);
8494 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8496 btrfs_remove_free_space_cache(cache
);
8497 btrfs_put_block_group(cache
);
8501 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8502 btrfs_block_group_used(&cache
->item
),
8505 btrfs_remove_free_space_cache(cache
);
8506 spin_lock(&info
->block_group_cache_lock
);
8507 rb_erase(&cache
->cache_node
,
8508 &info
->block_group_cache_tree
);
8509 spin_unlock(&info
->block_group_cache_lock
);
8510 btrfs_put_block_group(cache
);
8514 cache
->space_info
= space_info
;
8515 spin_lock(&cache
->space_info
->lock
);
8516 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8517 spin_unlock(&cache
->space_info
->lock
);
8519 __link_block_group(space_info
, cache
);
8521 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8522 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8523 set_block_group_ro(cache
, 1);
8526 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8527 if (!(get_alloc_profile(root
, space_info
->flags
) &
8528 (BTRFS_BLOCK_GROUP_RAID10
|
8529 BTRFS_BLOCK_GROUP_RAID1
|
8530 BTRFS_BLOCK_GROUP_RAID5
|
8531 BTRFS_BLOCK_GROUP_RAID6
|
8532 BTRFS_BLOCK_GROUP_DUP
)))
8535 * avoid allocating from un-mirrored block group if there are
8536 * mirrored block groups.
8538 list_for_each_entry(cache
,
8539 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
8541 set_block_group_ro(cache
, 1);
8542 list_for_each_entry(cache
,
8543 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
8545 set_block_group_ro(cache
, 1);
8548 init_global_block_rsv(info
);
8551 btrfs_free_path(path
);
8555 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
8556 struct btrfs_root
*root
)
8558 struct btrfs_block_group_cache
*block_group
, *tmp
;
8559 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
8560 struct btrfs_block_group_item item
;
8561 struct btrfs_key key
;
8564 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
,
8566 list_del_init(&block_group
->new_bg_list
);
8571 spin_lock(&block_group
->lock
);
8572 memcpy(&item
, &block_group
->item
, sizeof(item
));
8573 memcpy(&key
, &block_group
->key
, sizeof(key
));
8574 spin_unlock(&block_group
->lock
);
8576 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
8579 btrfs_abort_transaction(trans
, extent_root
, ret
);
8580 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
8581 key
.objectid
, key
.offset
);
8583 btrfs_abort_transaction(trans
, extent_root
, ret
);
8587 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8588 struct btrfs_root
*root
, u64 bytes_used
,
8589 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8593 struct btrfs_root
*extent_root
;
8594 struct btrfs_block_group_cache
*cache
;
8596 extent_root
= root
->fs_info
->extent_root
;
8598 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8600 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8603 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8605 if (!cache
->free_space_ctl
) {
8610 cache
->key
.objectid
= chunk_offset
;
8611 cache
->key
.offset
= size
;
8612 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8613 cache
->sectorsize
= root
->sectorsize
;
8614 cache
->fs_info
= root
->fs_info
;
8615 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8616 &root
->fs_info
->mapping_tree
,
8619 atomic_set(&cache
->count
, 1);
8620 spin_lock_init(&cache
->lock
);
8621 INIT_LIST_HEAD(&cache
->list
);
8622 INIT_LIST_HEAD(&cache
->cluster_list
);
8623 INIT_LIST_HEAD(&cache
->new_bg_list
);
8625 btrfs_init_free_space_ctl(cache
);
8627 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8628 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8629 cache
->flags
= type
;
8630 btrfs_set_block_group_flags(&cache
->item
, type
);
8632 cache
->last_byte_to_unpin
= (u64
)-1;
8633 cache
->cached
= BTRFS_CACHE_FINISHED
;
8634 ret
= exclude_super_stripes(root
, cache
);
8637 * We may have excluded something, so call this just in
8640 free_excluded_extents(root
, cache
);
8641 kfree(cache
->free_space_ctl
);
8646 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8647 chunk_offset
+ size
);
8649 free_excluded_extents(root
, cache
);
8651 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8653 btrfs_remove_free_space_cache(cache
);
8654 btrfs_put_block_group(cache
);
8658 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8659 &cache
->space_info
);
8661 btrfs_remove_free_space_cache(cache
);
8662 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8663 rb_erase(&cache
->cache_node
,
8664 &root
->fs_info
->block_group_cache_tree
);
8665 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8666 btrfs_put_block_group(cache
);
8669 update_global_block_rsv(root
->fs_info
);
8671 spin_lock(&cache
->space_info
->lock
);
8672 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8673 spin_unlock(&cache
->space_info
->lock
);
8675 __link_block_group(cache
->space_info
, cache
);
8677 list_add_tail(&cache
->new_bg_list
, &trans
->new_bgs
);
8679 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8684 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
8686 u64 extra_flags
= chunk_to_extended(flags
) &
8687 BTRFS_EXTENDED_PROFILE_MASK
;
8689 write_seqlock(&fs_info
->profiles_lock
);
8690 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
8691 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
8692 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
8693 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
8694 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
8695 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
8696 write_sequnlock(&fs_info
->profiles_lock
);
8699 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8700 struct btrfs_root
*root
, u64 group_start
)
8702 struct btrfs_path
*path
;
8703 struct btrfs_block_group_cache
*block_group
;
8704 struct btrfs_free_cluster
*cluster
;
8705 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8706 struct btrfs_key key
;
8707 struct inode
*inode
;
8712 root
= root
->fs_info
->extent_root
;
8714 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8715 BUG_ON(!block_group
);
8716 BUG_ON(!block_group
->ro
);
8719 * Free the reserved super bytes from this block group before
8722 free_excluded_extents(root
, block_group
);
8724 memcpy(&key
, &block_group
->key
, sizeof(key
));
8725 index
= get_block_group_index(block_group
);
8726 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8727 BTRFS_BLOCK_GROUP_RAID1
|
8728 BTRFS_BLOCK_GROUP_RAID10
))
8733 /* make sure this block group isn't part of an allocation cluster */
8734 cluster
= &root
->fs_info
->data_alloc_cluster
;
8735 spin_lock(&cluster
->refill_lock
);
8736 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8737 spin_unlock(&cluster
->refill_lock
);
8740 * make sure this block group isn't part of a metadata
8741 * allocation cluster
8743 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8744 spin_lock(&cluster
->refill_lock
);
8745 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8746 spin_unlock(&cluster
->refill_lock
);
8748 path
= btrfs_alloc_path();
8754 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
8755 if (!IS_ERR(inode
)) {
8756 ret
= btrfs_orphan_add(trans
, inode
);
8758 btrfs_add_delayed_iput(inode
);
8762 /* One for the block groups ref */
8763 spin_lock(&block_group
->lock
);
8764 if (block_group
->iref
) {
8765 block_group
->iref
= 0;
8766 block_group
->inode
= NULL
;
8767 spin_unlock(&block_group
->lock
);
8770 spin_unlock(&block_group
->lock
);
8772 /* One for our lookup ref */
8773 btrfs_add_delayed_iput(inode
);
8776 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8777 key
.offset
= block_group
->key
.objectid
;
8780 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8784 btrfs_release_path(path
);
8786 ret
= btrfs_del_item(trans
, tree_root
, path
);
8789 btrfs_release_path(path
);
8792 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8793 rb_erase(&block_group
->cache_node
,
8794 &root
->fs_info
->block_group_cache_tree
);
8796 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
8797 root
->fs_info
->first_logical_byte
= (u64
)-1;
8798 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8800 down_write(&block_group
->space_info
->groups_sem
);
8802 * we must use list_del_init so people can check to see if they
8803 * are still on the list after taking the semaphore
8805 list_del_init(&block_group
->list
);
8806 if (list_empty(&block_group
->space_info
->block_groups
[index
]))
8807 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
8808 up_write(&block_group
->space_info
->groups_sem
);
8810 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8811 wait_block_group_cache_done(block_group
);
8813 btrfs_remove_free_space_cache(block_group
);
8815 spin_lock(&block_group
->space_info
->lock
);
8816 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8817 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8818 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8819 spin_unlock(&block_group
->space_info
->lock
);
8821 memcpy(&key
, &block_group
->key
, sizeof(key
));
8823 btrfs_clear_space_info_full(root
->fs_info
);
8825 btrfs_put_block_group(block_group
);
8826 btrfs_put_block_group(block_group
);
8828 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8834 ret
= btrfs_del_item(trans
, root
, path
);
8836 btrfs_free_path(path
);
8840 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
8842 struct btrfs_space_info
*space_info
;
8843 struct btrfs_super_block
*disk_super
;
8849 disk_super
= fs_info
->super_copy
;
8850 if (!btrfs_super_root(disk_super
))
8853 features
= btrfs_super_incompat_flags(disk_super
);
8854 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
8857 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
8858 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8863 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
8864 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8866 flags
= BTRFS_BLOCK_GROUP_METADATA
;
8867 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8871 flags
= BTRFS_BLOCK_GROUP_DATA
;
8872 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8878 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8880 return unpin_extent_range(root
, start
, end
);
8883 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8884 u64 num_bytes
, u64
*actual_bytes
)
8886 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
8889 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
8891 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8892 struct btrfs_block_group_cache
*cache
= NULL
;
8897 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
8901 * try to trim all FS space, our block group may start from non-zero.
8903 if (range
->len
== total_bytes
)
8904 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
8906 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
8909 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
8910 btrfs_put_block_group(cache
);
8914 start
= max(range
->start
, cache
->key
.objectid
);
8915 end
= min(range
->start
+ range
->len
,
8916 cache
->key
.objectid
+ cache
->key
.offset
);
8918 if (end
- start
>= range
->minlen
) {
8919 if (!block_group_cache_done(cache
)) {
8920 ret
= cache_block_group(cache
, 0);
8922 btrfs_put_block_group(cache
);
8925 ret
= wait_block_group_cache_done(cache
);
8927 btrfs_put_block_group(cache
);
8931 ret
= btrfs_trim_block_group(cache
,
8937 trimmed
+= group_trimmed
;
8939 btrfs_put_block_group(cache
);
8944 cache
= next_block_group(fs_info
->tree_root
, cache
);
8947 range
->len
= trimmed
;