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 struct btrfs_delayed_ref_node
*node
, 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
);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
87 struct extent_buffer
*leaf
,
88 struct btrfs_extent_item
*ei
);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
90 struct btrfs_root
*root
,
91 u64 parent
, u64 root_objectid
,
92 u64 flags
, u64 owner
, u64 offset
,
93 struct btrfs_key
*ins
, int ref_mod
);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
95 struct btrfs_root
*root
,
96 u64 parent
, u64 root_objectid
,
97 u64 flags
, struct btrfs_disk_key
*key
,
98 int level
, struct btrfs_key
*ins
);
99 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
100 struct btrfs_root
*extent_root
, u64 flags
,
102 static int find_next_key(struct btrfs_path
*path
, int level
,
103 struct btrfs_key
*key
);
104 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
105 int dump_block_groups
);
106 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
107 u64 num_bytes
, int reserve
,
109 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
111 int btrfs_pin_extent(struct btrfs_root
*root
,
112 u64 bytenr
, u64 num_bytes
, int reserved
);
115 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
118 return cache
->cached
== BTRFS_CACHE_FINISHED
||
119 cache
->cached
== BTRFS_CACHE_ERROR
;
122 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
124 return (cache
->flags
& bits
) == bits
;
127 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
129 atomic_inc(&cache
->count
);
132 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
134 if (atomic_dec_and_test(&cache
->count
)) {
135 WARN_ON(cache
->pinned
> 0);
136 WARN_ON(cache
->reserved
> 0);
137 kfree(cache
->free_space_ctl
);
143 * this adds the block group to the fs_info rb tree for the block group
146 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
147 struct btrfs_block_group_cache
*block_group
)
150 struct rb_node
*parent
= NULL
;
151 struct btrfs_block_group_cache
*cache
;
153 spin_lock(&info
->block_group_cache_lock
);
154 p
= &info
->block_group_cache_tree
.rb_node
;
158 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
160 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
162 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
165 spin_unlock(&info
->block_group_cache_lock
);
170 rb_link_node(&block_group
->cache_node
, parent
, p
);
171 rb_insert_color(&block_group
->cache_node
,
172 &info
->block_group_cache_tree
);
174 if (info
->first_logical_byte
> block_group
->key
.objectid
)
175 info
->first_logical_byte
= block_group
->key
.objectid
;
177 spin_unlock(&info
->block_group_cache_lock
);
183 * This will return the block group at or after bytenr if contains is 0, else
184 * it will return the block group that contains the bytenr
186 static struct btrfs_block_group_cache
*
187 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
190 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
194 spin_lock(&info
->block_group_cache_lock
);
195 n
= info
->block_group_cache_tree
.rb_node
;
198 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
200 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
201 start
= cache
->key
.objectid
;
203 if (bytenr
< start
) {
204 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
207 } else if (bytenr
> start
) {
208 if (contains
&& bytenr
<= end
) {
219 btrfs_get_block_group(ret
);
220 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
221 info
->first_logical_byte
= ret
->key
.objectid
;
223 spin_unlock(&info
->block_group_cache_lock
);
228 static int add_excluded_extent(struct btrfs_root
*root
,
229 u64 start
, u64 num_bytes
)
231 u64 end
= start
+ num_bytes
- 1;
232 set_extent_bits(&root
->fs_info
->freed_extents
[0],
233 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
234 set_extent_bits(&root
->fs_info
->freed_extents
[1],
235 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
239 static void free_excluded_extents(struct btrfs_root
*root
,
240 struct btrfs_block_group_cache
*cache
)
244 start
= cache
->key
.objectid
;
245 end
= start
+ cache
->key
.offset
- 1;
247 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
248 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
249 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
250 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
253 static int exclude_super_stripes(struct btrfs_root
*root
,
254 struct btrfs_block_group_cache
*cache
)
261 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
262 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
263 cache
->bytes_super
+= stripe_len
;
264 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
270 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
271 bytenr
= btrfs_sb_offset(i
);
272 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
273 cache
->key
.objectid
, bytenr
,
274 0, &logical
, &nr
, &stripe_len
);
281 if (logical
[nr
] > cache
->key
.objectid
+
285 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
289 if (start
< cache
->key
.objectid
) {
290 start
= cache
->key
.objectid
;
291 len
= (logical
[nr
] + stripe_len
) - start
;
293 len
= min_t(u64
, stripe_len
,
294 cache
->key
.objectid
+
295 cache
->key
.offset
- start
);
298 cache
->bytes_super
+= len
;
299 ret
= add_excluded_extent(root
, start
, len
);
311 static struct btrfs_caching_control
*
312 get_caching_control(struct btrfs_block_group_cache
*cache
)
314 struct btrfs_caching_control
*ctl
;
316 spin_lock(&cache
->lock
);
317 if (!cache
->caching_ctl
) {
318 spin_unlock(&cache
->lock
);
322 ctl
= cache
->caching_ctl
;
323 atomic_inc(&ctl
->count
);
324 spin_unlock(&cache
->lock
);
328 static void put_caching_control(struct btrfs_caching_control
*ctl
)
330 if (atomic_dec_and_test(&ctl
->count
))
334 #ifdef CONFIG_BTRFS_DEBUG
335 static void fragment_free_space(struct btrfs_root
*root
,
336 struct btrfs_block_group_cache
*block_group
)
338 u64 start
= block_group
->key
.objectid
;
339 u64 len
= block_group
->key
.offset
;
340 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
341 root
->nodesize
: root
->sectorsize
;
342 u64 step
= chunk
<< 1;
344 while (len
> chunk
) {
345 btrfs_remove_free_space(block_group
, start
, chunk
);
356 * this is only called by cache_block_group, since we could have freed extents
357 * we need to check the pinned_extents for any extents that can't be used yet
358 * since their free space will be released as soon as the transaction commits.
360 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
361 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
363 u64 extent_start
, extent_end
, size
, total_added
= 0;
366 while (start
< end
) {
367 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
368 &extent_start
, &extent_end
,
369 EXTENT_DIRTY
| EXTENT_UPTODATE
,
374 if (extent_start
<= start
) {
375 start
= extent_end
+ 1;
376 } else if (extent_start
> start
&& extent_start
< end
) {
377 size
= extent_start
- start
;
379 ret
= btrfs_add_free_space(block_group
, start
,
381 BUG_ON(ret
); /* -ENOMEM or logic error */
382 start
= extent_end
+ 1;
391 ret
= btrfs_add_free_space(block_group
, start
, size
);
392 BUG_ON(ret
); /* -ENOMEM or logic error */
398 static noinline
void caching_thread(struct btrfs_work
*work
)
400 struct btrfs_block_group_cache
*block_group
;
401 struct btrfs_fs_info
*fs_info
;
402 struct btrfs_caching_control
*caching_ctl
;
403 struct btrfs_root
*extent_root
;
404 struct btrfs_path
*path
;
405 struct extent_buffer
*leaf
;
406 struct btrfs_key key
;
413 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
414 block_group
= caching_ctl
->block_group
;
415 fs_info
= block_group
->fs_info
;
416 extent_root
= fs_info
->extent_root
;
418 path
= btrfs_alloc_path();
422 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
424 #ifdef CONFIG_BTRFS_DEBUG
426 * If we're fragmenting we don't want to make anybody think we can
427 * allocate from this block group until we've had a chance to fragment
430 if (btrfs_should_fragment_free_space(extent_root
, block_group
))
434 * We don't want to deadlock with somebody trying to allocate a new
435 * extent for the extent root while also trying to search the extent
436 * root to add free space. So we skip locking and search the commit
437 * root, since its read-only
439 path
->skip_locking
= 1;
440 path
->search_commit_root
= 1;
445 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
447 mutex_lock(&caching_ctl
->mutex
);
448 /* need to make sure the commit_root doesn't disappear */
449 down_read(&fs_info
->commit_root_sem
);
452 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
456 leaf
= path
->nodes
[0];
457 nritems
= btrfs_header_nritems(leaf
);
460 if (btrfs_fs_closing(fs_info
) > 1) {
465 if (path
->slots
[0] < nritems
) {
466 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
468 ret
= find_next_key(path
, 0, &key
);
472 if (need_resched() ||
473 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
475 caching_ctl
->progress
= last
;
476 btrfs_release_path(path
);
477 up_read(&fs_info
->commit_root_sem
);
478 mutex_unlock(&caching_ctl
->mutex
);
483 ret
= btrfs_next_leaf(extent_root
, path
);
488 leaf
= path
->nodes
[0];
489 nritems
= btrfs_header_nritems(leaf
);
493 if (key
.objectid
< last
) {
496 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
499 caching_ctl
->progress
= last
;
500 btrfs_release_path(path
);
504 if (key
.objectid
< block_group
->key
.objectid
) {
509 if (key
.objectid
>= block_group
->key
.objectid
+
510 block_group
->key
.offset
)
513 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
514 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
515 total_found
+= add_new_free_space(block_group
,
518 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
519 last
= key
.objectid
+
520 fs_info
->tree_root
->nodesize
;
522 last
= key
.objectid
+ key
.offset
;
524 if (total_found
> (1024 * 1024 * 2)) {
527 wake_up(&caching_ctl
->wait
);
534 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
535 block_group
->key
.objectid
+
536 block_group
->key
.offset
);
537 spin_lock(&block_group
->lock
);
538 block_group
->caching_ctl
= NULL
;
539 block_group
->cached
= BTRFS_CACHE_FINISHED
;
540 spin_unlock(&block_group
->lock
);
542 #ifdef CONFIG_BTRFS_DEBUG
543 if (btrfs_should_fragment_free_space(extent_root
, block_group
)) {
546 spin_lock(&block_group
->space_info
->lock
);
547 spin_lock(&block_group
->lock
);
548 bytes_used
= block_group
->key
.offset
-
549 btrfs_block_group_used(&block_group
->item
);
550 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
551 spin_unlock(&block_group
->lock
);
552 spin_unlock(&block_group
->space_info
->lock
);
553 fragment_free_space(extent_root
, block_group
);
557 caching_ctl
->progress
= (u64
)-1;
559 btrfs_free_path(path
);
560 up_read(&fs_info
->commit_root_sem
);
562 free_excluded_extents(extent_root
, block_group
);
564 mutex_unlock(&caching_ctl
->mutex
);
567 spin_lock(&block_group
->lock
);
568 block_group
->caching_ctl
= NULL
;
569 block_group
->cached
= BTRFS_CACHE_ERROR
;
570 spin_unlock(&block_group
->lock
);
572 wake_up(&caching_ctl
->wait
);
574 put_caching_control(caching_ctl
);
575 btrfs_put_block_group(block_group
);
578 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
582 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
583 struct btrfs_caching_control
*caching_ctl
;
586 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
590 INIT_LIST_HEAD(&caching_ctl
->list
);
591 mutex_init(&caching_ctl
->mutex
);
592 init_waitqueue_head(&caching_ctl
->wait
);
593 caching_ctl
->block_group
= cache
;
594 caching_ctl
->progress
= cache
->key
.objectid
;
595 atomic_set(&caching_ctl
->count
, 1);
596 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
597 caching_thread
, NULL
, NULL
);
599 spin_lock(&cache
->lock
);
601 * This should be a rare occasion, but this could happen I think in the
602 * case where one thread starts to load the space cache info, and then
603 * some other thread starts a transaction commit which tries to do an
604 * allocation while the other thread is still loading the space cache
605 * info. The previous loop should have kept us from choosing this block
606 * group, but if we've moved to the state where we will wait on caching
607 * block groups we need to first check if we're doing a fast load here,
608 * so we can wait for it to finish, otherwise we could end up allocating
609 * from a block group who's cache gets evicted for one reason or
612 while (cache
->cached
== BTRFS_CACHE_FAST
) {
613 struct btrfs_caching_control
*ctl
;
615 ctl
= cache
->caching_ctl
;
616 atomic_inc(&ctl
->count
);
617 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
618 spin_unlock(&cache
->lock
);
622 finish_wait(&ctl
->wait
, &wait
);
623 put_caching_control(ctl
);
624 spin_lock(&cache
->lock
);
627 if (cache
->cached
!= BTRFS_CACHE_NO
) {
628 spin_unlock(&cache
->lock
);
632 WARN_ON(cache
->caching_ctl
);
633 cache
->caching_ctl
= caching_ctl
;
634 cache
->cached
= BTRFS_CACHE_FAST
;
635 spin_unlock(&cache
->lock
);
637 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
638 mutex_lock(&caching_ctl
->mutex
);
639 ret
= load_free_space_cache(fs_info
, cache
);
641 spin_lock(&cache
->lock
);
643 cache
->caching_ctl
= NULL
;
644 cache
->cached
= BTRFS_CACHE_FINISHED
;
645 cache
->last_byte_to_unpin
= (u64
)-1;
646 caching_ctl
->progress
= (u64
)-1;
648 if (load_cache_only
) {
649 cache
->caching_ctl
= NULL
;
650 cache
->cached
= BTRFS_CACHE_NO
;
652 cache
->cached
= BTRFS_CACHE_STARTED
;
653 cache
->has_caching_ctl
= 1;
656 spin_unlock(&cache
->lock
);
657 #ifdef CONFIG_BTRFS_DEBUG
659 btrfs_should_fragment_free_space(fs_info
->extent_root
,
663 spin_lock(&cache
->space_info
->lock
);
664 spin_lock(&cache
->lock
);
665 bytes_used
= cache
->key
.offset
-
666 btrfs_block_group_used(&cache
->item
);
667 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
668 spin_unlock(&cache
->lock
);
669 spin_unlock(&cache
->space_info
->lock
);
670 fragment_free_space(fs_info
->extent_root
, cache
);
673 mutex_unlock(&caching_ctl
->mutex
);
675 wake_up(&caching_ctl
->wait
);
677 put_caching_control(caching_ctl
);
678 free_excluded_extents(fs_info
->extent_root
, cache
);
683 * We are not going to do the fast caching, set cached to the
684 * appropriate value and wakeup any waiters.
686 spin_lock(&cache
->lock
);
687 if (load_cache_only
) {
688 cache
->caching_ctl
= NULL
;
689 cache
->cached
= BTRFS_CACHE_NO
;
691 cache
->cached
= BTRFS_CACHE_STARTED
;
692 cache
->has_caching_ctl
= 1;
694 spin_unlock(&cache
->lock
);
695 wake_up(&caching_ctl
->wait
);
698 if (load_cache_only
) {
699 put_caching_control(caching_ctl
);
703 down_write(&fs_info
->commit_root_sem
);
704 atomic_inc(&caching_ctl
->count
);
705 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
706 up_write(&fs_info
->commit_root_sem
);
708 btrfs_get_block_group(cache
);
710 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
716 * return the block group that starts at or after bytenr
718 static struct btrfs_block_group_cache
*
719 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
721 struct btrfs_block_group_cache
*cache
;
723 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
729 * return the block group that contains the given bytenr
731 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
732 struct btrfs_fs_info
*info
,
735 struct btrfs_block_group_cache
*cache
;
737 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
742 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
745 struct list_head
*head
= &info
->space_info
;
746 struct btrfs_space_info
*found
;
748 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
751 list_for_each_entry_rcu(found
, head
, list
) {
752 if (found
->flags
& flags
) {
762 * after adding space to the filesystem, we need to clear the full flags
763 * on all the space infos.
765 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
767 struct list_head
*head
= &info
->space_info
;
768 struct btrfs_space_info
*found
;
771 list_for_each_entry_rcu(found
, head
, list
)
776 /* simple helper to search for an existing data extent at a given offset */
777 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
780 struct btrfs_key key
;
781 struct btrfs_path
*path
;
783 path
= btrfs_alloc_path();
787 key
.objectid
= start
;
789 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
790 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
792 btrfs_free_path(path
);
797 * helper function to lookup reference count and flags of a tree block.
799 * the head node for delayed ref is used to store the sum of all the
800 * reference count modifications queued up in the rbtree. the head
801 * node may also store the extent flags to set. This way you can check
802 * to see what the reference count and extent flags would be if all of
803 * the delayed refs are not processed.
805 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
806 struct btrfs_root
*root
, u64 bytenr
,
807 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
809 struct btrfs_delayed_ref_head
*head
;
810 struct btrfs_delayed_ref_root
*delayed_refs
;
811 struct btrfs_path
*path
;
812 struct btrfs_extent_item
*ei
;
813 struct extent_buffer
*leaf
;
814 struct btrfs_key key
;
821 * If we don't have skinny metadata, don't bother doing anything
824 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
825 offset
= root
->nodesize
;
829 path
= btrfs_alloc_path();
834 path
->skip_locking
= 1;
835 path
->search_commit_root
= 1;
839 key
.objectid
= bytenr
;
842 key
.type
= BTRFS_METADATA_ITEM_KEY
;
844 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
846 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
851 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
852 if (path
->slots
[0]) {
854 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
856 if (key
.objectid
== bytenr
&&
857 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
858 key
.offset
== root
->nodesize
)
864 leaf
= path
->nodes
[0];
865 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
866 if (item_size
>= sizeof(*ei
)) {
867 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
868 struct btrfs_extent_item
);
869 num_refs
= btrfs_extent_refs(leaf
, ei
);
870 extent_flags
= btrfs_extent_flags(leaf
, ei
);
872 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
873 struct btrfs_extent_item_v0
*ei0
;
874 BUG_ON(item_size
!= sizeof(*ei0
));
875 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
876 struct btrfs_extent_item_v0
);
877 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
878 /* FIXME: this isn't correct for data */
879 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
884 BUG_ON(num_refs
== 0);
894 delayed_refs
= &trans
->transaction
->delayed_refs
;
895 spin_lock(&delayed_refs
->lock
);
896 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
898 if (!mutex_trylock(&head
->mutex
)) {
899 atomic_inc(&head
->node
.refs
);
900 spin_unlock(&delayed_refs
->lock
);
902 btrfs_release_path(path
);
905 * Mutex was contended, block until it's released and try
908 mutex_lock(&head
->mutex
);
909 mutex_unlock(&head
->mutex
);
910 btrfs_put_delayed_ref(&head
->node
);
913 spin_lock(&head
->lock
);
914 if (head
->extent_op
&& head
->extent_op
->update_flags
)
915 extent_flags
|= head
->extent_op
->flags_to_set
;
917 BUG_ON(num_refs
== 0);
919 num_refs
+= head
->node
.ref_mod
;
920 spin_unlock(&head
->lock
);
921 mutex_unlock(&head
->mutex
);
923 spin_unlock(&delayed_refs
->lock
);
925 WARN_ON(num_refs
== 0);
929 *flags
= extent_flags
;
931 btrfs_free_path(path
);
936 * Back reference rules. Back refs have three main goals:
938 * 1) differentiate between all holders of references to an extent so that
939 * when a reference is dropped we can make sure it was a valid reference
940 * before freeing the extent.
942 * 2) Provide enough information to quickly find the holders of an extent
943 * if we notice a given block is corrupted or bad.
945 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
946 * maintenance. This is actually the same as #2, but with a slightly
947 * different use case.
949 * There are two kinds of back refs. The implicit back refs is optimized
950 * for pointers in non-shared tree blocks. For a given pointer in a block,
951 * back refs of this kind provide information about the block's owner tree
952 * and the pointer's key. These information allow us to find the block by
953 * b-tree searching. The full back refs is for pointers in tree blocks not
954 * referenced by their owner trees. The location of tree block is recorded
955 * in the back refs. Actually the full back refs is generic, and can be
956 * used in all cases the implicit back refs is used. The major shortcoming
957 * of the full back refs is its overhead. Every time a tree block gets
958 * COWed, we have to update back refs entry for all pointers in it.
960 * For a newly allocated tree block, we use implicit back refs for
961 * pointers in it. This means most tree related operations only involve
962 * implicit back refs. For a tree block created in old transaction, the
963 * only way to drop a reference to it is COW it. So we can detect the
964 * event that tree block loses its owner tree's reference and do the
965 * back refs conversion.
967 * When a tree block is COW'd through a tree, there are four cases:
969 * The reference count of the block is one and the tree is the block's
970 * owner tree. Nothing to do in this case.
972 * The reference count of the block is one and the tree is not the
973 * block's owner tree. In this case, full back refs is used for pointers
974 * in the block. Remove these full back refs, add implicit back refs for
975 * every pointers in the new block.
977 * The reference count of the block is greater than one and the tree is
978 * the block's owner tree. In this case, implicit back refs is used for
979 * pointers in the block. Add full back refs for every pointers in the
980 * block, increase lower level extents' reference counts. The original
981 * implicit back refs are entailed to the new block.
983 * The reference count of the block is greater than one and the tree is
984 * not the block's owner tree. Add implicit back refs for every pointer in
985 * the new block, increase lower level extents' reference count.
987 * Back Reference Key composing:
989 * The key objectid corresponds to the first byte in the extent,
990 * The key type is used to differentiate between types of back refs.
991 * There are different meanings of the key offset for different types
994 * File extents can be referenced by:
996 * - multiple snapshots, subvolumes, or different generations in one subvol
997 * - different files inside a single subvolume
998 * - different offsets inside a file (bookend extents in file.c)
1000 * The extent ref structure for the implicit back refs has fields for:
1002 * - Objectid of the subvolume root
1003 * - objectid of the file holding the reference
1004 * - original offset in the file
1005 * - how many bookend extents
1007 * The key offset for the implicit back refs is hash of the first
1010 * The extent ref structure for the full back refs has field for:
1012 * - number of pointers in the tree leaf
1014 * The key offset for the implicit back refs is the first byte of
1017 * When a file extent is allocated, The implicit back refs is used.
1018 * the fields are filled in:
1020 * (root_key.objectid, inode objectid, offset in file, 1)
1022 * When a file extent is removed file truncation, we find the
1023 * corresponding implicit back refs and check the following fields:
1025 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1027 * Btree extents can be referenced by:
1029 * - Different subvolumes
1031 * Both the implicit back refs and the full back refs for tree blocks
1032 * only consist of key. The key offset for the implicit back refs is
1033 * objectid of block's owner tree. The key offset for the full back refs
1034 * is the first byte of parent block.
1036 * When implicit back refs is used, information about the lowest key and
1037 * level of the tree block are required. These information are stored in
1038 * tree block info structure.
1041 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1042 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1043 struct btrfs_root
*root
,
1044 struct btrfs_path
*path
,
1045 u64 owner
, u32 extra_size
)
1047 struct btrfs_extent_item
*item
;
1048 struct btrfs_extent_item_v0
*ei0
;
1049 struct btrfs_extent_ref_v0
*ref0
;
1050 struct btrfs_tree_block_info
*bi
;
1051 struct extent_buffer
*leaf
;
1052 struct btrfs_key key
;
1053 struct btrfs_key found_key
;
1054 u32 new_size
= sizeof(*item
);
1058 leaf
= path
->nodes
[0];
1059 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1061 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1062 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1063 struct btrfs_extent_item_v0
);
1064 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1066 if (owner
== (u64
)-1) {
1068 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1069 ret
= btrfs_next_leaf(root
, path
);
1072 BUG_ON(ret
> 0); /* Corruption */
1073 leaf
= path
->nodes
[0];
1075 btrfs_item_key_to_cpu(leaf
, &found_key
,
1077 BUG_ON(key
.objectid
!= found_key
.objectid
);
1078 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1082 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1083 struct btrfs_extent_ref_v0
);
1084 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1088 btrfs_release_path(path
);
1090 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1091 new_size
+= sizeof(*bi
);
1093 new_size
-= sizeof(*ei0
);
1094 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1095 new_size
+ extra_size
, 1);
1098 BUG_ON(ret
); /* Corruption */
1100 btrfs_extend_item(root
, path
, new_size
);
1102 leaf
= path
->nodes
[0];
1103 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1104 btrfs_set_extent_refs(leaf
, item
, refs
);
1105 /* FIXME: get real generation */
1106 btrfs_set_extent_generation(leaf
, item
, 0);
1107 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1108 btrfs_set_extent_flags(leaf
, item
,
1109 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1110 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1111 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1112 /* FIXME: get first key of the block */
1113 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1114 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1116 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1118 btrfs_mark_buffer_dirty(leaf
);
1123 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1125 u32 high_crc
= ~(u32
)0;
1126 u32 low_crc
= ~(u32
)0;
1129 lenum
= cpu_to_le64(root_objectid
);
1130 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1131 lenum
= cpu_to_le64(owner
);
1132 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1133 lenum
= cpu_to_le64(offset
);
1134 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1136 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1139 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1140 struct btrfs_extent_data_ref
*ref
)
1142 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1143 btrfs_extent_data_ref_objectid(leaf
, ref
),
1144 btrfs_extent_data_ref_offset(leaf
, ref
));
1147 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1148 struct btrfs_extent_data_ref
*ref
,
1149 u64 root_objectid
, u64 owner
, u64 offset
)
1151 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1152 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1153 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1158 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1159 struct btrfs_root
*root
,
1160 struct btrfs_path
*path
,
1161 u64 bytenr
, u64 parent
,
1163 u64 owner
, u64 offset
)
1165 struct btrfs_key key
;
1166 struct btrfs_extent_data_ref
*ref
;
1167 struct extent_buffer
*leaf
;
1173 key
.objectid
= bytenr
;
1175 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1176 key
.offset
= parent
;
1178 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1179 key
.offset
= hash_extent_data_ref(root_objectid
,
1184 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1193 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1194 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1195 btrfs_release_path(path
);
1196 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1207 leaf
= path
->nodes
[0];
1208 nritems
= btrfs_header_nritems(leaf
);
1210 if (path
->slots
[0] >= nritems
) {
1211 ret
= btrfs_next_leaf(root
, path
);
1217 leaf
= path
->nodes
[0];
1218 nritems
= btrfs_header_nritems(leaf
);
1222 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1223 if (key
.objectid
!= bytenr
||
1224 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1227 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1228 struct btrfs_extent_data_ref
);
1230 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1233 btrfs_release_path(path
);
1245 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1246 struct btrfs_root
*root
,
1247 struct btrfs_path
*path
,
1248 u64 bytenr
, u64 parent
,
1249 u64 root_objectid
, u64 owner
,
1250 u64 offset
, int refs_to_add
)
1252 struct btrfs_key key
;
1253 struct extent_buffer
*leaf
;
1258 key
.objectid
= bytenr
;
1260 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1261 key
.offset
= parent
;
1262 size
= sizeof(struct btrfs_shared_data_ref
);
1264 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1265 key
.offset
= hash_extent_data_ref(root_objectid
,
1267 size
= sizeof(struct btrfs_extent_data_ref
);
1270 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1271 if (ret
&& ret
!= -EEXIST
)
1274 leaf
= path
->nodes
[0];
1276 struct btrfs_shared_data_ref
*ref
;
1277 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1278 struct btrfs_shared_data_ref
);
1280 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1282 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1283 num_refs
+= refs_to_add
;
1284 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1287 struct btrfs_extent_data_ref
*ref
;
1288 while (ret
== -EEXIST
) {
1289 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1290 struct btrfs_extent_data_ref
);
1291 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1294 btrfs_release_path(path
);
1296 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1298 if (ret
&& ret
!= -EEXIST
)
1301 leaf
= path
->nodes
[0];
1303 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1304 struct btrfs_extent_data_ref
);
1306 btrfs_set_extent_data_ref_root(leaf
, ref
,
1308 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1309 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1310 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1312 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1313 num_refs
+= refs_to_add
;
1314 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1317 btrfs_mark_buffer_dirty(leaf
);
1320 btrfs_release_path(path
);
1324 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1325 struct btrfs_root
*root
,
1326 struct btrfs_path
*path
,
1327 int refs_to_drop
, int *last_ref
)
1329 struct btrfs_key key
;
1330 struct btrfs_extent_data_ref
*ref1
= NULL
;
1331 struct btrfs_shared_data_ref
*ref2
= NULL
;
1332 struct extent_buffer
*leaf
;
1336 leaf
= path
->nodes
[0];
1337 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1339 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1340 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1341 struct btrfs_extent_data_ref
);
1342 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1343 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1344 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1345 struct btrfs_shared_data_ref
);
1346 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1349 struct btrfs_extent_ref_v0
*ref0
;
1350 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1351 struct btrfs_extent_ref_v0
);
1352 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1358 BUG_ON(num_refs
< refs_to_drop
);
1359 num_refs
-= refs_to_drop
;
1361 if (num_refs
== 0) {
1362 ret
= btrfs_del_item(trans
, root
, path
);
1365 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1366 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1367 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1368 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1369 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1371 struct btrfs_extent_ref_v0
*ref0
;
1372 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1373 struct btrfs_extent_ref_v0
);
1374 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1377 btrfs_mark_buffer_dirty(leaf
);
1382 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1383 struct btrfs_extent_inline_ref
*iref
)
1385 struct btrfs_key key
;
1386 struct extent_buffer
*leaf
;
1387 struct btrfs_extent_data_ref
*ref1
;
1388 struct btrfs_shared_data_ref
*ref2
;
1391 leaf
= path
->nodes
[0];
1392 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1394 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1395 BTRFS_EXTENT_DATA_REF_KEY
) {
1396 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1397 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1399 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1400 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1402 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1403 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1404 struct btrfs_extent_data_ref
);
1405 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1406 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1407 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1408 struct btrfs_shared_data_ref
);
1409 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1410 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1411 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1412 struct btrfs_extent_ref_v0
*ref0
;
1413 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1414 struct btrfs_extent_ref_v0
);
1415 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1423 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1424 struct btrfs_root
*root
,
1425 struct btrfs_path
*path
,
1426 u64 bytenr
, u64 parent
,
1429 struct btrfs_key key
;
1432 key
.objectid
= bytenr
;
1434 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1435 key
.offset
= parent
;
1437 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1438 key
.offset
= root_objectid
;
1441 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1444 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1445 if (ret
== -ENOENT
&& parent
) {
1446 btrfs_release_path(path
);
1447 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1448 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1456 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1457 struct btrfs_root
*root
,
1458 struct btrfs_path
*path
,
1459 u64 bytenr
, u64 parent
,
1462 struct btrfs_key key
;
1465 key
.objectid
= bytenr
;
1467 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1468 key
.offset
= parent
;
1470 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1471 key
.offset
= root_objectid
;
1474 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1475 btrfs_release_path(path
);
1479 static inline int extent_ref_type(u64 parent
, u64 owner
)
1482 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1484 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1486 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1489 type
= BTRFS_SHARED_DATA_REF_KEY
;
1491 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1496 static int find_next_key(struct btrfs_path
*path
, int level
,
1497 struct btrfs_key
*key
)
1500 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1501 if (!path
->nodes
[level
])
1503 if (path
->slots
[level
] + 1 >=
1504 btrfs_header_nritems(path
->nodes
[level
]))
1507 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1508 path
->slots
[level
] + 1);
1510 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1511 path
->slots
[level
] + 1);
1518 * look for inline back ref. if back ref is found, *ref_ret is set
1519 * to the address of inline back ref, and 0 is returned.
1521 * if back ref isn't found, *ref_ret is set to the address where it
1522 * should be inserted, and -ENOENT is returned.
1524 * if insert is true and there are too many inline back refs, the path
1525 * points to the extent item, and -EAGAIN is returned.
1527 * NOTE: inline back refs are ordered in the same way that back ref
1528 * items in the tree are ordered.
1530 static noinline_for_stack
1531 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1532 struct btrfs_root
*root
,
1533 struct btrfs_path
*path
,
1534 struct btrfs_extent_inline_ref
**ref_ret
,
1535 u64 bytenr
, u64 num_bytes
,
1536 u64 parent
, u64 root_objectid
,
1537 u64 owner
, u64 offset
, int insert
)
1539 struct btrfs_key key
;
1540 struct extent_buffer
*leaf
;
1541 struct btrfs_extent_item
*ei
;
1542 struct btrfs_extent_inline_ref
*iref
;
1552 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1555 key
.objectid
= bytenr
;
1556 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1557 key
.offset
= num_bytes
;
1559 want
= extent_ref_type(parent
, owner
);
1561 extra_size
= btrfs_extent_inline_ref_size(want
);
1562 path
->keep_locks
= 1;
1567 * Owner is our parent level, so we can just add one to get the level
1568 * for the block we are interested in.
1570 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1571 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1576 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1583 * We may be a newly converted file system which still has the old fat
1584 * extent entries for metadata, so try and see if we have one of those.
1586 if (ret
> 0 && skinny_metadata
) {
1587 skinny_metadata
= false;
1588 if (path
->slots
[0]) {
1590 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1592 if (key
.objectid
== bytenr
&&
1593 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1594 key
.offset
== num_bytes
)
1598 key
.objectid
= bytenr
;
1599 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1600 key
.offset
= num_bytes
;
1601 btrfs_release_path(path
);
1606 if (ret
&& !insert
) {
1609 } else if (WARN_ON(ret
)) {
1614 leaf
= path
->nodes
[0];
1615 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1616 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1617 if (item_size
< sizeof(*ei
)) {
1622 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1628 leaf
= path
->nodes
[0];
1629 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1632 BUG_ON(item_size
< sizeof(*ei
));
1634 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1635 flags
= btrfs_extent_flags(leaf
, ei
);
1637 ptr
= (unsigned long)(ei
+ 1);
1638 end
= (unsigned long)ei
+ item_size
;
1640 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1641 ptr
+= sizeof(struct btrfs_tree_block_info
);
1651 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1652 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1656 ptr
+= btrfs_extent_inline_ref_size(type
);
1660 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1661 struct btrfs_extent_data_ref
*dref
;
1662 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1663 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1668 if (hash_extent_data_ref_item(leaf
, dref
) <
1669 hash_extent_data_ref(root_objectid
, owner
, offset
))
1673 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1675 if (parent
== ref_offset
) {
1679 if (ref_offset
< parent
)
1682 if (root_objectid
== ref_offset
) {
1686 if (ref_offset
< root_objectid
)
1690 ptr
+= btrfs_extent_inline_ref_size(type
);
1692 if (err
== -ENOENT
&& insert
) {
1693 if (item_size
+ extra_size
>=
1694 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1699 * To add new inline back ref, we have to make sure
1700 * there is no corresponding back ref item.
1701 * For simplicity, we just do not add new inline back
1702 * ref if there is any kind of item for this block
1704 if (find_next_key(path
, 0, &key
) == 0 &&
1705 key
.objectid
== bytenr
&&
1706 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1711 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1714 path
->keep_locks
= 0;
1715 btrfs_unlock_up_safe(path
, 1);
1721 * helper to add new inline back ref
1723 static noinline_for_stack
1724 void setup_inline_extent_backref(struct btrfs_root
*root
,
1725 struct btrfs_path
*path
,
1726 struct btrfs_extent_inline_ref
*iref
,
1727 u64 parent
, u64 root_objectid
,
1728 u64 owner
, u64 offset
, int refs_to_add
,
1729 struct btrfs_delayed_extent_op
*extent_op
)
1731 struct extent_buffer
*leaf
;
1732 struct btrfs_extent_item
*ei
;
1735 unsigned long item_offset
;
1740 leaf
= path
->nodes
[0];
1741 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1742 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1744 type
= extent_ref_type(parent
, owner
);
1745 size
= btrfs_extent_inline_ref_size(type
);
1747 btrfs_extend_item(root
, path
, size
);
1749 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1750 refs
= btrfs_extent_refs(leaf
, ei
);
1751 refs
+= refs_to_add
;
1752 btrfs_set_extent_refs(leaf
, ei
, refs
);
1754 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1756 ptr
= (unsigned long)ei
+ item_offset
;
1757 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1758 if (ptr
< end
- size
)
1759 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1762 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1763 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1764 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1765 struct btrfs_extent_data_ref
*dref
;
1766 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1767 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1768 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1769 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1770 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1771 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1772 struct btrfs_shared_data_ref
*sref
;
1773 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1774 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1775 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1776 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1777 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1779 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1781 btrfs_mark_buffer_dirty(leaf
);
1784 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1785 struct btrfs_root
*root
,
1786 struct btrfs_path
*path
,
1787 struct btrfs_extent_inline_ref
**ref_ret
,
1788 u64 bytenr
, u64 num_bytes
, u64 parent
,
1789 u64 root_objectid
, u64 owner
, u64 offset
)
1793 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1794 bytenr
, num_bytes
, parent
,
1795 root_objectid
, owner
, offset
, 0);
1799 btrfs_release_path(path
);
1802 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1803 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1806 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1807 root_objectid
, owner
, offset
);
1813 * helper to update/remove inline back ref
1815 static noinline_for_stack
1816 void update_inline_extent_backref(struct btrfs_root
*root
,
1817 struct btrfs_path
*path
,
1818 struct btrfs_extent_inline_ref
*iref
,
1820 struct btrfs_delayed_extent_op
*extent_op
,
1823 struct extent_buffer
*leaf
;
1824 struct btrfs_extent_item
*ei
;
1825 struct btrfs_extent_data_ref
*dref
= NULL
;
1826 struct btrfs_shared_data_ref
*sref
= NULL
;
1834 leaf
= path
->nodes
[0];
1835 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1836 refs
= btrfs_extent_refs(leaf
, ei
);
1837 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1838 refs
+= refs_to_mod
;
1839 btrfs_set_extent_refs(leaf
, ei
, refs
);
1841 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1843 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1845 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1846 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1847 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1848 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1849 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1850 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1853 BUG_ON(refs_to_mod
!= -1);
1856 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1857 refs
+= refs_to_mod
;
1860 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1861 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1863 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1866 size
= btrfs_extent_inline_ref_size(type
);
1867 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1868 ptr
= (unsigned long)iref
;
1869 end
= (unsigned long)ei
+ item_size
;
1870 if (ptr
+ size
< end
)
1871 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1874 btrfs_truncate_item(root
, path
, item_size
, 1);
1876 btrfs_mark_buffer_dirty(leaf
);
1879 static noinline_for_stack
1880 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1881 struct btrfs_root
*root
,
1882 struct btrfs_path
*path
,
1883 u64 bytenr
, u64 num_bytes
, u64 parent
,
1884 u64 root_objectid
, u64 owner
,
1885 u64 offset
, int refs_to_add
,
1886 struct btrfs_delayed_extent_op
*extent_op
)
1888 struct btrfs_extent_inline_ref
*iref
;
1891 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1892 bytenr
, num_bytes
, parent
,
1893 root_objectid
, owner
, offset
, 1);
1895 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1896 update_inline_extent_backref(root
, path
, iref
,
1897 refs_to_add
, extent_op
, NULL
);
1898 } else if (ret
== -ENOENT
) {
1899 setup_inline_extent_backref(root
, path
, iref
, parent
,
1900 root_objectid
, owner
, offset
,
1901 refs_to_add
, extent_op
);
1907 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1908 struct btrfs_root
*root
,
1909 struct btrfs_path
*path
,
1910 u64 bytenr
, u64 parent
, u64 root_objectid
,
1911 u64 owner
, u64 offset
, int refs_to_add
)
1914 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1915 BUG_ON(refs_to_add
!= 1);
1916 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1917 parent
, root_objectid
);
1919 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1920 parent
, root_objectid
,
1921 owner
, offset
, refs_to_add
);
1926 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1927 struct btrfs_root
*root
,
1928 struct btrfs_path
*path
,
1929 struct btrfs_extent_inline_ref
*iref
,
1930 int refs_to_drop
, int is_data
, int *last_ref
)
1934 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1936 update_inline_extent_backref(root
, path
, iref
,
1937 -refs_to_drop
, NULL
, last_ref
);
1938 } else if (is_data
) {
1939 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1943 ret
= btrfs_del_item(trans
, root
, path
);
1948 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1949 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1950 u64
*discarded_bytes
)
1953 u64 bytes_left
, end
;
1954 u64 aligned_start
= ALIGN(start
, 1 << 9);
1956 if (WARN_ON(start
!= aligned_start
)) {
1957 len
-= aligned_start
- start
;
1958 len
= round_down(len
, 1 << 9);
1959 start
= aligned_start
;
1962 *discarded_bytes
= 0;
1970 /* Skip any superblocks on this device. */
1971 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1972 u64 sb_start
= btrfs_sb_offset(j
);
1973 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1974 u64 size
= sb_start
- start
;
1976 if (!in_range(sb_start
, start
, bytes_left
) &&
1977 !in_range(sb_end
, start
, bytes_left
) &&
1978 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1982 * Superblock spans beginning of range. Adjust start and
1985 if (sb_start
<= start
) {
1986 start
+= sb_end
- start
;
1991 bytes_left
= end
- start
;
1996 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
1999 *discarded_bytes
+= size
;
2000 else if (ret
!= -EOPNOTSUPP
)
2009 bytes_left
= end
- start
;
2013 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2016 *discarded_bytes
+= bytes_left
;
2021 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
2022 u64 num_bytes
, u64
*actual_bytes
)
2025 u64 discarded_bytes
= 0;
2026 struct btrfs_bio
*bbio
= NULL
;
2029 /* Tell the block device(s) that the sectors can be discarded */
2030 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
2031 bytenr
, &num_bytes
, &bbio
, 0);
2032 /* Error condition is -ENOMEM */
2034 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2038 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2040 if (!stripe
->dev
->can_discard
)
2043 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2048 discarded_bytes
+= bytes
;
2049 else if (ret
!= -EOPNOTSUPP
)
2050 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2053 * Just in case we get back EOPNOTSUPP for some reason,
2054 * just ignore the return value so we don't screw up
2055 * people calling discard_extent.
2059 btrfs_put_bbio(bbio
);
2063 *actual_bytes
= discarded_bytes
;
2066 if (ret
== -EOPNOTSUPP
)
2071 /* Can return -ENOMEM */
2072 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2073 struct btrfs_root
*root
,
2074 u64 bytenr
, u64 num_bytes
, u64 parent
,
2075 u64 root_objectid
, u64 owner
, u64 offset
)
2078 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2080 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2081 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2083 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2084 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2086 parent
, root_objectid
, (int)owner
,
2087 BTRFS_ADD_DELAYED_REF
, NULL
);
2089 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2090 num_bytes
, parent
, root_objectid
,
2092 BTRFS_ADD_DELAYED_REF
, NULL
);
2097 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2098 struct btrfs_root
*root
,
2099 struct btrfs_delayed_ref_node
*node
,
2100 u64 parent
, u64 root_objectid
,
2101 u64 owner
, u64 offset
, int refs_to_add
,
2102 struct btrfs_delayed_extent_op
*extent_op
)
2104 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2105 struct btrfs_path
*path
;
2106 struct extent_buffer
*leaf
;
2107 struct btrfs_extent_item
*item
;
2108 struct btrfs_key key
;
2109 u64 bytenr
= node
->bytenr
;
2110 u64 num_bytes
= node
->num_bytes
;
2114 path
= btrfs_alloc_path();
2119 path
->leave_spinning
= 1;
2120 /* this will setup the path even if it fails to insert the back ref */
2121 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
2122 bytenr
, num_bytes
, parent
,
2123 root_objectid
, owner
, offset
,
2124 refs_to_add
, extent_op
);
2125 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2129 * Ok we had -EAGAIN which means we didn't have space to insert and
2130 * inline extent ref, so just update the reference count and add a
2133 leaf
= path
->nodes
[0];
2134 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2135 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2136 refs
= btrfs_extent_refs(leaf
, item
);
2137 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2139 __run_delayed_extent_op(extent_op
, leaf
, item
);
2141 btrfs_mark_buffer_dirty(leaf
);
2142 btrfs_release_path(path
);
2145 path
->leave_spinning
= 1;
2146 /* now insert the actual backref */
2147 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2148 path
, bytenr
, parent
, root_objectid
,
2149 owner
, offset
, refs_to_add
);
2151 btrfs_abort_transaction(trans
, root
, ret
);
2153 btrfs_free_path(path
);
2157 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2158 struct btrfs_root
*root
,
2159 struct btrfs_delayed_ref_node
*node
,
2160 struct btrfs_delayed_extent_op
*extent_op
,
2161 int insert_reserved
)
2164 struct btrfs_delayed_data_ref
*ref
;
2165 struct btrfs_key ins
;
2170 ins
.objectid
= node
->bytenr
;
2171 ins
.offset
= node
->num_bytes
;
2172 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2174 ref
= btrfs_delayed_node_to_data_ref(node
);
2175 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2177 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2178 parent
= ref
->parent
;
2179 ref_root
= ref
->root
;
2181 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2183 flags
|= extent_op
->flags_to_set
;
2184 ret
= alloc_reserved_file_extent(trans
, root
,
2185 parent
, ref_root
, flags
,
2186 ref
->objectid
, ref
->offset
,
2187 &ins
, node
->ref_mod
);
2188 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2189 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2190 ref_root
, ref
->objectid
,
2191 ref
->offset
, node
->ref_mod
,
2193 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2194 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2195 ref_root
, ref
->objectid
,
2196 ref
->offset
, node
->ref_mod
,
2204 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2205 struct extent_buffer
*leaf
,
2206 struct btrfs_extent_item
*ei
)
2208 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2209 if (extent_op
->update_flags
) {
2210 flags
|= extent_op
->flags_to_set
;
2211 btrfs_set_extent_flags(leaf
, ei
, flags
);
2214 if (extent_op
->update_key
) {
2215 struct btrfs_tree_block_info
*bi
;
2216 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2217 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2218 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2222 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2223 struct btrfs_root
*root
,
2224 struct btrfs_delayed_ref_node
*node
,
2225 struct btrfs_delayed_extent_op
*extent_op
)
2227 struct btrfs_key key
;
2228 struct btrfs_path
*path
;
2229 struct btrfs_extent_item
*ei
;
2230 struct extent_buffer
*leaf
;
2234 int metadata
= !extent_op
->is_data
;
2239 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2242 path
= btrfs_alloc_path();
2246 key
.objectid
= node
->bytenr
;
2249 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2250 key
.offset
= extent_op
->level
;
2252 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2253 key
.offset
= node
->num_bytes
;
2258 path
->leave_spinning
= 1;
2259 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2267 if (path
->slots
[0] > 0) {
2269 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2271 if (key
.objectid
== node
->bytenr
&&
2272 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2273 key
.offset
== node
->num_bytes
)
2277 btrfs_release_path(path
);
2280 key
.objectid
= node
->bytenr
;
2281 key
.offset
= node
->num_bytes
;
2282 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2291 leaf
= path
->nodes
[0];
2292 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2294 if (item_size
< sizeof(*ei
)) {
2295 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2301 leaf
= path
->nodes
[0];
2302 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2305 BUG_ON(item_size
< sizeof(*ei
));
2306 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2307 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2309 btrfs_mark_buffer_dirty(leaf
);
2311 btrfs_free_path(path
);
2315 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2316 struct btrfs_root
*root
,
2317 struct btrfs_delayed_ref_node
*node
,
2318 struct btrfs_delayed_extent_op
*extent_op
,
2319 int insert_reserved
)
2322 struct btrfs_delayed_tree_ref
*ref
;
2323 struct btrfs_key ins
;
2326 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2329 ref
= btrfs_delayed_node_to_tree_ref(node
);
2330 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2332 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2333 parent
= ref
->parent
;
2334 ref_root
= ref
->root
;
2336 ins
.objectid
= node
->bytenr
;
2337 if (skinny_metadata
) {
2338 ins
.offset
= ref
->level
;
2339 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2341 ins
.offset
= node
->num_bytes
;
2342 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2345 BUG_ON(node
->ref_mod
!= 1);
2346 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2347 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2348 ret
= alloc_reserved_tree_block(trans
, root
,
2350 extent_op
->flags_to_set
,
2353 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2354 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2358 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2359 ret
= __btrfs_free_extent(trans
, root
, node
,
2361 ref
->level
, 0, 1, extent_op
);
2368 /* helper function to actually process a single delayed ref entry */
2369 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2370 struct btrfs_root
*root
,
2371 struct btrfs_delayed_ref_node
*node
,
2372 struct btrfs_delayed_extent_op
*extent_op
,
2373 int insert_reserved
)
2377 if (trans
->aborted
) {
2378 if (insert_reserved
)
2379 btrfs_pin_extent(root
, node
->bytenr
,
2380 node
->num_bytes
, 1);
2384 if (btrfs_delayed_ref_is_head(node
)) {
2385 struct btrfs_delayed_ref_head
*head
;
2387 * we've hit the end of the chain and we were supposed
2388 * to insert this extent into the tree. But, it got
2389 * deleted before we ever needed to insert it, so all
2390 * we have to do is clean up the accounting
2393 head
= btrfs_delayed_node_to_head(node
);
2394 trace_run_delayed_ref_head(node
, head
, node
->action
);
2396 if (insert_reserved
) {
2397 btrfs_pin_extent(root
, node
->bytenr
,
2398 node
->num_bytes
, 1);
2399 if (head
->is_data
) {
2400 ret
= btrfs_del_csums(trans
, root
,
2406 /* Also free its reserved qgroup space */
2407 btrfs_qgroup_free_delayed_ref(root
->fs_info
,
2408 head
->qgroup_ref_root
,
2409 head
->qgroup_reserved
);
2413 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2414 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2415 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2417 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2418 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2419 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2426 static inline struct btrfs_delayed_ref_node
*
2427 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2429 struct btrfs_delayed_ref_node
*ref
;
2431 if (list_empty(&head
->ref_list
))
2435 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2436 * This is to prevent a ref count from going down to zero, which deletes
2437 * the extent item from the extent tree, when there still are references
2438 * to add, which would fail because they would not find the extent item.
2440 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2441 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2445 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2450 * Returns 0 on success or if called with an already aborted transaction.
2451 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2453 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2454 struct btrfs_root
*root
,
2457 struct btrfs_delayed_ref_root
*delayed_refs
;
2458 struct btrfs_delayed_ref_node
*ref
;
2459 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2460 struct btrfs_delayed_extent_op
*extent_op
;
2461 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2462 ktime_t start
= ktime_get();
2464 unsigned long count
= 0;
2465 unsigned long actual_count
= 0;
2466 int must_insert_reserved
= 0;
2468 delayed_refs
= &trans
->transaction
->delayed_refs
;
2474 spin_lock(&delayed_refs
->lock
);
2475 locked_ref
= btrfs_select_ref_head(trans
);
2477 spin_unlock(&delayed_refs
->lock
);
2481 /* grab the lock that says we are going to process
2482 * all the refs for this head */
2483 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2484 spin_unlock(&delayed_refs
->lock
);
2486 * we may have dropped the spin lock to get the head
2487 * mutex lock, and that might have given someone else
2488 * time to free the head. If that's true, it has been
2489 * removed from our list and we can move on.
2491 if (ret
== -EAGAIN
) {
2499 * We need to try and merge add/drops of the same ref since we
2500 * can run into issues with relocate dropping the implicit ref
2501 * and then it being added back again before the drop can
2502 * finish. If we merged anything we need to re-loop so we can
2504 * Or we can get node references of the same type that weren't
2505 * merged when created due to bumps in the tree mod seq, and
2506 * we need to merge them to prevent adding an inline extent
2507 * backref before dropping it (triggering a BUG_ON at
2508 * insert_inline_extent_backref()).
2510 spin_lock(&locked_ref
->lock
);
2511 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2515 * locked_ref is the head node, so we have to go one
2516 * node back for any delayed ref updates
2518 ref
= select_delayed_ref(locked_ref
);
2520 if (ref
&& ref
->seq
&&
2521 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2522 spin_unlock(&locked_ref
->lock
);
2523 spin_lock(&delayed_refs
->lock
);
2524 locked_ref
->processing
= 0;
2525 delayed_refs
->num_heads_ready
++;
2526 spin_unlock(&delayed_refs
->lock
);
2527 btrfs_delayed_ref_unlock(locked_ref
);
2535 * record the must insert reserved flag before we
2536 * drop the spin lock.
2538 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2539 locked_ref
->must_insert_reserved
= 0;
2541 extent_op
= locked_ref
->extent_op
;
2542 locked_ref
->extent_op
= NULL
;
2547 /* All delayed refs have been processed, Go ahead
2548 * and send the head node to run_one_delayed_ref,
2549 * so that any accounting fixes can happen
2551 ref
= &locked_ref
->node
;
2553 if (extent_op
&& must_insert_reserved
) {
2554 btrfs_free_delayed_extent_op(extent_op
);
2559 spin_unlock(&locked_ref
->lock
);
2560 ret
= run_delayed_extent_op(trans
, root
,
2562 btrfs_free_delayed_extent_op(extent_op
);
2566 * Need to reset must_insert_reserved if
2567 * there was an error so the abort stuff
2568 * can cleanup the reserved space
2571 if (must_insert_reserved
)
2572 locked_ref
->must_insert_reserved
= 1;
2573 spin_lock(&delayed_refs
->lock
);
2574 locked_ref
->processing
= 0;
2575 delayed_refs
->num_heads_ready
++;
2576 spin_unlock(&delayed_refs
->lock
);
2577 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2578 btrfs_delayed_ref_unlock(locked_ref
);
2585 * Need to drop our head ref lock and re-aqcuire the
2586 * delayed ref lock and then re-check to make sure
2589 spin_unlock(&locked_ref
->lock
);
2590 spin_lock(&delayed_refs
->lock
);
2591 spin_lock(&locked_ref
->lock
);
2592 if (!list_empty(&locked_ref
->ref_list
) ||
2593 locked_ref
->extent_op
) {
2594 spin_unlock(&locked_ref
->lock
);
2595 spin_unlock(&delayed_refs
->lock
);
2599 delayed_refs
->num_heads
--;
2600 rb_erase(&locked_ref
->href_node
,
2601 &delayed_refs
->href_root
);
2602 spin_unlock(&delayed_refs
->lock
);
2606 list_del(&ref
->list
);
2608 atomic_dec(&delayed_refs
->num_entries
);
2610 if (!btrfs_delayed_ref_is_head(ref
)) {
2612 * when we play the delayed ref, also correct the
2615 switch (ref
->action
) {
2616 case BTRFS_ADD_DELAYED_REF
:
2617 case BTRFS_ADD_DELAYED_EXTENT
:
2618 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2620 case BTRFS_DROP_DELAYED_REF
:
2621 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2627 spin_unlock(&locked_ref
->lock
);
2629 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2630 must_insert_reserved
);
2632 btrfs_free_delayed_extent_op(extent_op
);
2634 locked_ref
->processing
= 0;
2635 btrfs_delayed_ref_unlock(locked_ref
);
2636 btrfs_put_delayed_ref(ref
);
2637 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2642 * If this node is a head, that means all the refs in this head
2643 * have been dealt with, and we will pick the next head to deal
2644 * with, so we must unlock the head and drop it from the cluster
2645 * list before we release it.
2647 if (btrfs_delayed_ref_is_head(ref
)) {
2648 if (locked_ref
->is_data
&&
2649 locked_ref
->total_ref_mod
< 0) {
2650 spin_lock(&delayed_refs
->lock
);
2651 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2652 spin_unlock(&delayed_refs
->lock
);
2654 btrfs_delayed_ref_unlock(locked_ref
);
2657 btrfs_put_delayed_ref(ref
);
2663 * We don't want to include ref heads since we can have empty ref heads
2664 * and those will drastically skew our runtime down since we just do
2665 * accounting, no actual extent tree updates.
2667 if (actual_count
> 0) {
2668 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2672 * We weigh the current average higher than our current runtime
2673 * to avoid large swings in the average.
2675 spin_lock(&delayed_refs
->lock
);
2676 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2677 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2678 spin_unlock(&delayed_refs
->lock
);
2683 #ifdef SCRAMBLE_DELAYED_REFS
2685 * Normally delayed refs get processed in ascending bytenr order. This
2686 * correlates in most cases to the order added. To expose dependencies on this
2687 * order, we start to process the tree in the middle instead of the beginning
2689 static u64
find_middle(struct rb_root
*root
)
2691 struct rb_node
*n
= root
->rb_node
;
2692 struct btrfs_delayed_ref_node
*entry
;
2695 u64 first
= 0, last
= 0;
2699 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2700 first
= entry
->bytenr
;
2704 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2705 last
= entry
->bytenr
;
2710 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2711 WARN_ON(!entry
->in_tree
);
2713 middle
= entry
->bytenr
;
2726 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2730 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2731 sizeof(struct btrfs_extent_inline_ref
));
2732 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2733 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2736 * We don't ever fill up leaves all the way so multiply by 2 just to be
2737 * closer to what we're really going to want to ouse.
2739 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2743 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2744 * would require to store the csums for that many bytes.
2746 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2749 u64 num_csums_per_leaf
;
2752 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2753 num_csums_per_leaf
= div64_u64(csum_size
,
2754 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2755 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2756 num_csums
+= num_csums_per_leaf
- 1;
2757 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2761 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2762 struct btrfs_root
*root
)
2764 struct btrfs_block_rsv
*global_rsv
;
2765 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2766 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2767 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2768 u64 num_bytes
, num_dirty_bgs_bytes
;
2771 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2772 num_heads
= heads_to_leaves(root
, num_heads
);
2774 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2776 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2777 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2779 global_rsv
= &root
->fs_info
->global_block_rsv
;
2782 * If we can't allocate any more chunks lets make sure we have _lots_ of
2783 * wiggle room since running delayed refs can create more delayed refs.
2785 if (global_rsv
->space_info
->full
) {
2786 num_dirty_bgs_bytes
<<= 1;
2790 spin_lock(&global_rsv
->lock
);
2791 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2793 spin_unlock(&global_rsv
->lock
);
2797 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2798 struct btrfs_root
*root
)
2800 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2802 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2807 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2808 val
= num_entries
* avg_runtime
;
2809 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2811 if (val
>= NSEC_PER_SEC
/ 2)
2814 return btrfs_check_space_for_delayed_refs(trans
, root
);
2817 struct async_delayed_refs
{
2818 struct btrfs_root
*root
;
2822 struct completion wait
;
2823 struct btrfs_work work
;
2826 static void delayed_ref_async_start(struct btrfs_work
*work
)
2828 struct async_delayed_refs
*async
;
2829 struct btrfs_trans_handle
*trans
;
2832 async
= container_of(work
, struct async_delayed_refs
, work
);
2834 trans
= btrfs_join_transaction(async
->root
);
2835 if (IS_ERR(trans
)) {
2836 async
->error
= PTR_ERR(trans
);
2841 * trans->sync means that when we call end_transaciton, we won't
2842 * wait on delayed refs
2845 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2849 ret
= btrfs_end_transaction(trans
, async
->root
);
2850 if (ret
&& !async
->error
)
2854 complete(&async
->wait
);
2859 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2860 unsigned long count
, int wait
)
2862 struct async_delayed_refs
*async
;
2865 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2869 async
->root
= root
->fs_info
->tree_root
;
2870 async
->count
= count
;
2876 init_completion(&async
->wait
);
2878 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2879 delayed_ref_async_start
, NULL
, NULL
);
2881 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2884 wait_for_completion(&async
->wait
);
2893 * this starts processing the delayed reference count updates and
2894 * extent insertions we have queued up so far. count can be
2895 * 0, which means to process everything in the tree at the start
2896 * of the run (but not newly added entries), or it can be some target
2897 * number you'd like to process.
2899 * Returns 0 on success or if called with an aborted transaction
2900 * Returns <0 on error and aborts the transaction
2902 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2903 struct btrfs_root
*root
, unsigned long count
)
2905 struct rb_node
*node
;
2906 struct btrfs_delayed_ref_root
*delayed_refs
;
2907 struct btrfs_delayed_ref_head
*head
;
2909 int run_all
= count
== (unsigned long)-1;
2910 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2912 /* We'll clean this up in btrfs_cleanup_transaction */
2916 if (root
== root
->fs_info
->extent_root
)
2917 root
= root
->fs_info
->tree_root
;
2919 delayed_refs
= &trans
->transaction
->delayed_refs
;
2921 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2924 #ifdef SCRAMBLE_DELAYED_REFS
2925 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2927 trans
->can_flush_pending_bgs
= false;
2928 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2930 btrfs_abort_transaction(trans
, root
, ret
);
2935 if (!list_empty(&trans
->new_bgs
))
2936 btrfs_create_pending_block_groups(trans
, root
);
2938 spin_lock(&delayed_refs
->lock
);
2939 node
= rb_first(&delayed_refs
->href_root
);
2941 spin_unlock(&delayed_refs
->lock
);
2944 count
= (unsigned long)-1;
2947 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2949 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2950 struct btrfs_delayed_ref_node
*ref
;
2953 atomic_inc(&ref
->refs
);
2955 spin_unlock(&delayed_refs
->lock
);
2957 * Mutex was contended, block until it's
2958 * released and try again
2960 mutex_lock(&head
->mutex
);
2961 mutex_unlock(&head
->mutex
);
2963 btrfs_put_delayed_ref(ref
);
2969 node
= rb_next(node
);
2971 spin_unlock(&delayed_refs
->lock
);
2976 assert_qgroups_uptodate(trans
);
2977 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
2981 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2982 struct btrfs_root
*root
,
2983 u64 bytenr
, u64 num_bytes
, u64 flags
,
2984 int level
, int is_data
)
2986 struct btrfs_delayed_extent_op
*extent_op
;
2989 extent_op
= btrfs_alloc_delayed_extent_op();
2993 extent_op
->flags_to_set
= flags
;
2994 extent_op
->update_flags
= 1;
2995 extent_op
->update_key
= 0;
2996 extent_op
->is_data
= is_data
? 1 : 0;
2997 extent_op
->level
= level
;
2999 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
3000 num_bytes
, extent_op
);
3002 btrfs_free_delayed_extent_op(extent_op
);
3006 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
3007 struct btrfs_root
*root
,
3008 struct btrfs_path
*path
,
3009 u64 objectid
, u64 offset
, u64 bytenr
)
3011 struct btrfs_delayed_ref_head
*head
;
3012 struct btrfs_delayed_ref_node
*ref
;
3013 struct btrfs_delayed_data_ref
*data_ref
;
3014 struct btrfs_delayed_ref_root
*delayed_refs
;
3017 delayed_refs
= &trans
->transaction
->delayed_refs
;
3018 spin_lock(&delayed_refs
->lock
);
3019 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3021 spin_unlock(&delayed_refs
->lock
);
3025 if (!mutex_trylock(&head
->mutex
)) {
3026 atomic_inc(&head
->node
.refs
);
3027 spin_unlock(&delayed_refs
->lock
);
3029 btrfs_release_path(path
);
3032 * Mutex was contended, block until it's released and let
3035 mutex_lock(&head
->mutex
);
3036 mutex_unlock(&head
->mutex
);
3037 btrfs_put_delayed_ref(&head
->node
);
3040 spin_unlock(&delayed_refs
->lock
);
3042 spin_lock(&head
->lock
);
3043 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3044 /* If it's a shared ref we know a cross reference exists */
3045 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3050 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3053 * If our ref doesn't match the one we're currently looking at
3054 * then we have a cross reference.
3056 if (data_ref
->root
!= root
->root_key
.objectid
||
3057 data_ref
->objectid
!= objectid
||
3058 data_ref
->offset
!= offset
) {
3063 spin_unlock(&head
->lock
);
3064 mutex_unlock(&head
->mutex
);
3068 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
3069 struct btrfs_root
*root
,
3070 struct btrfs_path
*path
,
3071 u64 objectid
, u64 offset
, u64 bytenr
)
3073 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3074 struct extent_buffer
*leaf
;
3075 struct btrfs_extent_data_ref
*ref
;
3076 struct btrfs_extent_inline_ref
*iref
;
3077 struct btrfs_extent_item
*ei
;
3078 struct btrfs_key key
;
3082 key
.objectid
= bytenr
;
3083 key
.offset
= (u64
)-1;
3084 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3086 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3089 BUG_ON(ret
== 0); /* Corruption */
3092 if (path
->slots
[0] == 0)
3096 leaf
= path
->nodes
[0];
3097 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3099 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3103 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3104 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3105 if (item_size
< sizeof(*ei
)) {
3106 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3110 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3112 if (item_size
!= sizeof(*ei
) +
3113 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3116 if (btrfs_extent_generation(leaf
, ei
) <=
3117 btrfs_root_last_snapshot(&root
->root_item
))
3120 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3121 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3122 BTRFS_EXTENT_DATA_REF_KEY
)
3125 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3126 if (btrfs_extent_refs(leaf
, ei
) !=
3127 btrfs_extent_data_ref_count(leaf
, ref
) ||
3128 btrfs_extent_data_ref_root(leaf
, ref
) !=
3129 root
->root_key
.objectid
||
3130 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3131 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3139 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3140 struct btrfs_root
*root
,
3141 u64 objectid
, u64 offset
, u64 bytenr
)
3143 struct btrfs_path
*path
;
3147 path
= btrfs_alloc_path();
3152 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3154 if (ret
&& ret
!= -ENOENT
)
3157 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3159 } while (ret2
== -EAGAIN
);
3161 if (ret2
&& ret2
!= -ENOENT
) {
3166 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3169 btrfs_free_path(path
);
3170 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3175 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3176 struct btrfs_root
*root
,
3177 struct extent_buffer
*buf
,
3178 int full_backref
, int inc
)
3185 struct btrfs_key key
;
3186 struct btrfs_file_extent_item
*fi
;
3190 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3191 u64
, u64
, u64
, u64
, u64
, u64
);
3194 if (btrfs_test_is_dummy_root(root
))
3197 ref_root
= btrfs_header_owner(buf
);
3198 nritems
= btrfs_header_nritems(buf
);
3199 level
= btrfs_header_level(buf
);
3201 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3205 process_func
= btrfs_inc_extent_ref
;
3207 process_func
= btrfs_free_extent
;
3210 parent
= buf
->start
;
3214 for (i
= 0; i
< nritems
; i
++) {
3216 btrfs_item_key_to_cpu(buf
, &key
, i
);
3217 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3219 fi
= btrfs_item_ptr(buf
, i
,
3220 struct btrfs_file_extent_item
);
3221 if (btrfs_file_extent_type(buf
, fi
) ==
3222 BTRFS_FILE_EXTENT_INLINE
)
3224 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3228 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3229 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3230 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3231 parent
, ref_root
, key
.objectid
,
3236 bytenr
= btrfs_node_blockptr(buf
, i
);
3237 num_bytes
= root
->nodesize
;
3238 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3239 parent
, ref_root
, level
- 1, 0);
3249 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3250 struct extent_buffer
*buf
, int full_backref
)
3252 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3255 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3256 struct extent_buffer
*buf
, int full_backref
)
3258 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3261 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3262 struct btrfs_root
*root
,
3263 struct btrfs_path
*path
,
3264 struct btrfs_block_group_cache
*cache
)
3267 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3269 struct extent_buffer
*leaf
;
3271 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3278 leaf
= path
->nodes
[0];
3279 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3280 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3281 btrfs_mark_buffer_dirty(leaf
);
3283 btrfs_release_path(path
);
3288 static struct btrfs_block_group_cache
*
3289 next_block_group(struct btrfs_root
*root
,
3290 struct btrfs_block_group_cache
*cache
)
3292 struct rb_node
*node
;
3294 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3296 /* If our block group was removed, we need a full search. */
3297 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3298 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3300 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3301 btrfs_put_block_group(cache
);
3302 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3306 node
= rb_next(&cache
->cache_node
);
3307 btrfs_put_block_group(cache
);
3309 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3311 btrfs_get_block_group(cache
);
3314 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3318 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3319 struct btrfs_trans_handle
*trans
,
3320 struct btrfs_path
*path
)
3322 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3323 struct inode
*inode
= NULL
;
3325 int dcs
= BTRFS_DC_ERROR
;
3331 * If this block group is smaller than 100 megs don't bother caching the
3334 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3335 spin_lock(&block_group
->lock
);
3336 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3337 spin_unlock(&block_group
->lock
);
3344 inode
= lookup_free_space_inode(root
, block_group
, path
);
3345 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3346 ret
= PTR_ERR(inode
);
3347 btrfs_release_path(path
);
3351 if (IS_ERR(inode
)) {
3355 if (block_group
->ro
)
3358 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3365 * We want to set the generation to 0, that way if anything goes wrong
3366 * from here on out we know not to trust this cache when we load up next
3369 BTRFS_I(inode
)->generation
= 0;
3370 ret
= btrfs_update_inode(trans
, root
, inode
);
3373 * So theoretically we could recover from this, simply set the
3374 * super cache generation to 0 so we know to invalidate the
3375 * cache, but then we'd have to keep track of the block groups
3376 * that fail this way so we know we _have_ to reset this cache
3377 * before the next commit or risk reading stale cache. So to
3378 * limit our exposure to horrible edge cases lets just abort the
3379 * transaction, this only happens in really bad situations
3382 btrfs_abort_transaction(trans
, root
, ret
);
3387 /* We've already setup this transaction, go ahead and exit */
3388 if (block_group
->cache_generation
== trans
->transid
&&
3389 i_size_read(inode
)) {
3390 dcs
= BTRFS_DC_SETUP
;
3394 if (i_size_read(inode
) > 0) {
3395 ret
= btrfs_check_trunc_cache_free_space(root
,
3396 &root
->fs_info
->global_block_rsv
);
3400 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3405 spin_lock(&block_group
->lock
);
3406 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3407 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3409 * don't bother trying to write stuff out _if_
3410 * a) we're not cached,
3411 * b) we're with nospace_cache mount option.
3413 dcs
= BTRFS_DC_WRITTEN
;
3414 spin_unlock(&block_group
->lock
);
3417 spin_unlock(&block_group
->lock
);
3420 * We hit an ENOSPC when setting up the cache in this transaction, just
3421 * skip doing the setup, we've already cleared the cache so we're safe.
3423 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3429 * Try to preallocate enough space based on how big the block group is.
3430 * Keep in mind this has to include any pinned space which could end up
3431 * taking up quite a bit since it's not folded into the other space
3434 num_pages
= div_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3439 num_pages
*= PAGE_CACHE_SIZE
;
3441 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3445 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3446 num_pages
, num_pages
,
3449 * Our cache requires contiguous chunks so that we don't modify a bunch
3450 * of metadata or split extents when writing the cache out, which means
3451 * we can enospc if we are heavily fragmented in addition to just normal
3452 * out of space conditions. So if we hit this just skip setting up any
3453 * other block groups for this transaction, maybe we'll unpin enough
3454 * space the next time around.
3457 dcs
= BTRFS_DC_SETUP
;
3458 else if (ret
== -ENOSPC
)
3459 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3460 btrfs_free_reserved_data_space(inode
, 0, num_pages
);
3465 btrfs_release_path(path
);
3467 spin_lock(&block_group
->lock
);
3468 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3469 block_group
->cache_generation
= trans
->transid
;
3470 block_group
->disk_cache_state
= dcs
;
3471 spin_unlock(&block_group
->lock
);
3476 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3477 struct btrfs_root
*root
)
3479 struct btrfs_block_group_cache
*cache
, *tmp
;
3480 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3481 struct btrfs_path
*path
;
3483 if (list_empty(&cur_trans
->dirty_bgs
) ||
3484 !btrfs_test_opt(root
, SPACE_CACHE
))
3487 path
= btrfs_alloc_path();
3491 /* Could add new block groups, use _safe just in case */
3492 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3494 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3495 cache_save_setup(cache
, trans
, path
);
3498 btrfs_free_path(path
);
3503 * transaction commit does final block group cache writeback during a
3504 * critical section where nothing is allowed to change the FS. This is
3505 * required in order for the cache to actually match the block group,
3506 * but can introduce a lot of latency into the commit.
3508 * So, btrfs_start_dirty_block_groups is here to kick off block group
3509 * cache IO. There's a chance we'll have to redo some of it if the
3510 * block group changes again during the commit, but it greatly reduces
3511 * the commit latency by getting rid of the easy block groups while
3512 * we're still allowing others to join the commit.
3514 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3515 struct btrfs_root
*root
)
3517 struct btrfs_block_group_cache
*cache
;
3518 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3521 struct btrfs_path
*path
= NULL
;
3523 struct list_head
*io
= &cur_trans
->io_bgs
;
3524 int num_started
= 0;
3527 spin_lock(&cur_trans
->dirty_bgs_lock
);
3528 if (list_empty(&cur_trans
->dirty_bgs
)) {
3529 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3532 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3533 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3537 * make sure all the block groups on our dirty list actually
3540 btrfs_create_pending_block_groups(trans
, root
);
3543 path
= btrfs_alloc_path();
3549 * cache_write_mutex is here only to save us from balance or automatic
3550 * removal of empty block groups deleting this block group while we are
3551 * writing out the cache
3553 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3554 while (!list_empty(&dirty
)) {
3555 cache
= list_first_entry(&dirty
,
3556 struct btrfs_block_group_cache
,
3559 * this can happen if something re-dirties a block
3560 * group that is already under IO. Just wait for it to
3561 * finish and then do it all again
3563 if (!list_empty(&cache
->io_list
)) {
3564 list_del_init(&cache
->io_list
);
3565 btrfs_wait_cache_io(root
, trans
, cache
,
3566 &cache
->io_ctl
, path
,
3567 cache
->key
.objectid
);
3568 btrfs_put_block_group(cache
);
3573 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3574 * if it should update the cache_state. Don't delete
3575 * until after we wait.
3577 * Since we're not running in the commit critical section
3578 * we need the dirty_bgs_lock to protect from update_block_group
3580 spin_lock(&cur_trans
->dirty_bgs_lock
);
3581 list_del_init(&cache
->dirty_list
);
3582 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3586 cache_save_setup(cache
, trans
, path
);
3588 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3589 cache
->io_ctl
.inode
= NULL
;
3590 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3591 if (ret
== 0 && cache
->io_ctl
.inode
) {
3596 * the cache_write_mutex is protecting
3599 list_add_tail(&cache
->io_list
, io
);
3602 * if we failed to write the cache, the
3603 * generation will be bad and life goes on
3609 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3611 * Our block group might still be attached to the list
3612 * of new block groups in the transaction handle of some
3613 * other task (struct btrfs_trans_handle->new_bgs). This
3614 * means its block group item isn't yet in the extent
3615 * tree. If this happens ignore the error, as we will
3616 * try again later in the critical section of the
3617 * transaction commit.
3619 if (ret
== -ENOENT
) {
3621 spin_lock(&cur_trans
->dirty_bgs_lock
);
3622 if (list_empty(&cache
->dirty_list
)) {
3623 list_add_tail(&cache
->dirty_list
,
3624 &cur_trans
->dirty_bgs
);
3625 btrfs_get_block_group(cache
);
3627 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3629 btrfs_abort_transaction(trans
, root
, ret
);
3633 /* if its not on the io list, we need to put the block group */
3635 btrfs_put_block_group(cache
);
3641 * Avoid blocking other tasks for too long. It might even save
3642 * us from writing caches for block groups that are going to be
3645 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3646 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3648 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3651 * go through delayed refs for all the stuff we've just kicked off
3652 * and then loop back (just once)
3654 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3655 if (!ret
&& loops
== 0) {
3657 spin_lock(&cur_trans
->dirty_bgs_lock
);
3658 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3660 * dirty_bgs_lock protects us from concurrent block group
3661 * deletes too (not just cache_write_mutex).
3663 if (!list_empty(&dirty
)) {
3664 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3667 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3670 btrfs_free_path(path
);
3674 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3675 struct btrfs_root
*root
)
3677 struct btrfs_block_group_cache
*cache
;
3678 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3681 struct btrfs_path
*path
;
3682 struct list_head
*io
= &cur_trans
->io_bgs
;
3683 int num_started
= 0;
3685 path
= btrfs_alloc_path();
3690 * We don't need the lock here since we are protected by the transaction
3691 * commit. We want to do the cache_save_setup first and then run the
3692 * delayed refs to make sure we have the best chance at doing this all
3695 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3696 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3697 struct btrfs_block_group_cache
,
3701 * this can happen if cache_save_setup re-dirties a block
3702 * group that is already under IO. Just wait for it to
3703 * finish and then do it all again
3705 if (!list_empty(&cache
->io_list
)) {
3706 list_del_init(&cache
->io_list
);
3707 btrfs_wait_cache_io(root
, trans
, cache
,
3708 &cache
->io_ctl
, path
,
3709 cache
->key
.objectid
);
3710 btrfs_put_block_group(cache
);
3714 * don't remove from the dirty list until after we've waited
3717 list_del_init(&cache
->dirty_list
);
3720 cache_save_setup(cache
, trans
, path
);
3723 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3725 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3726 cache
->io_ctl
.inode
= NULL
;
3727 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3728 if (ret
== 0 && cache
->io_ctl
.inode
) {
3731 list_add_tail(&cache
->io_list
, io
);
3734 * if we failed to write the cache, the
3735 * generation will be bad and life goes on
3741 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3743 btrfs_abort_transaction(trans
, root
, ret
);
3746 /* if its not on the io list, we need to put the block group */
3748 btrfs_put_block_group(cache
);
3751 while (!list_empty(io
)) {
3752 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3754 list_del_init(&cache
->io_list
);
3755 btrfs_wait_cache_io(root
, trans
, cache
,
3756 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3757 btrfs_put_block_group(cache
);
3760 btrfs_free_path(path
);
3764 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3766 struct btrfs_block_group_cache
*block_group
;
3769 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3770 if (!block_group
|| block_group
->ro
)
3773 btrfs_put_block_group(block_group
);
3777 static const char *alloc_name(u64 flags
)
3780 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3782 case BTRFS_BLOCK_GROUP_METADATA
:
3784 case BTRFS_BLOCK_GROUP_DATA
:
3786 case BTRFS_BLOCK_GROUP_SYSTEM
:
3790 return "invalid-combination";
3794 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3795 u64 total_bytes
, u64 bytes_used
,
3796 struct btrfs_space_info
**space_info
)
3798 struct btrfs_space_info
*found
;
3803 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3804 BTRFS_BLOCK_GROUP_RAID10
))
3809 found
= __find_space_info(info
, flags
);
3811 spin_lock(&found
->lock
);
3812 found
->total_bytes
+= total_bytes
;
3813 found
->disk_total
+= total_bytes
* factor
;
3814 found
->bytes_used
+= bytes_used
;
3815 found
->disk_used
+= bytes_used
* factor
;
3816 if (total_bytes
> 0)
3818 spin_unlock(&found
->lock
);
3819 *space_info
= found
;
3822 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3826 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3832 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3833 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3834 init_rwsem(&found
->groups_sem
);
3835 spin_lock_init(&found
->lock
);
3836 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3837 found
->total_bytes
= total_bytes
;
3838 found
->disk_total
= total_bytes
* factor
;
3839 found
->bytes_used
= bytes_used
;
3840 found
->disk_used
= bytes_used
* factor
;
3841 found
->bytes_pinned
= 0;
3842 found
->bytes_reserved
= 0;
3843 found
->bytes_readonly
= 0;
3844 found
->bytes_may_use
= 0;
3846 found
->max_extent_size
= 0;
3847 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3848 found
->chunk_alloc
= 0;
3850 init_waitqueue_head(&found
->wait
);
3851 INIT_LIST_HEAD(&found
->ro_bgs
);
3853 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3854 info
->space_info_kobj
, "%s",
3855 alloc_name(found
->flags
));
3857 percpu_counter_destroy(&found
->total_bytes_pinned
);
3862 *space_info
= found
;
3863 list_add_rcu(&found
->list
, &info
->space_info
);
3864 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3865 info
->data_sinfo
= found
;
3870 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3872 u64 extra_flags
= chunk_to_extended(flags
) &
3873 BTRFS_EXTENDED_PROFILE_MASK
;
3875 write_seqlock(&fs_info
->profiles_lock
);
3876 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3877 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3878 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3879 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3880 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3881 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3882 write_sequnlock(&fs_info
->profiles_lock
);
3886 * returns target flags in extended format or 0 if restripe for this
3887 * chunk_type is not in progress
3889 * should be called with either volume_mutex or balance_lock held
3891 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3893 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3899 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3900 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3901 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3902 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3903 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3904 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3905 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3906 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3907 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3914 * @flags: available profiles in extended format (see ctree.h)
3916 * Returns reduced profile in chunk format. If profile changing is in
3917 * progress (either running or paused) picks the target profile (if it's
3918 * already available), otherwise falls back to plain reducing.
3920 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3922 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3928 * see if restripe for this chunk_type is in progress, if so
3929 * try to reduce to the target profile
3931 spin_lock(&root
->fs_info
->balance_lock
);
3932 target
= get_restripe_target(root
->fs_info
, flags
);
3934 /* pick target profile only if it's already available */
3935 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3936 spin_unlock(&root
->fs_info
->balance_lock
);
3937 return extended_to_chunk(target
);
3940 spin_unlock(&root
->fs_info
->balance_lock
);
3942 /* First, mask out the RAID levels which aren't possible */
3943 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3944 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
3945 allowed
|= btrfs_raid_group
[raid_type
];
3949 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
3950 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
3951 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
3952 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
3953 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
3954 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
3955 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
3956 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
3957 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
3958 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
3960 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
3962 return extended_to_chunk(flags
| allowed
);
3965 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3972 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3974 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3975 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3976 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3977 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3978 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3979 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3980 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3982 return btrfs_reduce_alloc_profile(root
, flags
);
3985 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3991 flags
= BTRFS_BLOCK_GROUP_DATA
;
3992 else if (root
== root
->fs_info
->chunk_root
)
3993 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3995 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3997 ret
= get_alloc_profile(root
, flags
);
4001 int btrfs_alloc_data_chunk_ondemand(struct inode
*inode
, u64 bytes
)
4003 struct btrfs_space_info
*data_sinfo
;
4004 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4005 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4008 int need_commit
= 2;
4009 int have_pinned_space
;
4011 /* make sure bytes are sectorsize aligned */
4012 bytes
= ALIGN(bytes
, root
->sectorsize
);
4014 if (btrfs_is_free_space_inode(inode
)) {
4016 ASSERT(current
->journal_info
);
4019 data_sinfo
= fs_info
->data_sinfo
;
4024 /* make sure we have enough space to handle the data first */
4025 spin_lock(&data_sinfo
->lock
);
4026 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
4027 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
4028 data_sinfo
->bytes_may_use
;
4030 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4031 struct btrfs_trans_handle
*trans
;
4034 * if we don't have enough free bytes in this space then we need
4035 * to alloc a new chunk.
4037 if (!data_sinfo
->full
) {
4040 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4041 spin_unlock(&data_sinfo
->lock
);
4043 alloc_target
= btrfs_get_alloc_profile(root
, 1);
4045 * It is ugly that we don't call nolock join
4046 * transaction for the free space inode case here.
4047 * But it is safe because we only do the data space
4048 * reservation for the free space cache in the
4049 * transaction context, the common join transaction
4050 * just increase the counter of the current transaction
4051 * handler, doesn't try to acquire the trans_lock of
4054 trans
= btrfs_join_transaction(root
);
4056 return PTR_ERR(trans
);
4058 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4060 CHUNK_ALLOC_NO_FORCE
);
4061 btrfs_end_transaction(trans
, root
);
4066 have_pinned_space
= 1;
4072 data_sinfo
= fs_info
->data_sinfo
;
4078 * If we don't have enough pinned space to deal with this
4079 * allocation, and no removed chunk in current transaction,
4080 * don't bother committing the transaction.
4082 have_pinned_space
= percpu_counter_compare(
4083 &data_sinfo
->total_bytes_pinned
,
4084 used
+ bytes
- data_sinfo
->total_bytes
);
4085 spin_unlock(&data_sinfo
->lock
);
4087 /* commit the current transaction and try again */
4090 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
4093 if (need_commit
> 0) {
4094 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4095 btrfs_wait_ordered_roots(fs_info
, -1);
4098 trans
= btrfs_join_transaction(root
);
4100 return PTR_ERR(trans
);
4101 if (have_pinned_space
>= 0 ||
4102 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4103 &trans
->transaction
->flags
) ||
4105 ret
= btrfs_commit_transaction(trans
, root
);
4109 * The cleaner kthread might still be doing iput
4110 * operations. Wait for it to finish so that
4111 * more space is released.
4113 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4114 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4117 btrfs_end_transaction(trans
, root
);
4121 trace_btrfs_space_reservation(root
->fs_info
,
4122 "space_info:enospc",
4123 data_sinfo
->flags
, bytes
, 1);
4126 data_sinfo
->bytes_may_use
+= bytes
;
4127 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4128 data_sinfo
->flags
, bytes
, 1);
4129 spin_unlock(&data_sinfo
->lock
);
4135 * New check_data_free_space() with ability for precious data reservation
4136 * Will replace old btrfs_check_data_free_space(), but for patch split,
4137 * add a new function first and then replace it.
4139 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4141 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4144 /* align the range */
4145 len
= round_up(start
+ len
, root
->sectorsize
) -
4146 round_down(start
, root
->sectorsize
);
4147 start
= round_down(start
, root
->sectorsize
);
4149 ret
= btrfs_alloc_data_chunk_ondemand(inode
, len
);
4154 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4156 * TODO: Find a good method to avoid reserve data space for NOCOW
4157 * range, but don't impact performance on quota disable case.
4159 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4164 * Called if we need to clear a data reservation for this inode
4165 * Normally in a error case.
4167 * This one will *NOT* use accurate qgroup reserved space API, just for case
4168 * which we can't sleep and is sure it won't affect qgroup reserved space.
4169 * Like clear_bit_hook().
4171 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4174 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4175 struct btrfs_space_info
*data_sinfo
;
4177 /* Make sure the range is aligned to sectorsize */
4178 len
= round_up(start
+ len
, root
->sectorsize
) -
4179 round_down(start
, root
->sectorsize
);
4180 start
= round_down(start
, root
->sectorsize
);
4182 data_sinfo
= root
->fs_info
->data_sinfo
;
4183 spin_lock(&data_sinfo
->lock
);
4184 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4185 data_sinfo
->bytes_may_use
= 0;
4187 data_sinfo
->bytes_may_use
-= len
;
4188 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4189 data_sinfo
->flags
, len
, 0);
4190 spin_unlock(&data_sinfo
->lock
);
4194 * Called if we need to clear a data reservation for this inode
4195 * Normally in a error case.
4197 * This one will handle the per-indoe data rsv map for accurate reserved
4200 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4202 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4203 btrfs_qgroup_free_data(inode
, start
, len
);
4206 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4208 struct list_head
*head
= &info
->space_info
;
4209 struct btrfs_space_info
*found
;
4212 list_for_each_entry_rcu(found
, head
, list
) {
4213 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4214 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4219 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4221 return (global
->size
<< 1);
4224 static int should_alloc_chunk(struct btrfs_root
*root
,
4225 struct btrfs_space_info
*sinfo
, int force
)
4227 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4228 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4229 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4232 if (force
== CHUNK_ALLOC_FORCE
)
4236 * We need to take into account the global rsv because for all intents
4237 * and purposes it's used space. Don't worry about locking the
4238 * global_rsv, it doesn't change except when the transaction commits.
4240 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4241 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4244 * in limited mode, we want to have some free space up to
4245 * about 1% of the FS size.
4247 if (force
== CHUNK_ALLOC_LIMITED
) {
4248 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4249 thresh
= max_t(u64
, 64 * 1024 * 1024,
4250 div_factor_fine(thresh
, 1));
4252 if (num_bytes
- num_allocated
< thresh
)
4256 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
4261 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4265 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4266 BTRFS_BLOCK_GROUP_RAID0
|
4267 BTRFS_BLOCK_GROUP_RAID5
|
4268 BTRFS_BLOCK_GROUP_RAID6
))
4269 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4270 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4273 num_dev
= 1; /* DUP or single */
4279 * If @is_allocation is true, reserve space in the system space info necessary
4280 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4283 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4284 struct btrfs_root
*root
,
4287 struct btrfs_space_info
*info
;
4294 * Needed because we can end up allocating a system chunk and for an
4295 * atomic and race free space reservation in the chunk block reserve.
4297 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4299 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4300 spin_lock(&info
->lock
);
4301 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4302 info
->bytes_reserved
- info
->bytes_readonly
-
4303 info
->bytes_may_use
;
4304 spin_unlock(&info
->lock
);
4306 num_devs
= get_profile_num_devs(root
, type
);
4308 /* num_devs device items to update and 1 chunk item to add or remove */
4309 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4310 btrfs_calc_trans_metadata_size(root
, 1);
4312 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4313 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4314 left
, thresh
, type
);
4315 dump_space_info(info
, 0, 0);
4318 if (left
< thresh
) {
4321 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4323 * Ignore failure to create system chunk. We might end up not
4324 * needing it, as we might not need to COW all nodes/leafs from
4325 * the paths we visit in the chunk tree (they were already COWed
4326 * or created in the current transaction for example).
4328 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4332 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4333 &root
->fs_info
->chunk_block_rsv
,
4334 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4336 trans
->chunk_bytes_reserved
+= thresh
;
4340 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4341 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4343 struct btrfs_space_info
*space_info
;
4344 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4345 int wait_for_alloc
= 0;
4348 /* Don't re-enter if we're already allocating a chunk */
4349 if (trans
->allocating_chunk
)
4352 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4354 ret
= update_space_info(extent_root
->fs_info
, flags
,
4356 BUG_ON(ret
); /* -ENOMEM */
4358 BUG_ON(!space_info
); /* Logic error */
4361 spin_lock(&space_info
->lock
);
4362 if (force
< space_info
->force_alloc
)
4363 force
= space_info
->force_alloc
;
4364 if (space_info
->full
) {
4365 if (should_alloc_chunk(extent_root
, space_info
, force
))
4369 spin_unlock(&space_info
->lock
);
4373 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4374 spin_unlock(&space_info
->lock
);
4376 } else if (space_info
->chunk_alloc
) {
4379 space_info
->chunk_alloc
= 1;
4382 spin_unlock(&space_info
->lock
);
4384 mutex_lock(&fs_info
->chunk_mutex
);
4387 * The chunk_mutex is held throughout the entirety of a chunk
4388 * allocation, so once we've acquired the chunk_mutex we know that the
4389 * other guy is done and we need to recheck and see if we should
4392 if (wait_for_alloc
) {
4393 mutex_unlock(&fs_info
->chunk_mutex
);
4399 trans
->allocating_chunk
= true;
4402 * If we have mixed data/metadata chunks we want to make sure we keep
4403 * allocating mixed chunks instead of individual chunks.
4405 if (btrfs_mixed_space_info(space_info
))
4406 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4409 * if we're doing a data chunk, go ahead and make sure that
4410 * we keep a reasonable number of metadata chunks allocated in the
4413 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4414 fs_info
->data_chunk_allocations
++;
4415 if (!(fs_info
->data_chunk_allocations
%
4416 fs_info
->metadata_ratio
))
4417 force_metadata_allocation(fs_info
);
4421 * Check if we have enough space in SYSTEM chunk because we may need
4422 * to update devices.
4424 check_system_chunk(trans
, extent_root
, flags
);
4426 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4427 trans
->allocating_chunk
= false;
4429 spin_lock(&space_info
->lock
);
4430 if (ret
< 0 && ret
!= -ENOSPC
)
4433 space_info
->full
= 1;
4437 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4439 space_info
->chunk_alloc
= 0;
4440 spin_unlock(&space_info
->lock
);
4441 mutex_unlock(&fs_info
->chunk_mutex
);
4443 * When we allocate a new chunk we reserve space in the chunk block
4444 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4445 * add new nodes/leafs to it if we end up needing to do it when
4446 * inserting the chunk item and updating device items as part of the
4447 * second phase of chunk allocation, performed by
4448 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4449 * large number of new block groups to create in our transaction
4450 * handle's new_bgs list to avoid exhausting the chunk block reserve
4451 * in extreme cases - like having a single transaction create many new
4452 * block groups when starting to write out the free space caches of all
4453 * the block groups that were made dirty during the lifetime of the
4456 if (trans
->can_flush_pending_bgs
&&
4457 trans
->chunk_bytes_reserved
>= (2 * 1024 * 1024ull)) {
4458 btrfs_create_pending_block_groups(trans
, trans
->root
);
4459 btrfs_trans_release_chunk_metadata(trans
);
4464 static int can_overcommit(struct btrfs_root
*root
,
4465 struct btrfs_space_info
*space_info
, u64 bytes
,
4466 enum btrfs_reserve_flush_enum flush
)
4468 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4469 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4474 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4475 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4478 * We only want to allow over committing if we have lots of actual space
4479 * free, but if we don't have enough space to handle the global reserve
4480 * space then we could end up having a real enospc problem when trying
4481 * to allocate a chunk or some other such important allocation.
4483 spin_lock(&global_rsv
->lock
);
4484 space_size
= calc_global_rsv_need_space(global_rsv
);
4485 spin_unlock(&global_rsv
->lock
);
4486 if (used
+ space_size
>= space_info
->total_bytes
)
4489 used
+= space_info
->bytes_may_use
;
4491 spin_lock(&root
->fs_info
->free_chunk_lock
);
4492 avail
= root
->fs_info
->free_chunk_space
;
4493 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4496 * If we have dup, raid1 or raid10 then only half of the free
4497 * space is actually useable. For raid56, the space info used
4498 * doesn't include the parity drive, so we don't have to
4501 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4502 BTRFS_BLOCK_GROUP_RAID1
|
4503 BTRFS_BLOCK_GROUP_RAID10
))
4507 * If we aren't flushing all things, let us overcommit up to
4508 * 1/2th of the space. If we can flush, don't let us overcommit
4509 * too much, let it overcommit up to 1/8 of the space.
4511 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4516 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4521 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4522 unsigned long nr_pages
, int nr_items
)
4524 struct super_block
*sb
= root
->fs_info
->sb
;
4526 if (down_read_trylock(&sb
->s_umount
)) {
4527 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4528 up_read(&sb
->s_umount
);
4531 * We needn't worry the filesystem going from r/w to r/o though
4532 * we don't acquire ->s_umount mutex, because the filesystem
4533 * should guarantee the delalloc inodes list be empty after
4534 * the filesystem is readonly(all dirty pages are written to
4537 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4538 if (!current
->journal_info
)
4539 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4543 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4548 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4549 nr
= (int)div64_u64(to_reclaim
, bytes
);
4555 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4558 * shrink metadata reservation for delalloc
4560 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4563 struct btrfs_block_rsv
*block_rsv
;
4564 struct btrfs_space_info
*space_info
;
4565 struct btrfs_trans_handle
*trans
;
4569 unsigned long nr_pages
;
4572 enum btrfs_reserve_flush_enum flush
;
4574 /* Calc the number of the pages we need flush for space reservation */
4575 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4576 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4578 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4579 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4580 space_info
= block_rsv
->space_info
;
4582 delalloc_bytes
= percpu_counter_sum_positive(
4583 &root
->fs_info
->delalloc_bytes
);
4584 if (delalloc_bytes
== 0) {
4588 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4593 while (delalloc_bytes
&& loops
< 3) {
4594 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4595 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4596 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4598 * We need to wait for the async pages to actually start before
4601 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4605 if (max_reclaim
<= nr_pages
)
4608 max_reclaim
-= nr_pages
;
4610 wait_event(root
->fs_info
->async_submit_wait
,
4611 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4615 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4617 flush
= BTRFS_RESERVE_NO_FLUSH
;
4618 spin_lock(&space_info
->lock
);
4619 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4620 spin_unlock(&space_info
->lock
);
4623 spin_unlock(&space_info
->lock
);
4626 if (wait_ordered
&& !trans
) {
4627 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4629 time_left
= schedule_timeout_killable(1);
4633 delalloc_bytes
= percpu_counter_sum_positive(
4634 &root
->fs_info
->delalloc_bytes
);
4639 * maybe_commit_transaction - possibly commit the transaction if its ok to
4640 * @root - the root we're allocating for
4641 * @bytes - the number of bytes we want to reserve
4642 * @force - force the commit
4644 * This will check to make sure that committing the transaction will actually
4645 * get us somewhere and then commit the transaction if it does. Otherwise it
4646 * will return -ENOSPC.
4648 static int may_commit_transaction(struct btrfs_root
*root
,
4649 struct btrfs_space_info
*space_info
,
4650 u64 bytes
, int force
)
4652 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4653 struct btrfs_trans_handle
*trans
;
4655 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4662 /* See if there is enough pinned space to make this reservation */
4663 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4668 * See if there is some space in the delayed insertion reservation for
4671 if (space_info
!= delayed_rsv
->space_info
)
4674 spin_lock(&delayed_rsv
->lock
);
4675 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4676 bytes
- delayed_rsv
->size
) >= 0) {
4677 spin_unlock(&delayed_rsv
->lock
);
4680 spin_unlock(&delayed_rsv
->lock
);
4683 trans
= btrfs_join_transaction(root
);
4687 return btrfs_commit_transaction(trans
, root
);
4691 FLUSH_DELAYED_ITEMS_NR
= 1,
4692 FLUSH_DELAYED_ITEMS
= 2,
4694 FLUSH_DELALLOC_WAIT
= 4,
4699 static int flush_space(struct btrfs_root
*root
,
4700 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4701 u64 orig_bytes
, int state
)
4703 struct btrfs_trans_handle
*trans
;
4708 case FLUSH_DELAYED_ITEMS_NR
:
4709 case FLUSH_DELAYED_ITEMS
:
4710 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4711 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4715 trans
= btrfs_join_transaction(root
);
4716 if (IS_ERR(trans
)) {
4717 ret
= PTR_ERR(trans
);
4720 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4721 btrfs_end_transaction(trans
, root
);
4723 case FLUSH_DELALLOC
:
4724 case FLUSH_DELALLOC_WAIT
:
4725 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4726 state
== FLUSH_DELALLOC_WAIT
);
4729 trans
= btrfs_join_transaction(root
);
4730 if (IS_ERR(trans
)) {
4731 ret
= PTR_ERR(trans
);
4734 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4735 btrfs_get_alloc_profile(root
, 0),
4736 CHUNK_ALLOC_NO_FORCE
);
4737 btrfs_end_transaction(trans
, root
);
4742 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4753 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4754 struct btrfs_space_info
*space_info
)
4760 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4762 spin_lock(&space_info
->lock
);
4763 if (can_overcommit(root
, space_info
, to_reclaim
,
4764 BTRFS_RESERVE_FLUSH_ALL
)) {
4769 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4770 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4771 space_info
->bytes_may_use
;
4772 if (can_overcommit(root
, space_info
, 1024 * 1024,
4773 BTRFS_RESERVE_FLUSH_ALL
))
4774 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4776 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4778 if (used
> expected
)
4779 to_reclaim
= used
- expected
;
4782 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4783 space_info
->bytes_reserved
);
4785 spin_unlock(&space_info
->lock
);
4790 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4791 struct btrfs_fs_info
*fs_info
, u64 used
)
4793 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4795 /* If we're just plain full then async reclaim just slows us down. */
4796 if (space_info
->bytes_used
>= thresh
)
4799 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4800 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4803 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4804 struct btrfs_fs_info
*fs_info
,
4809 spin_lock(&space_info
->lock
);
4811 * We run out of space and have not got any free space via flush_space,
4812 * so don't bother doing async reclaim.
4814 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4815 spin_unlock(&space_info
->lock
);
4819 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4820 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4821 space_info
->bytes_may_use
;
4822 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4823 spin_unlock(&space_info
->lock
);
4826 spin_unlock(&space_info
->lock
);
4831 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4833 struct btrfs_fs_info
*fs_info
;
4834 struct btrfs_space_info
*space_info
;
4838 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4839 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4841 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4846 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4848 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4849 to_reclaim
, flush_state
);
4851 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4854 } while (flush_state
< COMMIT_TRANS
);
4857 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4859 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4863 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4864 * @root - the root we're allocating for
4865 * @block_rsv - the block_rsv we're allocating for
4866 * @orig_bytes - the number of bytes we want
4867 * @flush - whether or not we can flush to make our reservation
4869 * This will reserve orgi_bytes number of bytes from the space info associated
4870 * with the block_rsv. If there is not enough space it will make an attempt to
4871 * flush out space to make room. It will do this by flushing delalloc if
4872 * possible or committing the transaction. If flush is 0 then no attempts to
4873 * regain reservations will be made and this will fail if there is not enough
4876 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4877 struct btrfs_block_rsv
*block_rsv
,
4879 enum btrfs_reserve_flush_enum flush
)
4881 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4883 u64 num_bytes
= orig_bytes
;
4884 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4886 bool flushing
= false;
4890 spin_lock(&space_info
->lock
);
4892 * We only want to wait if somebody other than us is flushing and we
4893 * are actually allowed to flush all things.
4895 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4896 space_info
->flush
) {
4897 spin_unlock(&space_info
->lock
);
4899 * If we have a trans handle we can't wait because the flusher
4900 * may have to commit the transaction, which would mean we would
4901 * deadlock since we are waiting for the flusher to finish, but
4902 * hold the current transaction open.
4904 if (current
->journal_info
)
4906 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4907 /* Must have been killed, return */
4911 spin_lock(&space_info
->lock
);
4915 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4916 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4917 space_info
->bytes_may_use
;
4920 * The idea here is that we've not already over-reserved the block group
4921 * then we can go ahead and save our reservation first and then start
4922 * flushing if we need to. Otherwise if we've already overcommitted
4923 * lets start flushing stuff first and then come back and try to make
4926 if (used
<= space_info
->total_bytes
) {
4927 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4928 space_info
->bytes_may_use
+= orig_bytes
;
4929 trace_btrfs_space_reservation(root
->fs_info
,
4930 "space_info", space_info
->flags
, orig_bytes
, 1);
4934 * Ok set num_bytes to orig_bytes since we aren't
4935 * overocmmitted, this way we only try and reclaim what
4938 num_bytes
= orig_bytes
;
4942 * Ok we're over committed, set num_bytes to the overcommitted
4943 * amount plus the amount of bytes that we need for this
4946 num_bytes
= used
- space_info
->total_bytes
+
4950 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4951 space_info
->bytes_may_use
+= orig_bytes
;
4952 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4953 space_info
->flags
, orig_bytes
,
4959 * Couldn't make our reservation, save our place so while we're trying
4960 * to reclaim space we can actually use it instead of somebody else
4961 * stealing it from us.
4963 * We make the other tasks wait for the flush only when we can flush
4966 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4968 space_info
->flush
= 1;
4969 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4972 * We will do the space reservation dance during log replay,
4973 * which means we won't have fs_info->fs_root set, so don't do
4974 * the async reclaim as we will panic.
4976 if (!root
->fs_info
->log_root_recovering
&&
4977 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4978 !work_busy(&root
->fs_info
->async_reclaim_work
))
4979 queue_work(system_unbound_wq
,
4980 &root
->fs_info
->async_reclaim_work
);
4982 spin_unlock(&space_info
->lock
);
4984 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4987 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4992 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4993 * would happen. So skip delalloc flush.
4995 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4996 (flush_state
== FLUSH_DELALLOC
||
4997 flush_state
== FLUSH_DELALLOC_WAIT
))
4998 flush_state
= ALLOC_CHUNK
;
5002 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
5003 flush_state
< COMMIT_TRANS
)
5005 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
5006 flush_state
<= COMMIT_TRANS
)
5010 if (ret
== -ENOSPC
&&
5011 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5012 struct btrfs_block_rsv
*global_rsv
=
5013 &root
->fs_info
->global_block_rsv
;
5015 if (block_rsv
!= global_rsv
&&
5016 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5020 trace_btrfs_space_reservation(root
->fs_info
,
5021 "space_info:enospc",
5022 space_info
->flags
, orig_bytes
, 1);
5024 spin_lock(&space_info
->lock
);
5025 space_info
->flush
= 0;
5026 wake_up_all(&space_info
->wait
);
5027 spin_unlock(&space_info
->lock
);
5032 static struct btrfs_block_rsv
*get_block_rsv(
5033 const struct btrfs_trans_handle
*trans
,
5034 const struct btrfs_root
*root
)
5036 struct btrfs_block_rsv
*block_rsv
= NULL
;
5038 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5039 (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
) ||
5040 (root
== root
->fs_info
->uuid_root
))
5041 block_rsv
= trans
->block_rsv
;
5044 block_rsv
= root
->block_rsv
;
5047 block_rsv
= &root
->fs_info
->empty_block_rsv
;
5052 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5056 spin_lock(&block_rsv
->lock
);
5057 if (block_rsv
->reserved
>= num_bytes
) {
5058 block_rsv
->reserved
-= num_bytes
;
5059 if (block_rsv
->reserved
< block_rsv
->size
)
5060 block_rsv
->full
= 0;
5063 spin_unlock(&block_rsv
->lock
);
5067 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5068 u64 num_bytes
, int update_size
)
5070 spin_lock(&block_rsv
->lock
);
5071 block_rsv
->reserved
+= num_bytes
;
5073 block_rsv
->size
+= num_bytes
;
5074 else if (block_rsv
->reserved
>= block_rsv
->size
)
5075 block_rsv
->full
= 1;
5076 spin_unlock(&block_rsv
->lock
);
5079 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5080 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5083 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5086 if (global_rsv
->space_info
!= dest
->space_info
)
5089 spin_lock(&global_rsv
->lock
);
5090 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5091 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5092 spin_unlock(&global_rsv
->lock
);
5095 global_rsv
->reserved
-= num_bytes
;
5096 if (global_rsv
->reserved
< global_rsv
->size
)
5097 global_rsv
->full
= 0;
5098 spin_unlock(&global_rsv
->lock
);
5100 block_rsv_add_bytes(dest
, num_bytes
, 1);
5104 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5105 struct btrfs_block_rsv
*block_rsv
,
5106 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5108 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5110 spin_lock(&block_rsv
->lock
);
5111 if (num_bytes
== (u64
)-1)
5112 num_bytes
= block_rsv
->size
;
5113 block_rsv
->size
-= num_bytes
;
5114 if (block_rsv
->reserved
>= block_rsv
->size
) {
5115 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5116 block_rsv
->reserved
= block_rsv
->size
;
5117 block_rsv
->full
= 1;
5121 spin_unlock(&block_rsv
->lock
);
5123 if (num_bytes
> 0) {
5125 spin_lock(&dest
->lock
);
5129 bytes_to_add
= dest
->size
- dest
->reserved
;
5130 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5131 dest
->reserved
+= bytes_to_add
;
5132 if (dest
->reserved
>= dest
->size
)
5134 num_bytes
-= bytes_to_add
;
5136 spin_unlock(&dest
->lock
);
5139 spin_lock(&space_info
->lock
);
5140 space_info
->bytes_may_use
-= num_bytes
;
5141 trace_btrfs_space_reservation(fs_info
, "space_info",
5142 space_info
->flags
, num_bytes
, 0);
5143 spin_unlock(&space_info
->lock
);
5148 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
5149 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
5153 ret
= block_rsv_use_bytes(src
, num_bytes
);
5157 block_rsv_add_bytes(dst
, num_bytes
, 1);
5161 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5163 memset(rsv
, 0, sizeof(*rsv
));
5164 spin_lock_init(&rsv
->lock
);
5168 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
5169 unsigned short type
)
5171 struct btrfs_block_rsv
*block_rsv
;
5172 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5174 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5178 btrfs_init_block_rsv(block_rsv
, type
);
5179 block_rsv
->space_info
= __find_space_info(fs_info
,
5180 BTRFS_BLOCK_GROUP_METADATA
);
5184 void btrfs_free_block_rsv(struct btrfs_root
*root
,
5185 struct btrfs_block_rsv
*rsv
)
5189 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5193 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5198 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5199 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5200 enum btrfs_reserve_flush_enum flush
)
5207 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5209 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5216 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5217 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5225 spin_lock(&block_rsv
->lock
);
5226 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5227 if (block_rsv
->reserved
>= num_bytes
)
5229 spin_unlock(&block_rsv
->lock
);
5234 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5235 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5236 enum btrfs_reserve_flush_enum flush
)
5244 spin_lock(&block_rsv
->lock
);
5245 num_bytes
= min_reserved
;
5246 if (block_rsv
->reserved
>= num_bytes
)
5249 num_bytes
-= block_rsv
->reserved
;
5250 spin_unlock(&block_rsv
->lock
);
5255 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5257 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5264 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
5265 struct btrfs_block_rsv
*dst_rsv
,
5268 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5271 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5272 struct btrfs_block_rsv
*block_rsv
,
5275 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5276 if (global_rsv
== block_rsv
||
5277 block_rsv
->space_info
!= global_rsv
->space_info
)
5279 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5284 * helper to calculate size of global block reservation.
5285 * the desired value is sum of space used by extent tree,
5286 * checksum tree and root tree
5288 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5290 struct btrfs_space_info
*sinfo
;
5294 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5296 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5297 spin_lock(&sinfo
->lock
);
5298 data_used
= sinfo
->bytes_used
;
5299 spin_unlock(&sinfo
->lock
);
5301 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5302 spin_lock(&sinfo
->lock
);
5303 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5305 meta_used
= sinfo
->bytes_used
;
5306 spin_unlock(&sinfo
->lock
);
5308 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5310 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5312 if (num_bytes
* 3 > meta_used
)
5313 num_bytes
= div_u64(meta_used
, 3);
5315 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5318 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5320 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5321 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5324 num_bytes
= calc_global_metadata_size(fs_info
);
5326 spin_lock(&sinfo
->lock
);
5327 spin_lock(&block_rsv
->lock
);
5329 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
5331 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5332 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5333 sinfo
->bytes_may_use
;
5335 if (sinfo
->total_bytes
> num_bytes
) {
5336 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5337 block_rsv
->reserved
+= num_bytes
;
5338 sinfo
->bytes_may_use
+= num_bytes
;
5339 trace_btrfs_space_reservation(fs_info
, "space_info",
5340 sinfo
->flags
, num_bytes
, 1);
5343 if (block_rsv
->reserved
>= block_rsv
->size
) {
5344 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5345 sinfo
->bytes_may_use
-= num_bytes
;
5346 trace_btrfs_space_reservation(fs_info
, "space_info",
5347 sinfo
->flags
, num_bytes
, 0);
5348 block_rsv
->reserved
= block_rsv
->size
;
5349 block_rsv
->full
= 1;
5352 spin_unlock(&block_rsv
->lock
);
5353 spin_unlock(&sinfo
->lock
);
5356 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5358 struct btrfs_space_info
*space_info
;
5360 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5361 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5363 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5364 fs_info
->global_block_rsv
.space_info
= space_info
;
5365 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5366 fs_info
->trans_block_rsv
.space_info
= space_info
;
5367 fs_info
->empty_block_rsv
.space_info
= space_info
;
5368 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5370 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5371 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5372 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5373 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5374 if (fs_info
->quota_root
)
5375 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5376 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5378 update_global_block_rsv(fs_info
);
5381 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5383 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5385 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5386 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5387 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5388 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5389 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5390 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5391 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5392 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5395 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5396 struct btrfs_root
*root
)
5398 if (!trans
->block_rsv
)
5401 if (!trans
->bytes_reserved
)
5404 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5405 trans
->transid
, trans
->bytes_reserved
, 0);
5406 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5407 trans
->bytes_reserved
= 0;
5411 * To be called after all the new block groups attached to the transaction
5412 * handle have been created (btrfs_create_pending_block_groups()).
5414 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5416 struct btrfs_fs_info
*fs_info
= trans
->root
->fs_info
;
5418 if (!trans
->chunk_bytes_reserved
)
5421 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5423 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5424 trans
->chunk_bytes_reserved
);
5425 trans
->chunk_bytes_reserved
= 0;
5428 /* Can only return 0 or -ENOSPC */
5429 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5430 struct inode
*inode
)
5432 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5433 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5434 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5437 * We need to hold space in order to delete our orphan item once we've
5438 * added it, so this takes the reservation so we can release it later
5439 * when we are truly done with the orphan item.
5441 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5442 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5443 btrfs_ino(inode
), num_bytes
, 1);
5444 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5447 void btrfs_orphan_release_metadata(struct inode
*inode
)
5449 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5450 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5451 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5452 btrfs_ino(inode
), num_bytes
, 0);
5453 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5457 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5458 * root: the root of the parent directory
5459 * rsv: block reservation
5460 * items: the number of items that we need do reservation
5461 * qgroup_reserved: used to return the reserved size in qgroup
5463 * This function is used to reserve the space for snapshot/subvolume
5464 * creation and deletion. Those operations are different with the
5465 * common file/directory operations, they change two fs/file trees
5466 * and root tree, the number of items that the qgroup reserves is
5467 * different with the free space reservation. So we can not use
5468 * the space reseravtion mechanism in start_transaction().
5470 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5471 struct btrfs_block_rsv
*rsv
,
5473 u64
*qgroup_reserved
,
5474 bool use_global_rsv
)
5478 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5480 if (root
->fs_info
->quota_enabled
) {
5481 /* One for parent inode, two for dir entries */
5482 num_bytes
= 3 * root
->nodesize
;
5483 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
);
5490 *qgroup_reserved
= num_bytes
;
5492 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5493 rsv
->space_info
= __find_space_info(root
->fs_info
,
5494 BTRFS_BLOCK_GROUP_METADATA
);
5495 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5496 BTRFS_RESERVE_FLUSH_ALL
);
5498 if (ret
== -ENOSPC
&& use_global_rsv
)
5499 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5501 if (ret
&& *qgroup_reserved
)
5502 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5507 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5508 struct btrfs_block_rsv
*rsv
,
5509 u64 qgroup_reserved
)
5511 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5515 * drop_outstanding_extent - drop an outstanding extent
5516 * @inode: the inode we're dropping the extent for
5517 * @num_bytes: the number of bytes we're relaseing.
5519 * This is called when we are freeing up an outstanding extent, either called
5520 * after an error or after an extent is written. This will return the number of
5521 * reserved extents that need to be freed. This must be called with
5522 * BTRFS_I(inode)->lock held.
5524 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5526 unsigned drop_inode_space
= 0;
5527 unsigned dropped_extents
= 0;
5528 unsigned num_extents
= 0;
5530 num_extents
= (unsigned)div64_u64(num_bytes
+
5531 BTRFS_MAX_EXTENT_SIZE
- 1,
5532 BTRFS_MAX_EXTENT_SIZE
);
5533 ASSERT(num_extents
);
5534 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5535 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5537 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5538 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5539 &BTRFS_I(inode
)->runtime_flags
))
5540 drop_inode_space
= 1;
5543 * If we have more or the same amount of outsanding extents than we have
5544 * reserved then we need to leave the reserved extents count alone.
5546 if (BTRFS_I(inode
)->outstanding_extents
>=
5547 BTRFS_I(inode
)->reserved_extents
)
5548 return drop_inode_space
;
5550 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5551 BTRFS_I(inode
)->outstanding_extents
;
5552 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5553 return dropped_extents
+ drop_inode_space
;
5557 * calc_csum_metadata_size - return the amount of metada space that must be
5558 * reserved/free'd for the given bytes.
5559 * @inode: the inode we're manipulating
5560 * @num_bytes: the number of bytes in question
5561 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5563 * This adjusts the number of csum_bytes in the inode and then returns the
5564 * correct amount of metadata that must either be reserved or freed. We
5565 * calculate how many checksums we can fit into one leaf and then divide the
5566 * number of bytes that will need to be checksumed by this value to figure out
5567 * how many checksums will be required. If we are adding bytes then the number
5568 * may go up and we will return the number of additional bytes that must be
5569 * reserved. If it is going down we will return the number of bytes that must
5572 * This must be called with BTRFS_I(inode)->lock held.
5574 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5577 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5578 u64 old_csums
, num_csums
;
5580 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5581 BTRFS_I(inode
)->csum_bytes
== 0)
5584 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5586 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5588 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5589 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5591 /* No change, no need to reserve more */
5592 if (old_csums
== num_csums
)
5596 return btrfs_calc_trans_metadata_size(root
,
5597 num_csums
- old_csums
);
5599 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5602 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5604 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5605 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5608 unsigned nr_extents
= 0;
5609 int extra_reserve
= 0;
5610 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5612 bool delalloc_lock
= true;
5616 /* If we are a free space inode we need to not flush since we will be in
5617 * the middle of a transaction commit. We also don't need the delalloc
5618 * mutex since we won't race with anybody. We need this mostly to make
5619 * lockdep shut its filthy mouth.
5621 if (btrfs_is_free_space_inode(inode
)) {
5622 flush
= BTRFS_RESERVE_NO_FLUSH
;
5623 delalloc_lock
= false;
5626 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5627 btrfs_transaction_in_commit(root
->fs_info
))
5628 schedule_timeout(1);
5631 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5633 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5635 spin_lock(&BTRFS_I(inode
)->lock
);
5636 nr_extents
= (unsigned)div64_u64(num_bytes
+
5637 BTRFS_MAX_EXTENT_SIZE
- 1,
5638 BTRFS_MAX_EXTENT_SIZE
);
5639 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5642 if (BTRFS_I(inode
)->outstanding_extents
>
5643 BTRFS_I(inode
)->reserved_extents
)
5644 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5645 BTRFS_I(inode
)->reserved_extents
;
5648 * Add an item to reserve for updating the inode when we complete the
5651 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5652 &BTRFS_I(inode
)->runtime_flags
)) {
5657 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5658 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5659 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5660 spin_unlock(&BTRFS_I(inode
)->lock
);
5662 if (root
->fs_info
->quota_enabled
) {
5663 ret
= btrfs_qgroup_reserve_meta(root
,
5664 nr_extents
* root
->nodesize
);
5669 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5670 if (unlikely(ret
)) {
5671 btrfs_qgroup_free_meta(root
, nr_extents
* root
->nodesize
);
5675 spin_lock(&BTRFS_I(inode
)->lock
);
5676 if (extra_reserve
) {
5677 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5678 &BTRFS_I(inode
)->runtime_flags
);
5681 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5682 spin_unlock(&BTRFS_I(inode
)->lock
);
5685 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5688 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5689 btrfs_ino(inode
), to_reserve
, 1);
5690 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5695 spin_lock(&BTRFS_I(inode
)->lock
);
5696 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5698 * If the inodes csum_bytes is the same as the original
5699 * csum_bytes then we know we haven't raced with any free()ers
5700 * so we can just reduce our inodes csum bytes and carry on.
5702 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5703 calc_csum_metadata_size(inode
, num_bytes
, 0);
5705 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5709 * This is tricky, but first we need to figure out how much we
5710 * free'd from any free-ers that occured during this
5711 * reservation, so we reset ->csum_bytes to the csum_bytes
5712 * before we dropped our lock, and then call the free for the
5713 * number of bytes that were freed while we were trying our
5716 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5717 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5718 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5722 * Now we need to see how much we would have freed had we not
5723 * been making this reservation and our ->csum_bytes were not
5724 * artificially inflated.
5726 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5727 bytes
= csum_bytes
- orig_csum_bytes
;
5728 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5731 * Now reset ->csum_bytes to what it should be. If bytes is
5732 * more than to_free then we would have free'd more space had we
5733 * not had an artificially high ->csum_bytes, so we need to free
5734 * the remainder. If bytes is the same or less then we don't
5735 * need to do anything, the other free-ers did the correct
5738 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5739 if (bytes
> to_free
)
5740 to_free
= bytes
- to_free
;
5744 spin_unlock(&BTRFS_I(inode
)->lock
);
5746 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5749 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5750 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5751 btrfs_ino(inode
), to_free
, 0);
5754 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5759 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5760 * @inode: the inode to release the reservation for
5761 * @num_bytes: the number of bytes we're releasing
5763 * This will release the metadata reservation for an inode. This can be called
5764 * once we complete IO for a given set of bytes to release their metadata
5767 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5769 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5773 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5774 spin_lock(&BTRFS_I(inode
)->lock
);
5775 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5778 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5779 spin_unlock(&BTRFS_I(inode
)->lock
);
5781 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5783 if (btrfs_test_is_dummy_root(root
))
5786 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5787 btrfs_ino(inode
), to_free
, 0);
5789 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5794 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5796 * @inode: inode we're writing to
5797 * @start: start range we are writing to
5798 * @len: how long the range we are writing to
5800 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5802 * This will do the following things
5804 * o reserve space in data space info for num bytes
5805 * and reserve precious corresponding qgroup space
5806 * (Done in check_data_free_space)
5808 * o reserve space for metadata space, based on the number of outstanding
5809 * extents and how much csums will be needed
5810 * also reserve metadata space in a per root over-reserve method.
5811 * o add to the inodes->delalloc_bytes
5812 * o add it to the fs_info's delalloc inodes list.
5813 * (Above 3 all done in delalloc_reserve_metadata)
5815 * Return 0 for success
5816 * Return <0 for error(-ENOSPC or -EQUOT)
5818 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
5822 ret
= btrfs_check_data_free_space(inode
, start
, len
);
5825 ret
= btrfs_delalloc_reserve_metadata(inode
, len
);
5827 btrfs_free_reserved_data_space(inode
, start
, len
);
5832 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5833 * @inode: inode we're releasing space for
5834 * @start: start position of the space already reserved
5835 * @len: the len of the space already reserved
5837 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5838 * called in the case that we don't need the metadata AND data reservations
5839 * anymore. So if there is an error or we insert an inline extent.
5841 * This function will release the metadata space that was not used and will
5842 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5843 * list if there are no delalloc bytes left.
5844 * Also it will handle the qgroup reserved space.
5846 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
5848 btrfs_delalloc_release_metadata(inode
, len
);
5849 btrfs_free_reserved_data_space(inode
, start
, len
);
5852 static int update_block_group(struct btrfs_trans_handle
*trans
,
5853 struct btrfs_root
*root
, u64 bytenr
,
5854 u64 num_bytes
, int alloc
)
5856 struct btrfs_block_group_cache
*cache
= NULL
;
5857 struct btrfs_fs_info
*info
= root
->fs_info
;
5858 u64 total
= num_bytes
;
5863 /* block accounting for super block */
5864 spin_lock(&info
->delalloc_root_lock
);
5865 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5867 old_val
+= num_bytes
;
5869 old_val
-= num_bytes
;
5870 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5871 spin_unlock(&info
->delalloc_root_lock
);
5874 cache
= btrfs_lookup_block_group(info
, bytenr
);
5877 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5878 BTRFS_BLOCK_GROUP_RAID1
|
5879 BTRFS_BLOCK_GROUP_RAID10
))
5884 * If this block group has free space cache written out, we
5885 * need to make sure to load it if we are removing space. This
5886 * is because we need the unpinning stage to actually add the
5887 * space back to the block group, otherwise we will leak space.
5889 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5890 cache_block_group(cache
, 1);
5892 byte_in_group
= bytenr
- cache
->key
.objectid
;
5893 WARN_ON(byte_in_group
> cache
->key
.offset
);
5895 spin_lock(&cache
->space_info
->lock
);
5896 spin_lock(&cache
->lock
);
5898 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5899 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5900 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5902 old_val
= btrfs_block_group_used(&cache
->item
);
5903 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5905 old_val
+= num_bytes
;
5906 btrfs_set_block_group_used(&cache
->item
, old_val
);
5907 cache
->reserved
-= num_bytes
;
5908 cache
->space_info
->bytes_reserved
-= num_bytes
;
5909 cache
->space_info
->bytes_used
+= num_bytes
;
5910 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5911 spin_unlock(&cache
->lock
);
5912 spin_unlock(&cache
->space_info
->lock
);
5914 old_val
-= num_bytes
;
5915 btrfs_set_block_group_used(&cache
->item
, old_val
);
5916 cache
->pinned
+= num_bytes
;
5917 cache
->space_info
->bytes_pinned
+= num_bytes
;
5918 cache
->space_info
->bytes_used
-= num_bytes
;
5919 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5920 spin_unlock(&cache
->lock
);
5921 spin_unlock(&cache
->space_info
->lock
);
5923 set_extent_dirty(info
->pinned_extents
,
5924 bytenr
, bytenr
+ num_bytes
- 1,
5925 GFP_NOFS
| __GFP_NOFAIL
);
5928 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5929 if (list_empty(&cache
->dirty_list
)) {
5930 list_add_tail(&cache
->dirty_list
,
5931 &trans
->transaction
->dirty_bgs
);
5932 trans
->transaction
->num_dirty_bgs
++;
5933 btrfs_get_block_group(cache
);
5935 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5938 * No longer have used bytes in this block group, queue it for
5939 * deletion. We do this after adding the block group to the
5940 * dirty list to avoid races between cleaner kthread and space
5943 if (!alloc
&& old_val
== 0) {
5944 spin_lock(&info
->unused_bgs_lock
);
5945 if (list_empty(&cache
->bg_list
)) {
5946 btrfs_get_block_group(cache
);
5947 list_add_tail(&cache
->bg_list
,
5950 spin_unlock(&info
->unused_bgs_lock
);
5953 btrfs_put_block_group(cache
);
5955 bytenr
+= num_bytes
;
5960 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5962 struct btrfs_block_group_cache
*cache
;
5965 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5966 bytenr
= root
->fs_info
->first_logical_byte
;
5967 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5969 if (bytenr
< (u64
)-1)
5972 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5976 bytenr
= cache
->key
.objectid
;
5977 btrfs_put_block_group(cache
);
5982 static int pin_down_extent(struct btrfs_root
*root
,
5983 struct btrfs_block_group_cache
*cache
,
5984 u64 bytenr
, u64 num_bytes
, int reserved
)
5986 spin_lock(&cache
->space_info
->lock
);
5987 spin_lock(&cache
->lock
);
5988 cache
->pinned
+= num_bytes
;
5989 cache
->space_info
->bytes_pinned
+= num_bytes
;
5991 cache
->reserved
-= num_bytes
;
5992 cache
->space_info
->bytes_reserved
-= num_bytes
;
5994 spin_unlock(&cache
->lock
);
5995 spin_unlock(&cache
->space_info
->lock
);
5997 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5998 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6000 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
6005 * this function must be called within transaction
6007 int btrfs_pin_extent(struct btrfs_root
*root
,
6008 u64 bytenr
, u64 num_bytes
, int reserved
)
6010 struct btrfs_block_group_cache
*cache
;
6012 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6013 BUG_ON(!cache
); /* Logic error */
6015 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
6017 btrfs_put_block_group(cache
);
6022 * this function must be called within transaction
6024 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
6025 u64 bytenr
, u64 num_bytes
)
6027 struct btrfs_block_group_cache
*cache
;
6030 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6035 * pull in the free space cache (if any) so that our pin
6036 * removes the free space from the cache. We have load_only set
6037 * to one because the slow code to read in the free extents does check
6038 * the pinned extents.
6040 cache_block_group(cache
, 1);
6042 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
6044 /* remove us from the free space cache (if we're there at all) */
6045 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6046 btrfs_put_block_group(cache
);
6050 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
6053 struct btrfs_block_group_cache
*block_group
;
6054 struct btrfs_caching_control
*caching_ctl
;
6056 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
6060 cache_block_group(block_group
, 0);
6061 caching_ctl
= get_caching_control(block_group
);
6065 BUG_ON(!block_group_cache_done(block_group
));
6066 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6068 mutex_lock(&caching_ctl
->mutex
);
6070 if (start
>= caching_ctl
->progress
) {
6071 ret
= add_excluded_extent(root
, start
, num_bytes
);
6072 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6073 ret
= btrfs_remove_free_space(block_group
,
6076 num_bytes
= caching_ctl
->progress
- start
;
6077 ret
= btrfs_remove_free_space(block_group
,
6082 num_bytes
= (start
+ num_bytes
) -
6083 caching_ctl
->progress
;
6084 start
= caching_ctl
->progress
;
6085 ret
= add_excluded_extent(root
, start
, num_bytes
);
6088 mutex_unlock(&caching_ctl
->mutex
);
6089 put_caching_control(caching_ctl
);
6091 btrfs_put_block_group(block_group
);
6095 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
6096 struct extent_buffer
*eb
)
6098 struct btrfs_file_extent_item
*item
;
6099 struct btrfs_key key
;
6103 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
6106 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6107 btrfs_item_key_to_cpu(eb
, &key
, i
);
6108 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6110 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6111 found_type
= btrfs_file_extent_type(eb
, item
);
6112 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6114 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6116 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6117 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6118 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
6125 * btrfs_update_reserved_bytes - update the block_group and space info counters
6126 * @cache: The cache we are manipulating
6127 * @num_bytes: The number of bytes in question
6128 * @reserve: One of the reservation enums
6129 * @delalloc: The blocks are allocated for the delalloc write
6131 * This is called by the allocator when it reserves space, or by somebody who is
6132 * freeing space that was never actually used on disk. For example if you
6133 * reserve some space for a new leaf in transaction A and before transaction A
6134 * commits you free that leaf, you call this with reserve set to 0 in order to
6135 * clear the reservation.
6137 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6138 * ENOSPC accounting. For data we handle the reservation through clearing the
6139 * delalloc bits in the io_tree. We have to do this since we could end up
6140 * allocating less disk space for the amount of data we have reserved in the
6141 * case of compression.
6143 * If this is a reservation and the block group has become read only we cannot
6144 * make the reservation and return -EAGAIN, otherwise this function always
6147 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6148 u64 num_bytes
, int reserve
, int delalloc
)
6150 struct btrfs_space_info
*space_info
= cache
->space_info
;
6153 spin_lock(&space_info
->lock
);
6154 spin_lock(&cache
->lock
);
6155 if (reserve
!= RESERVE_FREE
) {
6159 cache
->reserved
+= num_bytes
;
6160 space_info
->bytes_reserved
+= num_bytes
;
6161 if (reserve
== RESERVE_ALLOC
) {
6162 trace_btrfs_space_reservation(cache
->fs_info
,
6163 "space_info", space_info
->flags
,
6165 space_info
->bytes_may_use
-= num_bytes
;
6169 cache
->delalloc_bytes
+= num_bytes
;
6173 space_info
->bytes_readonly
+= num_bytes
;
6174 cache
->reserved
-= num_bytes
;
6175 space_info
->bytes_reserved
-= num_bytes
;
6178 cache
->delalloc_bytes
-= num_bytes
;
6180 spin_unlock(&cache
->lock
);
6181 spin_unlock(&space_info
->lock
);
6185 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
6186 struct btrfs_root
*root
)
6188 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6189 struct btrfs_caching_control
*next
;
6190 struct btrfs_caching_control
*caching_ctl
;
6191 struct btrfs_block_group_cache
*cache
;
6193 down_write(&fs_info
->commit_root_sem
);
6195 list_for_each_entry_safe(caching_ctl
, next
,
6196 &fs_info
->caching_block_groups
, list
) {
6197 cache
= caching_ctl
->block_group
;
6198 if (block_group_cache_done(cache
)) {
6199 cache
->last_byte_to_unpin
= (u64
)-1;
6200 list_del_init(&caching_ctl
->list
);
6201 put_caching_control(caching_ctl
);
6203 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6207 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6208 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6210 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6212 up_write(&fs_info
->commit_root_sem
);
6214 update_global_block_rsv(fs_info
);
6218 * Returns the free cluster for the given space info and sets empty_cluster to
6219 * what it should be based on the mount options.
6221 static struct btrfs_free_cluster
*
6222 fetch_cluster_info(struct btrfs_root
*root
, struct btrfs_space_info
*space_info
,
6225 struct btrfs_free_cluster
*ret
= NULL
;
6226 bool ssd
= btrfs_test_opt(root
, SSD
);
6229 if (btrfs_mixed_space_info(space_info
))
6233 *empty_cluster
= 2 * 1024 * 1024;
6234 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6235 ret
= &root
->fs_info
->meta_alloc_cluster
;
6237 *empty_cluster
= 64 * 1024;
6238 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6239 ret
= &root
->fs_info
->data_alloc_cluster
;
6245 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
6246 const bool return_free_space
)
6248 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6249 struct btrfs_block_group_cache
*cache
= NULL
;
6250 struct btrfs_space_info
*space_info
;
6251 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6252 struct btrfs_free_cluster
*cluster
= NULL
;
6254 u64 total_unpinned
= 0;
6255 u64 empty_cluster
= 0;
6258 while (start
<= end
) {
6261 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6263 btrfs_put_block_group(cache
);
6265 cache
= btrfs_lookup_block_group(fs_info
, start
);
6266 BUG_ON(!cache
); /* Logic error */
6268 cluster
= fetch_cluster_info(root
,
6271 empty_cluster
<<= 1;
6274 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6275 len
= min(len
, end
+ 1 - start
);
6277 if (start
< cache
->last_byte_to_unpin
) {
6278 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6279 if (return_free_space
)
6280 btrfs_add_free_space(cache
, start
, len
);
6284 total_unpinned
+= len
;
6285 space_info
= cache
->space_info
;
6288 * If this space cluster has been marked as fragmented and we've
6289 * unpinned enough in this block group to potentially allow a
6290 * cluster to be created inside of it go ahead and clear the
6293 if (cluster
&& cluster
->fragmented
&&
6294 total_unpinned
> empty_cluster
) {
6295 spin_lock(&cluster
->lock
);
6296 cluster
->fragmented
= 0;
6297 spin_unlock(&cluster
->lock
);
6300 spin_lock(&space_info
->lock
);
6301 spin_lock(&cache
->lock
);
6302 cache
->pinned
-= len
;
6303 space_info
->bytes_pinned
-= len
;
6304 space_info
->max_extent_size
= 0;
6305 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6307 space_info
->bytes_readonly
+= len
;
6310 spin_unlock(&cache
->lock
);
6311 if (!readonly
&& global_rsv
->space_info
== space_info
) {
6312 spin_lock(&global_rsv
->lock
);
6313 if (!global_rsv
->full
) {
6314 len
= min(len
, global_rsv
->size
-
6315 global_rsv
->reserved
);
6316 global_rsv
->reserved
+= len
;
6317 space_info
->bytes_may_use
+= len
;
6318 if (global_rsv
->reserved
>= global_rsv
->size
)
6319 global_rsv
->full
= 1;
6321 spin_unlock(&global_rsv
->lock
);
6323 spin_unlock(&space_info
->lock
);
6327 btrfs_put_block_group(cache
);
6331 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6332 struct btrfs_root
*root
)
6334 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6335 struct btrfs_block_group_cache
*block_group
, *tmp
;
6336 struct list_head
*deleted_bgs
;
6337 struct extent_io_tree
*unpin
;
6342 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6343 unpin
= &fs_info
->freed_extents
[1];
6345 unpin
= &fs_info
->freed_extents
[0];
6347 while (!trans
->aborted
) {
6348 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6349 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6350 EXTENT_DIRTY
, NULL
);
6352 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6356 if (btrfs_test_opt(root
, DISCARD
))
6357 ret
= btrfs_discard_extent(root
, start
,
6358 end
+ 1 - start
, NULL
);
6360 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
6361 unpin_extent_range(root
, start
, end
, true);
6362 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6367 * Transaction is finished. We don't need the lock anymore. We
6368 * do need to clean up the block groups in case of a transaction
6371 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6372 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6376 if (!trans
->aborted
)
6377 ret
= btrfs_discard_extent(root
,
6378 block_group
->key
.objectid
,
6379 block_group
->key
.offset
,
6382 list_del_init(&block_group
->bg_list
);
6383 btrfs_put_block_group_trimming(block_group
);
6384 btrfs_put_block_group(block_group
);
6387 const char *errstr
= btrfs_decode_error(ret
);
6389 "Discard failed while removing blockgroup: errno=%d %s\n",
6397 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6398 u64 owner
, u64 root_objectid
)
6400 struct btrfs_space_info
*space_info
;
6403 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6404 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6405 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6407 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6409 flags
= BTRFS_BLOCK_GROUP_DATA
;
6412 space_info
= __find_space_info(fs_info
, flags
);
6413 BUG_ON(!space_info
); /* Logic bug */
6414 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6418 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6419 struct btrfs_root
*root
,
6420 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6421 u64 root_objectid
, u64 owner_objectid
,
6422 u64 owner_offset
, int refs_to_drop
,
6423 struct btrfs_delayed_extent_op
*extent_op
)
6425 struct btrfs_key key
;
6426 struct btrfs_path
*path
;
6427 struct btrfs_fs_info
*info
= root
->fs_info
;
6428 struct btrfs_root
*extent_root
= info
->extent_root
;
6429 struct extent_buffer
*leaf
;
6430 struct btrfs_extent_item
*ei
;
6431 struct btrfs_extent_inline_ref
*iref
;
6434 int extent_slot
= 0;
6435 int found_extent
= 0;
6439 u64 bytenr
= node
->bytenr
;
6440 u64 num_bytes
= node
->num_bytes
;
6442 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6445 path
= btrfs_alloc_path();
6450 path
->leave_spinning
= 1;
6452 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6453 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6456 skinny_metadata
= 0;
6458 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6459 bytenr
, num_bytes
, parent
,
6460 root_objectid
, owner_objectid
,
6463 extent_slot
= path
->slots
[0];
6464 while (extent_slot
>= 0) {
6465 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6467 if (key
.objectid
!= bytenr
)
6469 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6470 key
.offset
== num_bytes
) {
6474 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6475 key
.offset
== owner_objectid
) {
6479 if (path
->slots
[0] - extent_slot
> 5)
6483 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6484 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6485 if (found_extent
&& item_size
< sizeof(*ei
))
6488 if (!found_extent
) {
6490 ret
= remove_extent_backref(trans
, extent_root
, path
,
6492 is_data
, &last_ref
);
6494 btrfs_abort_transaction(trans
, extent_root
, ret
);
6497 btrfs_release_path(path
);
6498 path
->leave_spinning
= 1;
6500 key
.objectid
= bytenr
;
6501 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6502 key
.offset
= num_bytes
;
6504 if (!is_data
&& skinny_metadata
) {
6505 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6506 key
.offset
= owner_objectid
;
6509 ret
= btrfs_search_slot(trans
, extent_root
,
6511 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6513 * Couldn't find our skinny metadata item,
6514 * see if we have ye olde extent item.
6517 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6519 if (key
.objectid
== bytenr
&&
6520 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6521 key
.offset
== num_bytes
)
6525 if (ret
> 0 && skinny_metadata
) {
6526 skinny_metadata
= false;
6527 key
.objectid
= bytenr
;
6528 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6529 key
.offset
= num_bytes
;
6530 btrfs_release_path(path
);
6531 ret
= btrfs_search_slot(trans
, extent_root
,
6536 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6539 btrfs_print_leaf(extent_root
,
6543 btrfs_abort_transaction(trans
, extent_root
, ret
);
6546 extent_slot
= path
->slots
[0];
6548 } else if (WARN_ON(ret
== -ENOENT
)) {
6549 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6551 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6552 bytenr
, parent
, root_objectid
, owner_objectid
,
6554 btrfs_abort_transaction(trans
, extent_root
, ret
);
6557 btrfs_abort_transaction(trans
, extent_root
, ret
);
6561 leaf
= path
->nodes
[0];
6562 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6563 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6564 if (item_size
< sizeof(*ei
)) {
6565 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6566 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6569 btrfs_abort_transaction(trans
, extent_root
, ret
);
6573 btrfs_release_path(path
);
6574 path
->leave_spinning
= 1;
6576 key
.objectid
= bytenr
;
6577 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6578 key
.offset
= num_bytes
;
6580 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6583 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6585 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6588 btrfs_abort_transaction(trans
, extent_root
, ret
);
6592 extent_slot
= path
->slots
[0];
6593 leaf
= path
->nodes
[0];
6594 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6597 BUG_ON(item_size
< sizeof(*ei
));
6598 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6599 struct btrfs_extent_item
);
6600 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6601 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6602 struct btrfs_tree_block_info
*bi
;
6603 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6604 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6605 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6608 refs
= btrfs_extent_refs(leaf
, ei
);
6609 if (refs
< refs_to_drop
) {
6610 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6611 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6613 btrfs_abort_transaction(trans
, extent_root
, ret
);
6616 refs
-= refs_to_drop
;
6620 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6622 * In the case of inline back ref, reference count will
6623 * be updated by remove_extent_backref
6626 BUG_ON(!found_extent
);
6628 btrfs_set_extent_refs(leaf
, ei
, refs
);
6629 btrfs_mark_buffer_dirty(leaf
);
6632 ret
= remove_extent_backref(trans
, extent_root
, path
,
6634 is_data
, &last_ref
);
6636 btrfs_abort_transaction(trans
, extent_root
, ret
);
6640 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6644 BUG_ON(is_data
&& refs_to_drop
!=
6645 extent_data_ref_count(path
, iref
));
6647 BUG_ON(path
->slots
[0] != extent_slot
);
6649 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6650 path
->slots
[0] = extent_slot
;
6656 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6659 btrfs_abort_transaction(trans
, extent_root
, ret
);
6662 btrfs_release_path(path
);
6665 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6667 btrfs_abort_transaction(trans
, extent_root
, ret
);
6672 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6674 btrfs_abort_transaction(trans
, extent_root
, ret
);
6678 btrfs_release_path(path
);
6681 btrfs_free_path(path
);
6686 * when we free an block, it is possible (and likely) that we free the last
6687 * delayed ref for that extent as well. This searches the delayed ref tree for
6688 * a given extent, and if there are no other delayed refs to be processed, it
6689 * removes it from the tree.
6691 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6692 struct btrfs_root
*root
, u64 bytenr
)
6694 struct btrfs_delayed_ref_head
*head
;
6695 struct btrfs_delayed_ref_root
*delayed_refs
;
6698 delayed_refs
= &trans
->transaction
->delayed_refs
;
6699 spin_lock(&delayed_refs
->lock
);
6700 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6702 goto out_delayed_unlock
;
6704 spin_lock(&head
->lock
);
6705 if (!list_empty(&head
->ref_list
))
6708 if (head
->extent_op
) {
6709 if (!head
->must_insert_reserved
)
6711 btrfs_free_delayed_extent_op(head
->extent_op
);
6712 head
->extent_op
= NULL
;
6716 * waiting for the lock here would deadlock. If someone else has it
6717 * locked they are already in the process of dropping it anyway
6719 if (!mutex_trylock(&head
->mutex
))
6723 * at this point we have a head with no other entries. Go
6724 * ahead and process it.
6726 head
->node
.in_tree
= 0;
6727 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6729 atomic_dec(&delayed_refs
->num_entries
);
6732 * we don't take a ref on the node because we're removing it from the
6733 * tree, so we just steal the ref the tree was holding.
6735 delayed_refs
->num_heads
--;
6736 if (head
->processing
== 0)
6737 delayed_refs
->num_heads_ready
--;
6738 head
->processing
= 0;
6739 spin_unlock(&head
->lock
);
6740 spin_unlock(&delayed_refs
->lock
);
6742 BUG_ON(head
->extent_op
);
6743 if (head
->must_insert_reserved
)
6746 mutex_unlock(&head
->mutex
);
6747 btrfs_put_delayed_ref(&head
->node
);
6750 spin_unlock(&head
->lock
);
6753 spin_unlock(&delayed_refs
->lock
);
6757 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6758 struct btrfs_root
*root
,
6759 struct extent_buffer
*buf
,
6760 u64 parent
, int last_ref
)
6765 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6766 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6767 buf
->start
, buf
->len
,
6768 parent
, root
->root_key
.objectid
,
6769 btrfs_header_level(buf
),
6770 BTRFS_DROP_DELAYED_REF
, NULL
);
6771 BUG_ON(ret
); /* -ENOMEM */
6777 if (btrfs_header_generation(buf
) == trans
->transid
) {
6778 struct btrfs_block_group_cache
*cache
;
6780 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6781 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6786 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6788 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6789 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6790 btrfs_put_block_group(cache
);
6794 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6796 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6797 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6798 btrfs_put_block_group(cache
);
6799 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6804 add_pinned_bytes(root
->fs_info
, buf
->len
,
6805 btrfs_header_level(buf
),
6806 root
->root_key
.objectid
);
6809 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6812 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6815 /* Can return -ENOMEM */
6816 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6817 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6818 u64 owner
, u64 offset
)
6821 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6823 if (btrfs_test_is_dummy_root(root
))
6826 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6829 * tree log blocks never actually go into the extent allocation
6830 * tree, just update pinning info and exit early.
6832 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6833 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6834 /* unlocks the pinned mutex */
6835 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6837 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6838 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6840 parent
, root_objectid
, (int)owner
,
6841 BTRFS_DROP_DELAYED_REF
, NULL
);
6843 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6845 parent
, root_objectid
, owner
,
6847 BTRFS_DROP_DELAYED_REF
, NULL
);
6853 * when we wait for progress in the block group caching, its because
6854 * our allocation attempt failed at least once. So, we must sleep
6855 * and let some progress happen before we try again.
6857 * This function will sleep at least once waiting for new free space to
6858 * show up, and then it will check the block group free space numbers
6859 * for our min num_bytes. Another option is to have it go ahead
6860 * and look in the rbtree for a free extent of a given size, but this
6863 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6864 * any of the information in this block group.
6866 static noinline
void
6867 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6870 struct btrfs_caching_control
*caching_ctl
;
6872 caching_ctl
= get_caching_control(cache
);
6876 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6877 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6879 put_caching_control(caching_ctl
);
6883 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6885 struct btrfs_caching_control
*caching_ctl
;
6888 caching_ctl
= get_caching_control(cache
);
6890 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6892 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6893 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6895 put_caching_control(caching_ctl
);
6899 int __get_raid_index(u64 flags
)
6901 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6902 return BTRFS_RAID_RAID10
;
6903 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6904 return BTRFS_RAID_RAID1
;
6905 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6906 return BTRFS_RAID_DUP
;
6907 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6908 return BTRFS_RAID_RAID0
;
6909 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6910 return BTRFS_RAID_RAID5
;
6911 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6912 return BTRFS_RAID_RAID6
;
6914 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6917 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6919 return __get_raid_index(cache
->flags
);
6922 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6923 [BTRFS_RAID_RAID10
] = "raid10",
6924 [BTRFS_RAID_RAID1
] = "raid1",
6925 [BTRFS_RAID_DUP
] = "dup",
6926 [BTRFS_RAID_RAID0
] = "raid0",
6927 [BTRFS_RAID_SINGLE
] = "single",
6928 [BTRFS_RAID_RAID5
] = "raid5",
6929 [BTRFS_RAID_RAID6
] = "raid6",
6932 static const char *get_raid_name(enum btrfs_raid_types type
)
6934 if (type
>= BTRFS_NR_RAID_TYPES
)
6937 return btrfs_raid_type_names
[type
];
6940 enum btrfs_loop_type
{
6941 LOOP_CACHING_NOWAIT
= 0,
6942 LOOP_CACHING_WAIT
= 1,
6943 LOOP_ALLOC_CHUNK
= 2,
6944 LOOP_NO_EMPTY_SIZE
= 3,
6948 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6952 down_read(&cache
->data_rwsem
);
6956 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6959 btrfs_get_block_group(cache
);
6961 down_read(&cache
->data_rwsem
);
6964 static struct btrfs_block_group_cache
*
6965 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6966 struct btrfs_free_cluster
*cluster
,
6969 struct btrfs_block_group_cache
*used_bg
;
6970 bool locked
= false;
6972 spin_lock(&cluster
->refill_lock
);
6974 if (used_bg
== cluster
->block_group
)
6977 up_read(&used_bg
->data_rwsem
);
6978 btrfs_put_block_group(used_bg
);
6981 used_bg
= cluster
->block_group
;
6985 if (used_bg
== block_group
)
6988 btrfs_get_block_group(used_bg
);
6993 if (down_read_trylock(&used_bg
->data_rwsem
))
6996 spin_unlock(&cluster
->refill_lock
);
6997 down_read(&used_bg
->data_rwsem
);
7003 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7007 up_read(&cache
->data_rwsem
);
7008 btrfs_put_block_group(cache
);
7012 * walks the btree of allocated extents and find a hole of a given size.
7013 * The key ins is changed to record the hole:
7014 * ins->objectid == start position
7015 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7016 * ins->offset == the size of the hole.
7017 * Any available blocks before search_start are skipped.
7019 * If there is no suitable free space, we will record the max size of
7020 * the free space extent currently.
7022 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
7023 u64 num_bytes
, u64 empty_size
,
7024 u64 hint_byte
, struct btrfs_key
*ins
,
7025 u64 flags
, int delalloc
)
7028 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
7029 struct btrfs_free_cluster
*last_ptr
= NULL
;
7030 struct btrfs_block_group_cache
*block_group
= NULL
;
7031 u64 search_start
= 0;
7032 u64 max_extent_size
= 0;
7033 u64 empty_cluster
= 0;
7034 struct btrfs_space_info
*space_info
;
7036 int index
= __get_raid_index(flags
);
7037 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
7038 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
7039 bool failed_cluster_refill
= false;
7040 bool failed_alloc
= false;
7041 bool use_cluster
= true;
7042 bool have_caching_bg
= false;
7043 bool orig_have_caching_bg
= false;
7044 bool full_search
= false;
7046 WARN_ON(num_bytes
< root
->sectorsize
);
7047 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7051 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
7053 space_info
= __find_space_info(root
->fs_info
, flags
);
7055 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
7060 * If our free space is heavily fragmented we may not be able to make
7061 * big contiguous allocations, so instead of doing the expensive search
7062 * for free space, simply return ENOSPC with our max_extent_size so we
7063 * can go ahead and search for a more manageable chunk.
7065 * If our max_extent_size is large enough for our allocation simply
7066 * disable clustering since we will likely not be able to find enough
7067 * space to create a cluster and induce latency trying.
7069 if (unlikely(space_info
->max_extent_size
)) {
7070 spin_lock(&space_info
->lock
);
7071 if (space_info
->max_extent_size
&&
7072 num_bytes
> space_info
->max_extent_size
) {
7073 ins
->offset
= space_info
->max_extent_size
;
7074 spin_unlock(&space_info
->lock
);
7076 } else if (space_info
->max_extent_size
) {
7077 use_cluster
= false;
7079 spin_unlock(&space_info
->lock
);
7082 last_ptr
= fetch_cluster_info(orig_root
, space_info
, &empty_cluster
);
7084 spin_lock(&last_ptr
->lock
);
7085 if (last_ptr
->block_group
)
7086 hint_byte
= last_ptr
->window_start
;
7087 if (last_ptr
->fragmented
) {
7089 * We still set window_start so we can keep track of the
7090 * last place we found an allocation to try and save
7093 hint_byte
= last_ptr
->window_start
;
7094 use_cluster
= false;
7096 spin_unlock(&last_ptr
->lock
);
7099 search_start
= max(search_start
, first_logical_byte(root
, 0));
7100 search_start
= max(search_start
, hint_byte
);
7101 if (search_start
== hint_byte
) {
7102 block_group
= btrfs_lookup_block_group(root
->fs_info
,
7105 * we don't want to use the block group if it doesn't match our
7106 * allocation bits, or if its not cached.
7108 * However if we are re-searching with an ideal block group
7109 * picked out then we don't care that the block group is cached.
7111 if (block_group
&& block_group_bits(block_group
, flags
) &&
7112 block_group
->cached
!= BTRFS_CACHE_NO
) {
7113 down_read(&space_info
->groups_sem
);
7114 if (list_empty(&block_group
->list
) ||
7117 * someone is removing this block group,
7118 * we can't jump into the have_block_group
7119 * target because our list pointers are not
7122 btrfs_put_block_group(block_group
);
7123 up_read(&space_info
->groups_sem
);
7125 index
= get_block_group_index(block_group
);
7126 btrfs_lock_block_group(block_group
, delalloc
);
7127 goto have_block_group
;
7129 } else if (block_group
) {
7130 btrfs_put_block_group(block_group
);
7134 have_caching_bg
= false;
7135 if (index
== 0 || index
== __get_raid_index(flags
))
7137 down_read(&space_info
->groups_sem
);
7138 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7143 btrfs_grab_block_group(block_group
, delalloc
);
7144 search_start
= block_group
->key
.objectid
;
7147 * this can happen if we end up cycling through all the
7148 * raid types, but we want to make sure we only allocate
7149 * for the proper type.
7151 if (!block_group_bits(block_group
, flags
)) {
7152 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7153 BTRFS_BLOCK_GROUP_RAID1
|
7154 BTRFS_BLOCK_GROUP_RAID5
|
7155 BTRFS_BLOCK_GROUP_RAID6
|
7156 BTRFS_BLOCK_GROUP_RAID10
;
7159 * if they asked for extra copies and this block group
7160 * doesn't provide them, bail. This does allow us to
7161 * fill raid0 from raid1.
7163 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7168 cached
= block_group_cache_done(block_group
);
7169 if (unlikely(!cached
)) {
7170 have_caching_bg
= true;
7171 ret
= cache_block_group(block_group
, 0);
7176 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7178 if (unlikely(block_group
->ro
))
7182 * Ok we want to try and use the cluster allocator, so
7185 if (last_ptr
&& use_cluster
) {
7186 struct btrfs_block_group_cache
*used_block_group
;
7187 unsigned long aligned_cluster
;
7189 * the refill lock keeps out other
7190 * people trying to start a new cluster
7192 used_block_group
= btrfs_lock_cluster(block_group
,
7195 if (!used_block_group
)
7196 goto refill_cluster
;
7198 if (used_block_group
!= block_group
&&
7199 (used_block_group
->ro
||
7200 !block_group_bits(used_block_group
, flags
)))
7201 goto release_cluster
;
7203 offset
= btrfs_alloc_from_cluster(used_block_group
,
7206 used_block_group
->key
.objectid
,
7209 /* we have a block, we're done */
7210 spin_unlock(&last_ptr
->refill_lock
);
7211 trace_btrfs_reserve_extent_cluster(root
,
7213 search_start
, num_bytes
);
7214 if (used_block_group
!= block_group
) {
7215 btrfs_release_block_group(block_group
,
7217 block_group
= used_block_group
;
7222 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7224 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7225 * set up a new clusters, so lets just skip it
7226 * and let the allocator find whatever block
7227 * it can find. If we reach this point, we
7228 * will have tried the cluster allocator
7229 * plenty of times and not have found
7230 * anything, so we are likely way too
7231 * fragmented for the clustering stuff to find
7234 * However, if the cluster is taken from the
7235 * current block group, release the cluster
7236 * first, so that we stand a better chance of
7237 * succeeding in the unclustered
7239 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7240 used_block_group
!= block_group
) {
7241 spin_unlock(&last_ptr
->refill_lock
);
7242 btrfs_release_block_group(used_block_group
,
7244 goto unclustered_alloc
;
7248 * this cluster didn't work out, free it and
7251 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7253 if (used_block_group
!= block_group
)
7254 btrfs_release_block_group(used_block_group
,
7257 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7258 spin_unlock(&last_ptr
->refill_lock
);
7259 goto unclustered_alloc
;
7262 aligned_cluster
= max_t(unsigned long,
7263 empty_cluster
+ empty_size
,
7264 block_group
->full_stripe_len
);
7266 /* allocate a cluster in this block group */
7267 ret
= btrfs_find_space_cluster(root
, block_group
,
7268 last_ptr
, search_start
,
7273 * now pull our allocation out of this
7276 offset
= btrfs_alloc_from_cluster(block_group
,
7282 /* we found one, proceed */
7283 spin_unlock(&last_ptr
->refill_lock
);
7284 trace_btrfs_reserve_extent_cluster(root
,
7285 block_group
, search_start
,
7289 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7290 && !failed_cluster_refill
) {
7291 spin_unlock(&last_ptr
->refill_lock
);
7293 failed_cluster_refill
= true;
7294 wait_block_group_cache_progress(block_group
,
7295 num_bytes
+ empty_cluster
+ empty_size
);
7296 goto have_block_group
;
7300 * at this point we either didn't find a cluster
7301 * or we weren't able to allocate a block from our
7302 * cluster. Free the cluster we've been trying
7303 * to use, and go to the next block group
7305 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7306 spin_unlock(&last_ptr
->refill_lock
);
7312 * We are doing an unclustered alloc, set the fragmented flag so
7313 * we don't bother trying to setup a cluster again until we get
7316 if (unlikely(last_ptr
)) {
7317 spin_lock(&last_ptr
->lock
);
7318 last_ptr
->fragmented
= 1;
7319 spin_unlock(&last_ptr
->lock
);
7321 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7323 block_group
->free_space_ctl
->free_space
<
7324 num_bytes
+ empty_cluster
+ empty_size
) {
7325 if (block_group
->free_space_ctl
->free_space
>
7328 block_group
->free_space_ctl
->free_space
;
7329 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7332 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7334 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7335 num_bytes
, empty_size
,
7338 * If we didn't find a chunk, and we haven't failed on this
7339 * block group before, and this block group is in the middle of
7340 * caching and we are ok with waiting, then go ahead and wait
7341 * for progress to be made, and set failed_alloc to true.
7343 * If failed_alloc is true then we've already waited on this
7344 * block group once and should move on to the next block group.
7346 if (!offset
&& !failed_alloc
&& !cached
&&
7347 loop
> LOOP_CACHING_NOWAIT
) {
7348 wait_block_group_cache_progress(block_group
,
7349 num_bytes
+ empty_size
);
7350 failed_alloc
= true;
7351 goto have_block_group
;
7352 } else if (!offset
) {
7356 search_start
= ALIGN(offset
, root
->stripesize
);
7358 /* move on to the next group */
7359 if (search_start
+ num_bytes
>
7360 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7361 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7365 if (offset
< search_start
)
7366 btrfs_add_free_space(block_group
, offset
,
7367 search_start
- offset
);
7368 BUG_ON(offset
> search_start
);
7370 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7371 alloc_type
, delalloc
);
7372 if (ret
== -EAGAIN
) {
7373 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7377 /* we are all good, lets return */
7378 ins
->objectid
= search_start
;
7379 ins
->offset
= num_bytes
;
7381 trace_btrfs_reserve_extent(orig_root
, block_group
,
7382 search_start
, num_bytes
);
7383 btrfs_release_block_group(block_group
, delalloc
);
7386 failed_cluster_refill
= false;
7387 failed_alloc
= false;
7388 BUG_ON(index
!= get_block_group_index(block_group
));
7389 btrfs_release_block_group(block_group
, delalloc
);
7391 up_read(&space_info
->groups_sem
);
7393 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7394 && !orig_have_caching_bg
)
7395 orig_have_caching_bg
= true;
7397 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7400 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7404 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7405 * caching kthreads as we move along
7406 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7407 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7408 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7411 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7413 if (loop
== LOOP_CACHING_NOWAIT
) {
7415 * We want to skip the LOOP_CACHING_WAIT step if we
7416 * don't have any unached bgs and we've alrelady done a
7417 * full search through.
7419 if (orig_have_caching_bg
|| !full_search
)
7420 loop
= LOOP_CACHING_WAIT
;
7422 loop
= LOOP_ALLOC_CHUNK
;
7427 if (loop
== LOOP_ALLOC_CHUNK
) {
7428 struct btrfs_trans_handle
*trans
;
7431 trans
= current
->journal_info
;
7435 trans
= btrfs_join_transaction(root
);
7437 if (IS_ERR(trans
)) {
7438 ret
= PTR_ERR(trans
);
7442 ret
= do_chunk_alloc(trans
, root
, flags
,
7446 * If we can't allocate a new chunk we've already looped
7447 * through at least once, move on to the NO_EMPTY_SIZE
7451 loop
= LOOP_NO_EMPTY_SIZE
;
7454 * Do not bail out on ENOSPC since we
7455 * can do more things.
7457 if (ret
< 0 && ret
!= -ENOSPC
)
7458 btrfs_abort_transaction(trans
,
7463 btrfs_end_transaction(trans
, root
);
7468 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7470 * Don't loop again if we already have no empty_size and
7473 if (empty_size
== 0 &&
7474 empty_cluster
== 0) {
7483 } else if (!ins
->objectid
) {
7485 } else if (ins
->objectid
) {
7486 if (!use_cluster
&& last_ptr
) {
7487 spin_lock(&last_ptr
->lock
);
7488 last_ptr
->window_start
= ins
->objectid
;
7489 spin_unlock(&last_ptr
->lock
);
7494 if (ret
== -ENOSPC
) {
7495 spin_lock(&space_info
->lock
);
7496 space_info
->max_extent_size
= max_extent_size
;
7497 spin_unlock(&space_info
->lock
);
7498 ins
->offset
= max_extent_size
;
7503 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7504 int dump_block_groups
)
7506 struct btrfs_block_group_cache
*cache
;
7509 spin_lock(&info
->lock
);
7510 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7512 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7513 info
->bytes_reserved
- info
->bytes_readonly
,
7514 (info
->full
) ? "" : "not ");
7515 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7516 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7517 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7518 info
->bytes_reserved
, info
->bytes_may_use
,
7519 info
->bytes_readonly
);
7520 spin_unlock(&info
->lock
);
7522 if (!dump_block_groups
)
7525 down_read(&info
->groups_sem
);
7527 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7528 spin_lock(&cache
->lock
);
7529 printk(KERN_INFO
"BTRFS: "
7530 "block group %llu has %llu bytes, "
7531 "%llu used %llu pinned %llu reserved %s\n",
7532 cache
->key
.objectid
, cache
->key
.offset
,
7533 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7534 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7535 btrfs_dump_free_space(cache
, bytes
);
7536 spin_unlock(&cache
->lock
);
7538 if (++index
< BTRFS_NR_RAID_TYPES
)
7540 up_read(&info
->groups_sem
);
7543 int btrfs_reserve_extent(struct btrfs_root
*root
,
7544 u64 num_bytes
, u64 min_alloc_size
,
7545 u64 empty_size
, u64 hint_byte
,
7546 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7548 bool final_tried
= num_bytes
== min_alloc_size
;
7552 flags
= btrfs_get_alloc_profile(root
, is_data
);
7554 WARN_ON(num_bytes
< root
->sectorsize
);
7555 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7558 if (ret
== -ENOSPC
) {
7559 if (!final_tried
&& ins
->offset
) {
7560 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7561 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7562 num_bytes
= max(num_bytes
, min_alloc_size
);
7563 if (num_bytes
== min_alloc_size
)
7566 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7567 struct btrfs_space_info
*sinfo
;
7569 sinfo
= __find_space_info(root
->fs_info
, flags
);
7570 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7573 dump_space_info(sinfo
, num_bytes
, 1);
7580 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7582 int pin
, int delalloc
)
7584 struct btrfs_block_group_cache
*cache
;
7587 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7589 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7595 pin_down_extent(root
, cache
, start
, len
, 1);
7597 if (btrfs_test_opt(root
, DISCARD
))
7598 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7599 btrfs_add_free_space(cache
, start
, len
);
7600 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7603 btrfs_put_block_group(cache
);
7605 trace_btrfs_reserved_extent_free(root
, start
, len
);
7610 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7611 u64 start
, u64 len
, int delalloc
)
7613 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7616 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7619 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7622 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7623 struct btrfs_root
*root
,
7624 u64 parent
, u64 root_objectid
,
7625 u64 flags
, u64 owner
, u64 offset
,
7626 struct btrfs_key
*ins
, int ref_mod
)
7629 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7630 struct btrfs_extent_item
*extent_item
;
7631 struct btrfs_extent_inline_ref
*iref
;
7632 struct btrfs_path
*path
;
7633 struct extent_buffer
*leaf
;
7638 type
= BTRFS_SHARED_DATA_REF_KEY
;
7640 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7642 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7644 path
= btrfs_alloc_path();
7648 path
->leave_spinning
= 1;
7649 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7652 btrfs_free_path(path
);
7656 leaf
= path
->nodes
[0];
7657 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7658 struct btrfs_extent_item
);
7659 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7660 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7661 btrfs_set_extent_flags(leaf
, extent_item
,
7662 flags
| BTRFS_EXTENT_FLAG_DATA
);
7664 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7665 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7667 struct btrfs_shared_data_ref
*ref
;
7668 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7669 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7670 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7672 struct btrfs_extent_data_ref
*ref
;
7673 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7674 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7675 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7676 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7677 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7680 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7681 btrfs_free_path(path
);
7683 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7684 if (ret
) { /* -ENOENT, logic error */
7685 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7686 ins
->objectid
, ins
->offset
);
7689 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7693 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7694 struct btrfs_root
*root
,
7695 u64 parent
, u64 root_objectid
,
7696 u64 flags
, struct btrfs_disk_key
*key
,
7697 int level
, struct btrfs_key
*ins
)
7700 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7701 struct btrfs_extent_item
*extent_item
;
7702 struct btrfs_tree_block_info
*block_info
;
7703 struct btrfs_extent_inline_ref
*iref
;
7704 struct btrfs_path
*path
;
7705 struct extent_buffer
*leaf
;
7706 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7707 u64 num_bytes
= ins
->offset
;
7708 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7711 if (!skinny_metadata
)
7712 size
+= sizeof(*block_info
);
7714 path
= btrfs_alloc_path();
7716 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7721 path
->leave_spinning
= 1;
7722 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7725 btrfs_free_path(path
);
7726 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7731 leaf
= path
->nodes
[0];
7732 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7733 struct btrfs_extent_item
);
7734 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7735 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7736 btrfs_set_extent_flags(leaf
, extent_item
,
7737 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7739 if (skinny_metadata
) {
7740 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7741 num_bytes
= root
->nodesize
;
7743 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7744 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7745 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7746 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7750 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7751 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7752 BTRFS_SHARED_BLOCK_REF_KEY
);
7753 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7755 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7756 BTRFS_TREE_BLOCK_REF_KEY
);
7757 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7760 btrfs_mark_buffer_dirty(leaf
);
7761 btrfs_free_path(path
);
7763 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7765 if (ret
) { /* -ENOENT, logic error */
7766 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7767 ins
->objectid
, ins
->offset
);
7771 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7775 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7776 struct btrfs_root
*root
,
7777 u64 root_objectid
, u64 owner
,
7778 u64 offset
, u64 ram_bytes
,
7779 struct btrfs_key
*ins
)
7783 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7785 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7787 root_objectid
, owner
, offset
,
7788 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
,
7794 * this is used by the tree logging recovery code. It records that
7795 * an extent has been allocated and makes sure to clear the free
7796 * space cache bits as well
7798 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7799 struct btrfs_root
*root
,
7800 u64 root_objectid
, u64 owner
, u64 offset
,
7801 struct btrfs_key
*ins
)
7804 struct btrfs_block_group_cache
*block_group
;
7807 * Mixed block groups will exclude before processing the log so we only
7808 * need to do the exlude dance if this fs isn't mixed.
7810 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7811 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7816 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7820 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7821 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7822 BUG_ON(ret
); /* logic error */
7823 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7824 0, owner
, offset
, ins
, 1);
7825 btrfs_put_block_group(block_group
);
7829 static struct extent_buffer
*
7830 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7831 u64 bytenr
, int level
)
7833 struct extent_buffer
*buf
;
7835 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7837 return ERR_PTR(-ENOMEM
);
7838 btrfs_set_header_generation(buf
, trans
->transid
);
7839 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7840 btrfs_tree_lock(buf
);
7841 clean_tree_block(trans
, root
->fs_info
, buf
);
7842 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7844 btrfs_set_lock_blocking(buf
);
7845 btrfs_set_buffer_uptodate(buf
);
7847 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7848 buf
->log_index
= root
->log_transid
% 2;
7850 * we allow two log transactions at a time, use different
7851 * EXENT bit to differentiate dirty pages.
7853 if (buf
->log_index
== 0)
7854 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7855 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7857 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7858 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7860 buf
->log_index
= -1;
7861 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7862 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7864 trans
->dirty
= true;
7865 /* this returns a buffer locked for blocking */
7869 static struct btrfs_block_rsv
*
7870 use_block_rsv(struct btrfs_trans_handle
*trans
,
7871 struct btrfs_root
*root
, u32 blocksize
)
7873 struct btrfs_block_rsv
*block_rsv
;
7874 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7876 bool global_updated
= false;
7878 block_rsv
= get_block_rsv(trans
, root
);
7880 if (unlikely(block_rsv
->size
== 0))
7883 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7887 if (block_rsv
->failfast
)
7888 return ERR_PTR(ret
);
7890 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7891 global_updated
= true;
7892 update_global_block_rsv(root
->fs_info
);
7896 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7897 static DEFINE_RATELIMIT_STATE(_rs
,
7898 DEFAULT_RATELIMIT_INTERVAL
* 10,
7899 /*DEFAULT_RATELIMIT_BURST*/ 1);
7900 if (__ratelimit(&_rs
))
7902 "BTRFS: block rsv returned %d\n", ret
);
7905 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7906 BTRFS_RESERVE_NO_FLUSH
);
7910 * If we couldn't reserve metadata bytes try and use some from
7911 * the global reserve if its space type is the same as the global
7914 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7915 block_rsv
->space_info
== global_rsv
->space_info
) {
7916 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7920 return ERR_PTR(ret
);
7923 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7924 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7926 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7927 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7931 * finds a free extent and does all the dirty work required for allocation
7932 * returns the tree buffer or an ERR_PTR on error.
7934 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7935 struct btrfs_root
*root
,
7936 u64 parent
, u64 root_objectid
,
7937 struct btrfs_disk_key
*key
, int level
,
7938 u64 hint
, u64 empty_size
)
7940 struct btrfs_key ins
;
7941 struct btrfs_block_rsv
*block_rsv
;
7942 struct extent_buffer
*buf
;
7943 struct btrfs_delayed_extent_op
*extent_op
;
7946 u32 blocksize
= root
->nodesize
;
7947 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7950 if (btrfs_test_is_dummy_root(root
)) {
7951 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7954 root
->alloc_bytenr
+= blocksize
;
7958 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7959 if (IS_ERR(block_rsv
))
7960 return ERR_CAST(block_rsv
);
7962 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7963 empty_size
, hint
, &ins
, 0, 0);
7967 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7970 goto out_free_reserved
;
7973 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7975 parent
= ins
.objectid
;
7976 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7980 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7981 extent_op
= btrfs_alloc_delayed_extent_op();
7987 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7989 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7990 extent_op
->flags_to_set
= flags
;
7991 if (skinny_metadata
)
7992 extent_op
->update_key
= 0;
7994 extent_op
->update_key
= 1;
7995 extent_op
->update_flags
= 1;
7996 extent_op
->is_data
= 0;
7997 extent_op
->level
= level
;
7999 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
8000 ins
.objectid
, ins
.offset
,
8001 parent
, root_objectid
, level
,
8002 BTRFS_ADD_DELAYED_EXTENT
,
8005 goto out_free_delayed
;
8010 btrfs_free_delayed_extent_op(extent_op
);
8012 free_extent_buffer(buf
);
8014 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
8016 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
8017 return ERR_PTR(ret
);
8020 struct walk_control
{
8021 u64 refs
[BTRFS_MAX_LEVEL
];
8022 u64 flags
[BTRFS_MAX_LEVEL
];
8023 struct btrfs_key update_progress
;
8034 #define DROP_REFERENCE 1
8035 #define UPDATE_BACKREF 2
8037 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8038 struct btrfs_root
*root
,
8039 struct walk_control
*wc
,
8040 struct btrfs_path
*path
)
8048 struct btrfs_key key
;
8049 struct extent_buffer
*eb
;
8054 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8055 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8056 wc
->reada_count
= max(wc
->reada_count
, 2);
8058 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8059 wc
->reada_count
= min_t(int, wc
->reada_count
,
8060 BTRFS_NODEPTRS_PER_BLOCK(root
));
8063 eb
= path
->nodes
[wc
->level
];
8064 nritems
= btrfs_header_nritems(eb
);
8065 blocksize
= root
->nodesize
;
8067 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8068 if (nread
>= wc
->reada_count
)
8072 bytenr
= btrfs_node_blockptr(eb
, slot
);
8073 generation
= btrfs_node_ptr_generation(eb
, slot
);
8075 if (slot
== path
->slots
[wc
->level
])
8078 if (wc
->stage
== UPDATE_BACKREF
&&
8079 generation
<= root
->root_key
.offset
)
8082 /* We don't lock the tree block, it's OK to be racy here */
8083 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
8084 wc
->level
- 1, 1, &refs
,
8086 /* We don't care about errors in readahead. */
8091 if (wc
->stage
== DROP_REFERENCE
) {
8095 if (wc
->level
== 1 &&
8096 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8098 if (!wc
->update_ref
||
8099 generation
<= root
->root_key
.offset
)
8101 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8102 ret
= btrfs_comp_cpu_keys(&key
,
8103 &wc
->update_progress
);
8107 if (wc
->level
== 1 &&
8108 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8112 readahead_tree_block(root
, bytenr
);
8115 wc
->reada_slot
= slot
;
8119 * These may not be seen by the usual inc/dec ref code so we have to
8122 static int record_one_subtree_extent(struct btrfs_trans_handle
*trans
,
8123 struct btrfs_root
*root
, u64 bytenr
,
8126 struct btrfs_qgroup_extent_record
*qrecord
;
8127 struct btrfs_delayed_ref_root
*delayed_refs
;
8129 qrecord
= kmalloc(sizeof(*qrecord
), GFP_NOFS
);
8133 qrecord
->bytenr
= bytenr
;
8134 qrecord
->num_bytes
= num_bytes
;
8135 qrecord
->old_roots
= NULL
;
8137 delayed_refs
= &trans
->transaction
->delayed_refs
;
8138 spin_lock(&delayed_refs
->lock
);
8139 if (btrfs_qgroup_insert_dirty_extent(delayed_refs
, qrecord
))
8141 spin_unlock(&delayed_refs
->lock
);
8146 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
8147 struct btrfs_root
*root
,
8148 struct extent_buffer
*eb
)
8150 int nr
= btrfs_header_nritems(eb
);
8151 int i
, extent_type
, ret
;
8152 struct btrfs_key key
;
8153 struct btrfs_file_extent_item
*fi
;
8154 u64 bytenr
, num_bytes
;
8156 /* We can be called directly from walk_up_proc() */
8157 if (!root
->fs_info
->quota_enabled
)
8160 for (i
= 0; i
< nr
; i
++) {
8161 btrfs_item_key_to_cpu(eb
, &key
, i
);
8163 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
8166 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
8167 /* filter out non qgroup-accountable extents */
8168 extent_type
= btrfs_file_extent_type(eb
, fi
);
8170 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
8173 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
8177 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
8179 ret
= record_one_subtree_extent(trans
, root
, bytenr
, num_bytes
);
8187 * Walk up the tree from the bottom, freeing leaves and any interior
8188 * nodes which have had all slots visited. If a node (leaf or
8189 * interior) is freed, the node above it will have it's slot
8190 * incremented. The root node will never be freed.
8192 * At the end of this function, we should have a path which has all
8193 * slots incremented to the next position for a search. If we need to
8194 * read a new node it will be NULL and the node above it will have the
8195 * correct slot selected for a later read.
8197 * If we increment the root nodes slot counter past the number of
8198 * elements, 1 is returned to signal completion of the search.
8200 static int adjust_slots_upwards(struct btrfs_root
*root
,
8201 struct btrfs_path
*path
, int root_level
)
8205 struct extent_buffer
*eb
;
8207 if (root_level
== 0)
8210 while (level
<= root_level
) {
8211 eb
= path
->nodes
[level
];
8212 nr
= btrfs_header_nritems(eb
);
8213 path
->slots
[level
]++;
8214 slot
= path
->slots
[level
];
8215 if (slot
>= nr
|| level
== 0) {
8217 * Don't free the root - we will detect this
8218 * condition after our loop and return a
8219 * positive value for caller to stop walking the tree.
8221 if (level
!= root_level
) {
8222 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8223 path
->locks
[level
] = 0;
8225 free_extent_buffer(eb
);
8226 path
->nodes
[level
] = NULL
;
8227 path
->slots
[level
] = 0;
8231 * We have a valid slot to walk back down
8232 * from. Stop here so caller can process these
8241 eb
= path
->nodes
[root_level
];
8242 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
8249 * root_eb is the subtree root and is locked before this function is called.
8251 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
8252 struct btrfs_root
*root
,
8253 struct extent_buffer
*root_eb
,
8259 struct extent_buffer
*eb
= root_eb
;
8260 struct btrfs_path
*path
= NULL
;
8262 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
8263 BUG_ON(root_eb
== NULL
);
8265 if (!root
->fs_info
->quota_enabled
)
8268 if (!extent_buffer_uptodate(root_eb
)) {
8269 ret
= btrfs_read_buffer(root_eb
, root_gen
);
8274 if (root_level
== 0) {
8275 ret
= account_leaf_items(trans
, root
, root_eb
);
8279 path
= btrfs_alloc_path();
8284 * Walk down the tree. Missing extent blocks are filled in as
8285 * we go. Metadata is accounted every time we read a new
8288 * When we reach a leaf, we account for file extent items in it,
8289 * walk back up the tree (adjusting slot pointers as we go)
8290 * and restart the search process.
8292 extent_buffer_get(root_eb
); /* For path */
8293 path
->nodes
[root_level
] = root_eb
;
8294 path
->slots
[root_level
] = 0;
8295 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
8298 while (level
>= 0) {
8299 if (path
->nodes
[level
] == NULL
) {
8304 /* We need to get child blockptr/gen from
8305 * parent before we can read it. */
8306 eb
= path
->nodes
[level
+ 1];
8307 parent_slot
= path
->slots
[level
+ 1];
8308 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
8309 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
8311 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
8315 } else if (!extent_buffer_uptodate(eb
)) {
8316 free_extent_buffer(eb
);
8321 path
->nodes
[level
] = eb
;
8322 path
->slots
[level
] = 0;
8324 btrfs_tree_read_lock(eb
);
8325 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
8326 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
8328 ret
= record_one_subtree_extent(trans
, root
, child_bytenr
,
8335 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
8339 /* Nonzero return here means we completed our search */
8340 ret
= adjust_slots_upwards(root
, path
, root_level
);
8344 /* Restart search with new slots */
8353 btrfs_free_path(path
);
8359 * helper to process tree block while walking down the tree.
8361 * when wc->stage == UPDATE_BACKREF, this function updates
8362 * back refs for pointers in the block.
8364 * NOTE: return value 1 means we should stop walking down.
8366 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8367 struct btrfs_root
*root
,
8368 struct btrfs_path
*path
,
8369 struct walk_control
*wc
, int lookup_info
)
8371 int level
= wc
->level
;
8372 struct extent_buffer
*eb
= path
->nodes
[level
];
8373 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8376 if (wc
->stage
== UPDATE_BACKREF
&&
8377 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8381 * when reference count of tree block is 1, it won't increase
8382 * again. once full backref flag is set, we never clear it.
8385 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8386 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8387 BUG_ON(!path
->locks
[level
]);
8388 ret
= btrfs_lookup_extent_info(trans
, root
,
8389 eb
->start
, level
, 1,
8392 BUG_ON(ret
== -ENOMEM
);
8395 BUG_ON(wc
->refs
[level
] == 0);
8398 if (wc
->stage
== DROP_REFERENCE
) {
8399 if (wc
->refs
[level
] > 1)
8402 if (path
->locks
[level
] && !wc
->keep_locks
) {
8403 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8404 path
->locks
[level
] = 0;
8409 /* wc->stage == UPDATE_BACKREF */
8410 if (!(wc
->flags
[level
] & flag
)) {
8411 BUG_ON(!path
->locks
[level
]);
8412 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8413 BUG_ON(ret
); /* -ENOMEM */
8414 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8415 BUG_ON(ret
); /* -ENOMEM */
8416 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8418 btrfs_header_level(eb
), 0);
8419 BUG_ON(ret
); /* -ENOMEM */
8420 wc
->flags
[level
] |= flag
;
8424 * the block is shared by multiple trees, so it's not good to
8425 * keep the tree lock
8427 if (path
->locks
[level
] && level
> 0) {
8428 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8429 path
->locks
[level
] = 0;
8435 * helper to process tree block pointer.
8437 * when wc->stage == DROP_REFERENCE, this function checks
8438 * reference count of the block pointed to. if the block
8439 * is shared and we need update back refs for the subtree
8440 * rooted at the block, this function changes wc->stage to
8441 * UPDATE_BACKREF. if the block is shared and there is no
8442 * need to update back, this function drops the reference
8445 * NOTE: return value 1 means we should stop walking down.
8447 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8448 struct btrfs_root
*root
,
8449 struct btrfs_path
*path
,
8450 struct walk_control
*wc
, int *lookup_info
)
8456 struct btrfs_key key
;
8457 struct extent_buffer
*next
;
8458 int level
= wc
->level
;
8461 bool need_account
= false;
8463 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8464 path
->slots
[level
]);
8466 * if the lower level block was created before the snapshot
8467 * was created, we know there is no need to update back refs
8470 if (wc
->stage
== UPDATE_BACKREF
&&
8471 generation
<= root
->root_key
.offset
) {
8476 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8477 blocksize
= root
->nodesize
;
8479 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8481 next
= btrfs_find_create_tree_block(root
, bytenr
);
8484 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8488 btrfs_tree_lock(next
);
8489 btrfs_set_lock_blocking(next
);
8491 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8492 &wc
->refs
[level
- 1],
8493 &wc
->flags
[level
- 1]);
8497 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8498 btrfs_err(root
->fs_info
, "Missing references.");
8504 if (wc
->stage
== DROP_REFERENCE
) {
8505 if (wc
->refs
[level
- 1] > 1) {
8506 need_account
= true;
8508 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8511 if (!wc
->update_ref
||
8512 generation
<= root
->root_key
.offset
)
8515 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8516 path
->slots
[level
]);
8517 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8521 wc
->stage
= UPDATE_BACKREF
;
8522 wc
->shared_level
= level
- 1;
8526 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8530 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8531 btrfs_tree_unlock(next
);
8532 free_extent_buffer(next
);
8538 if (reada
&& level
== 1)
8539 reada_walk_down(trans
, root
, wc
, path
);
8540 next
= read_tree_block(root
, bytenr
, generation
);
8542 return PTR_ERR(next
);
8543 } else if (!extent_buffer_uptodate(next
)) {
8544 free_extent_buffer(next
);
8547 btrfs_tree_lock(next
);
8548 btrfs_set_lock_blocking(next
);
8552 ASSERT(level
== btrfs_header_level(next
));
8553 if (level
!= btrfs_header_level(next
)) {
8554 btrfs_err(root
->fs_info
, "mismatched level");
8558 path
->nodes
[level
] = next
;
8559 path
->slots
[level
] = 0;
8560 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8566 wc
->refs
[level
- 1] = 0;
8567 wc
->flags
[level
- 1] = 0;
8568 if (wc
->stage
== DROP_REFERENCE
) {
8569 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8570 parent
= path
->nodes
[level
]->start
;
8572 ASSERT(root
->root_key
.objectid
==
8573 btrfs_header_owner(path
->nodes
[level
]));
8574 if (root
->root_key
.objectid
!=
8575 btrfs_header_owner(path
->nodes
[level
])) {
8576 btrfs_err(root
->fs_info
,
8577 "mismatched block owner");
8585 ret
= account_shared_subtree(trans
, root
, next
,
8586 generation
, level
- 1);
8588 btrfs_err_rl(root
->fs_info
,
8590 "%d accounting shared subtree. Quota "
8591 "is out of sync, rescan required.",
8595 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8596 root
->root_key
.objectid
, level
- 1, 0);
8605 btrfs_tree_unlock(next
);
8606 free_extent_buffer(next
);
8612 * helper to process tree block while walking up the tree.
8614 * when wc->stage == DROP_REFERENCE, this function drops
8615 * reference count on the block.
8617 * when wc->stage == UPDATE_BACKREF, this function changes
8618 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8619 * to UPDATE_BACKREF previously while processing the block.
8621 * NOTE: return value 1 means we should stop walking up.
8623 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8624 struct btrfs_root
*root
,
8625 struct btrfs_path
*path
,
8626 struct walk_control
*wc
)
8629 int level
= wc
->level
;
8630 struct extent_buffer
*eb
= path
->nodes
[level
];
8633 if (wc
->stage
== UPDATE_BACKREF
) {
8634 BUG_ON(wc
->shared_level
< level
);
8635 if (level
< wc
->shared_level
)
8638 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8642 wc
->stage
= DROP_REFERENCE
;
8643 wc
->shared_level
= -1;
8644 path
->slots
[level
] = 0;
8647 * check reference count again if the block isn't locked.
8648 * we should start walking down the tree again if reference
8651 if (!path
->locks
[level
]) {
8653 btrfs_tree_lock(eb
);
8654 btrfs_set_lock_blocking(eb
);
8655 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8657 ret
= btrfs_lookup_extent_info(trans
, root
,
8658 eb
->start
, level
, 1,
8662 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8663 path
->locks
[level
] = 0;
8666 BUG_ON(wc
->refs
[level
] == 0);
8667 if (wc
->refs
[level
] == 1) {
8668 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8669 path
->locks
[level
] = 0;
8675 /* wc->stage == DROP_REFERENCE */
8676 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8678 if (wc
->refs
[level
] == 1) {
8680 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8681 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8683 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8684 BUG_ON(ret
); /* -ENOMEM */
8685 ret
= account_leaf_items(trans
, root
, eb
);
8687 btrfs_err_rl(root
->fs_info
,
8689 "%d accounting leaf items. Quota "
8690 "is out of sync, rescan required.",
8694 /* make block locked assertion in clean_tree_block happy */
8695 if (!path
->locks
[level
] &&
8696 btrfs_header_generation(eb
) == trans
->transid
) {
8697 btrfs_tree_lock(eb
);
8698 btrfs_set_lock_blocking(eb
);
8699 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8701 clean_tree_block(trans
, root
->fs_info
, eb
);
8704 if (eb
== root
->node
) {
8705 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8708 BUG_ON(root
->root_key
.objectid
!=
8709 btrfs_header_owner(eb
));
8711 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8712 parent
= path
->nodes
[level
+ 1]->start
;
8714 BUG_ON(root
->root_key
.objectid
!=
8715 btrfs_header_owner(path
->nodes
[level
+ 1]));
8718 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8720 wc
->refs
[level
] = 0;
8721 wc
->flags
[level
] = 0;
8725 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8726 struct btrfs_root
*root
,
8727 struct btrfs_path
*path
,
8728 struct walk_control
*wc
)
8730 int level
= wc
->level
;
8731 int lookup_info
= 1;
8734 while (level
>= 0) {
8735 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8742 if (path
->slots
[level
] >=
8743 btrfs_header_nritems(path
->nodes
[level
]))
8746 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8748 path
->slots
[level
]++;
8757 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8758 struct btrfs_root
*root
,
8759 struct btrfs_path
*path
,
8760 struct walk_control
*wc
, int max_level
)
8762 int level
= wc
->level
;
8765 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8766 while (level
< max_level
&& path
->nodes
[level
]) {
8768 if (path
->slots
[level
] + 1 <
8769 btrfs_header_nritems(path
->nodes
[level
])) {
8770 path
->slots
[level
]++;
8773 ret
= walk_up_proc(trans
, root
, path
, wc
);
8777 if (path
->locks
[level
]) {
8778 btrfs_tree_unlock_rw(path
->nodes
[level
],
8779 path
->locks
[level
]);
8780 path
->locks
[level
] = 0;
8782 free_extent_buffer(path
->nodes
[level
]);
8783 path
->nodes
[level
] = NULL
;
8791 * drop a subvolume tree.
8793 * this function traverses the tree freeing any blocks that only
8794 * referenced by the tree.
8796 * when a shared tree block is found. this function decreases its
8797 * reference count by one. if update_ref is true, this function
8798 * also make sure backrefs for the shared block and all lower level
8799 * blocks are properly updated.
8801 * If called with for_reloc == 0, may exit early with -EAGAIN
8803 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8804 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8807 struct btrfs_path
*path
;
8808 struct btrfs_trans_handle
*trans
;
8809 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8810 struct btrfs_root_item
*root_item
= &root
->root_item
;
8811 struct walk_control
*wc
;
8812 struct btrfs_key key
;
8816 bool root_dropped
= false;
8818 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8820 path
= btrfs_alloc_path();
8826 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8828 btrfs_free_path(path
);
8833 trans
= btrfs_start_transaction(tree_root
, 0);
8834 if (IS_ERR(trans
)) {
8835 err
= PTR_ERR(trans
);
8840 trans
->block_rsv
= block_rsv
;
8842 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8843 level
= btrfs_header_level(root
->node
);
8844 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8845 btrfs_set_lock_blocking(path
->nodes
[level
]);
8846 path
->slots
[level
] = 0;
8847 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8848 memset(&wc
->update_progress
, 0,
8849 sizeof(wc
->update_progress
));
8851 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8852 memcpy(&wc
->update_progress
, &key
,
8853 sizeof(wc
->update_progress
));
8855 level
= root_item
->drop_level
;
8857 path
->lowest_level
= level
;
8858 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8859 path
->lowest_level
= 0;
8867 * unlock our path, this is safe because only this
8868 * function is allowed to delete this snapshot
8870 btrfs_unlock_up_safe(path
, 0);
8872 level
= btrfs_header_level(root
->node
);
8874 btrfs_tree_lock(path
->nodes
[level
]);
8875 btrfs_set_lock_blocking(path
->nodes
[level
]);
8876 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8878 ret
= btrfs_lookup_extent_info(trans
, root
,
8879 path
->nodes
[level
]->start
,
8880 level
, 1, &wc
->refs
[level
],
8886 BUG_ON(wc
->refs
[level
] == 0);
8888 if (level
== root_item
->drop_level
)
8891 btrfs_tree_unlock(path
->nodes
[level
]);
8892 path
->locks
[level
] = 0;
8893 WARN_ON(wc
->refs
[level
] != 1);
8899 wc
->shared_level
= -1;
8900 wc
->stage
= DROP_REFERENCE
;
8901 wc
->update_ref
= update_ref
;
8903 wc
->for_reloc
= for_reloc
;
8904 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8908 ret
= walk_down_tree(trans
, root
, path
, wc
);
8914 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8921 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8925 if (wc
->stage
== DROP_REFERENCE
) {
8927 btrfs_node_key(path
->nodes
[level
],
8928 &root_item
->drop_progress
,
8929 path
->slots
[level
]);
8930 root_item
->drop_level
= level
;
8933 BUG_ON(wc
->level
== 0);
8934 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8935 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8936 ret
= btrfs_update_root(trans
, tree_root
,
8940 btrfs_abort_transaction(trans
, tree_root
, ret
);
8945 btrfs_end_transaction_throttle(trans
, tree_root
);
8946 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8947 pr_debug("BTRFS: drop snapshot early exit\n");
8952 trans
= btrfs_start_transaction(tree_root
, 0);
8953 if (IS_ERR(trans
)) {
8954 err
= PTR_ERR(trans
);
8958 trans
->block_rsv
= block_rsv
;
8961 btrfs_release_path(path
);
8965 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8967 btrfs_abort_transaction(trans
, tree_root
, ret
);
8971 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8972 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8975 btrfs_abort_transaction(trans
, tree_root
, ret
);
8978 } else if (ret
> 0) {
8979 /* if we fail to delete the orphan item this time
8980 * around, it'll get picked up the next time.
8982 * The most common failure here is just -ENOENT.
8984 btrfs_del_orphan_item(trans
, tree_root
,
8985 root
->root_key
.objectid
);
8989 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8990 btrfs_add_dropped_root(trans
, root
);
8992 free_extent_buffer(root
->node
);
8993 free_extent_buffer(root
->commit_root
);
8994 btrfs_put_fs_root(root
);
8996 root_dropped
= true;
8998 btrfs_end_transaction_throttle(trans
, tree_root
);
9001 btrfs_free_path(path
);
9004 * So if we need to stop dropping the snapshot for whatever reason we
9005 * need to make sure to add it back to the dead root list so that we
9006 * keep trying to do the work later. This also cleans up roots if we
9007 * don't have it in the radix (like when we recover after a power fail
9008 * or unmount) so we don't leak memory.
9010 if (!for_reloc
&& root_dropped
== false)
9011 btrfs_add_dead_root(root
);
9012 if (err
&& err
!= -EAGAIN
)
9013 btrfs_std_error(root
->fs_info
, err
, NULL
);
9018 * drop subtree rooted at tree block 'node'.
9020 * NOTE: this function will unlock and release tree block 'node'
9021 * only used by relocation code
9023 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9024 struct btrfs_root
*root
,
9025 struct extent_buffer
*node
,
9026 struct extent_buffer
*parent
)
9028 struct btrfs_path
*path
;
9029 struct walk_control
*wc
;
9035 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9037 path
= btrfs_alloc_path();
9041 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9043 btrfs_free_path(path
);
9047 btrfs_assert_tree_locked(parent
);
9048 parent_level
= btrfs_header_level(parent
);
9049 extent_buffer_get(parent
);
9050 path
->nodes
[parent_level
] = parent
;
9051 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9053 btrfs_assert_tree_locked(node
);
9054 level
= btrfs_header_level(node
);
9055 path
->nodes
[level
] = node
;
9056 path
->slots
[level
] = 0;
9057 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9059 wc
->refs
[parent_level
] = 1;
9060 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9062 wc
->shared_level
= -1;
9063 wc
->stage
= DROP_REFERENCE
;
9067 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9070 wret
= walk_down_tree(trans
, root
, path
, wc
);
9076 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9084 btrfs_free_path(path
);
9088 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
9094 * if restripe for this chunk_type is on pick target profile and
9095 * return, otherwise do the usual balance
9097 stripped
= get_restripe_target(root
->fs_info
, flags
);
9099 return extended_to_chunk(stripped
);
9101 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
9103 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9104 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9105 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9107 if (num_devices
== 1) {
9108 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9109 stripped
= flags
& ~stripped
;
9111 /* turn raid0 into single device chunks */
9112 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9115 /* turn mirroring into duplication */
9116 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9117 BTRFS_BLOCK_GROUP_RAID10
))
9118 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9120 /* they already had raid on here, just return */
9121 if (flags
& stripped
)
9124 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9125 stripped
= flags
& ~stripped
;
9127 /* switch duplicated blocks with raid1 */
9128 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9129 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9131 /* this is drive concat, leave it alone */
9137 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9139 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9141 u64 min_allocable_bytes
;
9145 * We need some metadata space and system metadata space for
9146 * allocating chunks in some corner cases until we force to set
9147 * it to be readonly.
9150 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9152 min_allocable_bytes
= 1 * 1024 * 1024;
9154 min_allocable_bytes
= 0;
9156 spin_lock(&sinfo
->lock
);
9157 spin_lock(&cache
->lock
);
9165 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9166 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9168 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
9169 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
9170 min_allocable_bytes
<= sinfo
->total_bytes
) {
9171 sinfo
->bytes_readonly
+= num_bytes
;
9173 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9177 spin_unlock(&cache
->lock
);
9178 spin_unlock(&sinfo
->lock
);
9182 int btrfs_inc_block_group_ro(struct btrfs_root
*root
,
9183 struct btrfs_block_group_cache
*cache
)
9186 struct btrfs_trans_handle
*trans
;
9191 trans
= btrfs_join_transaction(root
);
9193 return PTR_ERR(trans
);
9196 * we're not allowed to set block groups readonly after the dirty
9197 * block groups cache has started writing. If it already started,
9198 * back off and let this transaction commit
9200 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
9201 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9202 u64 transid
= trans
->transid
;
9204 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9205 btrfs_end_transaction(trans
, root
);
9207 ret
= btrfs_wait_for_commit(root
, transid
);
9214 * if we are changing raid levels, try to allocate a corresponding
9215 * block group with the new raid level.
9217 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9218 if (alloc_flags
!= cache
->flags
) {
9219 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9222 * ENOSPC is allowed here, we may have enough space
9223 * already allocated at the new raid level to
9232 ret
= inc_block_group_ro(cache
, 0);
9235 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
9236 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9240 ret
= inc_block_group_ro(cache
, 0);
9242 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9243 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9244 lock_chunks(root
->fs_info
->chunk_root
);
9245 check_system_chunk(trans
, root
, alloc_flags
);
9246 unlock_chunks(root
->fs_info
->chunk_root
);
9248 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9250 btrfs_end_transaction(trans
, root
);
9254 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9255 struct btrfs_root
*root
, u64 type
)
9257 u64 alloc_flags
= get_alloc_profile(root
, type
);
9258 return do_chunk_alloc(trans
, root
, alloc_flags
,
9263 * helper to account the unused space of all the readonly block group in the
9264 * space_info. takes mirrors into account.
9266 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9268 struct btrfs_block_group_cache
*block_group
;
9272 /* It's df, we don't care if it's racey */
9273 if (list_empty(&sinfo
->ro_bgs
))
9276 spin_lock(&sinfo
->lock
);
9277 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9278 spin_lock(&block_group
->lock
);
9280 if (!block_group
->ro
) {
9281 spin_unlock(&block_group
->lock
);
9285 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9286 BTRFS_BLOCK_GROUP_RAID10
|
9287 BTRFS_BLOCK_GROUP_DUP
))
9292 free_bytes
+= (block_group
->key
.offset
-
9293 btrfs_block_group_used(&block_group
->item
)) *
9296 spin_unlock(&block_group
->lock
);
9298 spin_unlock(&sinfo
->lock
);
9303 void btrfs_dec_block_group_ro(struct btrfs_root
*root
,
9304 struct btrfs_block_group_cache
*cache
)
9306 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9311 spin_lock(&sinfo
->lock
);
9312 spin_lock(&cache
->lock
);
9314 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9315 cache
->pinned
- cache
->bytes_super
-
9316 btrfs_block_group_used(&cache
->item
);
9317 sinfo
->bytes_readonly
-= num_bytes
;
9318 list_del_init(&cache
->ro_list
);
9320 spin_unlock(&cache
->lock
);
9321 spin_unlock(&sinfo
->lock
);
9325 * checks to see if its even possible to relocate this block group.
9327 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9328 * ok to go ahead and try.
9330 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
9332 struct btrfs_block_group_cache
*block_group
;
9333 struct btrfs_space_info
*space_info
;
9334 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
9335 struct btrfs_device
*device
;
9336 struct btrfs_trans_handle
*trans
;
9345 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
9347 /* odd, couldn't find the block group, leave it alone */
9351 min_free
= btrfs_block_group_used(&block_group
->item
);
9353 /* no bytes used, we're good */
9357 space_info
= block_group
->space_info
;
9358 spin_lock(&space_info
->lock
);
9360 full
= space_info
->full
;
9363 * if this is the last block group we have in this space, we can't
9364 * relocate it unless we're able to allocate a new chunk below.
9366 * Otherwise, we need to make sure we have room in the space to handle
9367 * all of the extents from this block group. If we can, we're good
9369 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9370 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9371 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9372 min_free
< space_info
->total_bytes
)) {
9373 spin_unlock(&space_info
->lock
);
9376 spin_unlock(&space_info
->lock
);
9379 * ok we don't have enough space, but maybe we have free space on our
9380 * devices to allocate new chunks for relocation, so loop through our
9381 * alloc devices and guess if we have enough space. if this block
9382 * group is going to be restriped, run checks against the target
9383 * profile instead of the current one.
9395 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9397 index
= __get_raid_index(extended_to_chunk(target
));
9400 * this is just a balance, so if we were marked as full
9401 * we know there is no space for a new chunk
9406 index
= get_block_group_index(block_group
);
9409 if (index
== BTRFS_RAID_RAID10
) {
9413 } else if (index
== BTRFS_RAID_RAID1
) {
9415 } else if (index
== BTRFS_RAID_DUP
) {
9418 } else if (index
== BTRFS_RAID_RAID0
) {
9419 dev_min
= fs_devices
->rw_devices
;
9420 min_free
= div64_u64(min_free
, dev_min
);
9423 /* We need to do this so that we can look at pending chunks */
9424 trans
= btrfs_join_transaction(root
);
9425 if (IS_ERR(trans
)) {
9426 ret
= PTR_ERR(trans
);
9430 mutex_lock(&root
->fs_info
->chunk_mutex
);
9431 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9435 * check to make sure we can actually find a chunk with enough
9436 * space to fit our block group in.
9438 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9439 !device
->is_tgtdev_for_dev_replace
) {
9440 ret
= find_free_dev_extent(trans
, device
, min_free
,
9445 if (dev_nr
>= dev_min
)
9451 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9452 btrfs_end_transaction(trans
, root
);
9454 btrfs_put_block_group(block_group
);
9458 static int find_first_block_group(struct btrfs_root
*root
,
9459 struct btrfs_path
*path
, struct btrfs_key
*key
)
9462 struct btrfs_key found_key
;
9463 struct extent_buffer
*leaf
;
9466 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9471 slot
= path
->slots
[0];
9472 leaf
= path
->nodes
[0];
9473 if (slot
>= btrfs_header_nritems(leaf
)) {
9474 ret
= btrfs_next_leaf(root
, path
);
9481 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9483 if (found_key
.objectid
>= key
->objectid
&&
9484 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9494 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9496 struct btrfs_block_group_cache
*block_group
;
9500 struct inode
*inode
;
9502 block_group
= btrfs_lookup_first_block_group(info
, last
);
9503 while (block_group
) {
9504 spin_lock(&block_group
->lock
);
9505 if (block_group
->iref
)
9507 spin_unlock(&block_group
->lock
);
9508 block_group
= next_block_group(info
->tree_root
,
9518 inode
= block_group
->inode
;
9519 block_group
->iref
= 0;
9520 block_group
->inode
= NULL
;
9521 spin_unlock(&block_group
->lock
);
9523 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9524 btrfs_put_block_group(block_group
);
9528 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9530 struct btrfs_block_group_cache
*block_group
;
9531 struct btrfs_space_info
*space_info
;
9532 struct btrfs_caching_control
*caching_ctl
;
9535 down_write(&info
->commit_root_sem
);
9536 while (!list_empty(&info
->caching_block_groups
)) {
9537 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9538 struct btrfs_caching_control
, list
);
9539 list_del(&caching_ctl
->list
);
9540 put_caching_control(caching_ctl
);
9542 up_write(&info
->commit_root_sem
);
9544 spin_lock(&info
->unused_bgs_lock
);
9545 while (!list_empty(&info
->unused_bgs
)) {
9546 block_group
= list_first_entry(&info
->unused_bgs
,
9547 struct btrfs_block_group_cache
,
9549 list_del_init(&block_group
->bg_list
);
9550 btrfs_put_block_group(block_group
);
9552 spin_unlock(&info
->unused_bgs_lock
);
9554 spin_lock(&info
->block_group_cache_lock
);
9555 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9556 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9558 rb_erase(&block_group
->cache_node
,
9559 &info
->block_group_cache_tree
);
9560 RB_CLEAR_NODE(&block_group
->cache_node
);
9561 spin_unlock(&info
->block_group_cache_lock
);
9563 down_write(&block_group
->space_info
->groups_sem
);
9564 list_del(&block_group
->list
);
9565 up_write(&block_group
->space_info
->groups_sem
);
9567 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9568 wait_block_group_cache_done(block_group
);
9571 * We haven't cached this block group, which means we could
9572 * possibly have excluded extents on this block group.
9574 if (block_group
->cached
== BTRFS_CACHE_NO
||
9575 block_group
->cached
== BTRFS_CACHE_ERROR
)
9576 free_excluded_extents(info
->extent_root
, block_group
);
9578 btrfs_remove_free_space_cache(block_group
);
9579 btrfs_put_block_group(block_group
);
9581 spin_lock(&info
->block_group_cache_lock
);
9583 spin_unlock(&info
->block_group_cache_lock
);
9585 /* now that all the block groups are freed, go through and
9586 * free all the space_info structs. This is only called during
9587 * the final stages of unmount, and so we know nobody is
9588 * using them. We call synchronize_rcu() once before we start,
9589 * just to be on the safe side.
9593 release_global_block_rsv(info
);
9595 while (!list_empty(&info
->space_info
)) {
9598 space_info
= list_entry(info
->space_info
.next
,
9599 struct btrfs_space_info
,
9601 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
9602 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9603 space_info
->bytes_reserved
> 0 ||
9604 space_info
->bytes_may_use
> 0)) {
9605 dump_space_info(space_info
, 0, 0);
9608 list_del(&space_info
->list
);
9609 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9610 struct kobject
*kobj
;
9611 kobj
= space_info
->block_group_kobjs
[i
];
9612 space_info
->block_group_kobjs
[i
] = NULL
;
9618 kobject_del(&space_info
->kobj
);
9619 kobject_put(&space_info
->kobj
);
9624 static void __link_block_group(struct btrfs_space_info
*space_info
,
9625 struct btrfs_block_group_cache
*cache
)
9627 int index
= get_block_group_index(cache
);
9630 down_write(&space_info
->groups_sem
);
9631 if (list_empty(&space_info
->block_groups
[index
]))
9633 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9634 up_write(&space_info
->groups_sem
);
9637 struct raid_kobject
*rkobj
;
9640 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9643 rkobj
->raid_type
= index
;
9644 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9645 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9646 "%s", get_raid_name(index
));
9648 kobject_put(&rkobj
->kobj
);
9651 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9656 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9659 static struct btrfs_block_group_cache
*
9660 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9662 struct btrfs_block_group_cache
*cache
;
9664 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9668 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9670 if (!cache
->free_space_ctl
) {
9675 cache
->key
.objectid
= start
;
9676 cache
->key
.offset
= size
;
9677 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9679 cache
->sectorsize
= root
->sectorsize
;
9680 cache
->fs_info
= root
->fs_info
;
9681 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9682 &root
->fs_info
->mapping_tree
,
9684 atomic_set(&cache
->count
, 1);
9685 spin_lock_init(&cache
->lock
);
9686 init_rwsem(&cache
->data_rwsem
);
9687 INIT_LIST_HEAD(&cache
->list
);
9688 INIT_LIST_HEAD(&cache
->cluster_list
);
9689 INIT_LIST_HEAD(&cache
->bg_list
);
9690 INIT_LIST_HEAD(&cache
->ro_list
);
9691 INIT_LIST_HEAD(&cache
->dirty_list
);
9692 INIT_LIST_HEAD(&cache
->io_list
);
9693 btrfs_init_free_space_ctl(cache
);
9694 atomic_set(&cache
->trimming
, 0);
9699 int btrfs_read_block_groups(struct btrfs_root
*root
)
9701 struct btrfs_path
*path
;
9703 struct btrfs_block_group_cache
*cache
;
9704 struct btrfs_fs_info
*info
= root
->fs_info
;
9705 struct btrfs_space_info
*space_info
;
9706 struct btrfs_key key
;
9707 struct btrfs_key found_key
;
9708 struct extent_buffer
*leaf
;
9714 feature
= btrfs_super_incompat_flags(info
->super_copy
);
9715 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
9717 root
= info
->extent_root
;
9720 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9721 path
= btrfs_alloc_path();
9726 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9727 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9728 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9730 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9734 ret
= find_first_block_group(root
, path
, &key
);
9740 leaf
= path
->nodes
[0];
9741 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9743 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9752 * When we mount with old space cache, we need to
9753 * set BTRFS_DC_CLEAR and set dirty flag.
9755 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9756 * truncate the old free space cache inode and
9758 * b) Setting 'dirty flag' makes sure that we flush
9759 * the new space cache info onto disk.
9761 if (btrfs_test_opt(root
, SPACE_CACHE
))
9762 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9765 read_extent_buffer(leaf
, &cache
->item
,
9766 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9767 sizeof(cache
->item
));
9768 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9770 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
9771 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
9773 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
9774 cache
->key
.objectid
);
9779 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9780 btrfs_release_path(path
);
9783 * We need to exclude the super stripes now so that the space
9784 * info has super bytes accounted for, otherwise we'll think
9785 * we have more space than we actually do.
9787 ret
= exclude_super_stripes(root
, cache
);
9790 * We may have excluded something, so call this just in
9793 free_excluded_extents(root
, cache
);
9794 btrfs_put_block_group(cache
);
9799 * check for two cases, either we are full, and therefore
9800 * don't need to bother with the caching work since we won't
9801 * find any space, or we are empty, and we can just add all
9802 * the space in and be done with it. This saves us _alot_ of
9803 * time, particularly in the full case.
9805 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9806 cache
->last_byte_to_unpin
= (u64
)-1;
9807 cache
->cached
= BTRFS_CACHE_FINISHED
;
9808 free_excluded_extents(root
, cache
);
9809 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9810 cache
->last_byte_to_unpin
= (u64
)-1;
9811 cache
->cached
= BTRFS_CACHE_FINISHED
;
9812 add_new_free_space(cache
, root
->fs_info
,
9814 found_key
.objectid
+
9816 free_excluded_extents(root
, cache
);
9819 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9821 btrfs_remove_free_space_cache(cache
);
9822 btrfs_put_block_group(cache
);
9826 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9827 btrfs_block_group_used(&cache
->item
),
9830 btrfs_remove_free_space_cache(cache
);
9831 spin_lock(&info
->block_group_cache_lock
);
9832 rb_erase(&cache
->cache_node
,
9833 &info
->block_group_cache_tree
);
9834 RB_CLEAR_NODE(&cache
->cache_node
);
9835 spin_unlock(&info
->block_group_cache_lock
);
9836 btrfs_put_block_group(cache
);
9840 cache
->space_info
= space_info
;
9841 spin_lock(&cache
->space_info
->lock
);
9842 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9843 spin_unlock(&cache
->space_info
->lock
);
9845 __link_block_group(space_info
, cache
);
9847 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9848 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9849 inc_block_group_ro(cache
, 1);
9850 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9851 spin_lock(&info
->unused_bgs_lock
);
9852 /* Should always be true but just in case. */
9853 if (list_empty(&cache
->bg_list
)) {
9854 btrfs_get_block_group(cache
);
9855 list_add_tail(&cache
->bg_list
,
9858 spin_unlock(&info
->unused_bgs_lock
);
9862 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9863 if (!(get_alloc_profile(root
, space_info
->flags
) &
9864 (BTRFS_BLOCK_GROUP_RAID10
|
9865 BTRFS_BLOCK_GROUP_RAID1
|
9866 BTRFS_BLOCK_GROUP_RAID5
|
9867 BTRFS_BLOCK_GROUP_RAID6
|
9868 BTRFS_BLOCK_GROUP_DUP
)))
9871 * avoid allocating from un-mirrored block group if there are
9872 * mirrored block groups.
9874 list_for_each_entry(cache
,
9875 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9877 inc_block_group_ro(cache
, 1);
9878 list_for_each_entry(cache
,
9879 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9881 inc_block_group_ro(cache
, 1);
9884 init_global_block_rsv(info
);
9887 btrfs_free_path(path
);
9891 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9892 struct btrfs_root
*root
)
9894 struct btrfs_block_group_cache
*block_group
, *tmp
;
9895 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9896 struct btrfs_block_group_item item
;
9897 struct btrfs_key key
;
9899 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
9901 trans
->can_flush_pending_bgs
= false;
9902 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9906 spin_lock(&block_group
->lock
);
9907 memcpy(&item
, &block_group
->item
, sizeof(item
));
9908 memcpy(&key
, &block_group
->key
, sizeof(key
));
9909 spin_unlock(&block_group
->lock
);
9911 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9914 btrfs_abort_transaction(trans
, extent_root
, ret
);
9915 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9916 key
.objectid
, key
.offset
);
9918 btrfs_abort_transaction(trans
, extent_root
, ret
);
9920 list_del_init(&block_group
->bg_list
);
9922 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
9925 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9926 struct btrfs_root
*root
, u64 bytes_used
,
9927 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9931 struct btrfs_root
*extent_root
;
9932 struct btrfs_block_group_cache
*cache
;
9934 extent_root
= root
->fs_info
->extent_root
;
9936 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9938 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9942 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9943 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9944 btrfs_set_block_group_flags(&cache
->item
, type
);
9946 cache
->flags
= type
;
9947 cache
->last_byte_to_unpin
= (u64
)-1;
9948 cache
->cached
= BTRFS_CACHE_FINISHED
;
9949 ret
= exclude_super_stripes(root
, cache
);
9952 * We may have excluded something, so call this just in
9955 free_excluded_extents(root
, cache
);
9956 btrfs_put_block_group(cache
);
9960 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9961 chunk_offset
+ size
);
9963 free_excluded_extents(root
, cache
);
9965 #ifdef CONFIG_BTRFS_DEBUG
9966 if (btrfs_should_fragment_free_space(root
, cache
)) {
9967 u64 new_bytes_used
= size
- bytes_used
;
9969 bytes_used
+= new_bytes_used
>> 1;
9970 fragment_free_space(root
, cache
);
9974 * Call to ensure the corresponding space_info object is created and
9975 * assigned to our block group, but don't update its counters just yet.
9976 * We want our bg to be added to the rbtree with its ->space_info set.
9978 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0,
9979 &cache
->space_info
);
9981 btrfs_remove_free_space_cache(cache
);
9982 btrfs_put_block_group(cache
);
9986 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9988 btrfs_remove_free_space_cache(cache
);
9989 btrfs_put_block_group(cache
);
9994 * Now that our block group has its ->space_info set and is inserted in
9995 * the rbtree, update the space info's counters.
9997 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9998 &cache
->space_info
);
10000 btrfs_remove_free_space_cache(cache
);
10001 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10002 rb_erase(&cache
->cache_node
,
10003 &root
->fs_info
->block_group_cache_tree
);
10004 RB_CLEAR_NODE(&cache
->cache_node
);
10005 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10006 btrfs_put_block_group(cache
);
10009 update_global_block_rsv(root
->fs_info
);
10011 spin_lock(&cache
->space_info
->lock
);
10012 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
10013 spin_unlock(&cache
->space_info
->lock
);
10015 __link_block_group(cache
->space_info
, cache
);
10017 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10019 set_avail_alloc_bits(extent_root
->fs_info
, type
);
10024 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10026 u64 extra_flags
= chunk_to_extended(flags
) &
10027 BTRFS_EXTENDED_PROFILE_MASK
;
10029 write_seqlock(&fs_info
->profiles_lock
);
10030 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10031 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10032 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10033 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10034 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10035 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10036 write_sequnlock(&fs_info
->profiles_lock
);
10039 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10040 struct btrfs_root
*root
, u64 group_start
,
10041 struct extent_map
*em
)
10043 struct btrfs_path
*path
;
10044 struct btrfs_block_group_cache
*block_group
;
10045 struct btrfs_free_cluster
*cluster
;
10046 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
10047 struct btrfs_key key
;
10048 struct inode
*inode
;
10049 struct kobject
*kobj
= NULL
;
10053 struct btrfs_caching_control
*caching_ctl
= NULL
;
10056 root
= root
->fs_info
->extent_root
;
10058 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
10059 BUG_ON(!block_group
);
10060 BUG_ON(!block_group
->ro
);
10063 * Free the reserved super bytes from this block group before
10066 free_excluded_extents(root
, block_group
);
10068 memcpy(&key
, &block_group
->key
, sizeof(key
));
10069 index
= get_block_group_index(block_group
);
10070 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10071 BTRFS_BLOCK_GROUP_RAID1
|
10072 BTRFS_BLOCK_GROUP_RAID10
))
10077 /* make sure this block group isn't part of an allocation cluster */
10078 cluster
= &root
->fs_info
->data_alloc_cluster
;
10079 spin_lock(&cluster
->refill_lock
);
10080 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10081 spin_unlock(&cluster
->refill_lock
);
10084 * make sure this block group isn't part of a metadata
10085 * allocation cluster
10087 cluster
= &root
->fs_info
->meta_alloc_cluster
;
10088 spin_lock(&cluster
->refill_lock
);
10089 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10090 spin_unlock(&cluster
->refill_lock
);
10092 path
= btrfs_alloc_path();
10099 * get the inode first so any iput calls done for the io_list
10100 * aren't the final iput (no unlinks allowed now)
10102 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
10104 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10106 * make sure our free spache cache IO is done before remove the
10109 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10110 if (!list_empty(&block_group
->io_list
)) {
10111 list_del_init(&block_group
->io_list
);
10113 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10115 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10116 btrfs_wait_cache_io(root
, trans
, block_group
,
10117 &block_group
->io_ctl
, path
,
10118 block_group
->key
.objectid
);
10119 btrfs_put_block_group(block_group
);
10120 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10123 if (!list_empty(&block_group
->dirty_list
)) {
10124 list_del_init(&block_group
->dirty_list
);
10125 btrfs_put_block_group(block_group
);
10127 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10128 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10130 if (!IS_ERR(inode
)) {
10131 ret
= btrfs_orphan_add(trans
, inode
);
10133 btrfs_add_delayed_iput(inode
);
10136 clear_nlink(inode
);
10137 /* One for the block groups ref */
10138 spin_lock(&block_group
->lock
);
10139 if (block_group
->iref
) {
10140 block_group
->iref
= 0;
10141 block_group
->inode
= NULL
;
10142 spin_unlock(&block_group
->lock
);
10145 spin_unlock(&block_group
->lock
);
10147 /* One for our lookup ref */
10148 btrfs_add_delayed_iput(inode
);
10151 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10152 key
.offset
= block_group
->key
.objectid
;
10155 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10159 btrfs_release_path(path
);
10161 ret
= btrfs_del_item(trans
, tree_root
, path
);
10164 btrfs_release_path(path
);
10167 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10168 rb_erase(&block_group
->cache_node
,
10169 &root
->fs_info
->block_group_cache_tree
);
10170 RB_CLEAR_NODE(&block_group
->cache_node
);
10172 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10173 root
->fs_info
->first_logical_byte
= (u64
)-1;
10174 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10176 down_write(&block_group
->space_info
->groups_sem
);
10178 * we must use list_del_init so people can check to see if they
10179 * are still on the list after taking the semaphore
10181 list_del_init(&block_group
->list
);
10182 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10183 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10184 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10185 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
10187 up_write(&block_group
->space_info
->groups_sem
);
10193 if (block_group
->has_caching_ctl
)
10194 caching_ctl
= get_caching_control(block_group
);
10195 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10196 wait_block_group_cache_done(block_group
);
10197 if (block_group
->has_caching_ctl
) {
10198 down_write(&root
->fs_info
->commit_root_sem
);
10199 if (!caching_ctl
) {
10200 struct btrfs_caching_control
*ctl
;
10202 list_for_each_entry(ctl
,
10203 &root
->fs_info
->caching_block_groups
, list
)
10204 if (ctl
->block_group
== block_group
) {
10206 atomic_inc(&caching_ctl
->count
);
10211 list_del_init(&caching_ctl
->list
);
10212 up_write(&root
->fs_info
->commit_root_sem
);
10214 /* Once for the caching bgs list and once for us. */
10215 put_caching_control(caching_ctl
);
10216 put_caching_control(caching_ctl
);
10220 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10221 if (!list_empty(&block_group
->dirty_list
)) {
10224 if (!list_empty(&block_group
->io_list
)) {
10227 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10228 btrfs_remove_free_space_cache(block_group
);
10230 spin_lock(&block_group
->space_info
->lock
);
10231 list_del_init(&block_group
->ro_list
);
10233 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
10234 WARN_ON(block_group
->space_info
->total_bytes
10235 < block_group
->key
.offset
);
10236 WARN_ON(block_group
->space_info
->bytes_readonly
10237 < block_group
->key
.offset
);
10238 WARN_ON(block_group
->space_info
->disk_total
10239 < block_group
->key
.offset
* factor
);
10241 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10242 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10243 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10245 spin_unlock(&block_group
->space_info
->lock
);
10247 memcpy(&key
, &block_group
->key
, sizeof(key
));
10250 if (!list_empty(&em
->list
)) {
10251 /* We're in the transaction->pending_chunks list. */
10252 free_extent_map(em
);
10254 spin_lock(&block_group
->lock
);
10255 block_group
->removed
= 1;
10257 * At this point trimming can't start on this block group, because we
10258 * removed the block group from the tree fs_info->block_group_cache_tree
10259 * so no one can't find it anymore and even if someone already got this
10260 * block group before we removed it from the rbtree, they have already
10261 * incremented block_group->trimming - if they didn't, they won't find
10262 * any free space entries because we already removed them all when we
10263 * called btrfs_remove_free_space_cache().
10265 * And we must not remove the extent map from the fs_info->mapping_tree
10266 * to prevent the same logical address range and physical device space
10267 * ranges from being reused for a new block group. This is because our
10268 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10269 * completely transactionless, so while it is trimming a range the
10270 * currently running transaction might finish and a new one start,
10271 * allowing for new block groups to be created that can reuse the same
10272 * physical device locations unless we take this special care.
10274 * There may also be an implicit trim operation if the file system
10275 * is mounted with -odiscard. The same protections must remain
10276 * in place until the extents have been discarded completely when
10277 * the transaction commit has completed.
10279 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10281 * Make sure a trimmer task always sees the em in the pinned_chunks list
10282 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10283 * before checking block_group->removed).
10287 * Our em might be in trans->transaction->pending_chunks which
10288 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10289 * and so is the fs_info->pinned_chunks list.
10291 * So at this point we must be holding the chunk_mutex to avoid
10292 * any races with chunk allocation (more specifically at
10293 * volumes.c:contains_pending_extent()), to ensure it always
10294 * sees the em, either in the pending_chunks list or in the
10295 * pinned_chunks list.
10297 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
10299 spin_unlock(&block_group
->lock
);
10302 struct extent_map_tree
*em_tree
;
10304 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
10305 write_lock(&em_tree
->lock
);
10307 * The em might be in the pending_chunks list, so make sure the
10308 * chunk mutex is locked, since remove_extent_mapping() will
10309 * delete us from that list.
10311 remove_extent_mapping(em_tree
, em
);
10312 write_unlock(&em_tree
->lock
);
10313 /* once for the tree */
10314 free_extent_map(em
);
10317 unlock_chunks(root
);
10319 btrfs_put_block_group(block_group
);
10320 btrfs_put_block_group(block_group
);
10322 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10328 ret
= btrfs_del_item(trans
, root
, path
);
10330 btrfs_free_path(path
);
10334 struct btrfs_trans_handle
*
10335 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10336 const u64 chunk_offset
)
10338 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10339 struct extent_map
*em
;
10340 struct map_lookup
*map
;
10341 unsigned int num_items
;
10343 read_lock(&em_tree
->lock
);
10344 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10345 read_unlock(&em_tree
->lock
);
10346 ASSERT(em
&& em
->start
== chunk_offset
);
10349 * We need to reserve 3 + N units from the metadata space info in order
10350 * to remove a block group (done at btrfs_remove_chunk() and at
10351 * btrfs_remove_block_group()), which are used for:
10353 * 1 unit for adding the free space inode's orphan (located in the tree
10355 * 1 unit for deleting the block group item (located in the extent
10357 * 1 unit for deleting the free space item (located in tree of tree
10359 * N units for deleting N device extent items corresponding to each
10360 * stripe (located in the device tree).
10362 * In order to remove a block group we also need to reserve units in the
10363 * system space info in order to update the chunk tree (update one or
10364 * more device items and remove one chunk item), but this is done at
10365 * btrfs_remove_chunk() through a call to check_system_chunk().
10367 map
= (struct map_lookup
*)em
->bdev
;
10368 num_items
= 3 + map
->num_stripes
;
10369 free_extent_map(em
);
10371 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10376 * Process the unused_bgs list and remove any that don't have any allocated
10377 * space inside of them.
10379 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10381 struct btrfs_block_group_cache
*block_group
;
10382 struct btrfs_space_info
*space_info
;
10383 struct btrfs_root
*root
= fs_info
->extent_root
;
10384 struct btrfs_trans_handle
*trans
;
10387 if (!fs_info
->open
)
10390 spin_lock(&fs_info
->unused_bgs_lock
);
10391 while (!list_empty(&fs_info
->unused_bgs
)) {
10395 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10396 struct btrfs_block_group_cache
,
10398 list_del_init(&block_group
->bg_list
);
10400 space_info
= block_group
->space_info
;
10402 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10403 btrfs_put_block_group(block_group
);
10406 spin_unlock(&fs_info
->unused_bgs_lock
);
10408 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10410 /* Don't want to race with allocators so take the groups_sem */
10411 down_write(&space_info
->groups_sem
);
10412 spin_lock(&block_group
->lock
);
10413 if (block_group
->reserved
|| block_group
->pinned
||
10414 btrfs_block_group_used(&block_group
->item
) ||
10416 list_is_singular(&block_group
->list
)) {
10418 * We want to bail if we made new allocations or have
10419 * outstanding allocations in this block group. We do
10420 * the ro check in case balance is currently acting on
10421 * this block group.
10423 spin_unlock(&block_group
->lock
);
10424 up_write(&space_info
->groups_sem
);
10427 spin_unlock(&block_group
->lock
);
10429 /* We don't want to force the issue, only flip if it's ok. */
10430 ret
= inc_block_group_ro(block_group
, 0);
10431 up_write(&space_info
->groups_sem
);
10438 * Want to do this before we do anything else so we can recover
10439 * properly if we fail to join the transaction.
10441 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10442 block_group
->key
.objectid
);
10443 if (IS_ERR(trans
)) {
10444 btrfs_dec_block_group_ro(root
, block_group
);
10445 ret
= PTR_ERR(trans
);
10450 * We could have pending pinned extents for this block group,
10451 * just delete them, we don't care about them anymore.
10453 start
= block_group
->key
.objectid
;
10454 end
= start
+ block_group
->key
.offset
- 1;
10456 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10457 * btrfs_finish_extent_commit(). If we are at transaction N,
10458 * another task might be running finish_extent_commit() for the
10459 * previous transaction N - 1, and have seen a range belonging
10460 * to the block group in freed_extents[] before we were able to
10461 * clear the whole block group range from freed_extents[]. This
10462 * means that task can lookup for the block group after we
10463 * unpinned it from freed_extents[] and removed it, leading to
10464 * a BUG_ON() at btrfs_unpin_extent_range().
10466 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10467 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10468 EXTENT_DIRTY
, GFP_NOFS
);
10470 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10471 btrfs_dec_block_group_ro(root
, block_group
);
10474 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10475 EXTENT_DIRTY
, GFP_NOFS
);
10477 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10478 btrfs_dec_block_group_ro(root
, block_group
);
10481 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10483 /* Reset pinned so btrfs_put_block_group doesn't complain */
10484 spin_lock(&space_info
->lock
);
10485 spin_lock(&block_group
->lock
);
10487 space_info
->bytes_pinned
-= block_group
->pinned
;
10488 space_info
->bytes_readonly
+= block_group
->pinned
;
10489 percpu_counter_add(&space_info
->total_bytes_pinned
,
10490 -block_group
->pinned
);
10491 block_group
->pinned
= 0;
10493 spin_unlock(&block_group
->lock
);
10494 spin_unlock(&space_info
->lock
);
10496 /* DISCARD can flip during remount */
10497 trimming
= btrfs_test_opt(root
, DISCARD
);
10499 /* Implicit trim during transaction commit. */
10501 btrfs_get_block_group_trimming(block_group
);
10504 * Btrfs_remove_chunk will abort the transaction if things go
10507 ret
= btrfs_remove_chunk(trans
, root
,
10508 block_group
->key
.objectid
);
10512 btrfs_put_block_group_trimming(block_group
);
10517 * If we're not mounted with -odiscard, we can just forget
10518 * about this block group. Otherwise we'll need to wait
10519 * until transaction commit to do the actual discard.
10522 spin_lock(&fs_info
->unused_bgs_lock
);
10524 * A concurrent scrub might have added us to the list
10525 * fs_info->unused_bgs, so use a list_move operation
10526 * to add the block group to the deleted_bgs list.
10528 list_move(&block_group
->bg_list
,
10529 &trans
->transaction
->deleted_bgs
);
10530 spin_unlock(&fs_info
->unused_bgs_lock
);
10531 btrfs_get_block_group(block_group
);
10534 btrfs_end_transaction(trans
, root
);
10536 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10537 btrfs_put_block_group(block_group
);
10538 spin_lock(&fs_info
->unused_bgs_lock
);
10540 spin_unlock(&fs_info
->unused_bgs_lock
);
10543 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10545 struct btrfs_space_info
*space_info
;
10546 struct btrfs_super_block
*disk_super
;
10552 disk_super
= fs_info
->super_copy
;
10553 if (!btrfs_super_root(disk_super
))
10556 features
= btrfs_super_incompat_flags(disk_super
);
10557 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10560 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10561 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10566 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10567 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10569 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10570 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10574 flags
= BTRFS_BLOCK_GROUP_DATA
;
10575 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10581 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10583 return unpin_extent_range(root
, start
, end
, false);
10587 * It used to be that old block groups would be left around forever.
10588 * Iterating over them would be enough to trim unused space. Since we
10589 * now automatically remove them, we also need to iterate over unallocated
10592 * We don't want a transaction for this since the discard may take a
10593 * substantial amount of time. We don't require that a transaction be
10594 * running, but we do need to take a running transaction into account
10595 * to ensure that we're not discarding chunks that were released in
10596 * the current transaction.
10598 * Holding the chunks lock will prevent other threads from allocating
10599 * or releasing chunks, but it won't prevent a running transaction
10600 * from committing and releasing the memory that the pending chunks
10601 * list head uses. For that, we need to take a reference to the
10604 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10605 u64 minlen
, u64
*trimmed
)
10607 u64 start
= 0, len
= 0;
10612 /* Not writeable = nothing to do. */
10613 if (!device
->writeable
)
10616 /* No free space = nothing to do. */
10617 if (device
->total_bytes
<= device
->bytes_used
)
10623 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
10624 struct btrfs_transaction
*trans
;
10627 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10631 down_read(&fs_info
->commit_root_sem
);
10633 spin_lock(&fs_info
->trans_lock
);
10634 trans
= fs_info
->running_transaction
;
10636 atomic_inc(&trans
->use_count
);
10637 spin_unlock(&fs_info
->trans_lock
);
10639 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10642 btrfs_put_transaction(trans
);
10645 up_read(&fs_info
->commit_root_sem
);
10646 mutex_unlock(&fs_info
->chunk_mutex
);
10647 if (ret
== -ENOSPC
)
10652 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10653 up_read(&fs_info
->commit_root_sem
);
10654 mutex_unlock(&fs_info
->chunk_mutex
);
10662 if (fatal_signal_pending(current
)) {
10663 ret
= -ERESTARTSYS
;
10673 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
10675 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10676 struct btrfs_block_group_cache
*cache
= NULL
;
10677 struct btrfs_device
*device
;
10678 struct list_head
*devices
;
10683 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10687 * try to trim all FS space, our block group may start from non-zero.
10689 if (range
->len
== total_bytes
)
10690 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10692 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10695 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10696 btrfs_put_block_group(cache
);
10700 start
= max(range
->start
, cache
->key
.objectid
);
10701 end
= min(range
->start
+ range
->len
,
10702 cache
->key
.objectid
+ cache
->key
.offset
);
10704 if (end
- start
>= range
->minlen
) {
10705 if (!block_group_cache_done(cache
)) {
10706 ret
= cache_block_group(cache
, 0);
10708 btrfs_put_block_group(cache
);
10711 ret
= wait_block_group_cache_done(cache
);
10713 btrfs_put_block_group(cache
);
10717 ret
= btrfs_trim_block_group(cache
,
10723 trimmed
+= group_trimmed
;
10725 btrfs_put_block_group(cache
);
10730 cache
= next_block_group(fs_info
->tree_root
, cache
);
10733 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10734 devices
= &root
->fs_info
->fs_devices
->alloc_list
;
10735 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10736 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10741 trimmed
+= group_trimmed
;
10743 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10745 range
->len
= trimmed
;
10750 * btrfs_{start,end}_write_no_snapshoting() are similar to
10751 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10752 * data into the page cache through nocow before the subvolume is snapshoted,
10753 * but flush the data into disk after the snapshot creation, or to prevent
10754 * operations while snapshoting is ongoing and that cause the snapshot to be
10755 * inconsistent (writes followed by expanding truncates for example).
10757 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10759 percpu_counter_dec(&root
->subv_writers
->counter
);
10761 * Make sure counter is updated before we wake up waiters.
10764 if (waitqueue_active(&root
->subv_writers
->wait
))
10765 wake_up(&root
->subv_writers
->wait
);
10768 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10770 if (atomic_read(&root
->will_be_snapshoted
))
10773 percpu_counter_inc(&root
->subv_writers
->counter
);
10775 * Make sure counter is updated before we check for snapshot creation.
10778 if (atomic_read(&root
->will_be_snapshoted
)) {
10779 btrfs_end_write_no_snapshoting(root
);