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"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE
= 0,
58 CHUNK_ALLOC_LIMITED
= 1,
59 CHUNK_ALLOC_FORCE
= 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT
= 2,
77 static int update_block_group(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*root
, u64 bytenr
,
79 u64 num_bytes
, int alloc
);
80 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
81 struct btrfs_root
*root
,
82 u64 bytenr
, u64 num_bytes
, u64 parent
,
83 u64 root_objectid
, u64 owner_objectid
,
84 u64 owner_offset
, int refs_to_drop
,
85 struct btrfs_delayed_extent_op
*extra_op
,
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
88 struct extent_buffer
*leaf
,
89 struct btrfs_extent_item
*ei
);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
91 struct btrfs_root
*root
,
92 u64 parent
, u64 root_objectid
,
93 u64 flags
, u64 owner
, u64 offset
,
94 struct btrfs_key
*ins
, int ref_mod
);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
96 struct btrfs_root
*root
,
97 u64 parent
, u64 root_objectid
,
98 u64 flags
, struct btrfs_disk_key
*key
,
99 int level
, struct btrfs_key
*ins
,
101 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
102 struct btrfs_root
*extent_root
, u64 flags
,
104 static int find_next_key(struct btrfs_path
*path
, int level
,
105 struct btrfs_key
*key
);
106 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
107 int dump_block_groups
);
108 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
109 u64 num_bytes
, int reserve
,
111 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
113 int btrfs_pin_extent(struct btrfs_root
*root
,
114 u64 bytenr
, u64 num_bytes
, int reserved
);
117 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
120 return cache
->cached
== BTRFS_CACHE_FINISHED
||
121 cache
->cached
== BTRFS_CACHE_ERROR
;
124 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
126 return (cache
->flags
& bits
) == bits
;
129 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
131 atomic_inc(&cache
->count
);
134 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
136 if (atomic_dec_and_test(&cache
->count
)) {
137 WARN_ON(cache
->pinned
> 0);
138 WARN_ON(cache
->reserved
> 0);
139 kfree(cache
->free_space_ctl
);
145 * this adds the block group to the fs_info rb tree for the block group
148 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
149 struct btrfs_block_group_cache
*block_group
)
152 struct rb_node
*parent
= NULL
;
153 struct btrfs_block_group_cache
*cache
;
155 spin_lock(&info
->block_group_cache_lock
);
156 p
= &info
->block_group_cache_tree
.rb_node
;
160 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
162 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
164 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
167 spin_unlock(&info
->block_group_cache_lock
);
172 rb_link_node(&block_group
->cache_node
, parent
, p
);
173 rb_insert_color(&block_group
->cache_node
,
174 &info
->block_group_cache_tree
);
176 if (info
->first_logical_byte
> block_group
->key
.objectid
)
177 info
->first_logical_byte
= block_group
->key
.objectid
;
179 spin_unlock(&info
->block_group_cache_lock
);
185 * This will return the block group at or after bytenr if contains is 0, else
186 * it will return the block group that contains the bytenr
188 static struct btrfs_block_group_cache
*
189 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
192 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
196 spin_lock(&info
->block_group_cache_lock
);
197 n
= info
->block_group_cache_tree
.rb_node
;
200 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
202 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
203 start
= cache
->key
.objectid
;
205 if (bytenr
< start
) {
206 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
209 } else if (bytenr
> start
) {
210 if (contains
&& bytenr
<= end
) {
221 btrfs_get_block_group(ret
);
222 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
223 info
->first_logical_byte
= ret
->key
.objectid
;
225 spin_unlock(&info
->block_group_cache_lock
);
230 static int add_excluded_extent(struct btrfs_root
*root
,
231 u64 start
, u64 num_bytes
)
233 u64 end
= start
+ num_bytes
- 1;
234 set_extent_bits(&root
->fs_info
->freed_extents
[0],
235 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
236 set_extent_bits(&root
->fs_info
->freed_extents
[1],
237 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
241 static void free_excluded_extents(struct btrfs_root
*root
,
242 struct btrfs_block_group_cache
*cache
)
246 start
= cache
->key
.objectid
;
247 end
= start
+ cache
->key
.offset
- 1;
249 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
250 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
251 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
252 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
255 static int exclude_super_stripes(struct btrfs_root
*root
,
256 struct btrfs_block_group_cache
*cache
)
263 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
264 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
265 cache
->bytes_super
+= stripe_len
;
266 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
272 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
273 bytenr
= btrfs_sb_offset(i
);
274 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
275 cache
->key
.objectid
, bytenr
,
276 0, &logical
, &nr
, &stripe_len
);
283 if (logical
[nr
] > cache
->key
.objectid
+
287 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
291 if (start
< cache
->key
.objectid
) {
292 start
= cache
->key
.objectid
;
293 len
= (logical
[nr
] + stripe_len
) - start
;
295 len
= min_t(u64
, stripe_len
,
296 cache
->key
.objectid
+
297 cache
->key
.offset
- start
);
300 cache
->bytes_super
+= len
;
301 ret
= add_excluded_extent(root
, start
, len
);
313 static struct btrfs_caching_control
*
314 get_caching_control(struct btrfs_block_group_cache
*cache
)
316 struct btrfs_caching_control
*ctl
;
318 spin_lock(&cache
->lock
);
319 if (!cache
->caching_ctl
) {
320 spin_unlock(&cache
->lock
);
324 ctl
= cache
->caching_ctl
;
325 atomic_inc(&ctl
->count
);
326 spin_unlock(&cache
->lock
);
330 static void put_caching_control(struct btrfs_caching_control
*ctl
)
332 if (atomic_dec_and_test(&ctl
->count
))
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
341 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
342 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
344 u64 extent_start
, extent_end
, size
, total_added
= 0;
347 while (start
< end
) {
348 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
349 &extent_start
, &extent_end
,
350 EXTENT_DIRTY
| EXTENT_UPTODATE
,
355 if (extent_start
<= start
) {
356 start
= extent_end
+ 1;
357 } else if (extent_start
> start
&& extent_start
< end
) {
358 size
= extent_start
- start
;
360 ret
= btrfs_add_free_space(block_group
, start
,
362 BUG_ON(ret
); /* -ENOMEM or logic error */
363 start
= extent_end
+ 1;
372 ret
= btrfs_add_free_space(block_group
, start
, size
);
373 BUG_ON(ret
); /* -ENOMEM or logic error */
379 static noinline
void caching_thread(struct btrfs_work
*work
)
381 struct btrfs_block_group_cache
*block_group
;
382 struct btrfs_fs_info
*fs_info
;
383 struct btrfs_caching_control
*caching_ctl
;
384 struct btrfs_root
*extent_root
;
385 struct btrfs_path
*path
;
386 struct extent_buffer
*leaf
;
387 struct btrfs_key key
;
393 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
394 block_group
= caching_ctl
->block_group
;
395 fs_info
= block_group
->fs_info
;
396 extent_root
= fs_info
->extent_root
;
398 path
= btrfs_alloc_path();
402 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
405 * We don't want to deadlock with somebody trying to allocate a new
406 * extent for the extent root while also trying to search the extent
407 * root to add free space. So we skip locking and search the commit
408 * root, since its read-only
410 path
->skip_locking
= 1;
411 path
->search_commit_root
= 1;
416 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
418 mutex_lock(&caching_ctl
->mutex
);
419 /* need to make sure the commit_root doesn't disappear */
420 down_read(&fs_info
->commit_root_sem
);
423 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
427 leaf
= path
->nodes
[0];
428 nritems
= btrfs_header_nritems(leaf
);
431 if (btrfs_fs_closing(fs_info
) > 1) {
436 if (path
->slots
[0] < nritems
) {
437 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
439 ret
= find_next_key(path
, 0, &key
);
443 if (need_resched() ||
444 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
445 caching_ctl
->progress
= last
;
446 btrfs_release_path(path
);
447 up_read(&fs_info
->commit_root_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
->nodesize
;
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
->commit_root_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 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
551 caching_thread
, NULL
, NULL
);
553 spin_lock(&cache
->lock
);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
566 while (cache
->cached
== BTRFS_CACHE_FAST
) {
567 struct btrfs_caching_control
*ctl
;
569 ctl
= cache
->caching_ctl
;
570 atomic_inc(&ctl
->count
);
571 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
572 spin_unlock(&cache
->lock
);
576 finish_wait(&ctl
->wait
, &wait
);
577 put_caching_control(ctl
);
578 spin_lock(&cache
->lock
);
581 if (cache
->cached
!= BTRFS_CACHE_NO
) {
582 spin_unlock(&cache
->lock
);
586 WARN_ON(cache
->caching_ctl
);
587 cache
->caching_ctl
= caching_ctl
;
588 cache
->cached
= BTRFS_CACHE_FAST
;
589 spin_unlock(&cache
->lock
);
591 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
592 mutex_lock(&caching_ctl
->mutex
);
593 ret
= load_free_space_cache(fs_info
, cache
);
595 spin_lock(&cache
->lock
);
597 cache
->caching_ctl
= NULL
;
598 cache
->cached
= BTRFS_CACHE_FINISHED
;
599 cache
->last_byte_to_unpin
= (u64
)-1;
600 caching_ctl
->progress
= (u64
)-1;
602 if (load_cache_only
) {
603 cache
->caching_ctl
= NULL
;
604 cache
->cached
= BTRFS_CACHE_NO
;
606 cache
->cached
= BTRFS_CACHE_STARTED
;
607 cache
->has_caching_ctl
= 1;
610 spin_unlock(&cache
->lock
);
611 mutex_unlock(&caching_ctl
->mutex
);
613 wake_up(&caching_ctl
->wait
);
615 put_caching_control(caching_ctl
);
616 free_excluded_extents(fs_info
->extent_root
, cache
);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache
->lock
);
625 if (load_cache_only
) {
626 cache
->caching_ctl
= NULL
;
627 cache
->cached
= BTRFS_CACHE_NO
;
629 cache
->cached
= BTRFS_CACHE_STARTED
;
630 cache
->has_caching_ctl
= 1;
632 spin_unlock(&cache
->lock
);
633 wake_up(&caching_ctl
->wait
);
636 if (load_cache_only
) {
637 put_caching_control(caching_ctl
);
641 down_write(&fs_info
->commit_root_sem
);
642 atomic_inc(&caching_ctl
->count
);
643 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
644 up_write(&fs_info
->commit_root_sem
);
646 btrfs_get_block_group(cache
);
648 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
654 * return the block group that starts at or after bytenr
656 static struct btrfs_block_group_cache
*
657 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
659 struct btrfs_block_group_cache
*cache
;
661 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
667 * return the block group that contains the given bytenr
669 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
670 struct btrfs_fs_info
*info
,
673 struct btrfs_block_group_cache
*cache
;
675 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
680 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
683 struct list_head
*head
= &info
->space_info
;
684 struct btrfs_space_info
*found
;
686 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
689 list_for_each_entry_rcu(found
, head
, list
) {
690 if (found
->flags
& flags
) {
700 * after adding space to the filesystem, we need to clear the full flags
701 * on all the space infos.
703 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
705 struct list_head
*head
= &info
->space_info
;
706 struct btrfs_space_info
*found
;
709 list_for_each_entry_rcu(found
, head
, list
)
714 /* simple helper to search for an existing data extent at a given offset */
715 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
718 struct btrfs_key key
;
719 struct btrfs_path
*path
;
721 path
= btrfs_alloc_path();
725 key
.objectid
= start
;
727 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
728 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
730 btrfs_free_path(path
);
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
744 struct btrfs_root
*root
, u64 bytenr
,
745 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
747 struct btrfs_delayed_ref_head
*head
;
748 struct btrfs_delayed_ref_root
*delayed_refs
;
749 struct btrfs_path
*path
;
750 struct btrfs_extent_item
*ei
;
751 struct extent_buffer
*leaf
;
752 struct btrfs_key key
;
759 * If we don't have skinny metadata, don't bother doing anything
762 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
763 offset
= root
->nodesize
;
767 path
= btrfs_alloc_path();
772 path
->skip_locking
= 1;
773 path
->search_commit_root
= 1;
777 key
.objectid
= bytenr
;
780 key
.type
= BTRFS_METADATA_ITEM_KEY
;
782 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
->nodesize
)
802 leaf
= path
->nodes
[0];
803 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
804 if (item_size
>= sizeof(*ei
)) {
805 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
806 struct btrfs_extent_item
);
807 num_refs
= btrfs_extent_refs(leaf
, ei
);
808 extent_flags
= btrfs_extent_flags(leaf
, ei
);
810 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
811 struct btrfs_extent_item_v0
*ei0
;
812 BUG_ON(item_size
!= sizeof(*ei0
));
813 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
814 struct btrfs_extent_item_v0
);
815 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
816 /* FIXME: this isn't correct for data */
817 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
822 BUG_ON(num_refs
== 0);
832 delayed_refs
= &trans
->transaction
->delayed_refs
;
833 spin_lock(&delayed_refs
->lock
);
834 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
836 if (!mutex_trylock(&head
->mutex
)) {
837 atomic_inc(&head
->node
.refs
);
838 spin_unlock(&delayed_refs
->lock
);
840 btrfs_release_path(path
);
843 * Mutex was contended, block until it's released and try
846 mutex_lock(&head
->mutex
);
847 mutex_unlock(&head
->mutex
);
848 btrfs_put_delayed_ref(&head
->node
);
851 spin_lock(&head
->lock
);
852 if (head
->extent_op
&& head
->extent_op
->update_flags
)
853 extent_flags
|= head
->extent_op
->flags_to_set
;
855 BUG_ON(num_refs
== 0);
857 num_refs
+= head
->node
.ref_mod
;
858 spin_unlock(&head
->lock
);
859 mutex_unlock(&head
->mutex
);
861 spin_unlock(&delayed_refs
->lock
);
863 WARN_ON(num_refs
== 0);
867 *flags
= extent_flags
;
869 btrfs_free_path(path
);
874 * Back reference rules. Back refs have three main goals:
876 * 1) differentiate between all holders of references to an extent so that
877 * when a reference is dropped we can make sure it was a valid reference
878 * before freeing the extent.
880 * 2) Provide enough information to quickly find the holders of an extent
881 * if we notice a given block is corrupted or bad.
883 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884 * maintenance. This is actually the same as #2, but with a slightly
885 * different use case.
887 * There are two kinds of back refs. The implicit back refs is optimized
888 * for pointers in non-shared tree blocks. For a given pointer in a block,
889 * back refs of this kind provide information about the block's owner tree
890 * and the pointer's key. These information allow us to find the block by
891 * b-tree searching. The full back refs is for pointers in tree blocks not
892 * referenced by their owner trees. The location of tree block is recorded
893 * in the back refs. Actually the full back refs is generic, and can be
894 * used in all cases the implicit back refs is used. The major shortcoming
895 * of the full back refs is its overhead. Every time a tree block gets
896 * COWed, we have to update back refs entry for all pointers in it.
898 * For a newly allocated tree block, we use implicit back refs for
899 * pointers in it. This means most tree related operations only involve
900 * implicit back refs. For a tree block created in old transaction, the
901 * only way to drop a reference to it is COW it. So we can detect the
902 * event that tree block loses its owner tree's reference and do the
903 * back refs conversion.
905 * When a tree block is COW'd through a tree, there are four cases:
907 * The reference count of the block is one and the tree is the block's
908 * owner tree. Nothing to do in this case.
910 * The reference count of the block is one and the tree is not the
911 * block's owner tree. In this case, full back refs is used for pointers
912 * in the block. Remove these full back refs, add implicit back refs for
913 * every pointers in the new block.
915 * The reference count of the block is greater than one and the tree is
916 * the block's owner tree. In this case, implicit back refs is used for
917 * pointers in the block. Add full back refs for every pointers in the
918 * block, increase lower level extents' reference counts. The original
919 * implicit back refs are entailed to the new block.
921 * The reference count of the block is greater than one and the tree is
922 * not the block's owner tree. Add implicit back refs for every pointer in
923 * the new block, increase lower level extents' reference count.
925 * Back Reference Key composing:
927 * The key objectid corresponds to the first byte in the extent,
928 * The key type is used to differentiate between types of back refs.
929 * There are different meanings of the key offset for different types
932 * File extents can be referenced by:
934 * - multiple snapshots, subvolumes, or different generations in one subvol
935 * - different files inside a single subvolume
936 * - different offsets inside a file (bookend extents in file.c)
938 * The extent ref structure for the implicit back refs has fields for:
940 * - Objectid of the subvolume root
941 * - objectid of the file holding the reference
942 * - original offset in the file
943 * - how many bookend extents
945 * The key offset for the implicit back refs is hash of the first
948 * The extent ref structure for the full back refs has field for:
950 * - number of pointers in the tree leaf
952 * The key offset for the implicit back refs is the first byte of
955 * When a file extent is allocated, The implicit back refs is used.
956 * the fields are filled in:
958 * (root_key.objectid, inode objectid, offset in file, 1)
960 * When a file extent is removed file truncation, we find the
961 * corresponding implicit back refs and check the following fields:
963 * (btrfs_header_owner(leaf), inode objectid, offset in file)
965 * Btree extents can be referenced by:
967 * - Different subvolumes
969 * Both the implicit back refs and the full back refs for tree blocks
970 * only consist of key. The key offset for the implicit back refs is
971 * objectid of block's owner tree. The key offset for the full back refs
972 * is the first byte of parent block.
974 * When implicit back refs is used, information about the lowest key and
975 * level of the tree block are required. These information are stored in
976 * tree block info structure.
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
981 struct btrfs_root
*root
,
982 struct btrfs_path
*path
,
983 u64 owner
, u32 extra_size
)
985 struct btrfs_extent_item
*item
;
986 struct btrfs_extent_item_v0
*ei0
;
987 struct btrfs_extent_ref_v0
*ref0
;
988 struct btrfs_tree_block_info
*bi
;
989 struct extent_buffer
*leaf
;
990 struct btrfs_key key
;
991 struct btrfs_key found_key
;
992 u32 new_size
= sizeof(*item
);
996 leaf
= path
->nodes
[0];
997 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
999 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1000 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1001 struct btrfs_extent_item_v0
);
1002 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1004 if (owner
== (u64
)-1) {
1006 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1007 ret
= btrfs_next_leaf(root
, path
);
1010 BUG_ON(ret
> 0); /* Corruption */
1011 leaf
= path
->nodes
[0];
1013 btrfs_item_key_to_cpu(leaf
, &found_key
,
1015 BUG_ON(key
.objectid
!= found_key
.objectid
);
1016 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1020 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1021 struct btrfs_extent_ref_v0
);
1022 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1026 btrfs_release_path(path
);
1028 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1029 new_size
+= sizeof(*bi
);
1031 new_size
-= sizeof(*ei0
);
1032 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1033 new_size
+ extra_size
, 1);
1036 BUG_ON(ret
); /* Corruption */
1038 btrfs_extend_item(root
, path
, new_size
);
1040 leaf
= path
->nodes
[0];
1041 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1042 btrfs_set_extent_refs(leaf
, item
, refs
);
1043 /* FIXME: get real generation */
1044 btrfs_set_extent_generation(leaf
, item
, 0);
1045 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1046 btrfs_set_extent_flags(leaf
, item
,
1047 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1048 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1049 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1050 /* FIXME: get first key of the block */
1051 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1052 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1054 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1056 btrfs_mark_buffer_dirty(leaf
);
1061 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1063 u32 high_crc
= ~(u32
)0;
1064 u32 low_crc
= ~(u32
)0;
1067 lenum
= cpu_to_le64(root_objectid
);
1068 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1069 lenum
= cpu_to_le64(owner
);
1070 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1071 lenum
= cpu_to_le64(offset
);
1072 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1074 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1077 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1078 struct btrfs_extent_data_ref
*ref
)
1080 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1081 btrfs_extent_data_ref_objectid(leaf
, ref
),
1082 btrfs_extent_data_ref_offset(leaf
, ref
));
1085 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1086 struct btrfs_extent_data_ref
*ref
,
1087 u64 root_objectid
, u64 owner
, u64 offset
)
1089 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1090 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1091 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1096 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1097 struct btrfs_root
*root
,
1098 struct btrfs_path
*path
,
1099 u64 bytenr
, u64 parent
,
1101 u64 owner
, u64 offset
)
1103 struct btrfs_key key
;
1104 struct btrfs_extent_data_ref
*ref
;
1105 struct extent_buffer
*leaf
;
1111 key
.objectid
= bytenr
;
1113 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1114 key
.offset
= parent
;
1116 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1117 key
.offset
= hash_extent_data_ref(root_objectid
,
1122 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1133 btrfs_release_path(path
);
1134 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1145 leaf
= path
->nodes
[0];
1146 nritems
= btrfs_header_nritems(leaf
);
1148 if (path
->slots
[0] >= nritems
) {
1149 ret
= btrfs_next_leaf(root
, path
);
1155 leaf
= path
->nodes
[0];
1156 nritems
= btrfs_header_nritems(leaf
);
1160 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1161 if (key
.objectid
!= bytenr
||
1162 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1165 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1166 struct btrfs_extent_data_ref
);
1168 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1171 btrfs_release_path(path
);
1183 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1184 struct btrfs_root
*root
,
1185 struct btrfs_path
*path
,
1186 u64 bytenr
, u64 parent
,
1187 u64 root_objectid
, u64 owner
,
1188 u64 offset
, int refs_to_add
)
1190 struct btrfs_key key
;
1191 struct extent_buffer
*leaf
;
1196 key
.objectid
= bytenr
;
1198 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1199 key
.offset
= parent
;
1200 size
= sizeof(struct btrfs_shared_data_ref
);
1202 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1203 key
.offset
= hash_extent_data_ref(root_objectid
,
1205 size
= sizeof(struct btrfs_extent_data_ref
);
1208 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1209 if (ret
&& ret
!= -EEXIST
)
1212 leaf
= path
->nodes
[0];
1214 struct btrfs_shared_data_ref
*ref
;
1215 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1216 struct btrfs_shared_data_ref
);
1218 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1220 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1221 num_refs
+= refs_to_add
;
1222 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1225 struct btrfs_extent_data_ref
*ref
;
1226 while (ret
== -EEXIST
) {
1227 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1228 struct btrfs_extent_data_ref
);
1229 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1232 btrfs_release_path(path
);
1234 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1236 if (ret
&& ret
!= -EEXIST
)
1239 leaf
= path
->nodes
[0];
1241 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1242 struct btrfs_extent_data_ref
);
1244 btrfs_set_extent_data_ref_root(leaf
, ref
,
1246 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1247 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1248 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1250 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1251 num_refs
+= refs_to_add
;
1252 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1255 btrfs_mark_buffer_dirty(leaf
);
1258 btrfs_release_path(path
);
1262 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1263 struct btrfs_root
*root
,
1264 struct btrfs_path
*path
,
1265 int refs_to_drop
, int *last_ref
)
1267 struct btrfs_key key
;
1268 struct btrfs_extent_data_ref
*ref1
= NULL
;
1269 struct btrfs_shared_data_ref
*ref2
= NULL
;
1270 struct extent_buffer
*leaf
;
1274 leaf
= path
->nodes
[0];
1275 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1277 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1278 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1279 struct btrfs_extent_data_ref
);
1280 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1281 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1282 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1283 struct btrfs_shared_data_ref
);
1284 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1287 struct btrfs_extent_ref_v0
*ref0
;
1288 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1289 struct btrfs_extent_ref_v0
);
1290 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1296 BUG_ON(num_refs
< refs_to_drop
);
1297 num_refs
-= refs_to_drop
;
1299 if (num_refs
== 0) {
1300 ret
= btrfs_del_item(trans
, root
, path
);
1303 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1304 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1305 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1306 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 struct btrfs_extent_ref_v0
*ref0
;
1310 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1311 struct btrfs_extent_ref_v0
);
1312 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1315 btrfs_mark_buffer_dirty(leaf
);
1320 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1321 struct btrfs_path
*path
,
1322 struct btrfs_extent_inline_ref
*iref
)
1324 struct btrfs_key key
;
1325 struct extent_buffer
*leaf
;
1326 struct btrfs_extent_data_ref
*ref1
;
1327 struct btrfs_shared_data_ref
*ref2
;
1330 leaf
= path
->nodes
[0];
1331 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1333 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1334 BTRFS_EXTENT_DATA_REF_KEY
) {
1335 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1336 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1338 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1339 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1341 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1342 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1343 struct btrfs_extent_data_ref
);
1344 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1345 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1346 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1347 struct btrfs_shared_data_ref
);
1348 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1349 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1350 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1351 struct btrfs_extent_ref_v0
*ref0
;
1352 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1353 struct btrfs_extent_ref_v0
);
1354 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1362 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1363 struct btrfs_root
*root
,
1364 struct btrfs_path
*path
,
1365 u64 bytenr
, u64 parent
,
1368 struct btrfs_key key
;
1371 key
.objectid
= bytenr
;
1373 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1374 key
.offset
= parent
;
1376 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1377 key
.offset
= root_objectid
;
1380 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1383 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1384 if (ret
== -ENOENT
&& parent
) {
1385 btrfs_release_path(path
);
1386 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1387 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1395 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1396 struct btrfs_root
*root
,
1397 struct btrfs_path
*path
,
1398 u64 bytenr
, u64 parent
,
1401 struct btrfs_key key
;
1404 key
.objectid
= bytenr
;
1406 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1407 key
.offset
= parent
;
1409 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1410 key
.offset
= root_objectid
;
1413 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1414 btrfs_release_path(path
);
1418 static inline int extent_ref_type(u64 parent
, u64 owner
)
1421 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1423 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1425 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1428 type
= BTRFS_SHARED_DATA_REF_KEY
;
1430 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1435 static int find_next_key(struct btrfs_path
*path
, int level
,
1436 struct btrfs_key
*key
)
1439 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1440 if (!path
->nodes
[level
])
1442 if (path
->slots
[level
] + 1 >=
1443 btrfs_header_nritems(path
->nodes
[level
]))
1446 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1447 path
->slots
[level
] + 1);
1449 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1450 path
->slots
[level
] + 1);
1457 * look for inline back ref. if back ref is found, *ref_ret is set
1458 * to the address of inline back ref, and 0 is returned.
1460 * if back ref isn't found, *ref_ret is set to the address where it
1461 * should be inserted, and -ENOENT is returned.
1463 * if insert is true and there are too many inline back refs, the path
1464 * points to the extent item, and -EAGAIN is returned.
1466 * NOTE: inline back refs are ordered in the same way that back ref
1467 * items in the tree are ordered.
1469 static noinline_for_stack
1470 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1471 struct btrfs_root
*root
,
1472 struct btrfs_path
*path
,
1473 struct btrfs_extent_inline_ref
**ref_ret
,
1474 u64 bytenr
, u64 num_bytes
,
1475 u64 parent
, u64 root_objectid
,
1476 u64 owner
, u64 offset
, int insert
)
1478 struct btrfs_key key
;
1479 struct extent_buffer
*leaf
;
1480 struct btrfs_extent_item
*ei
;
1481 struct btrfs_extent_inline_ref
*iref
;
1491 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1494 key
.objectid
= bytenr
;
1495 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1496 key
.offset
= num_bytes
;
1498 want
= extent_ref_type(parent
, owner
);
1500 extra_size
= btrfs_extent_inline_ref_size(want
);
1501 path
->keep_locks
= 1;
1506 * Owner is our parent level, so we can just add one to get the level
1507 * for the block we are interested in.
1509 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1510 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1515 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1522 * We may be a newly converted file system which still has the old fat
1523 * extent entries for metadata, so try and see if we have one of those.
1525 if (ret
> 0 && skinny_metadata
) {
1526 skinny_metadata
= false;
1527 if (path
->slots
[0]) {
1529 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1531 if (key
.objectid
== bytenr
&&
1532 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1533 key
.offset
== num_bytes
)
1537 key
.objectid
= bytenr
;
1538 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1539 key
.offset
= num_bytes
;
1540 btrfs_release_path(path
);
1545 if (ret
&& !insert
) {
1548 } else if (WARN_ON(ret
)) {
1553 leaf
= path
->nodes
[0];
1554 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1555 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1556 if (item_size
< sizeof(*ei
)) {
1561 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1567 leaf
= path
->nodes
[0];
1568 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1571 BUG_ON(item_size
< sizeof(*ei
));
1573 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1574 flags
= btrfs_extent_flags(leaf
, ei
);
1576 ptr
= (unsigned long)(ei
+ 1);
1577 end
= (unsigned long)ei
+ item_size
;
1579 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1580 ptr
+= sizeof(struct btrfs_tree_block_info
);
1590 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1591 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1595 ptr
+= btrfs_extent_inline_ref_size(type
);
1599 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1600 struct btrfs_extent_data_ref
*dref
;
1601 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1602 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1607 if (hash_extent_data_ref_item(leaf
, dref
) <
1608 hash_extent_data_ref(root_objectid
, owner
, offset
))
1612 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1614 if (parent
== ref_offset
) {
1618 if (ref_offset
< parent
)
1621 if (root_objectid
== ref_offset
) {
1625 if (ref_offset
< root_objectid
)
1629 ptr
+= btrfs_extent_inline_ref_size(type
);
1631 if (err
== -ENOENT
&& insert
) {
1632 if (item_size
+ extra_size
>=
1633 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1638 * To add new inline back ref, we have to make sure
1639 * there is no corresponding back ref item.
1640 * For simplicity, we just do not add new inline back
1641 * ref if there is any kind of item for this block
1643 if (find_next_key(path
, 0, &key
) == 0 &&
1644 key
.objectid
== bytenr
&&
1645 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1650 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1653 path
->keep_locks
= 0;
1654 btrfs_unlock_up_safe(path
, 1);
1660 * helper to add new inline back ref
1662 static noinline_for_stack
1663 void setup_inline_extent_backref(struct btrfs_root
*root
,
1664 struct btrfs_path
*path
,
1665 struct btrfs_extent_inline_ref
*iref
,
1666 u64 parent
, u64 root_objectid
,
1667 u64 owner
, u64 offset
, int refs_to_add
,
1668 struct btrfs_delayed_extent_op
*extent_op
)
1670 struct extent_buffer
*leaf
;
1671 struct btrfs_extent_item
*ei
;
1674 unsigned long item_offset
;
1679 leaf
= path
->nodes
[0];
1680 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1681 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1683 type
= extent_ref_type(parent
, owner
);
1684 size
= btrfs_extent_inline_ref_size(type
);
1686 btrfs_extend_item(root
, path
, size
);
1688 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1689 refs
= btrfs_extent_refs(leaf
, ei
);
1690 refs
+= refs_to_add
;
1691 btrfs_set_extent_refs(leaf
, ei
, refs
);
1693 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1695 ptr
= (unsigned long)ei
+ item_offset
;
1696 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1697 if (ptr
< end
- size
)
1698 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1701 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1702 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1703 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1704 struct btrfs_extent_data_ref
*dref
;
1705 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1706 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1707 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1708 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1709 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1710 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1711 struct btrfs_shared_data_ref
*sref
;
1712 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1713 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1714 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1715 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1716 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1718 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1720 btrfs_mark_buffer_dirty(leaf
);
1723 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1724 struct btrfs_root
*root
,
1725 struct btrfs_path
*path
,
1726 struct btrfs_extent_inline_ref
**ref_ret
,
1727 u64 bytenr
, u64 num_bytes
, u64 parent
,
1728 u64 root_objectid
, u64 owner
, u64 offset
)
1732 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1733 bytenr
, num_bytes
, parent
,
1734 root_objectid
, owner
, offset
, 0);
1738 btrfs_release_path(path
);
1741 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1742 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1745 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1746 root_objectid
, owner
, offset
);
1752 * helper to update/remove inline back ref
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_root
*root
,
1756 struct btrfs_path
*path
,
1757 struct btrfs_extent_inline_ref
*iref
,
1759 struct btrfs_delayed_extent_op
*extent_op
,
1762 struct extent_buffer
*leaf
;
1763 struct btrfs_extent_item
*ei
;
1764 struct btrfs_extent_data_ref
*dref
= NULL
;
1765 struct btrfs_shared_data_ref
*sref
= NULL
;
1773 leaf
= path
->nodes
[0];
1774 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1775 refs
= btrfs_extent_refs(leaf
, ei
);
1776 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1777 refs
+= refs_to_mod
;
1778 btrfs_set_extent_refs(leaf
, ei
, refs
);
1780 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1782 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1784 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1785 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1786 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1787 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1788 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1789 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1792 BUG_ON(refs_to_mod
!= -1);
1795 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1796 refs
+= refs_to_mod
;
1799 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1800 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1802 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1805 size
= btrfs_extent_inline_ref_size(type
);
1806 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1807 ptr
= (unsigned long)iref
;
1808 end
= (unsigned long)ei
+ item_size
;
1809 if (ptr
+ size
< end
)
1810 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1813 btrfs_truncate_item(root
, path
, item_size
, 1);
1815 btrfs_mark_buffer_dirty(leaf
);
1818 static noinline_for_stack
1819 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1820 struct btrfs_root
*root
,
1821 struct btrfs_path
*path
,
1822 u64 bytenr
, u64 num_bytes
, u64 parent
,
1823 u64 root_objectid
, u64 owner
,
1824 u64 offset
, int refs_to_add
,
1825 struct btrfs_delayed_extent_op
*extent_op
)
1827 struct btrfs_extent_inline_ref
*iref
;
1830 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1831 bytenr
, num_bytes
, parent
,
1832 root_objectid
, owner
, offset
, 1);
1834 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1835 update_inline_extent_backref(root
, path
, iref
,
1836 refs_to_add
, extent_op
, NULL
);
1837 } else if (ret
== -ENOENT
) {
1838 setup_inline_extent_backref(root
, path
, iref
, parent
,
1839 root_objectid
, owner
, offset
,
1840 refs_to_add
, extent_op
);
1846 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1847 struct btrfs_root
*root
,
1848 struct btrfs_path
*path
,
1849 u64 bytenr
, u64 parent
, u64 root_objectid
,
1850 u64 owner
, u64 offset
, int refs_to_add
)
1853 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1854 BUG_ON(refs_to_add
!= 1);
1855 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1856 parent
, root_objectid
);
1858 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1859 parent
, root_objectid
,
1860 owner
, offset
, refs_to_add
);
1865 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1866 struct btrfs_root
*root
,
1867 struct btrfs_path
*path
,
1868 struct btrfs_extent_inline_ref
*iref
,
1869 int refs_to_drop
, int is_data
, int *last_ref
)
1873 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1875 update_inline_extent_backref(root
, path
, iref
,
1876 -refs_to_drop
, NULL
, last_ref
);
1877 } else if (is_data
) {
1878 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1882 ret
= btrfs_del_item(trans
, root
, path
);
1887 static int btrfs_issue_discard(struct block_device
*bdev
,
1890 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1893 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1894 u64 num_bytes
, u64
*actual_bytes
)
1897 u64 discarded_bytes
= 0;
1898 struct btrfs_bio
*bbio
= NULL
;
1901 /* Tell the block device(s) that the sectors can be discarded */
1902 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1903 bytenr
, &num_bytes
, &bbio
, 0);
1904 /* Error condition is -ENOMEM */
1906 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1910 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1911 if (!stripe
->dev
->can_discard
)
1914 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1918 discarded_bytes
+= stripe
->length
;
1919 else if (ret
!= -EOPNOTSUPP
)
1920 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923 * Just in case we get back EOPNOTSUPP for some reason,
1924 * just ignore the return value so we don't screw up
1925 * people calling discard_extent.
1929 btrfs_put_bbio(bbio
);
1933 *actual_bytes
= discarded_bytes
;
1936 if (ret
== -EOPNOTSUPP
)
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1943 struct btrfs_root
*root
,
1944 u64 bytenr
, u64 num_bytes
, u64 parent
,
1945 u64 root_objectid
, u64 owner
, u64 offset
,
1949 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1951 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1952 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1954 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1955 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1957 parent
, root_objectid
, (int)owner
,
1958 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1960 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1962 parent
, root_objectid
, owner
, offset
,
1963 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1969 struct btrfs_root
*root
,
1970 u64 bytenr
, u64 num_bytes
,
1971 u64 parent
, u64 root_objectid
,
1972 u64 owner
, u64 offset
, int refs_to_add
,
1974 struct btrfs_delayed_extent_op
*extent_op
)
1976 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1977 struct btrfs_path
*path
;
1978 struct extent_buffer
*leaf
;
1979 struct btrfs_extent_item
*item
;
1980 struct btrfs_key key
;
1983 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_ADD_EXCL
;
1985 path
= btrfs_alloc_path();
1989 if (!is_fstree(root_objectid
) || !root
->fs_info
->quota_enabled
)
1993 path
->leave_spinning
= 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
1996 bytenr
, num_bytes
, parent
,
1997 root_objectid
, owner
, offset
,
1998 refs_to_add
, extent_op
);
1999 if ((ret
< 0 && ret
!= -EAGAIN
) || (!ret
&& no_quota
))
2002 * Ok we were able to insert an inline extent and it appears to be a new
2003 * reference, deal with the qgroup accounting.
2005 if (!ret
&& !no_quota
) {
2006 ASSERT(root
->fs_info
->quota_enabled
);
2007 leaf
= path
->nodes
[0];
2008 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2009 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2010 struct btrfs_extent_item
);
2011 if (btrfs_extent_refs(leaf
, item
) > (u64
)refs_to_add
)
2012 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2013 btrfs_release_path(path
);
2015 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2016 bytenr
, num_bytes
, type
, 0);
2021 * Ok we had -EAGAIN which means we didn't have space to insert and
2022 * inline extent ref, so just update the reference count and add a
2025 leaf
= path
->nodes
[0];
2026 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2027 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2028 refs
= btrfs_extent_refs(leaf
, item
);
2030 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2031 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2033 __run_delayed_extent_op(extent_op
, leaf
, item
);
2035 btrfs_mark_buffer_dirty(leaf
);
2036 btrfs_release_path(path
);
2039 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2040 bytenr
, num_bytes
, type
, 0);
2046 path
->leave_spinning
= 1;
2047 /* now insert the actual backref */
2048 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2049 path
, bytenr
, parent
, root_objectid
,
2050 owner
, offset
, refs_to_add
);
2052 btrfs_abort_transaction(trans
, root
, ret
);
2054 btrfs_free_path(path
);
2058 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2059 struct btrfs_root
*root
,
2060 struct btrfs_delayed_ref_node
*node
,
2061 struct btrfs_delayed_extent_op
*extent_op
,
2062 int insert_reserved
)
2065 struct btrfs_delayed_data_ref
*ref
;
2066 struct btrfs_key ins
;
2071 ins
.objectid
= node
->bytenr
;
2072 ins
.offset
= node
->num_bytes
;
2073 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2075 ref
= btrfs_delayed_node_to_data_ref(node
);
2076 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2078 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2079 parent
= ref
->parent
;
2080 ref_root
= ref
->root
;
2082 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2084 flags
|= extent_op
->flags_to_set
;
2085 ret
= alloc_reserved_file_extent(trans
, root
,
2086 parent
, ref_root
, flags
,
2087 ref
->objectid
, ref
->offset
,
2088 &ins
, node
->ref_mod
);
2089 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2090 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2091 node
->num_bytes
, parent
,
2092 ref_root
, ref
->objectid
,
2093 ref
->offset
, node
->ref_mod
,
2094 node
->no_quota
, extent_op
);
2095 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2096 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2097 node
->num_bytes
, parent
,
2098 ref_root
, ref
->objectid
,
2099 ref
->offset
, node
->ref_mod
,
2100 extent_op
, node
->no_quota
);
2107 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2108 struct extent_buffer
*leaf
,
2109 struct btrfs_extent_item
*ei
)
2111 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2112 if (extent_op
->update_flags
) {
2113 flags
|= extent_op
->flags_to_set
;
2114 btrfs_set_extent_flags(leaf
, ei
, flags
);
2117 if (extent_op
->update_key
) {
2118 struct btrfs_tree_block_info
*bi
;
2119 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2120 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2121 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2125 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2126 struct btrfs_root
*root
,
2127 struct btrfs_delayed_ref_node
*node
,
2128 struct btrfs_delayed_extent_op
*extent_op
)
2130 struct btrfs_key key
;
2131 struct btrfs_path
*path
;
2132 struct btrfs_extent_item
*ei
;
2133 struct extent_buffer
*leaf
;
2137 int metadata
= !extent_op
->is_data
;
2142 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2145 path
= btrfs_alloc_path();
2149 key
.objectid
= node
->bytenr
;
2152 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2153 key
.offset
= extent_op
->level
;
2155 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2156 key
.offset
= node
->num_bytes
;
2161 path
->leave_spinning
= 1;
2162 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2170 if (path
->slots
[0] > 0) {
2172 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2174 if (key
.objectid
== node
->bytenr
&&
2175 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2176 key
.offset
== node
->num_bytes
)
2180 btrfs_release_path(path
);
2183 key
.objectid
= node
->bytenr
;
2184 key
.offset
= node
->num_bytes
;
2185 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2194 leaf
= path
->nodes
[0];
2195 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2196 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2197 if (item_size
< sizeof(*ei
)) {
2198 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2204 leaf
= path
->nodes
[0];
2205 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2208 BUG_ON(item_size
< sizeof(*ei
));
2209 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2210 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2212 btrfs_mark_buffer_dirty(leaf
);
2214 btrfs_free_path(path
);
2218 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2219 struct btrfs_root
*root
,
2220 struct btrfs_delayed_ref_node
*node
,
2221 struct btrfs_delayed_extent_op
*extent_op
,
2222 int insert_reserved
)
2225 struct btrfs_delayed_tree_ref
*ref
;
2226 struct btrfs_key ins
;
2229 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2232 ref
= btrfs_delayed_node_to_tree_ref(node
);
2233 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2235 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2236 parent
= ref
->parent
;
2237 ref_root
= ref
->root
;
2239 ins
.objectid
= node
->bytenr
;
2240 if (skinny_metadata
) {
2241 ins
.offset
= ref
->level
;
2242 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2244 ins
.offset
= node
->num_bytes
;
2245 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2248 BUG_ON(node
->ref_mod
!= 1);
2249 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2250 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2251 ret
= alloc_reserved_tree_block(trans
, root
,
2253 extent_op
->flags_to_set
,
2257 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2258 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2259 node
->num_bytes
, parent
, ref_root
,
2260 ref
->level
, 0, 1, node
->no_quota
,
2262 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2263 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2264 node
->num_bytes
, parent
, ref_root
,
2265 ref
->level
, 0, 1, extent_op
,
2273 /* helper function to actually process a single delayed ref entry */
2274 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2275 struct btrfs_root
*root
,
2276 struct btrfs_delayed_ref_node
*node
,
2277 struct btrfs_delayed_extent_op
*extent_op
,
2278 int insert_reserved
)
2282 if (trans
->aborted
) {
2283 if (insert_reserved
)
2284 btrfs_pin_extent(root
, node
->bytenr
,
2285 node
->num_bytes
, 1);
2289 if (btrfs_delayed_ref_is_head(node
)) {
2290 struct btrfs_delayed_ref_head
*head
;
2292 * we've hit the end of the chain and we were supposed
2293 * to insert this extent into the tree. But, it got
2294 * deleted before we ever needed to insert it, so all
2295 * we have to do is clean up the accounting
2298 head
= btrfs_delayed_node_to_head(node
);
2299 trace_run_delayed_ref_head(node
, head
, node
->action
);
2301 if (insert_reserved
) {
2302 btrfs_pin_extent(root
, node
->bytenr
,
2303 node
->num_bytes
, 1);
2304 if (head
->is_data
) {
2305 ret
= btrfs_del_csums(trans
, root
,
2313 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2314 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2315 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2317 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2318 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2319 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2326 static noinline
struct btrfs_delayed_ref_node
*
2327 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2329 struct rb_node
*node
;
2330 struct btrfs_delayed_ref_node
*ref
, *last
= NULL
;;
2333 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2334 * this prevents ref count from going down to zero when
2335 * there still are pending delayed ref.
2337 node
= rb_first(&head
->ref_root
);
2339 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2341 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2343 else if (last
== NULL
)
2345 node
= rb_next(node
);
2351 * Returns 0 on success or if called with an already aborted transaction.
2352 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2354 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2355 struct btrfs_root
*root
,
2358 struct btrfs_delayed_ref_root
*delayed_refs
;
2359 struct btrfs_delayed_ref_node
*ref
;
2360 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2361 struct btrfs_delayed_extent_op
*extent_op
;
2362 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2363 ktime_t start
= ktime_get();
2365 unsigned long count
= 0;
2366 unsigned long actual_count
= 0;
2367 int must_insert_reserved
= 0;
2369 delayed_refs
= &trans
->transaction
->delayed_refs
;
2375 spin_lock(&delayed_refs
->lock
);
2376 locked_ref
= btrfs_select_ref_head(trans
);
2378 spin_unlock(&delayed_refs
->lock
);
2382 /* grab the lock that says we are going to process
2383 * all the refs for this head */
2384 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2385 spin_unlock(&delayed_refs
->lock
);
2387 * we may have dropped the spin lock to get the head
2388 * mutex lock, and that might have given someone else
2389 * time to free the head. If that's true, it has been
2390 * removed from our list and we can move on.
2392 if (ret
== -EAGAIN
) {
2400 * We need to try and merge add/drops of the same ref since we
2401 * can run into issues with relocate dropping the implicit ref
2402 * and then it being added back again before the drop can
2403 * finish. If we merged anything we need to re-loop so we can
2406 spin_lock(&locked_ref
->lock
);
2407 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2411 * locked_ref is the head node, so we have to go one
2412 * node back for any delayed ref updates
2414 ref
= select_delayed_ref(locked_ref
);
2416 if (ref
&& ref
->seq
&&
2417 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2418 spin_unlock(&locked_ref
->lock
);
2419 btrfs_delayed_ref_unlock(locked_ref
);
2420 spin_lock(&delayed_refs
->lock
);
2421 locked_ref
->processing
= 0;
2422 delayed_refs
->num_heads_ready
++;
2423 spin_unlock(&delayed_refs
->lock
);
2431 * record the must insert reserved flag before we
2432 * drop the spin lock.
2434 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2435 locked_ref
->must_insert_reserved
= 0;
2437 extent_op
= locked_ref
->extent_op
;
2438 locked_ref
->extent_op
= NULL
;
2443 /* All delayed refs have been processed, Go ahead
2444 * and send the head node to run_one_delayed_ref,
2445 * so that any accounting fixes can happen
2447 ref
= &locked_ref
->node
;
2449 if (extent_op
&& must_insert_reserved
) {
2450 btrfs_free_delayed_extent_op(extent_op
);
2455 spin_unlock(&locked_ref
->lock
);
2456 ret
= run_delayed_extent_op(trans
, root
,
2458 btrfs_free_delayed_extent_op(extent_op
);
2462 * Need to reset must_insert_reserved if
2463 * there was an error so the abort stuff
2464 * can cleanup the reserved space
2467 if (must_insert_reserved
)
2468 locked_ref
->must_insert_reserved
= 1;
2469 locked_ref
->processing
= 0;
2470 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2471 btrfs_delayed_ref_unlock(locked_ref
);
2478 * Need to drop our head ref lock and re-aqcuire the
2479 * delayed ref lock and then re-check to make sure
2482 spin_unlock(&locked_ref
->lock
);
2483 spin_lock(&delayed_refs
->lock
);
2484 spin_lock(&locked_ref
->lock
);
2485 if (rb_first(&locked_ref
->ref_root
) ||
2486 locked_ref
->extent_op
) {
2487 spin_unlock(&locked_ref
->lock
);
2488 spin_unlock(&delayed_refs
->lock
);
2492 delayed_refs
->num_heads
--;
2493 rb_erase(&locked_ref
->href_node
,
2494 &delayed_refs
->href_root
);
2495 spin_unlock(&delayed_refs
->lock
);
2499 rb_erase(&ref
->rb_node
, &locked_ref
->ref_root
);
2501 atomic_dec(&delayed_refs
->num_entries
);
2503 if (!btrfs_delayed_ref_is_head(ref
)) {
2505 * when we play the delayed ref, also correct the
2508 switch (ref
->action
) {
2509 case BTRFS_ADD_DELAYED_REF
:
2510 case BTRFS_ADD_DELAYED_EXTENT
:
2511 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2513 case BTRFS_DROP_DELAYED_REF
:
2514 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2520 spin_unlock(&locked_ref
->lock
);
2522 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2523 must_insert_reserved
);
2525 btrfs_free_delayed_extent_op(extent_op
);
2527 locked_ref
->processing
= 0;
2528 btrfs_delayed_ref_unlock(locked_ref
);
2529 btrfs_put_delayed_ref(ref
);
2530 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2535 * If this node is a head, that means all the refs in this head
2536 * have been dealt with, and we will pick the next head to deal
2537 * with, so we must unlock the head and drop it from the cluster
2538 * list before we release it.
2540 if (btrfs_delayed_ref_is_head(ref
)) {
2541 if (locked_ref
->is_data
&&
2542 locked_ref
->total_ref_mod
< 0) {
2543 spin_lock(&delayed_refs
->lock
);
2544 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2545 spin_unlock(&delayed_refs
->lock
);
2547 btrfs_delayed_ref_unlock(locked_ref
);
2550 btrfs_put_delayed_ref(ref
);
2556 * We don't want to include ref heads since we can have empty ref heads
2557 * and those will drastically skew our runtime down since we just do
2558 * accounting, no actual extent tree updates.
2560 if (actual_count
> 0) {
2561 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2565 * We weigh the current average higher than our current runtime
2566 * to avoid large swings in the average.
2568 spin_lock(&delayed_refs
->lock
);
2569 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2570 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2571 spin_unlock(&delayed_refs
->lock
);
2576 #ifdef SCRAMBLE_DELAYED_REFS
2578 * Normally delayed refs get processed in ascending bytenr order. This
2579 * correlates in most cases to the order added. To expose dependencies on this
2580 * order, we start to process the tree in the middle instead of the beginning
2582 static u64
find_middle(struct rb_root
*root
)
2584 struct rb_node
*n
= root
->rb_node
;
2585 struct btrfs_delayed_ref_node
*entry
;
2588 u64 first
= 0, last
= 0;
2592 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2593 first
= entry
->bytenr
;
2597 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2598 last
= entry
->bytenr
;
2603 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2604 WARN_ON(!entry
->in_tree
);
2606 middle
= entry
->bytenr
;
2619 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2623 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2624 sizeof(struct btrfs_extent_inline_ref
));
2625 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2626 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2629 * We don't ever fill up leaves all the way so multiply by 2 just to be
2630 * closer to what we're really going to want to ouse.
2632 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2636 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2637 * would require to store the csums for that many bytes.
2639 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2642 u64 num_csums_per_leaf
;
2645 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2646 num_csums_per_leaf
= div64_u64(csum_size
,
2647 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2648 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2649 num_csums
+= num_csums_per_leaf
- 1;
2650 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2654 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2655 struct btrfs_root
*root
)
2657 struct btrfs_block_rsv
*global_rsv
;
2658 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2659 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2660 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2661 u64 num_bytes
, num_dirty_bgs_bytes
;
2664 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2665 num_heads
= heads_to_leaves(root
, num_heads
);
2667 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2669 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2670 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2672 global_rsv
= &root
->fs_info
->global_block_rsv
;
2675 * If we can't allocate any more chunks lets make sure we have _lots_ of
2676 * wiggle room since running delayed refs can create more delayed refs.
2678 if (global_rsv
->space_info
->full
) {
2679 num_dirty_bgs_bytes
<<= 1;
2683 spin_lock(&global_rsv
->lock
);
2684 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2686 spin_unlock(&global_rsv
->lock
);
2690 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2691 struct btrfs_root
*root
)
2693 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2695 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2700 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2701 val
= num_entries
* avg_runtime
;
2702 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2704 if (val
>= NSEC_PER_SEC
/ 2)
2707 return btrfs_check_space_for_delayed_refs(trans
, root
);
2710 struct async_delayed_refs
{
2711 struct btrfs_root
*root
;
2715 struct completion wait
;
2716 struct btrfs_work work
;
2719 static void delayed_ref_async_start(struct btrfs_work
*work
)
2721 struct async_delayed_refs
*async
;
2722 struct btrfs_trans_handle
*trans
;
2725 async
= container_of(work
, struct async_delayed_refs
, work
);
2727 trans
= btrfs_join_transaction(async
->root
);
2728 if (IS_ERR(trans
)) {
2729 async
->error
= PTR_ERR(trans
);
2734 * trans->sync means that when we call end_transaciton, we won't
2735 * wait on delayed refs
2738 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2742 ret
= btrfs_end_transaction(trans
, async
->root
);
2743 if (ret
&& !async
->error
)
2747 complete(&async
->wait
);
2752 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2753 unsigned long count
, int wait
)
2755 struct async_delayed_refs
*async
;
2758 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2762 async
->root
= root
->fs_info
->tree_root
;
2763 async
->count
= count
;
2769 init_completion(&async
->wait
);
2771 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2772 delayed_ref_async_start
, NULL
, NULL
);
2774 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2777 wait_for_completion(&async
->wait
);
2786 * this starts processing the delayed reference count updates and
2787 * extent insertions we have queued up so far. count can be
2788 * 0, which means to process everything in the tree at the start
2789 * of the run (but not newly added entries), or it can be some target
2790 * number you'd like to process.
2792 * Returns 0 on success or if called with an aborted transaction
2793 * Returns <0 on error and aborts the transaction
2795 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2796 struct btrfs_root
*root
, unsigned long count
)
2798 struct rb_node
*node
;
2799 struct btrfs_delayed_ref_root
*delayed_refs
;
2800 struct btrfs_delayed_ref_head
*head
;
2802 int run_all
= count
== (unsigned long)-1;
2804 /* We'll clean this up in btrfs_cleanup_transaction */
2808 if (root
== root
->fs_info
->extent_root
)
2809 root
= root
->fs_info
->tree_root
;
2811 delayed_refs
= &trans
->transaction
->delayed_refs
;
2813 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2816 #ifdef SCRAMBLE_DELAYED_REFS
2817 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2819 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2821 btrfs_abort_transaction(trans
, root
, ret
);
2826 if (!list_empty(&trans
->new_bgs
))
2827 btrfs_create_pending_block_groups(trans
, root
);
2829 spin_lock(&delayed_refs
->lock
);
2830 node
= rb_first(&delayed_refs
->href_root
);
2832 spin_unlock(&delayed_refs
->lock
);
2835 count
= (unsigned long)-1;
2838 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2840 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2841 struct btrfs_delayed_ref_node
*ref
;
2844 atomic_inc(&ref
->refs
);
2846 spin_unlock(&delayed_refs
->lock
);
2848 * Mutex was contended, block until it's
2849 * released and try again
2851 mutex_lock(&head
->mutex
);
2852 mutex_unlock(&head
->mutex
);
2854 btrfs_put_delayed_ref(ref
);
2860 node
= rb_next(node
);
2862 spin_unlock(&delayed_refs
->lock
);
2867 ret
= btrfs_delayed_qgroup_accounting(trans
, root
->fs_info
);
2870 assert_qgroups_uptodate(trans
);
2874 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2875 struct btrfs_root
*root
,
2876 u64 bytenr
, u64 num_bytes
, u64 flags
,
2877 int level
, int is_data
)
2879 struct btrfs_delayed_extent_op
*extent_op
;
2882 extent_op
= btrfs_alloc_delayed_extent_op();
2886 extent_op
->flags_to_set
= flags
;
2887 extent_op
->update_flags
= 1;
2888 extent_op
->update_key
= 0;
2889 extent_op
->is_data
= is_data
? 1 : 0;
2890 extent_op
->level
= level
;
2892 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2893 num_bytes
, extent_op
);
2895 btrfs_free_delayed_extent_op(extent_op
);
2899 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2900 struct btrfs_root
*root
,
2901 struct btrfs_path
*path
,
2902 u64 objectid
, u64 offset
, u64 bytenr
)
2904 struct btrfs_delayed_ref_head
*head
;
2905 struct btrfs_delayed_ref_node
*ref
;
2906 struct btrfs_delayed_data_ref
*data_ref
;
2907 struct btrfs_delayed_ref_root
*delayed_refs
;
2908 struct rb_node
*node
;
2911 delayed_refs
= &trans
->transaction
->delayed_refs
;
2912 spin_lock(&delayed_refs
->lock
);
2913 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2915 spin_unlock(&delayed_refs
->lock
);
2919 if (!mutex_trylock(&head
->mutex
)) {
2920 atomic_inc(&head
->node
.refs
);
2921 spin_unlock(&delayed_refs
->lock
);
2923 btrfs_release_path(path
);
2926 * Mutex was contended, block until it's released and let
2929 mutex_lock(&head
->mutex
);
2930 mutex_unlock(&head
->mutex
);
2931 btrfs_put_delayed_ref(&head
->node
);
2934 spin_unlock(&delayed_refs
->lock
);
2936 spin_lock(&head
->lock
);
2937 node
= rb_first(&head
->ref_root
);
2939 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2940 node
= rb_next(node
);
2942 /* If it's a shared ref we know a cross reference exists */
2943 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2948 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2951 * If our ref doesn't match the one we're currently looking at
2952 * then we have a cross reference.
2954 if (data_ref
->root
!= root
->root_key
.objectid
||
2955 data_ref
->objectid
!= objectid
||
2956 data_ref
->offset
!= offset
) {
2961 spin_unlock(&head
->lock
);
2962 mutex_unlock(&head
->mutex
);
2966 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2967 struct btrfs_root
*root
,
2968 struct btrfs_path
*path
,
2969 u64 objectid
, u64 offset
, u64 bytenr
)
2971 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2972 struct extent_buffer
*leaf
;
2973 struct btrfs_extent_data_ref
*ref
;
2974 struct btrfs_extent_inline_ref
*iref
;
2975 struct btrfs_extent_item
*ei
;
2976 struct btrfs_key key
;
2980 key
.objectid
= bytenr
;
2981 key
.offset
= (u64
)-1;
2982 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2984 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2987 BUG_ON(ret
== 0); /* Corruption */
2990 if (path
->slots
[0] == 0)
2994 leaf
= path
->nodes
[0];
2995 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2997 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3001 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3002 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3003 if (item_size
< sizeof(*ei
)) {
3004 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3008 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3010 if (item_size
!= sizeof(*ei
) +
3011 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3014 if (btrfs_extent_generation(leaf
, ei
) <=
3015 btrfs_root_last_snapshot(&root
->root_item
))
3018 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3019 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3020 BTRFS_EXTENT_DATA_REF_KEY
)
3023 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3024 if (btrfs_extent_refs(leaf
, ei
) !=
3025 btrfs_extent_data_ref_count(leaf
, ref
) ||
3026 btrfs_extent_data_ref_root(leaf
, ref
) !=
3027 root
->root_key
.objectid
||
3028 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3029 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3037 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3038 struct btrfs_root
*root
,
3039 u64 objectid
, u64 offset
, u64 bytenr
)
3041 struct btrfs_path
*path
;
3045 path
= btrfs_alloc_path();
3050 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3052 if (ret
&& ret
!= -ENOENT
)
3055 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3057 } while (ret2
== -EAGAIN
);
3059 if (ret2
&& ret2
!= -ENOENT
) {
3064 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3067 btrfs_free_path(path
);
3068 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3073 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3074 struct btrfs_root
*root
,
3075 struct extent_buffer
*buf
,
3076 int full_backref
, int inc
)
3083 struct btrfs_key key
;
3084 struct btrfs_file_extent_item
*fi
;
3088 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3089 u64
, u64
, u64
, u64
, u64
, u64
, int);
3092 if (btrfs_test_is_dummy_root(root
))
3095 ref_root
= btrfs_header_owner(buf
);
3096 nritems
= btrfs_header_nritems(buf
);
3097 level
= btrfs_header_level(buf
);
3099 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3103 process_func
= btrfs_inc_extent_ref
;
3105 process_func
= btrfs_free_extent
;
3108 parent
= buf
->start
;
3112 for (i
= 0; i
< nritems
; i
++) {
3114 btrfs_item_key_to_cpu(buf
, &key
, i
);
3115 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3117 fi
= btrfs_item_ptr(buf
, i
,
3118 struct btrfs_file_extent_item
);
3119 if (btrfs_file_extent_type(buf
, fi
) ==
3120 BTRFS_FILE_EXTENT_INLINE
)
3122 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3126 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3127 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3128 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3129 parent
, ref_root
, key
.objectid
,
3134 bytenr
= btrfs_node_blockptr(buf
, i
);
3135 num_bytes
= root
->nodesize
;
3136 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3137 parent
, ref_root
, level
- 1, 0,
3148 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3149 struct extent_buffer
*buf
, int full_backref
)
3151 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3154 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3155 struct extent_buffer
*buf
, int full_backref
)
3157 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3160 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3161 struct btrfs_root
*root
,
3162 struct btrfs_path
*path
,
3163 struct btrfs_block_group_cache
*cache
)
3166 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3168 struct extent_buffer
*leaf
;
3170 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3177 leaf
= path
->nodes
[0];
3178 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3179 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3180 btrfs_mark_buffer_dirty(leaf
);
3181 btrfs_release_path(path
);
3184 btrfs_abort_transaction(trans
, root
, ret
);
3189 static struct btrfs_block_group_cache
*
3190 next_block_group(struct btrfs_root
*root
,
3191 struct btrfs_block_group_cache
*cache
)
3193 struct rb_node
*node
;
3195 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3197 /* If our block group was removed, we need a full search. */
3198 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3199 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3201 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3202 btrfs_put_block_group(cache
);
3203 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3207 node
= rb_next(&cache
->cache_node
);
3208 btrfs_put_block_group(cache
);
3210 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3212 btrfs_get_block_group(cache
);
3215 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3219 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3220 struct btrfs_trans_handle
*trans
,
3221 struct btrfs_path
*path
)
3223 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3224 struct inode
*inode
= NULL
;
3226 int dcs
= BTRFS_DC_ERROR
;
3232 * If this block group is smaller than 100 megs don't bother caching the
3235 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3236 spin_lock(&block_group
->lock
);
3237 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3238 spin_unlock(&block_group
->lock
);
3245 inode
= lookup_free_space_inode(root
, block_group
, path
);
3246 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3247 ret
= PTR_ERR(inode
);
3248 btrfs_release_path(path
);
3252 if (IS_ERR(inode
)) {
3256 if (block_group
->ro
)
3259 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3265 /* We've already setup this transaction, go ahead and exit */
3266 if (block_group
->cache_generation
== trans
->transid
&&
3267 i_size_read(inode
)) {
3268 dcs
= BTRFS_DC_SETUP
;
3273 * We want to set the generation to 0, that way if anything goes wrong
3274 * from here on out we know not to trust this cache when we load up next
3277 BTRFS_I(inode
)->generation
= 0;
3278 ret
= btrfs_update_inode(trans
, root
, inode
);
3281 * So theoretically we could recover from this, simply set the
3282 * super cache generation to 0 so we know to invalidate the
3283 * cache, but then we'd have to keep track of the block groups
3284 * that fail this way so we know we _have_ to reset this cache
3285 * before the next commit or risk reading stale cache. So to
3286 * limit our exposure to horrible edge cases lets just abort the
3287 * transaction, this only happens in really bad situations
3290 btrfs_abort_transaction(trans
, root
, ret
);
3295 if (i_size_read(inode
) > 0) {
3296 ret
= btrfs_check_trunc_cache_free_space(root
,
3297 &root
->fs_info
->global_block_rsv
);
3301 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3306 spin_lock(&block_group
->lock
);
3307 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3308 !btrfs_test_opt(root
, SPACE_CACHE
) ||
3309 block_group
->delalloc_bytes
) {
3311 * don't bother trying to write stuff out _if_
3312 * a) we're not cached,
3313 * b) we're with nospace_cache mount option.
3315 dcs
= BTRFS_DC_WRITTEN
;
3316 spin_unlock(&block_group
->lock
);
3319 spin_unlock(&block_group
->lock
);
3322 * Try to preallocate enough space based on how big the block group is.
3323 * Keep in mind this has to include any pinned space which could end up
3324 * taking up quite a bit since it's not folded into the other space
3327 num_pages
= div_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3332 num_pages
*= PAGE_CACHE_SIZE
;
3334 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3338 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3339 num_pages
, num_pages
,
3342 dcs
= BTRFS_DC_SETUP
;
3343 btrfs_free_reserved_data_space(inode
, num_pages
);
3348 btrfs_release_path(path
);
3350 spin_lock(&block_group
->lock
);
3351 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3352 block_group
->cache_generation
= trans
->transid
;
3353 block_group
->disk_cache_state
= dcs
;
3354 spin_unlock(&block_group
->lock
);
3359 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3360 struct btrfs_root
*root
)
3362 struct btrfs_block_group_cache
*cache
, *tmp
;
3363 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3364 struct btrfs_path
*path
;
3366 if (list_empty(&cur_trans
->dirty_bgs
) ||
3367 !btrfs_test_opt(root
, SPACE_CACHE
))
3370 path
= btrfs_alloc_path();
3374 /* Could add new block groups, use _safe just in case */
3375 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3377 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3378 cache_save_setup(cache
, trans
, path
);
3381 btrfs_free_path(path
);
3386 * transaction commit does final block group cache writeback during a
3387 * critical section where nothing is allowed to change the FS. This is
3388 * required in order for the cache to actually match the block group,
3389 * but can introduce a lot of latency into the commit.
3391 * So, btrfs_start_dirty_block_groups is here to kick off block group
3392 * cache IO. There's a chance we'll have to redo some of it if the
3393 * block group changes again during the commit, but it greatly reduces
3394 * the commit latency by getting rid of the easy block groups while
3395 * we're still allowing others to join the commit.
3397 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3398 struct btrfs_root
*root
)
3400 struct btrfs_block_group_cache
*cache
;
3401 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3404 struct btrfs_path
*path
= NULL
;
3406 struct list_head
*io
= &cur_trans
->io_bgs
;
3407 int num_started
= 0;
3410 spin_lock(&cur_trans
->dirty_bgs_lock
);
3411 if (!list_empty(&cur_trans
->dirty_bgs
)) {
3412 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3414 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3417 if (list_empty(&dirty
)) {
3418 btrfs_free_path(path
);
3423 * make sure all the block groups on our dirty list actually
3426 btrfs_create_pending_block_groups(trans
, root
);
3429 path
= btrfs_alloc_path();
3434 while (!list_empty(&dirty
)) {
3435 cache
= list_first_entry(&dirty
,
3436 struct btrfs_block_group_cache
,
3440 * cache_write_mutex is here only to save us from balance
3441 * deleting this block group while we are writing out the
3444 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3447 * this can happen if something re-dirties a block
3448 * group that is already under IO. Just wait for it to
3449 * finish and then do it all again
3451 if (!list_empty(&cache
->io_list
)) {
3452 list_del_init(&cache
->io_list
);
3453 btrfs_wait_cache_io(root
, trans
, cache
,
3454 &cache
->io_ctl
, path
,
3455 cache
->key
.objectid
);
3456 btrfs_put_block_group(cache
);
3461 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3462 * if it should update the cache_state. Don't delete
3463 * until after we wait.
3465 * Since we're not running in the commit critical section
3466 * we need the dirty_bgs_lock to protect from update_block_group
3468 spin_lock(&cur_trans
->dirty_bgs_lock
);
3469 list_del_init(&cache
->dirty_list
);
3470 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3474 cache_save_setup(cache
, trans
, path
);
3476 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3477 cache
->io_ctl
.inode
= NULL
;
3478 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3479 if (ret
== 0 && cache
->io_ctl
.inode
) {
3484 * the cache_write_mutex is protecting
3487 list_add_tail(&cache
->io_list
, io
);
3490 * if we failed to write the cache, the
3491 * generation will be bad and life goes on
3497 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3498 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3500 /* if its not on the io list, we need to put the block group */
3502 btrfs_put_block_group(cache
);
3509 * go through delayed refs for all the stuff we've just kicked off
3510 * and then loop back (just once)
3512 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3513 if (!ret
&& loops
== 0) {
3515 spin_lock(&cur_trans
->dirty_bgs_lock
);
3516 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3517 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3521 btrfs_free_path(path
);
3525 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3526 struct btrfs_root
*root
)
3528 struct btrfs_block_group_cache
*cache
;
3529 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3532 struct btrfs_path
*path
;
3533 struct list_head
*io
= &cur_trans
->io_bgs
;
3534 int num_started
= 0;
3536 path
= btrfs_alloc_path();
3541 * We don't need the lock here since we are protected by the transaction
3542 * commit. We want to do the cache_save_setup first and then run the
3543 * delayed refs to make sure we have the best chance at doing this all
3546 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3547 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3548 struct btrfs_block_group_cache
,
3552 * this can happen if cache_save_setup re-dirties a block
3553 * group that is already under IO. Just wait for it to
3554 * finish and then do it all again
3556 if (!list_empty(&cache
->io_list
)) {
3557 list_del_init(&cache
->io_list
);
3558 btrfs_wait_cache_io(root
, trans
, cache
,
3559 &cache
->io_ctl
, path
,
3560 cache
->key
.objectid
);
3561 btrfs_put_block_group(cache
);
3565 * don't remove from the dirty list until after we've waited
3568 list_del_init(&cache
->dirty_list
);
3571 cache_save_setup(cache
, trans
, path
);
3574 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3576 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3577 cache
->io_ctl
.inode
= NULL
;
3578 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3579 if (ret
== 0 && cache
->io_ctl
.inode
) {
3582 list_add_tail(&cache
->io_list
, io
);
3585 * if we failed to write the cache, the
3586 * generation will be bad and life goes on
3592 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3594 /* if its not on the io list, we need to put the block group */
3596 btrfs_put_block_group(cache
);
3599 while (!list_empty(io
)) {
3600 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3602 list_del_init(&cache
->io_list
);
3603 btrfs_wait_cache_io(root
, trans
, cache
,
3604 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3605 btrfs_put_block_group(cache
);
3608 btrfs_free_path(path
);
3612 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3614 struct btrfs_block_group_cache
*block_group
;
3617 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3618 if (!block_group
|| block_group
->ro
)
3621 btrfs_put_block_group(block_group
);
3625 static const char *alloc_name(u64 flags
)
3628 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3630 case BTRFS_BLOCK_GROUP_METADATA
:
3632 case BTRFS_BLOCK_GROUP_DATA
:
3634 case BTRFS_BLOCK_GROUP_SYSTEM
:
3638 return "invalid-combination";
3642 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3643 u64 total_bytes
, u64 bytes_used
,
3644 struct btrfs_space_info
**space_info
)
3646 struct btrfs_space_info
*found
;
3651 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3652 BTRFS_BLOCK_GROUP_RAID10
))
3657 found
= __find_space_info(info
, flags
);
3659 spin_lock(&found
->lock
);
3660 found
->total_bytes
+= total_bytes
;
3661 found
->disk_total
+= total_bytes
* factor
;
3662 found
->bytes_used
+= bytes_used
;
3663 found
->disk_used
+= bytes_used
* factor
;
3665 spin_unlock(&found
->lock
);
3666 *space_info
= found
;
3669 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3673 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3679 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3680 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3681 init_rwsem(&found
->groups_sem
);
3682 spin_lock_init(&found
->lock
);
3683 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3684 found
->total_bytes
= total_bytes
;
3685 found
->disk_total
= total_bytes
* factor
;
3686 found
->bytes_used
= bytes_used
;
3687 found
->disk_used
= bytes_used
* factor
;
3688 found
->bytes_pinned
= 0;
3689 found
->bytes_reserved
= 0;
3690 found
->bytes_readonly
= 0;
3691 found
->bytes_may_use
= 0;
3693 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3694 found
->chunk_alloc
= 0;
3696 init_waitqueue_head(&found
->wait
);
3697 INIT_LIST_HEAD(&found
->ro_bgs
);
3699 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3700 info
->space_info_kobj
, "%s",
3701 alloc_name(found
->flags
));
3707 *space_info
= found
;
3708 list_add_rcu(&found
->list
, &info
->space_info
);
3709 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3710 info
->data_sinfo
= found
;
3715 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3717 u64 extra_flags
= chunk_to_extended(flags
) &
3718 BTRFS_EXTENDED_PROFILE_MASK
;
3720 write_seqlock(&fs_info
->profiles_lock
);
3721 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3722 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3723 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3724 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3725 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3726 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3727 write_sequnlock(&fs_info
->profiles_lock
);
3731 * returns target flags in extended format or 0 if restripe for this
3732 * chunk_type is not in progress
3734 * should be called with either volume_mutex or balance_lock held
3736 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3738 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3744 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3745 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3746 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3747 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3748 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3749 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3750 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3751 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3752 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3759 * @flags: available profiles in extended format (see ctree.h)
3761 * Returns reduced profile in chunk format. If profile changing is in
3762 * progress (either running or paused) picks the target profile (if it's
3763 * already available), otherwise falls back to plain reducing.
3765 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3767 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3772 * see if restripe for this chunk_type is in progress, if so
3773 * try to reduce to the target profile
3775 spin_lock(&root
->fs_info
->balance_lock
);
3776 target
= get_restripe_target(root
->fs_info
, flags
);
3778 /* pick target profile only if it's already available */
3779 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3780 spin_unlock(&root
->fs_info
->balance_lock
);
3781 return extended_to_chunk(target
);
3784 spin_unlock(&root
->fs_info
->balance_lock
);
3786 /* First, mask out the RAID levels which aren't possible */
3787 if (num_devices
== 1)
3788 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3789 BTRFS_BLOCK_GROUP_RAID5
);
3790 if (num_devices
< 3)
3791 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3792 if (num_devices
< 4)
3793 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3795 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3796 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3797 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3800 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3801 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3802 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3803 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3804 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3805 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3806 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3807 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3808 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3809 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3811 return extended_to_chunk(flags
| tmp
);
3814 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3821 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3823 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3824 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3825 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3826 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3827 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3828 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3829 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3831 return btrfs_reduce_alloc_profile(root
, flags
);
3834 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3840 flags
= BTRFS_BLOCK_GROUP_DATA
;
3841 else if (root
== root
->fs_info
->chunk_root
)
3842 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3844 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3846 ret
= get_alloc_profile(root
, flags
);
3851 * This will check the space that the inode allocates from to make sure we have
3852 * enough space for bytes.
3854 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3856 struct btrfs_space_info
*data_sinfo
;
3857 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3858 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3862 int have_pinned_space
= 1;
3864 /* make sure bytes are sectorsize aligned */
3865 bytes
= ALIGN(bytes
, root
->sectorsize
);
3867 if (btrfs_is_free_space_inode(inode
)) {
3869 ASSERT(current
->journal_info
);
3872 data_sinfo
= fs_info
->data_sinfo
;
3877 /* make sure we have enough space to handle the data first */
3878 spin_lock(&data_sinfo
->lock
);
3879 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3880 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3881 data_sinfo
->bytes_may_use
;
3883 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3884 struct btrfs_trans_handle
*trans
;
3887 * if we don't have enough free bytes in this space then we need
3888 * to alloc a new chunk.
3890 if (!data_sinfo
->full
) {
3893 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3894 spin_unlock(&data_sinfo
->lock
);
3896 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3898 * It is ugly that we don't call nolock join
3899 * transaction for the free space inode case here.
3900 * But it is safe because we only do the data space
3901 * reservation for the free space cache in the
3902 * transaction context, the common join transaction
3903 * just increase the counter of the current transaction
3904 * handler, doesn't try to acquire the trans_lock of
3907 trans
= btrfs_join_transaction(root
);
3909 return PTR_ERR(trans
);
3911 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3913 CHUNK_ALLOC_NO_FORCE
);
3914 btrfs_end_transaction(trans
, root
);
3923 data_sinfo
= fs_info
->data_sinfo
;
3929 * If we don't have enough pinned space to deal with this
3930 * allocation, and no removed chunk in current transaction,
3931 * don't bother committing the transaction.
3933 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3935 have_pinned_space
= 0;
3936 spin_unlock(&data_sinfo
->lock
);
3938 /* commit the current transaction and try again */
3941 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3944 trans
= btrfs_join_transaction(root
);
3946 return PTR_ERR(trans
);
3947 if (have_pinned_space
||
3948 trans
->transaction
->have_free_bgs
) {
3949 ret
= btrfs_commit_transaction(trans
, root
);
3954 btrfs_end_transaction(trans
, root
);
3958 trace_btrfs_space_reservation(root
->fs_info
,
3959 "space_info:enospc",
3960 data_sinfo
->flags
, bytes
, 1);
3963 data_sinfo
->bytes_may_use
+= bytes
;
3964 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3965 data_sinfo
->flags
, bytes
, 1);
3966 spin_unlock(&data_sinfo
->lock
);
3972 * Called if we need to clear a data reservation for this inode.
3974 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3976 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3977 struct btrfs_space_info
*data_sinfo
;
3979 /* make sure bytes are sectorsize aligned */
3980 bytes
= ALIGN(bytes
, root
->sectorsize
);
3982 data_sinfo
= root
->fs_info
->data_sinfo
;
3983 spin_lock(&data_sinfo
->lock
);
3984 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3985 data_sinfo
->bytes_may_use
-= bytes
;
3986 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3987 data_sinfo
->flags
, bytes
, 0);
3988 spin_unlock(&data_sinfo
->lock
);
3991 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3993 struct list_head
*head
= &info
->space_info
;
3994 struct btrfs_space_info
*found
;
3997 list_for_each_entry_rcu(found
, head
, list
) {
3998 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3999 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4004 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4006 return (global
->size
<< 1);
4009 static int should_alloc_chunk(struct btrfs_root
*root
,
4010 struct btrfs_space_info
*sinfo
, int force
)
4012 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4013 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4014 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4017 if (force
== CHUNK_ALLOC_FORCE
)
4021 * We need to take into account the global rsv because for all intents
4022 * and purposes it's used space. Don't worry about locking the
4023 * global_rsv, it doesn't change except when the transaction commits.
4025 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4026 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4029 * in limited mode, we want to have some free space up to
4030 * about 1% of the FS size.
4032 if (force
== CHUNK_ALLOC_LIMITED
) {
4033 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4034 thresh
= max_t(u64
, 64 * 1024 * 1024,
4035 div_factor_fine(thresh
, 1));
4037 if (num_bytes
- num_allocated
< thresh
)
4041 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
4046 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
4050 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4051 BTRFS_BLOCK_GROUP_RAID0
|
4052 BTRFS_BLOCK_GROUP_RAID5
|
4053 BTRFS_BLOCK_GROUP_RAID6
))
4054 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4055 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4058 num_dev
= 1; /* DUP or single */
4060 /* metadata for updaing devices and chunk tree */
4061 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
4064 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
4065 struct btrfs_root
*root
, u64 type
)
4067 struct btrfs_space_info
*info
;
4071 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4072 spin_lock(&info
->lock
);
4073 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4074 info
->bytes_reserved
- info
->bytes_readonly
;
4075 spin_unlock(&info
->lock
);
4077 thresh
= get_system_chunk_thresh(root
, type
);
4078 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4079 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4080 left
, thresh
, type
);
4081 dump_space_info(info
, 0, 0);
4084 if (left
< thresh
) {
4087 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4088 btrfs_alloc_chunk(trans
, root
, flags
);
4092 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4093 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4095 struct btrfs_space_info
*space_info
;
4096 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4097 int wait_for_alloc
= 0;
4100 /* Don't re-enter if we're already allocating a chunk */
4101 if (trans
->allocating_chunk
)
4104 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4106 ret
= update_space_info(extent_root
->fs_info
, flags
,
4108 BUG_ON(ret
); /* -ENOMEM */
4110 BUG_ON(!space_info
); /* Logic error */
4113 spin_lock(&space_info
->lock
);
4114 if (force
< space_info
->force_alloc
)
4115 force
= space_info
->force_alloc
;
4116 if (space_info
->full
) {
4117 if (should_alloc_chunk(extent_root
, space_info
, force
))
4121 spin_unlock(&space_info
->lock
);
4125 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4126 spin_unlock(&space_info
->lock
);
4128 } else if (space_info
->chunk_alloc
) {
4131 space_info
->chunk_alloc
= 1;
4134 spin_unlock(&space_info
->lock
);
4136 mutex_lock(&fs_info
->chunk_mutex
);
4139 * The chunk_mutex is held throughout the entirety of a chunk
4140 * allocation, so once we've acquired the chunk_mutex we know that the
4141 * other guy is done and we need to recheck and see if we should
4144 if (wait_for_alloc
) {
4145 mutex_unlock(&fs_info
->chunk_mutex
);
4150 trans
->allocating_chunk
= true;
4153 * If we have mixed data/metadata chunks we want to make sure we keep
4154 * allocating mixed chunks instead of individual chunks.
4156 if (btrfs_mixed_space_info(space_info
))
4157 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4160 * if we're doing a data chunk, go ahead and make sure that
4161 * we keep a reasonable number of metadata chunks allocated in the
4164 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4165 fs_info
->data_chunk_allocations
++;
4166 if (!(fs_info
->data_chunk_allocations
%
4167 fs_info
->metadata_ratio
))
4168 force_metadata_allocation(fs_info
);
4172 * Check if we have enough space in SYSTEM chunk because we may need
4173 * to update devices.
4175 check_system_chunk(trans
, extent_root
, flags
);
4177 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4178 trans
->allocating_chunk
= false;
4180 spin_lock(&space_info
->lock
);
4181 if (ret
< 0 && ret
!= -ENOSPC
)
4184 space_info
->full
= 1;
4188 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4190 space_info
->chunk_alloc
= 0;
4191 spin_unlock(&space_info
->lock
);
4192 mutex_unlock(&fs_info
->chunk_mutex
);
4196 static int can_overcommit(struct btrfs_root
*root
,
4197 struct btrfs_space_info
*space_info
, u64 bytes
,
4198 enum btrfs_reserve_flush_enum flush
)
4200 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4201 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4206 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4207 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4210 * We only want to allow over committing if we have lots of actual space
4211 * free, but if we don't have enough space to handle the global reserve
4212 * space then we could end up having a real enospc problem when trying
4213 * to allocate a chunk or some other such important allocation.
4215 spin_lock(&global_rsv
->lock
);
4216 space_size
= calc_global_rsv_need_space(global_rsv
);
4217 spin_unlock(&global_rsv
->lock
);
4218 if (used
+ space_size
>= space_info
->total_bytes
)
4221 used
+= space_info
->bytes_may_use
;
4223 spin_lock(&root
->fs_info
->free_chunk_lock
);
4224 avail
= root
->fs_info
->free_chunk_space
;
4225 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4228 * If we have dup, raid1 or raid10 then only half of the free
4229 * space is actually useable. For raid56, the space info used
4230 * doesn't include the parity drive, so we don't have to
4233 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4234 BTRFS_BLOCK_GROUP_RAID1
|
4235 BTRFS_BLOCK_GROUP_RAID10
))
4239 * If we aren't flushing all things, let us overcommit up to
4240 * 1/2th of the space. If we can flush, don't let us overcommit
4241 * too much, let it overcommit up to 1/8 of the space.
4243 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4248 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4253 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4254 unsigned long nr_pages
, int nr_items
)
4256 struct super_block
*sb
= root
->fs_info
->sb
;
4258 if (down_read_trylock(&sb
->s_umount
)) {
4259 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4260 up_read(&sb
->s_umount
);
4263 * We needn't worry the filesystem going from r/w to r/o though
4264 * we don't acquire ->s_umount mutex, because the filesystem
4265 * should guarantee the delalloc inodes list be empty after
4266 * the filesystem is readonly(all dirty pages are written to
4269 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4270 if (!current
->journal_info
)
4271 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4275 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4280 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4281 nr
= (int)div64_u64(to_reclaim
, bytes
);
4287 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4290 * shrink metadata reservation for delalloc
4292 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4295 struct btrfs_block_rsv
*block_rsv
;
4296 struct btrfs_space_info
*space_info
;
4297 struct btrfs_trans_handle
*trans
;
4301 unsigned long nr_pages
;
4304 enum btrfs_reserve_flush_enum flush
;
4306 /* Calc the number of the pages we need flush for space reservation */
4307 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4308 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4310 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4311 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4312 space_info
= block_rsv
->space_info
;
4314 delalloc_bytes
= percpu_counter_sum_positive(
4315 &root
->fs_info
->delalloc_bytes
);
4316 if (delalloc_bytes
== 0) {
4320 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4325 while (delalloc_bytes
&& loops
< 3) {
4326 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4327 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4328 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4330 * We need to wait for the async pages to actually start before
4333 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4337 if (max_reclaim
<= nr_pages
)
4340 max_reclaim
-= nr_pages
;
4342 wait_event(root
->fs_info
->async_submit_wait
,
4343 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4347 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4349 flush
= BTRFS_RESERVE_NO_FLUSH
;
4350 spin_lock(&space_info
->lock
);
4351 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4352 spin_unlock(&space_info
->lock
);
4355 spin_unlock(&space_info
->lock
);
4358 if (wait_ordered
&& !trans
) {
4359 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4361 time_left
= schedule_timeout_killable(1);
4365 delalloc_bytes
= percpu_counter_sum_positive(
4366 &root
->fs_info
->delalloc_bytes
);
4371 * maybe_commit_transaction - possibly commit the transaction if its ok to
4372 * @root - the root we're allocating for
4373 * @bytes - the number of bytes we want to reserve
4374 * @force - force the commit
4376 * This will check to make sure that committing the transaction will actually
4377 * get us somewhere and then commit the transaction if it does. Otherwise it
4378 * will return -ENOSPC.
4380 static int may_commit_transaction(struct btrfs_root
*root
,
4381 struct btrfs_space_info
*space_info
,
4382 u64 bytes
, int force
)
4384 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4385 struct btrfs_trans_handle
*trans
;
4387 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4394 /* See if there is enough pinned space to make this reservation */
4395 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4400 * See if there is some space in the delayed insertion reservation for
4403 if (space_info
!= delayed_rsv
->space_info
)
4406 spin_lock(&delayed_rsv
->lock
);
4407 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4408 bytes
- delayed_rsv
->size
) >= 0) {
4409 spin_unlock(&delayed_rsv
->lock
);
4412 spin_unlock(&delayed_rsv
->lock
);
4415 trans
= btrfs_join_transaction(root
);
4419 return btrfs_commit_transaction(trans
, root
);
4423 FLUSH_DELAYED_ITEMS_NR
= 1,
4424 FLUSH_DELAYED_ITEMS
= 2,
4426 FLUSH_DELALLOC_WAIT
= 4,
4431 static int flush_space(struct btrfs_root
*root
,
4432 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4433 u64 orig_bytes
, int state
)
4435 struct btrfs_trans_handle
*trans
;
4440 case FLUSH_DELAYED_ITEMS_NR
:
4441 case FLUSH_DELAYED_ITEMS
:
4442 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4443 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4447 trans
= btrfs_join_transaction(root
);
4448 if (IS_ERR(trans
)) {
4449 ret
= PTR_ERR(trans
);
4452 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4453 btrfs_end_transaction(trans
, root
);
4455 case FLUSH_DELALLOC
:
4456 case FLUSH_DELALLOC_WAIT
:
4457 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4458 state
== FLUSH_DELALLOC_WAIT
);
4461 trans
= btrfs_join_transaction(root
);
4462 if (IS_ERR(trans
)) {
4463 ret
= PTR_ERR(trans
);
4466 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4467 btrfs_get_alloc_profile(root
, 0),
4468 CHUNK_ALLOC_NO_FORCE
);
4469 btrfs_end_transaction(trans
, root
);
4474 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4485 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4486 struct btrfs_space_info
*space_info
)
4492 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4494 spin_lock(&space_info
->lock
);
4495 if (can_overcommit(root
, space_info
, to_reclaim
,
4496 BTRFS_RESERVE_FLUSH_ALL
)) {
4501 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4502 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4503 space_info
->bytes_may_use
;
4504 if (can_overcommit(root
, space_info
, 1024 * 1024,
4505 BTRFS_RESERVE_FLUSH_ALL
))
4506 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4508 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4510 if (used
> expected
)
4511 to_reclaim
= used
- expected
;
4514 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4515 space_info
->bytes_reserved
);
4517 spin_unlock(&space_info
->lock
);
4522 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4523 struct btrfs_fs_info
*fs_info
, u64 used
)
4525 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4527 /* If we're just plain full then async reclaim just slows us down. */
4528 if (space_info
->bytes_used
>= thresh
)
4531 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4532 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4535 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4536 struct btrfs_fs_info
*fs_info
,
4541 spin_lock(&space_info
->lock
);
4543 * We run out of space and have not got any free space via flush_space,
4544 * so don't bother doing async reclaim.
4546 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4547 spin_unlock(&space_info
->lock
);
4551 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4552 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4553 space_info
->bytes_may_use
;
4554 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4555 spin_unlock(&space_info
->lock
);
4558 spin_unlock(&space_info
->lock
);
4563 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4565 struct btrfs_fs_info
*fs_info
;
4566 struct btrfs_space_info
*space_info
;
4570 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4571 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4573 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4578 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4580 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4581 to_reclaim
, flush_state
);
4583 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4586 } while (flush_state
< COMMIT_TRANS
);
4589 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4591 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4595 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4596 * @root - the root we're allocating for
4597 * @block_rsv - the block_rsv we're allocating for
4598 * @orig_bytes - the number of bytes we want
4599 * @flush - whether or not we can flush to make our reservation
4601 * This will reserve orgi_bytes number of bytes from the space info associated
4602 * with the block_rsv. If there is not enough space it will make an attempt to
4603 * flush out space to make room. It will do this by flushing delalloc if
4604 * possible or committing the transaction. If flush is 0 then no attempts to
4605 * regain reservations will be made and this will fail if there is not enough
4608 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4609 struct btrfs_block_rsv
*block_rsv
,
4611 enum btrfs_reserve_flush_enum flush
)
4613 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4615 u64 num_bytes
= orig_bytes
;
4616 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4618 bool flushing
= false;
4622 spin_lock(&space_info
->lock
);
4624 * We only want to wait if somebody other than us is flushing and we
4625 * are actually allowed to flush all things.
4627 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4628 space_info
->flush
) {
4629 spin_unlock(&space_info
->lock
);
4631 * If we have a trans handle we can't wait because the flusher
4632 * may have to commit the transaction, which would mean we would
4633 * deadlock since we are waiting for the flusher to finish, but
4634 * hold the current transaction open.
4636 if (current
->journal_info
)
4638 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4639 /* Must have been killed, return */
4643 spin_lock(&space_info
->lock
);
4647 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4648 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4649 space_info
->bytes_may_use
;
4652 * The idea here is that we've not already over-reserved the block group
4653 * then we can go ahead and save our reservation first and then start
4654 * flushing if we need to. Otherwise if we've already overcommitted
4655 * lets start flushing stuff first and then come back and try to make
4658 if (used
<= space_info
->total_bytes
) {
4659 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4660 space_info
->bytes_may_use
+= orig_bytes
;
4661 trace_btrfs_space_reservation(root
->fs_info
,
4662 "space_info", space_info
->flags
, orig_bytes
, 1);
4666 * Ok set num_bytes to orig_bytes since we aren't
4667 * overocmmitted, this way we only try and reclaim what
4670 num_bytes
= orig_bytes
;
4674 * Ok we're over committed, set num_bytes to the overcommitted
4675 * amount plus the amount of bytes that we need for this
4678 num_bytes
= used
- space_info
->total_bytes
+
4682 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4683 space_info
->bytes_may_use
+= orig_bytes
;
4684 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4685 space_info
->flags
, orig_bytes
,
4691 * Couldn't make our reservation, save our place so while we're trying
4692 * to reclaim space we can actually use it instead of somebody else
4693 * stealing it from us.
4695 * We make the other tasks wait for the flush only when we can flush
4698 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4700 space_info
->flush
= 1;
4701 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4704 * We will do the space reservation dance during log replay,
4705 * which means we won't have fs_info->fs_root set, so don't do
4706 * the async reclaim as we will panic.
4708 if (!root
->fs_info
->log_root_recovering
&&
4709 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4710 !work_busy(&root
->fs_info
->async_reclaim_work
))
4711 queue_work(system_unbound_wq
,
4712 &root
->fs_info
->async_reclaim_work
);
4714 spin_unlock(&space_info
->lock
);
4716 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4719 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4724 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4725 * would happen. So skip delalloc flush.
4727 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4728 (flush_state
== FLUSH_DELALLOC
||
4729 flush_state
== FLUSH_DELALLOC_WAIT
))
4730 flush_state
= ALLOC_CHUNK
;
4734 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4735 flush_state
< COMMIT_TRANS
)
4737 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4738 flush_state
<= COMMIT_TRANS
)
4742 if (ret
== -ENOSPC
&&
4743 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4744 struct btrfs_block_rsv
*global_rsv
=
4745 &root
->fs_info
->global_block_rsv
;
4747 if (block_rsv
!= global_rsv
&&
4748 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4752 trace_btrfs_space_reservation(root
->fs_info
,
4753 "space_info:enospc",
4754 space_info
->flags
, orig_bytes
, 1);
4756 spin_lock(&space_info
->lock
);
4757 space_info
->flush
= 0;
4758 wake_up_all(&space_info
->wait
);
4759 spin_unlock(&space_info
->lock
);
4764 static struct btrfs_block_rsv
*get_block_rsv(
4765 const struct btrfs_trans_handle
*trans
,
4766 const struct btrfs_root
*root
)
4768 struct btrfs_block_rsv
*block_rsv
= NULL
;
4770 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4771 block_rsv
= trans
->block_rsv
;
4773 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4774 block_rsv
= trans
->block_rsv
;
4776 if (root
== root
->fs_info
->uuid_root
)
4777 block_rsv
= trans
->block_rsv
;
4780 block_rsv
= root
->block_rsv
;
4783 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4788 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4792 spin_lock(&block_rsv
->lock
);
4793 if (block_rsv
->reserved
>= num_bytes
) {
4794 block_rsv
->reserved
-= num_bytes
;
4795 if (block_rsv
->reserved
< block_rsv
->size
)
4796 block_rsv
->full
= 0;
4799 spin_unlock(&block_rsv
->lock
);
4803 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4804 u64 num_bytes
, int update_size
)
4806 spin_lock(&block_rsv
->lock
);
4807 block_rsv
->reserved
+= num_bytes
;
4809 block_rsv
->size
+= num_bytes
;
4810 else if (block_rsv
->reserved
>= block_rsv
->size
)
4811 block_rsv
->full
= 1;
4812 spin_unlock(&block_rsv
->lock
);
4815 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4816 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4819 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4822 if (global_rsv
->space_info
!= dest
->space_info
)
4825 spin_lock(&global_rsv
->lock
);
4826 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4827 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4828 spin_unlock(&global_rsv
->lock
);
4831 global_rsv
->reserved
-= num_bytes
;
4832 if (global_rsv
->reserved
< global_rsv
->size
)
4833 global_rsv
->full
= 0;
4834 spin_unlock(&global_rsv
->lock
);
4836 block_rsv_add_bytes(dest
, num_bytes
, 1);
4840 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4841 struct btrfs_block_rsv
*block_rsv
,
4842 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4844 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4846 spin_lock(&block_rsv
->lock
);
4847 if (num_bytes
== (u64
)-1)
4848 num_bytes
= block_rsv
->size
;
4849 block_rsv
->size
-= num_bytes
;
4850 if (block_rsv
->reserved
>= block_rsv
->size
) {
4851 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4852 block_rsv
->reserved
= block_rsv
->size
;
4853 block_rsv
->full
= 1;
4857 spin_unlock(&block_rsv
->lock
);
4859 if (num_bytes
> 0) {
4861 spin_lock(&dest
->lock
);
4865 bytes_to_add
= dest
->size
- dest
->reserved
;
4866 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4867 dest
->reserved
+= bytes_to_add
;
4868 if (dest
->reserved
>= dest
->size
)
4870 num_bytes
-= bytes_to_add
;
4872 spin_unlock(&dest
->lock
);
4875 spin_lock(&space_info
->lock
);
4876 space_info
->bytes_may_use
-= num_bytes
;
4877 trace_btrfs_space_reservation(fs_info
, "space_info",
4878 space_info
->flags
, num_bytes
, 0);
4879 spin_unlock(&space_info
->lock
);
4884 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4885 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4889 ret
= block_rsv_use_bytes(src
, num_bytes
);
4893 block_rsv_add_bytes(dst
, num_bytes
, 1);
4897 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4899 memset(rsv
, 0, sizeof(*rsv
));
4900 spin_lock_init(&rsv
->lock
);
4904 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4905 unsigned short type
)
4907 struct btrfs_block_rsv
*block_rsv
;
4908 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4910 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4914 btrfs_init_block_rsv(block_rsv
, type
);
4915 block_rsv
->space_info
= __find_space_info(fs_info
,
4916 BTRFS_BLOCK_GROUP_METADATA
);
4920 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4921 struct btrfs_block_rsv
*rsv
)
4925 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4929 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
4934 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4935 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4936 enum btrfs_reserve_flush_enum flush
)
4943 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4945 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4952 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4953 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4961 spin_lock(&block_rsv
->lock
);
4962 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4963 if (block_rsv
->reserved
>= num_bytes
)
4965 spin_unlock(&block_rsv
->lock
);
4970 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4971 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4972 enum btrfs_reserve_flush_enum flush
)
4980 spin_lock(&block_rsv
->lock
);
4981 num_bytes
= min_reserved
;
4982 if (block_rsv
->reserved
>= num_bytes
)
4985 num_bytes
-= block_rsv
->reserved
;
4986 spin_unlock(&block_rsv
->lock
);
4991 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4993 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5000 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
5001 struct btrfs_block_rsv
*dst_rsv
,
5004 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5007 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5008 struct btrfs_block_rsv
*block_rsv
,
5011 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5012 if (global_rsv
== block_rsv
||
5013 block_rsv
->space_info
!= global_rsv
->space_info
)
5015 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5020 * helper to calculate size of global block reservation.
5021 * the desired value is sum of space used by extent tree,
5022 * checksum tree and root tree
5024 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5026 struct btrfs_space_info
*sinfo
;
5030 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5032 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5033 spin_lock(&sinfo
->lock
);
5034 data_used
= sinfo
->bytes_used
;
5035 spin_unlock(&sinfo
->lock
);
5037 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5038 spin_lock(&sinfo
->lock
);
5039 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5041 meta_used
= sinfo
->bytes_used
;
5042 spin_unlock(&sinfo
->lock
);
5044 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5046 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5048 if (num_bytes
* 3 > meta_used
)
5049 num_bytes
= div_u64(meta_used
, 3);
5051 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5054 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5056 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5057 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5060 num_bytes
= calc_global_metadata_size(fs_info
);
5062 spin_lock(&sinfo
->lock
);
5063 spin_lock(&block_rsv
->lock
);
5065 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
5067 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5068 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5069 sinfo
->bytes_may_use
;
5071 if (sinfo
->total_bytes
> num_bytes
) {
5072 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5073 block_rsv
->reserved
+= num_bytes
;
5074 sinfo
->bytes_may_use
+= num_bytes
;
5075 trace_btrfs_space_reservation(fs_info
, "space_info",
5076 sinfo
->flags
, num_bytes
, 1);
5079 if (block_rsv
->reserved
>= block_rsv
->size
) {
5080 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5081 sinfo
->bytes_may_use
-= num_bytes
;
5082 trace_btrfs_space_reservation(fs_info
, "space_info",
5083 sinfo
->flags
, num_bytes
, 0);
5084 block_rsv
->reserved
= block_rsv
->size
;
5085 block_rsv
->full
= 1;
5088 spin_unlock(&block_rsv
->lock
);
5089 spin_unlock(&sinfo
->lock
);
5092 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5094 struct btrfs_space_info
*space_info
;
5096 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5097 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5099 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5100 fs_info
->global_block_rsv
.space_info
= space_info
;
5101 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5102 fs_info
->trans_block_rsv
.space_info
= space_info
;
5103 fs_info
->empty_block_rsv
.space_info
= space_info
;
5104 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5106 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5107 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5108 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5109 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5110 if (fs_info
->quota_root
)
5111 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5112 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5114 update_global_block_rsv(fs_info
);
5117 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5119 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5121 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5122 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5123 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5124 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5125 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5126 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5127 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5128 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5131 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5132 struct btrfs_root
*root
)
5134 if (!trans
->block_rsv
)
5137 if (!trans
->bytes_reserved
)
5140 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5141 trans
->transid
, trans
->bytes_reserved
, 0);
5142 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5143 trans
->bytes_reserved
= 0;
5146 /* Can only return 0 or -ENOSPC */
5147 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5148 struct inode
*inode
)
5150 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5151 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5152 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5155 * We need to hold space in order to delete our orphan item once we've
5156 * added it, so this takes the reservation so we can release it later
5157 * when we are truly done with the orphan item.
5159 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5160 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5161 btrfs_ino(inode
), num_bytes
, 1);
5162 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5165 void btrfs_orphan_release_metadata(struct inode
*inode
)
5167 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5168 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5169 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5170 btrfs_ino(inode
), num_bytes
, 0);
5171 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5175 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5176 * root: the root of the parent directory
5177 * rsv: block reservation
5178 * items: the number of items that we need do reservation
5179 * qgroup_reserved: used to return the reserved size in qgroup
5181 * This function is used to reserve the space for snapshot/subvolume
5182 * creation and deletion. Those operations are different with the
5183 * common file/directory operations, they change two fs/file trees
5184 * and root tree, the number of items that the qgroup reserves is
5185 * different with the free space reservation. So we can not use
5186 * the space reseravtion mechanism in start_transaction().
5188 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5189 struct btrfs_block_rsv
*rsv
,
5191 u64
*qgroup_reserved
,
5192 bool use_global_rsv
)
5196 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5198 if (root
->fs_info
->quota_enabled
) {
5199 /* One for parent inode, two for dir entries */
5200 num_bytes
= 3 * root
->nodesize
;
5201 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
5208 *qgroup_reserved
= num_bytes
;
5210 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5211 rsv
->space_info
= __find_space_info(root
->fs_info
,
5212 BTRFS_BLOCK_GROUP_METADATA
);
5213 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5214 BTRFS_RESERVE_FLUSH_ALL
);
5216 if (ret
== -ENOSPC
&& use_global_rsv
)
5217 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5220 if (*qgroup_reserved
)
5221 btrfs_qgroup_free(root
, *qgroup_reserved
);
5227 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5228 struct btrfs_block_rsv
*rsv
,
5229 u64 qgroup_reserved
)
5231 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5232 if (qgroup_reserved
)
5233 btrfs_qgroup_free(root
, qgroup_reserved
);
5237 * drop_outstanding_extent - drop an outstanding extent
5238 * @inode: the inode we're dropping the extent for
5239 * @num_bytes: the number of bytes we're relaseing.
5241 * This is called when we are freeing up an outstanding extent, either called
5242 * after an error or after an extent is written. This will return the number of
5243 * reserved extents that need to be freed. This must be called with
5244 * BTRFS_I(inode)->lock held.
5246 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5248 unsigned drop_inode_space
= 0;
5249 unsigned dropped_extents
= 0;
5250 unsigned num_extents
= 0;
5252 num_extents
= (unsigned)div64_u64(num_bytes
+
5253 BTRFS_MAX_EXTENT_SIZE
- 1,
5254 BTRFS_MAX_EXTENT_SIZE
);
5255 ASSERT(num_extents
);
5256 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5257 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5259 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5260 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5261 &BTRFS_I(inode
)->runtime_flags
))
5262 drop_inode_space
= 1;
5265 * If we have more or the same amount of outsanding extents than we have
5266 * reserved then we need to leave the reserved extents count alone.
5268 if (BTRFS_I(inode
)->outstanding_extents
>=
5269 BTRFS_I(inode
)->reserved_extents
)
5270 return drop_inode_space
;
5272 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5273 BTRFS_I(inode
)->outstanding_extents
;
5274 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5275 return dropped_extents
+ drop_inode_space
;
5279 * calc_csum_metadata_size - return the amount of metada space that must be
5280 * reserved/free'd for the given bytes.
5281 * @inode: the inode we're manipulating
5282 * @num_bytes: the number of bytes in question
5283 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5285 * This adjusts the number of csum_bytes in the inode and then returns the
5286 * correct amount of metadata that must either be reserved or freed. We
5287 * calculate how many checksums we can fit into one leaf and then divide the
5288 * number of bytes that will need to be checksumed by this value to figure out
5289 * how many checksums will be required. If we are adding bytes then the number
5290 * may go up and we will return the number of additional bytes that must be
5291 * reserved. If it is going down we will return the number of bytes that must
5294 * This must be called with BTRFS_I(inode)->lock held.
5296 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5299 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5300 u64 old_csums
, num_csums
;
5302 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5303 BTRFS_I(inode
)->csum_bytes
== 0)
5306 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5308 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5310 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5311 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5313 /* No change, no need to reserve more */
5314 if (old_csums
== num_csums
)
5318 return btrfs_calc_trans_metadata_size(root
,
5319 num_csums
- old_csums
);
5321 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5324 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5326 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5327 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5330 unsigned nr_extents
= 0;
5331 int extra_reserve
= 0;
5332 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5334 bool delalloc_lock
= true;
5338 /* If we are a free space inode we need to not flush since we will be in
5339 * the middle of a transaction commit. We also don't need the delalloc
5340 * mutex since we won't race with anybody. We need this mostly to make
5341 * lockdep shut its filthy mouth.
5343 if (btrfs_is_free_space_inode(inode
)) {
5344 flush
= BTRFS_RESERVE_NO_FLUSH
;
5345 delalloc_lock
= false;
5348 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5349 btrfs_transaction_in_commit(root
->fs_info
))
5350 schedule_timeout(1);
5353 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5355 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5357 spin_lock(&BTRFS_I(inode
)->lock
);
5358 nr_extents
= (unsigned)div64_u64(num_bytes
+
5359 BTRFS_MAX_EXTENT_SIZE
- 1,
5360 BTRFS_MAX_EXTENT_SIZE
);
5361 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5364 if (BTRFS_I(inode
)->outstanding_extents
>
5365 BTRFS_I(inode
)->reserved_extents
)
5366 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5367 BTRFS_I(inode
)->reserved_extents
;
5370 * Add an item to reserve for updating the inode when we complete the
5373 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5374 &BTRFS_I(inode
)->runtime_flags
)) {
5379 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5380 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5381 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5382 spin_unlock(&BTRFS_I(inode
)->lock
);
5384 if (root
->fs_info
->quota_enabled
) {
5385 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
5386 nr_extents
* root
->nodesize
);
5391 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5392 if (unlikely(ret
)) {
5393 if (root
->fs_info
->quota_enabled
)
5394 btrfs_qgroup_free(root
, num_bytes
+
5395 nr_extents
* root
->nodesize
);
5399 spin_lock(&BTRFS_I(inode
)->lock
);
5400 if (extra_reserve
) {
5401 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5402 &BTRFS_I(inode
)->runtime_flags
);
5405 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5406 spin_unlock(&BTRFS_I(inode
)->lock
);
5409 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5412 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5413 btrfs_ino(inode
), to_reserve
, 1);
5414 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5419 spin_lock(&BTRFS_I(inode
)->lock
);
5420 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5422 * If the inodes csum_bytes is the same as the original
5423 * csum_bytes then we know we haven't raced with any free()ers
5424 * so we can just reduce our inodes csum bytes and carry on.
5426 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5427 calc_csum_metadata_size(inode
, num_bytes
, 0);
5429 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5433 * This is tricky, but first we need to figure out how much we
5434 * free'd from any free-ers that occured during this
5435 * reservation, so we reset ->csum_bytes to the csum_bytes
5436 * before we dropped our lock, and then call the free for the
5437 * number of bytes that were freed while we were trying our
5440 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5441 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5442 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5446 * Now we need to see how much we would have freed had we not
5447 * been making this reservation and our ->csum_bytes were not
5448 * artificially inflated.
5450 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5451 bytes
= csum_bytes
- orig_csum_bytes
;
5452 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5455 * Now reset ->csum_bytes to what it should be. If bytes is
5456 * more than to_free then we would have free'd more space had we
5457 * not had an artificially high ->csum_bytes, so we need to free
5458 * the remainder. If bytes is the same or less then we don't
5459 * need to do anything, the other free-ers did the correct
5462 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5463 if (bytes
> to_free
)
5464 to_free
= bytes
- to_free
;
5468 spin_unlock(&BTRFS_I(inode
)->lock
);
5470 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5473 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5474 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5475 btrfs_ino(inode
), to_free
, 0);
5478 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5483 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5484 * @inode: the inode to release the reservation for
5485 * @num_bytes: the number of bytes we're releasing
5487 * This will release the metadata reservation for an inode. This can be called
5488 * once we complete IO for a given set of bytes to release their metadata
5491 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5493 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5497 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5498 spin_lock(&BTRFS_I(inode
)->lock
);
5499 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5502 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5503 spin_unlock(&BTRFS_I(inode
)->lock
);
5505 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5507 if (btrfs_test_is_dummy_root(root
))
5510 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5511 btrfs_ino(inode
), to_free
, 0);
5512 if (root
->fs_info
->quota_enabled
) {
5513 btrfs_qgroup_free(root
, num_bytes
+
5514 dropped
* root
->nodesize
);
5517 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5522 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5523 * @inode: inode we're writing to
5524 * @num_bytes: the number of bytes we want to allocate
5526 * This will do the following things
5528 * o reserve space in the data space info for num_bytes
5529 * o reserve space in the metadata space info based on number of outstanding
5530 * extents and how much csums will be needed
5531 * o add to the inodes ->delalloc_bytes
5532 * o add it to the fs_info's delalloc inodes list.
5534 * This will return 0 for success and -ENOSPC if there is no space left.
5536 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5540 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5544 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5546 btrfs_free_reserved_data_space(inode
, num_bytes
);
5554 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5555 * @inode: inode we're releasing space for
5556 * @num_bytes: the number of bytes we want to free up
5558 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5559 * called in the case that we don't need the metadata AND data reservations
5560 * anymore. So if there is an error or we insert an inline extent.
5562 * This function will release the metadata space that was not used and will
5563 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5564 * list if there are no delalloc bytes left.
5566 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5568 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5569 btrfs_free_reserved_data_space(inode
, num_bytes
);
5572 static int update_block_group(struct btrfs_trans_handle
*trans
,
5573 struct btrfs_root
*root
, u64 bytenr
,
5574 u64 num_bytes
, int alloc
)
5576 struct btrfs_block_group_cache
*cache
= NULL
;
5577 struct btrfs_fs_info
*info
= root
->fs_info
;
5578 u64 total
= num_bytes
;
5583 /* block accounting for super block */
5584 spin_lock(&info
->delalloc_root_lock
);
5585 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5587 old_val
+= num_bytes
;
5589 old_val
-= num_bytes
;
5590 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5591 spin_unlock(&info
->delalloc_root_lock
);
5594 cache
= btrfs_lookup_block_group(info
, bytenr
);
5597 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5598 BTRFS_BLOCK_GROUP_RAID1
|
5599 BTRFS_BLOCK_GROUP_RAID10
))
5604 * If this block group has free space cache written out, we
5605 * need to make sure to load it if we are removing space. This
5606 * is because we need the unpinning stage to actually add the
5607 * space back to the block group, otherwise we will leak space.
5609 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5610 cache_block_group(cache
, 1);
5612 byte_in_group
= bytenr
- cache
->key
.objectid
;
5613 WARN_ON(byte_in_group
> cache
->key
.offset
);
5615 spin_lock(&cache
->space_info
->lock
);
5616 spin_lock(&cache
->lock
);
5618 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5619 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5620 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5622 old_val
= btrfs_block_group_used(&cache
->item
);
5623 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5625 old_val
+= num_bytes
;
5626 btrfs_set_block_group_used(&cache
->item
, old_val
);
5627 cache
->reserved
-= num_bytes
;
5628 cache
->space_info
->bytes_reserved
-= num_bytes
;
5629 cache
->space_info
->bytes_used
+= num_bytes
;
5630 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5631 spin_unlock(&cache
->lock
);
5632 spin_unlock(&cache
->space_info
->lock
);
5634 old_val
-= num_bytes
;
5635 btrfs_set_block_group_used(&cache
->item
, old_val
);
5636 cache
->pinned
+= num_bytes
;
5637 cache
->space_info
->bytes_pinned
+= num_bytes
;
5638 cache
->space_info
->bytes_used
-= num_bytes
;
5639 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5640 spin_unlock(&cache
->lock
);
5641 spin_unlock(&cache
->space_info
->lock
);
5643 set_extent_dirty(info
->pinned_extents
,
5644 bytenr
, bytenr
+ num_bytes
- 1,
5645 GFP_NOFS
| __GFP_NOFAIL
);
5647 * No longer have used bytes in this block group, queue
5651 spin_lock(&info
->unused_bgs_lock
);
5652 if (list_empty(&cache
->bg_list
)) {
5653 btrfs_get_block_group(cache
);
5654 list_add_tail(&cache
->bg_list
,
5657 spin_unlock(&info
->unused_bgs_lock
);
5661 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5662 if (list_empty(&cache
->dirty_list
)) {
5663 list_add_tail(&cache
->dirty_list
,
5664 &trans
->transaction
->dirty_bgs
);
5665 trans
->transaction
->num_dirty_bgs
++;
5666 btrfs_get_block_group(cache
);
5668 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5670 btrfs_put_block_group(cache
);
5672 bytenr
+= num_bytes
;
5677 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5679 struct btrfs_block_group_cache
*cache
;
5682 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5683 bytenr
= root
->fs_info
->first_logical_byte
;
5684 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5686 if (bytenr
< (u64
)-1)
5689 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5693 bytenr
= cache
->key
.objectid
;
5694 btrfs_put_block_group(cache
);
5699 static int pin_down_extent(struct btrfs_root
*root
,
5700 struct btrfs_block_group_cache
*cache
,
5701 u64 bytenr
, u64 num_bytes
, int reserved
)
5703 spin_lock(&cache
->space_info
->lock
);
5704 spin_lock(&cache
->lock
);
5705 cache
->pinned
+= num_bytes
;
5706 cache
->space_info
->bytes_pinned
+= num_bytes
;
5708 cache
->reserved
-= num_bytes
;
5709 cache
->space_info
->bytes_reserved
-= num_bytes
;
5711 spin_unlock(&cache
->lock
);
5712 spin_unlock(&cache
->space_info
->lock
);
5714 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5715 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5717 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5722 * this function must be called within transaction
5724 int btrfs_pin_extent(struct btrfs_root
*root
,
5725 u64 bytenr
, u64 num_bytes
, int reserved
)
5727 struct btrfs_block_group_cache
*cache
;
5729 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5730 BUG_ON(!cache
); /* Logic error */
5732 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5734 btrfs_put_block_group(cache
);
5739 * this function must be called within transaction
5741 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5742 u64 bytenr
, u64 num_bytes
)
5744 struct btrfs_block_group_cache
*cache
;
5747 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5752 * pull in the free space cache (if any) so that our pin
5753 * removes the free space from the cache. We have load_only set
5754 * to one because the slow code to read in the free extents does check
5755 * the pinned extents.
5757 cache_block_group(cache
, 1);
5759 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5761 /* remove us from the free space cache (if we're there at all) */
5762 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5763 btrfs_put_block_group(cache
);
5767 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5770 struct btrfs_block_group_cache
*block_group
;
5771 struct btrfs_caching_control
*caching_ctl
;
5773 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5777 cache_block_group(block_group
, 0);
5778 caching_ctl
= get_caching_control(block_group
);
5782 BUG_ON(!block_group_cache_done(block_group
));
5783 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5785 mutex_lock(&caching_ctl
->mutex
);
5787 if (start
>= caching_ctl
->progress
) {
5788 ret
= add_excluded_extent(root
, start
, num_bytes
);
5789 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5790 ret
= btrfs_remove_free_space(block_group
,
5793 num_bytes
= caching_ctl
->progress
- start
;
5794 ret
= btrfs_remove_free_space(block_group
,
5799 num_bytes
= (start
+ num_bytes
) -
5800 caching_ctl
->progress
;
5801 start
= caching_ctl
->progress
;
5802 ret
= add_excluded_extent(root
, start
, num_bytes
);
5805 mutex_unlock(&caching_ctl
->mutex
);
5806 put_caching_control(caching_ctl
);
5808 btrfs_put_block_group(block_group
);
5812 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5813 struct extent_buffer
*eb
)
5815 struct btrfs_file_extent_item
*item
;
5816 struct btrfs_key key
;
5820 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5823 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5824 btrfs_item_key_to_cpu(eb
, &key
, i
);
5825 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5827 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5828 found_type
= btrfs_file_extent_type(eb
, item
);
5829 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5831 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5833 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5834 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5835 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5842 * btrfs_update_reserved_bytes - update the block_group and space info counters
5843 * @cache: The cache we are manipulating
5844 * @num_bytes: The number of bytes in question
5845 * @reserve: One of the reservation enums
5846 * @delalloc: The blocks are allocated for the delalloc write
5848 * This is called by the allocator when it reserves space, or by somebody who is
5849 * freeing space that was never actually used on disk. For example if you
5850 * reserve some space for a new leaf in transaction A and before transaction A
5851 * commits you free that leaf, you call this with reserve set to 0 in order to
5852 * clear the reservation.
5854 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5855 * ENOSPC accounting. For data we handle the reservation through clearing the
5856 * delalloc bits in the io_tree. We have to do this since we could end up
5857 * allocating less disk space for the amount of data we have reserved in the
5858 * case of compression.
5860 * If this is a reservation and the block group has become read only we cannot
5861 * make the reservation and return -EAGAIN, otherwise this function always
5864 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5865 u64 num_bytes
, int reserve
, int delalloc
)
5867 struct btrfs_space_info
*space_info
= cache
->space_info
;
5870 spin_lock(&space_info
->lock
);
5871 spin_lock(&cache
->lock
);
5872 if (reserve
!= RESERVE_FREE
) {
5876 cache
->reserved
+= num_bytes
;
5877 space_info
->bytes_reserved
+= num_bytes
;
5878 if (reserve
== RESERVE_ALLOC
) {
5879 trace_btrfs_space_reservation(cache
->fs_info
,
5880 "space_info", space_info
->flags
,
5882 space_info
->bytes_may_use
-= num_bytes
;
5886 cache
->delalloc_bytes
+= num_bytes
;
5890 space_info
->bytes_readonly
+= num_bytes
;
5891 cache
->reserved
-= num_bytes
;
5892 space_info
->bytes_reserved
-= num_bytes
;
5895 cache
->delalloc_bytes
-= num_bytes
;
5897 spin_unlock(&cache
->lock
);
5898 spin_unlock(&space_info
->lock
);
5902 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5903 struct btrfs_root
*root
)
5905 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5906 struct btrfs_caching_control
*next
;
5907 struct btrfs_caching_control
*caching_ctl
;
5908 struct btrfs_block_group_cache
*cache
;
5910 down_write(&fs_info
->commit_root_sem
);
5912 list_for_each_entry_safe(caching_ctl
, next
,
5913 &fs_info
->caching_block_groups
, list
) {
5914 cache
= caching_ctl
->block_group
;
5915 if (block_group_cache_done(cache
)) {
5916 cache
->last_byte_to_unpin
= (u64
)-1;
5917 list_del_init(&caching_ctl
->list
);
5918 put_caching_control(caching_ctl
);
5920 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5924 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5925 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5927 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5929 up_write(&fs_info
->commit_root_sem
);
5931 update_global_block_rsv(fs_info
);
5934 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
5935 const bool return_free_space
)
5937 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5938 struct btrfs_block_group_cache
*cache
= NULL
;
5939 struct btrfs_space_info
*space_info
;
5940 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5944 while (start
<= end
) {
5947 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5949 btrfs_put_block_group(cache
);
5950 cache
= btrfs_lookup_block_group(fs_info
, start
);
5951 BUG_ON(!cache
); /* Logic error */
5954 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5955 len
= min(len
, end
+ 1 - start
);
5957 if (start
< cache
->last_byte_to_unpin
) {
5958 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5959 if (return_free_space
)
5960 btrfs_add_free_space(cache
, start
, len
);
5964 space_info
= cache
->space_info
;
5966 spin_lock(&space_info
->lock
);
5967 spin_lock(&cache
->lock
);
5968 cache
->pinned
-= len
;
5969 space_info
->bytes_pinned
-= len
;
5970 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
5972 space_info
->bytes_readonly
+= len
;
5975 spin_unlock(&cache
->lock
);
5976 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5977 spin_lock(&global_rsv
->lock
);
5978 if (!global_rsv
->full
) {
5979 len
= min(len
, global_rsv
->size
-
5980 global_rsv
->reserved
);
5981 global_rsv
->reserved
+= len
;
5982 space_info
->bytes_may_use
+= len
;
5983 if (global_rsv
->reserved
>= global_rsv
->size
)
5984 global_rsv
->full
= 1;
5986 spin_unlock(&global_rsv
->lock
);
5988 spin_unlock(&space_info
->lock
);
5992 btrfs_put_block_group(cache
);
5996 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5997 struct btrfs_root
*root
)
5999 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6000 struct extent_io_tree
*unpin
;
6008 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6009 unpin
= &fs_info
->freed_extents
[1];
6011 unpin
= &fs_info
->freed_extents
[0];
6014 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6015 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6016 EXTENT_DIRTY
, NULL
);
6018 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6022 if (btrfs_test_opt(root
, DISCARD
))
6023 ret
= btrfs_discard_extent(root
, start
,
6024 end
+ 1 - start
, NULL
);
6026 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
6027 unpin_extent_range(root
, start
, end
, true);
6028 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6035 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6036 u64 owner
, u64 root_objectid
)
6038 struct btrfs_space_info
*space_info
;
6041 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6042 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6043 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6045 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6047 flags
= BTRFS_BLOCK_GROUP_DATA
;
6050 space_info
= __find_space_info(fs_info
, flags
);
6051 BUG_ON(!space_info
); /* Logic bug */
6052 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6056 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6057 struct btrfs_root
*root
,
6058 u64 bytenr
, u64 num_bytes
, u64 parent
,
6059 u64 root_objectid
, u64 owner_objectid
,
6060 u64 owner_offset
, int refs_to_drop
,
6061 struct btrfs_delayed_extent_op
*extent_op
,
6064 struct btrfs_key key
;
6065 struct btrfs_path
*path
;
6066 struct btrfs_fs_info
*info
= root
->fs_info
;
6067 struct btrfs_root
*extent_root
= info
->extent_root
;
6068 struct extent_buffer
*leaf
;
6069 struct btrfs_extent_item
*ei
;
6070 struct btrfs_extent_inline_ref
*iref
;
6073 int extent_slot
= 0;
6074 int found_extent
= 0;
6079 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_SUB_EXCL
;
6080 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6083 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
6086 path
= btrfs_alloc_path();
6091 path
->leave_spinning
= 1;
6093 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6094 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6097 skinny_metadata
= 0;
6099 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6100 bytenr
, num_bytes
, parent
,
6101 root_objectid
, owner_objectid
,
6104 extent_slot
= path
->slots
[0];
6105 while (extent_slot
>= 0) {
6106 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6108 if (key
.objectid
!= bytenr
)
6110 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6111 key
.offset
== num_bytes
) {
6115 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6116 key
.offset
== owner_objectid
) {
6120 if (path
->slots
[0] - extent_slot
> 5)
6124 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6125 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6126 if (found_extent
&& item_size
< sizeof(*ei
))
6129 if (!found_extent
) {
6131 ret
= remove_extent_backref(trans
, extent_root
, path
,
6133 is_data
, &last_ref
);
6135 btrfs_abort_transaction(trans
, extent_root
, ret
);
6138 btrfs_release_path(path
);
6139 path
->leave_spinning
= 1;
6141 key
.objectid
= bytenr
;
6142 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6143 key
.offset
= num_bytes
;
6145 if (!is_data
&& skinny_metadata
) {
6146 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6147 key
.offset
= owner_objectid
;
6150 ret
= btrfs_search_slot(trans
, extent_root
,
6152 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6154 * Couldn't find our skinny metadata item,
6155 * see if we have ye olde extent item.
6158 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6160 if (key
.objectid
== bytenr
&&
6161 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6162 key
.offset
== num_bytes
)
6166 if (ret
> 0 && skinny_metadata
) {
6167 skinny_metadata
= false;
6168 key
.objectid
= bytenr
;
6169 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6170 key
.offset
= num_bytes
;
6171 btrfs_release_path(path
);
6172 ret
= btrfs_search_slot(trans
, extent_root
,
6177 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6180 btrfs_print_leaf(extent_root
,
6184 btrfs_abort_transaction(trans
, extent_root
, ret
);
6187 extent_slot
= path
->slots
[0];
6189 } else if (WARN_ON(ret
== -ENOENT
)) {
6190 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6192 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6193 bytenr
, parent
, root_objectid
, owner_objectid
,
6195 btrfs_abort_transaction(trans
, extent_root
, ret
);
6198 btrfs_abort_transaction(trans
, extent_root
, ret
);
6202 leaf
= path
->nodes
[0];
6203 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6204 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6205 if (item_size
< sizeof(*ei
)) {
6206 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6207 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6210 btrfs_abort_transaction(trans
, extent_root
, ret
);
6214 btrfs_release_path(path
);
6215 path
->leave_spinning
= 1;
6217 key
.objectid
= bytenr
;
6218 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6219 key
.offset
= num_bytes
;
6221 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6224 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6226 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6229 btrfs_abort_transaction(trans
, extent_root
, ret
);
6233 extent_slot
= path
->slots
[0];
6234 leaf
= path
->nodes
[0];
6235 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6238 BUG_ON(item_size
< sizeof(*ei
));
6239 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6240 struct btrfs_extent_item
);
6241 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6242 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6243 struct btrfs_tree_block_info
*bi
;
6244 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6245 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6246 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6249 refs
= btrfs_extent_refs(leaf
, ei
);
6250 if (refs
< refs_to_drop
) {
6251 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6252 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6254 btrfs_abort_transaction(trans
, extent_root
, ret
);
6257 refs
-= refs_to_drop
;
6260 type
= BTRFS_QGROUP_OPER_SUB_SHARED
;
6262 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6264 * In the case of inline back ref, reference count will
6265 * be updated by remove_extent_backref
6268 BUG_ON(!found_extent
);
6270 btrfs_set_extent_refs(leaf
, ei
, refs
);
6271 btrfs_mark_buffer_dirty(leaf
);
6274 ret
= remove_extent_backref(trans
, extent_root
, path
,
6276 is_data
, &last_ref
);
6278 btrfs_abort_transaction(trans
, extent_root
, ret
);
6282 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6286 BUG_ON(is_data
&& refs_to_drop
!=
6287 extent_data_ref_count(root
, path
, iref
));
6289 BUG_ON(path
->slots
[0] != extent_slot
);
6291 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6292 path
->slots
[0] = extent_slot
;
6298 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6301 btrfs_abort_transaction(trans
, extent_root
, ret
);
6304 btrfs_release_path(path
);
6307 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6309 btrfs_abort_transaction(trans
, extent_root
, ret
);
6314 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6316 btrfs_abort_transaction(trans
, extent_root
, ret
);
6320 btrfs_release_path(path
);
6322 /* Deal with the quota accounting */
6323 if (!ret
&& last_ref
&& !no_quota
) {
6326 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
6327 type
== BTRFS_QGROUP_OPER_SUB_SHARED
)
6330 ret
= btrfs_qgroup_record_ref(trans
, info
, root_objectid
,
6331 bytenr
, num_bytes
, type
,
6335 btrfs_free_path(path
);
6340 * when we free an block, it is possible (and likely) that we free the last
6341 * delayed ref for that extent as well. This searches the delayed ref tree for
6342 * a given extent, and if there are no other delayed refs to be processed, it
6343 * removes it from the tree.
6345 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6346 struct btrfs_root
*root
, u64 bytenr
)
6348 struct btrfs_delayed_ref_head
*head
;
6349 struct btrfs_delayed_ref_root
*delayed_refs
;
6352 delayed_refs
= &trans
->transaction
->delayed_refs
;
6353 spin_lock(&delayed_refs
->lock
);
6354 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6356 goto out_delayed_unlock
;
6358 spin_lock(&head
->lock
);
6359 if (rb_first(&head
->ref_root
))
6362 if (head
->extent_op
) {
6363 if (!head
->must_insert_reserved
)
6365 btrfs_free_delayed_extent_op(head
->extent_op
);
6366 head
->extent_op
= NULL
;
6370 * waiting for the lock here would deadlock. If someone else has it
6371 * locked they are already in the process of dropping it anyway
6373 if (!mutex_trylock(&head
->mutex
))
6377 * at this point we have a head with no other entries. Go
6378 * ahead and process it.
6380 head
->node
.in_tree
= 0;
6381 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6383 atomic_dec(&delayed_refs
->num_entries
);
6386 * we don't take a ref on the node because we're removing it from the
6387 * tree, so we just steal the ref the tree was holding.
6389 delayed_refs
->num_heads
--;
6390 if (head
->processing
== 0)
6391 delayed_refs
->num_heads_ready
--;
6392 head
->processing
= 0;
6393 spin_unlock(&head
->lock
);
6394 spin_unlock(&delayed_refs
->lock
);
6396 BUG_ON(head
->extent_op
);
6397 if (head
->must_insert_reserved
)
6400 mutex_unlock(&head
->mutex
);
6401 btrfs_put_delayed_ref(&head
->node
);
6404 spin_unlock(&head
->lock
);
6407 spin_unlock(&delayed_refs
->lock
);
6411 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6412 struct btrfs_root
*root
,
6413 struct extent_buffer
*buf
,
6414 u64 parent
, int last_ref
)
6419 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6420 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6421 buf
->start
, buf
->len
,
6422 parent
, root
->root_key
.objectid
,
6423 btrfs_header_level(buf
),
6424 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6425 BUG_ON(ret
); /* -ENOMEM */
6431 if (btrfs_header_generation(buf
) == trans
->transid
) {
6432 struct btrfs_block_group_cache
*cache
;
6434 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6435 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6440 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6442 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6443 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6444 btrfs_put_block_group(cache
);
6448 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6450 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6451 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6452 btrfs_put_block_group(cache
);
6453 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6458 add_pinned_bytes(root
->fs_info
, buf
->len
,
6459 btrfs_header_level(buf
),
6460 root
->root_key
.objectid
);
6463 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6466 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6469 /* Can return -ENOMEM */
6470 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6471 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6472 u64 owner
, u64 offset
, int no_quota
)
6475 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6477 if (btrfs_test_is_dummy_root(root
))
6480 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6483 * tree log blocks never actually go into the extent allocation
6484 * tree, just update pinning info and exit early.
6486 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6487 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6488 /* unlocks the pinned mutex */
6489 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6491 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6492 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6494 parent
, root_objectid
, (int)owner
,
6495 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6497 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6499 parent
, root_objectid
, owner
,
6500 offset
, BTRFS_DROP_DELAYED_REF
,
6507 * when we wait for progress in the block group caching, its because
6508 * our allocation attempt failed at least once. So, we must sleep
6509 * and let some progress happen before we try again.
6511 * This function will sleep at least once waiting for new free space to
6512 * show up, and then it will check the block group free space numbers
6513 * for our min num_bytes. Another option is to have it go ahead
6514 * and look in the rbtree for a free extent of a given size, but this
6517 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6518 * any of the information in this block group.
6520 static noinline
void
6521 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6524 struct btrfs_caching_control
*caching_ctl
;
6526 caching_ctl
= get_caching_control(cache
);
6530 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6531 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6533 put_caching_control(caching_ctl
);
6537 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6539 struct btrfs_caching_control
*caching_ctl
;
6542 caching_ctl
= get_caching_control(cache
);
6544 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6546 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6547 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6549 put_caching_control(caching_ctl
);
6553 int __get_raid_index(u64 flags
)
6555 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6556 return BTRFS_RAID_RAID10
;
6557 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6558 return BTRFS_RAID_RAID1
;
6559 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6560 return BTRFS_RAID_DUP
;
6561 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6562 return BTRFS_RAID_RAID0
;
6563 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6564 return BTRFS_RAID_RAID5
;
6565 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6566 return BTRFS_RAID_RAID6
;
6568 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6571 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6573 return __get_raid_index(cache
->flags
);
6576 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6577 [BTRFS_RAID_RAID10
] = "raid10",
6578 [BTRFS_RAID_RAID1
] = "raid1",
6579 [BTRFS_RAID_DUP
] = "dup",
6580 [BTRFS_RAID_RAID0
] = "raid0",
6581 [BTRFS_RAID_SINGLE
] = "single",
6582 [BTRFS_RAID_RAID5
] = "raid5",
6583 [BTRFS_RAID_RAID6
] = "raid6",
6586 static const char *get_raid_name(enum btrfs_raid_types type
)
6588 if (type
>= BTRFS_NR_RAID_TYPES
)
6591 return btrfs_raid_type_names
[type
];
6594 enum btrfs_loop_type
{
6595 LOOP_CACHING_NOWAIT
= 0,
6596 LOOP_CACHING_WAIT
= 1,
6597 LOOP_ALLOC_CHUNK
= 2,
6598 LOOP_NO_EMPTY_SIZE
= 3,
6602 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6606 down_read(&cache
->data_rwsem
);
6610 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6613 btrfs_get_block_group(cache
);
6615 down_read(&cache
->data_rwsem
);
6618 static struct btrfs_block_group_cache
*
6619 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6620 struct btrfs_free_cluster
*cluster
,
6623 struct btrfs_block_group_cache
*used_bg
;
6624 bool locked
= false;
6626 spin_lock(&cluster
->refill_lock
);
6628 if (used_bg
== cluster
->block_group
)
6631 up_read(&used_bg
->data_rwsem
);
6632 btrfs_put_block_group(used_bg
);
6635 used_bg
= cluster
->block_group
;
6639 if (used_bg
== block_group
)
6642 btrfs_get_block_group(used_bg
);
6647 if (down_read_trylock(&used_bg
->data_rwsem
))
6650 spin_unlock(&cluster
->refill_lock
);
6651 down_read(&used_bg
->data_rwsem
);
6657 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6661 up_read(&cache
->data_rwsem
);
6662 btrfs_put_block_group(cache
);
6666 * walks the btree of allocated extents and find a hole of a given size.
6667 * The key ins is changed to record the hole:
6668 * ins->objectid == start position
6669 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6670 * ins->offset == the size of the hole.
6671 * Any available blocks before search_start are skipped.
6673 * If there is no suitable free space, we will record the max size of
6674 * the free space extent currently.
6676 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6677 u64 num_bytes
, u64 empty_size
,
6678 u64 hint_byte
, struct btrfs_key
*ins
,
6679 u64 flags
, int delalloc
)
6682 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6683 struct btrfs_free_cluster
*last_ptr
= NULL
;
6684 struct btrfs_block_group_cache
*block_group
= NULL
;
6685 u64 search_start
= 0;
6686 u64 max_extent_size
= 0;
6687 int empty_cluster
= 2 * 1024 * 1024;
6688 struct btrfs_space_info
*space_info
;
6690 int index
= __get_raid_index(flags
);
6691 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6692 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6693 bool failed_cluster_refill
= false;
6694 bool failed_alloc
= false;
6695 bool use_cluster
= true;
6696 bool have_caching_bg
= false;
6698 WARN_ON(num_bytes
< root
->sectorsize
);
6699 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6703 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6705 space_info
= __find_space_info(root
->fs_info
, flags
);
6707 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6712 * If the space info is for both data and metadata it means we have a
6713 * small filesystem and we can't use the clustering stuff.
6715 if (btrfs_mixed_space_info(space_info
))
6716 use_cluster
= false;
6718 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6719 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6720 if (!btrfs_test_opt(root
, SSD
))
6721 empty_cluster
= 64 * 1024;
6724 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6725 btrfs_test_opt(root
, SSD
)) {
6726 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6730 spin_lock(&last_ptr
->lock
);
6731 if (last_ptr
->block_group
)
6732 hint_byte
= last_ptr
->window_start
;
6733 spin_unlock(&last_ptr
->lock
);
6736 search_start
= max(search_start
, first_logical_byte(root
, 0));
6737 search_start
= max(search_start
, hint_byte
);
6742 if (search_start
== hint_byte
) {
6743 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6746 * we don't want to use the block group if it doesn't match our
6747 * allocation bits, or if its not cached.
6749 * However if we are re-searching with an ideal block group
6750 * picked out then we don't care that the block group is cached.
6752 if (block_group
&& block_group_bits(block_group
, flags
) &&
6753 block_group
->cached
!= BTRFS_CACHE_NO
) {
6754 down_read(&space_info
->groups_sem
);
6755 if (list_empty(&block_group
->list
) ||
6758 * someone is removing this block group,
6759 * we can't jump into the have_block_group
6760 * target because our list pointers are not
6763 btrfs_put_block_group(block_group
);
6764 up_read(&space_info
->groups_sem
);
6766 index
= get_block_group_index(block_group
);
6767 btrfs_lock_block_group(block_group
, delalloc
);
6768 goto have_block_group
;
6770 } else if (block_group
) {
6771 btrfs_put_block_group(block_group
);
6775 have_caching_bg
= false;
6776 down_read(&space_info
->groups_sem
);
6777 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6782 btrfs_grab_block_group(block_group
, delalloc
);
6783 search_start
= block_group
->key
.objectid
;
6786 * this can happen if we end up cycling through all the
6787 * raid types, but we want to make sure we only allocate
6788 * for the proper type.
6790 if (!block_group_bits(block_group
, flags
)) {
6791 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6792 BTRFS_BLOCK_GROUP_RAID1
|
6793 BTRFS_BLOCK_GROUP_RAID5
|
6794 BTRFS_BLOCK_GROUP_RAID6
|
6795 BTRFS_BLOCK_GROUP_RAID10
;
6798 * if they asked for extra copies and this block group
6799 * doesn't provide them, bail. This does allow us to
6800 * fill raid0 from raid1.
6802 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6807 cached
= block_group_cache_done(block_group
);
6808 if (unlikely(!cached
)) {
6809 ret
= cache_block_group(block_group
, 0);
6814 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6816 if (unlikely(block_group
->ro
))
6820 * Ok we want to try and use the cluster allocator, so
6824 struct btrfs_block_group_cache
*used_block_group
;
6825 unsigned long aligned_cluster
;
6827 * the refill lock keeps out other
6828 * people trying to start a new cluster
6830 used_block_group
= btrfs_lock_cluster(block_group
,
6833 if (!used_block_group
)
6834 goto refill_cluster
;
6836 if (used_block_group
!= block_group
&&
6837 (used_block_group
->ro
||
6838 !block_group_bits(used_block_group
, flags
)))
6839 goto release_cluster
;
6841 offset
= btrfs_alloc_from_cluster(used_block_group
,
6844 used_block_group
->key
.objectid
,
6847 /* we have a block, we're done */
6848 spin_unlock(&last_ptr
->refill_lock
);
6849 trace_btrfs_reserve_extent_cluster(root
,
6851 search_start
, num_bytes
);
6852 if (used_block_group
!= block_group
) {
6853 btrfs_release_block_group(block_group
,
6855 block_group
= used_block_group
;
6860 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6862 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6863 * set up a new clusters, so lets just skip it
6864 * and let the allocator find whatever block
6865 * it can find. If we reach this point, we
6866 * will have tried the cluster allocator
6867 * plenty of times and not have found
6868 * anything, so we are likely way too
6869 * fragmented for the clustering stuff to find
6872 * However, if the cluster is taken from the
6873 * current block group, release the cluster
6874 * first, so that we stand a better chance of
6875 * succeeding in the unclustered
6877 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6878 used_block_group
!= block_group
) {
6879 spin_unlock(&last_ptr
->refill_lock
);
6880 btrfs_release_block_group(used_block_group
,
6882 goto unclustered_alloc
;
6886 * this cluster didn't work out, free it and
6889 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6891 if (used_block_group
!= block_group
)
6892 btrfs_release_block_group(used_block_group
,
6895 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6896 spin_unlock(&last_ptr
->refill_lock
);
6897 goto unclustered_alloc
;
6900 aligned_cluster
= max_t(unsigned long,
6901 empty_cluster
+ empty_size
,
6902 block_group
->full_stripe_len
);
6904 /* allocate a cluster in this block group */
6905 ret
= btrfs_find_space_cluster(root
, block_group
,
6906 last_ptr
, search_start
,
6911 * now pull our allocation out of this
6914 offset
= btrfs_alloc_from_cluster(block_group
,
6920 /* we found one, proceed */
6921 spin_unlock(&last_ptr
->refill_lock
);
6922 trace_btrfs_reserve_extent_cluster(root
,
6923 block_group
, search_start
,
6927 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6928 && !failed_cluster_refill
) {
6929 spin_unlock(&last_ptr
->refill_lock
);
6931 failed_cluster_refill
= true;
6932 wait_block_group_cache_progress(block_group
,
6933 num_bytes
+ empty_cluster
+ empty_size
);
6934 goto have_block_group
;
6938 * at this point we either didn't find a cluster
6939 * or we weren't able to allocate a block from our
6940 * cluster. Free the cluster we've been trying
6941 * to use, and go to the next block group
6943 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6944 spin_unlock(&last_ptr
->refill_lock
);
6949 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6951 block_group
->free_space_ctl
->free_space
<
6952 num_bytes
+ empty_cluster
+ empty_size
) {
6953 if (block_group
->free_space_ctl
->free_space
>
6956 block_group
->free_space_ctl
->free_space
;
6957 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6960 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6962 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6963 num_bytes
, empty_size
,
6966 * If we didn't find a chunk, and we haven't failed on this
6967 * block group before, and this block group is in the middle of
6968 * caching and we are ok with waiting, then go ahead and wait
6969 * for progress to be made, and set failed_alloc to true.
6971 * If failed_alloc is true then we've already waited on this
6972 * block group once and should move on to the next block group.
6974 if (!offset
&& !failed_alloc
&& !cached
&&
6975 loop
> LOOP_CACHING_NOWAIT
) {
6976 wait_block_group_cache_progress(block_group
,
6977 num_bytes
+ empty_size
);
6978 failed_alloc
= true;
6979 goto have_block_group
;
6980 } else if (!offset
) {
6982 have_caching_bg
= true;
6986 search_start
= ALIGN(offset
, root
->stripesize
);
6988 /* move on to the next group */
6989 if (search_start
+ num_bytes
>
6990 block_group
->key
.objectid
+ block_group
->key
.offset
) {
6991 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6995 if (offset
< search_start
)
6996 btrfs_add_free_space(block_group
, offset
,
6997 search_start
- offset
);
6998 BUG_ON(offset
> search_start
);
7000 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7001 alloc_type
, delalloc
);
7002 if (ret
== -EAGAIN
) {
7003 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7007 /* we are all good, lets return */
7008 ins
->objectid
= search_start
;
7009 ins
->offset
= num_bytes
;
7011 trace_btrfs_reserve_extent(orig_root
, block_group
,
7012 search_start
, num_bytes
);
7013 btrfs_release_block_group(block_group
, delalloc
);
7016 failed_cluster_refill
= false;
7017 failed_alloc
= false;
7018 BUG_ON(index
!= get_block_group_index(block_group
));
7019 btrfs_release_block_group(block_group
, delalloc
);
7021 up_read(&space_info
->groups_sem
);
7023 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7026 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7030 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7031 * caching kthreads as we move along
7032 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7033 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7034 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7037 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7040 if (loop
== LOOP_ALLOC_CHUNK
) {
7041 struct btrfs_trans_handle
*trans
;
7044 trans
= current
->journal_info
;
7048 trans
= btrfs_join_transaction(root
);
7050 if (IS_ERR(trans
)) {
7051 ret
= PTR_ERR(trans
);
7055 ret
= do_chunk_alloc(trans
, root
, flags
,
7058 * Do not bail out on ENOSPC since we
7059 * can do more things.
7061 if (ret
< 0 && ret
!= -ENOSPC
)
7062 btrfs_abort_transaction(trans
,
7067 btrfs_end_transaction(trans
, root
);
7072 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7078 } else if (!ins
->objectid
) {
7080 } else if (ins
->objectid
) {
7085 ins
->offset
= max_extent_size
;
7089 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7090 int dump_block_groups
)
7092 struct btrfs_block_group_cache
*cache
;
7095 spin_lock(&info
->lock
);
7096 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7098 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7099 info
->bytes_reserved
- info
->bytes_readonly
,
7100 (info
->full
) ? "" : "not ");
7101 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7102 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7103 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7104 info
->bytes_reserved
, info
->bytes_may_use
,
7105 info
->bytes_readonly
);
7106 spin_unlock(&info
->lock
);
7108 if (!dump_block_groups
)
7111 down_read(&info
->groups_sem
);
7113 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7114 spin_lock(&cache
->lock
);
7115 printk(KERN_INFO
"BTRFS: "
7116 "block group %llu has %llu bytes, "
7117 "%llu used %llu pinned %llu reserved %s\n",
7118 cache
->key
.objectid
, cache
->key
.offset
,
7119 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7120 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7121 btrfs_dump_free_space(cache
, bytes
);
7122 spin_unlock(&cache
->lock
);
7124 if (++index
< BTRFS_NR_RAID_TYPES
)
7126 up_read(&info
->groups_sem
);
7129 int btrfs_reserve_extent(struct btrfs_root
*root
,
7130 u64 num_bytes
, u64 min_alloc_size
,
7131 u64 empty_size
, u64 hint_byte
,
7132 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7134 bool final_tried
= false;
7138 flags
= btrfs_get_alloc_profile(root
, is_data
);
7140 WARN_ON(num_bytes
< root
->sectorsize
);
7141 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7144 if (ret
== -ENOSPC
) {
7145 if (!final_tried
&& ins
->offset
) {
7146 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7147 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7148 num_bytes
= max(num_bytes
, min_alloc_size
);
7149 if (num_bytes
== min_alloc_size
)
7152 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7153 struct btrfs_space_info
*sinfo
;
7155 sinfo
= __find_space_info(root
->fs_info
, flags
);
7156 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7159 dump_space_info(sinfo
, num_bytes
, 1);
7166 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7168 int pin
, int delalloc
)
7170 struct btrfs_block_group_cache
*cache
;
7173 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7175 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7181 pin_down_extent(root
, cache
, start
, len
, 1);
7183 if (btrfs_test_opt(root
, DISCARD
))
7184 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7185 btrfs_add_free_space(cache
, start
, len
);
7186 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7188 btrfs_put_block_group(cache
);
7190 trace_btrfs_reserved_extent_free(root
, start
, len
);
7195 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7196 u64 start
, u64 len
, int delalloc
)
7198 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7201 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7204 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7207 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7208 struct btrfs_root
*root
,
7209 u64 parent
, u64 root_objectid
,
7210 u64 flags
, u64 owner
, u64 offset
,
7211 struct btrfs_key
*ins
, int ref_mod
)
7214 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7215 struct btrfs_extent_item
*extent_item
;
7216 struct btrfs_extent_inline_ref
*iref
;
7217 struct btrfs_path
*path
;
7218 struct extent_buffer
*leaf
;
7223 type
= BTRFS_SHARED_DATA_REF_KEY
;
7225 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7227 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7229 path
= btrfs_alloc_path();
7233 path
->leave_spinning
= 1;
7234 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7237 btrfs_free_path(path
);
7241 leaf
= path
->nodes
[0];
7242 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7243 struct btrfs_extent_item
);
7244 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7245 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7246 btrfs_set_extent_flags(leaf
, extent_item
,
7247 flags
| BTRFS_EXTENT_FLAG_DATA
);
7249 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7250 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7252 struct btrfs_shared_data_ref
*ref
;
7253 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7254 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7255 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7257 struct btrfs_extent_data_ref
*ref
;
7258 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7259 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7260 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7261 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7262 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7265 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7266 btrfs_free_path(path
);
7268 /* Always set parent to 0 here since its exclusive anyway. */
7269 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7270 ins
->objectid
, ins
->offset
,
7271 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7275 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7276 if (ret
) { /* -ENOENT, logic error */
7277 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7278 ins
->objectid
, ins
->offset
);
7281 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7285 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7286 struct btrfs_root
*root
,
7287 u64 parent
, u64 root_objectid
,
7288 u64 flags
, struct btrfs_disk_key
*key
,
7289 int level
, struct btrfs_key
*ins
,
7293 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7294 struct btrfs_extent_item
*extent_item
;
7295 struct btrfs_tree_block_info
*block_info
;
7296 struct btrfs_extent_inline_ref
*iref
;
7297 struct btrfs_path
*path
;
7298 struct extent_buffer
*leaf
;
7299 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7300 u64 num_bytes
= ins
->offset
;
7301 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7304 if (!skinny_metadata
)
7305 size
+= sizeof(*block_info
);
7307 path
= btrfs_alloc_path();
7309 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7314 path
->leave_spinning
= 1;
7315 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7318 btrfs_free_path(path
);
7319 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7324 leaf
= path
->nodes
[0];
7325 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7326 struct btrfs_extent_item
);
7327 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7328 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7329 btrfs_set_extent_flags(leaf
, extent_item
,
7330 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7332 if (skinny_metadata
) {
7333 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7334 num_bytes
= root
->nodesize
;
7336 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7337 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7338 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7339 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7343 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7344 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7345 BTRFS_SHARED_BLOCK_REF_KEY
);
7346 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7348 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7349 BTRFS_TREE_BLOCK_REF_KEY
);
7350 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7353 btrfs_mark_buffer_dirty(leaf
);
7354 btrfs_free_path(path
);
7357 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7358 ins
->objectid
, num_bytes
,
7359 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7364 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7366 if (ret
) { /* -ENOENT, logic error */
7367 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7368 ins
->objectid
, ins
->offset
);
7372 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7376 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7377 struct btrfs_root
*root
,
7378 u64 root_objectid
, u64 owner
,
7379 u64 offset
, struct btrfs_key
*ins
)
7383 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7385 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7387 root_objectid
, owner
, offset
,
7388 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7393 * this is used by the tree logging recovery code. It records that
7394 * an extent has been allocated and makes sure to clear the free
7395 * space cache bits as well
7397 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7398 struct btrfs_root
*root
,
7399 u64 root_objectid
, u64 owner
, u64 offset
,
7400 struct btrfs_key
*ins
)
7403 struct btrfs_block_group_cache
*block_group
;
7406 * Mixed block groups will exclude before processing the log so we only
7407 * need to do the exlude dance if this fs isn't mixed.
7409 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7410 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7415 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7419 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7420 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7421 BUG_ON(ret
); /* logic error */
7422 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7423 0, owner
, offset
, ins
, 1);
7424 btrfs_put_block_group(block_group
);
7428 static struct extent_buffer
*
7429 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7430 u64 bytenr
, int level
)
7432 struct extent_buffer
*buf
;
7434 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7436 return ERR_PTR(-ENOMEM
);
7437 btrfs_set_header_generation(buf
, trans
->transid
);
7438 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7439 btrfs_tree_lock(buf
);
7440 clean_tree_block(trans
, root
->fs_info
, buf
);
7441 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7443 btrfs_set_lock_blocking(buf
);
7444 btrfs_set_buffer_uptodate(buf
);
7446 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7447 buf
->log_index
= root
->log_transid
% 2;
7449 * we allow two log transactions at a time, use different
7450 * EXENT bit to differentiate dirty pages.
7452 if (buf
->log_index
== 0)
7453 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7454 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7456 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7457 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7459 buf
->log_index
= -1;
7460 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7461 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7463 trans
->blocks_used
++;
7464 /* this returns a buffer locked for blocking */
7468 static struct btrfs_block_rsv
*
7469 use_block_rsv(struct btrfs_trans_handle
*trans
,
7470 struct btrfs_root
*root
, u32 blocksize
)
7472 struct btrfs_block_rsv
*block_rsv
;
7473 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7475 bool global_updated
= false;
7477 block_rsv
= get_block_rsv(trans
, root
);
7479 if (unlikely(block_rsv
->size
== 0))
7482 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7486 if (block_rsv
->failfast
)
7487 return ERR_PTR(ret
);
7489 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7490 global_updated
= true;
7491 update_global_block_rsv(root
->fs_info
);
7495 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7496 static DEFINE_RATELIMIT_STATE(_rs
,
7497 DEFAULT_RATELIMIT_INTERVAL
* 10,
7498 /*DEFAULT_RATELIMIT_BURST*/ 1);
7499 if (__ratelimit(&_rs
))
7501 "BTRFS: block rsv returned %d\n", ret
);
7504 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7505 BTRFS_RESERVE_NO_FLUSH
);
7509 * If we couldn't reserve metadata bytes try and use some from
7510 * the global reserve if its space type is the same as the global
7513 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7514 block_rsv
->space_info
== global_rsv
->space_info
) {
7515 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7519 return ERR_PTR(ret
);
7522 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7523 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7525 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7526 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7530 * finds a free extent and does all the dirty work required for allocation
7531 * returns the key for the extent through ins, and a tree buffer for
7532 * the first block of the extent through buf.
7534 * returns the tree buffer or NULL.
7536 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7537 struct btrfs_root
*root
,
7538 u64 parent
, u64 root_objectid
,
7539 struct btrfs_disk_key
*key
, int level
,
7540 u64 hint
, u64 empty_size
)
7542 struct btrfs_key ins
;
7543 struct btrfs_block_rsv
*block_rsv
;
7544 struct extent_buffer
*buf
;
7547 u32 blocksize
= root
->nodesize
;
7548 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7551 if (btrfs_test_is_dummy_root(root
)) {
7552 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7555 root
->alloc_bytenr
+= blocksize
;
7559 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7560 if (IS_ERR(block_rsv
))
7561 return ERR_CAST(block_rsv
);
7563 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7564 empty_size
, hint
, &ins
, 0, 0);
7566 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7567 return ERR_PTR(ret
);
7570 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7571 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7573 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7575 parent
= ins
.objectid
;
7576 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7580 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7581 struct btrfs_delayed_extent_op
*extent_op
;
7582 extent_op
= btrfs_alloc_delayed_extent_op();
7583 BUG_ON(!extent_op
); /* -ENOMEM */
7585 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7587 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7588 extent_op
->flags_to_set
= flags
;
7589 if (skinny_metadata
)
7590 extent_op
->update_key
= 0;
7592 extent_op
->update_key
= 1;
7593 extent_op
->update_flags
= 1;
7594 extent_op
->is_data
= 0;
7595 extent_op
->level
= level
;
7597 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7599 ins
.offset
, parent
, root_objectid
,
7600 level
, BTRFS_ADD_DELAYED_EXTENT
,
7602 BUG_ON(ret
); /* -ENOMEM */
7607 struct walk_control
{
7608 u64 refs
[BTRFS_MAX_LEVEL
];
7609 u64 flags
[BTRFS_MAX_LEVEL
];
7610 struct btrfs_key update_progress
;
7621 #define DROP_REFERENCE 1
7622 #define UPDATE_BACKREF 2
7624 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7625 struct btrfs_root
*root
,
7626 struct walk_control
*wc
,
7627 struct btrfs_path
*path
)
7635 struct btrfs_key key
;
7636 struct extent_buffer
*eb
;
7641 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7642 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7643 wc
->reada_count
= max(wc
->reada_count
, 2);
7645 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7646 wc
->reada_count
= min_t(int, wc
->reada_count
,
7647 BTRFS_NODEPTRS_PER_BLOCK(root
));
7650 eb
= path
->nodes
[wc
->level
];
7651 nritems
= btrfs_header_nritems(eb
);
7652 blocksize
= root
->nodesize
;
7654 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7655 if (nread
>= wc
->reada_count
)
7659 bytenr
= btrfs_node_blockptr(eb
, slot
);
7660 generation
= btrfs_node_ptr_generation(eb
, slot
);
7662 if (slot
== path
->slots
[wc
->level
])
7665 if (wc
->stage
== UPDATE_BACKREF
&&
7666 generation
<= root
->root_key
.offset
)
7669 /* We don't lock the tree block, it's OK to be racy here */
7670 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7671 wc
->level
- 1, 1, &refs
,
7673 /* We don't care about errors in readahead. */
7678 if (wc
->stage
== DROP_REFERENCE
) {
7682 if (wc
->level
== 1 &&
7683 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7685 if (!wc
->update_ref
||
7686 generation
<= root
->root_key
.offset
)
7688 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7689 ret
= btrfs_comp_cpu_keys(&key
,
7690 &wc
->update_progress
);
7694 if (wc
->level
== 1 &&
7695 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7699 readahead_tree_block(root
, bytenr
);
7702 wc
->reada_slot
= slot
;
7705 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7706 struct btrfs_root
*root
,
7707 struct extent_buffer
*eb
)
7709 int nr
= btrfs_header_nritems(eb
);
7710 int i
, extent_type
, ret
;
7711 struct btrfs_key key
;
7712 struct btrfs_file_extent_item
*fi
;
7713 u64 bytenr
, num_bytes
;
7715 for (i
= 0; i
< nr
; i
++) {
7716 btrfs_item_key_to_cpu(eb
, &key
, i
);
7718 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7721 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7722 /* filter out non qgroup-accountable extents */
7723 extent_type
= btrfs_file_extent_type(eb
, fi
);
7725 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7728 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7732 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7734 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7737 BTRFS_QGROUP_OPER_SUB_SUBTREE
, 0);
7745 * Walk up the tree from the bottom, freeing leaves and any interior
7746 * nodes which have had all slots visited. If a node (leaf or
7747 * interior) is freed, the node above it will have it's slot
7748 * incremented. The root node will never be freed.
7750 * At the end of this function, we should have a path which has all
7751 * slots incremented to the next position for a search. If we need to
7752 * read a new node it will be NULL and the node above it will have the
7753 * correct slot selected for a later read.
7755 * If we increment the root nodes slot counter past the number of
7756 * elements, 1 is returned to signal completion of the search.
7758 static int adjust_slots_upwards(struct btrfs_root
*root
,
7759 struct btrfs_path
*path
, int root_level
)
7763 struct extent_buffer
*eb
;
7765 if (root_level
== 0)
7768 while (level
<= root_level
) {
7769 eb
= path
->nodes
[level
];
7770 nr
= btrfs_header_nritems(eb
);
7771 path
->slots
[level
]++;
7772 slot
= path
->slots
[level
];
7773 if (slot
>= nr
|| level
== 0) {
7775 * Don't free the root - we will detect this
7776 * condition after our loop and return a
7777 * positive value for caller to stop walking the tree.
7779 if (level
!= root_level
) {
7780 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7781 path
->locks
[level
] = 0;
7783 free_extent_buffer(eb
);
7784 path
->nodes
[level
] = NULL
;
7785 path
->slots
[level
] = 0;
7789 * We have a valid slot to walk back down
7790 * from. Stop here so caller can process these
7799 eb
= path
->nodes
[root_level
];
7800 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7807 * root_eb is the subtree root and is locked before this function is called.
7809 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7810 struct btrfs_root
*root
,
7811 struct extent_buffer
*root_eb
,
7817 struct extent_buffer
*eb
= root_eb
;
7818 struct btrfs_path
*path
= NULL
;
7820 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7821 BUG_ON(root_eb
== NULL
);
7823 if (!root
->fs_info
->quota_enabled
)
7826 if (!extent_buffer_uptodate(root_eb
)) {
7827 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7832 if (root_level
== 0) {
7833 ret
= account_leaf_items(trans
, root
, root_eb
);
7837 path
= btrfs_alloc_path();
7842 * Walk down the tree. Missing extent blocks are filled in as
7843 * we go. Metadata is accounted every time we read a new
7846 * When we reach a leaf, we account for file extent items in it,
7847 * walk back up the tree (adjusting slot pointers as we go)
7848 * and restart the search process.
7850 extent_buffer_get(root_eb
); /* For path */
7851 path
->nodes
[root_level
] = root_eb
;
7852 path
->slots
[root_level
] = 0;
7853 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7856 while (level
>= 0) {
7857 if (path
->nodes
[level
] == NULL
) {
7862 /* We need to get child blockptr/gen from
7863 * parent before we can read it. */
7864 eb
= path
->nodes
[level
+ 1];
7865 parent_slot
= path
->slots
[level
+ 1];
7866 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7867 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7869 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
7870 if (!eb
|| !extent_buffer_uptodate(eb
)) {
7875 path
->nodes
[level
] = eb
;
7876 path
->slots
[level
] = 0;
7878 btrfs_tree_read_lock(eb
);
7879 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7880 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7882 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7886 BTRFS_QGROUP_OPER_SUB_SUBTREE
,
7894 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7898 /* Nonzero return here means we completed our search */
7899 ret
= adjust_slots_upwards(root
, path
, root_level
);
7903 /* Restart search with new slots */
7912 btrfs_free_path(path
);
7918 * helper to process tree block while walking down the tree.
7920 * when wc->stage == UPDATE_BACKREF, this function updates
7921 * back refs for pointers in the block.
7923 * NOTE: return value 1 means we should stop walking down.
7925 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7926 struct btrfs_root
*root
,
7927 struct btrfs_path
*path
,
7928 struct walk_control
*wc
, int lookup_info
)
7930 int level
= wc
->level
;
7931 struct extent_buffer
*eb
= path
->nodes
[level
];
7932 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7935 if (wc
->stage
== UPDATE_BACKREF
&&
7936 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7940 * when reference count of tree block is 1, it won't increase
7941 * again. once full backref flag is set, we never clear it.
7944 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7945 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7946 BUG_ON(!path
->locks
[level
]);
7947 ret
= btrfs_lookup_extent_info(trans
, root
,
7948 eb
->start
, level
, 1,
7951 BUG_ON(ret
== -ENOMEM
);
7954 BUG_ON(wc
->refs
[level
] == 0);
7957 if (wc
->stage
== DROP_REFERENCE
) {
7958 if (wc
->refs
[level
] > 1)
7961 if (path
->locks
[level
] && !wc
->keep_locks
) {
7962 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7963 path
->locks
[level
] = 0;
7968 /* wc->stage == UPDATE_BACKREF */
7969 if (!(wc
->flags
[level
] & flag
)) {
7970 BUG_ON(!path
->locks
[level
]);
7971 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
7972 BUG_ON(ret
); /* -ENOMEM */
7973 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7974 BUG_ON(ret
); /* -ENOMEM */
7975 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7977 btrfs_header_level(eb
), 0);
7978 BUG_ON(ret
); /* -ENOMEM */
7979 wc
->flags
[level
] |= flag
;
7983 * the block is shared by multiple trees, so it's not good to
7984 * keep the tree lock
7986 if (path
->locks
[level
] && level
> 0) {
7987 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7988 path
->locks
[level
] = 0;
7994 * helper to process tree block pointer.
7996 * when wc->stage == DROP_REFERENCE, this function checks
7997 * reference count of the block pointed to. if the block
7998 * is shared and we need update back refs for the subtree
7999 * rooted at the block, this function changes wc->stage to
8000 * UPDATE_BACKREF. if the block is shared and there is no
8001 * need to update back, this function drops the reference
8004 * NOTE: return value 1 means we should stop walking down.
8006 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8007 struct btrfs_root
*root
,
8008 struct btrfs_path
*path
,
8009 struct walk_control
*wc
, int *lookup_info
)
8015 struct btrfs_key key
;
8016 struct extent_buffer
*next
;
8017 int level
= wc
->level
;
8020 bool need_account
= false;
8022 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8023 path
->slots
[level
]);
8025 * if the lower level block was created before the snapshot
8026 * was created, we know there is no need to update back refs
8029 if (wc
->stage
== UPDATE_BACKREF
&&
8030 generation
<= root
->root_key
.offset
) {
8035 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8036 blocksize
= root
->nodesize
;
8038 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8040 next
= btrfs_find_create_tree_block(root
, bytenr
);
8043 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8047 btrfs_tree_lock(next
);
8048 btrfs_set_lock_blocking(next
);
8050 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8051 &wc
->refs
[level
- 1],
8052 &wc
->flags
[level
- 1]);
8054 btrfs_tree_unlock(next
);
8058 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8059 btrfs_err(root
->fs_info
, "Missing references.");
8064 if (wc
->stage
== DROP_REFERENCE
) {
8065 if (wc
->refs
[level
- 1] > 1) {
8066 need_account
= true;
8068 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8071 if (!wc
->update_ref
||
8072 generation
<= root
->root_key
.offset
)
8075 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8076 path
->slots
[level
]);
8077 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8081 wc
->stage
= UPDATE_BACKREF
;
8082 wc
->shared_level
= level
- 1;
8086 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8090 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8091 btrfs_tree_unlock(next
);
8092 free_extent_buffer(next
);
8098 if (reada
&& level
== 1)
8099 reada_walk_down(trans
, root
, wc
, path
);
8100 next
= read_tree_block(root
, bytenr
, generation
);
8101 if (!next
|| !extent_buffer_uptodate(next
)) {
8102 free_extent_buffer(next
);
8105 btrfs_tree_lock(next
);
8106 btrfs_set_lock_blocking(next
);
8110 BUG_ON(level
!= btrfs_header_level(next
));
8111 path
->nodes
[level
] = next
;
8112 path
->slots
[level
] = 0;
8113 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8119 wc
->refs
[level
- 1] = 0;
8120 wc
->flags
[level
- 1] = 0;
8121 if (wc
->stage
== DROP_REFERENCE
) {
8122 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8123 parent
= path
->nodes
[level
]->start
;
8125 BUG_ON(root
->root_key
.objectid
!=
8126 btrfs_header_owner(path
->nodes
[level
]));
8131 ret
= account_shared_subtree(trans
, root
, next
,
8132 generation
, level
- 1);
8134 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8135 "%d accounting shared subtree. Quota "
8136 "is out of sync, rescan required.\n",
8137 root
->fs_info
->sb
->s_id
, ret
);
8140 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8141 root
->root_key
.objectid
, level
- 1, 0, 0);
8142 BUG_ON(ret
); /* -ENOMEM */
8144 btrfs_tree_unlock(next
);
8145 free_extent_buffer(next
);
8151 * helper to process tree block while walking up the tree.
8153 * when wc->stage == DROP_REFERENCE, this function drops
8154 * reference count on the block.
8156 * when wc->stage == UPDATE_BACKREF, this function changes
8157 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8158 * to UPDATE_BACKREF previously while processing the block.
8160 * NOTE: return value 1 means we should stop walking up.
8162 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8163 struct btrfs_root
*root
,
8164 struct btrfs_path
*path
,
8165 struct walk_control
*wc
)
8168 int level
= wc
->level
;
8169 struct extent_buffer
*eb
= path
->nodes
[level
];
8172 if (wc
->stage
== UPDATE_BACKREF
) {
8173 BUG_ON(wc
->shared_level
< level
);
8174 if (level
< wc
->shared_level
)
8177 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8181 wc
->stage
= DROP_REFERENCE
;
8182 wc
->shared_level
= -1;
8183 path
->slots
[level
] = 0;
8186 * check reference count again if the block isn't locked.
8187 * we should start walking down the tree again if reference
8190 if (!path
->locks
[level
]) {
8192 btrfs_tree_lock(eb
);
8193 btrfs_set_lock_blocking(eb
);
8194 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8196 ret
= btrfs_lookup_extent_info(trans
, root
,
8197 eb
->start
, level
, 1,
8201 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8202 path
->locks
[level
] = 0;
8205 BUG_ON(wc
->refs
[level
] == 0);
8206 if (wc
->refs
[level
] == 1) {
8207 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8208 path
->locks
[level
] = 0;
8214 /* wc->stage == DROP_REFERENCE */
8215 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8217 if (wc
->refs
[level
] == 1) {
8219 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8220 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8222 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8223 BUG_ON(ret
); /* -ENOMEM */
8224 ret
= account_leaf_items(trans
, root
, eb
);
8226 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8227 "%d accounting leaf items. Quota "
8228 "is out of sync, rescan required.\n",
8229 root
->fs_info
->sb
->s_id
, ret
);
8232 /* make block locked assertion in clean_tree_block happy */
8233 if (!path
->locks
[level
] &&
8234 btrfs_header_generation(eb
) == trans
->transid
) {
8235 btrfs_tree_lock(eb
);
8236 btrfs_set_lock_blocking(eb
);
8237 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8239 clean_tree_block(trans
, root
->fs_info
, eb
);
8242 if (eb
== root
->node
) {
8243 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8246 BUG_ON(root
->root_key
.objectid
!=
8247 btrfs_header_owner(eb
));
8249 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8250 parent
= path
->nodes
[level
+ 1]->start
;
8252 BUG_ON(root
->root_key
.objectid
!=
8253 btrfs_header_owner(path
->nodes
[level
+ 1]));
8256 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8258 wc
->refs
[level
] = 0;
8259 wc
->flags
[level
] = 0;
8263 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8264 struct btrfs_root
*root
,
8265 struct btrfs_path
*path
,
8266 struct walk_control
*wc
)
8268 int level
= wc
->level
;
8269 int lookup_info
= 1;
8272 while (level
>= 0) {
8273 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8280 if (path
->slots
[level
] >=
8281 btrfs_header_nritems(path
->nodes
[level
]))
8284 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8286 path
->slots
[level
]++;
8295 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8296 struct btrfs_root
*root
,
8297 struct btrfs_path
*path
,
8298 struct walk_control
*wc
, int max_level
)
8300 int level
= wc
->level
;
8303 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8304 while (level
< max_level
&& path
->nodes
[level
]) {
8306 if (path
->slots
[level
] + 1 <
8307 btrfs_header_nritems(path
->nodes
[level
])) {
8308 path
->slots
[level
]++;
8311 ret
= walk_up_proc(trans
, root
, path
, wc
);
8315 if (path
->locks
[level
]) {
8316 btrfs_tree_unlock_rw(path
->nodes
[level
],
8317 path
->locks
[level
]);
8318 path
->locks
[level
] = 0;
8320 free_extent_buffer(path
->nodes
[level
]);
8321 path
->nodes
[level
] = NULL
;
8329 * drop a subvolume tree.
8331 * this function traverses the tree freeing any blocks that only
8332 * referenced by the tree.
8334 * when a shared tree block is found. this function decreases its
8335 * reference count by one. if update_ref is true, this function
8336 * also make sure backrefs for the shared block and all lower level
8337 * blocks are properly updated.
8339 * If called with for_reloc == 0, may exit early with -EAGAIN
8341 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8342 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8345 struct btrfs_path
*path
;
8346 struct btrfs_trans_handle
*trans
;
8347 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8348 struct btrfs_root_item
*root_item
= &root
->root_item
;
8349 struct walk_control
*wc
;
8350 struct btrfs_key key
;
8354 bool root_dropped
= false;
8356 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8358 path
= btrfs_alloc_path();
8364 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8366 btrfs_free_path(path
);
8371 trans
= btrfs_start_transaction(tree_root
, 0);
8372 if (IS_ERR(trans
)) {
8373 err
= PTR_ERR(trans
);
8378 trans
->block_rsv
= block_rsv
;
8380 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8381 level
= btrfs_header_level(root
->node
);
8382 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8383 btrfs_set_lock_blocking(path
->nodes
[level
]);
8384 path
->slots
[level
] = 0;
8385 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8386 memset(&wc
->update_progress
, 0,
8387 sizeof(wc
->update_progress
));
8389 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8390 memcpy(&wc
->update_progress
, &key
,
8391 sizeof(wc
->update_progress
));
8393 level
= root_item
->drop_level
;
8395 path
->lowest_level
= level
;
8396 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8397 path
->lowest_level
= 0;
8405 * unlock our path, this is safe because only this
8406 * function is allowed to delete this snapshot
8408 btrfs_unlock_up_safe(path
, 0);
8410 level
= btrfs_header_level(root
->node
);
8412 btrfs_tree_lock(path
->nodes
[level
]);
8413 btrfs_set_lock_blocking(path
->nodes
[level
]);
8414 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8416 ret
= btrfs_lookup_extent_info(trans
, root
,
8417 path
->nodes
[level
]->start
,
8418 level
, 1, &wc
->refs
[level
],
8424 BUG_ON(wc
->refs
[level
] == 0);
8426 if (level
== root_item
->drop_level
)
8429 btrfs_tree_unlock(path
->nodes
[level
]);
8430 path
->locks
[level
] = 0;
8431 WARN_ON(wc
->refs
[level
] != 1);
8437 wc
->shared_level
= -1;
8438 wc
->stage
= DROP_REFERENCE
;
8439 wc
->update_ref
= update_ref
;
8441 wc
->for_reloc
= for_reloc
;
8442 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8446 ret
= walk_down_tree(trans
, root
, path
, wc
);
8452 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8459 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8463 if (wc
->stage
== DROP_REFERENCE
) {
8465 btrfs_node_key(path
->nodes
[level
],
8466 &root_item
->drop_progress
,
8467 path
->slots
[level
]);
8468 root_item
->drop_level
= level
;
8471 BUG_ON(wc
->level
== 0);
8472 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8473 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8474 ret
= btrfs_update_root(trans
, tree_root
,
8478 btrfs_abort_transaction(trans
, tree_root
, ret
);
8484 * Qgroup update accounting is run from
8485 * delayed ref handling. This usually works
8486 * out because delayed refs are normally the
8487 * only way qgroup updates are added. However,
8488 * we may have added updates during our tree
8489 * walk so run qgroups here to make sure we
8490 * don't lose any updates.
8492 ret
= btrfs_delayed_qgroup_accounting(trans
,
8495 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8496 "running qgroup updates "
8497 "during snapshot delete. "
8498 "Quota is out of sync, "
8499 "rescan required.\n", ret
);
8501 btrfs_end_transaction_throttle(trans
, tree_root
);
8502 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8503 pr_debug("BTRFS: drop snapshot early exit\n");
8508 trans
= btrfs_start_transaction(tree_root
, 0);
8509 if (IS_ERR(trans
)) {
8510 err
= PTR_ERR(trans
);
8514 trans
->block_rsv
= block_rsv
;
8517 btrfs_release_path(path
);
8521 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8523 btrfs_abort_transaction(trans
, tree_root
, ret
);
8527 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8528 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8531 btrfs_abort_transaction(trans
, tree_root
, ret
);
8534 } else if (ret
> 0) {
8535 /* if we fail to delete the orphan item this time
8536 * around, it'll get picked up the next time.
8538 * The most common failure here is just -ENOENT.
8540 btrfs_del_orphan_item(trans
, tree_root
,
8541 root
->root_key
.objectid
);
8545 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8546 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8548 free_extent_buffer(root
->node
);
8549 free_extent_buffer(root
->commit_root
);
8550 btrfs_put_fs_root(root
);
8552 root_dropped
= true;
8554 ret
= btrfs_delayed_qgroup_accounting(trans
, tree_root
->fs_info
);
8556 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8557 "running qgroup updates "
8558 "during snapshot delete. "
8559 "Quota is out of sync, "
8560 "rescan required.\n", ret
);
8562 btrfs_end_transaction_throttle(trans
, tree_root
);
8565 btrfs_free_path(path
);
8568 * So if we need to stop dropping the snapshot for whatever reason we
8569 * need to make sure to add it back to the dead root list so that we
8570 * keep trying to do the work later. This also cleans up roots if we
8571 * don't have it in the radix (like when we recover after a power fail
8572 * or unmount) so we don't leak memory.
8574 if (!for_reloc
&& root_dropped
== false)
8575 btrfs_add_dead_root(root
);
8576 if (err
&& err
!= -EAGAIN
)
8577 btrfs_std_error(root
->fs_info
, err
);
8582 * drop subtree rooted at tree block 'node'.
8584 * NOTE: this function will unlock and release tree block 'node'
8585 * only used by relocation code
8587 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8588 struct btrfs_root
*root
,
8589 struct extent_buffer
*node
,
8590 struct extent_buffer
*parent
)
8592 struct btrfs_path
*path
;
8593 struct walk_control
*wc
;
8599 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8601 path
= btrfs_alloc_path();
8605 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8607 btrfs_free_path(path
);
8611 btrfs_assert_tree_locked(parent
);
8612 parent_level
= btrfs_header_level(parent
);
8613 extent_buffer_get(parent
);
8614 path
->nodes
[parent_level
] = parent
;
8615 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8617 btrfs_assert_tree_locked(node
);
8618 level
= btrfs_header_level(node
);
8619 path
->nodes
[level
] = node
;
8620 path
->slots
[level
] = 0;
8621 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8623 wc
->refs
[parent_level
] = 1;
8624 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8626 wc
->shared_level
= -1;
8627 wc
->stage
= DROP_REFERENCE
;
8631 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8634 wret
= walk_down_tree(trans
, root
, path
, wc
);
8640 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8648 btrfs_free_path(path
);
8652 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8658 * if restripe for this chunk_type is on pick target profile and
8659 * return, otherwise do the usual balance
8661 stripped
= get_restripe_target(root
->fs_info
, flags
);
8663 return extended_to_chunk(stripped
);
8665 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8667 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8668 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8669 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8671 if (num_devices
== 1) {
8672 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8673 stripped
= flags
& ~stripped
;
8675 /* turn raid0 into single device chunks */
8676 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8679 /* turn mirroring into duplication */
8680 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8681 BTRFS_BLOCK_GROUP_RAID10
))
8682 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8684 /* they already had raid on here, just return */
8685 if (flags
& stripped
)
8688 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8689 stripped
= flags
& ~stripped
;
8691 /* switch duplicated blocks with raid1 */
8692 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8693 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8695 /* this is drive concat, leave it alone */
8701 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8703 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8705 u64 min_allocable_bytes
;
8710 * We need some metadata space and system metadata space for
8711 * allocating chunks in some corner cases until we force to set
8712 * it to be readonly.
8715 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8717 min_allocable_bytes
= 1 * 1024 * 1024;
8719 min_allocable_bytes
= 0;
8721 spin_lock(&sinfo
->lock
);
8722 spin_lock(&cache
->lock
);
8729 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8730 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8732 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8733 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8734 min_allocable_bytes
<= sinfo
->total_bytes
) {
8735 sinfo
->bytes_readonly
+= num_bytes
;
8737 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8741 spin_unlock(&cache
->lock
);
8742 spin_unlock(&sinfo
->lock
);
8746 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8747 struct btrfs_block_group_cache
*cache
)
8750 struct btrfs_trans_handle
*trans
;
8757 trans
= btrfs_join_transaction(root
);
8759 return PTR_ERR(trans
);
8762 * we're not allowed to set block groups readonly after the dirty
8763 * block groups cache has started writing. If it already started,
8764 * back off and let this transaction commit
8766 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
8767 if (trans
->transaction
->dirty_bg_run
) {
8768 u64 transid
= trans
->transid
;
8770 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8771 btrfs_end_transaction(trans
, root
);
8773 ret
= btrfs_wait_for_commit(root
, transid
);
8780 ret
= set_block_group_ro(cache
, 0);
8783 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8784 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8788 ret
= set_block_group_ro(cache
, 0);
8790 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
8791 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8792 check_system_chunk(trans
, root
, alloc_flags
);
8794 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8796 btrfs_end_transaction(trans
, root
);
8800 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8801 struct btrfs_root
*root
, u64 type
)
8803 u64 alloc_flags
= get_alloc_profile(root
, type
);
8804 return do_chunk_alloc(trans
, root
, alloc_flags
,
8809 * helper to account the unused space of all the readonly block group in the
8810 * space_info. takes mirrors into account.
8812 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8814 struct btrfs_block_group_cache
*block_group
;
8818 /* It's df, we don't care if it's racey */
8819 if (list_empty(&sinfo
->ro_bgs
))
8822 spin_lock(&sinfo
->lock
);
8823 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8824 spin_lock(&block_group
->lock
);
8826 if (!block_group
->ro
) {
8827 spin_unlock(&block_group
->lock
);
8831 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8832 BTRFS_BLOCK_GROUP_RAID10
|
8833 BTRFS_BLOCK_GROUP_DUP
))
8838 free_bytes
+= (block_group
->key
.offset
-
8839 btrfs_block_group_used(&block_group
->item
)) *
8842 spin_unlock(&block_group
->lock
);
8844 spin_unlock(&sinfo
->lock
);
8849 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8850 struct btrfs_block_group_cache
*cache
)
8852 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8857 spin_lock(&sinfo
->lock
);
8858 spin_lock(&cache
->lock
);
8859 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8860 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8861 sinfo
->bytes_readonly
-= num_bytes
;
8863 list_del_init(&cache
->ro_list
);
8864 spin_unlock(&cache
->lock
);
8865 spin_unlock(&sinfo
->lock
);
8869 * checks to see if its even possible to relocate this block group.
8871 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8872 * ok to go ahead and try.
8874 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8876 struct btrfs_block_group_cache
*block_group
;
8877 struct btrfs_space_info
*space_info
;
8878 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8879 struct btrfs_device
*device
;
8880 struct btrfs_trans_handle
*trans
;
8889 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8891 /* odd, couldn't find the block group, leave it alone */
8895 min_free
= btrfs_block_group_used(&block_group
->item
);
8897 /* no bytes used, we're good */
8901 space_info
= block_group
->space_info
;
8902 spin_lock(&space_info
->lock
);
8904 full
= space_info
->full
;
8907 * if this is the last block group we have in this space, we can't
8908 * relocate it unless we're able to allocate a new chunk below.
8910 * Otherwise, we need to make sure we have room in the space to handle
8911 * all of the extents from this block group. If we can, we're good
8913 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8914 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8915 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8916 min_free
< space_info
->total_bytes
)) {
8917 spin_unlock(&space_info
->lock
);
8920 spin_unlock(&space_info
->lock
);
8923 * ok we don't have enough space, but maybe we have free space on our
8924 * devices to allocate new chunks for relocation, so loop through our
8925 * alloc devices and guess if we have enough space. if this block
8926 * group is going to be restriped, run checks against the target
8927 * profile instead of the current one.
8939 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8941 index
= __get_raid_index(extended_to_chunk(target
));
8944 * this is just a balance, so if we were marked as full
8945 * we know there is no space for a new chunk
8950 index
= get_block_group_index(block_group
);
8953 if (index
== BTRFS_RAID_RAID10
) {
8957 } else if (index
== BTRFS_RAID_RAID1
) {
8959 } else if (index
== BTRFS_RAID_DUP
) {
8962 } else if (index
== BTRFS_RAID_RAID0
) {
8963 dev_min
= fs_devices
->rw_devices
;
8964 min_free
= div64_u64(min_free
, dev_min
);
8967 /* We need to do this so that we can look at pending chunks */
8968 trans
= btrfs_join_transaction(root
);
8969 if (IS_ERR(trans
)) {
8970 ret
= PTR_ERR(trans
);
8974 mutex_lock(&root
->fs_info
->chunk_mutex
);
8975 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8979 * check to make sure we can actually find a chunk with enough
8980 * space to fit our block group in.
8982 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8983 !device
->is_tgtdev_for_dev_replace
) {
8984 ret
= find_free_dev_extent(trans
, device
, min_free
,
8989 if (dev_nr
>= dev_min
)
8995 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8996 btrfs_end_transaction(trans
, root
);
8998 btrfs_put_block_group(block_group
);
9002 static int find_first_block_group(struct btrfs_root
*root
,
9003 struct btrfs_path
*path
, struct btrfs_key
*key
)
9006 struct btrfs_key found_key
;
9007 struct extent_buffer
*leaf
;
9010 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9015 slot
= path
->slots
[0];
9016 leaf
= path
->nodes
[0];
9017 if (slot
>= btrfs_header_nritems(leaf
)) {
9018 ret
= btrfs_next_leaf(root
, path
);
9025 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9027 if (found_key
.objectid
>= key
->objectid
&&
9028 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9038 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9040 struct btrfs_block_group_cache
*block_group
;
9044 struct inode
*inode
;
9046 block_group
= btrfs_lookup_first_block_group(info
, last
);
9047 while (block_group
) {
9048 spin_lock(&block_group
->lock
);
9049 if (block_group
->iref
)
9051 spin_unlock(&block_group
->lock
);
9052 block_group
= next_block_group(info
->tree_root
,
9062 inode
= block_group
->inode
;
9063 block_group
->iref
= 0;
9064 block_group
->inode
= NULL
;
9065 spin_unlock(&block_group
->lock
);
9067 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9068 btrfs_put_block_group(block_group
);
9072 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9074 struct btrfs_block_group_cache
*block_group
;
9075 struct btrfs_space_info
*space_info
;
9076 struct btrfs_caching_control
*caching_ctl
;
9079 down_write(&info
->commit_root_sem
);
9080 while (!list_empty(&info
->caching_block_groups
)) {
9081 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9082 struct btrfs_caching_control
, list
);
9083 list_del(&caching_ctl
->list
);
9084 put_caching_control(caching_ctl
);
9086 up_write(&info
->commit_root_sem
);
9088 spin_lock(&info
->unused_bgs_lock
);
9089 while (!list_empty(&info
->unused_bgs
)) {
9090 block_group
= list_first_entry(&info
->unused_bgs
,
9091 struct btrfs_block_group_cache
,
9093 list_del_init(&block_group
->bg_list
);
9094 btrfs_put_block_group(block_group
);
9096 spin_unlock(&info
->unused_bgs_lock
);
9098 spin_lock(&info
->block_group_cache_lock
);
9099 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9100 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9102 rb_erase(&block_group
->cache_node
,
9103 &info
->block_group_cache_tree
);
9104 RB_CLEAR_NODE(&block_group
->cache_node
);
9105 spin_unlock(&info
->block_group_cache_lock
);
9107 down_write(&block_group
->space_info
->groups_sem
);
9108 list_del(&block_group
->list
);
9109 up_write(&block_group
->space_info
->groups_sem
);
9111 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9112 wait_block_group_cache_done(block_group
);
9115 * We haven't cached this block group, which means we could
9116 * possibly have excluded extents on this block group.
9118 if (block_group
->cached
== BTRFS_CACHE_NO
||
9119 block_group
->cached
== BTRFS_CACHE_ERROR
)
9120 free_excluded_extents(info
->extent_root
, block_group
);
9122 btrfs_remove_free_space_cache(block_group
);
9123 btrfs_put_block_group(block_group
);
9125 spin_lock(&info
->block_group_cache_lock
);
9127 spin_unlock(&info
->block_group_cache_lock
);
9129 /* now that all the block groups are freed, go through and
9130 * free all the space_info structs. This is only called during
9131 * the final stages of unmount, and so we know nobody is
9132 * using them. We call synchronize_rcu() once before we start,
9133 * just to be on the safe side.
9137 release_global_block_rsv(info
);
9139 while (!list_empty(&info
->space_info
)) {
9142 space_info
= list_entry(info
->space_info
.next
,
9143 struct btrfs_space_info
,
9145 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
9146 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9147 space_info
->bytes_reserved
> 0 ||
9148 space_info
->bytes_may_use
> 0)) {
9149 dump_space_info(space_info
, 0, 0);
9152 list_del(&space_info
->list
);
9153 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9154 struct kobject
*kobj
;
9155 kobj
= space_info
->block_group_kobjs
[i
];
9156 space_info
->block_group_kobjs
[i
] = NULL
;
9162 kobject_del(&space_info
->kobj
);
9163 kobject_put(&space_info
->kobj
);
9168 static void __link_block_group(struct btrfs_space_info
*space_info
,
9169 struct btrfs_block_group_cache
*cache
)
9171 int index
= get_block_group_index(cache
);
9174 down_write(&space_info
->groups_sem
);
9175 if (list_empty(&space_info
->block_groups
[index
]))
9177 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9178 up_write(&space_info
->groups_sem
);
9181 struct raid_kobject
*rkobj
;
9184 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9187 rkobj
->raid_type
= index
;
9188 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9189 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9190 "%s", get_raid_name(index
));
9192 kobject_put(&rkobj
->kobj
);
9195 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9200 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9203 static struct btrfs_block_group_cache
*
9204 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9206 struct btrfs_block_group_cache
*cache
;
9208 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9212 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9214 if (!cache
->free_space_ctl
) {
9219 cache
->key
.objectid
= start
;
9220 cache
->key
.offset
= size
;
9221 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9223 cache
->sectorsize
= root
->sectorsize
;
9224 cache
->fs_info
= root
->fs_info
;
9225 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9226 &root
->fs_info
->mapping_tree
,
9228 atomic_set(&cache
->count
, 1);
9229 spin_lock_init(&cache
->lock
);
9230 init_rwsem(&cache
->data_rwsem
);
9231 INIT_LIST_HEAD(&cache
->list
);
9232 INIT_LIST_HEAD(&cache
->cluster_list
);
9233 INIT_LIST_HEAD(&cache
->bg_list
);
9234 INIT_LIST_HEAD(&cache
->ro_list
);
9235 INIT_LIST_HEAD(&cache
->dirty_list
);
9236 INIT_LIST_HEAD(&cache
->io_list
);
9237 btrfs_init_free_space_ctl(cache
);
9238 atomic_set(&cache
->trimming
, 0);
9243 int btrfs_read_block_groups(struct btrfs_root
*root
)
9245 struct btrfs_path
*path
;
9247 struct btrfs_block_group_cache
*cache
;
9248 struct btrfs_fs_info
*info
= root
->fs_info
;
9249 struct btrfs_space_info
*space_info
;
9250 struct btrfs_key key
;
9251 struct btrfs_key found_key
;
9252 struct extent_buffer
*leaf
;
9256 root
= info
->extent_root
;
9259 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9260 path
= btrfs_alloc_path();
9265 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9266 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9267 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9269 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9273 ret
= find_first_block_group(root
, path
, &key
);
9279 leaf
= path
->nodes
[0];
9280 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9282 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9291 * When we mount with old space cache, we need to
9292 * set BTRFS_DC_CLEAR and set dirty flag.
9294 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9295 * truncate the old free space cache inode and
9297 * b) Setting 'dirty flag' makes sure that we flush
9298 * the new space cache info onto disk.
9300 if (btrfs_test_opt(root
, SPACE_CACHE
))
9301 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9304 read_extent_buffer(leaf
, &cache
->item
,
9305 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9306 sizeof(cache
->item
));
9307 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9309 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9310 btrfs_release_path(path
);
9313 * We need to exclude the super stripes now so that the space
9314 * info has super bytes accounted for, otherwise we'll think
9315 * we have more space than we actually do.
9317 ret
= exclude_super_stripes(root
, cache
);
9320 * We may have excluded something, so call this just in
9323 free_excluded_extents(root
, cache
);
9324 btrfs_put_block_group(cache
);
9329 * check for two cases, either we are full, and therefore
9330 * don't need to bother with the caching work since we won't
9331 * find any space, or we are empty, and we can just add all
9332 * the space in and be done with it. This saves us _alot_ of
9333 * time, particularly in the full case.
9335 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9336 cache
->last_byte_to_unpin
= (u64
)-1;
9337 cache
->cached
= BTRFS_CACHE_FINISHED
;
9338 free_excluded_extents(root
, cache
);
9339 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9340 cache
->last_byte_to_unpin
= (u64
)-1;
9341 cache
->cached
= BTRFS_CACHE_FINISHED
;
9342 add_new_free_space(cache
, root
->fs_info
,
9344 found_key
.objectid
+
9346 free_excluded_extents(root
, cache
);
9349 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9351 btrfs_remove_free_space_cache(cache
);
9352 btrfs_put_block_group(cache
);
9356 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9357 btrfs_block_group_used(&cache
->item
),
9360 btrfs_remove_free_space_cache(cache
);
9361 spin_lock(&info
->block_group_cache_lock
);
9362 rb_erase(&cache
->cache_node
,
9363 &info
->block_group_cache_tree
);
9364 RB_CLEAR_NODE(&cache
->cache_node
);
9365 spin_unlock(&info
->block_group_cache_lock
);
9366 btrfs_put_block_group(cache
);
9370 cache
->space_info
= space_info
;
9371 spin_lock(&cache
->space_info
->lock
);
9372 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9373 spin_unlock(&cache
->space_info
->lock
);
9375 __link_block_group(space_info
, cache
);
9377 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9378 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9379 set_block_group_ro(cache
, 1);
9380 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9381 spin_lock(&info
->unused_bgs_lock
);
9382 /* Should always be true but just in case. */
9383 if (list_empty(&cache
->bg_list
)) {
9384 btrfs_get_block_group(cache
);
9385 list_add_tail(&cache
->bg_list
,
9388 spin_unlock(&info
->unused_bgs_lock
);
9392 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9393 if (!(get_alloc_profile(root
, space_info
->flags
) &
9394 (BTRFS_BLOCK_GROUP_RAID10
|
9395 BTRFS_BLOCK_GROUP_RAID1
|
9396 BTRFS_BLOCK_GROUP_RAID5
|
9397 BTRFS_BLOCK_GROUP_RAID6
|
9398 BTRFS_BLOCK_GROUP_DUP
)))
9401 * avoid allocating from un-mirrored block group if there are
9402 * mirrored block groups.
9404 list_for_each_entry(cache
,
9405 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9407 set_block_group_ro(cache
, 1);
9408 list_for_each_entry(cache
,
9409 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9411 set_block_group_ro(cache
, 1);
9414 init_global_block_rsv(info
);
9417 btrfs_free_path(path
);
9421 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9422 struct btrfs_root
*root
)
9424 struct btrfs_block_group_cache
*block_group
, *tmp
;
9425 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9426 struct btrfs_block_group_item item
;
9427 struct btrfs_key key
;
9430 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9434 spin_lock(&block_group
->lock
);
9435 memcpy(&item
, &block_group
->item
, sizeof(item
));
9436 memcpy(&key
, &block_group
->key
, sizeof(key
));
9437 spin_unlock(&block_group
->lock
);
9439 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9442 btrfs_abort_transaction(trans
, extent_root
, ret
);
9443 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9444 key
.objectid
, key
.offset
);
9446 btrfs_abort_transaction(trans
, extent_root
, ret
);
9448 list_del_init(&block_group
->bg_list
);
9452 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9453 struct btrfs_root
*root
, u64 bytes_used
,
9454 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9458 struct btrfs_root
*extent_root
;
9459 struct btrfs_block_group_cache
*cache
;
9461 extent_root
= root
->fs_info
->extent_root
;
9463 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9465 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9469 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9470 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9471 btrfs_set_block_group_flags(&cache
->item
, type
);
9473 cache
->flags
= type
;
9474 cache
->last_byte_to_unpin
= (u64
)-1;
9475 cache
->cached
= BTRFS_CACHE_FINISHED
;
9476 ret
= exclude_super_stripes(root
, cache
);
9479 * We may have excluded something, so call this just in
9482 free_excluded_extents(root
, cache
);
9483 btrfs_put_block_group(cache
);
9487 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9488 chunk_offset
+ size
);
9490 free_excluded_extents(root
, cache
);
9492 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9494 btrfs_remove_free_space_cache(cache
);
9495 btrfs_put_block_group(cache
);
9499 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9500 &cache
->space_info
);
9502 btrfs_remove_free_space_cache(cache
);
9503 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9504 rb_erase(&cache
->cache_node
,
9505 &root
->fs_info
->block_group_cache_tree
);
9506 RB_CLEAR_NODE(&cache
->cache_node
);
9507 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9508 btrfs_put_block_group(cache
);
9511 update_global_block_rsv(root
->fs_info
);
9513 spin_lock(&cache
->space_info
->lock
);
9514 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9515 spin_unlock(&cache
->space_info
->lock
);
9517 __link_block_group(cache
->space_info
, cache
);
9519 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9521 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9526 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9528 u64 extra_flags
= chunk_to_extended(flags
) &
9529 BTRFS_EXTENDED_PROFILE_MASK
;
9531 write_seqlock(&fs_info
->profiles_lock
);
9532 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9533 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9534 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9535 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9536 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9537 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9538 write_sequnlock(&fs_info
->profiles_lock
);
9541 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9542 struct btrfs_root
*root
, u64 group_start
,
9543 struct extent_map
*em
)
9545 struct btrfs_path
*path
;
9546 struct btrfs_block_group_cache
*block_group
;
9547 struct btrfs_free_cluster
*cluster
;
9548 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9549 struct btrfs_key key
;
9550 struct inode
*inode
;
9551 struct kobject
*kobj
= NULL
;
9555 struct btrfs_caching_control
*caching_ctl
= NULL
;
9558 root
= root
->fs_info
->extent_root
;
9560 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9561 BUG_ON(!block_group
);
9562 BUG_ON(!block_group
->ro
);
9565 * Free the reserved super bytes from this block group before
9568 free_excluded_extents(root
, block_group
);
9570 memcpy(&key
, &block_group
->key
, sizeof(key
));
9571 index
= get_block_group_index(block_group
);
9572 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9573 BTRFS_BLOCK_GROUP_RAID1
|
9574 BTRFS_BLOCK_GROUP_RAID10
))
9579 /* make sure this block group isn't part of an allocation cluster */
9580 cluster
= &root
->fs_info
->data_alloc_cluster
;
9581 spin_lock(&cluster
->refill_lock
);
9582 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9583 spin_unlock(&cluster
->refill_lock
);
9586 * make sure this block group isn't part of a metadata
9587 * allocation cluster
9589 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9590 spin_lock(&cluster
->refill_lock
);
9591 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9592 spin_unlock(&cluster
->refill_lock
);
9594 path
= btrfs_alloc_path();
9601 * get the inode first so any iput calls done for the io_list
9602 * aren't the final iput (no unlinks allowed now)
9604 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9606 mutex_lock(&trans
->transaction
->cache_write_mutex
);
9608 * make sure our free spache cache IO is done before remove the
9611 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9612 if (!list_empty(&block_group
->io_list
)) {
9613 list_del_init(&block_group
->io_list
);
9615 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
9617 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9618 btrfs_wait_cache_io(root
, trans
, block_group
,
9619 &block_group
->io_ctl
, path
,
9620 block_group
->key
.objectid
);
9621 btrfs_put_block_group(block_group
);
9622 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9625 if (!list_empty(&block_group
->dirty_list
)) {
9626 list_del_init(&block_group
->dirty_list
);
9627 btrfs_put_block_group(block_group
);
9629 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9630 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
9632 if (!IS_ERR(inode
)) {
9633 ret
= btrfs_orphan_add(trans
, inode
);
9635 btrfs_add_delayed_iput(inode
);
9639 /* One for the block groups ref */
9640 spin_lock(&block_group
->lock
);
9641 if (block_group
->iref
) {
9642 block_group
->iref
= 0;
9643 block_group
->inode
= NULL
;
9644 spin_unlock(&block_group
->lock
);
9647 spin_unlock(&block_group
->lock
);
9649 /* One for our lookup ref */
9650 btrfs_add_delayed_iput(inode
);
9653 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9654 key
.offset
= block_group
->key
.objectid
;
9657 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9661 btrfs_release_path(path
);
9663 ret
= btrfs_del_item(trans
, tree_root
, path
);
9666 btrfs_release_path(path
);
9669 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9670 rb_erase(&block_group
->cache_node
,
9671 &root
->fs_info
->block_group_cache_tree
);
9672 RB_CLEAR_NODE(&block_group
->cache_node
);
9674 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9675 root
->fs_info
->first_logical_byte
= (u64
)-1;
9676 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9678 down_write(&block_group
->space_info
->groups_sem
);
9680 * we must use list_del_init so people can check to see if they
9681 * are still on the list after taking the semaphore
9683 list_del_init(&block_group
->list
);
9684 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9685 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9686 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9687 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9689 up_write(&block_group
->space_info
->groups_sem
);
9695 if (block_group
->has_caching_ctl
)
9696 caching_ctl
= get_caching_control(block_group
);
9697 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9698 wait_block_group_cache_done(block_group
);
9699 if (block_group
->has_caching_ctl
) {
9700 down_write(&root
->fs_info
->commit_root_sem
);
9702 struct btrfs_caching_control
*ctl
;
9704 list_for_each_entry(ctl
,
9705 &root
->fs_info
->caching_block_groups
, list
)
9706 if (ctl
->block_group
== block_group
) {
9708 atomic_inc(&caching_ctl
->count
);
9713 list_del_init(&caching_ctl
->list
);
9714 up_write(&root
->fs_info
->commit_root_sem
);
9716 /* Once for the caching bgs list and once for us. */
9717 put_caching_control(caching_ctl
);
9718 put_caching_control(caching_ctl
);
9722 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9723 if (!list_empty(&block_group
->dirty_list
)) {
9726 if (!list_empty(&block_group
->io_list
)) {
9729 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9730 btrfs_remove_free_space_cache(block_group
);
9732 spin_lock(&block_group
->space_info
->lock
);
9733 list_del_init(&block_group
->ro_list
);
9734 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9735 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9736 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9737 spin_unlock(&block_group
->space_info
->lock
);
9739 memcpy(&key
, &block_group
->key
, sizeof(key
));
9742 if (!list_empty(&em
->list
)) {
9743 /* We're in the transaction->pending_chunks list. */
9744 free_extent_map(em
);
9746 spin_lock(&block_group
->lock
);
9747 block_group
->removed
= 1;
9749 * At this point trimming can't start on this block group, because we
9750 * removed the block group from the tree fs_info->block_group_cache_tree
9751 * so no one can't find it anymore and even if someone already got this
9752 * block group before we removed it from the rbtree, they have already
9753 * incremented block_group->trimming - if they didn't, they won't find
9754 * any free space entries because we already removed them all when we
9755 * called btrfs_remove_free_space_cache().
9757 * And we must not remove the extent map from the fs_info->mapping_tree
9758 * to prevent the same logical address range and physical device space
9759 * ranges from being reused for a new block group. This is because our
9760 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9761 * completely transactionless, so while it is trimming a range the
9762 * currently running transaction might finish and a new one start,
9763 * allowing for new block groups to be created that can reuse the same
9764 * physical device locations unless we take this special care.
9766 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9768 * Make sure a trimmer task always sees the em in the pinned_chunks list
9769 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9770 * before checking block_group->removed).
9774 * Our em might be in trans->transaction->pending_chunks which
9775 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9776 * and so is the fs_info->pinned_chunks list.
9778 * So at this point we must be holding the chunk_mutex to avoid
9779 * any races with chunk allocation (more specifically at
9780 * volumes.c:contains_pending_extent()), to ensure it always
9781 * sees the em, either in the pending_chunks list or in the
9782 * pinned_chunks list.
9784 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9786 spin_unlock(&block_group
->lock
);
9789 struct extent_map_tree
*em_tree
;
9791 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9792 write_lock(&em_tree
->lock
);
9794 * The em might be in the pending_chunks list, so make sure the
9795 * chunk mutex is locked, since remove_extent_mapping() will
9796 * delete us from that list.
9798 remove_extent_mapping(em_tree
, em
);
9799 write_unlock(&em_tree
->lock
);
9800 /* once for the tree */
9801 free_extent_map(em
);
9804 unlock_chunks(root
);
9806 btrfs_put_block_group(block_group
);
9807 btrfs_put_block_group(block_group
);
9809 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9815 ret
= btrfs_del_item(trans
, root
, path
);
9817 btrfs_free_path(path
);
9822 * Process the unused_bgs list and remove any that don't have any allocated
9823 * space inside of them.
9825 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9827 struct btrfs_block_group_cache
*block_group
;
9828 struct btrfs_space_info
*space_info
;
9829 struct btrfs_root
*root
= fs_info
->extent_root
;
9830 struct btrfs_trans_handle
*trans
;
9836 spin_lock(&fs_info
->unused_bgs_lock
);
9837 while (!list_empty(&fs_info
->unused_bgs
)) {
9840 block_group
= list_first_entry(&fs_info
->unused_bgs
,
9841 struct btrfs_block_group_cache
,
9843 space_info
= block_group
->space_info
;
9844 list_del_init(&block_group
->bg_list
);
9845 if (ret
|| btrfs_mixed_space_info(space_info
)) {
9846 btrfs_put_block_group(block_group
);
9849 spin_unlock(&fs_info
->unused_bgs_lock
);
9851 /* Don't want to race with allocators so take the groups_sem */
9852 down_write(&space_info
->groups_sem
);
9853 spin_lock(&block_group
->lock
);
9854 if (block_group
->reserved
||
9855 btrfs_block_group_used(&block_group
->item
) ||
9858 * We want to bail if we made new allocations or have
9859 * outstanding allocations in this block group. We do
9860 * the ro check in case balance is currently acting on
9863 spin_unlock(&block_group
->lock
);
9864 up_write(&space_info
->groups_sem
);
9867 spin_unlock(&block_group
->lock
);
9869 /* We don't want to force the issue, only flip if it's ok. */
9870 ret
= set_block_group_ro(block_group
, 0);
9871 up_write(&space_info
->groups_sem
);
9878 * Want to do this before we do anything else so we can recover
9879 * properly if we fail to join the transaction.
9881 /* 1 for btrfs_orphan_reserve_metadata() */
9882 trans
= btrfs_start_transaction(root
, 1);
9883 if (IS_ERR(trans
)) {
9884 btrfs_set_block_group_rw(root
, block_group
);
9885 ret
= PTR_ERR(trans
);
9890 * We could have pending pinned extents for this block group,
9891 * just delete them, we don't care about them anymore.
9893 start
= block_group
->key
.objectid
;
9894 end
= start
+ block_group
->key
.offset
- 1;
9896 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9897 * btrfs_finish_extent_commit(). If we are at transaction N,
9898 * another task might be running finish_extent_commit() for the
9899 * previous transaction N - 1, and have seen a range belonging
9900 * to the block group in freed_extents[] before we were able to
9901 * clear the whole block group range from freed_extents[]. This
9902 * means that task can lookup for the block group after we
9903 * unpinned it from freed_extents[] and removed it, leading to
9904 * a BUG_ON() at btrfs_unpin_extent_range().
9906 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
9907 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
9908 EXTENT_DIRTY
, GFP_NOFS
);
9910 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9911 btrfs_set_block_group_rw(root
, block_group
);
9914 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
9915 EXTENT_DIRTY
, GFP_NOFS
);
9917 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9918 btrfs_set_block_group_rw(root
, block_group
);
9921 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9923 /* Reset pinned so btrfs_put_block_group doesn't complain */
9924 block_group
->pinned
= 0;
9927 * Btrfs_remove_chunk will abort the transaction if things go
9930 ret
= btrfs_remove_chunk(trans
, root
,
9931 block_group
->key
.objectid
);
9933 btrfs_end_transaction(trans
, root
);
9935 btrfs_put_block_group(block_group
);
9936 spin_lock(&fs_info
->unused_bgs_lock
);
9938 spin_unlock(&fs_info
->unused_bgs_lock
);
9941 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
9943 struct btrfs_space_info
*space_info
;
9944 struct btrfs_super_block
*disk_super
;
9950 disk_super
= fs_info
->super_copy
;
9951 if (!btrfs_super_root(disk_super
))
9954 features
= btrfs_super_incompat_flags(disk_super
);
9955 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
9958 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
9959 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9964 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
9965 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9967 flags
= BTRFS_BLOCK_GROUP_METADATA
;
9968 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9972 flags
= BTRFS_BLOCK_GROUP_DATA
;
9973 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9979 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
9981 return unpin_extent_range(root
, start
, end
, false);
9984 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
9986 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9987 struct btrfs_block_group_cache
*cache
= NULL
;
9992 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
9996 * try to trim all FS space, our block group may start from non-zero.
9998 if (range
->len
== total_bytes
)
9999 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10001 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10004 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10005 btrfs_put_block_group(cache
);
10009 start
= max(range
->start
, cache
->key
.objectid
);
10010 end
= min(range
->start
+ range
->len
,
10011 cache
->key
.objectid
+ cache
->key
.offset
);
10013 if (end
- start
>= range
->minlen
) {
10014 if (!block_group_cache_done(cache
)) {
10015 ret
= cache_block_group(cache
, 0);
10017 btrfs_put_block_group(cache
);
10020 ret
= wait_block_group_cache_done(cache
);
10022 btrfs_put_block_group(cache
);
10026 ret
= btrfs_trim_block_group(cache
,
10032 trimmed
+= group_trimmed
;
10034 btrfs_put_block_group(cache
);
10039 cache
= next_block_group(fs_info
->tree_root
, cache
);
10042 range
->len
= trimmed
;
10047 * btrfs_{start,end}_write_no_snapshoting() are similar to
10048 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10049 * data into the page cache through nocow before the subvolume is snapshoted,
10050 * but flush the data into disk after the snapshot creation, or to prevent
10051 * operations while snapshoting is ongoing and that cause the snapshot to be
10052 * inconsistent (writes followed by expanding truncates for example).
10054 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10056 percpu_counter_dec(&root
->subv_writers
->counter
);
10058 * Make sure counter is updated before we wake up
10062 if (waitqueue_active(&root
->subv_writers
->wait
))
10063 wake_up(&root
->subv_writers
->wait
);
10066 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10068 if (atomic_read(&root
->will_be_snapshoted
))
10071 percpu_counter_inc(&root
->subv_writers
->counter
);
10073 * Make sure counter is updated before we check for snapshot creation.
10076 if (atomic_read(&root
->will_be_snapshoted
)) {
10077 btrfs_end_write_no_snapshoting(root
);