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
;
3861 int need_commit
= 2;
3862 int have_pinned_space
;
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
);
3919 have_pinned_space
= 1;
3925 data_sinfo
= fs_info
->data_sinfo
;
3931 * If we don't have enough pinned space to deal with this
3932 * allocation, and no removed chunk in current transaction,
3933 * don't bother committing the transaction.
3935 have_pinned_space
= percpu_counter_compare(
3936 &data_sinfo
->total_bytes_pinned
,
3937 used
+ bytes
- data_sinfo
->total_bytes
);
3938 spin_unlock(&data_sinfo
->lock
);
3940 /* commit the current transaction and try again */
3943 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3946 trans
= btrfs_join_transaction(root
);
3948 return PTR_ERR(trans
);
3949 if (have_pinned_space
>= 0 ||
3950 trans
->transaction
->have_free_bgs
||
3952 ret
= btrfs_commit_transaction(trans
, root
);
3956 * make sure that all running delayed iput are
3959 down_write(&root
->fs_info
->delayed_iput_sem
);
3960 up_write(&root
->fs_info
->delayed_iput_sem
);
3963 btrfs_end_transaction(trans
, root
);
3967 trace_btrfs_space_reservation(root
->fs_info
,
3968 "space_info:enospc",
3969 data_sinfo
->flags
, bytes
, 1);
3972 data_sinfo
->bytes_may_use
+= bytes
;
3973 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3974 data_sinfo
->flags
, bytes
, 1);
3975 spin_unlock(&data_sinfo
->lock
);
3981 * Called if we need to clear a data reservation for this inode.
3983 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3985 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3986 struct btrfs_space_info
*data_sinfo
;
3988 /* make sure bytes are sectorsize aligned */
3989 bytes
= ALIGN(bytes
, root
->sectorsize
);
3991 data_sinfo
= root
->fs_info
->data_sinfo
;
3992 spin_lock(&data_sinfo
->lock
);
3993 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3994 data_sinfo
->bytes_may_use
-= bytes
;
3995 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3996 data_sinfo
->flags
, bytes
, 0);
3997 spin_unlock(&data_sinfo
->lock
);
4000 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4002 struct list_head
*head
= &info
->space_info
;
4003 struct btrfs_space_info
*found
;
4006 list_for_each_entry_rcu(found
, head
, list
) {
4007 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4008 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4013 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4015 return (global
->size
<< 1);
4018 static int should_alloc_chunk(struct btrfs_root
*root
,
4019 struct btrfs_space_info
*sinfo
, int force
)
4021 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4022 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4023 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4026 if (force
== CHUNK_ALLOC_FORCE
)
4030 * We need to take into account the global rsv because for all intents
4031 * and purposes it's used space. Don't worry about locking the
4032 * global_rsv, it doesn't change except when the transaction commits.
4034 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4035 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4038 * in limited mode, we want to have some free space up to
4039 * about 1% of the FS size.
4041 if (force
== CHUNK_ALLOC_LIMITED
) {
4042 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4043 thresh
= max_t(u64
, 64 * 1024 * 1024,
4044 div_factor_fine(thresh
, 1));
4046 if (num_bytes
- num_allocated
< thresh
)
4050 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
4055 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
4059 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4060 BTRFS_BLOCK_GROUP_RAID0
|
4061 BTRFS_BLOCK_GROUP_RAID5
|
4062 BTRFS_BLOCK_GROUP_RAID6
))
4063 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4064 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4067 num_dev
= 1; /* DUP or single */
4069 /* metadata for updaing devices and chunk tree */
4070 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
4073 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
4074 struct btrfs_root
*root
, u64 type
)
4076 struct btrfs_space_info
*info
;
4080 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4081 spin_lock(&info
->lock
);
4082 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4083 info
->bytes_reserved
- info
->bytes_readonly
;
4084 spin_unlock(&info
->lock
);
4086 thresh
= get_system_chunk_thresh(root
, type
);
4087 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4088 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4089 left
, thresh
, type
);
4090 dump_space_info(info
, 0, 0);
4093 if (left
< thresh
) {
4096 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4097 btrfs_alloc_chunk(trans
, root
, flags
);
4101 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4102 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4104 struct btrfs_space_info
*space_info
;
4105 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4106 int wait_for_alloc
= 0;
4109 /* Don't re-enter if we're already allocating a chunk */
4110 if (trans
->allocating_chunk
)
4113 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4115 ret
= update_space_info(extent_root
->fs_info
, flags
,
4117 BUG_ON(ret
); /* -ENOMEM */
4119 BUG_ON(!space_info
); /* Logic error */
4122 spin_lock(&space_info
->lock
);
4123 if (force
< space_info
->force_alloc
)
4124 force
= space_info
->force_alloc
;
4125 if (space_info
->full
) {
4126 if (should_alloc_chunk(extent_root
, space_info
, force
))
4130 spin_unlock(&space_info
->lock
);
4134 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4135 spin_unlock(&space_info
->lock
);
4137 } else if (space_info
->chunk_alloc
) {
4140 space_info
->chunk_alloc
= 1;
4143 spin_unlock(&space_info
->lock
);
4145 mutex_lock(&fs_info
->chunk_mutex
);
4148 * The chunk_mutex is held throughout the entirety of a chunk
4149 * allocation, so once we've acquired the chunk_mutex we know that the
4150 * other guy is done and we need to recheck and see if we should
4153 if (wait_for_alloc
) {
4154 mutex_unlock(&fs_info
->chunk_mutex
);
4159 trans
->allocating_chunk
= true;
4162 * If we have mixed data/metadata chunks we want to make sure we keep
4163 * allocating mixed chunks instead of individual chunks.
4165 if (btrfs_mixed_space_info(space_info
))
4166 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4169 * if we're doing a data chunk, go ahead and make sure that
4170 * we keep a reasonable number of metadata chunks allocated in the
4173 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4174 fs_info
->data_chunk_allocations
++;
4175 if (!(fs_info
->data_chunk_allocations
%
4176 fs_info
->metadata_ratio
))
4177 force_metadata_allocation(fs_info
);
4181 * Check if we have enough space in SYSTEM chunk because we may need
4182 * to update devices.
4184 check_system_chunk(trans
, extent_root
, flags
);
4186 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4187 trans
->allocating_chunk
= false;
4189 spin_lock(&space_info
->lock
);
4190 if (ret
< 0 && ret
!= -ENOSPC
)
4193 space_info
->full
= 1;
4197 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4199 space_info
->chunk_alloc
= 0;
4200 spin_unlock(&space_info
->lock
);
4201 mutex_unlock(&fs_info
->chunk_mutex
);
4205 static int can_overcommit(struct btrfs_root
*root
,
4206 struct btrfs_space_info
*space_info
, u64 bytes
,
4207 enum btrfs_reserve_flush_enum flush
)
4209 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4210 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4215 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4216 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4219 * We only want to allow over committing if we have lots of actual space
4220 * free, but if we don't have enough space to handle the global reserve
4221 * space then we could end up having a real enospc problem when trying
4222 * to allocate a chunk or some other such important allocation.
4224 spin_lock(&global_rsv
->lock
);
4225 space_size
= calc_global_rsv_need_space(global_rsv
);
4226 spin_unlock(&global_rsv
->lock
);
4227 if (used
+ space_size
>= space_info
->total_bytes
)
4230 used
+= space_info
->bytes_may_use
;
4232 spin_lock(&root
->fs_info
->free_chunk_lock
);
4233 avail
= root
->fs_info
->free_chunk_space
;
4234 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4237 * If we have dup, raid1 or raid10 then only half of the free
4238 * space is actually useable. For raid56, the space info used
4239 * doesn't include the parity drive, so we don't have to
4242 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4243 BTRFS_BLOCK_GROUP_RAID1
|
4244 BTRFS_BLOCK_GROUP_RAID10
))
4248 * If we aren't flushing all things, let us overcommit up to
4249 * 1/2th of the space. If we can flush, don't let us overcommit
4250 * too much, let it overcommit up to 1/8 of the space.
4252 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4257 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4262 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4263 unsigned long nr_pages
, int nr_items
)
4265 struct super_block
*sb
= root
->fs_info
->sb
;
4267 if (down_read_trylock(&sb
->s_umount
)) {
4268 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4269 up_read(&sb
->s_umount
);
4272 * We needn't worry the filesystem going from r/w to r/o though
4273 * we don't acquire ->s_umount mutex, because the filesystem
4274 * should guarantee the delalloc inodes list be empty after
4275 * the filesystem is readonly(all dirty pages are written to
4278 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4279 if (!current
->journal_info
)
4280 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4284 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4289 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4290 nr
= (int)div64_u64(to_reclaim
, bytes
);
4296 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4299 * shrink metadata reservation for delalloc
4301 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4304 struct btrfs_block_rsv
*block_rsv
;
4305 struct btrfs_space_info
*space_info
;
4306 struct btrfs_trans_handle
*trans
;
4310 unsigned long nr_pages
;
4313 enum btrfs_reserve_flush_enum flush
;
4315 /* Calc the number of the pages we need flush for space reservation */
4316 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4317 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4319 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4320 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4321 space_info
= block_rsv
->space_info
;
4323 delalloc_bytes
= percpu_counter_sum_positive(
4324 &root
->fs_info
->delalloc_bytes
);
4325 if (delalloc_bytes
== 0) {
4329 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4334 while (delalloc_bytes
&& loops
< 3) {
4335 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4336 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4337 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4339 * We need to wait for the async pages to actually start before
4342 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4346 if (max_reclaim
<= nr_pages
)
4349 max_reclaim
-= nr_pages
;
4351 wait_event(root
->fs_info
->async_submit_wait
,
4352 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4356 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4358 flush
= BTRFS_RESERVE_NO_FLUSH
;
4359 spin_lock(&space_info
->lock
);
4360 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4361 spin_unlock(&space_info
->lock
);
4364 spin_unlock(&space_info
->lock
);
4367 if (wait_ordered
&& !trans
) {
4368 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4370 time_left
= schedule_timeout_killable(1);
4374 delalloc_bytes
= percpu_counter_sum_positive(
4375 &root
->fs_info
->delalloc_bytes
);
4380 * maybe_commit_transaction - possibly commit the transaction if its ok to
4381 * @root - the root we're allocating for
4382 * @bytes - the number of bytes we want to reserve
4383 * @force - force the commit
4385 * This will check to make sure that committing the transaction will actually
4386 * get us somewhere and then commit the transaction if it does. Otherwise it
4387 * will return -ENOSPC.
4389 static int may_commit_transaction(struct btrfs_root
*root
,
4390 struct btrfs_space_info
*space_info
,
4391 u64 bytes
, int force
)
4393 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4394 struct btrfs_trans_handle
*trans
;
4396 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4403 /* See if there is enough pinned space to make this reservation */
4404 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4409 * See if there is some space in the delayed insertion reservation for
4412 if (space_info
!= delayed_rsv
->space_info
)
4415 spin_lock(&delayed_rsv
->lock
);
4416 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4417 bytes
- delayed_rsv
->size
) >= 0) {
4418 spin_unlock(&delayed_rsv
->lock
);
4421 spin_unlock(&delayed_rsv
->lock
);
4424 trans
= btrfs_join_transaction(root
);
4428 return btrfs_commit_transaction(trans
, root
);
4432 FLUSH_DELAYED_ITEMS_NR
= 1,
4433 FLUSH_DELAYED_ITEMS
= 2,
4435 FLUSH_DELALLOC_WAIT
= 4,
4440 static int flush_space(struct btrfs_root
*root
,
4441 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4442 u64 orig_bytes
, int state
)
4444 struct btrfs_trans_handle
*trans
;
4449 case FLUSH_DELAYED_ITEMS_NR
:
4450 case FLUSH_DELAYED_ITEMS
:
4451 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4452 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4456 trans
= btrfs_join_transaction(root
);
4457 if (IS_ERR(trans
)) {
4458 ret
= PTR_ERR(trans
);
4461 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4462 btrfs_end_transaction(trans
, root
);
4464 case FLUSH_DELALLOC
:
4465 case FLUSH_DELALLOC_WAIT
:
4466 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4467 state
== FLUSH_DELALLOC_WAIT
);
4470 trans
= btrfs_join_transaction(root
);
4471 if (IS_ERR(trans
)) {
4472 ret
= PTR_ERR(trans
);
4475 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4476 btrfs_get_alloc_profile(root
, 0),
4477 CHUNK_ALLOC_NO_FORCE
);
4478 btrfs_end_transaction(trans
, root
);
4483 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4494 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4495 struct btrfs_space_info
*space_info
)
4501 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4503 spin_lock(&space_info
->lock
);
4504 if (can_overcommit(root
, space_info
, to_reclaim
,
4505 BTRFS_RESERVE_FLUSH_ALL
)) {
4510 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4511 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4512 space_info
->bytes_may_use
;
4513 if (can_overcommit(root
, space_info
, 1024 * 1024,
4514 BTRFS_RESERVE_FLUSH_ALL
))
4515 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4517 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4519 if (used
> expected
)
4520 to_reclaim
= used
- expected
;
4523 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4524 space_info
->bytes_reserved
);
4526 spin_unlock(&space_info
->lock
);
4531 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4532 struct btrfs_fs_info
*fs_info
, u64 used
)
4534 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4536 /* If we're just plain full then async reclaim just slows us down. */
4537 if (space_info
->bytes_used
>= thresh
)
4540 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4541 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4544 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4545 struct btrfs_fs_info
*fs_info
,
4550 spin_lock(&space_info
->lock
);
4552 * We run out of space and have not got any free space via flush_space,
4553 * so don't bother doing async reclaim.
4555 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4556 spin_unlock(&space_info
->lock
);
4560 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4561 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4562 space_info
->bytes_may_use
;
4563 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4564 spin_unlock(&space_info
->lock
);
4567 spin_unlock(&space_info
->lock
);
4572 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4574 struct btrfs_fs_info
*fs_info
;
4575 struct btrfs_space_info
*space_info
;
4579 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4580 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4582 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4587 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4589 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4590 to_reclaim
, flush_state
);
4592 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4595 } while (flush_state
< COMMIT_TRANS
);
4598 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4600 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4604 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4605 * @root - the root we're allocating for
4606 * @block_rsv - the block_rsv we're allocating for
4607 * @orig_bytes - the number of bytes we want
4608 * @flush - whether or not we can flush to make our reservation
4610 * This will reserve orgi_bytes number of bytes from the space info associated
4611 * with the block_rsv. If there is not enough space it will make an attempt to
4612 * flush out space to make room. It will do this by flushing delalloc if
4613 * possible or committing the transaction. If flush is 0 then no attempts to
4614 * regain reservations will be made and this will fail if there is not enough
4617 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4618 struct btrfs_block_rsv
*block_rsv
,
4620 enum btrfs_reserve_flush_enum flush
)
4622 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4624 u64 num_bytes
= orig_bytes
;
4625 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4627 bool flushing
= false;
4631 spin_lock(&space_info
->lock
);
4633 * We only want to wait if somebody other than us is flushing and we
4634 * are actually allowed to flush all things.
4636 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4637 space_info
->flush
) {
4638 spin_unlock(&space_info
->lock
);
4640 * If we have a trans handle we can't wait because the flusher
4641 * may have to commit the transaction, which would mean we would
4642 * deadlock since we are waiting for the flusher to finish, but
4643 * hold the current transaction open.
4645 if (current
->journal_info
)
4647 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4648 /* Must have been killed, return */
4652 spin_lock(&space_info
->lock
);
4656 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4657 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4658 space_info
->bytes_may_use
;
4661 * The idea here is that we've not already over-reserved the block group
4662 * then we can go ahead and save our reservation first and then start
4663 * flushing if we need to. Otherwise if we've already overcommitted
4664 * lets start flushing stuff first and then come back and try to make
4667 if (used
<= space_info
->total_bytes
) {
4668 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4669 space_info
->bytes_may_use
+= orig_bytes
;
4670 trace_btrfs_space_reservation(root
->fs_info
,
4671 "space_info", space_info
->flags
, orig_bytes
, 1);
4675 * Ok set num_bytes to orig_bytes since we aren't
4676 * overocmmitted, this way we only try and reclaim what
4679 num_bytes
= orig_bytes
;
4683 * Ok we're over committed, set num_bytes to the overcommitted
4684 * amount plus the amount of bytes that we need for this
4687 num_bytes
= used
- space_info
->total_bytes
+
4691 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4692 space_info
->bytes_may_use
+= orig_bytes
;
4693 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4694 space_info
->flags
, orig_bytes
,
4700 * Couldn't make our reservation, save our place so while we're trying
4701 * to reclaim space we can actually use it instead of somebody else
4702 * stealing it from us.
4704 * We make the other tasks wait for the flush only when we can flush
4707 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4709 space_info
->flush
= 1;
4710 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4713 * We will do the space reservation dance during log replay,
4714 * which means we won't have fs_info->fs_root set, so don't do
4715 * the async reclaim as we will panic.
4717 if (!root
->fs_info
->log_root_recovering
&&
4718 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4719 !work_busy(&root
->fs_info
->async_reclaim_work
))
4720 queue_work(system_unbound_wq
,
4721 &root
->fs_info
->async_reclaim_work
);
4723 spin_unlock(&space_info
->lock
);
4725 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4728 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4733 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4734 * would happen. So skip delalloc flush.
4736 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4737 (flush_state
== FLUSH_DELALLOC
||
4738 flush_state
== FLUSH_DELALLOC_WAIT
))
4739 flush_state
= ALLOC_CHUNK
;
4743 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4744 flush_state
< COMMIT_TRANS
)
4746 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4747 flush_state
<= COMMIT_TRANS
)
4751 if (ret
== -ENOSPC
&&
4752 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4753 struct btrfs_block_rsv
*global_rsv
=
4754 &root
->fs_info
->global_block_rsv
;
4756 if (block_rsv
!= global_rsv
&&
4757 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4761 trace_btrfs_space_reservation(root
->fs_info
,
4762 "space_info:enospc",
4763 space_info
->flags
, orig_bytes
, 1);
4765 spin_lock(&space_info
->lock
);
4766 space_info
->flush
= 0;
4767 wake_up_all(&space_info
->wait
);
4768 spin_unlock(&space_info
->lock
);
4773 static struct btrfs_block_rsv
*get_block_rsv(
4774 const struct btrfs_trans_handle
*trans
,
4775 const struct btrfs_root
*root
)
4777 struct btrfs_block_rsv
*block_rsv
= NULL
;
4779 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4780 block_rsv
= trans
->block_rsv
;
4782 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4783 block_rsv
= trans
->block_rsv
;
4785 if (root
== root
->fs_info
->uuid_root
)
4786 block_rsv
= trans
->block_rsv
;
4789 block_rsv
= root
->block_rsv
;
4792 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4797 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4801 spin_lock(&block_rsv
->lock
);
4802 if (block_rsv
->reserved
>= num_bytes
) {
4803 block_rsv
->reserved
-= num_bytes
;
4804 if (block_rsv
->reserved
< block_rsv
->size
)
4805 block_rsv
->full
= 0;
4808 spin_unlock(&block_rsv
->lock
);
4812 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4813 u64 num_bytes
, int update_size
)
4815 spin_lock(&block_rsv
->lock
);
4816 block_rsv
->reserved
+= num_bytes
;
4818 block_rsv
->size
+= num_bytes
;
4819 else if (block_rsv
->reserved
>= block_rsv
->size
)
4820 block_rsv
->full
= 1;
4821 spin_unlock(&block_rsv
->lock
);
4824 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4825 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4828 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4831 if (global_rsv
->space_info
!= dest
->space_info
)
4834 spin_lock(&global_rsv
->lock
);
4835 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4836 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4837 spin_unlock(&global_rsv
->lock
);
4840 global_rsv
->reserved
-= num_bytes
;
4841 if (global_rsv
->reserved
< global_rsv
->size
)
4842 global_rsv
->full
= 0;
4843 spin_unlock(&global_rsv
->lock
);
4845 block_rsv_add_bytes(dest
, num_bytes
, 1);
4849 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4850 struct btrfs_block_rsv
*block_rsv
,
4851 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4853 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4855 spin_lock(&block_rsv
->lock
);
4856 if (num_bytes
== (u64
)-1)
4857 num_bytes
= block_rsv
->size
;
4858 block_rsv
->size
-= num_bytes
;
4859 if (block_rsv
->reserved
>= block_rsv
->size
) {
4860 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4861 block_rsv
->reserved
= block_rsv
->size
;
4862 block_rsv
->full
= 1;
4866 spin_unlock(&block_rsv
->lock
);
4868 if (num_bytes
> 0) {
4870 spin_lock(&dest
->lock
);
4874 bytes_to_add
= dest
->size
- dest
->reserved
;
4875 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4876 dest
->reserved
+= bytes_to_add
;
4877 if (dest
->reserved
>= dest
->size
)
4879 num_bytes
-= bytes_to_add
;
4881 spin_unlock(&dest
->lock
);
4884 spin_lock(&space_info
->lock
);
4885 space_info
->bytes_may_use
-= num_bytes
;
4886 trace_btrfs_space_reservation(fs_info
, "space_info",
4887 space_info
->flags
, num_bytes
, 0);
4888 spin_unlock(&space_info
->lock
);
4893 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4894 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4898 ret
= block_rsv_use_bytes(src
, num_bytes
);
4902 block_rsv_add_bytes(dst
, num_bytes
, 1);
4906 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4908 memset(rsv
, 0, sizeof(*rsv
));
4909 spin_lock_init(&rsv
->lock
);
4913 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4914 unsigned short type
)
4916 struct btrfs_block_rsv
*block_rsv
;
4917 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4919 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4923 btrfs_init_block_rsv(block_rsv
, type
);
4924 block_rsv
->space_info
= __find_space_info(fs_info
,
4925 BTRFS_BLOCK_GROUP_METADATA
);
4929 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4930 struct btrfs_block_rsv
*rsv
)
4934 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4938 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
4943 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4944 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4945 enum btrfs_reserve_flush_enum flush
)
4952 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4954 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4961 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4962 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4970 spin_lock(&block_rsv
->lock
);
4971 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4972 if (block_rsv
->reserved
>= num_bytes
)
4974 spin_unlock(&block_rsv
->lock
);
4979 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4980 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4981 enum btrfs_reserve_flush_enum flush
)
4989 spin_lock(&block_rsv
->lock
);
4990 num_bytes
= min_reserved
;
4991 if (block_rsv
->reserved
>= num_bytes
)
4994 num_bytes
-= block_rsv
->reserved
;
4995 spin_unlock(&block_rsv
->lock
);
5000 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5002 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5009 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
5010 struct btrfs_block_rsv
*dst_rsv
,
5013 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5016 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5017 struct btrfs_block_rsv
*block_rsv
,
5020 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5021 if (global_rsv
== block_rsv
||
5022 block_rsv
->space_info
!= global_rsv
->space_info
)
5024 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5029 * helper to calculate size of global block reservation.
5030 * the desired value is sum of space used by extent tree,
5031 * checksum tree and root tree
5033 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5035 struct btrfs_space_info
*sinfo
;
5039 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5041 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5042 spin_lock(&sinfo
->lock
);
5043 data_used
= sinfo
->bytes_used
;
5044 spin_unlock(&sinfo
->lock
);
5046 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5047 spin_lock(&sinfo
->lock
);
5048 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5050 meta_used
= sinfo
->bytes_used
;
5051 spin_unlock(&sinfo
->lock
);
5053 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5055 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5057 if (num_bytes
* 3 > meta_used
)
5058 num_bytes
= div_u64(meta_used
, 3);
5060 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5063 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5065 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5066 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5069 num_bytes
= calc_global_metadata_size(fs_info
);
5071 spin_lock(&sinfo
->lock
);
5072 spin_lock(&block_rsv
->lock
);
5074 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
5076 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5077 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5078 sinfo
->bytes_may_use
;
5080 if (sinfo
->total_bytes
> num_bytes
) {
5081 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5082 block_rsv
->reserved
+= num_bytes
;
5083 sinfo
->bytes_may_use
+= num_bytes
;
5084 trace_btrfs_space_reservation(fs_info
, "space_info",
5085 sinfo
->flags
, num_bytes
, 1);
5088 if (block_rsv
->reserved
>= block_rsv
->size
) {
5089 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5090 sinfo
->bytes_may_use
-= num_bytes
;
5091 trace_btrfs_space_reservation(fs_info
, "space_info",
5092 sinfo
->flags
, num_bytes
, 0);
5093 block_rsv
->reserved
= block_rsv
->size
;
5094 block_rsv
->full
= 1;
5097 spin_unlock(&block_rsv
->lock
);
5098 spin_unlock(&sinfo
->lock
);
5101 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5103 struct btrfs_space_info
*space_info
;
5105 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5106 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5108 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5109 fs_info
->global_block_rsv
.space_info
= space_info
;
5110 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5111 fs_info
->trans_block_rsv
.space_info
= space_info
;
5112 fs_info
->empty_block_rsv
.space_info
= space_info
;
5113 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5115 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5116 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5117 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5118 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5119 if (fs_info
->quota_root
)
5120 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5121 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5123 update_global_block_rsv(fs_info
);
5126 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5128 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5130 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5131 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5132 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5133 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5134 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5135 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5136 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5137 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5140 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5141 struct btrfs_root
*root
)
5143 if (!trans
->block_rsv
)
5146 if (!trans
->bytes_reserved
)
5149 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5150 trans
->transid
, trans
->bytes_reserved
, 0);
5151 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5152 trans
->bytes_reserved
= 0;
5155 /* Can only return 0 or -ENOSPC */
5156 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5157 struct inode
*inode
)
5159 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5160 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5161 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5164 * We need to hold space in order to delete our orphan item once we've
5165 * added it, so this takes the reservation so we can release it later
5166 * when we are truly done with the orphan item.
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
, 1);
5171 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5174 void btrfs_orphan_release_metadata(struct inode
*inode
)
5176 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5177 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5178 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5179 btrfs_ino(inode
), num_bytes
, 0);
5180 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5184 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5185 * root: the root of the parent directory
5186 * rsv: block reservation
5187 * items: the number of items that we need do reservation
5188 * qgroup_reserved: used to return the reserved size in qgroup
5190 * This function is used to reserve the space for snapshot/subvolume
5191 * creation and deletion. Those operations are different with the
5192 * common file/directory operations, they change two fs/file trees
5193 * and root tree, the number of items that the qgroup reserves is
5194 * different with the free space reservation. So we can not use
5195 * the space reseravtion mechanism in start_transaction().
5197 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5198 struct btrfs_block_rsv
*rsv
,
5200 u64
*qgroup_reserved
,
5201 bool use_global_rsv
)
5205 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5207 if (root
->fs_info
->quota_enabled
) {
5208 /* One for parent inode, two for dir entries */
5209 num_bytes
= 3 * root
->nodesize
;
5210 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
5217 *qgroup_reserved
= num_bytes
;
5219 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5220 rsv
->space_info
= __find_space_info(root
->fs_info
,
5221 BTRFS_BLOCK_GROUP_METADATA
);
5222 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5223 BTRFS_RESERVE_FLUSH_ALL
);
5225 if (ret
== -ENOSPC
&& use_global_rsv
)
5226 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5229 if (*qgroup_reserved
)
5230 btrfs_qgroup_free(root
, *qgroup_reserved
);
5236 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5237 struct btrfs_block_rsv
*rsv
,
5238 u64 qgroup_reserved
)
5240 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5241 if (qgroup_reserved
)
5242 btrfs_qgroup_free(root
, qgroup_reserved
);
5246 * drop_outstanding_extent - drop an outstanding extent
5247 * @inode: the inode we're dropping the extent for
5248 * @num_bytes: the number of bytes we're relaseing.
5250 * This is called when we are freeing up an outstanding extent, either called
5251 * after an error or after an extent is written. This will return the number of
5252 * reserved extents that need to be freed. This must be called with
5253 * BTRFS_I(inode)->lock held.
5255 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5257 unsigned drop_inode_space
= 0;
5258 unsigned dropped_extents
= 0;
5259 unsigned num_extents
= 0;
5261 num_extents
= (unsigned)div64_u64(num_bytes
+
5262 BTRFS_MAX_EXTENT_SIZE
- 1,
5263 BTRFS_MAX_EXTENT_SIZE
);
5264 ASSERT(num_extents
);
5265 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5266 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5268 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5269 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5270 &BTRFS_I(inode
)->runtime_flags
))
5271 drop_inode_space
= 1;
5274 * If we have more or the same amount of outsanding extents than we have
5275 * reserved then we need to leave the reserved extents count alone.
5277 if (BTRFS_I(inode
)->outstanding_extents
>=
5278 BTRFS_I(inode
)->reserved_extents
)
5279 return drop_inode_space
;
5281 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5282 BTRFS_I(inode
)->outstanding_extents
;
5283 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5284 return dropped_extents
+ drop_inode_space
;
5288 * calc_csum_metadata_size - return the amount of metada space that must be
5289 * reserved/free'd for the given bytes.
5290 * @inode: the inode we're manipulating
5291 * @num_bytes: the number of bytes in question
5292 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5294 * This adjusts the number of csum_bytes in the inode and then returns the
5295 * correct amount of metadata that must either be reserved or freed. We
5296 * calculate how many checksums we can fit into one leaf and then divide the
5297 * number of bytes that will need to be checksumed by this value to figure out
5298 * how many checksums will be required. If we are adding bytes then the number
5299 * may go up and we will return the number of additional bytes that must be
5300 * reserved. If it is going down we will return the number of bytes that must
5303 * This must be called with BTRFS_I(inode)->lock held.
5305 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5308 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5309 u64 old_csums
, num_csums
;
5311 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5312 BTRFS_I(inode
)->csum_bytes
== 0)
5315 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5317 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5319 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5320 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5322 /* No change, no need to reserve more */
5323 if (old_csums
== num_csums
)
5327 return btrfs_calc_trans_metadata_size(root
,
5328 num_csums
- old_csums
);
5330 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5333 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5335 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5336 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5339 unsigned nr_extents
= 0;
5340 int extra_reserve
= 0;
5341 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5343 bool delalloc_lock
= true;
5347 /* If we are a free space inode we need to not flush since we will be in
5348 * the middle of a transaction commit. We also don't need the delalloc
5349 * mutex since we won't race with anybody. We need this mostly to make
5350 * lockdep shut its filthy mouth.
5352 if (btrfs_is_free_space_inode(inode
)) {
5353 flush
= BTRFS_RESERVE_NO_FLUSH
;
5354 delalloc_lock
= false;
5357 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5358 btrfs_transaction_in_commit(root
->fs_info
))
5359 schedule_timeout(1);
5362 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5364 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5366 spin_lock(&BTRFS_I(inode
)->lock
);
5367 nr_extents
= (unsigned)div64_u64(num_bytes
+
5368 BTRFS_MAX_EXTENT_SIZE
- 1,
5369 BTRFS_MAX_EXTENT_SIZE
);
5370 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5373 if (BTRFS_I(inode
)->outstanding_extents
>
5374 BTRFS_I(inode
)->reserved_extents
)
5375 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5376 BTRFS_I(inode
)->reserved_extents
;
5379 * Add an item to reserve for updating the inode when we complete the
5382 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5383 &BTRFS_I(inode
)->runtime_flags
)) {
5388 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5389 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5390 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5391 spin_unlock(&BTRFS_I(inode
)->lock
);
5393 if (root
->fs_info
->quota_enabled
) {
5394 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
5395 nr_extents
* root
->nodesize
);
5400 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5401 if (unlikely(ret
)) {
5402 if (root
->fs_info
->quota_enabled
)
5403 btrfs_qgroup_free(root
, num_bytes
+
5404 nr_extents
* root
->nodesize
);
5408 spin_lock(&BTRFS_I(inode
)->lock
);
5409 if (extra_reserve
) {
5410 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5411 &BTRFS_I(inode
)->runtime_flags
);
5414 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5415 spin_unlock(&BTRFS_I(inode
)->lock
);
5418 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5421 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5422 btrfs_ino(inode
), to_reserve
, 1);
5423 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5428 spin_lock(&BTRFS_I(inode
)->lock
);
5429 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5431 * If the inodes csum_bytes is the same as the original
5432 * csum_bytes then we know we haven't raced with any free()ers
5433 * so we can just reduce our inodes csum bytes and carry on.
5435 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5436 calc_csum_metadata_size(inode
, num_bytes
, 0);
5438 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5442 * This is tricky, but first we need to figure out how much we
5443 * free'd from any free-ers that occured during this
5444 * reservation, so we reset ->csum_bytes to the csum_bytes
5445 * before we dropped our lock, and then call the free for the
5446 * number of bytes that were freed while we were trying our
5449 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5450 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5451 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5455 * Now we need to see how much we would have freed had we not
5456 * been making this reservation and our ->csum_bytes were not
5457 * artificially inflated.
5459 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5460 bytes
= csum_bytes
- orig_csum_bytes
;
5461 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5464 * Now reset ->csum_bytes to what it should be. If bytes is
5465 * more than to_free then we would have free'd more space had we
5466 * not had an artificially high ->csum_bytes, so we need to free
5467 * the remainder. If bytes is the same or less then we don't
5468 * need to do anything, the other free-ers did the correct
5471 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5472 if (bytes
> to_free
)
5473 to_free
= bytes
- to_free
;
5477 spin_unlock(&BTRFS_I(inode
)->lock
);
5479 if (root
->fs_info
->quota_enabled
)
5480 btrfs_qgroup_free(root
, dropped
* root
->nodesize
);
5481 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5485 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5486 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5487 btrfs_ino(inode
), to_free
, 0);
5490 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5495 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5496 * @inode: the inode to release the reservation for
5497 * @num_bytes: the number of bytes we're releasing
5499 * This will release the metadata reservation for an inode. This can be called
5500 * once we complete IO for a given set of bytes to release their metadata
5503 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5505 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5509 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5510 spin_lock(&BTRFS_I(inode
)->lock
);
5511 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5514 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5515 spin_unlock(&BTRFS_I(inode
)->lock
);
5517 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5519 if (btrfs_test_is_dummy_root(root
))
5522 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5523 btrfs_ino(inode
), to_free
, 0);
5524 if (root
->fs_info
->quota_enabled
) {
5525 btrfs_qgroup_free(root
, num_bytes
+
5526 dropped
* root
->nodesize
);
5529 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5534 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5535 * @inode: inode we're writing to
5536 * @num_bytes: the number of bytes we want to allocate
5538 * This will do the following things
5540 * o reserve space in the data space info for num_bytes
5541 * o reserve space in the metadata space info based on number of outstanding
5542 * extents and how much csums will be needed
5543 * o add to the inodes ->delalloc_bytes
5544 * o add it to the fs_info's delalloc inodes list.
5546 * This will return 0 for success and -ENOSPC if there is no space left.
5548 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5552 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5556 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5558 btrfs_free_reserved_data_space(inode
, num_bytes
);
5566 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5567 * @inode: inode we're releasing space for
5568 * @num_bytes: the number of bytes we want to free up
5570 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5571 * called in the case that we don't need the metadata AND data reservations
5572 * anymore. So if there is an error or we insert an inline extent.
5574 * This function will release the metadata space that was not used and will
5575 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5576 * list if there are no delalloc bytes left.
5578 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5580 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5581 btrfs_free_reserved_data_space(inode
, num_bytes
);
5584 static int update_block_group(struct btrfs_trans_handle
*trans
,
5585 struct btrfs_root
*root
, u64 bytenr
,
5586 u64 num_bytes
, int alloc
)
5588 struct btrfs_block_group_cache
*cache
= NULL
;
5589 struct btrfs_fs_info
*info
= root
->fs_info
;
5590 u64 total
= num_bytes
;
5595 /* block accounting for super block */
5596 spin_lock(&info
->delalloc_root_lock
);
5597 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5599 old_val
+= num_bytes
;
5601 old_val
-= num_bytes
;
5602 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5603 spin_unlock(&info
->delalloc_root_lock
);
5606 cache
= btrfs_lookup_block_group(info
, bytenr
);
5609 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5610 BTRFS_BLOCK_GROUP_RAID1
|
5611 BTRFS_BLOCK_GROUP_RAID10
))
5616 * If this block group has free space cache written out, we
5617 * need to make sure to load it if we are removing space. This
5618 * is because we need the unpinning stage to actually add the
5619 * space back to the block group, otherwise we will leak space.
5621 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5622 cache_block_group(cache
, 1);
5624 byte_in_group
= bytenr
- cache
->key
.objectid
;
5625 WARN_ON(byte_in_group
> cache
->key
.offset
);
5627 spin_lock(&cache
->space_info
->lock
);
5628 spin_lock(&cache
->lock
);
5630 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5631 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5632 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5634 old_val
= btrfs_block_group_used(&cache
->item
);
5635 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5637 old_val
+= num_bytes
;
5638 btrfs_set_block_group_used(&cache
->item
, old_val
);
5639 cache
->reserved
-= num_bytes
;
5640 cache
->space_info
->bytes_reserved
-= num_bytes
;
5641 cache
->space_info
->bytes_used
+= num_bytes
;
5642 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5643 spin_unlock(&cache
->lock
);
5644 spin_unlock(&cache
->space_info
->lock
);
5646 old_val
-= num_bytes
;
5647 btrfs_set_block_group_used(&cache
->item
, old_val
);
5648 cache
->pinned
+= num_bytes
;
5649 cache
->space_info
->bytes_pinned
+= num_bytes
;
5650 cache
->space_info
->bytes_used
-= num_bytes
;
5651 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5652 spin_unlock(&cache
->lock
);
5653 spin_unlock(&cache
->space_info
->lock
);
5655 set_extent_dirty(info
->pinned_extents
,
5656 bytenr
, bytenr
+ num_bytes
- 1,
5657 GFP_NOFS
| __GFP_NOFAIL
);
5659 * No longer have used bytes in this block group, queue
5663 spin_lock(&info
->unused_bgs_lock
);
5664 if (list_empty(&cache
->bg_list
)) {
5665 btrfs_get_block_group(cache
);
5666 list_add_tail(&cache
->bg_list
,
5669 spin_unlock(&info
->unused_bgs_lock
);
5673 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5674 if (list_empty(&cache
->dirty_list
)) {
5675 list_add_tail(&cache
->dirty_list
,
5676 &trans
->transaction
->dirty_bgs
);
5677 trans
->transaction
->num_dirty_bgs
++;
5678 btrfs_get_block_group(cache
);
5680 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5682 btrfs_put_block_group(cache
);
5684 bytenr
+= num_bytes
;
5689 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5691 struct btrfs_block_group_cache
*cache
;
5694 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5695 bytenr
= root
->fs_info
->first_logical_byte
;
5696 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5698 if (bytenr
< (u64
)-1)
5701 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5705 bytenr
= cache
->key
.objectid
;
5706 btrfs_put_block_group(cache
);
5711 static int pin_down_extent(struct btrfs_root
*root
,
5712 struct btrfs_block_group_cache
*cache
,
5713 u64 bytenr
, u64 num_bytes
, int reserved
)
5715 spin_lock(&cache
->space_info
->lock
);
5716 spin_lock(&cache
->lock
);
5717 cache
->pinned
+= num_bytes
;
5718 cache
->space_info
->bytes_pinned
+= num_bytes
;
5720 cache
->reserved
-= num_bytes
;
5721 cache
->space_info
->bytes_reserved
-= num_bytes
;
5723 spin_unlock(&cache
->lock
);
5724 spin_unlock(&cache
->space_info
->lock
);
5726 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5727 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5729 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5734 * this function must be called within transaction
5736 int btrfs_pin_extent(struct btrfs_root
*root
,
5737 u64 bytenr
, u64 num_bytes
, int reserved
)
5739 struct btrfs_block_group_cache
*cache
;
5741 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5742 BUG_ON(!cache
); /* Logic error */
5744 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5746 btrfs_put_block_group(cache
);
5751 * this function must be called within transaction
5753 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5754 u64 bytenr
, u64 num_bytes
)
5756 struct btrfs_block_group_cache
*cache
;
5759 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5764 * pull in the free space cache (if any) so that our pin
5765 * removes the free space from the cache. We have load_only set
5766 * to one because the slow code to read in the free extents does check
5767 * the pinned extents.
5769 cache_block_group(cache
, 1);
5771 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5773 /* remove us from the free space cache (if we're there at all) */
5774 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5775 btrfs_put_block_group(cache
);
5779 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5782 struct btrfs_block_group_cache
*block_group
;
5783 struct btrfs_caching_control
*caching_ctl
;
5785 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5789 cache_block_group(block_group
, 0);
5790 caching_ctl
= get_caching_control(block_group
);
5794 BUG_ON(!block_group_cache_done(block_group
));
5795 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5797 mutex_lock(&caching_ctl
->mutex
);
5799 if (start
>= caching_ctl
->progress
) {
5800 ret
= add_excluded_extent(root
, start
, num_bytes
);
5801 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5802 ret
= btrfs_remove_free_space(block_group
,
5805 num_bytes
= caching_ctl
->progress
- start
;
5806 ret
= btrfs_remove_free_space(block_group
,
5811 num_bytes
= (start
+ num_bytes
) -
5812 caching_ctl
->progress
;
5813 start
= caching_ctl
->progress
;
5814 ret
= add_excluded_extent(root
, start
, num_bytes
);
5817 mutex_unlock(&caching_ctl
->mutex
);
5818 put_caching_control(caching_ctl
);
5820 btrfs_put_block_group(block_group
);
5824 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5825 struct extent_buffer
*eb
)
5827 struct btrfs_file_extent_item
*item
;
5828 struct btrfs_key key
;
5832 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5835 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5836 btrfs_item_key_to_cpu(eb
, &key
, i
);
5837 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5839 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5840 found_type
= btrfs_file_extent_type(eb
, item
);
5841 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5843 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5845 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5846 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5847 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5854 * btrfs_update_reserved_bytes - update the block_group and space info counters
5855 * @cache: The cache we are manipulating
5856 * @num_bytes: The number of bytes in question
5857 * @reserve: One of the reservation enums
5858 * @delalloc: The blocks are allocated for the delalloc write
5860 * This is called by the allocator when it reserves space, or by somebody who is
5861 * freeing space that was never actually used on disk. For example if you
5862 * reserve some space for a new leaf in transaction A and before transaction A
5863 * commits you free that leaf, you call this with reserve set to 0 in order to
5864 * clear the reservation.
5866 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5867 * ENOSPC accounting. For data we handle the reservation through clearing the
5868 * delalloc bits in the io_tree. We have to do this since we could end up
5869 * allocating less disk space for the amount of data we have reserved in the
5870 * case of compression.
5872 * If this is a reservation and the block group has become read only we cannot
5873 * make the reservation and return -EAGAIN, otherwise this function always
5876 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5877 u64 num_bytes
, int reserve
, int delalloc
)
5879 struct btrfs_space_info
*space_info
= cache
->space_info
;
5882 spin_lock(&space_info
->lock
);
5883 spin_lock(&cache
->lock
);
5884 if (reserve
!= RESERVE_FREE
) {
5888 cache
->reserved
+= num_bytes
;
5889 space_info
->bytes_reserved
+= num_bytes
;
5890 if (reserve
== RESERVE_ALLOC
) {
5891 trace_btrfs_space_reservation(cache
->fs_info
,
5892 "space_info", space_info
->flags
,
5894 space_info
->bytes_may_use
-= num_bytes
;
5898 cache
->delalloc_bytes
+= num_bytes
;
5902 space_info
->bytes_readonly
+= num_bytes
;
5903 cache
->reserved
-= num_bytes
;
5904 space_info
->bytes_reserved
-= num_bytes
;
5907 cache
->delalloc_bytes
-= num_bytes
;
5909 spin_unlock(&cache
->lock
);
5910 spin_unlock(&space_info
->lock
);
5914 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5915 struct btrfs_root
*root
)
5917 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5918 struct btrfs_caching_control
*next
;
5919 struct btrfs_caching_control
*caching_ctl
;
5920 struct btrfs_block_group_cache
*cache
;
5922 down_write(&fs_info
->commit_root_sem
);
5924 list_for_each_entry_safe(caching_ctl
, next
,
5925 &fs_info
->caching_block_groups
, list
) {
5926 cache
= caching_ctl
->block_group
;
5927 if (block_group_cache_done(cache
)) {
5928 cache
->last_byte_to_unpin
= (u64
)-1;
5929 list_del_init(&caching_ctl
->list
);
5930 put_caching_control(caching_ctl
);
5932 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5936 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5937 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5939 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5941 up_write(&fs_info
->commit_root_sem
);
5943 update_global_block_rsv(fs_info
);
5946 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
5947 const bool return_free_space
)
5949 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5950 struct btrfs_block_group_cache
*cache
= NULL
;
5951 struct btrfs_space_info
*space_info
;
5952 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5956 while (start
<= end
) {
5959 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5961 btrfs_put_block_group(cache
);
5962 cache
= btrfs_lookup_block_group(fs_info
, start
);
5963 BUG_ON(!cache
); /* Logic error */
5966 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5967 len
= min(len
, end
+ 1 - start
);
5969 if (start
< cache
->last_byte_to_unpin
) {
5970 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5971 if (return_free_space
)
5972 btrfs_add_free_space(cache
, start
, len
);
5976 space_info
= cache
->space_info
;
5978 spin_lock(&space_info
->lock
);
5979 spin_lock(&cache
->lock
);
5980 cache
->pinned
-= len
;
5981 space_info
->bytes_pinned
-= len
;
5982 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
5984 space_info
->bytes_readonly
+= len
;
5987 spin_unlock(&cache
->lock
);
5988 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5989 spin_lock(&global_rsv
->lock
);
5990 if (!global_rsv
->full
) {
5991 len
= min(len
, global_rsv
->size
-
5992 global_rsv
->reserved
);
5993 global_rsv
->reserved
+= len
;
5994 space_info
->bytes_may_use
+= len
;
5995 if (global_rsv
->reserved
>= global_rsv
->size
)
5996 global_rsv
->full
= 1;
5998 spin_unlock(&global_rsv
->lock
);
6000 spin_unlock(&space_info
->lock
);
6004 btrfs_put_block_group(cache
);
6008 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6009 struct btrfs_root
*root
)
6011 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6012 struct extent_io_tree
*unpin
;
6020 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6021 unpin
= &fs_info
->freed_extents
[1];
6023 unpin
= &fs_info
->freed_extents
[0];
6026 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6027 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6028 EXTENT_DIRTY
, NULL
);
6030 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6034 if (btrfs_test_opt(root
, DISCARD
))
6035 ret
= btrfs_discard_extent(root
, start
,
6036 end
+ 1 - start
, NULL
);
6038 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
6039 unpin_extent_range(root
, start
, end
, true);
6040 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6047 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6048 u64 owner
, u64 root_objectid
)
6050 struct btrfs_space_info
*space_info
;
6053 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6054 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6055 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6057 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6059 flags
= BTRFS_BLOCK_GROUP_DATA
;
6062 space_info
= __find_space_info(fs_info
, flags
);
6063 BUG_ON(!space_info
); /* Logic bug */
6064 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6068 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6069 struct btrfs_root
*root
,
6070 u64 bytenr
, u64 num_bytes
, u64 parent
,
6071 u64 root_objectid
, u64 owner_objectid
,
6072 u64 owner_offset
, int refs_to_drop
,
6073 struct btrfs_delayed_extent_op
*extent_op
,
6076 struct btrfs_key key
;
6077 struct btrfs_path
*path
;
6078 struct btrfs_fs_info
*info
= root
->fs_info
;
6079 struct btrfs_root
*extent_root
= info
->extent_root
;
6080 struct extent_buffer
*leaf
;
6081 struct btrfs_extent_item
*ei
;
6082 struct btrfs_extent_inline_ref
*iref
;
6085 int extent_slot
= 0;
6086 int found_extent
= 0;
6091 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_SUB_EXCL
;
6092 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6095 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
6098 path
= btrfs_alloc_path();
6103 path
->leave_spinning
= 1;
6105 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6106 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6109 skinny_metadata
= 0;
6111 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6112 bytenr
, num_bytes
, parent
,
6113 root_objectid
, owner_objectid
,
6116 extent_slot
= path
->slots
[0];
6117 while (extent_slot
>= 0) {
6118 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6120 if (key
.objectid
!= bytenr
)
6122 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6123 key
.offset
== num_bytes
) {
6127 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6128 key
.offset
== owner_objectid
) {
6132 if (path
->slots
[0] - extent_slot
> 5)
6136 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6137 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6138 if (found_extent
&& item_size
< sizeof(*ei
))
6141 if (!found_extent
) {
6143 ret
= remove_extent_backref(trans
, extent_root
, path
,
6145 is_data
, &last_ref
);
6147 btrfs_abort_transaction(trans
, extent_root
, ret
);
6150 btrfs_release_path(path
);
6151 path
->leave_spinning
= 1;
6153 key
.objectid
= bytenr
;
6154 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6155 key
.offset
= num_bytes
;
6157 if (!is_data
&& skinny_metadata
) {
6158 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6159 key
.offset
= owner_objectid
;
6162 ret
= btrfs_search_slot(trans
, extent_root
,
6164 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6166 * Couldn't find our skinny metadata item,
6167 * see if we have ye olde extent item.
6170 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6172 if (key
.objectid
== bytenr
&&
6173 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6174 key
.offset
== num_bytes
)
6178 if (ret
> 0 && skinny_metadata
) {
6179 skinny_metadata
= false;
6180 key
.objectid
= bytenr
;
6181 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6182 key
.offset
= num_bytes
;
6183 btrfs_release_path(path
);
6184 ret
= btrfs_search_slot(trans
, extent_root
,
6189 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6192 btrfs_print_leaf(extent_root
,
6196 btrfs_abort_transaction(trans
, extent_root
, ret
);
6199 extent_slot
= path
->slots
[0];
6201 } else if (WARN_ON(ret
== -ENOENT
)) {
6202 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6204 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6205 bytenr
, parent
, root_objectid
, owner_objectid
,
6207 btrfs_abort_transaction(trans
, extent_root
, ret
);
6210 btrfs_abort_transaction(trans
, extent_root
, ret
);
6214 leaf
= path
->nodes
[0];
6215 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6216 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6217 if (item_size
< sizeof(*ei
)) {
6218 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6219 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6222 btrfs_abort_transaction(trans
, extent_root
, ret
);
6226 btrfs_release_path(path
);
6227 path
->leave_spinning
= 1;
6229 key
.objectid
= bytenr
;
6230 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6231 key
.offset
= num_bytes
;
6233 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6236 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6238 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6241 btrfs_abort_transaction(trans
, extent_root
, ret
);
6245 extent_slot
= path
->slots
[0];
6246 leaf
= path
->nodes
[0];
6247 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6250 BUG_ON(item_size
< sizeof(*ei
));
6251 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6252 struct btrfs_extent_item
);
6253 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6254 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6255 struct btrfs_tree_block_info
*bi
;
6256 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6257 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6258 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6261 refs
= btrfs_extent_refs(leaf
, ei
);
6262 if (refs
< refs_to_drop
) {
6263 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6264 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6266 btrfs_abort_transaction(trans
, extent_root
, ret
);
6269 refs
-= refs_to_drop
;
6272 type
= BTRFS_QGROUP_OPER_SUB_SHARED
;
6274 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6276 * In the case of inline back ref, reference count will
6277 * be updated by remove_extent_backref
6280 BUG_ON(!found_extent
);
6282 btrfs_set_extent_refs(leaf
, ei
, refs
);
6283 btrfs_mark_buffer_dirty(leaf
);
6286 ret
= remove_extent_backref(trans
, extent_root
, path
,
6288 is_data
, &last_ref
);
6290 btrfs_abort_transaction(trans
, extent_root
, ret
);
6294 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6298 BUG_ON(is_data
&& refs_to_drop
!=
6299 extent_data_ref_count(root
, path
, iref
));
6301 BUG_ON(path
->slots
[0] != extent_slot
);
6303 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6304 path
->slots
[0] = extent_slot
;
6310 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6313 btrfs_abort_transaction(trans
, extent_root
, ret
);
6316 btrfs_release_path(path
);
6319 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6321 btrfs_abort_transaction(trans
, extent_root
, ret
);
6326 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6328 btrfs_abort_transaction(trans
, extent_root
, ret
);
6332 btrfs_release_path(path
);
6334 /* Deal with the quota accounting */
6335 if (!ret
&& last_ref
&& !no_quota
) {
6338 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
6339 type
== BTRFS_QGROUP_OPER_SUB_SHARED
)
6342 ret
= btrfs_qgroup_record_ref(trans
, info
, root_objectid
,
6343 bytenr
, num_bytes
, type
,
6347 btrfs_free_path(path
);
6352 * when we free an block, it is possible (and likely) that we free the last
6353 * delayed ref for that extent as well. This searches the delayed ref tree for
6354 * a given extent, and if there are no other delayed refs to be processed, it
6355 * removes it from the tree.
6357 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6358 struct btrfs_root
*root
, u64 bytenr
)
6360 struct btrfs_delayed_ref_head
*head
;
6361 struct btrfs_delayed_ref_root
*delayed_refs
;
6364 delayed_refs
= &trans
->transaction
->delayed_refs
;
6365 spin_lock(&delayed_refs
->lock
);
6366 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6368 goto out_delayed_unlock
;
6370 spin_lock(&head
->lock
);
6371 if (rb_first(&head
->ref_root
))
6374 if (head
->extent_op
) {
6375 if (!head
->must_insert_reserved
)
6377 btrfs_free_delayed_extent_op(head
->extent_op
);
6378 head
->extent_op
= NULL
;
6382 * waiting for the lock here would deadlock. If someone else has it
6383 * locked they are already in the process of dropping it anyway
6385 if (!mutex_trylock(&head
->mutex
))
6389 * at this point we have a head with no other entries. Go
6390 * ahead and process it.
6392 head
->node
.in_tree
= 0;
6393 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6395 atomic_dec(&delayed_refs
->num_entries
);
6398 * we don't take a ref on the node because we're removing it from the
6399 * tree, so we just steal the ref the tree was holding.
6401 delayed_refs
->num_heads
--;
6402 if (head
->processing
== 0)
6403 delayed_refs
->num_heads_ready
--;
6404 head
->processing
= 0;
6405 spin_unlock(&head
->lock
);
6406 spin_unlock(&delayed_refs
->lock
);
6408 BUG_ON(head
->extent_op
);
6409 if (head
->must_insert_reserved
)
6412 mutex_unlock(&head
->mutex
);
6413 btrfs_put_delayed_ref(&head
->node
);
6416 spin_unlock(&head
->lock
);
6419 spin_unlock(&delayed_refs
->lock
);
6423 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6424 struct btrfs_root
*root
,
6425 struct extent_buffer
*buf
,
6426 u64 parent
, int last_ref
)
6431 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6432 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6433 buf
->start
, buf
->len
,
6434 parent
, root
->root_key
.objectid
,
6435 btrfs_header_level(buf
),
6436 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6437 BUG_ON(ret
); /* -ENOMEM */
6443 if (btrfs_header_generation(buf
) == trans
->transid
) {
6444 struct btrfs_block_group_cache
*cache
;
6446 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6447 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6452 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6454 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6455 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6456 btrfs_put_block_group(cache
);
6460 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6462 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6463 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6464 btrfs_put_block_group(cache
);
6465 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6470 add_pinned_bytes(root
->fs_info
, buf
->len
,
6471 btrfs_header_level(buf
),
6472 root
->root_key
.objectid
);
6475 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6478 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6481 /* Can return -ENOMEM */
6482 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6483 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6484 u64 owner
, u64 offset
, int no_quota
)
6487 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6489 if (btrfs_test_is_dummy_root(root
))
6492 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6495 * tree log blocks never actually go into the extent allocation
6496 * tree, just update pinning info and exit early.
6498 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6499 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6500 /* unlocks the pinned mutex */
6501 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6503 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6504 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6506 parent
, root_objectid
, (int)owner
,
6507 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6509 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6511 parent
, root_objectid
, owner
,
6512 offset
, BTRFS_DROP_DELAYED_REF
,
6519 * when we wait for progress in the block group caching, its because
6520 * our allocation attempt failed at least once. So, we must sleep
6521 * and let some progress happen before we try again.
6523 * This function will sleep at least once waiting for new free space to
6524 * show up, and then it will check the block group free space numbers
6525 * for our min num_bytes. Another option is to have it go ahead
6526 * and look in the rbtree for a free extent of a given size, but this
6529 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6530 * any of the information in this block group.
6532 static noinline
void
6533 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6536 struct btrfs_caching_control
*caching_ctl
;
6538 caching_ctl
= get_caching_control(cache
);
6542 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6543 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6545 put_caching_control(caching_ctl
);
6549 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6551 struct btrfs_caching_control
*caching_ctl
;
6554 caching_ctl
= get_caching_control(cache
);
6556 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6558 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6559 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6561 put_caching_control(caching_ctl
);
6565 int __get_raid_index(u64 flags
)
6567 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6568 return BTRFS_RAID_RAID10
;
6569 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6570 return BTRFS_RAID_RAID1
;
6571 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6572 return BTRFS_RAID_DUP
;
6573 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6574 return BTRFS_RAID_RAID0
;
6575 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6576 return BTRFS_RAID_RAID5
;
6577 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6578 return BTRFS_RAID_RAID6
;
6580 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6583 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6585 return __get_raid_index(cache
->flags
);
6588 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6589 [BTRFS_RAID_RAID10
] = "raid10",
6590 [BTRFS_RAID_RAID1
] = "raid1",
6591 [BTRFS_RAID_DUP
] = "dup",
6592 [BTRFS_RAID_RAID0
] = "raid0",
6593 [BTRFS_RAID_SINGLE
] = "single",
6594 [BTRFS_RAID_RAID5
] = "raid5",
6595 [BTRFS_RAID_RAID6
] = "raid6",
6598 static const char *get_raid_name(enum btrfs_raid_types type
)
6600 if (type
>= BTRFS_NR_RAID_TYPES
)
6603 return btrfs_raid_type_names
[type
];
6606 enum btrfs_loop_type
{
6607 LOOP_CACHING_NOWAIT
= 0,
6608 LOOP_CACHING_WAIT
= 1,
6609 LOOP_ALLOC_CHUNK
= 2,
6610 LOOP_NO_EMPTY_SIZE
= 3,
6614 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6618 down_read(&cache
->data_rwsem
);
6622 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6625 btrfs_get_block_group(cache
);
6627 down_read(&cache
->data_rwsem
);
6630 static struct btrfs_block_group_cache
*
6631 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6632 struct btrfs_free_cluster
*cluster
,
6635 struct btrfs_block_group_cache
*used_bg
;
6636 bool locked
= false;
6638 spin_lock(&cluster
->refill_lock
);
6640 if (used_bg
== cluster
->block_group
)
6643 up_read(&used_bg
->data_rwsem
);
6644 btrfs_put_block_group(used_bg
);
6647 used_bg
= cluster
->block_group
;
6651 if (used_bg
== block_group
)
6654 btrfs_get_block_group(used_bg
);
6659 if (down_read_trylock(&used_bg
->data_rwsem
))
6662 spin_unlock(&cluster
->refill_lock
);
6663 down_read(&used_bg
->data_rwsem
);
6669 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6673 up_read(&cache
->data_rwsem
);
6674 btrfs_put_block_group(cache
);
6678 * walks the btree of allocated extents and find a hole of a given size.
6679 * The key ins is changed to record the hole:
6680 * ins->objectid == start position
6681 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6682 * ins->offset == the size of the hole.
6683 * Any available blocks before search_start are skipped.
6685 * If there is no suitable free space, we will record the max size of
6686 * the free space extent currently.
6688 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6689 u64 num_bytes
, u64 empty_size
,
6690 u64 hint_byte
, struct btrfs_key
*ins
,
6691 u64 flags
, int delalloc
)
6694 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6695 struct btrfs_free_cluster
*last_ptr
= NULL
;
6696 struct btrfs_block_group_cache
*block_group
= NULL
;
6697 u64 search_start
= 0;
6698 u64 max_extent_size
= 0;
6699 int empty_cluster
= 2 * 1024 * 1024;
6700 struct btrfs_space_info
*space_info
;
6702 int index
= __get_raid_index(flags
);
6703 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6704 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6705 bool failed_cluster_refill
= false;
6706 bool failed_alloc
= false;
6707 bool use_cluster
= true;
6708 bool have_caching_bg
= false;
6710 WARN_ON(num_bytes
< root
->sectorsize
);
6711 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6715 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6717 space_info
= __find_space_info(root
->fs_info
, flags
);
6719 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6724 * If the space info is for both data and metadata it means we have a
6725 * small filesystem and we can't use the clustering stuff.
6727 if (btrfs_mixed_space_info(space_info
))
6728 use_cluster
= false;
6730 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6731 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6732 if (!btrfs_test_opt(root
, SSD
))
6733 empty_cluster
= 64 * 1024;
6736 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6737 btrfs_test_opt(root
, SSD
)) {
6738 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6742 spin_lock(&last_ptr
->lock
);
6743 if (last_ptr
->block_group
)
6744 hint_byte
= last_ptr
->window_start
;
6745 spin_unlock(&last_ptr
->lock
);
6748 search_start
= max(search_start
, first_logical_byte(root
, 0));
6749 search_start
= max(search_start
, hint_byte
);
6754 if (search_start
== hint_byte
) {
6755 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6758 * we don't want to use the block group if it doesn't match our
6759 * allocation bits, or if its not cached.
6761 * However if we are re-searching with an ideal block group
6762 * picked out then we don't care that the block group is cached.
6764 if (block_group
&& block_group_bits(block_group
, flags
) &&
6765 block_group
->cached
!= BTRFS_CACHE_NO
) {
6766 down_read(&space_info
->groups_sem
);
6767 if (list_empty(&block_group
->list
) ||
6770 * someone is removing this block group,
6771 * we can't jump into the have_block_group
6772 * target because our list pointers are not
6775 btrfs_put_block_group(block_group
);
6776 up_read(&space_info
->groups_sem
);
6778 index
= get_block_group_index(block_group
);
6779 btrfs_lock_block_group(block_group
, delalloc
);
6780 goto have_block_group
;
6782 } else if (block_group
) {
6783 btrfs_put_block_group(block_group
);
6787 have_caching_bg
= false;
6788 down_read(&space_info
->groups_sem
);
6789 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6794 btrfs_grab_block_group(block_group
, delalloc
);
6795 search_start
= block_group
->key
.objectid
;
6798 * this can happen if we end up cycling through all the
6799 * raid types, but we want to make sure we only allocate
6800 * for the proper type.
6802 if (!block_group_bits(block_group
, flags
)) {
6803 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6804 BTRFS_BLOCK_GROUP_RAID1
|
6805 BTRFS_BLOCK_GROUP_RAID5
|
6806 BTRFS_BLOCK_GROUP_RAID6
|
6807 BTRFS_BLOCK_GROUP_RAID10
;
6810 * if they asked for extra copies and this block group
6811 * doesn't provide them, bail. This does allow us to
6812 * fill raid0 from raid1.
6814 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6819 cached
= block_group_cache_done(block_group
);
6820 if (unlikely(!cached
)) {
6821 ret
= cache_block_group(block_group
, 0);
6826 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6828 if (unlikely(block_group
->ro
))
6832 * Ok we want to try and use the cluster allocator, so
6836 struct btrfs_block_group_cache
*used_block_group
;
6837 unsigned long aligned_cluster
;
6839 * the refill lock keeps out other
6840 * people trying to start a new cluster
6842 used_block_group
= btrfs_lock_cluster(block_group
,
6845 if (!used_block_group
)
6846 goto refill_cluster
;
6848 if (used_block_group
!= block_group
&&
6849 (used_block_group
->ro
||
6850 !block_group_bits(used_block_group
, flags
)))
6851 goto release_cluster
;
6853 offset
= btrfs_alloc_from_cluster(used_block_group
,
6856 used_block_group
->key
.objectid
,
6859 /* we have a block, we're done */
6860 spin_unlock(&last_ptr
->refill_lock
);
6861 trace_btrfs_reserve_extent_cluster(root
,
6863 search_start
, num_bytes
);
6864 if (used_block_group
!= block_group
) {
6865 btrfs_release_block_group(block_group
,
6867 block_group
= used_block_group
;
6872 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6874 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6875 * set up a new clusters, so lets just skip it
6876 * and let the allocator find whatever block
6877 * it can find. If we reach this point, we
6878 * will have tried the cluster allocator
6879 * plenty of times and not have found
6880 * anything, so we are likely way too
6881 * fragmented for the clustering stuff to find
6884 * However, if the cluster is taken from the
6885 * current block group, release the cluster
6886 * first, so that we stand a better chance of
6887 * succeeding in the unclustered
6889 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6890 used_block_group
!= block_group
) {
6891 spin_unlock(&last_ptr
->refill_lock
);
6892 btrfs_release_block_group(used_block_group
,
6894 goto unclustered_alloc
;
6898 * this cluster didn't work out, free it and
6901 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6903 if (used_block_group
!= block_group
)
6904 btrfs_release_block_group(used_block_group
,
6907 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6908 spin_unlock(&last_ptr
->refill_lock
);
6909 goto unclustered_alloc
;
6912 aligned_cluster
= max_t(unsigned long,
6913 empty_cluster
+ empty_size
,
6914 block_group
->full_stripe_len
);
6916 /* allocate a cluster in this block group */
6917 ret
= btrfs_find_space_cluster(root
, block_group
,
6918 last_ptr
, search_start
,
6923 * now pull our allocation out of this
6926 offset
= btrfs_alloc_from_cluster(block_group
,
6932 /* we found one, proceed */
6933 spin_unlock(&last_ptr
->refill_lock
);
6934 trace_btrfs_reserve_extent_cluster(root
,
6935 block_group
, search_start
,
6939 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6940 && !failed_cluster_refill
) {
6941 spin_unlock(&last_ptr
->refill_lock
);
6943 failed_cluster_refill
= true;
6944 wait_block_group_cache_progress(block_group
,
6945 num_bytes
+ empty_cluster
+ empty_size
);
6946 goto have_block_group
;
6950 * at this point we either didn't find a cluster
6951 * or we weren't able to allocate a block from our
6952 * cluster. Free the cluster we've been trying
6953 * to use, and go to the next block group
6955 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6956 spin_unlock(&last_ptr
->refill_lock
);
6961 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6963 block_group
->free_space_ctl
->free_space
<
6964 num_bytes
+ empty_cluster
+ empty_size
) {
6965 if (block_group
->free_space_ctl
->free_space
>
6968 block_group
->free_space_ctl
->free_space
;
6969 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6972 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6974 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6975 num_bytes
, empty_size
,
6978 * If we didn't find a chunk, and we haven't failed on this
6979 * block group before, and this block group is in the middle of
6980 * caching and we are ok with waiting, then go ahead and wait
6981 * for progress to be made, and set failed_alloc to true.
6983 * If failed_alloc is true then we've already waited on this
6984 * block group once and should move on to the next block group.
6986 if (!offset
&& !failed_alloc
&& !cached
&&
6987 loop
> LOOP_CACHING_NOWAIT
) {
6988 wait_block_group_cache_progress(block_group
,
6989 num_bytes
+ empty_size
);
6990 failed_alloc
= true;
6991 goto have_block_group
;
6992 } else if (!offset
) {
6994 have_caching_bg
= true;
6998 search_start
= ALIGN(offset
, root
->stripesize
);
7000 /* move on to the next group */
7001 if (search_start
+ num_bytes
>
7002 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7003 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7007 if (offset
< search_start
)
7008 btrfs_add_free_space(block_group
, offset
,
7009 search_start
- offset
);
7010 BUG_ON(offset
> search_start
);
7012 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7013 alloc_type
, delalloc
);
7014 if (ret
== -EAGAIN
) {
7015 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7019 /* we are all good, lets return */
7020 ins
->objectid
= search_start
;
7021 ins
->offset
= num_bytes
;
7023 trace_btrfs_reserve_extent(orig_root
, block_group
,
7024 search_start
, num_bytes
);
7025 btrfs_release_block_group(block_group
, delalloc
);
7028 failed_cluster_refill
= false;
7029 failed_alloc
= false;
7030 BUG_ON(index
!= get_block_group_index(block_group
));
7031 btrfs_release_block_group(block_group
, delalloc
);
7033 up_read(&space_info
->groups_sem
);
7035 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7038 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7042 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7043 * caching kthreads as we move along
7044 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7045 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7046 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7049 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7052 if (loop
== LOOP_ALLOC_CHUNK
) {
7053 struct btrfs_trans_handle
*trans
;
7056 trans
= current
->journal_info
;
7060 trans
= btrfs_join_transaction(root
);
7062 if (IS_ERR(trans
)) {
7063 ret
= PTR_ERR(trans
);
7067 ret
= do_chunk_alloc(trans
, root
, flags
,
7070 * Do not bail out on ENOSPC since we
7071 * can do more things.
7073 if (ret
< 0 && ret
!= -ENOSPC
)
7074 btrfs_abort_transaction(trans
,
7079 btrfs_end_transaction(trans
, root
);
7084 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7090 } else if (!ins
->objectid
) {
7092 } else if (ins
->objectid
) {
7097 ins
->offset
= max_extent_size
;
7101 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7102 int dump_block_groups
)
7104 struct btrfs_block_group_cache
*cache
;
7107 spin_lock(&info
->lock
);
7108 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7110 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7111 info
->bytes_reserved
- info
->bytes_readonly
,
7112 (info
->full
) ? "" : "not ");
7113 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7114 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7115 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7116 info
->bytes_reserved
, info
->bytes_may_use
,
7117 info
->bytes_readonly
);
7118 spin_unlock(&info
->lock
);
7120 if (!dump_block_groups
)
7123 down_read(&info
->groups_sem
);
7125 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7126 spin_lock(&cache
->lock
);
7127 printk(KERN_INFO
"BTRFS: "
7128 "block group %llu has %llu bytes, "
7129 "%llu used %llu pinned %llu reserved %s\n",
7130 cache
->key
.objectid
, cache
->key
.offset
,
7131 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7132 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7133 btrfs_dump_free_space(cache
, bytes
);
7134 spin_unlock(&cache
->lock
);
7136 if (++index
< BTRFS_NR_RAID_TYPES
)
7138 up_read(&info
->groups_sem
);
7141 int btrfs_reserve_extent(struct btrfs_root
*root
,
7142 u64 num_bytes
, u64 min_alloc_size
,
7143 u64 empty_size
, u64 hint_byte
,
7144 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7146 bool final_tried
= false;
7150 flags
= btrfs_get_alloc_profile(root
, is_data
);
7152 WARN_ON(num_bytes
< root
->sectorsize
);
7153 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7156 if (ret
== -ENOSPC
) {
7157 if (!final_tried
&& ins
->offset
) {
7158 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7159 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7160 num_bytes
= max(num_bytes
, min_alloc_size
);
7161 if (num_bytes
== min_alloc_size
)
7164 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7165 struct btrfs_space_info
*sinfo
;
7167 sinfo
= __find_space_info(root
->fs_info
, flags
);
7168 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7171 dump_space_info(sinfo
, num_bytes
, 1);
7178 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7180 int pin
, int delalloc
)
7182 struct btrfs_block_group_cache
*cache
;
7185 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7187 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7193 pin_down_extent(root
, cache
, start
, len
, 1);
7195 if (btrfs_test_opt(root
, DISCARD
))
7196 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7197 btrfs_add_free_space(cache
, start
, len
);
7198 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7200 btrfs_put_block_group(cache
);
7202 trace_btrfs_reserved_extent_free(root
, start
, len
);
7207 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7208 u64 start
, u64 len
, int delalloc
)
7210 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7213 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7216 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7219 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7220 struct btrfs_root
*root
,
7221 u64 parent
, u64 root_objectid
,
7222 u64 flags
, u64 owner
, u64 offset
,
7223 struct btrfs_key
*ins
, int ref_mod
)
7226 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7227 struct btrfs_extent_item
*extent_item
;
7228 struct btrfs_extent_inline_ref
*iref
;
7229 struct btrfs_path
*path
;
7230 struct extent_buffer
*leaf
;
7235 type
= BTRFS_SHARED_DATA_REF_KEY
;
7237 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7239 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7241 path
= btrfs_alloc_path();
7245 path
->leave_spinning
= 1;
7246 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7249 btrfs_free_path(path
);
7253 leaf
= path
->nodes
[0];
7254 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7255 struct btrfs_extent_item
);
7256 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7257 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7258 btrfs_set_extent_flags(leaf
, extent_item
,
7259 flags
| BTRFS_EXTENT_FLAG_DATA
);
7261 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7262 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7264 struct btrfs_shared_data_ref
*ref
;
7265 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7266 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7267 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7269 struct btrfs_extent_data_ref
*ref
;
7270 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7271 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7272 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7273 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7274 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7277 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7278 btrfs_free_path(path
);
7280 /* Always set parent to 0 here since its exclusive anyway. */
7281 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7282 ins
->objectid
, ins
->offset
,
7283 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7287 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7288 if (ret
) { /* -ENOENT, logic error */
7289 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7290 ins
->objectid
, ins
->offset
);
7293 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7297 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7298 struct btrfs_root
*root
,
7299 u64 parent
, u64 root_objectid
,
7300 u64 flags
, struct btrfs_disk_key
*key
,
7301 int level
, struct btrfs_key
*ins
,
7305 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7306 struct btrfs_extent_item
*extent_item
;
7307 struct btrfs_tree_block_info
*block_info
;
7308 struct btrfs_extent_inline_ref
*iref
;
7309 struct btrfs_path
*path
;
7310 struct extent_buffer
*leaf
;
7311 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7312 u64 num_bytes
= ins
->offset
;
7313 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7316 if (!skinny_metadata
)
7317 size
+= sizeof(*block_info
);
7319 path
= btrfs_alloc_path();
7321 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7326 path
->leave_spinning
= 1;
7327 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7330 btrfs_free_path(path
);
7331 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7336 leaf
= path
->nodes
[0];
7337 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7338 struct btrfs_extent_item
);
7339 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7340 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7341 btrfs_set_extent_flags(leaf
, extent_item
,
7342 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7344 if (skinny_metadata
) {
7345 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7346 num_bytes
= root
->nodesize
;
7348 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7349 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7350 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7351 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7355 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7356 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7357 BTRFS_SHARED_BLOCK_REF_KEY
);
7358 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7360 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7361 BTRFS_TREE_BLOCK_REF_KEY
);
7362 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7365 btrfs_mark_buffer_dirty(leaf
);
7366 btrfs_free_path(path
);
7369 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7370 ins
->objectid
, num_bytes
,
7371 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7376 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7378 if (ret
) { /* -ENOENT, logic error */
7379 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7380 ins
->objectid
, ins
->offset
);
7384 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7388 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7389 struct btrfs_root
*root
,
7390 u64 root_objectid
, u64 owner
,
7391 u64 offset
, struct btrfs_key
*ins
)
7395 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7397 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7399 root_objectid
, owner
, offset
,
7400 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7405 * this is used by the tree logging recovery code. It records that
7406 * an extent has been allocated and makes sure to clear the free
7407 * space cache bits as well
7409 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7410 struct btrfs_root
*root
,
7411 u64 root_objectid
, u64 owner
, u64 offset
,
7412 struct btrfs_key
*ins
)
7415 struct btrfs_block_group_cache
*block_group
;
7418 * Mixed block groups will exclude before processing the log so we only
7419 * need to do the exlude dance if this fs isn't mixed.
7421 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7422 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7427 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7431 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7432 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7433 BUG_ON(ret
); /* logic error */
7434 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7435 0, owner
, offset
, ins
, 1);
7436 btrfs_put_block_group(block_group
);
7440 static struct extent_buffer
*
7441 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7442 u64 bytenr
, int level
)
7444 struct extent_buffer
*buf
;
7446 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7448 return ERR_PTR(-ENOMEM
);
7449 btrfs_set_header_generation(buf
, trans
->transid
);
7450 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7451 btrfs_tree_lock(buf
);
7452 clean_tree_block(trans
, root
->fs_info
, buf
);
7453 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7455 btrfs_set_lock_blocking(buf
);
7456 btrfs_set_buffer_uptodate(buf
);
7458 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7459 buf
->log_index
= root
->log_transid
% 2;
7461 * we allow two log transactions at a time, use different
7462 * EXENT bit to differentiate dirty pages.
7464 if (buf
->log_index
== 0)
7465 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7466 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7468 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7469 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7471 buf
->log_index
= -1;
7472 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7473 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7475 trans
->blocks_used
++;
7476 /* this returns a buffer locked for blocking */
7480 static struct btrfs_block_rsv
*
7481 use_block_rsv(struct btrfs_trans_handle
*trans
,
7482 struct btrfs_root
*root
, u32 blocksize
)
7484 struct btrfs_block_rsv
*block_rsv
;
7485 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7487 bool global_updated
= false;
7489 block_rsv
= get_block_rsv(trans
, root
);
7491 if (unlikely(block_rsv
->size
== 0))
7494 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7498 if (block_rsv
->failfast
)
7499 return ERR_PTR(ret
);
7501 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7502 global_updated
= true;
7503 update_global_block_rsv(root
->fs_info
);
7507 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7508 static DEFINE_RATELIMIT_STATE(_rs
,
7509 DEFAULT_RATELIMIT_INTERVAL
* 10,
7510 /*DEFAULT_RATELIMIT_BURST*/ 1);
7511 if (__ratelimit(&_rs
))
7513 "BTRFS: block rsv returned %d\n", ret
);
7516 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7517 BTRFS_RESERVE_NO_FLUSH
);
7521 * If we couldn't reserve metadata bytes try and use some from
7522 * the global reserve if its space type is the same as the global
7525 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7526 block_rsv
->space_info
== global_rsv
->space_info
) {
7527 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7531 return ERR_PTR(ret
);
7534 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7535 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7537 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7538 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7542 * finds a free extent and does all the dirty work required for allocation
7543 * returns the key for the extent through ins, and a tree buffer for
7544 * the first block of the extent through buf.
7546 * returns the tree buffer or NULL.
7548 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7549 struct btrfs_root
*root
,
7550 u64 parent
, u64 root_objectid
,
7551 struct btrfs_disk_key
*key
, int level
,
7552 u64 hint
, u64 empty_size
)
7554 struct btrfs_key ins
;
7555 struct btrfs_block_rsv
*block_rsv
;
7556 struct extent_buffer
*buf
;
7559 u32 blocksize
= root
->nodesize
;
7560 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7563 if (btrfs_test_is_dummy_root(root
)) {
7564 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7567 root
->alloc_bytenr
+= blocksize
;
7571 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7572 if (IS_ERR(block_rsv
))
7573 return ERR_CAST(block_rsv
);
7575 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7576 empty_size
, hint
, &ins
, 0, 0);
7578 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7579 return ERR_PTR(ret
);
7582 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7583 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7585 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7587 parent
= ins
.objectid
;
7588 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7592 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7593 struct btrfs_delayed_extent_op
*extent_op
;
7594 extent_op
= btrfs_alloc_delayed_extent_op();
7595 BUG_ON(!extent_op
); /* -ENOMEM */
7597 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7599 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7600 extent_op
->flags_to_set
= flags
;
7601 if (skinny_metadata
)
7602 extent_op
->update_key
= 0;
7604 extent_op
->update_key
= 1;
7605 extent_op
->update_flags
= 1;
7606 extent_op
->is_data
= 0;
7607 extent_op
->level
= level
;
7609 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7611 ins
.offset
, parent
, root_objectid
,
7612 level
, BTRFS_ADD_DELAYED_EXTENT
,
7614 BUG_ON(ret
); /* -ENOMEM */
7619 struct walk_control
{
7620 u64 refs
[BTRFS_MAX_LEVEL
];
7621 u64 flags
[BTRFS_MAX_LEVEL
];
7622 struct btrfs_key update_progress
;
7633 #define DROP_REFERENCE 1
7634 #define UPDATE_BACKREF 2
7636 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7637 struct btrfs_root
*root
,
7638 struct walk_control
*wc
,
7639 struct btrfs_path
*path
)
7647 struct btrfs_key key
;
7648 struct extent_buffer
*eb
;
7653 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7654 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7655 wc
->reada_count
= max(wc
->reada_count
, 2);
7657 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7658 wc
->reada_count
= min_t(int, wc
->reada_count
,
7659 BTRFS_NODEPTRS_PER_BLOCK(root
));
7662 eb
= path
->nodes
[wc
->level
];
7663 nritems
= btrfs_header_nritems(eb
);
7664 blocksize
= root
->nodesize
;
7666 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7667 if (nread
>= wc
->reada_count
)
7671 bytenr
= btrfs_node_blockptr(eb
, slot
);
7672 generation
= btrfs_node_ptr_generation(eb
, slot
);
7674 if (slot
== path
->slots
[wc
->level
])
7677 if (wc
->stage
== UPDATE_BACKREF
&&
7678 generation
<= root
->root_key
.offset
)
7681 /* We don't lock the tree block, it's OK to be racy here */
7682 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7683 wc
->level
- 1, 1, &refs
,
7685 /* We don't care about errors in readahead. */
7690 if (wc
->stage
== DROP_REFERENCE
) {
7694 if (wc
->level
== 1 &&
7695 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7697 if (!wc
->update_ref
||
7698 generation
<= root
->root_key
.offset
)
7700 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7701 ret
= btrfs_comp_cpu_keys(&key
,
7702 &wc
->update_progress
);
7706 if (wc
->level
== 1 &&
7707 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7711 readahead_tree_block(root
, bytenr
);
7714 wc
->reada_slot
= slot
;
7717 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7718 struct btrfs_root
*root
,
7719 struct extent_buffer
*eb
)
7721 int nr
= btrfs_header_nritems(eb
);
7722 int i
, extent_type
, ret
;
7723 struct btrfs_key key
;
7724 struct btrfs_file_extent_item
*fi
;
7725 u64 bytenr
, num_bytes
;
7727 for (i
= 0; i
< nr
; i
++) {
7728 btrfs_item_key_to_cpu(eb
, &key
, i
);
7730 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7733 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7734 /* filter out non qgroup-accountable extents */
7735 extent_type
= btrfs_file_extent_type(eb
, fi
);
7737 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7740 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7744 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7746 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7749 BTRFS_QGROUP_OPER_SUB_SUBTREE
, 0);
7757 * Walk up the tree from the bottom, freeing leaves and any interior
7758 * nodes which have had all slots visited. If a node (leaf or
7759 * interior) is freed, the node above it will have it's slot
7760 * incremented. The root node will never be freed.
7762 * At the end of this function, we should have a path which has all
7763 * slots incremented to the next position for a search. If we need to
7764 * read a new node it will be NULL and the node above it will have the
7765 * correct slot selected for a later read.
7767 * If we increment the root nodes slot counter past the number of
7768 * elements, 1 is returned to signal completion of the search.
7770 static int adjust_slots_upwards(struct btrfs_root
*root
,
7771 struct btrfs_path
*path
, int root_level
)
7775 struct extent_buffer
*eb
;
7777 if (root_level
== 0)
7780 while (level
<= root_level
) {
7781 eb
= path
->nodes
[level
];
7782 nr
= btrfs_header_nritems(eb
);
7783 path
->slots
[level
]++;
7784 slot
= path
->slots
[level
];
7785 if (slot
>= nr
|| level
== 0) {
7787 * Don't free the root - we will detect this
7788 * condition after our loop and return a
7789 * positive value for caller to stop walking the tree.
7791 if (level
!= root_level
) {
7792 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7793 path
->locks
[level
] = 0;
7795 free_extent_buffer(eb
);
7796 path
->nodes
[level
] = NULL
;
7797 path
->slots
[level
] = 0;
7801 * We have a valid slot to walk back down
7802 * from. Stop here so caller can process these
7811 eb
= path
->nodes
[root_level
];
7812 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7819 * root_eb is the subtree root and is locked before this function is called.
7821 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7822 struct btrfs_root
*root
,
7823 struct extent_buffer
*root_eb
,
7829 struct extent_buffer
*eb
= root_eb
;
7830 struct btrfs_path
*path
= NULL
;
7832 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7833 BUG_ON(root_eb
== NULL
);
7835 if (!root
->fs_info
->quota_enabled
)
7838 if (!extent_buffer_uptodate(root_eb
)) {
7839 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7844 if (root_level
== 0) {
7845 ret
= account_leaf_items(trans
, root
, root_eb
);
7849 path
= btrfs_alloc_path();
7854 * Walk down the tree. Missing extent blocks are filled in as
7855 * we go. Metadata is accounted every time we read a new
7858 * When we reach a leaf, we account for file extent items in it,
7859 * walk back up the tree (adjusting slot pointers as we go)
7860 * and restart the search process.
7862 extent_buffer_get(root_eb
); /* For path */
7863 path
->nodes
[root_level
] = root_eb
;
7864 path
->slots
[root_level
] = 0;
7865 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7868 while (level
>= 0) {
7869 if (path
->nodes
[level
] == NULL
) {
7874 /* We need to get child blockptr/gen from
7875 * parent before we can read it. */
7876 eb
= path
->nodes
[level
+ 1];
7877 parent_slot
= path
->slots
[level
+ 1];
7878 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7879 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7881 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
7882 if (!eb
|| !extent_buffer_uptodate(eb
)) {
7887 path
->nodes
[level
] = eb
;
7888 path
->slots
[level
] = 0;
7890 btrfs_tree_read_lock(eb
);
7891 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7892 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7894 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7898 BTRFS_QGROUP_OPER_SUB_SUBTREE
,
7906 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7910 /* Nonzero return here means we completed our search */
7911 ret
= adjust_slots_upwards(root
, path
, root_level
);
7915 /* Restart search with new slots */
7924 btrfs_free_path(path
);
7930 * helper to process tree block while walking down the tree.
7932 * when wc->stage == UPDATE_BACKREF, this function updates
7933 * back refs for pointers in the block.
7935 * NOTE: return value 1 means we should stop walking down.
7937 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7938 struct btrfs_root
*root
,
7939 struct btrfs_path
*path
,
7940 struct walk_control
*wc
, int lookup_info
)
7942 int level
= wc
->level
;
7943 struct extent_buffer
*eb
= path
->nodes
[level
];
7944 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7947 if (wc
->stage
== UPDATE_BACKREF
&&
7948 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7952 * when reference count of tree block is 1, it won't increase
7953 * again. once full backref flag is set, we never clear it.
7956 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7957 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7958 BUG_ON(!path
->locks
[level
]);
7959 ret
= btrfs_lookup_extent_info(trans
, root
,
7960 eb
->start
, level
, 1,
7963 BUG_ON(ret
== -ENOMEM
);
7966 BUG_ON(wc
->refs
[level
] == 0);
7969 if (wc
->stage
== DROP_REFERENCE
) {
7970 if (wc
->refs
[level
] > 1)
7973 if (path
->locks
[level
] && !wc
->keep_locks
) {
7974 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7975 path
->locks
[level
] = 0;
7980 /* wc->stage == UPDATE_BACKREF */
7981 if (!(wc
->flags
[level
] & flag
)) {
7982 BUG_ON(!path
->locks
[level
]);
7983 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
7984 BUG_ON(ret
); /* -ENOMEM */
7985 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7986 BUG_ON(ret
); /* -ENOMEM */
7987 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7989 btrfs_header_level(eb
), 0);
7990 BUG_ON(ret
); /* -ENOMEM */
7991 wc
->flags
[level
] |= flag
;
7995 * the block is shared by multiple trees, so it's not good to
7996 * keep the tree lock
7998 if (path
->locks
[level
] && level
> 0) {
7999 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8000 path
->locks
[level
] = 0;
8006 * helper to process tree block pointer.
8008 * when wc->stage == DROP_REFERENCE, this function checks
8009 * reference count of the block pointed to. if the block
8010 * is shared and we need update back refs for the subtree
8011 * rooted at the block, this function changes wc->stage to
8012 * UPDATE_BACKREF. if the block is shared and there is no
8013 * need to update back, this function drops the reference
8016 * NOTE: return value 1 means we should stop walking down.
8018 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8019 struct btrfs_root
*root
,
8020 struct btrfs_path
*path
,
8021 struct walk_control
*wc
, int *lookup_info
)
8027 struct btrfs_key key
;
8028 struct extent_buffer
*next
;
8029 int level
= wc
->level
;
8032 bool need_account
= false;
8034 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8035 path
->slots
[level
]);
8037 * if the lower level block was created before the snapshot
8038 * was created, we know there is no need to update back refs
8041 if (wc
->stage
== UPDATE_BACKREF
&&
8042 generation
<= root
->root_key
.offset
) {
8047 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8048 blocksize
= root
->nodesize
;
8050 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8052 next
= btrfs_find_create_tree_block(root
, bytenr
);
8055 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8059 btrfs_tree_lock(next
);
8060 btrfs_set_lock_blocking(next
);
8062 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8063 &wc
->refs
[level
- 1],
8064 &wc
->flags
[level
- 1]);
8066 btrfs_tree_unlock(next
);
8070 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8071 btrfs_err(root
->fs_info
, "Missing references.");
8076 if (wc
->stage
== DROP_REFERENCE
) {
8077 if (wc
->refs
[level
- 1] > 1) {
8078 need_account
= true;
8080 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8083 if (!wc
->update_ref
||
8084 generation
<= root
->root_key
.offset
)
8087 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8088 path
->slots
[level
]);
8089 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8093 wc
->stage
= UPDATE_BACKREF
;
8094 wc
->shared_level
= level
- 1;
8098 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8102 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8103 btrfs_tree_unlock(next
);
8104 free_extent_buffer(next
);
8110 if (reada
&& level
== 1)
8111 reada_walk_down(trans
, root
, wc
, path
);
8112 next
= read_tree_block(root
, bytenr
, generation
);
8113 if (!next
|| !extent_buffer_uptodate(next
)) {
8114 free_extent_buffer(next
);
8117 btrfs_tree_lock(next
);
8118 btrfs_set_lock_blocking(next
);
8122 BUG_ON(level
!= btrfs_header_level(next
));
8123 path
->nodes
[level
] = next
;
8124 path
->slots
[level
] = 0;
8125 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8131 wc
->refs
[level
- 1] = 0;
8132 wc
->flags
[level
- 1] = 0;
8133 if (wc
->stage
== DROP_REFERENCE
) {
8134 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8135 parent
= path
->nodes
[level
]->start
;
8137 BUG_ON(root
->root_key
.objectid
!=
8138 btrfs_header_owner(path
->nodes
[level
]));
8143 ret
= account_shared_subtree(trans
, root
, next
,
8144 generation
, level
- 1);
8146 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8147 "%d accounting shared subtree. Quota "
8148 "is out of sync, rescan required.\n",
8149 root
->fs_info
->sb
->s_id
, ret
);
8152 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8153 root
->root_key
.objectid
, level
- 1, 0, 0);
8154 BUG_ON(ret
); /* -ENOMEM */
8156 btrfs_tree_unlock(next
);
8157 free_extent_buffer(next
);
8163 * helper to process tree block while walking up the tree.
8165 * when wc->stage == DROP_REFERENCE, this function drops
8166 * reference count on the block.
8168 * when wc->stage == UPDATE_BACKREF, this function changes
8169 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8170 * to UPDATE_BACKREF previously while processing the block.
8172 * NOTE: return value 1 means we should stop walking up.
8174 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8175 struct btrfs_root
*root
,
8176 struct btrfs_path
*path
,
8177 struct walk_control
*wc
)
8180 int level
= wc
->level
;
8181 struct extent_buffer
*eb
= path
->nodes
[level
];
8184 if (wc
->stage
== UPDATE_BACKREF
) {
8185 BUG_ON(wc
->shared_level
< level
);
8186 if (level
< wc
->shared_level
)
8189 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8193 wc
->stage
= DROP_REFERENCE
;
8194 wc
->shared_level
= -1;
8195 path
->slots
[level
] = 0;
8198 * check reference count again if the block isn't locked.
8199 * we should start walking down the tree again if reference
8202 if (!path
->locks
[level
]) {
8204 btrfs_tree_lock(eb
);
8205 btrfs_set_lock_blocking(eb
);
8206 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8208 ret
= btrfs_lookup_extent_info(trans
, root
,
8209 eb
->start
, level
, 1,
8213 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8214 path
->locks
[level
] = 0;
8217 BUG_ON(wc
->refs
[level
] == 0);
8218 if (wc
->refs
[level
] == 1) {
8219 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8220 path
->locks
[level
] = 0;
8226 /* wc->stage == DROP_REFERENCE */
8227 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8229 if (wc
->refs
[level
] == 1) {
8231 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8232 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8234 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8235 BUG_ON(ret
); /* -ENOMEM */
8236 ret
= account_leaf_items(trans
, root
, eb
);
8238 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8239 "%d accounting leaf items. Quota "
8240 "is out of sync, rescan required.\n",
8241 root
->fs_info
->sb
->s_id
, ret
);
8244 /* make block locked assertion in clean_tree_block happy */
8245 if (!path
->locks
[level
] &&
8246 btrfs_header_generation(eb
) == trans
->transid
) {
8247 btrfs_tree_lock(eb
);
8248 btrfs_set_lock_blocking(eb
);
8249 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8251 clean_tree_block(trans
, root
->fs_info
, eb
);
8254 if (eb
== root
->node
) {
8255 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8258 BUG_ON(root
->root_key
.objectid
!=
8259 btrfs_header_owner(eb
));
8261 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8262 parent
= path
->nodes
[level
+ 1]->start
;
8264 BUG_ON(root
->root_key
.objectid
!=
8265 btrfs_header_owner(path
->nodes
[level
+ 1]));
8268 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8270 wc
->refs
[level
] = 0;
8271 wc
->flags
[level
] = 0;
8275 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8276 struct btrfs_root
*root
,
8277 struct btrfs_path
*path
,
8278 struct walk_control
*wc
)
8280 int level
= wc
->level
;
8281 int lookup_info
= 1;
8284 while (level
>= 0) {
8285 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8292 if (path
->slots
[level
] >=
8293 btrfs_header_nritems(path
->nodes
[level
]))
8296 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8298 path
->slots
[level
]++;
8307 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8308 struct btrfs_root
*root
,
8309 struct btrfs_path
*path
,
8310 struct walk_control
*wc
, int max_level
)
8312 int level
= wc
->level
;
8315 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8316 while (level
< max_level
&& path
->nodes
[level
]) {
8318 if (path
->slots
[level
] + 1 <
8319 btrfs_header_nritems(path
->nodes
[level
])) {
8320 path
->slots
[level
]++;
8323 ret
= walk_up_proc(trans
, root
, path
, wc
);
8327 if (path
->locks
[level
]) {
8328 btrfs_tree_unlock_rw(path
->nodes
[level
],
8329 path
->locks
[level
]);
8330 path
->locks
[level
] = 0;
8332 free_extent_buffer(path
->nodes
[level
]);
8333 path
->nodes
[level
] = NULL
;
8341 * drop a subvolume tree.
8343 * this function traverses the tree freeing any blocks that only
8344 * referenced by the tree.
8346 * when a shared tree block is found. this function decreases its
8347 * reference count by one. if update_ref is true, this function
8348 * also make sure backrefs for the shared block and all lower level
8349 * blocks are properly updated.
8351 * If called with for_reloc == 0, may exit early with -EAGAIN
8353 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8354 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8357 struct btrfs_path
*path
;
8358 struct btrfs_trans_handle
*trans
;
8359 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8360 struct btrfs_root_item
*root_item
= &root
->root_item
;
8361 struct walk_control
*wc
;
8362 struct btrfs_key key
;
8366 bool root_dropped
= false;
8368 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8370 path
= btrfs_alloc_path();
8376 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8378 btrfs_free_path(path
);
8383 trans
= btrfs_start_transaction(tree_root
, 0);
8384 if (IS_ERR(trans
)) {
8385 err
= PTR_ERR(trans
);
8390 trans
->block_rsv
= block_rsv
;
8392 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8393 level
= btrfs_header_level(root
->node
);
8394 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8395 btrfs_set_lock_blocking(path
->nodes
[level
]);
8396 path
->slots
[level
] = 0;
8397 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8398 memset(&wc
->update_progress
, 0,
8399 sizeof(wc
->update_progress
));
8401 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8402 memcpy(&wc
->update_progress
, &key
,
8403 sizeof(wc
->update_progress
));
8405 level
= root_item
->drop_level
;
8407 path
->lowest_level
= level
;
8408 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8409 path
->lowest_level
= 0;
8417 * unlock our path, this is safe because only this
8418 * function is allowed to delete this snapshot
8420 btrfs_unlock_up_safe(path
, 0);
8422 level
= btrfs_header_level(root
->node
);
8424 btrfs_tree_lock(path
->nodes
[level
]);
8425 btrfs_set_lock_blocking(path
->nodes
[level
]);
8426 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8428 ret
= btrfs_lookup_extent_info(trans
, root
,
8429 path
->nodes
[level
]->start
,
8430 level
, 1, &wc
->refs
[level
],
8436 BUG_ON(wc
->refs
[level
] == 0);
8438 if (level
== root_item
->drop_level
)
8441 btrfs_tree_unlock(path
->nodes
[level
]);
8442 path
->locks
[level
] = 0;
8443 WARN_ON(wc
->refs
[level
] != 1);
8449 wc
->shared_level
= -1;
8450 wc
->stage
= DROP_REFERENCE
;
8451 wc
->update_ref
= update_ref
;
8453 wc
->for_reloc
= for_reloc
;
8454 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8458 ret
= walk_down_tree(trans
, root
, path
, wc
);
8464 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8471 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8475 if (wc
->stage
== DROP_REFERENCE
) {
8477 btrfs_node_key(path
->nodes
[level
],
8478 &root_item
->drop_progress
,
8479 path
->slots
[level
]);
8480 root_item
->drop_level
= level
;
8483 BUG_ON(wc
->level
== 0);
8484 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8485 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8486 ret
= btrfs_update_root(trans
, tree_root
,
8490 btrfs_abort_transaction(trans
, tree_root
, ret
);
8496 * Qgroup update accounting is run from
8497 * delayed ref handling. This usually works
8498 * out because delayed refs are normally the
8499 * only way qgroup updates are added. However,
8500 * we may have added updates during our tree
8501 * walk so run qgroups here to make sure we
8502 * don't lose any updates.
8504 ret
= btrfs_delayed_qgroup_accounting(trans
,
8507 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8508 "running qgroup updates "
8509 "during snapshot delete. "
8510 "Quota is out of sync, "
8511 "rescan required.\n", ret
);
8513 btrfs_end_transaction_throttle(trans
, tree_root
);
8514 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8515 pr_debug("BTRFS: drop snapshot early exit\n");
8520 trans
= btrfs_start_transaction(tree_root
, 0);
8521 if (IS_ERR(trans
)) {
8522 err
= PTR_ERR(trans
);
8526 trans
->block_rsv
= block_rsv
;
8529 btrfs_release_path(path
);
8533 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8535 btrfs_abort_transaction(trans
, tree_root
, ret
);
8539 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8540 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8543 btrfs_abort_transaction(trans
, tree_root
, ret
);
8546 } else if (ret
> 0) {
8547 /* if we fail to delete the orphan item this time
8548 * around, it'll get picked up the next time.
8550 * The most common failure here is just -ENOENT.
8552 btrfs_del_orphan_item(trans
, tree_root
,
8553 root
->root_key
.objectid
);
8557 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8558 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8560 free_extent_buffer(root
->node
);
8561 free_extent_buffer(root
->commit_root
);
8562 btrfs_put_fs_root(root
);
8564 root_dropped
= true;
8566 ret
= btrfs_delayed_qgroup_accounting(trans
, tree_root
->fs_info
);
8568 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8569 "running qgroup updates "
8570 "during snapshot delete. "
8571 "Quota is out of sync, "
8572 "rescan required.\n", ret
);
8574 btrfs_end_transaction_throttle(trans
, tree_root
);
8577 btrfs_free_path(path
);
8580 * So if we need to stop dropping the snapshot for whatever reason we
8581 * need to make sure to add it back to the dead root list so that we
8582 * keep trying to do the work later. This also cleans up roots if we
8583 * don't have it in the radix (like when we recover after a power fail
8584 * or unmount) so we don't leak memory.
8586 if (!for_reloc
&& root_dropped
== false)
8587 btrfs_add_dead_root(root
);
8588 if (err
&& err
!= -EAGAIN
)
8589 btrfs_std_error(root
->fs_info
, err
);
8594 * drop subtree rooted at tree block 'node'.
8596 * NOTE: this function will unlock and release tree block 'node'
8597 * only used by relocation code
8599 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8600 struct btrfs_root
*root
,
8601 struct extent_buffer
*node
,
8602 struct extent_buffer
*parent
)
8604 struct btrfs_path
*path
;
8605 struct walk_control
*wc
;
8611 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8613 path
= btrfs_alloc_path();
8617 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8619 btrfs_free_path(path
);
8623 btrfs_assert_tree_locked(parent
);
8624 parent_level
= btrfs_header_level(parent
);
8625 extent_buffer_get(parent
);
8626 path
->nodes
[parent_level
] = parent
;
8627 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8629 btrfs_assert_tree_locked(node
);
8630 level
= btrfs_header_level(node
);
8631 path
->nodes
[level
] = node
;
8632 path
->slots
[level
] = 0;
8633 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8635 wc
->refs
[parent_level
] = 1;
8636 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8638 wc
->shared_level
= -1;
8639 wc
->stage
= DROP_REFERENCE
;
8643 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8646 wret
= walk_down_tree(trans
, root
, path
, wc
);
8652 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8660 btrfs_free_path(path
);
8664 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8670 * if restripe for this chunk_type is on pick target profile and
8671 * return, otherwise do the usual balance
8673 stripped
= get_restripe_target(root
->fs_info
, flags
);
8675 return extended_to_chunk(stripped
);
8677 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8679 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8680 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8681 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8683 if (num_devices
== 1) {
8684 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8685 stripped
= flags
& ~stripped
;
8687 /* turn raid0 into single device chunks */
8688 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8691 /* turn mirroring into duplication */
8692 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8693 BTRFS_BLOCK_GROUP_RAID10
))
8694 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8696 /* they already had raid on here, just return */
8697 if (flags
& stripped
)
8700 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8701 stripped
= flags
& ~stripped
;
8703 /* switch duplicated blocks with raid1 */
8704 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8705 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8707 /* this is drive concat, leave it alone */
8713 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8715 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8717 u64 min_allocable_bytes
;
8722 * We need some metadata space and system metadata space for
8723 * allocating chunks in some corner cases until we force to set
8724 * it to be readonly.
8727 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8729 min_allocable_bytes
= 1 * 1024 * 1024;
8731 min_allocable_bytes
= 0;
8733 spin_lock(&sinfo
->lock
);
8734 spin_lock(&cache
->lock
);
8741 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8742 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8744 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8745 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8746 min_allocable_bytes
<= sinfo
->total_bytes
) {
8747 sinfo
->bytes_readonly
+= num_bytes
;
8749 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8753 spin_unlock(&cache
->lock
);
8754 spin_unlock(&sinfo
->lock
);
8758 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8759 struct btrfs_block_group_cache
*cache
)
8762 struct btrfs_trans_handle
*trans
;
8769 trans
= btrfs_join_transaction(root
);
8771 return PTR_ERR(trans
);
8774 * we're not allowed to set block groups readonly after the dirty
8775 * block groups cache has started writing. If it already started,
8776 * back off and let this transaction commit
8778 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
8779 if (trans
->transaction
->dirty_bg_run
) {
8780 u64 transid
= trans
->transid
;
8782 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8783 btrfs_end_transaction(trans
, root
);
8785 ret
= btrfs_wait_for_commit(root
, transid
);
8792 ret
= set_block_group_ro(cache
, 0);
8795 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8796 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8800 ret
= set_block_group_ro(cache
, 0);
8802 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
8803 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8804 check_system_chunk(trans
, root
, alloc_flags
);
8806 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8808 btrfs_end_transaction(trans
, root
);
8812 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8813 struct btrfs_root
*root
, u64 type
)
8815 u64 alloc_flags
= get_alloc_profile(root
, type
);
8816 return do_chunk_alloc(trans
, root
, alloc_flags
,
8821 * helper to account the unused space of all the readonly block group in the
8822 * space_info. takes mirrors into account.
8824 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8826 struct btrfs_block_group_cache
*block_group
;
8830 /* It's df, we don't care if it's racey */
8831 if (list_empty(&sinfo
->ro_bgs
))
8834 spin_lock(&sinfo
->lock
);
8835 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8836 spin_lock(&block_group
->lock
);
8838 if (!block_group
->ro
) {
8839 spin_unlock(&block_group
->lock
);
8843 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8844 BTRFS_BLOCK_GROUP_RAID10
|
8845 BTRFS_BLOCK_GROUP_DUP
))
8850 free_bytes
+= (block_group
->key
.offset
-
8851 btrfs_block_group_used(&block_group
->item
)) *
8854 spin_unlock(&block_group
->lock
);
8856 spin_unlock(&sinfo
->lock
);
8861 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8862 struct btrfs_block_group_cache
*cache
)
8864 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8869 spin_lock(&sinfo
->lock
);
8870 spin_lock(&cache
->lock
);
8871 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8872 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8873 sinfo
->bytes_readonly
-= num_bytes
;
8875 list_del_init(&cache
->ro_list
);
8876 spin_unlock(&cache
->lock
);
8877 spin_unlock(&sinfo
->lock
);
8881 * checks to see if its even possible to relocate this block group.
8883 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8884 * ok to go ahead and try.
8886 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8888 struct btrfs_block_group_cache
*block_group
;
8889 struct btrfs_space_info
*space_info
;
8890 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8891 struct btrfs_device
*device
;
8892 struct btrfs_trans_handle
*trans
;
8901 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8903 /* odd, couldn't find the block group, leave it alone */
8907 min_free
= btrfs_block_group_used(&block_group
->item
);
8909 /* no bytes used, we're good */
8913 space_info
= block_group
->space_info
;
8914 spin_lock(&space_info
->lock
);
8916 full
= space_info
->full
;
8919 * if this is the last block group we have in this space, we can't
8920 * relocate it unless we're able to allocate a new chunk below.
8922 * Otherwise, we need to make sure we have room in the space to handle
8923 * all of the extents from this block group. If we can, we're good
8925 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8926 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8927 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8928 min_free
< space_info
->total_bytes
)) {
8929 spin_unlock(&space_info
->lock
);
8932 spin_unlock(&space_info
->lock
);
8935 * ok we don't have enough space, but maybe we have free space on our
8936 * devices to allocate new chunks for relocation, so loop through our
8937 * alloc devices and guess if we have enough space. if this block
8938 * group is going to be restriped, run checks against the target
8939 * profile instead of the current one.
8951 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8953 index
= __get_raid_index(extended_to_chunk(target
));
8956 * this is just a balance, so if we were marked as full
8957 * we know there is no space for a new chunk
8962 index
= get_block_group_index(block_group
);
8965 if (index
== BTRFS_RAID_RAID10
) {
8969 } else if (index
== BTRFS_RAID_RAID1
) {
8971 } else if (index
== BTRFS_RAID_DUP
) {
8974 } else if (index
== BTRFS_RAID_RAID0
) {
8975 dev_min
= fs_devices
->rw_devices
;
8976 min_free
= div64_u64(min_free
, dev_min
);
8979 /* We need to do this so that we can look at pending chunks */
8980 trans
= btrfs_join_transaction(root
);
8981 if (IS_ERR(trans
)) {
8982 ret
= PTR_ERR(trans
);
8986 mutex_lock(&root
->fs_info
->chunk_mutex
);
8987 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8991 * check to make sure we can actually find a chunk with enough
8992 * space to fit our block group in.
8994 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8995 !device
->is_tgtdev_for_dev_replace
) {
8996 ret
= find_free_dev_extent(trans
, device
, min_free
,
9001 if (dev_nr
>= dev_min
)
9007 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9008 btrfs_end_transaction(trans
, root
);
9010 btrfs_put_block_group(block_group
);
9014 static int find_first_block_group(struct btrfs_root
*root
,
9015 struct btrfs_path
*path
, struct btrfs_key
*key
)
9018 struct btrfs_key found_key
;
9019 struct extent_buffer
*leaf
;
9022 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9027 slot
= path
->slots
[0];
9028 leaf
= path
->nodes
[0];
9029 if (slot
>= btrfs_header_nritems(leaf
)) {
9030 ret
= btrfs_next_leaf(root
, path
);
9037 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9039 if (found_key
.objectid
>= key
->objectid
&&
9040 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9050 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9052 struct btrfs_block_group_cache
*block_group
;
9056 struct inode
*inode
;
9058 block_group
= btrfs_lookup_first_block_group(info
, last
);
9059 while (block_group
) {
9060 spin_lock(&block_group
->lock
);
9061 if (block_group
->iref
)
9063 spin_unlock(&block_group
->lock
);
9064 block_group
= next_block_group(info
->tree_root
,
9074 inode
= block_group
->inode
;
9075 block_group
->iref
= 0;
9076 block_group
->inode
= NULL
;
9077 spin_unlock(&block_group
->lock
);
9079 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9080 btrfs_put_block_group(block_group
);
9084 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9086 struct btrfs_block_group_cache
*block_group
;
9087 struct btrfs_space_info
*space_info
;
9088 struct btrfs_caching_control
*caching_ctl
;
9091 down_write(&info
->commit_root_sem
);
9092 while (!list_empty(&info
->caching_block_groups
)) {
9093 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9094 struct btrfs_caching_control
, list
);
9095 list_del(&caching_ctl
->list
);
9096 put_caching_control(caching_ctl
);
9098 up_write(&info
->commit_root_sem
);
9100 spin_lock(&info
->unused_bgs_lock
);
9101 while (!list_empty(&info
->unused_bgs
)) {
9102 block_group
= list_first_entry(&info
->unused_bgs
,
9103 struct btrfs_block_group_cache
,
9105 list_del_init(&block_group
->bg_list
);
9106 btrfs_put_block_group(block_group
);
9108 spin_unlock(&info
->unused_bgs_lock
);
9110 spin_lock(&info
->block_group_cache_lock
);
9111 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9112 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9114 rb_erase(&block_group
->cache_node
,
9115 &info
->block_group_cache_tree
);
9116 RB_CLEAR_NODE(&block_group
->cache_node
);
9117 spin_unlock(&info
->block_group_cache_lock
);
9119 down_write(&block_group
->space_info
->groups_sem
);
9120 list_del(&block_group
->list
);
9121 up_write(&block_group
->space_info
->groups_sem
);
9123 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9124 wait_block_group_cache_done(block_group
);
9127 * We haven't cached this block group, which means we could
9128 * possibly have excluded extents on this block group.
9130 if (block_group
->cached
== BTRFS_CACHE_NO
||
9131 block_group
->cached
== BTRFS_CACHE_ERROR
)
9132 free_excluded_extents(info
->extent_root
, block_group
);
9134 btrfs_remove_free_space_cache(block_group
);
9135 btrfs_put_block_group(block_group
);
9137 spin_lock(&info
->block_group_cache_lock
);
9139 spin_unlock(&info
->block_group_cache_lock
);
9141 /* now that all the block groups are freed, go through and
9142 * free all the space_info structs. This is only called during
9143 * the final stages of unmount, and so we know nobody is
9144 * using them. We call synchronize_rcu() once before we start,
9145 * just to be on the safe side.
9149 release_global_block_rsv(info
);
9151 while (!list_empty(&info
->space_info
)) {
9154 space_info
= list_entry(info
->space_info
.next
,
9155 struct btrfs_space_info
,
9157 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
9158 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9159 space_info
->bytes_reserved
> 0 ||
9160 space_info
->bytes_may_use
> 0)) {
9161 dump_space_info(space_info
, 0, 0);
9164 list_del(&space_info
->list
);
9165 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9166 struct kobject
*kobj
;
9167 kobj
= space_info
->block_group_kobjs
[i
];
9168 space_info
->block_group_kobjs
[i
] = NULL
;
9174 kobject_del(&space_info
->kobj
);
9175 kobject_put(&space_info
->kobj
);
9180 static void __link_block_group(struct btrfs_space_info
*space_info
,
9181 struct btrfs_block_group_cache
*cache
)
9183 int index
= get_block_group_index(cache
);
9186 down_write(&space_info
->groups_sem
);
9187 if (list_empty(&space_info
->block_groups
[index
]))
9189 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9190 up_write(&space_info
->groups_sem
);
9193 struct raid_kobject
*rkobj
;
9196 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9199 rkobj
->raid_type
= index
;
9200 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9201 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9202 "%s", get_raid_name(index
));
9204 kobject_put(&rkobj
->kobj
);
9207 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9212 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9215 static struct btrfs_block_group_cache
*
9216 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9218 struct btrfs_block_group_cache
*cache
;
9220 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9224 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9226 if (!cache
->free_space_ctl
) {
9231 cache
->key
.objectid
= start
;
9232 cache
->key
.offset
= size
;
9233 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9235 cache
->sectorsize
= root
->sectorsize
;
9236 cache
->fs_info
= root
->fs_info
;
9237 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9238 &root
->fs_info
->mapping_tree
,
9240 atomic_set(&cache
->count
, 1);
9241 spin_lock_init(&cache
->lock
);
9242 init_rwsem(&cache
->data_rwsem
);
9243 INIT_LIST_HEAD(&cache
->list
);
9244 INIT_LIST_HEAD(&cache
->cluster_list
);
9245 INIT_LIST_HEAD(&cache
->bg_list
);
9246 INIT_LIST_HEAD(&cache
->ro_list
);
9247 INIT_LIST_HEAD(&cache
->dirty_list
);
9248 INIT_LIST_HEAD(&cache
->io_list
);
9249 btrfs_init_free_space_ctl(cache
);
9250 atomic_set(&cache
->trimming
, 0);
9255 int btrfs_read_block_groups(struct btrfs_root
*root
)
9257 struct btrfs_path
*path
;
9259 struct btrfs_block_group_cache
*cache
;
9260 struct btrfs_fs_info
*info
= root
->fs_info
;
9261 struct btrfs_space_info
*space_info
;
9262 struct btrfs_key key
;
9263 struct btrfs_key found_key
;
9264 struct extent_buffer
*leaf
;
9268 root
= info
->extent_root
;
9271 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9272 path
= btrfs_alloc_path();
9277 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9278 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9279 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9281 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9285 ret
= find_first_block_group(root
, path
, &key
);
9291 leaf
= path
->nodes
[0];
9292 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9294 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9303 * When we mount with old space cache, we need to
9304 * set BTRFS_DC_CLEAR and set dirty flag.
9306 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9307 * truncate the old free space cache inode and
9309 * b) Setting 'dirty flag' makes sure that we flush
9310 * the new space cache info onto disk.
9312 if (btrfs_test_opt(root
, SPACE_CACHE
))
9313 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9316 read_extent_buffer(leaf
, &cache
->item
,
9317 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9318 sizeof(cache
->item
));
9319 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9321 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9322 btrfs_release_path(path
);
9325 * We need to exclude the super stripes now so that the space
9326 * info has super bytes accounted for, otherwise we'll think
9327 * we have more space than we actually do.
9329 ret
= exclude_super_stripes(root
, cache
);
9332 * We may have excluded something, so call this just in
9335 free_excluded_extents(root
, cache
);
9336 btrfs_put_block_group(cache
);
9341 * check for two cases, either we are full, and therefore
9342 * don't need to bother with the caching work since we won't
9343 * find any space, or we are empty, and we can just add all
9344 * the space in and be done with it. This saves us _alot_ of
9345 * time, particularly in the full case.
9347 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9348 cache
->last_byte_to_unpin
= (u64
)-1;
9349 cache
->cached
= BTRFS_CACHE_FINISHED
;
9350 free_excluded_extents(root
, cache
);
9351 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9352 cache
->last_byte_to_unpin
= (u64
)-1;
9353 cache
->cached
= BTRFS_CACHE_FINISHED
;
9354 add_new_free_space(cache
, root
->fs_info
,
9356 found_key
.objectid
+
9358 free_excluded_extents(root
, cache
);
9361 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9363 btrfs_remove_free_space_cache(cache
);
9364 btrfs_put_block_group(cache
);
9368 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9369 btrfs_block_group_used(&cache
->item
),
9372 btrfs_remove_free_space_cache(cache
);
9373 spin_lock(&info
->block_group_cache_lock
);
9374 rb_erase(&cache
->cache_node
,
9375 &info
->block_group_cache_tree
);
9376 RB_CLEAR_NODE(&cache
->cache_node
);
9377 spin_unlock(&info
->block_group_cache_lock
);
9378 btrfs_put_block_group(cache
);
9382 cache
->space_info
= space_info
;
9383 spin_lock(&cache
->space_info
->lock
);
9384 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9385 spin_unlock(&cache
->space_info
->lock
);
9387 __link_block_group(space_info
, cache
);
9389 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9390 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9391 set_block_group_ro(cache
, 1);
9392 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9393 spin_lock(&info
->unused_bgs_lock
);
9394 /* Should always be true but just in case. */
9395 if (list_empty(&cache
->bg_list
)) {
9396 btrfs_get_block_group(cache
);
9397 list_add_tail(&cache
->bg_list
,
9400 spin_unlock(&info
->unused_bgs_lock
);
9404 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9405 if (!(get_alloc_profile(root
, space_info
->flags
) &
9406 (BTRFS_BLOCK_GROUP_RAID10
|
9407 BTRFS_BLOCK_GROUP_RAID1
|
9408 BTRFS_BLOCK_GROUP_RAID5
|
9409 BTRFS_BLOCK_GROUP_RAID6
|
9410 BTRFS_BLOCK_GROUP_DUP
)))
9413 * avoid allocating from un-mirrored block group if there are
9414 * mirrored block groups.
9416 list_for_each_entry(cache
,
9417 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9419 set_block_group_ro(cache
, 1);
9420 list_for_each_entry(cache
,
9421 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9423 set_block_group_ro(cache
, 1);
9426 init_global_block_rsv(info
);
9429 btrfs_free_path(path
);
9433 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9434 struct btrfs_root
*root
)
9436 struct btrfs_block_group_cache
*block_group
, *tmp
;
9437 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9438 struct btrfs_block_group_item item
;
9439 struct btrfs_key key
;
9442 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9446 spin_lock(&block_group
->lock
);
9447 memcpy(&item
, &block_group
->item
, sizeof(item
));
9448 memcpy(&key
, &block_group
->key
, sizeof(key
));
9449 spin_unlock(&block_group
->lock
);
9451 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9454 btrfs_abort_transaction(trans
, extent_root
, ret
);
9455 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9456 key
.objectid
, key
.offset
);
9458 btrfs_abort_transaction(trans
, extent_root
, ret
);
9460 list_del_init(&block_group
->bg_list
);
9464 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9465 struct btrfs_root
*root
, u64 bytes_used
,
9466 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9470 struct btrfs_root
*extent_root
;
9471 struct btrfs_block_group_cache
*cache
;
9473 extent_root
= root
->fs_info
->extent_root
;
9475 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9477 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9481 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9482 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9483 btrfs_set_block_group_flags(&cache
->item
, type
);
9485 cache
->flags
= type
;
9486 cache
->last_byte_to_unpin
= (u64
)-1;
9487 cache
->cached
= BTRFS_CACHE_FINISHED
;
9488 ret
= exclude_super_stripes(root
, cache
);
9491 * We may have excluded something, so call this just in
9494 free_excluded_extents(root
, cache
);
9495 btrfs_put_block_group(cache
);
9499 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9500 chunk_offset
+ size
);
9502 free_excluded_extents(root
, cache
);
9504 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9506 btrfs_remove_free_space_cache(cache
);
9507 btrfs_put_block_group(cache
);
9511 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9512 &cache
->space_info
);
9514 btrfs_remove_free_space_cache(cache
);
9515 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9516 rb_erase(&cache
->cache_node
,
9517 &root
->fs_info
->block_group_cache_tree
);
9518 RB_CLEAR_NODE(&cache
->cache_node
);
9519 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9520 btrfs_put_block_group(cache
);
9523 update_global_block_rsv(root
->fs_info
);
9525 spin_lock(&cache
->space_info
->lock
);
9526 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9527 spin_unlock(&cache
->space_info
->lock
);
9529 __link_block_group(cache
->space_info
, cache
);
9531 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9533 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9538 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9540 u64 extra_flags
= chunk_to_extended(flags
) &
9541 BTRFS_EXTENDED_PROFILE_MASK
;
9543 write_seqlock(&fs_info
->profiles_lock
);
9544 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9545 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9546 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9547 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9548 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9549 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9550 write_sequnlock(&fs_info
->profiles_lock
);
9553 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9554 struct btrfs_root
*root
, u64 group_start
,
9555 struct extent_map
*em
)
9557 struct btrfs_path
*path
;
9558 struct btrfs_block_group_cache
*block_group
;
9559 struct btrfs_free_cluster
*cluster
;
9560 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9561 struct btrfs_key key
;
9562 struct inode
*inode
;
9563 struct kobject
*kobj
= NULL
;
9567 struct btrfs_caching_control
*caching_ctl
= NULL
;
9570 root
= root
->fs_info
->extent_root
;
9572 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9573 BUG_ON(!block_group
);
9574 BUG_ON(!block_group
->ro
);
9577 * Free the reserved super bytes from this block group before
9580 free_excluded_extents(root
, block_group
);
9582 memcpy(&key
, &block_group
->key
, sizeof(key
));
9583 index
= get_block_group_index(block_group
);
9584 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9585 BTRFS_BLOCK_GROUP_RAID1
|
9586 BTRFS_BLOCK_GROUP_RAID10
))
9591 /* make sure this block group isn't part of an allocation cluster */
9592 cluster
= &root
->fs_info
->data_alloc_cluster
;
9593 spin_lock(&cluster
->refill_lock
);
9594 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9595 spin_unlock(&cluster
->refill_lock
);
9598 * make sure this block group isn't part of a metadata
9599 * allocation cluster
9601 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9602 spin_lock(&cluster
->refill_lock
);
9603 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9604 spin_unlock(&cluster
->refill_lock
);
9606 path
= btrfs_alloc_path();
9613 * get the inode first so any iput calls done for the io_list
9614 * aren't the final iput (no unlinks allowed now)
9616 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9618 mutex_lock(&trans
->transaction
->cache_write_mutex
);
9620 * make sure our free spache cache IO is done before remove the
9623 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9624 if (!list_empty(&block_group
->io_list
)) {
9625 list_del_init(&block_group
->io_list
);
9627 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
9629 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9630 btrfs_wait_cache_io(root
, trans
, block_group
,
9631 &block_group
->io_ctl
, path
,
9632 block_group
->key
.objectid
);
9633 btrfs_put_block_group(block_group
);
9634 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9637 if (!list_empty(&block_group
->dirty_list
)) {
9638 list_del_init(&block_group
->dirty_list
);
9639 btrfs_put_block_group(block_group
);
9641 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9642 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
9644 if (!IS_ERR(inode
)) {
9645 ret
= btrfs_orphan_add(trans
, inode
);
9647 btrfs_add_delayed_iput(inode
);
9651 /* One for the block groups ref */
9652 spin_lock(&block_group
->lock
);
9653 if (block_group
->iref
) {
9654 block_group
->iref
= 0;
9655 block_group
->inode
= NULL
;
9656 spin_unlock(&block_group
->lock
);
9659 spin_unlock(&block_group
->lock
);
9661 /* One for our lookup ref */
9662 btrfs_add_delayed_iput(inode
);
9665 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9666 key
.offset
= block_group
->key
.objectid
;
9669 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9673 btrfs_release_path(path
);
9675 ret
= btrfs_del_item(trans
, tree_root
, path
);
9678 btrfs_release_path(path
);
9681 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9682 rb_erase(&block_group
->cache_node
,
9683 &root
->fs_info
->block_group_cache_tree
);
9684 RB_CLEAR_NODE(&block_group
->cache_node
);
9686 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9687 root
->fs_info
->first_logical_byte
= (u64
)-1;
9688 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9690 down_write(&block_group
->space_info
->groups_sem
);
9692 * we must use list_del_init so people can check to see if they
9693 * are still on the list after taking the semaphore
9695 list_del_init(&block_group
->list
);
9696 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9697 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9698 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9699 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9701 up_write(&block_group
->space_info
->groups_sem
);
9707 if (block_group
->has_caching_ctl
)
9708 caching_ctl
= get_caching_control(block_group
);
9709 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9710 wait_block_group_cache_done(block_group
);
9711 if (block_group
->has_caching_ctl
) {
9712 down_write(&root
->fs_info
->commit_root_sem
);
9714 struct btrfs_caching_control
*ctl
;
9716 list_for_each_entry(ctl
,
9717 &root
->fs_info
->caching_block_groups
, list
)
9718 if (ctl
->block_group
== block_group
) {
9720 atomic_inc(&caching_ctl
->count
);
9725 list_del_init(&caching_ctl
->list
);
9726 up_write(&root
->fs_info
->commit_root_sem
);
9728 /* Once for the caching bgs list and once for us. */
9729 put_caching_control(caching_ctl
);
9730 put_caching_control(caching_ctl
);
9734 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9735 if (!list_empty(&block_group
->dirty_list
)) {
9738 if (!list_empty(&block_group
->io_list
)) {
9741 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9742 btrfs_remove_free_space_cache(block_group
);
9744 spin_lock(&block_group
->space_info
->lock
);
9745 list_del_init(&block_group
->ro_list
);
9747 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
9748 WARN_ON(block_group
->space_info
->total_bytes
9749 < block_group
->key
.offset
);
9750 WARN_ON(block_group
->space_info
->bytes_readonly
9751 < block_group
->key
.offset
);
9752 WARN_ON(block_group
->space_info
->disk_total
9753 < block_group
->key
.offset
* factor
);
9755 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9756 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9757 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9759 spin_unlock(&block_group
->space_info
->lock
);
9761 memcpy(&key
, &block_group
->key
, sizeof(key
));
9764 if (!list_empty(&em
->list
)) {
9765 /* We're in the transaction->pending_chunks list. */
9766 free_extent_map(em
);
9768 spin_lock(&block_group
->lock
);
9769 block_group
->removed
= 1;
9771 * At this point trimming can't start on this block group, because we
9772 * removed the block group from the tree fs_info->block_group_cache_tree
9773 * so no one can't find it anymore and even if someone already got this
9774 * block group before we removed it from the rbtree, they have already
9775 * incremented block_group->trimming - if they didn't, they won't find
9776 * any free space entries because we already removed them all when we
9777 * called btrfs_remove_free_space_cache().
9779 * And we must not remove the extent map from the fs_info->mapping_tree
9780 * to prevent the same logical address range and physical device space
9781 * ranges from being reused for a new block group. This is because our
9782 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9783 * completely transactionless, so while it is trimming a range the
9784 * currently running transaction might finish and a new one start,
9785 * allowing for new block groups to be created that can reuse the same
9786 * physical device locations unless we take this special care.
9788 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9790 * Make sure a trimmer task always sees the em in the pinned_chunks list
9791 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9792 * before checking block_group->removed).
9796 * Our em might be in trans->transaction->pending_chunks which
9797 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9798 * and so is the fs_info->pinned_chunks list.
9800 * So at this point we must be holding the chunk_mutex to avoid
9801 * any races with chunk allocation (more specifically at
9802 * volumes.c:contains_pending_extent()), to ensure it always
9803 * sees the em, either in the pending_chunks list or in the
9804 * pinned_chunks list.
9806 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9808 spin_unlock(&block_group
->lock
);
9811 struct extent_map_tree
*em_tree
;
9813 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9814 write_lock(&em_tree
->lock
);
9816 * The em might be in the pending_chunks list, so make sure the
9817 * chunk mutex is locked, since remove_extent_mapping() will
9818 * delete us from that list.
9820 remove_extent_mapping(em_tree
, em
);
9821 write_unlock(&em_tree
->lock
);
9822 /* once for the tree */
9823 free_extent_map(em
);
9826 unlock_chunks(root
);
9828 btrfs_put_block_group(block_group
);
9829 btrfs_put_block_group(block_group
);
9831 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9837 ret
= btrfs_del_item(trans
, root
, path
);
9839 btrfs_free_path(path
);
9844 * Process the unused_bgs list and remove any that don't have any allocated
9845 * space inside of them.
9847 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9849 struct btrfs_block_group_cache
*block_group
;
9850 struct btrfs_space_info
*space_info
;
9851 struct btrfs_root
*root
= fs_info
->extent_root
;
9852 struct btrfs_trans_handle
*trans
;
9858 spin_lock(&fs_info
->unused_bgs_lock
);
9859 while (!list_empty(&fs_info
->unused_bgs
)) {
9862 block_group
= list_first_entry(&fs_info
->unused_bgs
,
9863 struct btrfs_block_group_cache
,
9865 space_info
= block_group
->space_info
;
9866 list_del_init(&block_group
->bg_list
);
9867 if (ret
|| btrfs_mixed_space_info(space_info
)) {
9868 btrfs_put_block_group(block_group
);
9871 spin_unlock(&fs_info
->unused_bgs_lock
);
9873 /* Don't want to race with allocators so take the groups_sem */
9874 down_write(&space_info
->groups_sem
);
9875 spin_lock(&block_group
->lock
);
9876 if (block_group
->reserved
||
9877 btrfs_block_group_used(&block_group
->item
) ||
9880 * We want to bail if we made new allocations or have
9881 * outstanding allocations in this block group. We do
9882 * the ro check in case balance is currently acting on
9885 spin_unlock(&block_group
->lock
);
9886 up_write(&space_info
->groups_sem
);
9889 spin_unlock(&block_group
->lock
);
9891 /* We don't want to force the issue, only flip if it's ok. */
9892 ret
= set_block_group_ro(block_group
, 0);
9893 up_write(&space_info
->groups_sem
);
9900 * Want to do this before we do anything else so we can recover
9901 * properly if we fail to join the transaction.
9903 /* 1 for btrfs_orphan_reserve_metadata() */
9904 trans
= btrfs_start_transaction(root
, 1);
9905 if (IS_ERR(trans
)) {
9906 btrfs_set_block_group_rw(root
, block_group
);
9907 ret
= PTR_ERR(trans
);
9912 * We could have pending pinned extents for this block group,
9913 * just delete them, we don't care about them anymore.
9915 start
= block_group
->key
.objectid
;
9916 end
= start
+ block_group
->key
.offset
- 1;
9918 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9919 * btrfs_finish_extent_commit(). If we are at transaction N,
9920 * another task might be running finish_extent_commit() for the
9921 * previous transaction N - 1, and have seen a range belonging
9922 * to the block group in freed_extents[] before we were able to
9923 * clear the whole block group range from freed_extents[]. This
9924 * means that task can lookup for the block group after we
9925 * unpinned it from freed_extents[] and removed it, leading to
9926 * a BUG_ON() at btrfs_unpin_extent_range().
9928 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
9929 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
9930 EXTENT_DIRTY
, GFP_NOFS
);
9932 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9933 btrfs_set_block_group_rw(root
, block_group
);
9936 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
9937 EXTENT_DIRTY
, GFP_NOFS
);
9939 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9940 btrfs_set_block_group_rw(root
, block_group
);
9943 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9945 /* Reset pinned so btrfs_put_block_group doesn't complain */
9946 spin_lock(&space_info
->lock
);
9947 spin_lock(&block_group
->lock
);
9949 space_info
->bytes_pinned
-= block_group
->pinned
;
9950 space_info
->bytes_readonly
+= block_group
->pinned
;
9951 percpu_counter_add(&space_info
->total_bytes_pinned
,
9952 -block_group
->pinned
);
9953 block_group
->pinned
= 0;
9955 spin_unlock(&block_group
->lock
);
9956 spin_unlock(&space_info
->lock
);
9959 * Btrfs_remove_chunk will abort the transaction if things go
9962 ret
= btrfs_remove_chunk(trans
, root
,
9963 block_group
->key
.objectid
);
9965 btrfs_end_transaction(trans
, root
);
9967 btrfs_put_block_group(block_group
);
9968 spin_lock(&fs_info
->unused_bgs_lock
);
9970 spin_unlock(&fs_info
->unused_bgs_lock
);
9973 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
9975 struct btrfs_space_info
*space_info
;
9976 struct btrfs_super_block
*disk_super
;
9982 disk_super
= fs_info
->super_copy
;
9983 if (!btrfs_super_root(disk_super
))
9986 features
= btrfs_super_incompat_flags(disk_super
);
9987 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
9990 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
9991 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9996 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
9997 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9999 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10000 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10004 flags
= BTRFS_BLOCK_GROUP_DATA
;
10005 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10011 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10013 return unpin_extent_range(root
, start
, end
, false);
10016 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
10018 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10019 struct btrfs_block_group_cache
*cache
= NULL
;
10024 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10028 * try to trim all FS space, our block group may start from non-zero.
10030 if (range
->len
== total_bytes
)
10031 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10033 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10036 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10037 btrfs_put_block_group(cache
);
10041 start
= max(range
->start
, cache
->key
.objectid
);
10042 end
= min(range
->start
+ range
->len
,
10043 cache
->key
.objectid
+ cache
->key
.offset
);
10045 if (end
- start
>= range
->minlen
) {
10046 if (!block_group_cache_done(cache
)) {
10047 ret
= cache_block_group(cache
, 0);
10049 btrfs_put_block_group(cache
);
10052 ret
= wait_block_group_cache_done(cache
);
10054 btrfs_put_block_group(cache
);
10058 ret
= btrfs_trim_block_group(cache
,
10064 trimmed
+= group_trimmed
;
10066 btrfs_put_block_group(cache
);
10071 cache
= next_block_group(fs_info
->tree_root
, cache
);
10074 range
->len
= trimmed
;
10079 * btrfs_{start,end}_write_no_snapshoting() are similar to
10080 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10081 * data into the page cache through nocow before the subvolume is snapshoted,
10082 * but flush the data into disk after the snapshot creation, or to prevent
10083 * operations while snapshoting is ongoing and that cause the snapshot to be
10084 * inconsistent (writes followed by expanding truncates for example).
10086 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10088 percpu_counter_dec(&root
->subv_writers
->counter
);
10090 * Make sure counter is updated before we wake up
10094 if (waitqueue_active(&root
->subv_writers
->wait
))
10095 wake_up(&root
->subv_writers
->wait
);
10098 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10100 if (atomic_read(&root
->will_be_snapshoted
))
10103 percpu_counter_inc(&root
->subv_writers
->counter
);
10105 * Make sure counter is updated before we check for snapshot creation.
10108 if (atomic_read(&root
->will_be_snapshoted
)) {
10109 btrfs_end_write_no_snapshoting(root
);