1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
21 #include "print-tree.h"
25 #include "free-space-cache.h"
26 #include "free-space-tree.h"
30 #include "ref-verify.h"
32 #undef SCRAMBLE_DELAYED_REFS
35 * control flags for do_chunk_alloc's force field
36 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
37 * if we really need one.
39 * CHUNK_ALLOC_LIMITED means to only try and allocate one
40 * if we have very few chunks already allocated. This is
41 * used as part of the clustering code to help make sure
42 * we have a good pool of storage to cluster in, without
43 * filling the FS with empty chunks
45 * CHUNK_ALLOC_FORCE means it must try to allocate one
49 CHUNK_ALLOC_NO_FORCE
= 0,
50 CHUNK_ALLOC_LIMITED
= 1,
51 CHUNK_ALLOC_FORCE
= 2,
54 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
55 struct btrfs_delayed_ref_node
*node
, u64 parent
,
56 u64 root_objectid
, u64 owner_objectid
,
57 u64 owner_offset
, int refs_to_drop
,
58 struct btrfs_delayed_extent_op
*extra_op
);
59 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
60 struct extent_buffer
*leaf
,
61 struct btrfs_extent_item
*ei
);
62 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
63 u64 parent
, u64 root_objectid
,
64 u64 flags
, u64 owner
, u64 offset
,
65 struct btrfs_key
*ins
, int ref_mod
);
66 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
67 struct btrfs_delayed_ref_node
*node
,
68 struct btrfs_delayed_extent_op
*extent_op
);
69 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
70 struct btrfs_fs_info
*fs_info
, u64 flags
,
72 static int find_next_key(struct btrfs_path
*path
, int level
,
73 struct btrfs_key
*key
);
74 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
75 struct btrfs_space_info
*info
, u64 bytes
,
76 int dump_block_groups
);
77 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
79 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
80 struct btrfs_space_info
*space_info
,
82 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
83 struct btrfs_space_info
*space_info
,
87 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
90 return cache
->cached
== BTRFS_CACHE_FINISHED
||
91 cache
->cached
== BTRFS_CACHE_ERROR
;
94 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
96 return (cache
->flags
& bits
) == bits
;
99 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
101 atomic_inc(&cache
->count
);
104 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
106 if (atomic_dec_and_test(&cache
->count
)) {
107 WARN_ON(cache
->pinned
> 0);
108 WARN_ON(cache
->reserved
> 0);
111 * If not empty, someone is still holding mutex of
112 * full_stripe_lock, which can only be released by caller.
113 * And it will definitely cause use-after-free when caller
114 * tries to release full stripe lock.
116 * No better way to resolve, but only to warn.
118 WARN_ON(!RB_EMPTY_ROOT(&cache
->full_stripe_locks_root
.root
));
119 kfree(cache
->free_space_ctl
);
125 * this adds the block group to the fs_info rb tree for the block group
128 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
129 struct btrfs_block_group_cache
*block_group
)
132 struct rb_node
*parent
= NULL
;
133 struct btrfs_block_group_cache
*cache
;
135 spin_lock(&info
->block_group_cache_lock
);
136 p
= &info
->block_group_cache_tree
.rb_node
;
140 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
142 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
144 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
147 spin_unlock(&info
->block_group_cache_lock
);
152 rb_link_node(&block_group
->cache_node
, parent
, p
);
153 rb_insert_color(&block_group
->cache_node
,
154 &info
->block_group_cache_tree
);
156 if (info
->first_logical_byte
> block_group
->key
.objectid
)
157 info
->first_logical_byte
= block_group
->key
.objectid
;
159 spin_unlock(&info
->block_group_cache_lock
);
165 * This will return the block group at or after bytenr if contains is 0, else
166 * it will return the block group that contains the bytenr
168 static struct btrfs_block_group_cache
*
169 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
172 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
176 spin_lock(&info
->block_group_cache_lock
);
177 n
= info
->block_group_cache_tree
.rb_node
;
180 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
182 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
183 start
= cache
->key
.objectid
;
185 if (bytenr
< start
) {
186 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
189 } else if (bytenr
> start
) {
190 if (contains
&& bytenr
<= end
) {
201 btrfs_get_block_group(ret
);
202 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
203 info
->first_logical_byte
= ret
->key
.objectid
;
205 spin_unlock(&info
->block_group_cache_lock
);
210 static int add_excluded_extent(struct btrfs_fs_info
*fs_info
,
211 u64 start
, u64 num_bytes
)
213 u64 end
= start
+ num_bytes
- 1;
214 set_extent_bits(&fs_info
->freed_extents
[0],
215 start
, end
, EXTENT_UPTODATE
);
216 set_extent_bits(&fs_info
->freed_extents
[1],
217 start
, end
, EXTENT_UPTODATE
);
221 static void free_excluded_extents(struct btrfs_fs_info
*fs_info
,
222 struct btrfs_block_group_cache
*cache
)
226 start
= cache
->key
.objectid
;
227 end
= start
+ cache
->key
.offset
- 1;
229 clear_extent_bits(&fs_info
->freed_extents
[0],
230 start
, end
, EXTENT_UPTODATE
);
231 clear_extent_bits(&fs_info
->freed_extents
[1],
232 start
, end
, EXTENT_UPTODATE
);
235 static int exclude_super_stripes(struct btrfs_fs_info
*fs_info
,
236 struct btrfs_block_group_cache
*cache
)
243 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
244 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
245 cache
->bytes_super
+= stripe_len
;
246 ret
= add_excluded_extent(fs_info
, cache
->key
.objectid
,
252 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
253 bytenr
= btrfs_sb_offset(i
);
254 ret
= btrfs_rmap_block(fs_info
, cache
->key
.objectid
,
255 bytenr
, &logical
, &nr
, &stripe_len
);
262 if (logical
[nr
] > cache
->key
.objectid
+
266 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
270 if (start
< cache
->key
.objectid
) {
271 start
= cache
->key
.objectid
;
272 len
= (logical
[nr
] + stripe_len
) - start
;
274 len
= min_t(u64
, stripe_len
,
275 cache
->key
.objectid
+
276 cache
->key
.offset
- start
);
279 cache
->bytes_super
+= len
;
280 ret
= add_excluded_extent(fs_info
, start
, len
);
292 static struct btrfs_caching_control
*
293 get_caching_control(struct btrfs_block_group_cache
*cache
)
295 struct btrfs_caching_control
*ctl
;
297 spin_lock(&cache
->lock
);
298 if (!cache
->caching_ctl
) {
299 spin_unlock(&cache
->lock
);
303 ctl
= cache
->caching_ctl
;
304 refcount_inc(&ctl
->count
);
305 spin_unlock(&cache
->lock
);
309 static void put_caching_control(struct btrfs_caching_control
*ctl
)
311 if (refcount_dec_and_test(&ctl
->count
))
315 #ifdef CONFIG_BTRFS_DEBUG
316 static void fragment_free_space(struct btrfs_block_group_cache
*block_group
)
318 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
319 u64 start
= block_group
->key
.objectid
;
320 u64 len
= block_group
->key
.offset
;
321 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
322 fs_info
->nodesize
: fs_info
->sectorsize
;
323 u64 step
= chunk
<< 1;
325 while (len
> chunk
) {
326 btrfs_remove_free_space(block_group
, start
, chunk
);
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
341 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
344 struct btrfs_fs_info
*info
= block_group
->fs_info
;
345 u64 extent_start
, extent_end
, size
, total_added
= 0;
348 while (start
< end
) {
349 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
350 &extent_start
, &extent_end
,
351 EXTENT_DIRTY
| EXTENT_UPTODATE
,
356 if (extent_start
<= start
) {
357 start
= extent_end
+ 1;
358 } else if (extent_start
> start
&& extent_start
< end
) {
359 size
= extent_start
- start
;
361 ret
= btrfs_add_free_space(block_group
, start
,
363 BUG_ON(ret
); /* -ENOMEM or logic error */
364 start
= extent_end
+ 1;
373 ret
= btrfs_add_free_space(block_group
, start
, size
);
374 BUG_ON(ret
); /* -ENOMEM or logic error */
380 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
382 struct btrfs_block_group_cache
*block_group
= caching_ctl
->block_group
;
383 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
384 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
385 struct btrfs_path
*path
;
386 struct extent_buffer
*leaf
;
387 struct btrfs_key key
;
394 path
= btrfs_alloc_path();
398 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
400 #ifdef CONFIG_BTRFS_DEBUG
402 * If we're fragmenting we don't want to make anybody think we can
403 * allocate from this block group until we've had a chance to fragment
406 if (btrfs_should_fragment_free_space(block_group
))
410 * We don't want to deadlock with somebody trying to allocate a new
411 * extent for the extent root while also trying to search the extent
412 * root to add free space. So we skip locking and search the commit
413 * root, since its read-only
415 path
->skip_locking
= 1;
416 path
->search_commit_root
= 1;
417 path
->reada
= READA_FORWARD
;
421 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
424 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
428 leaf
= path
->nodes
[0];
429 nritems
= btrfs_header_nritems(leaf
);
432 if (btrfs_fs_closing(fs_info
) > 1) {
437 if (path
->slots
[0] < nritems
) {
438 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
440 ret
= find_next_key(path
, 0, &key
);
444 if (need_resched() ||
445 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
447 caching_ctl
->progress
= last
;
448 btrfs_release_path(path
);
449 up_read(&fs_info
->commit_root_sem
);
450 mutex_unlock(&caching_ctl
->mutex
);
452 mutex_lock(&caching_ctl
->mutex
);
453 down_read(&fs_info
->commit_root_sem
);
457 ret
= btrfs_next_leaf(extent_root
, path
);
462 leaf
= path
->nodes
[0];
463 nritems
= btrfs_header_nritems(leaf
);
467 if (key
.objectid
< last
) {
470 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
473 caching_ctl
->progress
= last
;
474 btrfs_release_path(path
);
478 if (key
.objectid
< block_group
->key
.objectid
) {
483 if (key
.objectid
>= block_group
->key
.objectid
+
484 block_group
->key
.offset
)
487 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
488 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
489 total_found
+= add_new_free_space(block_group
, last
,
491 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
492 last
= key
.objectid
+
495 last
= key
.objectid
+ key
.offset
;
497 if (total_found
> CACHING_CTL_WAKE_UP
) {
500 wake_up(&caching_ctl
->wait
);
507 total_found
+= add_new_free_space(block_group
, last
,
508 block_group
->key
.objectid
+
509 block_group
->key
.offset
);
510 caching_ctl
->progress
= (u64
)-1;
513 btrfs_free_path(path
);
517 static noinline
void caching_thread(struct btrfs_work
*work
)
519 struct btrfs_block_group_cache
*block_group
;
520 struct btrfs_fs_info
*fs_info
;
521 struct btrfs_caching_control
*caching_ctl
;
524 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
525 block_group
= caching_ctl
->block_group
;
526 fs_info
= block_group
->fs_info
;
528 mutex_lock(&caching_ctl
->mutex
);
529 down_read(&fs_info
->commit_root_sem
);
531 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
532 ret
= load_free_space_tree(caching_ctl
);
534 ret
= load_extent_tree_free(caching_ctl
);
536 spin_lock(&block_group
->lock
);
537 block_group
->caching_ctl
= NULL
;
538 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
539 spin_unlock(&block_group
->lock
);
541 #ifdef CONFIG_BTRFS_DEBUG
542 if (btrfs_should_fragment_free_space(block_group
)) {
545 spin_lock(&block_group
->space_info
->lock
);
546 spin_lock(&block_group
->lock
);
547 bytes_used
= block_group
->key
.offset
-
548 btrfs_block_group_used(&block_group
->item
);
549 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
550 spin_unlock(&block_group
->lock
);
551 spin_unlock(&block_group
->space_info
->lock
);
552 fragment_free_space(block_group
);
556 caching_ctl
->progress
= (u64
)-1;
558 up_read(&fs_info
->commit_root_sem
);
559 free_excluded_extents(fs_info
, block_group
);
560 mutex_unlock(&caching_ctl
->mutex
);
562 wake_up(&caching_ctl
->wait
);
564 put_caching_control(caching_ctl
);
565 btrfs_put_block_group(block_group
);
568 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
572 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
573 struct btrfs_caching_control
*caching_ctl
;
576 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
580 INIT_LIST_HEAD(&caching_ctl
->list
);
581 mutex_init(&caching_ctl
->mutex
);
582 init_waitqueue_head(&caching_ctl
->wait
);
583 caching_ctl
->block_group
= cache
;
584 caching_ctl
->progress
= cache
->key
.objectid
;
585 refcount_set(&caching_ctl
->count
, 1);
586 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
587 caching_thread
, NULL
, NULL
);
589 spin_lock(&cache
->lock
);
591 * This should be a rare occasion, but this could happen I think in the
592 * case where one thread starts to load the space cache info, and then
593 * some other thread starts a transaction commit which tries to do an
594 * allocation while the other thread is still loading the space cache
595 * info. The previous loop should have kept us from choosing this block
596 * group, but if we've moved to the state where we will wait on caching
597 * block groups we need to first check if we're doing a fast load here,
598 * so we can wait for it to finish, otherwise we could end up allocating
599 * from a block group who's cache gets evicted for one reason or
602 while (cache
->cached
== BTRFS_CACHE_FAST
) {
603 struct btrfs_caching_control
*ctl
;
605 ctl
= cache
->caching_ctl
;
606 refcount_inc(&ctl
->count
);
607 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
608 spin_unlock(&cache
->lock
);
612 finish_wait(&ctl
->wait
, &wait
);
613 put_caching_control(ctl
);
614 spin_lock(&cache
->lock
);
617 if (cache
->cached
!= BTRFS_CACHE_NO
) {
618 spin_unlock(&cache
->lock
);
622 WARN_ON(cache
->caching_ctl
);
623 cache
->caching_ctl
= caching_ctl
;
624 cache
->cached
= BTRFS_CACHE_FAST
;
625 spin_unlock(&cache
->lock
);
627 if (btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
628 mutex_lock(&caching_ctl
->mutex
);
629 ret
= load_free_space_cache(fs_info
, cache
);
631 spin_lock(&cache
->lock
);
633 cache
->caching_ctl
= NULL
;
634 cache
->cached
= BTRFS_CACHE_FINISHED
;
635 cache
->last_byte_to_unpin
= (u64
)-1;
636 caching_ctl
->progress
= (u64
)-1;
638 if (load_cache_only
) {
639 cache
->caching_ctl
= NULL
;
640 cache
->cached
= BTRFS_CACHE_NO
;
642 cache
->cached
= BTRFS_CACHE_STARTED
;
643 cache
->has_caching_ctl
= 1;
646 spin_unlock(&cache
->lock
);
647 #ifdef CONFIG_BTRFS_DEBUG
649 btrfs_should_fragment_free_space(cache
)) {
652 spin_lock(&cache
->space_info
->lock
);
653 spin_lock(&cache
->lock
);
654 bytes_used
= cache
->key
.offset
-
655 btrfs_block_group_used(&cache
->item
);
656 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
657 spin_unlock(&cache
->lock
);
658 spin_unlock(&cache
->space_info
->lock
);
659 fragment_free_space(cache
);
662 mutex_unlock(&caching_ctl
->mutex
);
664 wake_up(&caching_ctl
->wait
);
666 put_caching_control(caching_ctl
);
667 free_excluded_extents(fs_info
, cache
);
672 * We're either using the free space tree or no caching at all.
673 * Set cached to the appropriate value and wakeup any waiters.
675 spin_lock(&cache
->lock
);
676 if (load_cache_only
) {
677 cache
->caching_ctl
= NULL
;
678 cache
->cached
= BTRFS_CACHE_NO
;
680 cache
->cached
= BTRFS_CACHE_STARTED
;
681 cache
->has_caching_ctl
= 1;
683 spin_unlock(&cache
->lock
);
684 wake_up(&caching_ctl
->wait
);
687 if (load_cache_only
) {
688 put_caching_control(caching_ctl
);
692 down_write(&fs_info
->commit_root_sem
);
693 refcount_inc(&caching_ctl
->count
);
694 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
695 up_write(&fs_info
->commit_root_sem
);
697 btrfs_get_block_group(cache
);
699 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
705 * return the block group that starts at or after bytenr
707 static struct btrfs_block_group_cache
*
708 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
710 return block_group_cache_tree_search(info
, bytenr
, 0);
714 * return the block group that contains the given bytenr
716 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
717 struct btrfs_fs_info
*info
,
720 return block_group_cache_tree_search(info
, bytenr
, 1);
723 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
726 struct list_head
*head
= &info
->space_info
;
727 struct btrfs_space_info
*found
;
729 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
732 list_for_each_entry_rcu(found
, head
, list
) {
733 if (found
->flags
& flags
) {
742 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, s64 num_bytes
,
743 bool metadata
, u64 root_objectid
)
745 struct btrfs_space_info
*space_info
;
749 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
750 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
752 flags
= BTRFS_BLOCK_GROUP_METADATA
;
754 flags
= BTRFS_BLOCK_GROUP_DATA
;
757 space_info
= __find_space_info(fs_info
, flags
);
759 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
763 * after adding space to the filesystem, we need to clear the full flags
764 * on all the space infos.
766 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
768 struct list_head
*head
= &info
->space_info
;
769 struct btrfs_space_info
*found
;
772 list_for_each_entry_rcu(found
, head
, list
)
777 /* simple helper to search for an existing data extent at a given offset */
778 int btrfs_lookup_data_extent(struct btrfs_fs_info
*fs_info
, u64 start
, u64 len
)
781 struct btrfs_key key
;
782 struct btrfs_path
*path
;
784 path
= btrfs_alloc_path();
788 key
.objectid
= start
;
790 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
791 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
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_fs_info
*fs_info
, 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(fs_info
, SKINNY_METADATA
)) {
825 offset
= fs_info
->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
, fs_info
->extent_root
, &key
, path
, 0, 0);
850 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
851 if (path
->slots
[0]) {
853 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
855 if (key
.objectid
== bytenr
&&
856 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
857 key
.offset
== fs_info
->nodesize
)
863 leaf
= path
->nodes
[0];
864 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
865 if (item_size
>= sizeof(*ei
)) {
866 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
867 struct btrfs_extent_item
);
868 num_refs
= btrfs_extent_refs(leaf
, ei
);
869 extent_flags
= btrfs_extent_flags(leaf
, ei
);
871 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
872 struct btrfs_extent_item_v0
*ei0
;
873 BUG_ON(item_size
!= sizeof(*ei0
));
874 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
875 struct btrfs_extent_item_v0
);
876 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
877 /* FIXME: this isn't correct for data */
878 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
883 BUG_ON(num_refs
== 0);
893 delayed_refs
= &trans
->transaction
->delayed_refs
;
894 spin_lock(&delayed_refs
->lock
);
895 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
897 if (!mutex_trylock(&head
->mutex
)) {
898 refcount_inc(&head
->refs
);
899 spin_unlock(&delayed_refs
->lock
);
901 btrfs_release_path(path
);
904 * Mutex was contended, block until it's released and try
907 mutex_lock(&head
->mutex
);
908 mutex_unlock(&head
->mutex
);
909 btrfs_put_delayed_ref_head(head
);
912 spin_lock(&head
->lock
);
913 if (head
->extent_op
&& head
->extent_op
->update_flags
)
914 extent_flags
|= head
->extent_op
->flags_to_set
;
916 BUG_ON(num_refs
== 0);
918 num_refs
+= head
->ref_mod
;
919 spin_unlock(&head
->lock
);
920 mutex_unlock(&head
->mutex
);
922 spin_unlock(&delayed_refs
->lock
);
924 WARN_ON(num_refs
== 0);
928 *flags
= extent_flags
;
930 btrfs_free_path(path
);
935 * Back reference rules. Back refs have three main goals:
937 * 1) differentiate between all holders of references to an extent so that
938 * when a reference is dropped we can make sure it was a valid reference
939 * before freeing the extent.
941 * 2) Provide enough information to quickly find the holders of an extent
942 * if we notice a given block is corrupted or bad.
944 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
945 * maintenance. This is actually the same as #2, but with a slightly
946 * different use case.
948 * There are two kinds of back refs. The implicit back refs is optimized
949 * for pointers in non-shared tree blocks. For a given pointer in a block,
950 * back refs of this kind provide information about the block's owner tree
951 * and the pointer's key. These information allow us to find the block by
952 * b-tree searching. The full back refs is for pointers in tree blocks not
953 * referenced by their owner trees. The location of tree block is recorded
954 * in the back refs. Actually the full back refs is generic, and can be
955 * used in all cases the implicit back refs is used. The major shortcoming
956 * of the full back refs is its overhead. Every time a tree block gets
957 * COWed, we have to update back refs entry for all pointers in it.
959 * For a newly allocated tree block, we use implicit back refs for
960 * pointers in it. This means most tree related operations only involve
961 * implicit back refs. For a tree block created in old transaction, the
962 * only way to drop a reference to it is COW it. So we can detect the
963 * event that tree block loses its owner tree's reference and do the
964 * back refs conversion.
966 * When a tree block is COWed through a tree, there are four cases:
968 * The reference count of the block is one and the tree is the block's
969 * owner tree. Nothing to do in this case.
971 * The reference count of the block is one and the tree is not the
972 * block's owner tree. In this case, full back refs is used for pointers
973 * in the block. Remove these full back refs, add implicit back refs for
974 * every pointers in the new block.
976 * The reference count of the block is greater than one and the tree is
977 * the block's owner tree. In this case, implicit back refs is used for
978 * pointers in the block. Add full back refs for every pointers in the
979 * block, increase lower level extents' reference counts. The original
980 * implicit back refs are entailed to the new block.
982 * The reference count of the block is greater than one and the tree is
983 * not the block's owner tree. Add implicit back refs for every pointer in
984 * the new block, increase lower level extents' reference count.
986 * Back Reference Key composing:
988 * The key objectid corresponds to the first byte in the extent,
989 * The key type is used to differentiate between types of back refs.
990 * There are different meanings of the key offset for different types
993 * File extents can be referenced by:
995 * - multiple snapshots, subvolumes, or different generations in one subvol
996 * - different files inside a single subvolume
997 * - different offsets inside a file (bookend extents in file.c)
999 * The extent ref structure for the implicit back refs has fields for:
1001 * - Objectid of the subvolume root
1002 * - objectid of the file holding the reference
1003 * - original offset in the file
1004 * - how many bookend extents
1006 * The key offset for the implicit back refs is hash of the first
1009 * The extent ref structure for the full back refs has field for:
1011 * - number of pointers in the tree leaf
1013 * The key offset for the implicit back refs is the first byte of
1016 * When a file extent is allocated, The implicit back refs is used.
1017 * the fields are filled in:
1019 * (root_key.objectid, inode objectid, offset in file, 1)
1021 * When a file extent is removed file truncation, we find the
1022 * corresponding implicit back refs and check the following fields:
1024 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1026 * Btree extents can be referenced by:
1028 * - Different subvolumes
1030 * Both the implicit back refs and the full back refs for tree blocks
1031 * only consist of key. The key offset for the implicit back refs is
1032 * objectid of block's owner tree. The key offset for the full back refs
1033 * is the first byte of parent block.
1035 * When implicit back refs is used, information about the lowest key and
1036 * level of the tree block are required. These information are stored in
1037 * tree block info structure.
1040 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1041 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1042 struct btrfs_fs_info
*fs_info
,
1043 struct btrfs_path
*path
,
1044 u64 owner
, u32 extra_size
)
1046 struct btrfs_root
*root
= fs_info
->extent_root
;
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(fs_info
, 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 memzero_extent_buffer(leaf
, (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
);
1124 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1125 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1126 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1128 int btrfs_get_extent_inline_ref_type(const struct extent_buffer
*eb
,
1129 struct btrfs_extent_inline_ref
*iref
,
1130 enum btrfs_inline_ref_type is_data
)
1132 int type
= btrfs_extent_inline_ref_type(eb
, iref
);
1133 u64 offset
= btrfs_extent_inline_ref_offset(eb
, iref
);
1135 if (type
== BTRFS_TREE_BLOCK_REF_KEY
||
1136 type
== BTRFS_SHARED_BLOCK_REF_KEY
||
1137 type
== BTRFS_SHARED_DATA_REF_KEY
||
1138 type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1139 if (is_data
== BTRFS_REF_TYPE_BLOCK
) {
1140 if (type
== BTRFS_TREE_BLOCK_REF_KEY
)
1142 if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1143 ASSERT(eb
->fs_info
);
1145 * Every shared one has parent tree
1146 * block, which must be aligned to
1150 IS_ALIGNED(offset
, eb
->fs_info
->nodesize
))
1153 } else if (is_data
== BTRFS_REF_TYPE_DATA
) {
1154 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1156 if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1157 ASSERT(eb
->fs_info
);
1159 * Every shared one has parent tree
1160 * block, which must be aligned to
1164 IS_ALIGNED(offset
, eb
->fs_info
->nodesize
))
1168 ASSERT(is_data
== BTRFS_REF_TYPE_ANY
);
1173 btrfs_print_leaf((struct extent_buffer
*)eb
);
1174 btrfs_err(eb
->fs_info
, "eb %llu invalid extent inline ref type %d",
1178 return BTRFS_REF_TYPE_INVALID
;
1181 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1183 u32 high_crc
= ~(u32
)0;
1184 u32 low_crc
= ~(u32
)0;
1187 lenum
= cpu_to_le64(root_objectid
);
1188 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1189 lenum
= cpu_to_le64(owner
);
1190 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1191 lenum
= cpu_to_le64(offset
);
1192 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1194 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1197 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1198 struct btrfs_extent_data_ref
*ref
)
1200 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1201 btrfs_extent_data_ref_objectid(leaf
, ref
),
1202 btrfs_extent_data_ref_offset(leaf
, ref
));
1205 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1206 struct btrfs_extent_data_ref
*ref
,
1207 u64 root_objectid
, u64 owner
, u64 offset
)
1209 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1210 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1211 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1216 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1217 struct btrfs_path
*path
,
1218 u64 bytenr
, u64 parent
,
1220 u64 owner
, u64 offset
)
1222 struct btrfs_root
*root
= trans
->fs_info
->extent_root
;
1223 struct btrfs_key key
;
1224 struct btrfs_extent_data_ref
*ref
;
1225 struct extent_buffer
*leaf
;
1231 key
.objectid
= bytenr
;
1233 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1234 key
.offset
= parent
;
1236 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1237 key
.offset
= hash_extent_data_ref(root_objectid
,
1242 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1251 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1252 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1253 btrfs_release_path(path
);
1254 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1265 leaf
= path
->nodes
[0];
1266 nritems
= btrfs_header_nritems(leaf
);
1268 if (path
->slots
[0] >= nritems
) {
1269 ret
= btrfs_next_leaf(root
, path
);
1275 leaf
= path
->nodes
[0];
1276 nritems
= btrfs_header_nritems(leaf
);
1280 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1281 if (key
.objectid
!= bytenr
||
1282 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1285 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1286 struct btrfs_extent_data_ref
);
1288 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1291 btrfs_release_path(path
);
1303 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1304 struct btrfs_path
*path
,
1305 u64 bytenr
, u64 parent
,
1306 u64 root_objectid
, u64 owner
,
1307 u64 offset
, int refs_to_add
)
1309 struct btrfs_root
*root
= trans
->fs_info
->extent_root
;
1310 struct btrfs_key key
;
1311 struct extent_buffer
*leaf
;
1316 key
.objectid
= bytenr
;
1318 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1319 key
.offset
= parent
;
1320 size
= sizeof(struct btrfs_shared_data_ref
);
1322 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1323 key
.offset
= hash_extent_data_ref(root_objectid
,
1325 size
= sizeof(struct btrfs_extent_data_ref
);
1328 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1329 if (ret
&& ret
!= -EEXIST
)
1332 leaf
= path
->nodes
[0];
1334 struct btrfs_shared_data_ref
*ref
;
1335 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1336 struct btrfs_shared_data_ref
);
1338 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1340 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1341 num_refs
+= refs_to_add
;
1342 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1345 struct btrfs_extent_data_ref
*ref
;
1346 while (ret
== -EEXIST
) {
1347 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1348 struct btrfs_extent_data_ref
);
1349 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1352 btrfs_release_path(path
);
1354 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1356 if (ret
&& ret
!= -EEXIST
)
1359 leaf
= path
->nodes
[0];
1361 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1362 struct btrfs_extent_data_ref
);
1364 btrfs_set_extent_data_ref_root(leaf
, ref
,
1366 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1367 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1368 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1370 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1371 num_refs
+= refs_to_add
;
1372 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1375 btrfs_mark_buffer_dirty(leaf
);
1378 btrfs_release_path(path
);
1382 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1383 struct btrfs_path
*path
,
1384 int refs_to_drop
, int *last_ref
)
1386 struct btrfs_key key
;
1387 struct btrfs_extent_data_ref
*ref1
= NULL
;
1388 struct btrfs_shared_data_ref
*ref2
= NULL
;
1389 struct extent_buffer
*leaf
;
1393 leaf
= path
->nodes
[0];
1394 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1396 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1397 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1398 struct btrfs_extent_data_ref
);
1399 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1400 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1401 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1402 struct btrfs_shared_data_ref
);
1403 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1404 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1405 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1406 struct btrfs_extent_ref_v0
*ref0
;
1407 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1408 struct btrfs_extent_ref_v0
);
1409 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1415 BUG_ON(num_refs
< refs_to_drop
);
1416 num_refs
-= refs_to_drop
;
1418 if (num_refs
== 0) {
1419 ret
= btrfs_del_item(trans
, trans
->fs_info
->extent_root
, path
);
1422 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1423 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1424 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1425 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1426 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1428 struct btrfs_extent_ref_v0
*ref0
;
1429 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1430 struct btrfs_extent_ref_v0
);
1431 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1434 btrfs_mark_buffer_dirty(leaf
);
1439 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1440 struct btrfs_extent_inline_ref
*iref
)
1442 struct btrfs_key key
;
1443 struct extent_buffer
*leaf
;
1444 struct btrfs_extent_data_ref
*ref1
;
1445 struct btrfs_shared_data_ref
*ref2
;
1449 leaf
= path
->nodes
[0];
1450 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1453 * If type is invalid, we should have bailed out earlier than
1456 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
1457 ASSERT(type
!= BTRFS_REF_TYPE_INVALID
);
1458 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1459 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1460 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1462 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1463 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1465 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1466 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1467 struct btrfs_extent_data_ref
);
1468 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1469 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1470 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1471 struct btrfs_shared_data_ref
);
1472 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1473 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1474 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1475 struct btrfs_extent_ref_v0
*ref0
;
1476 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1477 struct btrfs_extent_ref_v0
);
1478 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1486 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1487 struct btrfs_path
*path
,
1488 u64 bytenr
, u64 parent
,
1491 struct btrfs_root
*root
= trans
->fs_info
->extent_root
;
1492 struct btrfs_key key
;
1495 key
.objectid
= bytenr
;
1497 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1498 key
.offset
= parent
;
1500 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1501 key
.offset
= root_objectid
;
1504 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1507 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1508 if (ret
== -ENOENT
&& parent
) {
1509 btrfs_release_path(path
);
1510 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1511 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1519 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1520 struct btrfs_path
*path
,
1521 u64 bytenr
, u64 parent
,
1524 struct btrfs_key key
;
1527 key
.objectid
= bytenr
;
1529 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1530 key
.offset
= parent
;
1532 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1533 key
.offset
= root_objectid
;
1536 ret
= btrfs_insert_empty_item(trans
, trans
->fs_info
->extent_root
,
1538 btrfs_release_path(path
);
1542 static inline int extent_ref_type(u64 parent
, u64 owner
)
1545 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1547 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1549 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1552 type
= BTRFS_SHARED_DATA_REF_KEY
;
1554 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1559 static int find_next_key(struct btrfs_path
*path
, int level
,
1560 struct btrfs_key
*key
)
1563 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1564 if (!path
->nodes
[level
])
1566 if (path
->slots
[level
] + 1 >=
1567 btrfs_header_nritems(path
->nodes
[level
]))
1570 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1571 path
->slots
[level
] + 1);
1573 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1574 path
->slots
[level
] + 1);
1581 * look for inline back ref. if back ref is found, *ref_ret is set
1582 * to the address of inline back ref, and 0 is returned.
1584 * if back ref isn't found, *ref_ret is set to the address where it
1585 * should be inserted, and -ENOENT is returned.
1587 * if insert is true and there are too many inline back refs, the path
1588 * points to the extent item, and -EAGAIN is returned.
1590 * NOTE: inline back refs are ordered in the same way that back ref
1591 * items in the tree are ordered.
1593 static noinline_for_stack
1594 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1595 struct btrfs_path
*path
,
1596 struct btrfs_extent_inline_ref
**ref_ret
,
1597 u64 bytenr
, u64 num_bytes
,
1598 u64 parent
, u64 root_objectid
,
1599 u64 owner
, u64 offset
, int insert
)
1601 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1602 struct btrfs_root
*root
= fs_info
->extent_root
;
1603 struct btrfs_key key
;
1604 struct extent_buffer
*leaf
;
1605 struct btrfs_extent_item
*ei
;
1606 struct btrfs_extent_inline_ref
*iref
;
1616 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
1619 key
.objectid
= bytenr
;
1620 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1621 key
.offset
= num_bytes
;
1623 want
= extent_ref_type(parent
, owner
);
1625 extra_size
= btrfs_extent_inline_ref_size(want
);
1626 path
->keep_locks
= 1;
1631 * Owner is our level, so we can just add one to get the level for the
1632 * block we are interested in.
1634 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1635 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1640 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1647 * We may be a newly converted file system which still has the old fat
1648 * extent entries for metadata, so try and see if we have one of those.
1650 if (ret
> 0 && skinny_metadata
) {
1651 skinny_metadata
= false;
1652 if (path
->slots
[0]) {
1654 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1656 if (key
.objectid
== bytenr
&&
1657 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1658 key
.offset
== num_bytes
)
1662 key
.objectid
= bytenr
;
1663 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1664 key
.offset
= num_bytes
;
1665 btrfs_release_path(path
);
1670 if (ret
&& !insert
) {
1673 } else if (WARN_ON(ret
)) {
1678 leaf
= path
->nodes
[0];
1679 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1680 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1681 if (item_size
< sizeof(*ei
)) {
1686 ret
= convert_extent_item_v0(trans
, fs_info
, path
, owner
,
1692 leaf
= path
->nodes
[0];
1693 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1696 BUG_ON(item_size
< sizeof(*ei
));
1698 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1699 flags
= btrfs_extent_flags(leaf
, ei
);
1701 ptr
= (unsigned long)(ei
+ 1);
1702 end
= (unsigned long)ei
+ item_size
;
1704 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1705 ptr
+= sizeof(struct btrfs_tree_block_info
);
1709 if (owner
>= BTRFS_FIRST_FREE_OBJECTID
)
1710 needed
= BTRFS_REF_TYPE_DATA
;
1712 needed
= BTRFS_REF_TYPE_BLOCK
;
1720 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1721 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, needed
);
1722 if (type
== BTRFS_REF_TYPE_INVALID
) {
1730 ptr
+= btrfs_extent_inline_ref_size(type
);
1734 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1735 struct btrfs_extent_data_ref
*dref
;
1736 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1737 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1742 if (hash_extent_data_ref_item(leaf
, dref
) <
1743 hash_extent_data_ref(root_objectid
, owner
, offset
))
1747 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1749 if (parent
== ref_offset
) {
1753 if (ref_offset
< parent
)
1756 if (root_objectid
== ref_offset
) {
1760 if (ref_offset
< root_objectid
)
1764 ptr
+= btrfs_extent_inline_ref_size(type
);
1766 if (err
== -ENOENT
&& insert
) {
1767 if (item_size
+ extra_size
>=
1768 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1773 * To add new inline back ref, we have to make sure
1774 * there is no corresponding back ref item.
1775 * For simplicity, we just do not add new inline back
1776 * ref if there is any kind of item for this block
1778 if (find_next_key(path
, 0, &key
) == 0 &&
1779 key
.objectid
== bytenr
&&
1780 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1785 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1788 path
->keep_locks
= 0;
1789 btrfs_unlock_up_safe(path
, 1);
1795 * helper to add new inline back ref
1797 static noinline_for_stack
1798 void setup_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1799 struct btrfs_path
*path
,
1800 struct btrfs_extent_inline_ref
*iref
,
1801 u64 parent
, u64 root_objectid
,
1802 u64 owner
, u64 offset
, int refs_to_add
,
1803 struct btrfs_delayed_extent_op
*extent_op
)
1805 struct extent_buffer
*leaf
;
1806 struct btrfs_extent_item
*ei
;
1809 unsigned long item_offset
;
1814 leaf
= path
->nodes
[0];
1815 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1816 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1818 type
= extent_ref_type(parent
, owner
);
1819 size
= btrfs_extent_inline_ref_size(type
);
1821 btrfs_extend_item(fs_info
, path
, size
);
1823 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1824 refs
= btrfs_extent_refs(leaf
, ei
);
1825 refs
+= refs_to_add
;
1826 btrfs_set_extent_refs(leaf
, ei
, refs
);
1828 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1830 ptr
= (unsigned long)ei
+ item_offset
;
1831 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1832 if (ptr
< end
- size
)
1833 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1836 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1837 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1838 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1839 struct btrfs_extent_data_ref
*dref
;
1840 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1841 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1842 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1843 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1844 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1845 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1846 struct btrfs_shared_data_ref
*sref
;
1847 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1848 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1849 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1850 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1851 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1853 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1855 btrfs_mark_buffer_dirty(leaf
);
1858 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1859 struct btrfs_path
*path
,
1860 struct btrfs_extent_inline_ref
**ref_ret
,
1861 u64 bytenr
, u64 num_bytes
, u64 parent
,
1862 u64 root_objectid
, u64 owner
, u64 offset
)
1866 ret
= lookup_inline_extent_backref(trans
, path
, ref_ret
, bytenr
,
1867 num_bytes
, parent
, root_objectid
,
1872 btrfs_release_path(path
);
1875 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1876 ret
= lookup_tree_block_ref(trans
, path
, bytenr
, parent
,
1879 ret
= lookup_extent_data_ref(trans
, path
, bytenr
, parent
,
1880 root_objectid
, owner
, offset
);
1886 * helper to update/remove inline back ref
1888 static noinline_for_stack
1889 void update_inline_extent_backref(struct btrfs_path
*path
,
1890 struct btrfs_extent_inline_ref
*iref
,
1892 struct btrfs_delayed_extent_op
*extent_op
,
1895 struct extent_buffer
*leaf
= path
->nodes
[0];
1896 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
1897 struct btrfs_extent_item
*ei
;
1898 struct btrfs_extent_data_ref
*dref
= NULL
;
1899 struct btrfs_shared_data_ref
*sref
= NULL
;
1907 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1908 refs
= btrfs_extent_refs(leaf
, ei
);
1909 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1910 refs
+= refs_to_mod
;
1911 btrfs_set_extent_refs(leaf
, ei
, refs
);
1913 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1916 * If type is invalid, we should have bailed out after
1917 * lookup_inline_extent_backref().
1919 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_ANY
);
1920 ASSERT(type
!= BTRFS_REF_TYPE_INVALID
);
1922 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1923 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1924 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1925 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1926 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1927 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1930 BUG_ON(refs_to_mod
!= -1);
1933 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1934 refs
+= refs_to_mod
;
1937 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1938 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1940 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1943 size
= btrfs_extent_inline_ref_size(type
);
1944 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1945 ptr
= (unsigned long)iref
;
1946 end
= (unsigned long)ei
+ item_size
;
1947 if (ptr
+ size
< end
)
1948 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1951 btrfs_truncate_item(fs_info
, path
, item_size
, 1);
1953 btrfs_mark_buffer_dirty(leaf
);
1956 static noinline_for_stack
1957 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1958 struct btrfs_fs_info
*fs_info
,
1959 struct btrfs_path
*path
,
1960 u64 bytenr
, u64 num_bytes
, u64 parent
,
1961 u64 root_objectid
, u64 owner
,
1962 u64 offset
, int refs_to_add
,
1963 struct btrfs_delayed_extent_op
*extent_op
)
1965 struct btrfs_extent_inline_ref
*iref
;
1968 ret
= lookup_inline_extent_backref(trans
, path
, &iref
, bytenr
,
1969 num_bytes
, parent
, root_objectid
,
1972 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1973 update_inline_extent_backref(path
, iref
, refs_to_add
,
1975 } else if (ret
== -ENOENT
) {
1976 setup_inline_extent_backref(fs_info
, path
, iref
, parent
,
1977 root_objectid
, owner
, offset
,
1978 refs_to_add
, extent_op
);
1984 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1985 struct btrfs_path
*path
,
1986 u64 bytenr
, u64 parent
, u64 root_objectid
,
1987 u64 owner
, u64 offset
, int refs_to_add
)
1990 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1991 BUG_ON(refs_to_add
!= 1);
1992 ret
= insert_tree_block_ref(trans
, path
, bytenr
, parent
,
1995 ret
= insert_extent_data_ref(trans
, path
, bytenr
, parent
,
1996 root_objectid
, owner
, offset
,
2002 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
2003 struct btrfs_fs_info
*fs_info
,
2004 struct btrfs_path
*path
,
2005 struct btrfs_extent_inline_ref
*iref
,
2006 int refs_to_drop
, int is_data
, int *last_ref
)
2010 BUG_ON(!is_data
&& refs_to_drop
!= 1);
2012 update_inline_extent_backref(path
, iref
, -refs_to_drop
, NULL
,
2014 } else if (is_data
) {
2015 ret
= remove_extent_data_ref(trans
, path
, refs_to_drop
,
2019 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
2024 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2025 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
2026 u64
*discarded_bytes
)
2029 u64 bytes_left
, end
;
2030 u64 aligned_start
= ALIGN(start
, 1 << 9);
2032 if (WARN_ON(start
!= aligned_start
)) {
2033 len
-= aligned_start
- start
;
2034 len
= round_down(len
, 1 << 9);
2035 start
= aligned_start
;
2038 *discarded_bytes
= 0;
2046 /* Skip any superblocks on this device. */
2047 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
2048 u64 sb_start
= btrfs_sb_offset(j
);
2049 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
2050 u64 size
= sb_start
- start
;
2052 if (!in_range(sb_start
, start
, bytes_left
) &&
2053 !in_range(sb_end
, start
, bytes_left
) &&
2054 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
2058 * Superblock spans beginning of range. Adjust start and
2061 if (sb_start
<= start
) {
2062 start
+= sb_end
- start
;
2067 bytes_left
= end
- start
;
2072 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2075 *discarded_bytes
+= size
;
2076 else if (ret
!= -EOPNOTSUPP
)
2085 bytes_left
= end
- start
;
2089 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2092 *discarded_bytes
+= bytes_left
;
2097 int btrfs_discard_extent(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
2098 u64 num_bytes
, u64
*actual_bytes
)
2101 u64 discarded_bytes
= 0;
2102 struct btrfs_bio
*bbio
= NULL
;
2106 * Avoid races with device replace and make sure our bbio has devices
2107 * associated to its stripes that don't go away while we are discarding.
2109 btrfs_bio_counter_inc_blocked(fs_info
);
2110 /* Tell the block device(s) that the sectors can be discarded */
2111 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_DISCARD
, bytenr
, &num_bytes
,
2113 /* Error condition is -ENOMEM */
2115 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2119 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2121 struct request_queue
*req_q
;
2123 if (!stripe
->dev
->bdev
) {
2124 ASSERT(btrfs_test_opt(fs_info
, DEGRADED
));
2127 req_q
= bdev_get_queue(stripe
->dev
->bdev
);
2128 if (!blk_queue_discard(req_q
))
2131 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2136 discarded_bytes
+= bytes
;
2137 else if (ret
!= -EOPNOTSUPP
)
2138 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2141 * Just in case we get back EOPNOTSUPP for some reason,
2142 * just ignore the return value so we don't screw up
2143 * people calling discard_extent.
2147 btrfs_put_bbio(bbio
);
2149 btrfs_bio_counter_dec(fs_info
);
2152 *actual_bytes
= discarded_bytes
;
2155 if (ret
== -EOPNOTSUPP
)
2160 /* Can return -ENOMEM */
2161 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2162 struct btrfs_root
*root
,
2163 u64 bytenr
, u64 num_bytes
, u64 parent
,
2164 u64 root_objectid
, u64 owner
, u64 offset
)
2166 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2167 int old_ref_mod
, new_ref_mod
;
2170 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2171 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2173 btrfs_ref_tree_mod(root
, bytenr
, num_bytes
, parent
, root_objectid
,
2174 owner
, offset
, BTRFS_ADD_DELAYED_REF
);
2176 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2177 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
,
2179 root_objectid
, (int)owner
,
2180 BTRFS_ADD_DELAYED_REF
, NULL
,
2181 &old_ref_mod
, &new_ref_mod
);
2183 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
,
2185 root_objectid
, owner
, offset
,
2186 0, BTRFS_ADD_DELAYED_REF
,
2187 &old_ref_mod
, &new_ref_mod
);
2190 if (ret
== 0 && old_ref_mod
< 0 && new_ref_mod
>= 0) {
2191 bool metadata
= owner
< BTRFS_FIRST_FREE_OBJECTID
;
2193 add_pinned_bytes(fs_info
, -num_bytes
, metadata
, root_objectid
);
2200 * __btrfs_inc_extent_ref - insert backreference for a given extent
2202 * @trans: Handle of transaction
2204 * @node: The delayed ref node used to get the bytenr/length for
2205 * extent whose references are incremented.
2207 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2208 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2209 * bytenr of the parent block. Since new extents are always
2210 * created with indirect references, this will only be the case
2211 * when relocating a shared extent. In that case, root_objectid
2212 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2215 * @root_objectid: The id of the root where this modification has originated,
2216 * this can be either one of the well-known metadata trees or
2217 * the subvolume id which references this extent.
2219 * @owner: For data extents it is the inode number of the owning file.
2220 * For metadata extents this parameter holds the level in the
2221 * tree of the extent.
2223 * @offset: For metadata extents the offset is ignored and is currently
2224 * always passed as 0. For data extents it is the fileoffset
2225 * this extent belongs to.
2227 * @refs_to_add Number of references to add
2229 * @extent_op Pointer to a structure, holding information necessary when
2230 * updating a tree block's flags
2233 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2234 struct btrfs_delayed_ref_node
*node
,
2235 u64 parent
, u64 root_objectid
,
2236 u64 owner
, u64 offset
, int refs_to_add
,
2237 struct btrfs_delayed_extent_op
*extent_op
)
2239 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2240 struct btrfs_path
*path
;
2241 struct extent_buffer
*leaf
;
2242 struct btrfs_extent_item
*item
;
2243 struct btrfs_key key
;
2244 u64 bytenr
= node
->bytenr
;
2245 u64 num_bytes
= node
->num_bytes
;
2249 path
= btrfs_alloc_path();
2253 path
->reada
= READA_FORWARD
;
2254 path
->leave_spinning
= 1;
2255 /* this will setup the path even if it fails to insert the back ref */
2256 ret
= insert_inline_extent_backref(trans
, fs_info
, path
, bytenr
,
2257 num_bytes
, parent
, root_objectid
,
2259 refs_to_add
, extent_op
);
2260 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2264 * Ok we had -EAGAIN which means we didn't have space to insert and
2265 * inline extent ref, so just update the reference count and add a
2268 leaf
= path
->nodes
[0];
2269 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2270 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2271 refs
= btrfs_extent_refs(leaf
, item
);
2272 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2274 __run_delayed_extent_op(extent_op
, leaf
, item
);
2276 btrfs_mark_buffer_dirty(leaf
);
2277 btrfs_release_path(path
);
2279 path
->reada
= READA_FORWARD
;
2280 path
->leave_spinning
= 1;
2281 /* now insert the actual backref */
2282 ret
= insert_extent_backref(trans
, path
, bytenr
, parent
, root_objectid
,
2283 owner
, offset
, refs_to_add
);
2285 btrfs_abort_transaction(trans
, ret
);
2287 btrfs_free_path(path
);
2291 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2292 struct btrfs_delayed_ref_node
*node
,
2293 struct btrfs_delayed_extent_op
*extent_op
,
2294 int insert_reserved
)
2297 struct btrfs_delayed_data_ref
*ref
;
2298 struct btrfs_key ins
;
2303 ins
.objectid
= node
->bytenr
;
2304 ins
.offset
= node
->num_bytes
;
2305 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2307 ref
= btrfs_delayed_node_to_data_ref(node
);
2308 trace_run_delayed_data_ref(trans
->fs_info
, node
, ref
, node
->action
);
2310 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2311 parent
= ref
->parent
;
2312 ref_root
= ref
->root
;
2314 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2316 flags
|= extent_op
->flags_to_set
;
2317 ret
= alloc_reserved_file_extent(trans
, parent
, ref_root
,
2318 flags
, ref
->objectid
,
2321 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2322 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
2323 ref
->objectid
, ref
->offset
,
2324 node
->ref_mod
, extent_op
);
2325 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2326 ret
= __btrfs_free_extent(trans
, node
, parent
,
2327 ref_root
, ref
->objectid
,
2328 ref
->offset
, node
->ref_mod
,
2336 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2337 struct extent_buffer
*leaf
,
2338 struct btrfs_extent_item
*ei
)
2340 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2341 if (extent_op
->update_flags
) {
2342 flags
|= extent_op
->flags_to_set
;
2343 btrfs_set_extent_flags(leaf
, ei
, flags
);
2346 if (extent_op
->update_key
) {
2347 struct btrfs_tree_block_info
*bi
;
2348 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2349 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2350 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2354 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2355 struct btrfs_delayed_ref_head
*head
,
2356 struct btrfs_delayed_extent_op
*extent_op
)
2358 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2359 struct btrfs_key key
;
2360 struct btrfs_path
*path
;
2361 struct btrfs_extent_item
*ei
;
2362 struct extent_buffer
*leaf
;
2366 int metadata
= !extent_op
->is_data
;
2371 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2374 path
= btrfs_alloc_path();
2378 key
.objectid
= head
->bytenr
;
2381 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2382 key
.offset
= extent_op
->level
;
2384 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2385 key
.offset
= head
->num_bytes
;
2389 path
->reada
= READA_FORWARD
;
2390 path
->leave_spinning
= 1;
2391 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 1);
2398 if (path
->slots
[0] > 0) {
2400 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2402 if (key
.objectid
== head
->bytenr
&&
2403 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2404 key
.offset
== head
->num_bytes
)
2408 btrfs_release_path(path
);
2411 key
.objectid
= head
->bytenr
;
2412 key
.offset
= head
->num_bytes
;
2413 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2422 leaf
= path
->nodes
[0];
2423 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2424 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2425 if (item_size
< sizeof(*ei
)) {
2426 ret
= convert_extent_item_v0(trans
, fs_info
, path
, (u64
)-1, 0);
2431 leaf
= path
->nodes
[0];
2432 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2435 BUG_ON(item_size
< sizeof(*ei
));
2436 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2437 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2439 btrfs_mark_buffer_dirty(leaf
);
2441 btrfs_free_path(path
);
2445 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2446 struct btrfs_fs_info
*fs_info
,
2447 struct btrfs_delayed_ref_node
*node
,
2448 struct btrfs_delayed_extent_op
*extent_op
,
2449 int insert_reserved
)
2452 struct btrfs_delayed_tree_ref
*ref
;
2456 ref
= btrfs_delayed_node_to_tree_ref(node
);
2457 trace_run_delayed_tree_ref(fs_info
, node
, ref
, node
->action
);
2459 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2460 parent
= ref
->parent
;
2461 ref_root
= ref
->root
;
2463 if (node
->ref_mod
!= 1) {
2465 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2466 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
2470 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2471 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2472 ret
= alloc_reserved_tree_block(trans
, node
, extent_op
);
2473 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2474 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
2475 ref
->level
, 0, 1, extent_op
);
2476 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2477 ret
= __btrfs_free_extent(trans
, node
, parent
, ref_root
,
2478 ref
->level
, 0, 1, extent_op
);
2485 /* helper function to actually process a single delayed ref entry */
2486 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2487 struct btrfs_fs_info
*fs_info
,
2488 struct btrfs_delayed_ref_node
*node
,
2489 struct btrfs_delayed_extent_op
*extent_op
,
2490 int insert_reserved
)
2494 if (trans
->aborted
) {
2495 if (insert_reserved
)
2496 btrfs_pin_extent(fs_info
, node
->bytenr
,
2497 node
->num_bytes
, 1);
2501 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2502 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2503 ret
= run_delayed_tree_ref(trans
, fs_info
, node
, extent_op
,
2505 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2506 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2507 ret
= run_delayed_data_ref(trans
, node
, extent_op
,
2514 static inline struct btrfs_delayed_ref_node
*
2515 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2517 struct btrfs_delayed_ref_node
*ref
;
2519 if (RB_EMPTY_ROOT(&head
->ref_tree
))
2523 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2524 * This is to prevent a ref count from going down to zero, which deletes
2525 * the extent item from the extent tree, when there still are references
2526 * to add, which would fail because they would not find the extent item.
2528 if (!list_empty(&head
->ref_add_list
))
2529 return list_first_entry(&head
->ref_add_list
,
2530 struct btrfs_delayed_ref_node
, add_list
);
2532 ref
= rb_entry(rb_first(&head
->ref_tree
),
2533 struct btrfs_delayed_ref_node
, ref_node
);
2534 ASSERT(list_empty(&ref
->add_list
));
2538 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root
*delayed_refs
,
2539 struct btrfs_delayed_ref_head
*head
)
2541 spin_lock(&delayed_refs
->lock
);
2542 head
->processing
= 0;
2543 delayed_refs
->num_heads_ready
++;
2544 spin_unlock(&delayed_refs
->lock
);
2545 btrfs_delayed_ref_unlock(head
);
2548 static int cleanup_extent_op(struct btrfs_trans_handle
*trans
,
2549 struct btrfs_delayed_ref_head
*head
)
2551 struct btrfs_delayed_extent_op
*extent_op
= head
->extent_op
;
2556 head
->extent_op
= NULL
;
2557 if (head
->must_insert_reserved
) {
2558 btrfs_free_delayed_extent_op(extent_op
);
2561 spin_unlock(&head
->lock
);
2562 ret
= run_delayed_extent_op(trans
, head
, extent_op
);
2563 btrfs_free_delayed_extent_op(extent_op
);
2564 return ret
? ret
: 1;
2567 static int cleanup_ref_head(struct btrfs_trans_handle
*trans
,
2568 struct btrfs_delayed_ref_head
*head
)
2571 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2572 struct btrfs_delayed_ref_root
*delayed_refs
;
2575 delayed_refs
= &trans
->transaction
->delayed_refs
;
2577 ret
= cleanup_extent_op(trans
, head
);
2579 unselect_delayed_ref_head(delayed_refs
, head
);
2580 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2587 * Need to drop our head ref lock and re-acquire the delayed ref lock
2588 * and then re-check to make sure nobody got added.
2590 spin_unlock(&head
->lock
);
2591 spin_lock(&delayed_refs
->lock
);
2592 spin_lock(&head
->lock
);
2593 if (!RB_EMPTY_ROOT(&head
->ref_tree
) || head
->extent_op
) {
2594 spin_unlock(&head
->lock
);
2595 spin_unlock(&delayed_refs
->lock
);
2598 delayed_refs
->num_heads
--;
2599 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
2600 RB_CLEAR_NODE(&head
->href_node
);
2601 spin_unlock(&head
->lock
);
2602 spin_unlock(&delayed_refs
->lock
);
2603 atomic_dec(&delayed_refs
->num_entries
);
2605 trace_run_delayed_ref_head(fs_info
, head
, 0);
2607 if (head
->total_ref_mod
< 0) {
2608 struct btrfs_space_info
*space_info
;
2612 flags
= BTRFS_BLOCK_GROUP_DATA
;
2613 else if (head
->is_system
)
2614 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
2616 flags
= BTRFS_BLOCK_GROUP_METADATA
;
2617 space_info
= __find_space_info(fs_info
, flags
);
2619 percpu_counter_add(&space_info
->total_bytes_pinned
,
2622 if (head
->is_data
) {
2623 spin_lock(&delayed_refs
->lock
);
2624 delayed_refs
->pending_csums
-= head
->num_bytes
;
2625 spin_unlock(&delayed_refs
->lock
);
2629 if (head
->must_insert_reserved
) {
2630 btrfs_pin_extent(fs_info
, head
->bytenr
,
2631 head
->num_bytes
, 1);
2632 if (head
->is_data
) {
2633 ret
= btrfs_del_csums(trans
, fs_info
, head
->bytenr
,
2638 /* Also free its reserved qgroup space */
2639 btrfs_qgroup_free_delayed_ref(fs_info
, head
->qgroup_ref_root
,
2640 head
->qgroup_reserved
);
2641 btrfs_delayed_ref_unlock(head
);
2642 btrfs_put_delayed_ref_head(head
);
2647 * Returns 0 on success or if called with an already aborted transaction.
2648 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2650 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2653 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2654 struct btrfs_delayed_ref_root
*delayed_refs
;
2655 struct btrfs_delayed_ref_node
*ref
;
2656 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2657 struct btrfs_delayed_extent_op
*extent_op
;
2658 ktime_t start
= ktime_get();
2660 unsigned long count
= 0;
2661 unsigned long actual_count
= 0;
2662 int must_insert_reserved
= 0;
2664 delayed_refs
= &trans
->transaction
->delayed_refs
;
2670 spin_lock(&delayed_refs
->lock
);
2671 locked_ref
= btrfs_select_ref_head(trans
);
2673 spin_unlock(&delayed_refs
->lock
);
2677 /* grab the lock that says we are going to process
2678 * all the refs for this head */
2679 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2680 spin_unlock(&delayed_refs
->lock
);
2682 * we may have dropped the spin lock to get the head
2683 * mutex lock, and that might have given someone else
2684 * time to free the head. If that's true, it has been
2685 * removed from our list and we can move on.
2687 if (ret
== -EAGAIN
) {
2695 * We need to try and merge add/drops of the same ref since we
2696 * can run into issues with relocate dropping the implicit ref
2697 * and then it being added back again before the drop can
2698 * finish. If we merged anything we need to re-loop so we can
2700 * Or we can get node references of the same type that weren't
2701 * merged when created due to bumps in the tree mod seq, and
2702 * we need to merge them to prevent adding an inline extent
2703 * backref before dropping it (triggering a BUG_ON at
2704 * insert_inline_extent_backref()).
2706 spin_lock(&locked_ref
->lock
);
2707 btrfs_merge_delayed_refs(trans
, delayed_refs
, locked_ref
);
2709 ref
= select_delayed_ref(locked_ref
);
2711 if (ref
&& ref
->seq
&&
2712 btrfs_check_delayed_seq(fs_info
, ref
->seq
)) {
2713 spin_unlock(&locked_ref
->lock
);
2714 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2722 * We're done processing refs in this ref_head, clean everything
2723 * up and move on to the next ref_head.
2726 ret
= cleanup_ref_head(trans
, locked_ref
);
2728 /* We dropped our lock, we need to loop. */
2741 rb_erase(&ref
->ref_node
, &locked_ref
->ref_tree
);
2742 RB_CLEAR_NODE(&ref
->ref_node
);
2743 if (!list_empty(&ref
->add_list
))
2744 list_del(&ref
->add_list
);
2746 * When we play the delayed ref, also correct the ref_mod on
2749 switch (ref
->action
) {
2750 case BTRFS_ADD_DELAYED_REF
:
2751 case BTRFS_ADD_DELAYED_EXTENT
:
2752 locked_ref
->ref_mod
-= ref
->ref_mod
;
2754 case BTRFS_DROP_DELAYED_REF
:
2755 locked_ref
->ref_mod
+= ref
->ref_mod
;
2760 atomic_dec(&delayed_refs
->num_entries
);
2763 * Record the must-insert_reserved flag before we drop the spin
2766 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2767 locked_ref
->must_insert_reserved
= 0;
2769 extent_op
= locked_ref
->extent_op
;
2770 locked_ref
->extent_op
= NULL
;
2771 spin_unlock(&locked_ref
->lock
);
2773 ret
= run_one_delayed_ref(trans
, fs_info
, ref
, extent_op
,
2774 must_insert_reserved
);
2776 btrfs_free_delayed_extent_op(extent_op
);
2778 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2779 btrfs_put_delayed_ref(ref
);
2780 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d",
2785 btrfs_put_delayed_ref(ref
);
2791 * We don't want to include ref heads since we can have empty ref heads
2792 * and those will drastically skew our runtime down since we just do
2793 * accounting, no actual extent tree updates.
2795 if (actual_count
> 0) {
2796 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2800 * We weigh the current average higher than our current runtime
2801 * to avoid large swings in the average.
2803 spin_lock(&delayed_refs
->lock
);
2804 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2805 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2806 spin_unlock(&delayed_refs
->lock
);
2811 #ifdef SCRAMBLE_DELAYED_REFS
2813 * Normally delayed refs get processed in ascending bytenr order. This
2814 * correlates in most cases to the order added. To expose dependencies on this
2815 * order, we start to process the tree in the middle instead of the beginning
2817 static u64
find_middle(struct rb_root
*root
)
2819 struct rb_node
*n
= root
->rb_node
;
2820 struct btrfs_delayed_ref_node
*entry
;
2823 u64 first
= 0, last
= 0;
2827 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2828 first
= entry
->bytenr
;
2832 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2833 last
= entry
->bytenr
;
2838 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2839 WARN_ON(!entry
->in_tree
);
2841 middle
= entry
->bytenr
;
2854 static inline u64
heads_to_leaves(struct btrfs_fs_info
*fs_info
, u64 heads
)
2858 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2859 sizeof(struct btrfs_extent_inline_ref
));
2860 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2861 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2864 * We don't ever fill up leaves all the way so multiply by 2 just to be
2865 * closer to what we're really going to want to use.
2867 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(fs_info
));
2871 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2872 * would require to store the csums for that many bytes.
2874 u64
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info
*fs_info
, u64 csum_bytes
)
2877 u64 num_csums_per_leaf
;
2880 csum_size
= BTRFS_MAX_ITEM_SIZE(fs_info
);
2881 num_csums_per_leaf
= div64_u64(csum_size
,
2882 (u64
)btrfs_super_csum_size(fs_info
->super_copy
));
2883 num_csums
= div64_u64(csum_bytes
, fs_info
->sectorsize
);
2884 num_csums
+= num_csums_per_leaf
- 1;
2885 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2889 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2890 struct btrfs_fs_info
*fs_info
)
2892 struct btrfs_block_rsv
*global_rsv
;
2893 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2894 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2895 unsigned int num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2896 u64 num_bytes
, num_dirty_bgs_bytes
;
2899 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2900 num_heads
= heads_to_leaves(fs_info
, num_heads
);
2902 num_bytes
+= (num_heads
- 1) * fs_info
->nodesize
;
2904 num_bytes
+= btrfs_csum_bytes_to_leaves(fs_info
, csum_bytes
) *
2906 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(fs_info
,
2908 global_rsv
= &fs_info
->global_block_rsv
;
2911 * If we can't allocate any more chunks lets make sure we have _lots_ of
2912 * wiggle room since running delayed refs can create more delayed refs.
2914 if (global_rsv
->space_info
->full
) {
2915 num_dirty_bgs_bytes
<<= 1;
2919 spin_lock(&global_rsv
->lock
);
2920 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2922 spin_unlock(&global_rsv
->lock
);
2926 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2927 struct btrfs_fs_info
*fs_info
)
2930 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2935 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2936 val
= num_entries
* avg_runtime
;
2937 if (val
>= NSEC_PER_SEC
)
2939 if (val
>= NSEC_PER_SEC
/ 2)
2942 return btrfs_check_space_for_delayed_refs(trans
, fs_info
);
2945 struct async_delayed_refs
{
2946 struct btrfs_root
*root
;
2951 struct completion wait
;
2952 struct btrfs_work work
;
2955 static inline struct async_delayed_refs
*
2956 to_async_delayed_refs(struct btrfs_work
*work
)
2958 return container_of(work
, struct async_delayed_refs
, work
);
2961 static void delayed_ref_async_start(struct btrfs_work
*work
)
2963 struct async_delayed_refs
*async
= to_async_delayed_refs(work
);
2964 struct btrfs_trans_handle
*trans
;
2965 struct btrfs_fs_info
*fs_info
= async
->root
->fs_info
;
2968 /* if the commit is already started, we don't need to wait here */
2969 if (btrfs_transaction_blocked(fs_info
))
2972 trans
= btrfs_join_transaction(async
->root
);
2973 if (IS_ERR(trans
)) {
2974 async
->error
= PTR_ERR(trans
);
2979 * trans->sync means that when we call end_transaction, we won't
2980 * wait on delayed refs
2984 /* Don't bother flushing if we got into a different transaction */
2985 if (trans
->transid
> async
->transid
)
2988 ret
= btrfs_run_delayed_refs(trans
, async
->count
);
2992 ret
= btrfs_end_transaction(trans
);
2993 if (ret
&& !async
->error
)
2997 complete(&async
->wait
);
3002 int btrfs_async_run_delayed_refs(struct btrfs_fs_info
*fs_info
,
3003 unsigned long count
, u64 transid
, int wait
)
3005 struct async_delayed_refs
*async
;
3008 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
3012 async
->root
= fs_info
->tree_root
;
3013 async
->count
= count
;
3015 async
->transid
= transid
;
3020 init_completion(&async
->wait
);
3022 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
3023 delayed_ref_async_start
, NULL
, NULL
);
3025 btrfs_queue_work(fs_info
->extent_workers
, &async
->work
);
3028 wait_for_completion(&async
->wait
);
3037 * this starts processing the delayed reference count updates and
3038 * extent insertions we have queued up so far. count can be
3039 * 0, which means to process everything in the tree at the start
3040 * of the run (but not newly added entries), or it can be some target
3041 * number you'd like to process.
3043 * Returns 0 on success or if called with an aborted transaction
3044 * Returns <0 on error and aborts the transaction
3046 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
3047 unsigned long count
)
3049 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3050 struct rb_node
*node
;
3051 struct btrfs_delayed_ref_root
*delayed_refs
;
3052 struct btrfs_delayed_ref_head
*head
;
3054 int run_all
= count
== (unsigned long)-1;
3055 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
3057 /* We'll clean this up in btrfs_cleanup_transaction */
3061 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
3064 delayed_refs
= &trans
->transaction
->delayed_refs
;
3066 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
3069 #ifdef SCRAMBLE_DELAYED_REFS
3070 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
3072 trans
->can_flush_pending_bgs
= false;
3073 ret
= __btrfs_run_delayed_refs(trans
, count
);
3075 btrfs_abort_transaction(trans
, ret
);
3080 if (!list_empty(&trans
->new_bgs
))
3081 btrfs_create_pending_block_groups(trans
);
3083 spin_lock(&delayed_refs
->lock
);
3084 node
= rb_first(&delayed_refs
->href_root
);
3086 spin_unlock(&delayed_refs
->lock
);
3089 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
3091 refcount_inc(&head
->refs
);
3092 spin_unlock(&delayed_refs
->lock
);
3094 /* Mutex was contended, block until it's released and retry. */
3095 mutex_lock(&head
->mutex
);
3096 mutex_unlock(&head
->mutex
);
3098 btrfs_put_delayed_ref_head(head
);
3103 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3107 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3108 struct btrfs_fs_info
*fs_info
,
3109 u64 bytenr
, u64 num_bytes
, u64 flags
,
3110 int level
, int is_data
)
3112 struct btrfs_delayed_extent_op
*extent_op
;
3115 extent_op
= btrfs_alloc_delayed_extent_op();
3119 extent_op
->flags_to_set
= flags
;
3120 extent_op
->update_flags
= true;
3121 extent_op
->update_key
= false;
3122 extent_op
->is_data
= is_data
? true : false;
3123 extent_op
->level
= level
;
3125 ret
= btrfs_add_delayed_extent_op(fs_info
, trans
, bytenr
,
3126 num_bytes
, extent_op
);
3128 btrfs_free_delayed_extent_op(extent_op
);
3132 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
3133 struct btrfs_path
*path
,
3134 u64 objectid
, u64 offset
, u64 bytenr
)
3136 struct btrfs_delayed_ref_head
*head
;
3137 struct btrfs_delayed_ref_node
*ref
;
3138 struct btrfs_delayed_data_ref
*data_ref
;
3139 struct btrfs_delayed_ref_root
*delayed_refs
;
3140 struct btrfs_transaction
*cur_trans
;
3141 struct rb_node
*node
;
3144 spin_lock(&root
->fs_info
->trans_lock
);
3145 cur_trans
= root
->fs_info
->running_transaction
;
3147 refcount_inc(&cur_trans
->use_count
);
3148 spin_unlock(&root
->fs_info
->trans_lock
);
3152 delayed_refs
= &cur_trans
->delayed_refs
;
3153 spin_lock(&delayed_refs
->lock
);
3154 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3156 spin_unlock(&delayed_refs
->lock
);
3157 btrfs_put_transaction(cur_trans
);
3161 if (!mutex_trylock(&head
->mutex
)) {
3162 refcount_inc(&head
->refs
);
3163 spin_unlock(&delayed_refs
->lock
);
3165 btrfs_release_path(path
);
3168 * Mutex was contended, block until it's released and let
3171 mutex_lock(&head
->mutex
);
3172 mutex_unlock(&head
->mutex
);
3173 btrfs_put_delayed_ref_head(head
);
3174 btrfs_put_transaction(cur_trans
);
3177 spin_unlock(&delayed_refs
->lock
);
3179 spin_lock(&head
->lock
);
3181 * XXX: We should replace this with a proper search function in the
3184 for (node
= rb_first(&head
->ref_tree
); node
; node
= rb_next(node
)) {
3185 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, ref_node
);
3186 /* If it's a shared ref we know a cross reference exists */
3187 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3192 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3195 * If our ref doesn't match the one we're currently looking at
3196 * then we have a cross reference.
3198 if (data_ref
->root
!= root
->root_key
.objectid
||
3199 data_ref
->objectid
!= objectid
||
3200 data_ref
->offset
!= offset
) {
3205 spin_unlock(&head
->lock
);
3206 mutex_unlock(&head
->mutex
);
3207 btrfs_put_transaction(cur_trans
);
3211 static noinline
int check_committed_ref(struct btrfs_root
*root
,
3212 struct btrfs_path
*path
,
3213 u64 objectid
, u64 offset
, u64 bytenr
)
3215 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3216 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3217 struct extent_buffer
*leaf
;
3218 struct btrfs_extent_data_ref
*ref
;
3219 struct btrfs_extent_inline_ref
*iref
;
3220 struct btrfs_extent_item
*ei
;
3221 struct btrfs_key key
;
3226 key
.objectid
= bytenr
;
3227 key
.offset
= (u64
)-1;
3228 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3230 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3233 BUG_ON(ret
== 0); /* Corruption */
3236 if (path
->slots
[0] == 0)
3240 leaf
= path
->nodes
[0];
3241 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3243 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3247 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3248 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3249 if (item_size
< sizeof(*ei
)) {
3250 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3254 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3256 if (item_size
!= sizeof(*ei
) +
3257 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3260 if (btrfs_extent_generation(leaf
, ei
) <=
3261 btrfs_root_last_snapshot(&root
->root_item
))
3264 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3266 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
3267 if (type
!= BTRFS_EXTENT_DATA_REF_KEY
)
3270 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3271 if (btrfs_extent_refs(leaf
, ei
) !=
3272 btrfs_extent_data_ref_count(leaf
, ref
) ||
3273 btrfs_extent_data_ref_root(leaf
, ref
) !=
3274 root
->root_key
.objectid
||
3275 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3276 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3284 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
3287 struct btrfs_path
*path
;
3291 path
= btrfs_alloc_path();
3296 ret
= check_committed_ref(root
, path
, objectid
,
3298 if (ret
&& ret
!= -ENOENT
)
3301 ret2
= check_delayed_ref(root
, path
, objectid
,
3303 } while (ret2
== -EAGAIN
);
3305 if (ret2
&& ret2
!= -ENOENT
) {
3310 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3313 btrfs_free_path(path
);
3314 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3319 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3320 struct btrfs_root
*root
,
3321 struct extent_buffer
*buf
,
3322 int full_backref
, int inc
)
3324 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3330 struct btrfs_key key
;
3331 struct btrfs_file_extent_item
*fi
;
3335 int (*process_func
)(struct btrfs_trans_handle
*,
3336 struct btrfs_root
*,
3337 u64
, u64
, u64
, u64
, u64
, u64
);
3340 if (btrfs_is_testing(fs_info
))
3343 ref_root
= btrfs_header_owner(buf
);
3344 nritems
= btrfs_header_nritems(buf
);
3345 level
= btrfs_header_level(buf
);
3347 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3351 process_func
= btrfs_inc_extent_ref
;
3353 process_func
= btrfs_free_extent
;
3356 parent
= buf
->start
;
3360 for (i
= 0; i
< nritems
; i
++) {
3362 btrfs_item_key_to_cpu(buf
, &key
, i
);
3363 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3365 fi
= btrfs_item_ptr(buf
, i
,
3366 struct btrfs_file_extent_item
);
3367 if (btrfs_file_extent_type(buf
, fi
) ==
3368 BTRFS_FILE_EXTENT_INLINE
)
3370 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3374 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3375 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3376 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3377 parent
, ref_root
, key
.objectid
,
3382 bytenr
= btrfs_node_blockptr(buf
, i
);
3383 num_bytes
= fs_info
->nodesize
;
3384 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3385 parent
, ref_root
, level
- 1, 0);
3395 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3396 struct extent_buffer
*buf
, int full_backref
)
3398 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3401 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3402 struct extent_buffer
*buf
, int full_backref
)
3404 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3407 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3408 struct btrfs_fs_info
*fs_info
,
3409 struct btrfs_path
*path
,
3410 struct btrfs_block_group_cache
*cache
)
3413 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3415 struct extent_buffer
*leaf
;
3417 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3424 leaf
= path
->nodes
[0];
3425 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3426 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3427 btrfs_mark_buffer_dirty(leaf
);
3429 btrfs_release_path(path
);
3434 static struct btrfs_block_group_cache
*
3435 next_block_group(struct btrfs_fs_info
*fs_info
,
3436 struct btrfs_block_group_cache
*cache
)
3438 struct rb_node
*node
;
3440 spin_lock(&fs_info
->block_group_cache_lock
);
3442 /* If our block group was removed, we need a full search. */
3443 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3444 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3446 spin_unlock(&fs_info
->block_group_cache_lock
);
3447 btrfs_put_block_group(cache
);
3448 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
3450 node
= rb_next(&cache
->cache_node
);
3451 btrfs_put_block_group(cache
);
3453 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3455 btrfs_get_block_group(cache
);
3458 spin_unlock(&fs_info
->block_group_cache_lock
);
3462 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3463 struct btrfs_trans_handle
*trans
,
3464 struct btrfs_path
*path
)
3466 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3467 struct btrfs_root
*root
= fs_info
->tree_root
;
3468 struct inode
*inode
= NULL
;
3469 struct extent_changeset
*data_reserved
= NULL
;
3471 int dcs
= BTRFS_DC_ERROR
;
3477 * If this block group is smaller than 100 megs don't bother caching the
3480 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3481 spin_lock(&block_group
->lock
);
3482 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3483 spin_unlock(&block_group
->lock
);
3490 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
3491 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3492 ret
= PTR_ERR(inode
);
3493 btrfs_release_path(path
);
3497 if (IS_ERR(inode
)) {
3501 if (block_group
->ro
)
3504 ret
= create_free_space_inode(fs_info
, trans
, block_group
,
3512 * We want to set the generation to 0, that way if anything goes wrong
3513 * from here on out we know not to trust this cache when we load up next
3516 BTRFS_I(inode
)->generation
= 0;
3517 ret
= btrfs_update_inode(trans
, root
, inode
);
3520 * So theoretically we could recover from this, simply set the
3521 * super cache generation to 0 so we know to invalidate the
3522 * cache, but then we'd have to keep track of the block groups
3523 * that fail this way so we know we _have_ to reset this cache
3524 * before the next commit or risk reading stale cache. So to
3525 * limit our exposure to horrible edge cases lets just abort the
3526 * transaction, this only happens in really bad situations
3529 btrfs_abort_transaction(trans
, ret
);
3534 /* We've already setup this transaction, go ahead and exit */
3535 if (block_group
->cache_generation
== trans
->transid
&&
3536 i_size_read(inode
)) {
3537 dcs
= BTRFS_DC_SETUP
;
3541 if (i_size_read(inode
) > 0) {
3542 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3543 &fs_info
->global_block_rsv
);
3547 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3552 spin_lock(&block_group
->lock
);
3553 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3554 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3556 * don't bother trying to write stuff out _if_
3557 * a) we're not cached,
3558 * b) we're with nospace_cache mount option,
3559 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3561 dcs
= BTRFS_DC_WRITTEN
;
3562 spin_unlock(&block_group
->lock
);
3565 spin_unlock(&block_group
->lock
);
3568 * We hit an ENOSPC when setting up the cache in this transaction, just
3569 * skip doing the setup, we've already cleared the cache so we're safe.
3571 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3577 * Try to preallocate enough space based on how big the block group is.
3578 * Keep in mind this has to include any pinned space which could end up
3579 * taking up quite a bit since it's not folded into the other space
3582 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3587 num_pages
*= PAGE_SIZE
;
3589 ret
= btrfs_check_data_free_space(inode
, &data_reserved
, 0, num_pages
);
3593 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3594 num_pages
, num_pages
,
3597 * Our cache requires contiguous chunks so that we don't modify a bunch
3598 * of metadata or split extents when writing the cache out, which means
3599 * we can enospc if we are heavily fragmented in addition to just normal
3600 * out of space conditions. So if we hit this just skip setting up any
3601 * other block groups for this transaction, maybe we'll unpin enough
3602 * space the next time around.
3605 dcs
= BTRFS_DC_SETUP
;
3606 else if (ret
== -ENOSPC
)
3607 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3612 btrfs_release_path(path
);
3614 spin_lock(&block_group
->lock
);
3615 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3616 block_group
->cache_generation
= trans
->transid
;
3617 block_group
->disk_cache_state
= dcs
;
3618 spin_unlock(&block_group
->lock
);
3620 extent_changeset_free(data_reserved
);
3624 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3625 struct btrfs_fs_info
*fs_info
)
3627 struct btrfs_block_group_cache
*cache
, *tmp
;
3628 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3629 struct btrfs_path
*path
;
3631 if (list_empty(&cur_trans
->dirty_bgs
) ||
3632 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3635 path
= btrfs_alloc_path();
3639 /* Could add new block groups, use _safe just in case */
3640 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3642 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3643 cache_save_setup(cache
, trans
, path
);
3646 btrfs_free_path(path
);
3651 * transaction commit does final block group cache writeback during a
3652 * critical section where nothing is allowed to change the FS. This is
3653 * required in order for the cache to actually match the block group,
3654 * but can introduce a lot of latency into the commit.
3656 * So, btrfs_start_dirty_block_groups is here to kick off block group
3657 * cache IO. There's a chance we'll have to redo some of it if the
3658 * block group changes again during the commit, but it greatly reduces
3659 * the commit latency by getting rid of the easy block groups while
3660 * we're still allowing others to join the commit.
3662 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
)
3664 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3665 struct btrfs_block_group_cache
*cache
;
3666 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3669 struct btrfs_path
*path
= NULL
;
3671 struct list_head
*io
= &cur_trans
->io_bgs
;
3672 int num_started
= 0;
3675 spin_lock(&cur_trans
->dirty_bgs_lock
);
3676 if (list_empty(&cur_trans
->dirty_bgs
)) {
3677 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3680 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3681 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3685 * make sure all the block groups on our dirty list actually
3688 btrfs_create_pending_block_groups(trans
);
3691 path
= btrfs_alloc_path();
3697 * cache_write_mutex is here only to save us from balance or automatic
3698 * removal of empty block groups deleting this block group while we are
3699 * writing out the cache
3701 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3702 while (!list_empty(&dirty
)) {
3703 cache
= list_first_entry(&dirty
,
3704 struct btrfs_block_group_cache
,
3707 * this can happen if something re-dirties a block
3708 * group that is already under IO. Just wait for it to
3709 * finish and then do it all again
3711 if (!list_empty(&cache
->io_list
)) {
3712 list_del_init(&cache
->io_list
);
3713 btrfs_wait_cache_io(trans
, cache
, path
);
3714 btrfs_put_block_group(cache
);
3719 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3720 * if it should update the cache_state. Don't delete
3721 * until after we wait.
3723 * Since we're not running in the commit critical section
3724 * we need the dirty_bgs_lock to protect from update_block_group
3726 spin_lock(&cur_trans
->dirty_bgs_lock
);
3727 list_del_init(&cache
->dirty_list
);
3728 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3732 cache_save_setup(cache
, trans
, path
);
3734 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3735 cache
->io_ctl
.inode
= NULL
;
3736 ret
= btrfs_write_out_cache(fs_info
, trans
,
3738 if (ret
== 0 && cache
->io_ctl
.inode
) {
3743 * The cache_write_mutex is protecting the
3744 * io_list, also refer to the definition of
3745 * btrfs_transaction::io_bgs for more details
3747 list_add_tail(&cache
->io_list
, io
);
3750 * if we failed to write the cache, the
3751 * generation will be bad and life goes on
3757 ret
= write_one_cache_group(trans
, fs_info
,
3760 * Our block group might still be attached to the list
3761 * of new block groups in the transaction handle of some
3762 * other task (struct btrfs_trans_handle->new_bgs). This
3763 * means its block group item isn't yet in the extent
3764 * tree. If this happens ignore the error, as we will
3765 * try again later in the critical section of the
3766 * transaction commit.
3768 if (ret
== -ENOENT
) {
3770 spin_lock(&cur_trans
->dirty_bgs_lock
);
3771 if (list_empty(&cache
->dirty_list
)) {
3772 list_add_tail(&cache
->dirty_list
,
3773 &cur_trans
->dirty_bgs
);
3774 btrfs_get_block_group(cache
);
3776 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3778 btrfs_abort_transaction(trans
, ret
);
3782 /* if its not on the io list, we need to put the block group */
3784 btrfs_put_block_group(cache
);
3790 * Avoid blocking other tasks for too long. It might even save
3791 * us from writing caches for block groups that are going to be
3794 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3795 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3797 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3800 * go through delayed refs for all the stuff we've just kicked off
3801 * and then loop back (just once)
3803 ret
= btrfs_run_delayed_refs(trans
, 0);
3804 if (!ret
&& loops
== 0) {
3806 spin_lock(&cur_trans
->dirty_bgs_lock
);
3807 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3809 * dirty_bgs_lock protects us from concurrent block group
3810 * deletes too (not just cache_write_mutex).
3812 if (!list_empty(&dirty
)) {
3813 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3816 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3817 } else if (ret
< 0) {
3818 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3821 btrfs_free_path(path
);
3825 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3826 struct btrfs_fs_info
*fs_info
)
3828 struct btrfs_block_group_cache
*cache
;
3829 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3832 struct btrfs_path
*path
;
3833 struct list_head
*io
= &cur_trans
->io_bgs
;
3834 int num_started
= 0;
3836 path
= btrfs_alloc_path();
3841 * Even though we are in the critical section of the transaction commit,
3842 * we can still have concurrent tasks adding elements to this
3843 * transaction's list of dirty block groups. These tasks correspond to
3844 * endio free space workers started when writeback finishes for a
3845 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3846 * allocate new block groups as a result of COWing nodes of the root
3847 * tree when updating the free space inode. The writeback for the space
3848 * caches is triggered by an earlier call to
3849 * btrfs_start_dirty_block_groups() and iterations of the following
3851 * Also we want to do the cache_save_setup first and then run the
3852 * delayed refs to make sure we have the best chance at doing this all
3855 spin_lock(&cur_trans
->dirty_bgs_lock
);
3856 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3857 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3858 struct btrfs_block_group_cache
,
3862 * this can happen if cache_save_setup re-dirties a block
3863 * group that is already under IO. Just wait for it to
3864 * finish and then do it all again
3866 if (!list_empty(&cache
->io_list
)) {
3867 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3868 list_del_init(&cache
->io_list
);
3869 btrfs_wait_cache_io(trans
, cache
, path
);
3870 btrfs_put_block_group(cache
);
3871 spin_lock(&cur_trans
->dirty_bgs_lock
);
3875 * don't remove from the dirty list until after we've waited
3878 list_del_init(&cache
->dirty_list
);
3879 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3882 cache_save_setup(cache
, trans
, path
);
3885 ret
= btrfs_run_delayed_refs(trans
,
3886 (unsigned long) -1);
3888 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3889 cache
->io_ctl
.inode
= NULL
;
3890 ret
= btrfs_write_out_cache(fs_info
, trans
,
3892 if (ret
== 0 && cache
->io_ctl
.inode
) {
3895 list_add_tail(&cache
->io_list
, io
);
3898 * if we failed to write the cache, the
3899 * generation will be bad and life goes on
3905 ret
= write_one_cache_group(trans
, fs_info
,
3908 * One of the free space endio workers might have
3909 * created a new block group while updating a free space
3910 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3911 * and hasn't released its transaction handle yet, in
3912 * which case the new block group is still attached to
3913 * its transaction handle and its creation has not
3914 * finished yet (no block group item in the extent tree
3915 * yet, etc). If this is the case, wait for all free
3916 * space endio workers to finish and retry. This is a
3917 * a very rare case so no need for a more efficient and
3920 if (ret
== -ENOENT
) {
3921 wait_event(cur_trans
->writer_wait
,
3922 atomic_read(&cur_trans
->num_writers
) == 1);
3923 ret
= write_one_cache_group(trans
, fs_info
,
3927 btrfs_abort_transaction(trans
, ret
);
3930 /* if its not on the io list, we need to put the block group */
3932 btrfs_put_block_group(cache
);
3933 spin_lock(&cur_trans
->dirty_bgs_lock
);
3935 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3938 * Refer to the definition of io_bgs member for details why it's safe
3939 * to use it without any locking
3941 while (!list_empty(io
)) {
3942 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3944 list_del_init(&cache
->io_list
);
3945 btrfs_wait_cache_io(trans
, cache
, path
);
3946 btrfs_put_block_group(cache
);
3949 btrfs_free_path(path
);
3953 int btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3955 struct btrfs_block_group_cache
*block_group
;
3958 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
3959 if (!block_group
|| block_group
->ro
)
3962 btrfs_put_block_group(block_group
);
3966 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3968 struct btrfs_block_group_cache
*bg
;
3971 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3975 spin_lock(&bg
->lock
);
3979 atomic_inc(&bg
->nocow_writers
);
3980 spin_unlock(&bg
->lock
);
3982 /* no put on block group, done by btrfs_dec_nocow_writers */
3984 btrfs_put_block_group(bg
);
3990 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3992 struct btrfs_block_group_cache
*bg
;
3994 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3996 if (atomic_dec_and_test(&bg
->nocow_writers
))
3997 wake_up_var(&bg
->nocow_writers
);
3999 * Once for our lookup and once for the lookup done by a previous call
4000 * to btrfs_inc_nocow_writers()
4002 btrfs_put_block_group(bg
);
4003 btrfs_put_block_group(bg
);
4006 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
4008 wait_var_event(&bg
->nocow_writers
, !atomic_read(&bg
->nocow_writers
));
4011 static const char *alloc_name(u64 flags
)
4014 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
4016 case BTRFS_BLOCK_GROUP_METADATA
:
4018 case BTRFS_BLOCK_GROUP_DATA
:
4020 case BTRFS_BLOCK_GROUP_SYSTEM
:
4024 return "invalid-combination";
4028 static int create_space_info(struct btrfs_fs_info
*info
, u64 flags
)
4031 struct btrfs_space_info
*space_info
;
4035 space_info
= kzalloc(sizeof(*space_info
), GFP_NOFS
);
4039 ret
= percpu_counter_init(&space_info
->total_bytes_pinned
, 0,
4046 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
4047 INIT_LIST_HEAD(&space_info
->block_groups
[i
]);
4048 init_rwsem(&space_info
->groups_sem
);
4049 spin_lock_init(&space_info
->lock
);
4050 space_info
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
4051 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4052 init_waitqueue_head(&space_info
->wait
);
4053 INIT_LIST_HEAD(&space_info
->ro_bgs
);
4054 INIT_LIST_HEAD(&space_info
->tickets
);
4055 INIT_LIST_HEAD(&space_info
->priority_tickets
);
4057 ret
= kobject_init_and_add(&space_info
->kobj
, &space_info_ktype
,
4058 info
->space_info_kobj
, "%s",
4059 alloc_name(space_info
->flags
));
4061 percpu_counter_destroy(&space_info
->total_bytes_pinned
);
4066 list_add_rcu(&space_info
->list
, &info
->space_info
);
4067 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4068 info
->data_sinfo
= space_info
;
4073 static void update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
4074 u64 total_bytes
, u64 bytes_used
,
4076 struct btrfs_space_info
**space_info
)
4078 struct btrfs_space_info
*found
;
4081 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
4082 BTRFS_BLOCK_GROUP_RAID10
))
4087 found
= __find_space_info(info
, flags
);
4089 spin_lock(&found
->lock
);
4090 found
->total_bytes
+= total_bytes
;
4091 found
->disk_total
+= total_bytes
* factor
;
4092 found
->bytes_used
+= bytes_used
;
4093 found
->disk_used
+= bytes_used
* factor
;
4094 found
->bytes_readonly
+= bytes_readonly
;
4095 if (total_bytes
> 0)
4097 space_info_add_new_bytes(info
, found
, total_bytes
-
4098 bytes_used
- bytes_readonly
);
4099 spin_unlock(&found
->lock
);
4100 *space_info
= found
;
4103 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
4105 u64 extra_flags
= chunk_to_extended(flags
) &
4106 BTRFS_EXTENDED_PROFILE_MASK
;
4108 write_seqlock(&fs_info
->profiles_lock
);
4109 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4110 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4111 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4112 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4113 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4114 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4115 write_sequnlock(&fs_info
->profiles_lock
);
4119 * returns target flags in extended format or 0 if restripe for this
4120 * chunk_type is not in progress
4122 * should be called with balance_lock held
4124 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4126 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4132 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4133 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4134 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4135 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4136 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4137 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4138 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4139 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4140 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4147 * @flags: available profiles in extended format (see ctree.h)
4149 * Returns reduced profile in chunk format. If profile changing is in
4150 * progress (either running or paused) picks the target profile (if it's
4151 * already available), otherwise falls back to plain reducing.
4153 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
4155 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
4161 * see if restripe for this chunk_type is in progress, if so
4162 * try to reduce to the target profile
4164 spin_lock(&fs_info
->balance_lock
);
4165 target
= get_restripe_target(fs_info
, flags
);
4167 /* pick target profile only if it's already available */
4168 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4169 spin_unlock(&fs_info
->balance_lock
);
4170 return extended_to_chunk(target
);
4173 spin_unlock(&fs_info
->balance_lock
);
4175 /* First, mask out the RAID levels which aren't possible */
4176 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4177 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4178 allowed
|= btrfs_raid_array
[raid_type
].bg_flag
;
4182 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4183 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4184 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4185 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4186 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4187 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4188 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4189 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4190 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4191 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4193 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4195 return extended_to_chunk(flags
| allowed
);
4198 static u64
get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
4205 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4207 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4208 flags
|= fs_info
->avail_data_alloc_bits
;
4209 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4210 flags
|= fs_info
->avail_system_alloc_bits
;
4211 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4212 flags
|= fs_info
->avail_metadata_alloc_bits
;
4213 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4215 return btrfs_reduce_alloc_profile(fs_info
, flags
);
4218 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
4220 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4225 flags
= BTRFS_BLOCK_GROUP_DATA
;
4226 else if (root
== fs_info
->chunk_root
)
4227 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4229 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4231 ret
= get_alloc_profile(fs_info
, flags
);
4235 u64
btrfs_data_alloc_profile(struct btrfs_fs_info
*fs_info
)
4237 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4240 u64
btrfs_metadata_alloc_profile(struct btrfs_fs_info
*fs_info
)
4242 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4245 u64
btrfs_system_alloc_profile(struct btrfs_fs_info
*fs_info
)
4247 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4250 static u64
btrfs_space_info_used(struct btrfs_space_info
*s_info
,
4251 bool may_use_included
)
4254 return s_info
->bytes_used
+ s_info
->bytes_reserved
+
4255 s_info
->bytes_pinned
+ s_info
->bytes_readonly
+
4256 (may_use_included
? s_info
->bytes_may_use
: 0);
4259 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode
*inode
, u64 bytes
)
4261 struct btrfs_root
*root
= inode
->root
;
4262 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4263 struct btrfs_space_info
*data_sinfo
= fs_info
->data_sinfo
;
4266 int need_commit
= 2;
4267 int have_pinned_space
;
4269 /* make sure bytes are sectorsize aligned */
4270 bytes
= ALIGN(bytes
, fs_info
->sectorsize
);
4272 if (btrfs_is_free_space_inode(inode
)) {
4274 ASSERT(current
->journal_info
);
4278 /* make sure we have enough space to handle the data first */
4279 spin_lock(&data_sinfo
->lock
);
4280 used
= btrfs_space_info_used(data_sinfo
, true);
4282 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4283 struct btrfs_trans_handle
*trans
;
4286 * if we don't have enough free bytes in this space then we need
4287 * to alloc a new chunk.
4289 if (!data_sinfo
->full
) {
4292 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4293 spin_unlock(&data_sinfo
->lock
);
4295 alloc_target
= btrfs_data_alloc_profile(fs_info
);
4297 * It is ugly that we don't call nolock join
4298 * transaction for the free space inode case here.
4299 * But it is safe because we only do the data space
4300 * reservation for the free space cache in the
4301 * transaction context, the common join transaction
4302 * just increase the counter of the current transaction
4303 * handler, doesn't try to acquire the trans_lock of
4306 trans
= btrfs_join_transaction(root
);
4308 return PTR_ERR(trans
);
4310 ret
= do_chunk_alloc(trans
, fs_info
, alloc_target
,
4311 CHUNK_ALLOC_NO_FORCE
);
4312 btrfs_end_transaction(trans
);
4317 have_pinned_space
= 1;
4326 * If we don't have enough pinned space to deal with this
4327 * allocation, and no removed chunk in current transaction,
4328 * don't bother committing the transaction.
4330 have_pinned_space
= percpu_counter_compare(
4331 &data_sinfo
->total_bytes_pinned
,
4332 used
+ bytes
- data_sinfo
->total_bytes
);
4333 spin_unlock(&data_sinfo
->lock
);
4335 /* commit the current transaction and try again */
4340 if (need_commit
> 0) {
4341 btrfs_start_delalloc_roots(fs_info
, -1);
4342 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0,
4346 trans
= btrfs_join_transaction(root
);
4348 return PTR_ERR(trans
);
4349 if (have_pinned_space
>= 0 ||
4350 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4351 &trans
->transaction
->flags
) ||
4353 ret
= btrfs_commit_transaction(trans
);
4357 * The cleaner kthread might still be doing iput
4358 * operations. Wait for it to finish so that
4359 * more space is released.
4361 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
4362 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
4365 btrfs_end_transaction(trans
);
4369 trace_btrfs_space_reservation(fs_info
,
4370 "space_info:enospc",
4371 data_sinfo
->flags
, bytes
, 1);
4374 data_sinfo
->bytes_may_use
+= bytes
;
4375 trace_btrfs_space_reservation(fs_info
, "space_info",
4376 data_sinfo
->flags
, bytes
, 1);
4377 spin_unlock(&data_sinfo
->lock
);
4382 int btrfs_check_data_free_space(struct inode
*inode
,
4383 struct extent_changeset
**reserved
, u64 start
, u64 len
)
4385 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4388 /* align the range */
4389 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4390 round_down(start
, fs_info
->sectorsize
);
4391 start
= round_down(start
, fs_info
->sectorsize
);
4393 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
), len
);
4397 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4398 ret
= btrfs_qgroup_reserve_data(inode
, reserved
, start
, len
);
4400 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4407 * Called if we need to clear a data reservation for this inode
4408 * Normally in a error case.
4410 * This one will *NOT* use accurate qgroup reserved space API, just for case
4411 * which we can't sleep and is sure it won't affect qgroup reserved space.
4412 * Like clear_bit_hook().
4414 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4417 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4418 struct btrfs_space_info
*data_sinfo
;
4420 /* Make sure the range is aligned to sectorsize */
4421 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4422 round_down(start
, fs_info
->sectorsize
);
4423 start
= round_down(start
, fs_info
->sectorsize
);
4425 data_sinfo
= fs_info
->data_sinfo
;
4426 spin_lock(&data_sinfo
->lock
);
4427 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4428 data_sinfo
->bytes_may_use
= 0;
4430 data_sinfo
->bytes_may_use
-= len
;
4431 trace_btrfs_space_reservation(fs_info
, "space_info",
4432 data_sinfo
->flags
, len
, 0);
4433 spin_unlock(&data_sinfo
->lock
);
4437 * Called if we need to clear a data reservation for this inode
4438 * Normally in a error case.
4440 * This one will handle the per-inode data rsv map for accurate reserved
4443 void btrfs_free_reserved_data_space(struct inode
*inode
,
4444 struct extent_changeset
*reserved
, u64 start
, u64 len
)
4446 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4448 /* Make sure the range is aligned to sectorsize */
4449 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
) -
4450 round_down(start
, root
->fs_info
->sectorsize
);
4451 start
= round_down(start
, root
->fs_info
->sectorsize
);
4453 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4454 btrfs_qgroup_free_data(inode
, reserved
, start
, len
);
4457 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4459 struct list_head
*head
= &info
->space_info
;
4460 struct btrfs_space_info
*found
;
4463 list_for_each_entry_rcu(found
, head
, list
) {
4464 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4465 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4470 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4472 return (global
->size
<< 1);
4475 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
4476 struct btrfs_space_info
*sinfo
, int force
)
4478 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4479 u64 bytes_used
= btrfs_space_info_used(sinfo
, false);
4482 if (force
== CHUNK_ALLOC_FORCE
)
4486 * We need to take into account the global rsv because for all intents
4487 * and purposes it's used space. Don't worry about locking the
4488 * global_rsv, it doesn't change except when the transaction commits.
4490 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4491 bytes_used
+= calc_global_rsv_need_space(global_rsv
);
4494 * in limited mode, we want to have some free space up to
4495 * about 1% of the FS size.
4497 if (force
== CHUNK_ALLOC_LIMITED
) {
4498 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
4499 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4501 if (sinfo
->total_bytes
- bytes_used
< thresh
)
4505 if (bytes_used
+ SZ_2M
< div_factor(sinfo
->total_bytes
, 8))
4510 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4514 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4515 BTRFS_BLOCK_GROUP_RAID0
|
4516 BTRFS_BLOCK_GROUP_RAID5
|
4517 BTRFS_BLOCK_GROUP_RAID6
))
4518 num_dev
= fs_info
->fs_devices
->rw_devices
;
4519 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4522 num_dev
= 1; /* DUP or single */
4528 * If @is_allocation is true, reserve space in the system space info necessary
4529 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4532 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4533 struct btrfs_fs_info
*fs_info
, u64 type
)
4535 struct btrfs_space_info
*info
;
4542 * Needed because we can end up allocating a system chunk and for an
4543 * atomic and race free space reservation in the chunk block reserve.
4545 lockdep_assert_held(&fs_info
->chunk_mutex
);
4547 info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4548 spin_lock(&info
->lock
);
4549 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4550 spin_unlock(&info
->lock
);
4552 num_devs
= get_profile_num_devs(fs_info
, type
);
4554 /* num_devs device items to update and 1 chunk item to add or remove */
4555 thresh
= btrfs_calc_trunc_metadata_size(fs_info
, num_devs
) +
4556 btrfs_calc_trans_metadata_size(fs_info
, 1);
4558 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4559 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4560 left
, thresh
, type
);
4561 dump_space_info(fs_info
, info
, 0, 0);
4564 if (left
< thresh
) {
4565 u64 flags
= btrfs_system_alloc_profile(fs_info
);
4568 * Ignore failure to create system chunk. We might end up not
4569 * needing it, as we might not need to COW all nodes/leafs from
4570 * the paths we visit in the chunk tree (they were already COWed
4571 * or created in the current transaction for example).
4573 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4577 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
4578 &fs_info
->chunk_block_rsv
,
4579 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4581 trans
->chunk_bytes_reserved
+= thresh
;
4586 * If force is CHUNK_ALLOC_FORCE:
4587 * - return 1 if it successfully allocates a chunk,
4588 * - return errors including -ENOSPC otherwise.
4589 * If force is NOT CHUNK_ALLOC_FORCE:
4590 * - return 0 if it doesn't need to allocate a new chunk,
4591 * - return 1 if it successfully allocates a chunk,
4592 * - return errors including -ENOSPC otherwise.
4594 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4595 struct btrfs_fs_info
*fs_info
, u64 flags
, int force
)
4597 struct btrfs_space_info
*space_info
;
4598 int wait_for_alloc
= 0;
4601 /* Don't re-enter if we're already allocating a chunk */
4602 if (trans
->allocating_chunk
)
4605 space_info
= __find_space_info(fs_info
, flags
);
4609 spin_lock(&space_info
->lock
);
4610 if (force
< space_info
->force_alloc
)
4611 force
= space_info
->force_alloc
;
4612 if (space_info
->full
) {
4613 if (should_alloc_chunk(fs_info
, space_info
, force
))
4617 spin_unlock(&space_info
->lock
);
4621 if (!should_alloc_chunk(fs_info
, space_info
, force
)) {
4622 spin_unlock(&space_info
->lock
);
4624 } else if (space_info
->chunk_alloc
) {
4627 space_info
->chunk_alloc
= 1;
4630 spin_unlock(&space_info
->lock
);
4632 mutex_lock(&fs_info
->chunk_mutex
);
4635 * The chunk_mutex is held throughout the entirety of a chunk
4636 * allocation, so once we've acquired the chunk_mutex we know that the
4637 * other guy is done and we need to recheck and see if we should
4640 if (wait_for_alloc
) {
4641 mutex_unlock(&fs_info
->chunk_mutex
);
4647 trans
->allocating_chunk
= true;
4650 * If we have mixed data/metadata chunks we want to make sure we keep
4651 * allocating mixed chunks instead of individual chunks.
4653 if (btrfs_mixed_space_info(space_info
))
4654 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4657 * if we're doing a data chunk, go ahead and make sure that
4658 * we keep a reasonable number of metadata chunks allocated in the
4661 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4662 fs_info
->data_chunk_allocations
++;
4663 if (!(fs_info
->data_chunk_allocations
%
4664 fs_info
->metadata_ratio
))
4665 force_metadata_allocation(fs_info
);
4669 * Check if we have enough space in SYSTEM chunk because we may need
4670 * to update devices.
4672 check_system_chunk(trans
, fs_info
, flags
);
4674 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4675 trans
->allocating_chunk
= false;
4677 spin_lock(&space_info
->lock
);
4680 space_info
->full
= 1;
4687 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4689 space_info
->chunk_alloc
= 0;
4690 spin_unlock(&space_info
->lock
);
4691 mutex_unlock(&fs_info
->chunk_mutex
);
4693 * When we allocate a new chunk we reserve space in the chunk block
4694 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4695 * add new nodes/leafs to it if we end up needing to do it when
4696 * inserting the chunk item and updating device items as part of the
4697 * second phase of chunk allocation, performed by
4698 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4699 * large number of new block groups to create in our transaction
4700 * handle's new_bgs list to avoid exhausting the chunk block reserve
4701 * in extreme cases - like having a single transaction create many new
4702 * block groups when starting to write out the free space caches of all
4703 * the block groups that were made dirty during the lifetime of the
4706 if (trans
->can_flush_pending_bgs
&&
4707 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4708 btrfs_create_pending_block_groups(trans
);
4709 btrfs_trans_release_chunk_metadata(trans
);
4714 static int can_overcommit(struct btrfs_fs_info
*fs_info
,
4715 struct btrfs_space_info
*space_info
, u64 bytes
,
4716 enum btrfs_reserve_flush_enum flush
,
4719 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4725 /* Don't overcommit when in mixed mode. */
4726 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4730 profile
= btrfs_system_alloc_profile(fs_info
);
4732 profile
= btrfs_metadata_alloc_profile(fs_info
);
4734 used
= btrfs_space_info_used(space_info
, false);
4737 * We only want to allow over committing if we have lots of actual space
4738 * free, but if we don't have enough space to handle the global reserve
4739 * space then we could end up having a real enospc problem when trying
4740 * to allocate a chunk or some other such important allocation.
4742 spin_lock(&global_rsv
->lock
);
4743 space_size
= calc_global_rsv_need_space(global_rsv
);
4744 spin_unlock(&global_rsv
->lock
);
4745 if (used
+ space_size
>= space_info
->total_bytes
)
4748 used
+= space_info
->bytes_may_use
;
4750 avail
= atomic64_read(&fs_info
->free_chunk_space
);
4753 * If we have dup, raid1 or raid10 then only half of the free
4754 * space is actually useable. For raid56, the space info used
4755 * doesn't include the parity drive, so we don't have to
4758 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4759 BTRFS_BLOCK_GROUP_RAID1
|
4760 BTRFS_BLOCK_GROUP_RAID10
))
4764 * If we aren't flushing all things, let us overcommit up to
4765 * 1/2th of the space. If we can flush, don't let us overcommit
4766 * too much, let it overcommit up to 1/8 of the space.
4768 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4773 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4778 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info
*fs_info
,
4779 unsigned long nr_pages
, int nr_items
)
4781 struct super_block
*sb
= fs_info
->sb
;
4783 if (down_read_trylock(&sb
->s_umount
)) {
4784 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4785 up_read(&sb
->s_umount
);
4788 * We needn't worry the filesystem going from r/w to r/o though
4789 * we don't acquire ->s_umount mutex, because the filesystem
4790 * should guarantee the delalloc inodes list be empty after
4791 * the filesystem is readonly(all dirty pages are written to
4794 btrfs_start_delalloc_roots(fs_info
, nr_items
);
4795 if (!current
->journal_info
)
4796 btrfs_wait_ordered_roots(fs_info
, nr_items
, 0, (u64
)-1);
4800 static inline u64
calc_reclaim_items_nr(struct btrfs_fs_info
*fs_info
,
4806 bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
4807 nr
= div64_u64(to_reclaim
, bytes
);
4813 #define EXTENT_SIZE_PER_ITEM SZ_256K
4816 * shrink metadata reservation for delalloc
4818 static void shrink_delalloc(struct btrfs_fs_info
*fs_info
, u64 to_reclaim
,
4819 u64 orig
, bool wait_ordered
)
4821 struct btrfs_space_info
*space_info
;
4822 struct btrfs_trans_handle
*trans
;
4827 unsigned long nr_pages
;
4830 /* Calc the number of the pages we need flush for space reservation */
4831 items
= calc_reclaim_items_nr(fs_info
, to_reclaim
);
4832 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4834 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4835 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4837 delalloc_bytes
= percpu_counter_sum_positive(
4838 &fs_info
->delalloc_bytes
);
4839 if (delalloc_bytes
== 0) {
4843 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4848 while (delalloc_bytes
&& loops
< 3) {
4849 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4850 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4851 btrfs_writeback_inodes_sb_nr(fs_info
, nr_pages
, items
);
4853 * We need to wait for the async pages to actually start before
4856 max_reclaim
= atomic_read(&fs_info
->async_delalloc_pages
);
4860 if (max_reclaim
<= nr_pages
)
4863 max_reclaim
-= nr_pages
;
4865 wait_event(fs_info
->async_submit_wait
,
4866 atomic_read(&fs_info
->async_delalloc_pages
) <=
4869 spin_lock(&space_info
->lock
);
4870 if (list_empty(&space_info
->tickets
) &&
4871 list_empty(&space_info
->priority_tickets
)) {
4872 spin_unlock(&space_info
->lock
);
4875 spin_unlock(&space_info
->lock
);
4878 if (wait_ordered
&& !trans
) {
4879 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4881 time_left
= schedule_timeout_killable(1);
4885 delalloc_bytes
= percpu_counter_sum_positive(
4886 &fs_info
->delalloc_bytes
);
4890 struct reserve_ticket
{
4893 struct list_head list
;
4894 wait_queue_head_t wait
;
4898 * maybe_commit_transaction - possibly commit the transaction if its ok to
4899 * @root - the root we're allocating for
4900 * @bytes - the number of bytes we want to reserve
4901 * @force - force the commit
4903 * This will check to make sure that committing the transaction will actually
4904 * get us somewhere and then commit the transaction if it does. Otherwise it
4905 * will return -ENOSPC.
4907 static int may_commit_transaction(struct btrfs_fs_info
*fs_info
,
4908 struct btrfs_space_info
*space_info
)
4910 struct reserve_ticket
*ticket
= NULL
;
4911 struct btrfs_block_rsv
*delayed_rsv
= &fs_info
->delayed_block_rsv
;
4912 struct btrfs_trans_handle
*trans
;
4915 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4919 spin_lock(&space_info
->lock
);
4920 if (!list_empty(&space_info
->priority_tickets
))
4921 ticket
= list_first_entry(&space_info
->priority_tickets
,
4922 struct reserve_ticket
, list
);
4923 else if (!list_empty(&space_info
->tickets
))
4924 ticket
= list_first_entry(&space_info
->tickets
,
4925 struct reserve_ticket
, list
);
4926 bytes
= (ticket
) ? ticket
->bytes
: 0;
4927 spin_unlock(&space_info
->lock
);
4932 /* See if there is enough pinned space to make this reservation */
4933 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4938 * See if there is some space in the delayed insertion reservation for
4941 if (space_info
!= delayed_rsv
->space_info
)
4944 spin_lock(&delayed_rsv
->lock
);
4945 if (delayed_rsv
->size
> bytes
)
4948 bytes
-= delayed_rsv
->size
;
4949 spin_unlock(&delayed_rsv
->lock
);
4951 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4957 trans
= btrfs_join_transaction(fs_info
->extent_root
);
4961 return btrfs_commit_transaction(trans
);
4965 * Try to flush some data based on policy set by @state. This is only advisory
4966 * and may fail for various reasons. The caller is supposed to examine the
4967 * state of @space_info to detect the outcome.
4969 static void flush_space(struct btrfs_fs_info
*fs_info
,
4970 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4973 struct btrfs_root
*root
= fs_info
->extent_root
;
4974 struct btrfs_trans_handle
*trans
;
4979 case FLUSH_DELAYED_ITEMS_NR
:
4980 case FLUSH_DELAYED_ITEMS
:
4981 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4982 nr
= calc_reclaim_items_nr(fs_info
, num_bytes
) * 2;
4986 trans
= btrfs_join_transaction(root
);
4987 if (IS_ERR(trans
)) {
4988 ret
= PTR_ERR(trans
);
4991 ret
= btrfs_run_delayed_items_nr(trans
, nr
);
4992 btrfs_end_transaction(trans
);
4994 case FLUSH_DELALLOC
:
4995 case FLUSH_DELALLOC_WAIT
:
4996 shrink_delalloc(fs_info
, num_bytes
* 2, num_bytes
,
4997 state
== FLUSH_DELALLOC_WAIT
);
5000 trans
= btrfs_join_transaction(root
);
5001 if (IS_ERR(trans
)) {
5002 ret
= PTR_ERR(trans
);
5005 ret
= do_chunk_alloc(trans
, fs_info
,
5006 btrfs_metadata_alloc_profile(fs_info
),
5007 CHUNK_ALLOC_NO_FORCE
);
5008 btrfs_end_transaction(trans
);
5009 if (ret
> 0 || ret
== -ENOSPC
)
5013 ret
= may_commit_transaction(fs_info
, space_info
);
5020 trace_btrfs_flush_space(fs_info
, space_info
->flags
, num_bytes
, state
,
5026 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info
*fs_info
,
5027 struct btrfs_space_info
*space_info
,
5030 struct reserve_ticket
*ticket
;
5035 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
5036 to_reclaim
+= ticket
->bytes
;
5037 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
5038 to_reclaim
+= ticket
->bytes
;
5042 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
5043 if (can_overcommit(fs_info
, space_info
, to_reclaim
,
5044 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
5047 used
= btrfs_space_info_used(space_info
, true);
5049 if (can_overcommit(fs_info
, space_info
, SZ_1M
,
5050 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
5051 expected
= div_factor_fine(space_info
->total_bytes
, 95);
5053 expected
= div_factor_fine(space_info
->total_bytes
, 90);
5055 if (used
> expected
)
5056 to_reclaim
= used
- expected
;
5059 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
5060 space_info
->bytes_reserved
);
5064 static inline int need_do_async_reclaim(struct btrfs_fs_info
*fs_info
,
5065 struct btrfs_space_info
*space_info
,
5066 u64 used
, bool system_chunk
)
5068 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
5070 /* If we're just plain full then async reclaim just slows us down. */
5071 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
5074 if (!btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5078 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
5079 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
5082 static void wake_all_tickets(struct list_head
*head
)
5084 struct reserve_ticket
*ticket
;
5086 while (!list_empty(head
)) {
5087 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
5088 list_del_init(&ticket
->list
);
5089 ticket
->error
= -ENOSPC
;
5090 wake_up(&ticket
->wait
);
5095 * This is for normal flushers, we can wait all goddamned day if we want to. We
5096 * will loop and continuously try to flush as long as we are making progress.
5097 * We count progress as clearing off tickets each time we have to loop.
5099 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
5101 struct btrfs_fs_info
*fs_info
;
5102 struct btrfs_space_info
*space_info
;
5105 int commit_cycles
= 0;
5106 u64 last_tickets_id
;
5108 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
5109 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5111 spin_lock(&space_info
->lock
);
5112 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5115 space_info
->flush
= 0;
5116 spin_unlock(&space_info
->lock
);
5119 last_tickets_id
= space_info
->tickets_id
;
5120 spin_unlock(&space_info
->lock
);
5122 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5124 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
5125 spin_lock(&space_info
->lock
);
5126 if (list_empty(&space_info
->tickets
)) {
5127 space_info
->flush
= 0;
5128 spin_unlock(&space_info
->lock
);
5131 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
,
5134 if (last_tickets_id
== space_info
->tickets_id
) {
5137 last_tickets_id
= space_info
->tickets_id
;
5138 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5143 if (flush_state
> COMMIT_TRANS
) {
5145 if (commit_cycles
> 2) {
5146 wake_all_tickets(&space_info
->tickets
);
5147 space_info
->flush
= 0;
5149 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5152 spin_unlock(&space_info
->lock
);
5153 } while (flush_state
<= COMMIT_TRANS
);
5156 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5158 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5161 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5162 struct btrfs_space_info
*space_info
,
5163 struct reserve_ticket
*ticket
)
5166 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5168 spin_lock(&space_info
->lock
);
5169 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5172 spin_unlock(&space_info
->lock
);
5175 spin_unlock(&space_info
->lock
);
5178 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
5180 spin_lock(&space_info
->lock
);
5181 if (ticket
->bytes
== 0) {
5182 spin_unlock(&space_info
->lock
);
5185 spin_unlock(&space_info
->lock
);
5188 * Priority flushers can't wait on delalloc without
5191 if (flush_state
== FLUSH_DELALLOC
||
5192 flush_state
== FLUSH_DELALLOC_WAIT
)
5193 flush_state
= ALLOC_CHUNK
;
5194 } while (flush_state
< COMMIT_TRANS
);
5197 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5198 struct btrfs_space_info
*space_info
,
5199 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5205 spin_lock(&space_info
->lock
);
5206 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5207 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5212 spin_unlock(&space_info
->lock
);
5216 finish_wait(&ticket
->wait
, &wait
);
5217 spin_lock(&space_info
->lock
);
5220 ret
= ticket
->error
;
5221 if (!list_empty(&ticket
->list
))
5222 list_del_init(&ticket
->list
);
5223 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5224 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5225 space_info
->bytes_may_use
-= num_bytes
;
5226 trace_btrfs_space_reservation(fs_info
, "space_info",
5227 space_info
->flags
, num_bytes
, 0);
5229 spin_unlock(&space_info
->lock
);
5235 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5236 * @root - the root we're allocating for
5237 * @space_info - the space info we want to allocate from
5238 * @orig_bytes - the number of bytes we want
5239 * @flush - whether or not we can flush to make our reservation
5241 * This will reserve orig_bytes number of bytes from the space info associated
5242 * with the block_rsv. If there is not enough space it will make an attempt to
5243 * flush out space to make room. It will do this by flushing delalloc if
5244 * possible or committing the transaction. If flush is 0 then no attempts to
5245 * regain reservations will be made and this will fail if there is not enough
5248 static int __reserve_metadata_bytes(struct btrfs_fs_info
*fs_info
,
5249 struct btrfs_space_info
*space_info
,
5251 enum btrfs_reserve_flush_enum flush
,
5254 struct reserve_ticket ticket
;
5259 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5261 spin_lock(&space_info
->lock
);
5263 used
= btrfs_space_info_used(space_info
, true);
5266 * If we have enough space then hooray, make our reservation and carry
5267 * on. If not see if we can overcommit, and if we can, hooray carry on.
5268 * If not things get more complicated.
5270 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5271 space_info
->bytes_may_use
+= orig_bytes
;
5272 trace_btrfs_space_reservation(fs_info
, "space_info",
5273 space_info
->flags
, orig_bytes
, 1);
5275 } else if (can_overcommit(fs_info
, space_info
, orig_bytes
, flush
,
5277 space_info
->bytes_may_use
+= orig_bytes
;
5278 trace_btrfs_space_reservation(fs_info
, "space_info",
5279 space_info
->flags
, orig_bytes
, 1);
5284 * If we couldn't make a reservation then setup our reservation ticket
5285 * and kick the async worker if it's not already running.
5287 * If we are a priority flusher then we just need to add our ticket to
5288 * the list and we will do our own flushing further down.
5290 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5291 ticket
.bytes
= orig_bytes
;
5293 init_waitqueue_head(&ticket
.wait
);
5294 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5295 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5296 if (!space_info
->flush
) {
5297 space_info
->flush
= 1;
5298 trace_btrfs_trigger_flush(fs_info
,
5302 queue_work(system_unbound_wq
,
5303 &fs_info
->async_reclaim_work
);
5306 list_add_tail(&ticket
.list
,
5307 &space_info
->priority_tickets
);
5309 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5312 * We will do the space reservation dance during log replay,
5313 * which means we won't have fs_info->fs_root set, so don't do
5314 * the async reclaim as we will panic.
5316 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
) &&
5317 need_do_async_reclaim(fs_info
, space_info
,
5318 used
, system_chunk
) &&
5319 !work_busy(&fs_info
->async_reclaim_work
)) {
5320 trace_btrfs_trigger_flush(fs_info
, space_info
->flags
,
5321 orig_bytes
, flush
, "preempt");
5322 queue_work(system_unbound_wq
,
5323 &fs_info
->async_reclaim_work
);
5326 spin_unlock(&space_info
->lock
);
5327 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5330 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5331 return wait_reserve_ticket(fs_info
, space_info
, &ticket
,
5335 priority_reclaim_metadata_space(fs_info
, space_info
, &ticket
);
5336 spin_lock(&space_info
->lock
);
5338 if (ticket
.bytes
< orig_bytes
) {
5339 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5340 space_info
->bytes_may_use
-= num_bytes
;
5341 trace_btrfs_space_reservation(fs_info
, "space_info",
5346 list_del_init(&ticket
.list
);
5349 spin_unlock(&space_info
->lock
);
5350 ASSERT(list_empty(&ticket
.list
));
5355 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5356 * @root - the root we're allocating for
5357 * @block_rsv - the block_rsv we're allocating for
5358 * @orig_bytes - the number of bytes we want
5359 * @flush - whether or not we can flush to make our reservation
5361 * This will reserve orgi_bytes number of bytes from the space info associated
5362 * with the block_rsv. If there is not enough space it will make an attempt to
5363 * flush out space to make room. It will do this by flushing delalloc if
5364 * possible or committing the transaction. If flush is 0 then no attempts to
5365 * regain reservations will be made and this will fail if there is not enough
5368 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5369 struct btrfs_block_rsv
*block_rsv
,
5371 enum btrfs_reserve_flush_enum flush
)
5373 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5374 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5376 bool system_chunk
= (root
== fs_info
->chunk_root
);
5378 ret
= __reserve_metadata_bytes(fs_info
, block_rsv
->space_info
,
5379 orig_bytes
, flush
, system_chunk
);
5380 if (ret
== -ENOSPC
&&
5381 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5382 if (block_rsv
!= global_rsv
&&
5383 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5386 if (ret
== -ENOSPC
) {
5387 trace_btrfs_space_reservation(fs_info
, "space_info:enospc",
5388 block_rsv
->space_info
->flags
,
5391 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
))
5392 dump_space_info(fs_info
, block_rsv
->space_info
,
5398 static struct btrfs_block_rsv
*get_block_rsv(
5399 const struct btrfs_trans_handle
*trans
,
5400 const struct btrfs_root
*root
)
5402 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5403 struct btrfs_block_rsv
*block_rsv
= NULL
;
5405 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5406 (root
== fs_info
->csum_root
&& trans
->adding_csums
) ||
5407 (root
== fs_info
->uuid_root
))
5408 block_rsv
= trans
->block_rsv
;
5411 block_rsv
= root
->block_rsv
;
5414 block_rsv
= &fs_info
->empty_block_rsv
;
5419 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5423 spin_lock(&block_rsv
->lock
);
5424 if (block_rsv
->reserved
>= num_bytes
) {
5425 block_rsv
->reserved
-= num_bytes
;
5426 if (block_rsv
->reserved
< block_rsv
->size
)
5427 block_rsv
->full
= 0;
5430 spin_unlock(&block_rsv
->lock
);
5434 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5435 u64 num_bytes
, int update_size
)
5437 spin_lock(&block_rsv
->lock
);
5438 block_rsv
->reserved
+= num_bytes
;
5440 block_rsv
->size
+= num_bytes
;
5441 else if (block_rsv
->reserved
>= block_rsv
->size
)
5442 block_rsv
->full
= 1;
5443 spin_unlock(&block_rsv
->lock
);
5446 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5447 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5450 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5453 if (global_rsv
->space_info
!= dest
->space_info
)
5456 spin_lock(&global_rsv
->lock
);
5457 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5458 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5459 spin_unlock(&global_rsv
->lock
);
5462 global_rsv
->reserved
-= num_bytes
;
5463 if (global_rsv
->reserved
< global_rsv
->size
)
5464 global_rsv
->full
= 0;
5465 spin_unlock(&global_rsv
->lock
);
5467 block_rsv_add_bytes(dest
, num_bytes
, 1);
5472 * This is for space we already have accounted in space_info->bytes_may_use, so
5473 * basically when we're returning space from block_rsv's.
5475 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5476 struct btrfs_space_info
*space_info
,
5479 struct reserve_ticket
*ticket
;
5480 struct list_head
*head
;
5482 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5483 bool check_overcommit
= false;
5485 spin_lock(&space_info
->lock
);
5486 head
= &space_info
->priority_tickets
;
5489 * If we are over our limit then we need to check and see if we can
5490 * overcommit, and if we can't then we just need to free up our space
5491 * and not satisfy any requests.
5493 used
= btrfs_space_info_used(space_info
, true);
5494 if (used
- num_bytes
>= space_info
->total_bytes
)
5495 check_overcommit
= true;
5497 while (!list_empty(head
) && num_bytes
) {
5498 ticket
= list_first_entry(head
, struct reserve_ticket
,
5501 * We use 0 bytes because this space is already reserved, so
5502 * adding the ticket space would be a double count.
5504 if (check_overcommit
&&
5505 !can_overcommit(fs_info
, space_info
, 0, flush
, false))
5507 if (num_bytes
>= ticket
->bytes
) {
5508 list_del_init(&ticket
->list
);
5509 num_bytes
-= ticket
->bytes
;
5511 space_info
->tickets_id
++;
5512 wake_up(&ticket
->wait
);
5514 ticket
->bytes
-= num_bytes
;
5519 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5520 head
= &space_info
->tickets
;
5521 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5524 space_info
->bytes_may_use
-= num_bytes
;
5525 trace_btrfs_space_reservation(fs_info
, "space_info",
5526 space_info
->flags
, num_bytes
, 0);
5527 spin_unlock(&space_info
->lock
);
5531 * This is for newly allocated space that isn't accounted in
5532 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5533 * we use this helper.
5535 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5536 struct btrfs_space_info
*space_info
,
5539 struct reserve_ticket
*ticket
;
5540 struct list_head
*head
= &space_info
->priority_tickets
;
5543 while (!list_empty(head
) && num_bytes
) {
5544 ticket
= list_first_entry(head
, struct reserve_ticket
,
5546 if (num_bytes
>= ticket
->bytes
) {
5547 trace_btrfs_space_reservation(fs_info
, "space_info",
5550 list_del_init(&ticket
->list
);
5551 num_bytes
-= ticket
->bytes
;
5552 space_info
->bytes_may_use
+= ticket
->bytes
;
5554 space_info
->tickets_id
++;
5555 wake_up(&ticket
->wait
);
5557 trace_btrfs_space_reservation(fs_info
, "space_info",
5560 space_info
->bytes_may_use
+= num_bytes
;
5561 ticket
->bytes
-= num_bytes
;
5566 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5567 head
= &space_info
->tickets
;
5572 static u64
block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5573 struct btrfs_block_rsv
*block_rsv
,
5574 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5575 u64
*qgroup_to_release_ret
)
5577 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5578 u64 qgroup_to_release
= 0;
5581 spin_lock(&block_rsv
->lock
);
5582 if (num_bytes
== (u64
)-1) {
5583 num_bytes
= block_rsv
->size
;
5584 qgroup_to_release
= block_rsv
->qgroup_rsv_size
;
5586 block_rsv
->size
-= num_bytes
;
5587 if (block_rsv
->reserved
>= block_rsv
->size
) {
5588 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5589 block_rsv
->reserved
= block_rsv
->size
;
5590 block_rsv
->full
= 1;
5594 if (block_rsv
->qgroup_rsv_reserved
>= block_rsv
->qgroup_rsv_size
) {
5595 qgroup_to_release
= block_rsv
->qgroup_rsv_reserved
-
5596 block_rsv
->qgroup_rsv_size
;
5597 block_rsv
->qgroup_rsv_reserved
= block_rsv
->qgroup_rsv_size
;
5599 qgroup_to_release
= 0;
5601 spin_unlock(&block_rsv
->lock
);
5604 if (num_bytes
> 0) {
5606 spin_lock(&dest
->lock
);
5610 bytes_to_add
= dest
->size
- dest
->reserved
;
5611 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5612 dest
->reserved
+= bytes_to_add
;
5613 if (dest
->reserved
>= dest
->size
)
5615 num_bytes
-= bytes_to_add
;
5617 spin_unlock(&dest
->lock
);
5620 space_info_add_old_bytes(fs_info
, space_info
,
5623 if (qgroup_to_release_ret
)
5624 *qgroup_to_release_ret
= qgroup_to_release
;
5628 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5629 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5634 ret
= block_rsv_use_bytes(src
, num_bytes
);
5638 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5642 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5644 memset(rsv
, 0, sizeof(*rsv
));
5645 spin_lock_init(&rsv
->lock
);
5649 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info
*fs_info
,
5650 struct btrfs_block_rsv
*rsv
,
5651 unsigned short type
)
5653 btrfs_init_block_rsv(rsv
, type
);
5654 rsv
->space_info
= __find_space_info(fs_info
,
5655 BTRFS_BLOCK_GROUP_METADATA
);
5658 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_fs_info
*fs_info
,
5659 unsigned short type
)
5661 struct btrfs_block_rsv
*block_rsv
;
5663 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5667 btrfs_init_metadata_block_rsv(fs_info
, block_rsv
, type
);
5671 void btrfs_free_block_rsv(struct btrfs_fs_info
*fs_info
,
5672 struct btrfs_block_rsv
*rsv
)
5676 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5680 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5685 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5686 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5687 enum btrfs_reserve_flush_enum flush
)
5694 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5696 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5703 int btrfs_block_rsv_check(struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5711 spin_lock(&block_rsv
->lock
);
5712 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5713 if (block_rsv
->reserved
>= num_bytes
)
5715 spin_unlock(&block_rsv
->lock
);
5720 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5721 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5722 enum btrfs_reserve_flush_enum flush
)
5730 spin_lock(&block_rsv
->lock
);
5731 num_bytes
= min_reserved
;
5732 if (block_rsv
->reserved
>= num_bytes
)
5735 num_bytes
-= block_rsv
->reserved
;
5736 spin_unlock(&block_rsv
->lock
);
5741 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5743 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5751 * btrfs_inode_rsv_refill - refill the inode block rsv.
5752 * @inode - the inode we are refilling.
5753 * @flush - the flusing restriction.
5755 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5756 * block_rsv->size as the minimum size. We'll either refill the missing amount
5757 * or return if we already have enough space. This will also handle the resreve
5758 * tracepoint for the reserved amount.
5760 static int btrfs_inode_rsv_refill(struct btrfs_inode
*inode
,
5761 enum btrfs_reserve_flush_enum flush
)
5763 struct btrfs_root
*root
= inode
->root
;
5764 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5766 u64 qgroup_num_bytes
= 0;
5769 spin_lock(&block_rsv
->lock
);
5770 if (block_rsv
->reserved
< block_rsv
->size
)
5771 num_bytes
= block_rsv
->size
- block_rsv
->reserved
;
5772 if (block_rsv
->qgroup_rsv_reserved
< block_rsv
->qgroup_rsv_size
)
5773 qgroup_num_bytes
= block_rsv
->qgroup_rsv_size
-
5774 block_rsv
->qgroup_rsv_reserved
;
5775 spin_unlock(&block_rsv
->lock
);
5780 ret
= btrfs_qgroup_reserve_meta_prealloc(root
, qgroup_num_bytes
, true);
5783 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5785 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5786 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5787 btrfs_ino(inode
), num_bytes
, 1);
5789 /* Don't forget to increase qgroup_rsv_reserved */
5790 spin_lock(&block_rsv
->lock
);
5791 block_rsv
->qgroup_rsv_reserved
+= qgroup_num_bytes
;
5792 spin_unlock(&block_rsv
->lock
);
5794 btrfs_qgroup_free_meta_prealloc(root
, qgroup_num_bytes
);
5799 * btrfs_inode_rsv_release - release any excessive reservation.
5800 * @inode - the inode we need to release from.
5801 * @qgroup_free - free or convert qgroup meta.
5802 * Unlike normal operation, qgroup meta reservation needs to know if we are
5803 * freeing qgroup reservation or just converting it into per-trans. Normally
5804 * @qgroup_free is true for error handling, and false for normal release.
5806 * This is the same as btrfs_block_rsv_release, except that it handles the
5807 * tracepoint for the reservation.
5809 static void btrfs_inode_rsv_release(struct btrfs_inode
*inode
, bool qgroup_free
)
5811 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
5812 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5813 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5815 u64 qgroup_to_release
= 0;
5818 * Since we statically set the block_rsv->size we just want to say we
5819 * are releasing 0 bytes, and then we'll just get the reservation over
5822 released
= block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, 0,
5823 &qgroup_to_release
);
5825 trace_btrfs_space_reservation(fs_info
, "delalloc",
5826 btrfs_ino(inode
), released
, 0);
5828 btrfs_qgroup_free_meta_prealloc(inode
->root
, qgroup_to_release
);
5830 btrfs_qgroup_convert_reserved_meta(inode
->root
,
5834 void btrfs_block_rsv_release(struct btrfs_fs_info
*fs_info
,
5835 struct btrfs_block_rsv
*block_rsv
,
5838 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5840 if (global_rsv
== block_rsv
||
5841 block_rsv
->space_info
!= global_rsv
->space_info
)
5843 block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, num_bytes
, NULL
);
5846 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5848 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5849 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5853 * The global block rsv is based on the size of the extent tree, the
5854 * checksum tree and the root tree. If the fs is empty we want to set
5855 * it to a minimal amount for safety.
5857 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5858 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5859 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5860 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5862 spin_lock(&sinfo
->lock
);
5863 spin_lock(&block_rsv
->lock
);
5865 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5867 if (block_rsv
->reserved
< block_rsv
->size
) {
5868 num_bytes
= btrfs_space_info_used(sinfo
, true);
5869 if (sinfo
->total_bytes
> num_bytes
) {
5870 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5871 num_bytes
= min(num_bytes
,
5872 block_rsv
->size
- block_rsv
->reserved
);
5873 block_rsv
->reserved
+= num_bytes
;
5874 sinfo
->bytes_may_use
+= num_bytes
;
5875 trace_btrfs_space_reservation(fs_info
, "space_info",
5876 sinfo
->flags
, num_bytes
,
5879 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5880 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5881 sinfo
->bytes_may_use
-= num_bytes
;
5882 trace_btrfs_space_reservation(fs_info
, "space_info",
5883 sinfo
->flags
, num_bytes
, 0);
5884 block_rsv
->reserved
= block_rsv
->size
;
5887 if (block_rsv
->reserved
== block_rsv
->size
)
5888 block_rsv
->full
= 1;
5890 block_rsv
->full
= 0;
5892 spin_unlock(&block_rsv
->lock
);
5893 spin_unlock(&sinfo
->lock
);
5896 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5898 struct btrfs_space_info
*space_info
;
5900 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5901 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5903 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5904 fs_info
->global_block_rsv
.space_info
= space_info
;
5905 fs_info
->trans_block_rsv
.space_info
= space_info
;
5906 fs_info
->empty_block_rsv
.space_info
= space_info
;
5907 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5909 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5910 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5911 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5912 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5913 if (fs_info
->quota_root
)
5914 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5915 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5917 update_global_block_rsv(fs_info
);
5920 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5922 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5924 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5925 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5926 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5927 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5928 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5929 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5934 * To be called after all the new block groups attached to the transaction
5935 * handle have been created (btrfs_create_pending_block_groups()).
5937 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5939 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5941 if (!trans
->chunk_bytes_reserved
)
5944 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5946 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5947 trans
->chunk_bytes_reserved
, NULL
);
5948 trans
->chunk_bytes_reserved
= 0;
5952 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5953 * root: the root of the parent directory
5954 * rsv: block reservation
5955 * items: the number of items that we need do reservation
5956 * qgroup_reserved: used to return the reserved size in qgroup
5958 * This function is used to reserve the space for snapshot/subvolume
5959 * creation and deletion. Those operations are different with the
5960 * common file/directory operations, they change two fs/file trees
5961 * and root tree, the number of items that the qgroup reserves is
5962 * different with the free space reservation. So we can not use
5963 * the space reservation mechanism in start_transaction().
5965 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5966 struct btrfs_block_rsv
*rsv
,
5968 bool use_global_rsv
)
5972 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5973 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5975 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
5976 /* One for parent inode, two for dir entries */
5977 num_bytes
= 3 * fs_info
->nodesize
;
5978 ret
= btrfs_qgroup_reserve_meta_prealloc(root
, num_bytes
, true);
5985 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, items
);
5986 rsv
->space_info
= __find_space_info(fs_info
,
5987 BTRFS_BLOCK_GROUP_METADATA
);
5988 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5989 BTRFS_RESERVE_FLUSH_ALL
);
5991 if (ret
== -ENOSPC
&& use_global_rsv
)
5992 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5994 if (ret
&& num_bytes
)
5995 btrfs_qgroup_free_meta_prealloc(root
, num_bytes
);
6000 void btrfs_subvolume_release_metadata(struct btrfs_fs_info
*fs_info
,
6001 struct btrfs_block_rsv
*rsv
)
6003 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
6006 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info
*fs_info
,
6007 struct btrfs_inode
*inode
)
6009 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
6010 u64 reserve_size
= 0;
6011 u64 qgroup_rsv_size
= 0;
6013 unsigned outstanding_extents
;
6015 lockdep_assert_held(&inode
->lock
);
6016 outstanding_extents
= inode
->outstanding_extents
;
6017 if (outstanding_extents
)
6018 reserve_size
= btrfs_calc_trans_metadata_size(fs_info
,
6019 outstanding_extents
+ 1);
6020 csum_leaves
= btrfs_csum_bytes_to_leaves(fs_info
,
6022 reserve_size
+= btrfs_calc_trans_metadata_size(fs_info
,
6025 * For qgroup rsv, the calculation is very simple:
6026 * account one nodesize for each outstanding extent
6028 * This is overestimating in most cases.
6030 qgroup_rsv_size
= outstanding_extents
* fs_info
->nodesize
;
6032 spin_lock(&block_rsv
->lock
);
6033 block_rsv
->size
= reserve_size
;
6034 block_rsv
->qgroup_rsv_size
= qgroup_rsv_size
;
6035 spin_unlock(&block_rsv
->lock
);
6038 int btrfs_delalloc_reserve_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6040 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6041 unsigned nr_extents
;
6042 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
6044 bool delalloc_lock
= true;
6046 /* If we are a free space inode we need to not flush since we will be in
6047 * the middle of a transaction commit. We also don't need the delalloc
6048 * mutex since we won't race with anybody. We need this mostly to make
6049 * lockdep shut its filthy mouth.
6051 * If we have a transaction open (can happen if we call truncate_block
6052 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6054 if (btrfs_is_free_space_inode(inode
)) {
6055 flush
= BTRFS_RESERVE_NO_FLUSH
;
6056 delalloc_lock
= false;
6058 if (current
->journal_info
)
6059 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
6061 if (btrfs_transaction_in_commit(fs_info
))
6062 schedule_timeout(1);
6066 mutex_lock(&inode
->delalloc_mutex
);
6068 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6070 /* Add our new extents and calculate the new rsv size. */
6071 spin_lock(&inode
->lock
);
6072 nr_extents
= count_max_extents(num_bytes
);
6073 btrfs_mod_outstanding_extents(inode
, nr_extents
);
6074 inode
->csum_bytes
+= num_bytes
;
6075 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6076 spin_unlock(&inode
->lock
);
6078 ret
= btrfs_inode_rsv_refill(inode
, flush
);
6083 mutex_unlock(&inode
->delalloc_mutex
);
6087 spin_lock(&inode
->lock
);
6088 nr_extents
= count_max_extents(num_bytes
);
6089 btrfs_mod_outstanding_extents(inode
, -nr_extents
);
6090 inode
->csum_bytes
-= num_bytes
;
6091 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6092 spin_unlock(&inode
->lock
);
6094 btrfs_inode_rsv_release(inode
, true);
6096 mutex_unlock(&inode
->delalloc_mutex
);
6101 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6102 * @inode: the inode to release the reservation for.
6103 * @num_bytes: the number of bytes we are releasing.
6104 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6106 * This will release the metadata reservation for an inode. This can be called
6107 * once we complete IO for a given set of bytes to release their metadata
6108 * reservations, or on error for the same reason.
6110 void btrfs_delalloc_release_metadata(struct btrfs_inode
*inode
, u64 num_bytes
,
6113 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6115 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6116 spin_lock(&inode
->lock
);
6117 inode
->csum_bytes
-= num_bytes
;
6118 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6119 spin_unlock(&inode
->lock
);
6121 if (btrfs_is_testing(fs_info
))
6124 btrfs_inode_rsv_release(inode
, qgroup_free
);
6128 * btrfs_delalloc_release_extents - release our outstanding_extents
6129 * @inode: the inode to balance the reservation for.
6130 * @num_bytes: the number of bytes we originally reserved with
6131 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6133 * When we reserve space we increase outstanding_extents for the extents we may
6134 * add. Once we've set the range as delalloc or created our ordered extents we
6135 * have outstanding_extents to track the real usage, so we use this to free our
6136 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6137 * with btrfs_delalloc_reserve_metadata.
6139 void btrfs_delalloc_release_extents(struct btrfs_inode
*inode
, u64 num_bytes
,
6142 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6143 unsigned num_extents
;
6145 spin_lock(&inode
->lock
);
6146 num_extents
= count_max_extents(num_bytes
);
6147 btrfs_mod_outstanding_extents(inode
, -num_extents
);
6148 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6149 spin_unlock(&inode
->lock
);
6151 if (btrfs_is_testing(fs_info
))
6154 btrfs_inode_rsv_release(inode
, qgroup_free
);
6158 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6160 * @inode: inode we're writing to
6161 * @start: start range we are writing to
6162 * @len: how long the range we are writing to
6163 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6164 * current reservation.
6166 * This will do the following things
6168 * o reserve space in data space info for num bytes
6169 * and reserve precious corresponding qgroup space
6170 * (Done in check_data_free_space)
6172 * o reserve space for metadata space, based on the number of outstanding
6173 * extents and how much csums will be needed
6174 * also reserve metadata space in a per root over-reserve method.
6175 * o add to the inodes->delalloc_bytes
6176 * o add it to the fs_info's delalloc inodes list.
6177 * (Above 3 all done in delalloc_reserve_metadata)
6179 * Return 0 for success
6180 * Return <0 for error(-ENOSPC or -EQUOT)
6182 int btrfs_delalloc_reserve_space(struct inode
*inode
,
6183 struct extent_changeset
**reserved
, u64 start
, u64 len
)
6187 ret
= btrfs_check_data_free_space(inode
, reserved
, start
, len
);
6190 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
);
6192 btrfs_free_reserved_data_space(inode
, *reserved
, start
, len
);
6197 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6198 * @inode: inode we're releasing space for
6199 * @start: start position of the space already reserved
6200 * @len: the len of the space already reserved
6201 * @release_bytes: the len of the space we consumed or didn't use
6203 * This function will release the metadata space that was not used and will
6204 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6205 * list if there are no delalloc bytes left.
6206 * Also it will handle the qgroup reserved space.
6208 void btrfs_delalloc_release_space(struct inode
*inode
,
6209 struct extent_changeset
*reserved
,
6210 u64 start
, u64 len
, bool qgroup_free
)
6212 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
, qgroup_free
);
6213 btrfs_free_reserved_data_space(inode
, reserved
, start
, len
);
6216 static int update_block_group(struct btrfs_trans_handle
*trans
,
6217 struct btrfs_fs_info
*info
, u64 bytenr
,
6218 u64 num_bytes
, int alloc
)
6220 struct btrfs_block_group_cache
*cache
= NULL
;
6221 u64 total
= num_bytes
;
6226 /* block accounting for super block */
6227 spin_lock(&info
->delalloc_root_lock
);
6228 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6230 old_val
+= num_bytes
;
6232 old_val
-= num_bytes
;
6233 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6234 spin_unlock(&info
->delalloc_root_lock
);
6237 cache
= btrfs_lookup_block_group(info
, bytenr
);
6240 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6241 BTRFS_BLOCK_GROUP_RAID1
|
6242 BTRFS_BLOCK_GROUP_RAID10
))
6247 * If this block group has free space cache written out, we
6248 * need to make sure to load it if we are removing space. This
6249 * is because we need the unpinning stage to actually add the
6250 * space back to the block group, otherwise we will leak space.
6252 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6253 cache_block_group(cache
, 1);
6255 byte_in_group
= bytenr
- cache
->key
.objectid
;
6256 WARN_ON(byte_in_group
> cache
->key
.offset
);
6258 spin_lock(&cache
->space_info
->lock
);
6259 spin_lock(&cache
->lock
);
6261 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
6262 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6263 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6265 old_val
= btrfs_block_group_used(&cache
->item
);
6266 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6268 old_val
+= num_bytes
;
6269 btrfs_set_block_group_used(&cache
->item
, old_val
);
6270 cache
->reserved
-= num_bytes
;
6271 cache
->space_info
->bytes_reserved
-= num_bytes
;
6272 cache
->space_info
->bytes_used
+= num_bytes
;
6273 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6274 spin_unlock(&cache
->lock
);
6275 spin_unlock(&cache
->space_info
->lock
);
6277 old_val
-= num_bytes
;
6278 btrfs_set_block_group_used(&cache
->item
, old_val
);
6279 cache
->pinned
+= num_bytes
;
6280 cache
->space_info
->bytes_pinned
+= num_bytes
;
6281 cache
->space_info
->bytes_used
-= num_bytes
;
6282 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6283 spin_unlock(&cache
->lock
);
6284 spin_unlock(&cache
->space_info
->lock
);
6286 trace_btrfs_space_reservation(info
, "pinned",
6287 cache
->space_info
->flags
,
6289 percpu_counter_add(&cache
->space_info
->total_bytes_pinned
,
6291 set_extent_dirty(info
->pinned_extents
,
6292 bytenr
, bytenr
+ num_bytes
- 1,
6293 GFP_NOFS
| __GFP_NOFAIL
);
6296 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6297 if (list_empty(&cache
->dirty_list
)) {
6298 list_add_tail(&cache
->dirty_list
,
6299 &trans
->transaction
->dirty_bgs
);
6300 trans
->transaction
->num_dirty_bgs
++;
6301 btrfs_get_block_group(cache
);
6303 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6306 * No longer have used bytes in this block group, queue it for
6307 * deletion. We do this after adding the block group to the
6308 * dirty list to avoid races between cleaner kthread and space
6311 if (!alloc
&& old_val
== 0) {
6312 spin_lock(&info
->unused_bgs_lock
);
6313 if (list_empty(&cache
->bg_list
)) {
6314 btrfs_get_block_group(cache
);
6315 trace_btrfs_add_unused_block_group(cache
);
6316 list_add_tail(&cache
->bg_list
,
6319 spin_unlock(&info
->unused_bgs_lock
);
6322 btrfs_put_block_group(cache
);
6324 bytenr
+= num_bytes
;
6329 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
, u64 search_start
)
6331 struct btrfs_block_group_cache
*cache
;
6334 spin_lock(&fs_info
->block_group_cache_lock
);
6335 bytenr
= fs_info
->first_logical_byte
;
6336 spin_unlock(&fs_info
->block_group_cache_lock
);
6338 if (bytenr
< (u64
)-1)
6341 cache
= btrfs_lookup_first_block_group(fs_info
, search_start
);
6345 bytenr
= cache
->key
.objectid
;
6346 btrfs_put_block_group(cache
);
6351 static int pin_down_extent(struct btrfs_fs_info
*fs_info
,
6352 struct btrfs_block_group_cache
*cache
,
6353 u64 bytenr
, u64 num_bytes
, int reserved
)
6355 spin_lock(&cache
->space_info
->lock
);
6356 spin_lock(&cache
->lock
);
6357 cache
->pinned
+= num_bytes
;
6358 cache
->space_info
->bytes_pinned
+= num_bytes
;
6360 cache
->reserved
-= num_bytes
;
6361 cache
->space_info
->bytes_reserved
-= num_bytes
;
6363 spin_unlock(&cache
->lock
);
6364 spin_unlock(&cache
->space_info
->lock
);
6366 trace_btrfs_space_reservation(fs_info
, "pinned",
6367 cache
->space_info
->flags
, num_bytes
, 1);
6368 percpu_counter_add(&cache
->space_info
->total_bytes_pinned
, num_bytes
);
6369 set_extent_dirty(fs_info
->pinned_extents
, bytenr
,
6370 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6375 * this function must be called within transaction
6377 int btrfs_pin_extent(struct btrfs_fs_info
*fs_info
,
6378 u64 bytenr
, u64 num_bytes
, int reserved
)
6380 struct btrfs_block_group_cache
*cache
;
6382 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6383 BUG_ON(!cache
); /* Logic error */
6385 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, reserved
);
6387 btrfs_put_block_group(cache
);
6392 * this function must be called within transaction
6394 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info
*fs_info
,
6395 u64 bytenr
, u64 num_bytes
)
6397 struct btrfs_block_group_cache
*cache
;
6400 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6405 * pull in the free space cache (if any) so that our pin
6406 * removes the free space from the cache. We have load_only set
6407 * to one because the slow code to read in the free extents does check
6408 * the pinned extents.
6410 cache_block_group(cache
, 1);
6412 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, 0);
6414 /* remove us from the free space cache (if we're there at all) */
6415 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6416 btrfs_put_block_group(cache
);
6420 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
6421 u64 start
, u64 num_bytes
)
6424 struct btrfs_block_group_cache
*block_group
;
6425 struct btrfs_caching_control
*caching_ctl
;
6427 block_group
= btrfs_lookup_block_group(fs_info
, start
);
6431 cache_block_group(block_group
, 0);
6432 caching_ctl
= get_caching_control(block_group
);
6436 BUG_ON(!block_group_cache_done(block_group
));
6437 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6439 mutex_lock(&caching_ctl
->mutex
);
6441 if (start
>= caching_ctl
->progress
) {
6442 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6443 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6444 ret
= btrfs_remove_free_space(block_group
,
6447 num_bytes
= caching_ctl
->progress
- start
;
6448 ret
= btrfs_remove_free_space(block_group
,
6453 num_bytes
= (start
+ num_bytes
) -
6454 caching_ctl
->progress
;
6455 start
= caching_ctl
->progress
;
6456 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6459 mutex_unlock(&caching_ctl
->mutex
);
6460 put_caching_control(caching_ctl
);
6462 btrfs_put_block_group(block_group
);
6466 int btrfs_exclude_logged_extents(struct btrfs_fs_info
*fs_info
,
6467 struct extent_buffer
*eb
)
6469 struct btrfs_file_extent_item
*item
;
6470 struct btrfs_key key
;
6475 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
6478 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6479 btrfs_item_key_to_cpu(eb
, &key
, i
);
6480 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6482 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6483 found_type
= btrfs_file_extent_type(eb
, item
);
6484 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6486 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6488 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6489 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6490 ret
= __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
6499 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6501 atomic_inc(&bg
->reservations
);
6504 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6507 struct btrfs_block_group_cache
*bg
;
6509 bg
= btrfs_lookup_block_group(fs_info
, start
);
6511 if (atomic_dec_and_test(&bg
->reservations
))
6512 wake_up_var(&bg
->reservations
);
6513 btrfs_put_block_group(bg
);
6516 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6518 struct btrfs_space_info
*space_info
= bg
->space_info
;
6522 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6526 * Our block group is read only but before we set it to read only,
6527 * some task might have had allocated an extent from it already, but it
6528 * has not yet created a respective ordered extent (and added it to a
6529 * root's list of ordered extents).
6530 * Therefore wait for any task currently allocating extents, since the
6531 * block group's reservations counter is incremented while a read lock
6532 * on the groups' semaphore is held and decremented after releasing
6533 * the read access on that semaphore and creating the ordered extent.
6535 down_write(&space_info
->groups_sem
);
6536 up_write(&space_info
->groups_sem
);
6538 wait_var_event(&bg
->reservations
, !atomic_read(&bg
->reservations
));
6542 * btrfs_add_reserved_bytes - update the block_group and space info counters
6543 * @cache: The cache we are manipulating
6544 * @ram_bytes: The number of bytes of file content, and will be same to
6545 * @num_bytes except for the compress path.
6546 * @num_bytes: The number of bytes in question
6547 * @delalloc: The blocks are allocated for the delalloc write
6549 * This is called by the allocator when it reserves space. If this is a
6550 * reservation and the block group has become read only we cannot make the
6551 * reservation and return -EAGAIN, otherwise this function always succeeds.
6553 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6554 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6556 struct btrfs_space_info
*space_info
= cache
->space_info
;
6559 spin_lock(&space_info
->lock
);
6560 spin_lock(&cache
->lock
);
6564 cache
->reserved
+= num_bytes
;
6565 space_info
->bytes_reserved
+= num_bytes
;
6567 trace_btrfs_space_reservation(cache
->fs_info
,
6568 "space_info", space_info
->flags
,
6570 space_info
->bytes_may_use
-= ram_bytes
;
6572 cache
->delalloc_bytes
+= num_bytes
;
6574 spin_unlock(&cache
->lock
);
6575 spin_unlock(&space_info
->lock
);
6580 * btrfs_free_reserved_bytes - update the block_group and space info counters
6581 * @cache: The cache we are manipulating
6582 * @num_bytes: The number of bytes in question
6583 * @delalloc: The blocks are allocated for the delalloc write
6585 * This is called by somebody who is freeing space that was never actually used
6586 * on disk. For example if you reserve some space for a new leaf in transaction
6587 * A and before transaction A commits you free that leaf, you call this with
6588 * reserve set to 0 in order to clear the reservation.
6591 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6592 u64 num_bytes
, int delalloc
)
6594 struct btrfs_space_info
*space_info
= cache
->space_info
;
6597 spin_lock(&space_info
->lock
);
6598 spin_lock(&cache
->lock
);
6600 space_info
->bytes_readonly
+= num_bytes
;
6601 cache
->reserved
-= num_bytes
;
6602 space_info
->bytes_reserved
-= num_bytes
;
6605 cache
->delalloc_bytes
-= num_bytes
;
6606 spin_unlock(&cache
->lock
);
6607 spin_unlock(&space_info
->lock
);
6610 void btrfs_prepare_extent_commit(struct btrfs_fs_info
*fs_info
)
6612 struct btrfs_caching_control
*next
;
6613 struct btrfs_caching_control
*caching_ctl
;
6614 struct btrfs_block_group_cache
*cache
;
6616 down_write(&fs_info
->commit_root_sem
);
6618 list_for_each_entry_safe(caching_ctl
, next
,
6619 &fs_info
->caching_block_groups
, list
) {
6620 cache
= caching_ctl
->block_group
;
6621 if (block_group_cache_done(cache
)) {
6622 cache
->last_byte_to_unpin
= (u64
)-1;
6623 list_del_init(&caching_ctl
->list
);
6624 put_caching_control(caching_ctl
);
6626 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6630 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6631 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6633 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6635 up_write(&fs_info
->commit_root_sem
);
6637 update_global_block_rsv(fs_info
);
6641 * Returns the free cluster for the given space info and sets empty_cluster to
6642 * what it should be based on the mount options.
6644 static struct btrfs_free_cluster
*
6645 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
6646 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
6648 struct btrfs_free_cluster
*ret
= NULL
;
6651 if (btrfs_mixed_space_info(space_info
))
6654 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6655 ret
= &fs_info
->meta_alloc_cluster
;
6656 if (btrfs_test_opt(fs_info
, SSD
))
6657 *empty_cluster
= SZ_2M
;
6659 *empty_cluster
= SZ_64K
;
6660 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
6661 btrfs_test_opt(fs_info
, SSD_SPREAD
)) {
6662 *empty_cluster
= SZ_2M
;
6663 ret
= &fs_info
->data_alloc_cluster
;
6669 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
6671 const bool return_free_space
)
6673 struct btrfs_block_group_cache
*cache
= NULL
;
6674 struct btrfs_space_info
*space_info
;
6675 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6676 struct btrfs_free_cluster
*cluster
= NULL
;
6678 u64 total_unpinned
= 0;
6679 u64 empty_cluster
= 0;
6682 while (start
<= end
) {
6685 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6687 btrfs_put_block_group(cache
);
6689 cache
= btrfs_lookup_block_group(fs_info
, start
);
6690 BUG_ON(!cache
); /* Logic error */
6692 cluster
= fetch_cluster_info(fs_info
,
6695 empty_cluster
<<= 1;
6698 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6699 len
= min(len
, end
+ 1 - start
);
6701 if (start
< cache
->last_byte_to_unpin
) {
6702 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6703 if (return_free_space
)
6704 btrfs_add_free_space(cache
, start
, len
);
6708 total_unpinned
+= len
;
6709 space_info
= cache
->space_info
;
6712 * If this space cluster has been marked as fragmented and we've
6713 * unpinned enough in this block group to potentially allow a
6714 * cluster to be created inside of it go ahead and clear the
6717 if (cluster
&& cluster
->fragmented
&&
6718 total_unpinned
> empty_cluster
) {
6719 spin_lock(&cluster
->lock
);
6720 cluster
->fragmented
= 0;
6721 spin_unlock(&cluster
->lock
);
6724 spin_lock(&space_info
->lock
);
6725 spin_lock(&cache
->lock
);
6726 cache
->pinned
-= len
;
6727 space_info
->bytes_pinned
-= len
;
6729 trace_btrfs_space_reservation(fs_info
, "pinned",
6730 space_info
->flags
, len
, 0);
6731 space_info
->max_extent_size
= 0;
6732 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6734 space_info
->bytes_readonly
+= len
;
6737 spin_unlock(&cache
->lock
);
6738 if (!readonly
&& return_free_space
&&
6739 global_rsv
->space_info
== space_info
) {
6742 spin_lock(&global_rsv
->lock
);
6743 if (!global_rsv
->full
) {
6744 to_add
= min(len
, global_rsv
->size
-
6745 global_rsv
->reserved
);
6746 global_rsv
->reserved
+= to_add
;
6747 space_info
->bytes_may_use
+= to_add
;
6748 if (global_rsv
->reserved
>= global_rsv
->size
)
6749 global_rsv
->full
= 1;
6750 trace_btrfs_space_reservation(fs_info
,
6756 spin_unlock(&global_rsv
->lock
);
6757 /* Add to any tickets we may have */
6759 space_info_add_new_bytes(fs_info
, space_info
,
6762 spin_unlock(&space_info
->lock
);
6766 btrfs_put_block_group(cache
);
6770 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
)
6772 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
6773 struct btrfs_block_group_cache
*block_group
, *tmp
;
6774 struct list_head
*deleted_bgs
;
6775 struct extent_io_tree
*unpin
;
6780 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6781 unpin
= &fs_info
->freed_extents
[1];
6783 unpin
= &fs_info
->freed_extents
[0];
6785 while (!trans
->aborted
) {
6786 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6787 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6788 EXTENT_DIRTY
, NULL
);
6790 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6794 if (btrfs_test_opt(fs_info
, DISCARD
))
6795 ret
= btrfs_discard_extent(fs_info
, start
,
6796 end
+ 1 - start
, NULL
);
6798 clear_extent_dirty(unpin
, start
, end
);
6799 unpin_extent_range(fs_info
, start
, end
, true);
6800 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6805 * Transaction is finished. We don't need the lock anymore. We
6806 * do need to clean up the block groups in case of a transaction
6809 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6810 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6814 if (!trans
->aborted
)
6815 ret
= btrfs_discard_extent(fs_info
,
6816 block_group
->key
.objectid
,
6817 block_group
->key
.offset
,
6820 list_del_init(&block_group
->bg_list
);
6821 btrfs_put_block_group_trimming(block_group
);
6822 btrfs_put_block_group(block_group
);
6825 const char *errstr
= btrfs_decode_error(ret
);
6827 "discard failed while removing blockgroup: errno=%d %s",
6835 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6836 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6837 u64 root_objectid
, u64 owner_objectid
,
6838 u64 owner_offset
, int refs_to_drop
,
6839 struct btrfs_delayed_extent_op
*extent_op
)
6841 struct btrfs_fs_info
*info
= trans
->fs_info
;
6842 struct btrfs_key key
;
6843 struct btrfs_path
*path
;
6844 struct btrfs_root
*extent_root
= info
->extent_root
;
6845 struct extent_buffer
*leaf
;
6846 struct btrfs_extent_item
*ei
;
6847 struct btrfs_extent_inline_ref
*iref
;
6850 int extent_slot
= 0;
6851 int found_extent
= 0;
6855 u64 bytenr
= node
->bytenr
;
6856 u64 num_bytes
= node
->num_bytes
;
6858 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
6860 path
= btrfs_alloc_path();
6864 path
->reada
= READA_FORWARD
;
6865 path
->leave_spinning
= 1;
6867 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6868 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6871 skinny_metadata
= false;
6873 ret
= lookup_extent_backref(trans
, path
, &iref
, bytenr
, num_bytes
,
6874 parent
, root_objectid
, owner_objectid
,
6877 extent_slot
= path
->slots
[0];
6878 while (extent_slot
>= 0) {
6879 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6881 if (key
.objectid
!= bytenr
)
6883 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6884 key
.offset
== num_bytes
) {
6888 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6889 key
.offset
== owner_objectid
) {
6893 if (path
->slots
[0] - extent_slot
> 5)
6897 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6898 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6899 if (found_extent
&& item_size
< sizeof(*ei
))
6902 if (!found_extent
) {
6904 ret
= remove_extent_backref(trans
, info
, path
, NULL
,
6906 is_data
, &last_ref
);
6908 btrfs_abort_transaction(trans
, ret
);
6911 btrfs_release_path(path
);
6912 path
->leave_spinning
= 1;
6914 key
.objectid
= bytenr
;
6915 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6916 key
.offset
= num_bytes
;
6918 if (!is_data
&& skinny_metadata
) {
6919 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6920 key
.offset
= owner_objectid
;
6923 ret
= btrfs_search_slot(trans
, extent_root
,
6925 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6927 * Couldn't find our skinny metadata item,
6928 * see if we have ye olde extent item.
6931 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6933 if (key
.objectid
== bytenr
&&
6934 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6935 key
.offset
== num_bytes
)
6939 if (ret
> 0 && skinny_metadata
) {
6940 skinny_metadata
= false;
6941 key
.objectid
= bytenr
;
6942 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6943 key
.offset
= num_bytes
;
6944 btrfs_release_path(path
);
6945 ret
= btrfs_search_slot(trans
, extent_root
,
6951 "umm, got %d back from search, was looking for %llu",
6954 btrfs_print_leaf(path
->nodes
[0]);
6957 btrfs_abort_transaction(trans
, ret
);
6960 extent_slot
= path
->slots
[0];
6962 } else if (WARN_ON(ret
== -ENOENT
)) {
6963 btrfs_print_leaf(path
->nodes
[0]);
6965 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6966 bytenr
, parent
, root_objectid
, owner_objectid
,
6968 btrfs_abort_transaction(trans
, ret
);
6971 btrfs_abort_transaction(trans
, ret
);
6975 leaf
= path
->nodes
[0];
6976 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6977 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6978 if (item_size
< sizeof(*ei
)) {
6979 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6980 ret
= convert_extent_item_v0(trans
, info
, path
, owner_objectid
,
6983 btrfs_abort_transaction(trans
, ret
);
6987 btrfs_release_path(path
);
6988 path
->leave_spinning
= 1;
6990 key
.objectid
= bytenr
;
6991 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6992 key
.offset
= num_bytes
;
6994 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6998 "umm, got %d back from search, was looking for %llu",
7000 btrfs_print_leaf(path
->nodes
[0]);
7003 btrfs_abort_transaction(trans
, ret
);
7007 extent_slot
= path
->slots
[0];
7008 leaf
= path
->nodes
[0];
7009 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
7012 BUG_ON(item_size
< sizeof(*ei
));
7013 ei
= btrfs_item_ptr(leaf
, extent_slot
,
7014 struct btrfs_extent_item
);
7015 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
7016 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
7017 struct btrfs_tree_block_info
*bi
;
7018 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
7019 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
7020 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
7023 refs
= btrfs_extent_refs(leaf
, ei
);
7024 if (refs
< refs_to_drop
) {
7026 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7027 refs_to_drop
, refs
, bytenr
);
7029 btrfs_abort_transaction(trans
, ret
);
7032 refs
-= refs_to_drop
;
7036 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7038 * In the case of inline back ref, reference count will
7039 * be updated by remove_extent_backref
7042 BUG_ON(!found_extent
);
7044 btrfs_set_extent_refs(leaf
, ei
, refs
);
7045 btrfs_mark_buffer_dirty(leaf
);
7048 ret
= remove_extent_backref(trans
, info
, path
,
7050 is_data
, &last_ref
);
7052 btrfs_abort_transaction(trans
, ret
);
7058 BUG_ON(is_data
&& refs_to_drop
!=
7059 extent_data_ref_count(path
, iref
));
7061 BUG_ON(path
->slots
[0] != extent_slot
);
7063 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7064 path
->slots
[0] = extent_slot
;
7070 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7073 btrfs_abort_transaction(trans
, ret
);
7076 btrfs_release_path(path
);
7079 ret
= btrfs_del_csums(trans
, info
, bytenr
, num_bytes
);
7081 btrfs_abort_transaction(trans
, ret
);
7086 ret
= add_to_free_space_tree(trans
, bytenr
, num_bytes
);
7088 btrfs_abort_transaction(trans
, ret
);
7092 ret
= update_block_group(trans
, info
, bytenr
, num_bytes
, 0);
7094 btrfs_abort_transaction(trans
, ret
);
7098 btrfs_release_path(path
);
7101 btrfs_free_path(path
);
7106 * when we free an block, it is possible (and likely) that we free the last
7107 * delayed ref for that extent as well. This searches the delayed ref tree for
7108 * a given extent, and if there are no other delayed refs to be processed, it
7109 * removes it from the tree.
7111 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7114 struct btrfs_delayed_ref_head
*head
;
7115 struct btrfs_delayed_ref_root
*delayed_refs
;
7118 delayed_refs
= &trans
->transaction
->delayed_refs
;
7119 spin_lock(&delayed_refs
->lock
);
7120 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
7122 goto out_delayed_unlock
;
7124 spin_lock(&head
->lock
);
7125 if (!RB_EMPTY_ROOT(&head
->ref_tree
))
7128 if (head
->extent_op
) {
7129 if (!head
->must_insert_reserved
)
7131 btrfs_free_delayed_extent_op(head
->extent_op
);
7132 head
->extent_op
= NULL
;
7136 * waiting for the lock here would deadlock. If someone else has it
7137 * locked they are already in the process of dropping it anyway
7139 if (!mutex_trylock(&head
->mutex
))
7143 * at this point we have a head with no other entries. Go
7144 * ahead and process it.
7146 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7147 RB_CLEAR_NODE(&head
->href_node
);
7148 atomic_dec(&delayed_refs
->num_entries
);
7151 * we don't take a ref on the node because we're removing it from the
7152 * tree, so we just steal the ref the tree was holding.
7154 delayed_refs
->num_heads
--;
7155 if (head
->processing
== 0)
7156 delayed_refs
->num_heads_ready
--;
7157 head
->processing
= 0;
7158 spin_unlock(&head
->lock
);
7159 spin_unlock(&delayed_refs
->lock
);
7161 BUG_ON(head
->extent_op
);
7162 if (head
->must_insert_reserved
)
7165 mutex_unlock(&head
->mutex
);
7166 btrfs_put_delayed_ref_head(head
);
7169 spin_unlock(&head
->lock
);
7172 spin_unlock(&delayed_refs
->lock
);
7176 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7177 struct btrfs_root
*root
,
7178 struct extent_buffer
*buf
,
7179 u64 parent
, int last_ref
)
7181 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7185 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7186 int old_ref_mod
, new_ref_mod
;
7188 btrfs_ref_tree_mod(root
, buf
->start
, buf
->len
, parent
,
7189 root
->root_key
.objectid
,
7190 btrfs_header_level(buf
), 0,
7191 BTRFS_DROP_DELAYED_REF
);
7192 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
,
7194 root
->root_key
.objectid
,
7195 btrfs_header_level(buf
),
7196 BTRFS_DROP_DELAYED_REF
, NULL
,
7197 &old_ref_mod
, &new_ref_mod
);
7198 BUG_ON(ret
); /* -ENOMEM */
7199 pin
= old_ref_mod
>= 0 && new_ref_mod
< 0;
7202 if (last_ref
&& btrfs_header_generation(buf
) == trans
->transid
) {
7203 struct btrfs_block_group_cache
*cache
;
7205 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7206 ret
= check_ref_cleanup(trans
, buf
->start
);
7212 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
7214 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7215 pin_down_extent(fs_info
, cache
, buf
->start
,
7217 btrfs_put_block_group(cache
);
7221 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7223 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7224 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7225 btrfs_put_block_group(cache
);
7226 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
7230 add_pinned_bytes(fs_info
, buf
->len
, true,
7231 root
->root_key
.objectid
);
7235 * Deleting the buffer, clear the corrupt flag since it doesn't
7238 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7242 /* Can return -ENOMEM */
7243 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
7244 struct btrfs_root
*root
,
7245 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7246 u64 owner
, u64 offset
)
7248 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7249 int old_ref_mod
, new_ref_mod
;
7252 if (btrfs_is_testing(fs_info
))
7255 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
)
7256 btrfs_ref_tree_mod(root
, bytenr
, num_bytes
, parent
,
7257 root_objectid
, owner
, offset
,
7258 BTRFS_DROP_DELAYED_REF
);
7261 * tree log blocks never actually go into the extent allocation
7262 * tree, just update pinning info and exit early.
7264 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7265 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7266 /* unlocks the pinned mutex */
7267 btrfs_pin_extent(fs_info
, bytenr
, num_bytes
, 1);
7268 old_ref_mod
= new_ref_mod
= 0;
7270 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7271 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
,
7273 root_objectid
, (int)owner
,
7274 BTRFS_DROP_DELAYED_REF
, NULL
,
7275 &old_ref_mod
, &new_ref_mod
);
7277 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
,
7279 root_objectid
, owner
, offset
,
7280 0, BTRFS_DROP_DELAYED_REF
,
7281 &old_ref_mod
, &new_ref_mod
);
7284 if (ret
== 0 && old_ref_mod
>= 0 && new_ref_mod
< 0) {
7285 bool metadata
= owner
< BTRFS_FIRST_FREE_OBJECTID
;
7287 add_pinned_bytes(fs_info
, num_bytes
, metadata
, root_objectid
);
7294 * when we wait for progress in the block group caching, its because
7295 * our allocation attempt failed at least once. So, we must sleep
7296 * and let some progress happen before we try again.
7298 * This function will sleep at least once waiting for new free space to
7299 * show up, and then it will check the block group free space numbers
7300 * for our min num_bytes. Another option is to have it go ahead
7301 * and look in the rbtree for a free extent of a given size, but this
7304 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7305 * any of the information in this block group.
7307 static noinline
void
7308 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7311 struct btrfs_caching_control
*caching_ctl
;
7313 caching_ctl
= get_caching_control(cache
);
7317 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7318 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7320 put_caching_control(caching_ctl
);
7324 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7326 struct btrfs_caching_control
*caching_ctl
;
7329 caching_ctl
= get_caching_control(cache
);
7331 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7333 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7334 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7336 put_caching_control(caching_ctl
);
7340 enum btrfs_loop_type
{
7341 LOOP_CACHING_NOWAIT
= 0,
7342 LOOP_CACHING_WAIT
= 1,
7343 LOOP_ALLOC_CHUNK
= 2,
7344 LOOP_NO_EMPTY_SIZE
= 3,
7348 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7352 down_read(&cache
->data_rwsem
);
7356 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7359 btrfs_get_block_group(cache
);
7361 down_read(&cache
->data_rwsem
);
7364 static struct btrfs_block_group_cache
*
7365 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7366 struct btrfs_free_cluster
*cluster
,
7369 struct btrfs_block_group_cache
*used_bg
= NULL
;
7371 spin_lock(&cluster
->refill_lock
);
7373 used_bg
= cluster
->block_group
;
7377 if (used_bg
== block_group
)
7380 btrfs_get_block_group(used_bg
);
7385 if (down_read_trylock(&used_bg
->data_rwsem
))
7388 spin_unlock(&cluster
->refill_lock
);
7390 /* We should only have one-level nested. */
7391 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
7393 spin_lock(&cluster
->refill_lock
);
7394 if (used_bg
== cluster
->block_group
)
7397 up_read(&used_bg
->data_rwsem
);
7398 btrfs_put_block_group(used_bg
);
7403 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7407 up_read(&cache
->data_rwsem
);
7408 btrfs_put_block_group(cache
);
7412 * walks the btree of allocated extents and find a hole of a given size.
7413 * The key ins is changed to record the hole:
7414 * ins->objectid == start position
7415 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7416 * ins->offset == the size of the hole.
7417 * Any available blocks before search_start are skipped.
7419 * If there is no suitable free space, we will record the max size of
7420 * the free space extent currently.
7422 static noinline
int find_free_extent(struct btrfs_fs_info
*fs_info
,
7423 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7424 u64 hint_byte
, struct btrfs_key
*ins
,
7425 u64 flags
, int delalloc
)
7428 struct btrfs_root
*root
= fs_info
->extent_root
;
7429 struct btrfs_free_cluster
*last_ptr
= NULL
;
7430 struct btrfs_block_group_cache
*block_group
= NULL
;
7431 u64 search_start
= 0;
7432 u64 max_extent_size
= 0;
7433 u64 empty_cluster
= 0;
7434 struct btrfs_space_info
*space_info
;
7436 int index
= btrfs_bg_flags_to_raid_index(flags
);
7437 bool failed_cluster_refill
= false;
7438 bool failed_alloc
= false;
7439 bool use_cluster
= true;
7440 bool have_caching_bg
= false;
7441 bool orig_have_caching_bg
= false;
7442 bool full_search
= false;
7444 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7445 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7449 trace_find_free_extent(fs_info
, num_bytes
, empty_size
, flags
);
7451 space_info
= __find_space_info(fs_info
, flags
);
7453 btrfs_err(fs_info
, "No space info for %llu", flags
);
7458 * If our free space is heavily fragmented we may not be able to make
7459 * big contiguous allocations, so instead of doing the expensive search
7460 * for free space, simply return ENOSPC with our max_extent_size so we
7461 * can go ahead and search for a more manageable chunk.
7463 * If our max_extent_size is large enough for our allocation simply
7464 * disable clustering since we will likely not be able to find enough
7465 * space to create a cluster and induce latency trying.
7467 if (unlikely(space_info
->max_extent_size
)) {
7468 spin_lock(&space_info
->lock
);
7469 if (space_info
->max_extent_size
&&
7470 num_bytes
> space_info
->max_extent_size
) {
7471 ins
->offset
= space_info
->max_extent_size
;
7472 spin_unlock(&space_info
->lock
);
7474 } else if (space_info
->max_extent_size
) {
7475 use_cluster
= false;
7477 spin_unlock(&space_info
->lock
);
7480 last_ptr
= fetch_cluster_info(fs_info
, space_info
, &empty_cluster
);
7482 spin_lock(&last_ptr
->lock
);
7483 if (last_ptr
->block_group
)
7484 hint_byte
= last_ptr
->window_start
;
7485 if (last_ptr
->fragmented
) {
7487 * We still set window_start so we can keep track of the
7488 * last place we found an allocation to try and save
7491 hint_byte
= last_ptr
->window_start
;
7492 use_cluster
= false;
7494 spin_unlock(&last_ptr
->lock
);
7497 search_start
= max(search_start
, first_logical_byte(fs_info
, 0));
7498 search_start
= max(search_start
, hint_byte
);
7499 if (search_start
== hint_byte
) {
7500 block_group
= btrfs_lookup_block_group(fs_info
, search_start
);
7502 * we don't want to use the block group if it doesn't match our
7503 * allocation bits, or if its not cached.
7505 * However if we are re-searching with an ideal block group
7506 * picked out then we don't care that the block group is cached.
7508 if (block_group
&& block_group_bits(block_group
, flags
) &&
7509 block_group
->cached
!= BTRFS_CACHE_NO
) {
7510 down_read(&space_info
->groups_sem
);
7511 if (list_empty(&block_group
->list
) ||
7514 * someone is removing this block group,
7515 * we can't jump into the have_block_group
7516 * target because our list pointers are not
7519 btrfs_put_block_group(block_group
);
7520 up_read(&space_info
->groups_sem
);
7522 index
= btrfs_bg_flags_to_raid_index(
7523 block_group
->flags
);
7524 btrfs_lock_block_group(block_group
, delalloc
);
7525 goto have_block_group
;
7527 } else if (block_group
) {
7528 btrfs_put_block_group(block_group
);
7532 have_caching_bg
= false;
7533 if (index
== 0 || index
== btrfs_bg_flags_to_raid_index(flags
))
7535 down_read(&space_info
->groups_sem
);
7536 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7541 /* If the block group is read-only, we can skip it entirely. */
7542 if (unlikely(block_group
->ro
))
7545 btrfs_grab_block_group(block_group
, delalloc
);
7546 search_start
= block_group
->key
.objectid
;
7549 * this can happen if we end up cycling through all the
7550 * raid types, but we want to make sure we only allocate
7551 * for the proper type.
7553 if (!block_group_bits(block_group
, flags
)) {
7554 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7555 BTRFS_BLOCK_GROUP_RAID1
|
7556 BTRFS_BLOCK_GROUP_RAID5
|
7557 BTRFS_BLOCK_GROUP_RAID6
|
7558 BTRFS_BLOCK_GROUP_RAID10
;
7561 * if they asked for extra copies and this block group
7562 * doesn't provide them, bail. This does allow us to
7563 * fill raid0 from raid1.
7565 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7570 cached
= block_group_cache_done(block_group
);
7571 if (unlikely(!cached
)) {
7572 have_caching_bg
= true;
7573 ret
= cache_block_group(block_group
, 0);
7578 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7582 * Ok we want to try and use the cluster allocator, so
7585 if (last_ptr
&& use_cluster
) {
7586 struct btrfs_block_group_cache
*used_block_group
;
7587 unsigned long aligned_cluster
;
7589 * the refill lock keeps out other
7590 * people trying to start a new cluster
7592 used_block_group
= btrfs_lock_cluster(block_group
,
7595 if (!used_block_group
)
7596 goto refill_cluster
;
7598 if (used_block_group
!= block_group
&&
7599 (used_block_group
->ro
||
7600 !block_group_bits(used_block_group
, flags
)))
7601 goto release_cluster
;
7603 offset
= btrfs_alloc_from_cluster(used_block_group
,
7606 used_block_group
->key
.objectid
,
7609 /* we have a block, we're done */
7610 spin_unlock(&last_ptr
->refill_lock
);
7611 trace_btrfs_reserve_extent_cluster(
7613 search_start
, num_bytes
);
7614 if (used_block_group
!= block_group
) {
7615 btrfs_release_block_group(block_group
,
7617 block_group
= used_block_group
;
7622 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7624 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7625 * set up a new clusters, so lets just skip it
7626 * and let the allocator find whatever block
7627 * it can find. If we reach this point, we
7628 * will have tried the cluster allocator
7629 * plenty of times and not have found
7630 * anything, so we are likely way too
7631 * fragmented for the clustering stuff to find
7634 * However, if the cluster is taken from the
7635 * current block group, release the cluster
7636 * first, so that we stand a better chance of
7637 * succeeding in the unclustered
7639 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7640 used_block_group
!= block_group
) {
7641 spin_unlock(&last_ptr
->refill_lock
);
7642 btrfs_release_block_group(used_block_group
,
7644 goto unclustered_alloc
;
7648 * this cluster didn't work out, free it and
7651 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7653 if (used_block_group
!= block_group
)
7654 btrfs_release_block_group(used_block_group
,
7657 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7658 spin_unlock(&last_ptr
->refill_lock
);
7659 goto unclustered_alloc
;
7662 aligned_cluster
= max_t(unsigned long,
7663 empty_cluster
+ empty_size
,
7664 block_group
->full_stripe_len
);
7666 /* allocate a cluster in this block group */
7667 ret
= btrfs_find_space_cluster(fs_info
, block_group
,
7668 last_ptr
, search_start
,
7673 * now pull our allocation out of this
7676 offset
= btrfs_alloc_from_cluster(block_group
,
7682 /* we found one, proceed */
7683 spin_unlock(&last_ptr
->refill_lock
);
7684 trace_btrfs_reserve_extent_cluster(
7685 block_group
, search_start
,
7689 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7690 && !failed_cluster_refill
) {
7691 spin_unlock(&last_ptr
->refill_lock
);
7693 failed_cluster_refill
= true;
7694 wait_block_group_cache_progress(block_group
,
7695 num_bytes
+ empty_cluster
+ empty_size
);
7696 goto have_block_group
;
7700 * at this point we either didn't find a cluster
7701 * or we weren't able to allocate a block from our
7702 * cluster. Free the cluster we've been trying
7703 * to use, and go to the next block group
7705 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7706 spin_unlock(&last_ptr
->refill_lock
);
7712 * We are doing an unclustered alloc, set the fragmented flag so
7713 * we don't bother trying to setup a cluster again until we get
7716 if (unlikely(last_ptr
)) {
7717 spin_lock(&last_ptr
->lock
);
7718 last_ptr
->fragmented
= 1;
7719 spin_unlock(&last_ptr
->lock
);
7722 struct btrfs_free_space_ctl
*ctl
=
7723 block_group
->free_space_ctl
;
7725 spin_lock(&ctl
->tree_lock
);
7726 if (ctl
->free_space
<
7727 num_bytes
+ empty_cluster
+ empty_size
) {
7728 if (ctl
->free_space
> max_extent_size
)
7729 max_extent_size
= ctl
->free_space
;
7730 spin_unlock(&ctl
->tree_lock
);
7733 spin_unlock(&ctl
->tree_lock
);
7736 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7737 num_bytes
, empty_size
,
7740 * If we didn't find a chunk, and we haven't failed on this
7741 * block group before, and this block group is in the middle of
7742 * caching and we are ok with waiting, then go ahead and wait
7743 * for progress to be made, and set failed_alloc to true.
7745 * If failed_alloc is true then we've already waited on this
7746 * block group once and should move on to the next block group.
7748 if (!offset
&& !failed_alloc
&& !cached
&&
7749 loop
> LOOP_CACHING_NOWAIT
) {
7750 wait_block_group_cache_progress(block_group
,
7751 num_bytes
+ empty_size
);
7752 failed_alloc
= true;
7753 goto have_block_group
;
7754 } else if (!offset
) {
7758 search_start
= ALIGN(offset
, fs_info
->stripesize
);
7760 /* move on to the next group */
7761 if (search_start
+ num_bytes
>
7762 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7763 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7767 if (offset
< search_start
)
7768 btrfs_add_free_space(block_group
, offset
,
7769 search_start
- offset
);
7770 BUG_ON(offset
> search_start
);
7772 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7773 num_bytes
, delalloc
);
7774 if (ret
== -EAGAIN
) {
7775 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7778 btrfs_inc_block_group_reservations(block_group
);
7780 /* we are all good, lets return */
7781 ins
->objectid
= search_start
;
7782 ins
->offset
= num_bytes
;
7784 trace_btrfs_reserve_extent(block_group
, search_start
, num_bytes
);
7785 btrfs_release_block_group(block_group
, delalloc
);
7788 failed_cluster_refill
= false;
7789 failed_alloc
= false;
7790 BUG_ON(btrfs_bg_flags_to_raid_index(block_group
->flags
) !=
7792 btrfs_release_block_group(block_group
, delalloc
);
7795 up_read(&space_info
->groups_sem
);
7797 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7798 && !orig_have_caching_bg
)
7799 orig_have_caching_bg
= true;
7801 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7804 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7808 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7809 * caching kthreads as we move along
7810 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7811 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7812 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7815 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7817 if (loop
== LOOP_CACHING_NOWAIT
) {
7819 * We want to skip the LOOP_CACHING_WAIT step if we
7820 * don't have any uncached bgs and we've already done a
7821 * full search through.
7823 if (orig_have_caching_bg
|| !full_search
)
7824 loop
= LOOP_CACHING_WAIT
;
7826 loop
= LOOP_ALLOC_CHUNK
;
7831 if (loop
== LOOP_ALLOC_CHUNK
) {
7832 struct btrfs_trans_handle
*trans
;
7835 trans
= current
->journal_info
;
7839 trans
= btrfs_join_transaction(root
);
7841 if (IS_ERR(trans
)) {
7842 ret
= PTR_ERR(trans
);
7846 ret
= do_chunk_alloc(trans
, fs_info
, flags
,
7850 * If we can't allocate a new chunk we've already looped
7851 * through at least once, move on to the NO_EMPTY_SIZE
7855 loop
= LOOP_NO_EMPTY_SIZE
;
7858 * Do not bail out on ENOSPC since we
7859 * can do more things.
7861 if (ret
< 0 && ret
!= -ENOSPC
)
7862 btrfs_abort_transaction(trans
, ret
);
7866 btrfs_end_transaction(trans
);
7871 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7873 * Don't loop again if we already have no empty_size and
7876 if (empty_size
== 0 &&
7877 empty_cluster
== 0) {
7886 } else if (!ins
->objectid
) {
7888 } else if (ins
->objectid
) {
7889 if (!use_cluster
&& last_ptr
) {
7890 spin_lock(&last_ptr
->lock
);
7891 last_ptr
->window_start
= ins
->objectid
;
7892 spin_unlock(&last_ptr
->lock
);
7897 if (ret
== -ENOSPC
) {
7898 spin_lock(&space_info
->lock
);
7899 space_info
->max_extent_size
= max_extent_size
;
7900 spin_unlock(&space_info
->lock
);
7901 ins
->offset
= max_extent_size
;
7906 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
7907 struct btrfs_space_info
*info
, u64 bytes
,
7908 int dump_block_groups
)
7910 struct btrfs_block_group_cache
*cache
;
7913 spin_lock(&info
->lock
);
7914 btrfs_info(fs_info
, "space_info %llu has %llu free, is %sfull",
7916 info
->total_bytes
- btrfs_space_info_used(info
, true),
7917 info
->full
? "" : "not ");
7919 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7920 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7921 info
->bytes_reserved
, info
->bytes_may_use
,
7922 info
->bytes_readonly
);
7923 spin_unlock(&info
->lock
);
7925 if (!dump_block_groups
)
7928 down_read(&info
->groups_sem
);
7930 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7931 spin_lock(&cache
->lock
);
7933 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7934 cache
->key
.objectid
, cache
->key
.offset
,
7935 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7936 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7937 btrfs_dump_free_space(cache
, bytes
);
7938 spin_unlock(&cache
->lock
);
7940 if (++index
< BTRFS_NR_RAID_TYPES
)
7942 up_read(&info
->groups_sem
);
7946 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7947 * hole that is at least as big as @num_bytes.
7949 * @root - The root that will contain this extent
7951 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7952 * is used for accounting purposes. This value differs
7953 * from @num_bytes only in the case of compressed extents.
7955 * @num_bytes - Number of bytes to allocate on-disk.
7957 * @min_alloc_size - Indicates the minimum amount of space that the
7958 * allocator should try to satisfy. In some cases
7959 * @num_bytes may be larger than what is required and if
7960 * the filesystem is fragmented then allocation fails.
7961 * However, the presence of @min_alloc_size gives a
7962 * chance to try and satisfy the smaller allocation.
7964 * @empty_size - A hint that you plan on doing more COW. This is the
7965 * size in bytes the allocator should try to find free
7966 * next to the block it returns. This is just a hint and
7967 * may be ignored by the allocator.
7969 * @hint_byte - Hint to the allocator to start searching above the byte
7970 * address passed. It might be ignored.
7972 * @ins - This key is modified to record the found hole. It will
7973 * have the following values:
7974 * ins->objectid == start position
7975 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7976 * ins->offset == the size of the hole.
7978 * @is_data - Boolean flag indicating whether an extent is
7979 * allocated for data (true) or metadata (false)
7981 * @delalloc - Boolean flag indicating whether this allocation is for
7982 * delalloc or not. If 'true' data_rwsem of block groups
7983 * is going to be acquired.
7986 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7987 * case -ENOSPC is returned then @ins->offset will contain the size of the
7988 * largest available hole the allocator managed to find.
7990 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7991 u64 num_bytes
, u64 min_alloc_size
,
7992 u64 empty_size
, u64 hint_byte
,
7993 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7995 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7996 bool final_tried
= num_bytes
== min_alloc_size
;
8000 flags
= get_alloc_profile_by_root(root
, is_data
);
8002 WARN_ON(num_bytes
< fs_info
->sectorsize
);
8003 ret
= find_free_extent(fs_info
, ram_bytes
, num_bytes
, empty_size
,
8004 hint_byte
, ins
, flags
, delalloc
);
8005 if (!ret
&& !is_data
) {
8006 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
8007 } else if (ret
== -ENOSPC
) {
8008 if (!final_tried
&& ins
->offset
) {
8009 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
8010 num_bytes
= round_down(num_bytes
,
8011 fs_info
->sectorsize
);
8012 num_bytes
= max(num_bytes
, min_alloc_size
);
8013 ram_bytes
= num_bytes
;
8014 if (num_bytes
== min_alloc_size
)
8017 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8018 struct btrfs_space_info
*sinfo
;
8020 sinfo
= __find_space_info(fs_info
, flags
);
8022 "allocation failed flags %llu, wanted %llu",
8025 dump_space_info(fs_info
, sinfo
, num_bytes
, 1);
8032 static int __btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8034 int pin
, int delalloc
)
8036 struct btrfs_block_group_cache
*cache
;
8039 cache
= btrfs_lookup_block_group(fs_info
, start
);
8041 btrfs_err(fs_info
, "Unable to find block group for %llu",
8047 pin_down_extent(fs_info
, cache
, start
, len
, 1);
8049 if (btrfs_test_opt(fs_info
, DISCARD
))
8050 ret
= btrfs_discard_extent(fs_info
, start
, len
, NULL
);
8051 btrfs_add_free_space(cache
, start
, len
);
8052 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8053 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
8056 btrfs_put_block_group(cache
);
8060 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8061 u64 start
, u64 len
, int delalloc
)
8063 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 0, delalloc
);
8066 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info
*fs_info
,
8069 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 1, 0);
8072 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8073 u64 parent
, u64 root_objectid
,
8074 u64 flags
, u64 owner
, u64 offset
,
8075 struct btrfs_key
*ins
, int ref_mod
)
8077 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
8079 struct btrfs_extent_item
*extent_item
;
8080 struct btrfs_extent_inline_ref
*iref
;
8081 struct btrfs_path
*path
;
8082 struct extent_buffer
*leaf
;
8087 type
= BTRFS_SHARED_DATA_REF_KEY
;
8089 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8091 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8093 path
= btrfs_alloc_path();
8097 path
->leave_spinning
= 1;
8098 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8101 btrfs_free_path(path
);
8105 leaf
= path
->nodes
[0];
8106 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8107 struct btrfs_extent_item
);
8108 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8109 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8110 btrfs_set_extent_flags(leaf
, extent_item
,
8111 flags
| BTRFS_EXTENT_FLAG_DATA
);
8113 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8114 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8116 struct btrfs_shared_data_ref
*ref
;
8117 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8118 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8119 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8121 struct btrfs_extent_data_ref
*ref
;
8122 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8123 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8124 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8125 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8126 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8129 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8130 btrfs_free_path(path
);
8132 ret
= remove_from_free_space_tree(trans
, ins
->objectid
, ins
->offset
);
8136 ret
= update_block_group(trans
, fs_info
, ins
->objectid
, ins
->offset
, 1);
8137 if (ret
) { /* -ENOENT, logic error */
8138 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8139 ins
->objectid
, ins
->offset
);
8142 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
, ins
->offset
);
8146 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8147 struct btrfs_delayed_ref_node
*node
,
8148 struct btrfs_delayed_extent_op
*extent_op
)
8150 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
8152 struct btrfs_extent_item
*extent_item
;
8153 struct btrfs_key extent_key
;
8154 struct btrfs_tree_block_info
*block_info
;
8155 struct btrfs_extent_inline_ref
*iref
;
8156 struct btrfs_path
*path
;
8157 struct extent_buffer
*leaf
;
8158 struct btrfs_delayed_tree_ref
*ref
;
8159 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8161 u64 flags
= extent_op
->flags_to_set
;
8162 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8164 ref
= btrfs_delayed_node_to_tree_ref(node
);
8166 extent_key
.objectid
= node
->bytenr
;
8167 if (skinny_metadata
) {
8168 extent_key
.offset
= ref
->level
;
8169 extent_key
.type
= BTRFS_METADATA_ITEM_KEY
;
8170 num_bytes
= fs_info
->nodesize
;
8172 extent_key
.offset
= node
->num_bytes
;
8173 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
8174 size
+= sizeof(*block_info
);
8175 num_bytes
= node
->num_bytes
;
8178 path
= btrfs_alloc_path();
8180 btrfs_free_and_pin_reserved_extent(fs_info
,
8181 extent_key
.objectid
,
8186 path
->leave_spinning
= 1;
8187 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8190 btrfs_free_path(path
);
8191 btrfs_free_and_pin_reserved_extent(fs_info
,
8192 extent_key
.objectid
,
8197 leaf
= path
->nodes
[0];
8198 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8199 struct btrfs_extent_item
);
8200 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8201 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8202 btrfs_set_extent_flags(leaf
, extent_item
,
8203 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8205 if (skinny_metadata
) {
8206 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8208 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8209 btrfs_set_tree_block_key(leaf
, block_info
, &extent_op
->key
);
8210 btrfs_set_tree_block_level(leaf
, block_info
, ref
->level
);
8211 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8214 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
8215 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8216 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8217 BTRFS_SHARED_BLOCK_REF_KEY
);
8218 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->parent
);
8220 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8221 BTRFS_TREE_BLOCK_REF_KEY
);
8222 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->root
);
8225 btrfs_mark_buffer_dirty(leaf
);
8226 btrfs_free_path(path
);
8228 ret
= remove_from_free_space_tree(trans
, extent_key
.objectid
,
8233 ret
= update_block_group(trans
, fs_info
, extent_key
.objectid
,
8234 fs_info
->nodesize
, 1);
8235 if (ret
) { /* -ENOENT, logic error */
8236 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8237 extent_key
.objectid
, extent_key
.offset
);
8241 trace_btrfs_reserved_extent_alloc(fs_info
, extent_key
.objectid
,
8246 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8247 struct btrfs_root
*root
, u64 owner
,
8248 u64 offset
, u64 ram_bytes
,
8249 struct btrfs_key
*ins
)
8253 BUG_ON(root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
8255 btrfs_ref_tree_mod(root
, ins
->objectid
, ins
->offset
, 0,
8256 root
->root_key
.objectid
, owner
, offset
,
8257 BTRFS_ADD_DELAYED_EXTENT
);
8259 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
,
8261 root
->root_key
.objectid
, owner
,
8263 BTRFS_ADD_DELAYED_EXTENT
, NULL
, NULL
);
8268 * this is used by the tree logging recovery code. It records that
8269 * an extent has been allocated and makes sure to clear the free
8270 * space cache bits as well
8272 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8273 struct btrfs_fs_info
*fs_info
,
8274 u64 root_objectid
, u64 owner
, u64 offset
,
8275 struct btrfs_key
*ins
)
8278 struct btrfs_block_group_cache
*block_group
;
8279 struct btrfs_space_info
*space_info
;
8282 * Mixed block groups will exclude before processing the log so we only
8283 * need to do the exclude dance if this fs isn't mixed.
8285 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
8286 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
8292 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
8296 space_info
= block_group
->space_info
;
8297 spin_lock(&space_info
->lock
);
8298 spin_lock(&block_group
->lock
);
8299 space_info
->bytes_reserved
+= ins
->offset
;
8300 block_group
->reserved
+= ins
->offset
;
8301 spin_unlock(&block_group
->lock
);
8302 spin_unlock(&space_info
->lock
);
8304 ret
= alloc_reserved_file_extent(trans
, 0, root_objectid
, 0, owner
,
8306 btrfs_put_block_group(block_group
);
8310 static struct extent_buffer
*
8311 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8312 u64 bytenr
, int level
)
8314 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8315 struct extent_buffer
*buf
;
8317 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8321 btrfs_set_header_generation(buf
, trans
->transid
);
8322 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8323 btrfs_tree_lock(buf
);
8324 clean_tree_block(fs_info
, buf
);
8325 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8327 btrfs_set_lock_blocking(buf
);
8328 set_extent_buffer_uptodate(buf
);
8330 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8331 buf
->log_index
= root
->log_transid
% 2;
8333 * we allow two log transactions at a time, use different
8334 * EXENT bit to differentiate dirty pages.
8336 if (buf
->log_index
== 0)
8337 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8338 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8340 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8341 buf
->start
+ buf
->len
- 1);
8343 buf
->log_index
= -1;
8344 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8345 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8347 trans
->dirty
= true;
8348 /* this returns a buffer locked for blocking */
8352 static struct btrfs_block_rsv
*
8353 use_block_rsv(struct btrfs_trans_handle
*trans
,
8354 struct btrfs_root
*root
, u32 blocksize
)
8356 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8357 struct btrfs_block_rsv
*block_rsv
;
8358 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
8360 bool global_updated
= false;
8362 block_rsv
= get_block_rsv(trans
, root
);
8364 if (unlikely(block_rsv
->size
== 0))
8367 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8371 if (block_rsv
->failfast
)
8372 return ERR_PTR(ret
);
8374 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8375 global_updated
= true;
8376 update_global_block_rsv(fs_info
);
8380 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8381 static DEFINE_RATELIMIT_STATE(_rs
,
8382 DEFAULT_RATELIMIT_INTERVAL
* 10,
8383 /*DEFAULT_RATELIMIT_BURST*/ 1);
8384 if (__ratelimit(&_rs
))
8386 "BTRFS: block rsv returned %d\n", ret
);
8389 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8390 BTRFS_RESERVE_NO_FLUSH
);
8394 * If we couldn't reserve metadata bytes try and use some from
8395 * the global reserve if its space type is the same as the global
8398 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8399 block_rsv
->space_info
== global_rsv
->space_info
) {
8400 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8404 return ERR_PTR(ret
);
8407 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8408 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8410 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8411 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0, NULL
);
8415 * finds a free extent and does all the dirty work required for allocation
8416 * returns the tree buffer or an ERR_PTR on error.
8418 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8419 struct btrfs_root
*root
,
8420 u64 parent
, u64 root_objectid
,
8421 const struct btrfs_disk_key
*key
,
8422 int level
, u64 hint
,
8425 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8426 struct btrfs_key ins
;
8427 struct btrfs_block_rsv
*block_rsv
;
8428 struct extent_buffer
*buf
;
8429 struct btrfs_delayed_extent_op
*extent_op
;
8432 u32 blocksize
= fs_info
->nodesize
;
8433 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8435 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8436 if (btrfs_is_testing(fs_info
)) {
8437 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8440 root
->alloc_bytenr
+= blocksize
;
8445 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8446 if (IS_ERR(block_rsv
))
8447 return ERR_CAST(block_rsv
);
8449 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8450 empty_size
, hint
, &ins
, 0, 0);
8454 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8457 goto out_free_reserved
;
8460 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8462 parent
= ins
.objectid
;
8463 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8467 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8468 extent_op
= btrfs_alloc_delayed_extent_op();
8474 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8476 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8477 extent_op
->flags_to_set
= flags
;
8478 extent_op
->update_key
= skinny_metadata
? false : true;
8479 extent_op
->update_flags
= true;
8480 extent_op
->is_data
= false;
8481 extent_op
->level
= level
;
8483 btrfs_ref_tree_mod(root
, ins
.objectid
, ins
.offset
, parent
,
8484 root_objectid
, level
, 0,
8485 BTRFS_ADD_DELAYED_EXTENT
);
8486 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
8488 root_objectid
, level
,
8489 BTRFS_ADD_DELAYED_EXTENT
,
8490 extent_op
, NULL
, NULL
);
8492 goto out_free_delayed
;
8497 btrfs_free_delayed_extent_op(extent_op
);
8499 free_extent_buffer(buf
);
8501 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
8503 unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
8504 return ERR_PTR(ret
);
8507 struct walk_control
{
8508 u64 refs
[BTRFS_MAX_LEVEL
];
8509 u64 flags
[BTRFS_MAX_LEVEL
];
8510 struct btrfs_key update_progress
;
8521 #define DROP_REFERENCE 1
8522 #define UPDATE_BACKREF 2
8524 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8525 struct btrfs_root
*root
,
8526 struct walk_control
*wc
,
8527 struct btrfs_path
*path
)
8529 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8535 struct btrfs_key key
;
8536 struct extent_buffer
*eb
;
8541 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8542 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8543 wc
->reada_count
= max(wc
->reada_count
, 2);
8545 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8546 wc
->reada_count
= min_t(int, wc
->reada_count
,
8547 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
8550 eb
= path
->nodes
[wc
->level
];
8551 nritems
= btrfs_header_nritems(eb
);
8553 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8554 if (nread
>= wc
->reada_count
)
8558 bytenr
= btrfs_node_blockptr(eb
, slot
);
8559 generation
= btrfs_node_ptr_generation(eb
, slot
);
8561 if (slot
== path
->slots
[wc
->level
])
8564 if (wc
->stage
== UPDATE_BACKREF
&&
8565 generation
<= root
->root_key
.offset
)
8568 /* We don't lock the tree block, it's OK to be racy here */
8569 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
8570 wc
->level
- 1, 1, &refs
,
8572 /* We don't care about errors in readahead. */
8577 if (wc
->stage
== DROP_REFERENCE
) {
8581 if (wc
->level
== 1 &&
8582 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8584 if (!wc
->update_ref
||
8585 generation
<= root
->root_key
.offset
)
8587 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8588 ret
= btrfs_comp_cpu_keys(&key
,
8589 &wc
->update_progress
);
8593 if (wc
->level
== 1 &&
8594 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8598 readahead_tree_block(fs_info
, bytenr
);
8601 wc
->reada_slot
= slot
;
8605 * helper to process tree block while walking down the tree.
8607 * when wc->stage == UPDATE_BACKREF, this function updates
8608 * back refs for pointers in the block.
8610 * NOTE: return value 1 means we should stop walking down.
8612 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8613 struct btrfs_root
*root
,
8614 struct btrfs_path
*path
,
8615 struct walk_control
*wc
, int lookup_info
)
8617 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8618 int level
= wc
->level
;
8619 struct extent_buffer
*eb
= path
->nodes
[level
];
8620 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8623 if (wc
->stage
== UPDATE_BACKREF
&&
8624 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8628 * when reference count of tree block is 1, it won't increase
8629 * again. once full backref flag is set, we never clear it.
8632 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8633 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8634 BUG_ON(!path
->locks
[level
]);
8635 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8636 eb
->start
, level
, 1,
8639 BUG_ON(ret
== -ENOMEM
);
8642 BUG_ON(wc
->refs
[level
] == 0);
8645 if (wc
->stage
== DROP_REFERENCE
) {
8646 if (wc
->refs
[level
] > 1)
8649 if (path
->locks
[level
] && !wc
->keep_locks
) {
8650 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8651 path
->locks
[level
] = 0;
8656 /* wc->stage == UPDATE_BACKREF */
8657 if (!(wc
->flags
[level
] & flag
)) {
8658 BUG_ON(!path
->locks
[level
]);
8659 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8660 BUG_ON(ret
); /* -ENOMEM */
8661 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8662 BUG_ON(ret
); /* -ENOMEM */
8663 ret
= btrfs_set_disk_extent_flags(trans
, fs_info
, eb
->start
,
8665 btrfs_header_level(eb
), 0);
8666 BUG_ON(ret
); /* -ENOMEM */
8667 wc
->flags
[level
] |= flag
;
8671 * the block is shared by multiple trees, so it's not good to
8672 * keep the tree lock
8674 if (path
->locks
[level
] && level
> 0) {
8675 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8676 path
->locks
[level
] = 0;
8682 * helper to process tree block pointer.
8684 * when wc->stage == DROP_REFERENCE, this function checks
8685 * reference count of the block pointed to. if the block
8686 * is shared and we need update back refs for the subtree
8687 * rooted at the block, this function changes wc->stage to
8688 * UPDATE_BACKREF. if the block is shared and there is no
8689 * need to update back, this function drops the reference
8692 * NOTE: return value 1 means we should stop walking down.
8694 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8695 struct btrfs_root
*root
,
8696 struct btrfs_path
*path
,
8697 struct walk_control
*wc
, int *lookup_info
)
8699 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8704 struct btrfs_key key
;
8705 struct btrfs_key first_key
;
8706 struct extent_buffer
*next
;
8707 int level
= wc
->level
;
8710 bool need_account
= false;
8712 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8713 path
->slots
[level
]);
8715 * if the lower level block was created before the snapshot
8716 * was created, we know there is no need to update back refs
8719 if (wc
->stage
== UPDATE_BACKREF
&&
8720 generation
<= root
->root_key
.offset
) {
8725 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8726 btrfs_node_key_to_cpu(path
->nodes
[level
], &first_key
,
8727 path
->slots
[level
]);
8728 blocksize
= fs_info
->nodesize
;
8730 next
= find_extent_buffer(fs_info
, bytenr
);
8732 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8734 return PTR_ERR(next
);
8736 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8740 btrfs_tree_lock(next
);
8741 btrfs_set_lock_blocking(next
);
8743 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
8744 &wc
->refs
[level
- 1],
8745 &wc
->flags
[level
- 1]);
8749 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8750 btrfs_err(fs_info
, "Missing references.");
8756 if (wc
->stage
== DROP_REFERENCE
) {
8757 if (wc
->refs
[level
- 1] > 1) {
8758 need_account
= true;
8760 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8763 if (!wc
->update_ref
||
8764 generation
<= root
->root_key
.offset
)
8767 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8768 path
->slots
[level
]);
8769 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8773 wc
->stage
= UPDATE_BACKREF
;
8774 wc
->shared_level
= level
- 1;
8778 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8782 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8783 btrfs_tree_unlock(next
);
8784 free_extent_buffer(next
);
8790 if (reada
&& level
== 1)
8791 reada_walk_down(trans
, root
, wc
, path
);
8792 next
= read_tree_block(fs_info
, bytenr
, generation
, level
- 1,
8795 return PTR_ERR(next
);
8796 } else if (!extent_buffer_uptodate(next
)) {
8797 free_extent_buffer(next
);
8800 btrfs_tree_lock(next
);
8801 btrfs_set_lock_blocking(next
);
8805 ASSERT(level
== btrfs_header_level(next
));
8806 if (level
!= btrfs_header_level(next
)) {
8807 btrfs_err(root
->fs_info
, "mismatched level");
8811 path
->nodes
[level
] = next
;
8812 path
->slots
[level
] = 0;
8813 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8819 wc
->refs
[level
- 1] = 0;
8820 wc
->flags
[level
- 1] = 0;
8821 if (wc
->stage
== DROP_REFERENCE
) {
8822 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8823 parent
= path
->nodes
[level
]->start
;
8825 ASSERT(root
->root_key
.objectid
==
8826 btrfs_header_owner(path
->nodes
[level
]));
8827 if (root
->root_key
.objectid
!=
8828 btrfs_header_owner(path
->nodes
[level
])) {
8829 btrfs_err(root
->fs_info
,
8830 "mismatched block owner");
8838 ret
= btrfs_qgroup_trace_subtree(trans
, root
, next
,
8839 generation
, level
- 1);
8841 btrfs_err_rl(fs_info
,
8842 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8846 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
8847 parent
, root
->root_key
.objectid
,
8857 btrfs_tree_unlock(next
);
8858 free_extent_buffer(next
);
8864 * helper to process tree block while walking up the tree.
8866 * when wc->stage == DROP_REFERENCE, this function drops
8867 * reference count on the block.
8869 * when wc->stage == UPDATE_BACKREF, this function changes
8870 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8871 * to UPDATE_BACKREF previously while processing the block.
8873 * NOTE: return value 1 means we should stop walking up.
8875 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8876 struct btrfs_root
*root
,
8877 struct btrfs_path
*path
,
8878 struct walk_control
*wc
)
8880 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8882 int level
= wc
->level
;
8883 struct extent_buffer
*eb
= path
->nodes
[level
];
8886 if (wc
->stage
== UPDATE_BACKREF
) {
8887 BUG_ON(wc
->shared_level
< level
);
8888 if (level
< wc
->shared_level
)
8891 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8895 wc
->stage
= DROP_REFERENCE
;
8896 wc
->shared_level
= -1;
8897 path
->slots
[level
] = 0;
8900 * check reference count again if the block isn't locked.
8901 * we should start walking down the tree again if reference
8904 if (!path
->locks
[level
]) {
8906 btrfs_tree_lock(eb
);
8907 btrfs_set_lock_blocking(eb
);
8908 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8910 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8911 eb
->start
, level
, 1,
8915 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8916 path
->locks
[level
] = 0;
8919 BUG_ON(wc
->refs
[level
] == 0);
8920 if (wc
->refs
[level
] == 1) {
8921 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8922 path
->locks
[level
] = 0;
8928 /* wc->stage == DROP_REFERENCE */
8929 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8931 if (wc
->refs
[level
] == 1) {
8933 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8934 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8936 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8937 BUG_ON(ret
); /* -ENOMEM */
8938 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, eb
);
8940 btrfs_err_rl(fs_info
,
8941 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8945 /* make block locked assertion in clean_tree_block happy */
8946 if (!path
->locks
[level
] &&
8947 btrfs_header_generation(eb
) == trans
->transid
) {
8948 btrfs_tree_lock(eb
);
8949 btrfs_set_lock_blocking(eb
);
8950 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8952 clean_tree_block(fs_info
, eb
);
8955 if (eb
== root
->node
) {
8956 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8959 BUG_ON(root
->root_key
.objectid
!=
8960 btrfs_header_owner(eb
));
8962 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8963 parent
= path
->nodes
[level
+ 1]->start
;
8965 BUG_ON(root
->root_key
.objectid
!=
8966 btrfs_header_owner(path
->nodes
[level
+ 1]));
8969 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8971 wc
->refs
[level
] = 0;
8972 wc
->flags
[level
] = 0;
8976 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8977 struct btrfs_root
*root
,
8978 struct btrfs_path
*path
,
8979 struct walk_control
*wc
)
8981 int level
= wc
->level
;
8982 int lookup_info
= 1;
8985 while (level
>= 0) {
8986 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8993 if (path
->slots
[level
] >=
8994 btrfs_header_nritems(path
->nodes
[level
]))
8997 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8999 path
->slots
[level
]++;
9008 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
9009 struct btrfs_root
*root
,
9010 struct btrfs_path
*path
,
9011 struct walk_control
*wc
, int max_level
)
9013 int level
= wc
->level
;
9016 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
9017 while (level
< max_level
&& path
->nodes
[level
]) {
9019 if (path
->slots
[level
] + 1 <
9020 btrfs_header_nritems(path
->nodes
[level
])) {
9021 path
->slots
[level
]++;
9024 ret
= walk_up_proc(trans
, root
, path
, wc
);
9028 if (path
->locks
[level
]) {
9029 btrfs_tree_unlock_rw(path
->nodes
[level
],
9030 path
->locks
[level
]);
9031 path
->locks
[level
] = 0;
9033 free_extent_buffer(path
->nodes
[level
]);
9034 path
->nodes
[level
] = NULL
;
9042 * drop a subvolume tree.
9044 * this function traverses the tree freeing any blocks that only
9045 * referenced by the tree.
9047 * when a shared tree block is found. this function decreases its
9048 * reference count by one. if update_ref is true, this function
9049 * also make sure backrefs for the shared block and all lower level
9050 * blocks are properly updated.
9052 * If called with for_reloc == 0, may exit early with -EAGAIN
9054 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9055 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9058 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9059 struct btrfs_path
*path
;
9060 struct btrfs_trans_handle
*trans
;
9061 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
9062 struct btrfs_root_item
*root_item
= &root
->root_item
;
9063 struct walk_control
*wc
;
9064 struct btrfs_key key
;
9068 bool root_dropped
= false;
9070 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->objectid
);
9072 path
= btrfs_alloc_path();
9078 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9080 btrfs_free_path(path
);
9085 trans
= btrfs_start_transaction(tree_root
, 0);
9086 if (IS_ERR(trans
)) {
9087 err
= PTR_ERR(trans
);
9092 trans
->block_rsv
= block_rsv
;
9094 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9095 level
= btrfs_header_level(root
->node
);
9096 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9097 btrfs_set_lock_blocking(path
->nodes
[level
]);
9098 path
->slots
[level
] = 0;
9099 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9100 memset(&wc
->update_progress
, 0,
9101 sizeof(wc
->update_progress
));
9103 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9104 memcpy(&wc
->update_progress
, &key
,
9105 sizeof(wc
->update_progress
));
9107 level
= root_item
->drop_level
;
9109 path
->lowest_level
= level
;
9110 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9111 path
->lowest_level
= 0;
9119 * unlock our path, this is safe because only this
9120 * function is allowed to delete this snapshot
9122 btrfs_unlock_up_safe(path
, 0);
9124 level
= btrfs_header_level(root
->node
);
9126 btrfs_tree_lock(path
->nodes
[level
]);
9127 btrfs_set_lock_blocking(path
->nodes
[level
]);
9128 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9130 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
9131 path
->nodes
[level
]->start
,
9132 level
, 1, &wc
->refs
[level
],
9138 BUG_ON(wc
->refs
[level
] == 0);
9140 if (level
== root_item
->drop_level
)
9143 btrfs_tree_unlock(path
->nodes
[level
]);
9144 path
->locks
[level
] = 0;
9145 WARN_ON(wc
->refs
[level
] != 1);
9151 wc
->shared_level
= -1;
9152 wc
->stage
= DROP_REFERENCE
;
9153 wc
->update_ref
= update_ref
;
9155 wc
->for_reloc
= for_reloc
;
9156 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9160 ret
= walk_down_tree(trans
, root
, path
, wc
);
9166 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9173 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9177 if (wc
->stage
== DROP_REFERENCE
) {
9179 btrfs_node_key(path
->nodes
[level
],
9180 &root_item
->drop_progress
,
9181 path
->slots
[level
]);
9182 root_item
->drop_level
= level
;
9185 BUG_ON(wc
->level
== 0);
9186 if (btrfs_should_end_transaction(trans
) ||
9187 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
9188 ret
= btrfs_update_root(trans
, tree_root
,
9192 btrfs_abort_transaction(trans
, ret
);
9197 btrfs_end_transaction_throttle(trans
);
9198 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
9199 btrfs_debug(fs_info
,
9200 "drop snapshot early exit");
9205 trans
= btrfs_start_transaction(tree_root
, 0);
9206 if (IS_ERR(trans
)) {
9207 err
= PTR_ERR(trans
);
9211 trans
->block_rsv
= block_rsv
;
9214 btrfs_release_path(path
);
9218 ret
= btrfs_del_root(trans
, fs_info
, &root
->root_key
);
9220 btrfs_abort_transaction(trans
, ret
);
9225 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9226 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9229 btrfs_abort_transaction(trans
, ret
);
9232 } else if (ret
> 0) {
9233 /* if we fail to delete the orphan item this time
9234 * around, it'll get picked up the next time.
9236 * The most common failure here is just -ENOENT.
9238 btrfs_del_orphan_item(trans
, tree_root
,
9239 root
->root_key
.objectid
);
9243 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9244 btrfs_add_dropped_root(trans
, root
);
9246 free_extent_buffer(root
->node
);
9247 free_extent_buffer(root
->commit_root
);
9248 btrfs_put_fs_root(root
);
9250 root_dropped
= true;
9252 btrfs_end_transaction_throttle(trans
);
9255 btrfs_free_path(path
);
9258 * So if we need to stop dropping the snapshot for whatever reason we
9259 * need to make sure to add it back to the dead root list so that we
9260 * keep trying to do the work later. This also cleans up roots if we
9261 * don't have it in the radix (like when we recover after a power fail
9262 * or unmount) so we don't leak memory.
9264 if (!for_reloc
&& !root_dropped
)
9265 btrfs_add_dead_root(root
);
9266 if (err
&& err
!= -EAGAIN
)
9267 btrfs_handle_fs_error(fs_info
, err
, NULL
);
9272 * drop subtree rooted at tree block 'node'.
9274 * NOTE: this function will unlock and release tree block 'node'
9275 * only used by relocation code
9277 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9278 struct btrfs_root
*root
,
9279 struct extent_buffer
*node
,
9280 struct extent_buffer
*parent
)
9282 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9283 struct btrfs_path
*path
;
9284 struct walk_control
*wc
;
9290 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9292 path
= btrfs_alloc_path();
9296 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9298 btrfs_free_path(path
);
9302 btrfs_assert_tree_locked(parent
);
9303 parent_level
= btrfs_header_level(parent
);
9304 extent_buffer_get(parent
);
9305 path
->nodes
[parent_level
] = parent
;
9306 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9308 btrfs_assert_tree_locked(node
);
9309 level
= btrfs_header_level(node
);
9310 path
->nodes
[level
] = node
;
9311 path
->slots
[level
] = 0;
9312 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9314 wc
->refs
[parent_level
] = 1;
9315 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9317 wc
->shared_level
= -1;
9318 wc
->stage
= DROP_REFERENCE
;
9322 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9325 wret
= walk_down_tree(trans
, root
, path
, wc
);
9331 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9339 btrfs_free_path(path
);
9343 static u64
update_block_group_flags(struct btrfs_fs_info
*fs_info
, u64 flags
)
9349 * if restripe for this chunk_type is on pick target profile and
9350 * return, otherwise do the usual balance
9352 stripped
= get_restripe_target(fs_info
, flags
);
9354 return extended_to_chunk(stripped
);
9356 num_devices
= fs_info
->fs_devices
->rw_devices
;
9358 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9359 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9360 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9362 if (num_devices
== 1) {
9363 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9364 stripped
= flags
& ~stripped
;
9366 /* turn raid0 into single device chunks */
9367 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9370 /* turn mirroring into duplication */
9371 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9372 BTRFS_BLOCK_GROUP_RAID10
))
9373 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9375 /* they already had raid on here, just return */
9376 if (flags
& stripped
)
9379 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9380 stripped
= flags
& ~stripped
;
9382 /* switch duplicated blocks with raid1 */
9383 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9384 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9386 /* this is drive concat, leave it alone */
9392 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9394 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9396 u64 min_allocable_bytes
;
9400 * We need some metadata space and system metadata space for
9401 * allocating chunks in some corner cases until we force to set
9402 * it to be readonly.
9405 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9407 min_allocable_bytes
= SZ_1M
;
9409 min_allocable_bytes
= 0;
9411 spin_lock(&sinfo
->lock
);
9412 spin_lock(&cache
->lock
);
9420 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9421 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9423 if (btrfs_space_info_used(sinfo
, true) + num_bytes
+
9424 min_allocable_bytes
<= sinfo
->total_bytes
) {
9425 sinfo
->bytes_readonly
+= num_bytes
;
9427 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9431 spin_unlock(&cache
->lock
);
9432 spin_unlock(&sinfo
->lock
);
9436 int btrfs_inc_block_group_ro(struct btrfs_fs_info
*fs_info
,
9437 struct btrfs_block_group_cache
*cache
)
9440 struct btrfs_trans_handle
*trans
;
9445 trans
= btrfs_join_transaction(fs_info
->extent_root
);
9447 return PTR_ERR(trans
);
9450 * we're not allowed to set block groups readonly after the dirty
9451 * block groups cache has started writing. If it already started,
9452 * back off and let this transaction commit
9454 mutex_lock(&fs_info
->ro_block_group_mutex
);
9455 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9456 u64 transid
= trans
->transid
;
9458 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9459 btrfs_end_transaction(trans
);
9461 ret
= btrfs_wait_for_commit(fs_info
, transid
);
9468 * if we are changing raid levels, try to allocate a corresponding
9469 * block group with the new raid level.
9471 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9472 if (alloc_flags
!= cache
->flags
) {
9473 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9476 * ENOSPC is allowed here, we may have enough space
9477 * already allocated at the new raid level to
9486 ret
= inc_block_group_ro(cache
, 0);
9489 alloc_flags
= get_alloc_profile(fs_info
, cache
->space_info
->flags
);
9490 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9494 ret
= inc_block_group_ro(cache
, 0);
9496 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9497 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9498 mutex_lock(&fs_info
->chunk_mutex
);
9499 check_system_chunk(trans
, fs_info
, alloc_flags
);
9500 mutex_unlock(&fs_info
->chunk_mutex
);
9502 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9504 btrfs_end_transaction(trans
);
9508 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9509 struct btrfs_fs_info
*fs_info
, u64 type
)
9511 u64 alloc_flags
= get_alloc_profile(fs_info
, type
);
9513 return do_chunk_alloc(trans
, fs_info
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9517 * helper to account the unused space of all the readonly block group in the
9518 * space_info. takes mirrors into account.
9520 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9522 struct btrfs_block_group_cache
*block_group
;
9526 /* It's df, we don't care if it's racy */
9527 if (list_empty(&sinfo
->ro_bgs
))
9530 spin_lock(&sinfo
->lock
);
9531 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9532 spin_lock(&block_group
->lock
);
9534 if (!block_group
->ro
) {
9535 spin_unlock(&block_group
->lock
);
9539 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9540 BTRFS_BLOCK_GROUP_RAID10
|
9541 BTRFS_BLOCK_GROUP_DUP
))
9546 free_bytes
+= (block_group
->key
.offset
-
9547 btrfs_block_group_used(&block_group
->item
)) *
9550 spin_unlock(&block_group
->lock
);
9552 spin_unlock(&sinfo
->lock
);
9557 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache
*cache
)
9559 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9564 spin_lock(&sinfo
->lock
);
9565 spin_lock(&cache
->lock
);
9567 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9568 cache
->pinned
- cache
->bytes_super
-
9569 btrfs_block_group_used(&cache
->item
);
9570 sinfo
->bytes_readonly
-= num_bytes
;
9571 list_del_init(&cache
->ro_list
);
9573 spin_unlock(&cache
->lock
);
9574 spin_unlock(&sinfo
->lock
);
9578 * checks to see if its even possible to relocate this block group.
9580 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9581 * ok to go ahead and try.
9583 int btrfs_can_relocate(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
9585 struct btrfs_root
*root
= fs_info
->extent_root
;
9586 struct btrfs_block_group_cache
*block_group
;
9587 struct btrfs_space_info
*space_info
;
9588 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
9589 struct btrfs_device
*device
;
9590 struct btrfs_trans_handle
*trans
;
9600 debug
= btrfs_test_opt(fs_info
, ENOSPC_DEBUG
);
9602 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
9604 /* odd, couldn't find the block group, leave it alone */
9608 "can't find block group for bytenr %llu",
9613 min_free
= btrfs_block_group_used(&block_group
->item
);
9615 /* no bytes used, we're good */
9619 space_info
= block_group
->space_info
;
9620 spin_lock(&space_info
->lock
);
9622 full
= space_info
->full
;
9625 * if this is the last block group we have in this space, we can't
9626 * relocate it unless we're able to allocate a new chunk below.
9628 * Otherwise, we need to make sure we have room in the space to handle
9629 * all of the extents from this block group. If we can, we're good
9631 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9632 (btrfs_space_info_used(space_info
, false) + min_free
<
9633 space_info
->total_bytes
)) {
9634 spin_unlock(&space_info
->lock
);
9637 spin_unlock(&space_info
->lock
);
9640 * ok we don't have enough space, but maybe we have free space on our
9641 * devices to allocate new chunks for relocation, so loop through our
9642 * alloc devices and guess if we have enough space. if this block
9643 * group is going to be restriped, run checks against the target
9644 * profile instead of the current one.
9656 target
= get_restripe_target(fs_info
, block_group
->flags
);
9658 index
= btrfs_bg_flags_to_raid_index(extended_to_chunk(target
));
9661 * this is just a balance, so if we were marked as full
9662 * we know there is no space for a new chunk
9667 "no space to alloc new chunk for block group %llu",
9668 block_group
->key
.objectid
);
9672 index
= btrfs_bg_flags_to_raid_index(block_group
->flags
);
9675 if (index
== BTRFS_RAID_RAID10
) {
9679 } else if (index
== BTRFS_RAID_RAID1
) {
9681 } else if (index
== BTRFS_RAID_DUP
) {
9684 } else if (index
== BTRFS_RAID_RAID0
) {
9685 dev_min
= fs_devices
->rw_devices
;
9686 min_free
= div64_u64(min_free
, dev_min
);
9689 /* We need to do this so that we can look at pending chunks */
9690 trans
= btrfs_join_transaction(root
);
9691 if (IS_ERR(trans
)) {
9692 ret
= PTR_ERR(trans
);
9696 mutex_lock(&fs_info
->chunk_mutex
);
9697 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9701 * check to make sure we can actually find a chunk with enough
9702 * space to fit our block group in.
9704 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9705 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
9706 ret
= find_free_dev_extent(trans
, device
, min_free
,
9711 if (dev_nr
>= dev_min
)
9717 if (debug
&& ret
== -1)
9719 "no space to allocate a new chunk for block group %llu",
9720 block_group
->key
.objectid
);
9721 mutex_unlock(&fs_info
->chunk_mutex
);
9722 btrfs_end_transaction(trans
);
9724 btrfs_put_block_group(block_group
);
9728 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
9729 struct btrfs_path
*path
,
9730 struct btrfs_key
*key
)
9732 struct btrfs_root
*root
= fs_info
->extent_root
;
9734 struct btrfs_key found_key
;
9735 struct extent_buffer
*leaf
;
9738 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9743 slot
= path
->slots
[0];
9744 leaf
= path
->nodes
[0];
9745 if (slot
>= btrfs_header_nritems(leaf
)) {
9746 ret
= btrfs_next_leaf(root
, path
);
9753 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9755 if (found_key
.objectid
>= key
->objectid
&&
9756 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9757 struct extent_map_tree
*em_tree
;
9758 struct extent_map
*em
;
9760 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9761 read_lock(&em_tree
->lock
);
9762 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9764 read_unlock(&em_tree
->lock
);
9767 "logical %llu len %llu found bg but no related chunk",
9768 found_key
.objectid
, found_key
.offset
);
9773 free_extent_map(em
);
9782 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9784 struct btrfs_block_group_cache
*block_group
;
9788 struct inode
*inode
;
9790 block_group
= btrfs_lookup_first_block_group(info
, last
);
9791 while (block_group
) {
9792 spin_lock(&block_group
->lock
);
9793 if (block_group
->iref
)
9795 spin_unlock(&block_group
->lock
);
9796 block_group
= next_block_group(info
, block_group
);
9805 inode
= block_group
->inode
;
9806 block_group
->iref
= 0;
9807 block_group
->inode
= NULL
;
9808 spin_unlock(&block_group
->lock
);
9809 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9811 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9812 btrfs_put_block_group(block_group
);
9817 * Must be called only after stopping all workers, since we could have block
9818 * group caching kthreads running, and therefore they could race with us if we
9819 * freed the block groups before stopping them.
9821 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9823 struct btrfs_block_group_cache
*block_group
;
9824 struct btrfs_space_info
*space_info
;
9825 struct btrfs_caching_control
*caching_ctl
;
9828 down_write(&info
->commit_root_sem
);
9829 while (!list_empty(&info
->caching_block_groups
)) {
9830 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9831 struct btrfs_caching_control
, list
);
9832 list_del(&caching_ctl
->list
);
9833 put_caching_control(caching_ctl
);
9835 up_write(&info
->commit_root_sem
);
9837 spin_lock(&info
->unused_bgs_lock
);
9838 while (!list_empty(&info
->unused_bgs
)) {
9839 block_group
= list_first_entry(&info
->unused_bgs
,
9840 struct btrfs_block_group_cache
,
9842 list_del_init(&block_group
->bg_list
);
9843 btrfs_put_block_group(block_group
);
9845 spin_unlock(&info
->unused_bgs_lock
);
9847 spin_lock(&info
->block_group_cache_lock
);
9848 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9849 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9851 rb_erase(&block_group
->cache_node
,
9852 &info
->block_group_cache_tree
);
9853 RB_CLEAR_NODE(&block_group
->cache_node
);
9854 spin_unlock(&info
->block_group_cache_lock
);
9856 down_write(&block_group
->space_info
->groups_sem
);
9857 list_del(&block_group
->list
);
9858 up_write(&block_group
->space_info
->groups_sem
);
9861 * We haven't cached this block group, which means we could
9862 * possibly have excluded extents on this block group.
9864 if (block_group
->cached
== BTRFS_CACHE_NO
||
9865 block_group
->cached
== BTRFS_CACHE_ERROR
)
9866 free_excluded_extents(info
, block_group
);
9868 btrfs_remove_free_space_cache(block_group
);
9869 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
9870 ASSERT(list_empty(&block_group
->dirty_list
));
9871 ASSERT(list_empty(&block_group
->io_list
));
9872 ASSERT(list_empty(&block_group
->bg_list
));
9873 ASSERT(atomic_read(&block_group
->count
) == 1);
9874 btrfs_put_block_group(block_group
);
9876 spin_lock(&info
->block_group_cache_lock
);
9878 spin_unlock(&info
->block_group_cache_lock
);
9880 /* now that all the block groups are freed, go through and
9881 * free all the space_info structs. This is only called during
9882 * the final stages of unmount, and so we know nobody is
9883 * using them. We call synchronize_rcu() once before we start,
9884 * just to be on the safe side.
9888 release_global_block_rsv(info
);
9890 while (!list_empty(&info
->space_info
)) {
9893 space_info
= list_entry(info
->space_info
.next
,
9894 struct btrfs_space_info
,
9898 * Do not hide this behind enospc_debug, this is actually
9899 * important and indicates a real bug if this happens.
9901 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9902 space_info
->bytes_reserved
> 0 ||
9903 space_info
->bytes_may_use
> 0))
9904 dump_space_info(info
, space_info
, 0, 0);
9905 list_del(&space_info
->list
);
9906 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9907 struct kobject
*kobj
;
9908 kobj
= space_info
->block_group_kobjs
[i
];
9909 space_info
->block_group_kobjs
[i
] = NULL
;
9915 kobject_del(&space_info
->kobj
);
9916 kobject_put(&space_info
->kobj
);
9921 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9922 void btrfs_add_raid_kobjects(struct btrfs_fs_info
*fs_info
)
9924 struct btrfs_space_info
*space_info
;
9925 struct raid_kobject
*rkobj
;
9930 spin_lock(&fs_info
->pending_raid_kobjs_lock
);
9931 list_splice_init(&fs_info
->pending_raid_kobjs
, &list
);
9932 spin_unlock(&fs_info
->pending_raid_kobjs_lock
);
9934 list_for_each_entry(rkobj
, &list
, list
) {
9935 space_info
= __find_space_info(fs_info
, rkobj
->flags
);
9936 index
= btrfs_bg_flags_to_raid_index(rkobj
->flags
);
9938 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9939 "%s", get_raid_name(index
));
9941 kobject_put(&rkobj
->kobj
);
9947 "failed to add kobject for block cache, ignoring");
9950 static void link_block_group(struct btrfs_block_group_cache
*cache
)
9952 struct btrfs_space_info
*space_info
= cache
->space_info
;
9953 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
9954 int index
= btrfs_bg_flags_to_raid_index(cache
->flags
);
9957 down_write(&space_info
->groups_sem
);
9958 if (list_empty(&space_info
->block_groups
[index
]))
9960 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9961 up_write(&space_info
->groups_sem
);
9964 struct raid_kobject
*rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9966 btrfs_warn(cache
->fs_info
,
9967 "couldn't alloc memory for raid level kobject");
9970 rkobj
->flags
= cache
->flags
;
9971 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9973 spin_lock(&fs_info
->pending_raid_kobjs_lock
);
9974 list_add_tail(&rkobj
->list
, &fs_info
->pending_raid_kobjs
);
9975 spin_unlock(&fs_info
->pending_raid_kobjs_lock
);
9976 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9980 static struct btrfs_block_group_cache
*
9981 btrfs_create_block_group_cache(struct btrfs_fs_info
*fs_info
,
9982 u64 start
, u64 size
)
9984 struct btrfs_block_group_cache
*cache
;
9986 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9990 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9992 if (!cache
->free_space_ctl
) {
9997 cache
->key
.objectid
= start
;
9998 cache
->key
.offset
= size
;
9999 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10001 cache
->fs_info
= fs_info
;
10002 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
, start
);
10003 set_free_space_tree_thresholds(cache
);
10005 atomic_set(&cache
->count
, 1);
10006 spin_lock_init(&cache
->lock
);
10007 init_rwsem(&cache
->data_rwsem
);
10008 INIT_LIST_HEAD(&cache
->list
);
10009 INIT_LIST_HEAD(&cache
->cluster_list
);
10010 INIT_LIST_HEAD(&cache
->bg_list
);
10011 INIT_LIST_HEAD(&cache
->ro_list
);
10012 INIT_LIST_HEAD(&cache
->dirty_list
);
10013 INIT_LIST_HEAD(&cache
->io_list
);
10014 btrfs_init_free_space_ctl(cache
);
10015 atomic_set(&cache
->trimming
, 0);
10016 mutex_init(&cache
->free_space_lock
);
10017 btrfs_init_full_stripe_locks_tree(&cache
->full_stripe_locks_root
);
10022 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
10024 struct btrfs_path
*path
;
10026 struct btrfs_block_group_cache
*cache
;
10027 struct btrfs_space_info
*space_info
;
10028 struct btrfs_key key
;
10029 struct btrfs_key found_key
;
10030 struct extent_buffer
*leaf
;
10031 int need_clear
= 0;
10036 feature
= btrfs_super_incompat_flags(info
->super_copy
);
10037 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
10041 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10042 path
= btrfs_alloc_path();
10045 path
->reada
= READA_FORWARD
;
10047 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
10048 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
10049 btrfs_super_generation(info
->super_copy
) != cache_gen
)
10051 if (btrfs_test_opt(info
, CLEAR_CACHE
))
10055 ret
= find_first_block_group(info
, path
, &key
);
10061 leaf
= path
->nodes
[0];
10062 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
10064 cache
= btrfs_create_block_group_cache(info
, found_key
.objectid
,
10073 * When we mount with old space cache, we need to
10074 * set BTRFS_DC_CLEAR and set dirty flag.
10076 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10077 * truncate the old free space cache inode and
10079 * b) Setting 'dirty flag' makes sure that we flush
10080 * the new space cache info onto disk.
10082 if (btrfs_test_opt(info
, SPACE_CACHE
))
10083 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10086 read_extent_buffer(leaf
, &cache
->item
,
10087 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10088 sizeof(cache
->item
));
10089 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10091 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
10092 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
10094 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10095 cache
->key
.objectid
);
10100 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10101 btrfs_release_path(path
);
10104 * We need to exclude the super stripes now so that the space
10105 * info has super bytes accounted for, otherwise we'll think
10106 * we have more space than we actually do.
10108 ret
= exclude_super_stripes(info
, cache
);
10111 * We may have excluded something, so call this just in
10114 free_excluded_extents(info
, cache
);
10115 btrfs_put_block_group(cache
);
10120 * check for two cases, either we are full, and therefore
10121 * don't need to bother with the caching work since we won't
10122 * find any space, or we are empty, and we can just add all
10123 * the space in and be done with it. This saves us _alot_ of
10124 * time, particularly in the full case.
10126 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10127 cache
->last_byte_to_unpin
= (u64
)-1;
10128 cache
->cached
= BTRFS_CACHE_FINISHED
;
10129 free_excluded_extents(info
, cache
);
10130 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10131 cache
->last_byte_to_unpin
= (u64
)-1;
10132 cache
->cached
= BTRFS_CACHE_FINISHED
;
10133 add_new_free_space(cache
, found_key
.objectid
,
10134 found_key
.objectid
+
10136 free_excluded_extents(info
, cache
);
10139 ret
= btrfs_add_block_group_cache(info
, cache
);
10141 btrfs_remove_free_space_cache(cache
);
10142 btrfs_put_block_group(cache
);
10146 trace_btrfs_add_block_group(info
, cache
, 0);
10147 update_space_info(info
, cache
->flags
, found_key
.offset
,
10148 btrfs_block_group_used(&cache
->item
),
10149 cache
->bytes_super
, &space_info
);
10151 cache
->space_info
= space_info
;
10153 link_block_group(cache
);
10155 set_avail_alloc_bits(info
, cache
->flags
);
10156 if (btrfs_chunk_readonly(info
, cache
->key
.objectid
)) {
10157 inc_block_group_ro(cache
, 1);
10158 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10159 spin_lock(&info
->unused_bgs_lock
);
10160 /* Should always be true but just in case. */
10161 if (list_empty(&cache
->bg_list
)) {
10162 btrfs_get_block_group(cache
);
10163 trace_btrfs_add_unused_block_group(cache
);
10164 list_add_tail(&cache
->bg_list
,
10165 &info
->unused_bgs
);
10167 spin_unlock(&info
->unused_bgs_lock
);
10171 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
10172 if (!(get_alloc_profile(info
, space_info
->flags
) &
10173 (BTRFS_BLOCK_GROUP_RAID10
|
10174 BTRFS_BLOCK_GROUP_RAID1
|
10175 BTRFS_BLOCK_GROUP_RAID5
|
10176 BTRFS_BLOCK_GROUP_RAID6
|
10177 BTRFS_BLOCK_GROUP_DUP
)))
10180 * avoid allocating from un-mirrored block group if there are
10181 * mirrored block groups.
10183 list_for_each_entry(cache
,
10184 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10186 inc_block_group_ro(cache
, 1);
10187 list_for_each_entry(cache
,
10188 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10190 inc_block_group_ro(cache
, 1);
10193 btrfs_add_raid_kobjects(info
);
10194 init_global_block_rsv(info
);
10197 btrfs_free_path(path
);
10201 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
)
10203 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
10204 struct btrfs_block_group_cache
*block_group
, *tmp
;
10205 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
10206 struct btrfs_block_group_item item
;
10207 struct btrfs_key key
;
10209 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10211 trans
->can_flush_pending_bgs
= false;
10212 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10216 spin_lock(&block_group
->lock
);
10217 memcpy(&item
, &block_group
->item
, sizeof(item
));
10218 memcpy(&key
, &block_group
->key
, sizeof(key
));
10219 spin_unlock(&block_group
->lock
);
10221 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10224 btrfs_abort_transaction(trans
, ret
);
10225 ret
= btrfs_finish_chunk_alloc(trans
, fs_info
, key
.objectid
,
10228 btrfs_abort_transaction(trans
, ret
);
10229 add_block_group_free_space(trans
, block_group
);
10230 /* already aborted the transaction if it failed. */
10232 list_del_init(&block_group
->bg_list
);
10234 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10237 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
, u64 bytes_used
,
10238 u64 type
, u64 chunk_offset
, u64 size
)
10240 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
10241 struct btrfs_block_group_cache
*cache
;
10244 btrfs_set_log_full_commit(fs_info
, trans
);
10246 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
, size
);
10250 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10251 btrfs_set_block_group_chunk_objectid(&cache
->item
,
10252 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
10253 btrfs_set_block_group_flags(&cache
->item
, type
);
10255 cache
->flags
= type
;
10256 cache
->last_byte_to_unpin
= (u64
)-1;
10257 cache
->cached
= BTRFS_CACHE_FINISHED
;
10258 cache
->needs_free_space
= 1;
10259 ret
= exclude_super_stripes(fs_info
, cache
);
10262 * We may have excluded something, so call this just in
10265 free_excluded_extents(fs_info
, cache
);
10266 btrfs_put_block_group(cache
);
10270 add_new_free_space(cache
, chunk_offset
, chunk_offset
+ size
);
10272 free_excluded_extents(fs_info
, cache
);
10274 #ifdef CONFIG_BTRFS_DEBUG
10275 if (btrfs_should_fragment_free_space(cache
)) {
10276 u64 new_bytes_used
= size
- bytes_used
;
10278 bytes_used
+= new_bytes_used
>> 1;
10279 fragment_free_space(cache
);
10283 * Ensure the corresponding space_info object is created and
10284 * assigned to our block group. We want our bg to be added to the rbtree
10285 * with its ->space_info set.
10287 cache
->space_info
= __find_space_info(fs_info
, cache
->flags
);
10288 ASSERT(cache
->space_info
);
10290 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
10292 btrfs_remove_free_space_cache(cache
);
10293 btrfs_put_block_group(cache
);
10298 * Now that our block group has its ->space_info set and is inserted in
10299 * the rbtree, update the space info's counters.
10301 trace_btrfs_add_block_group(fs_info
, cache
, 1);
10302 update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
10303 cache
->bytes_super
, &cache
->space_info
);
10304 update_global_block_rsv(fs_info
);
10306 link_block_group(cache
);
10308 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10310 set_avail_alloc_bits(fs_info
, type
);
10314 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10316 u64 extra_flags
= chunk_to_extended(flags
) &
10317 BTRFS_EXTENDED_PROFILE_MASK
;
10319 write_seqlock(&fs_info
->profiles_lock
);
10320 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10321 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10322 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10323 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10324 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10325 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10326 write_sequnlock(&fs_info
->profiles_lock
);
10329 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10330 u64 group_start
, struct extent_map
*em
)
10332 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
10333 struct btrfs_root
*root
= fs_info
->extent_root
;
10334 struct btrfs_path
*path
;
10335 struct btrfs_block_group_cache
*block_group
;
10336 struct btrfs_free_cluster
*cluster
;
10337 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
10338 struct btrfs_key key
;
10339 struct inode
*inode
;
10340 struct kobject
*kobj
= NULL
;
10344 struct btrfs_caching_control
*caching_ctl
= NULL
;
10347 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
10348 BUG_ON(!block_group
);
10349 BUG_ON(!block_group
->ro
);
10351 trace_btrfs_remove_block_group(block_group
);
10353 * Free the reserved super bytes from this block group before
10356 free_excluded_extents(fs_info
, block_group
);
10357 btrfs_free_ref_tree_range(fs_info
, block_group
->key
.objectid
,
10358 block_group
->key
.offset
);
10360 memcpy(&key
, &block_group
->key
, sizeof(key
));
10361 index
= btrfs_bg_flags_to_raid_index(block_group
->flags
);
10362 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10363 BTRFS_BLOCK_GROUP_RAID1
|
10364 BTRFS_BLOCK_GROUP_RAID10
))
10369 /* make sure this block group isn't part of an allocation cluster */
10370 cluster
= &fs_info
->data_alloc_cluster
;
10371 spin_lock(&cluster
->refill_lock
);
10372 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10373 spin_unlock(&cluster
->refill_lock
);
10376 * make sure this block group isn't part of a metadata
10377 * allocation cluster
10379 cluster
= &fs_info
->meta_alloc_cluster
;
10380 spin_lock(&cluster
->refill_lock
);
10381 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10382 spin_unlock(&cluster
->refill_lock
);
10384 path
= btrfs_alloc_path();
10391 * get the inode first so any iput calls done for the io_list
10392 * aren't the final iput (no unlinks allowed now)
10394 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
10396 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10398 * make sure our free spache cache IO is done before remove the
10401 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10402 if (!list_empty(&block_group
->io_list
)) {
10403 list_del_init(&block_group
->io_list
);
10405 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10407 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10408 btrfs_wait_cache_io(trans
, block_group
, path
);
10409 btrfs_put_block_group(block_group
);
10410 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10413 if (!list_empty(&block_group
->dirty_list
)) {
10414 list_del_init(&block_group
->dirty_list
);
10415 btrfs_put_block_group(block_group
);
10417 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10418 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10420 if (!IS_ERR(inode
)) {
10421 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10423 btrfs_add_delayed_iput(inode
);
10426 clear_nlink(inode
);
10427 /* One for the block groups ref */
10428 spin_lock(&block_group
->lock
);
10429 if (block_group
->iref
) {
10430 block_group
->iref
= 0;
10431 block_group
->inode
= NULL
;
10432 spin_unlock(&block_group
->lock
);
10435 spin_unlock(&block_group
->lock
);
10437 /* One for our lookup ref */
10438 btrfs_add_delayed_iput(inode
);
10441 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10442 key
.offset
= block_group
->key
.objectid
;
10445 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10449 btrfs_release_path(path
);
10451 ret
= btrfs_del_item(trans
, tree_root
, path
);
10454 btrfs_release_path(path
);
10457 spin_lock(&fs_info
->block_group_cache_lock
);
10458 rb_erase(&block_group
->cache_node
,
10459 &fs_info
->block_group_cache_tree
);
10460 RB_CLEAR_NODE(&block_group
->cache_node
);
10462 if (fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10463 fs_info
->first_logical_byte
= (u64
)-1;
10464 spin_unlock(&fs_info
->block_group_cache_lock
);
10466 down_write(&block_group
->space_info
->groups_sem
);
10468 * we must use list_del_init so people can check to see if they
10469 * are still on the list after taking the semaphore
10471 list_del_init(&block_group
->list
);
10472 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10473 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10474 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10475 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
10477 up_write(&block_group
->space_info
->groups_sem
);
10483 if (block_group
->has_caching_ctl
)
10484 caching_ctl
= get_caching_control(block_group
);
10485 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10486 wait_block_group_cache_done(block_group
);
10487 if (block_group
->has_caching_ctl
) {
10488 down_write(&fs_info
->commit_root_sem
);
10489 if (!caching_ctl
) {
10490 struct btrfs_caching_control
*ctl
;
10492 list_for_each_entry(ctl
,
10493 &fs_info
->caching_block_groups
, list
)
10494 if (ctl
->block_group
== block_group
) {
10496 refcount_inc(&caching_ctl
->count
);
10501 list_del_init(&caching_ctl
->list
);
10502 up_write(&fs_info
->commit_root_sem
);
10504 /* Once for the caching bgs list and once for us. */
10505 put_caching_control(caching_ctl
);
10506 put_caching_control(caching_ctl
);
10510 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10511 if (!list_empty(&block_group
->dirty_list
)) {
10514 if (!list_empty(&block_group
->io_list
)) {
10517 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10518 btrfs_remove_free_space_cache(block_group
);
10520 spin_lock(&block_group
->space_info
->lock
);
10521 list_del_init(&block_group
->ro_list
);
10523 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
10524 WARN_ON(block_group
->space_info
->total_bytes
10525 < block_group
->key
.offset
);
10526 WARN_ON(block_group
->space_info
->bytes_readonly
10527 < block_group
->key
.offset
);
10528 WARN_ON(block_group
->space_info
->disk_total
10529 < block_group
->key
.offset
* factor
);
10531 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10532 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10533 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10535 spin_unlock(&block_group
->space_info
->lock
);
10537 memcpy(&key
, &block_group
->key
, sizeof(key
));
10539 mutex_lock(&fs_info
->chunk_mutex
);
10540 if (!list_empty(&em
->list
)) {
10541 /* We're in the transaction->pending_chunks list. */
10542 free_extent_map(em
);
10544 spin_lock(&block_group
->lock
);
10545 block_group
->removed
= 1;
10547 * At this point trimming can't start on this block group, because we
10548 * removed the block group from the tree fs_info->block_group_cache_tree
10549 * so no one can't find it anymore and even if someone already got this
10550 * block group before we removed it from the rbtree, they have already
10551 * incremented block_group->trimming - if they didn't, they won't find
10552 * any free space entries because we already removed them all when we
10553 * called btrfs_remove_free_space_cache().
10555 * And we must not remove the extent map from the fs_info->mapping_tree
10556 * to prevent the same logical address range and physical device space
10557 * ranges from being reused for a new block group. This is because our
10558 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10559 * completely transactionless, so while it is trimming a range the
10560 * currently running transaction might finish and a new one start,
10561 * allowing for new block groups to be created that can reuse the same
10562 * physical device locations unless we take this special care.
10564 * There may also be an implicit trim operation if the file system
10565 * is mounted with -odiscard. The same protections must remain
10566 * in place until the extents have been discarded completely when
10567 * the transaction commit has completed.
10569 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10571 * Make sure a trimmer task always sees the em in the pinned_chunks list
10572 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10573 * before checking block_group->removed).
10577 * Our em might be in trans->transaction->pending_chunks which
10578 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10579 * and so is the fs_info->pinned_chunks list.
10581 * So at this point we must be holding the chunk_mutex to avoid
10582 * any races with chunk allocation (more specifically at
10583 * volumes.c:contains_pending_extent()), to ensure it always
10584 * sees the em, either in the pending_chunks list or in the
10585 * pinned_chunks list.
10587 list_move_tail(&em
->list
, &fs_info
->pinned_chunks
);
10589 spin_unlock(&block_group
->lock
);
10592 struct extent_map_tree
*em_tree
;
10594 em_tree
= &fs_info
->mapping_tree
.map_tree
;
10595 write_lock(&em_tree
->lock
);
10597 * The em might be in the pending_chunks list, so make sure the
10598 * chunk mutex is locked, since remove_extent_mapping() will
10599 * delete us from that list.
10601 remove_extent_mapping(em_tree
, em
);
10602 write_unlock(&em_tree
->lock
);
10603 /* once for the tree */
10604 free_extent_map(em
);
10607 mutex_unlock(&fs_info
->chunk_mutex
);
10609 ret
= remove_block_group_free_space(trans
, block_group
);
10613 btrfs_put_block_group(block_group
);
10614 btrfs_put_block_group(block_group
);
10616 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10622 ret
= btrfs_del_item(trans
, root
, path
);
10624 btrfs_free_path(path
);
10628 struct btrfs_trans_handle
*
10629 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10630 const u64 chunk_offset
)
10632 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10633 struct extent_map
*em
;
10634 struct map_lookup
*map
;
10635 unsigned int num_items
;
10637 read_lock(&em_tree
->lock
);
10638 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10639 read_unlock(&em_tree
->lock
);
10640 ASSERT(em
&& em
->start
== chunk_offset
);
10643 * We need to reserve 3 + N units from the metadata space info in order
10644 * to remove a block group (done at btrfs_remove_chunk() and at
10645 * btrfs_remove_block_group()), which are used for:
10647 * 1 unit for adding the free space inode's orphan (located in the tree
10649 * 1 unit for deleting the block group item (located in the extent
10651 * 1 unit for deleting the free space item (located in tree of tree
10653 * N units for deleting N device extent items corresponding to each
10654 * stripe (located in the device tree).
10656 * In order to remove a block group we also need to reserve units in the
10657 * system space info in order to update the chunk tree (update one or
10658 * more device items and remove one chunk item), but this is done at
10659 * btrfs_remove_chunk() through a call to check_system_chunk().
10661 map
= em
->map_lookup
;
10662 num_items
= 3 + map
->num_stripes
;
10663 free_extent_map(em
);
10665 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10670 * Process the unused_bgs list and remove any that don't have any allocated
10671 * space inside of them.
10673 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10675 struct btrfs_block_group_cache
*block_group
;
10676 struct btrfs_space_info
*space_info
;
10677 struct btrfs_trans_handle
*trans
;
10680 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10683 spin_lock(&fs_info
->unused_bgs_lock
);
10684 while (!list_empty(&fs_info
->unused_bgs
)) {
10688 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10689 struct btrfs_block_group_cache
,
10691 list_del_init(&block_group
->bg_list
);
10693 space_info
= block_group
->space_info
;
10695 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10696 btrfs_put_block_group(block_group
);
10699 spin_unlock(&fs_info
->unused_bgs_lock
);
10701 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10703 /* Don't want to race with allocators so take the groups_sem */
10704 down_write(&space_info
->groups_sem
);
10705 spin_lock(&block_group
->lock
);
10706 if (block_group
->reserved
||
10707 btrfs_block_group_used(&block_group
->item
) ||
10709 list_is_singular(&block_group
->list
)) {
10711 * We want to bail if we made new allocations or have
10712 * outstanding allocations in this block group. We do
10713 * the ro check in case balance is currently acting on
10714 * this block group.
10716 trace_btrfs_skip_unused_block_group(block_group
);
10717 spin_unlock(&block_group
->lock
);
10718 up_write(&space_info
->groups_sem
);
10721 spin_unlock(&block_group
->lock
);
10723 /* We don't want to force the issue, only flip if it's ok. */
10724 ret
= inc_block_group_ro(block_group
, 0);
10725 up_write(&space_info
->groups_sem
);
10732 * Want to do this before we do anything else so we can recover
10733 * properly if we fail to join the transaction.
10735 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10736 block_group
->key
.objectid
);
10737 if (IS_ERR(trans
)) {
10738 btrfs_dec_block_group_ro(block_group
);
10739 ret
= PTR_ERR(trans
);
10744 * We could have pending pinned extents for this block group,
10745 * just delete them, we don't care about them anymore.
10747 start
= block_group
->key
.objectid
;
10748 end
= start
+ block_group
->key
.offset
- 1;
10750 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10751 * btrfs_finish_extent_commit(). If we are at transaction N,
10752 * another task might be running finish_extent_commit() for the
10753 * previous transaction N - 1, and have seen a range belonging
10754 * to the block group in freed_extents[] before we were able to
10755 * clear the whole block group range from freed_extents[]. This
10756 * means that task can lookup for the block group after we
10757 * unpinned it from freed_extents[] and removed it, leading to
10758 * a BUG_ON() at btrfs_unpin_extent_range().
10760 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10761 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10764 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10765 btrfs_dec_block_group_ro(block_group
);
10768 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10771 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10772 btrfs_dec_block_group_ro(block_group
);
10775 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10777 /* Reset pinned so btrfs_put_block_group doesn't complain */
10778 spin_lock(&space_info
->lock
);
10779 spin_lock(&block_group
->lock
);
10781 space_info
->bytes_pinned
-= block_group
->pinned
;
10782 space_info
->bytes_readonly
+= block_group
->pinned
;
10783 percpu_counter_add(&space_info
->total_bytes_pinned
,
10784 -block_group
->pinned
);
10785 block_group
->pinned
= 0;
10787 spin_unlock(&block_group
->lock
);
10788 spin_unlock(&space_info
->lock
);
10790 /* DISCARD can flip during remount */
10791 trimming
= btrfs_test_opt(fs_info
, DISCARD
);
10793 /* Implicit trim during transaction commit. */
10795 btrfs_get_block_group_trimming(block_group
);
10798 * Btrfs_remove_chunk will abort the transaction if things go
10801 ret
= btrfs_remove_chunk(trans
, fs_info
,
10802 block_group
->key
.objectid
);
10806 btrfs_put_block_group_trimming(block_group
);
10811 * If we're not mounted with -odiscard, we can just forget
10812 * about this block group. Otherwise we'll need to wait
10813 * until transaction commit to do the actual discard.
10816 spin_lock(&fs_info
->unused_bgs_lock
);
10818 * A concurrent scrub might have added us to the list
10819 * fs_info->unused_bgs, so use a list_move operation
10820 * to add the block group to the deleted_bgs list.
10822 list_move(&block_group
->bg_list
,
10823 &trans
->transaction
->deleted_bgs
);
10824 spin_unlock(&fs_info
->unused_bgs_lock
);
10825 btrfs_get_block_group(block_group
);
10828 btrfs_end_transaction(trans
);
10830 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10831 btrfs_put_block_group(block_group
);
10832 spin_lock(&fs_info
->unused_bgs_lock
);
10834 spin_unlock(&fs_info
->unused_bgs_lock
);
10837 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10839 struct btrfs_super_block
*disk_super
;
10845 disk_super
= fs_info
->super_copy
;
10846 if (!btrfs_super_root(disk_super
))
10849 features
= btrfs_super_incompat_flags(disk_super
);
10850 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10853 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10854 ret
= create_space_info(fs_info
, flags
);
10859 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10860 ret
= create_space_info(fs_info
, flags
);
10862 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10863 ret
= create_space_info(fs_info
, flags
);
10867 flags
= BTRFS_BLOCK_GROUP_DATA
;
10868 ret
= create_space_info(fs_info
, flags
);
10874 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
10875 u64 start
, u64 end
)
10877 return unpin_extent_range(fs_info
, start
, end
, false);
10881 * It used to be that old block groups would be left around forever.
10882 * Iterating over them would be enough to trim unused space. Since we
10883 * now automatically remove them, we also need to iterate over unallocated
10886 * We don't want a transaction for this since the discard may take a
10887 * substantial amount of time. We don't require that a transaction be
10888 * running, but we do need to take a running transaction into account
10889 * to ensure that we're not discarding chunks that were released in
10890 * the current transaction.
10892 * Holding the chunks lock will prevent other threads from allocating
10893 * or releasing chunks, but it won't prevent a running transaction
10894 * from committing and releasing the memory that the pending chunks
10895 * list head uses. For that, we need to take a reference to the
10898 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10899 u64 minlen
, u64
*trimmed
)
10901 u64 start
= 0, len
= 0;
10906 /* Not writeable = nothing to do. */
10907 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
10910 /* No free space = nothing to do. */
10911 if (device
->total_bytes
<= device
->bytes_used
)
10917 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
10918 struct btrfs_transaction
*trans
;
10921 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10925 down_read(&fs_info
->commit_root_sem
);
10927 spin_lock(&fs_info
->trans_lock
);
10928 trans
= fs_info
->running_transaction
;
10930 refcount_inc(&trans
->use_count
);
10931 spin_unlock(&fs_info
->trans_lock
);
10933 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10936 btrfs_put_transaction(trans
);
10939 up_read(&fs_info
->commit_root_sem
);
10940 mutex_unlock(&fs_info
->chunk_mutex
);
10941 if (ret
== -ENOSPC
)
10946 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10947 up_read(&fs_info
->commit_root_sem
);
10948 mutex_unlock(&fs_info
->chunk_mutex
);
10956 if (fatal_signal_pending(current
)) {
10957 ret
= -ERESTARTSYS
;
10967 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
10969 struct btrfs_block_group_cache
*cache
= NULL
;
10970 struct btrfs_device
*device
;
10971 struct list_head
*devices
;
10976 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10980 * try to trim all FS space, our block group may start from non-zero.
10982 if (range
->len
== total_bytes
)
10983 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10985 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10988 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10989 btrfs_put_block_group(cache
);
10993 start
= max(range
->start
, cache
->key
.objectid
);
10994 end
= min(range
->start
+ range
->len
,
10995 cache
->key
.objectid
+ cache
->key
.offset
);
10997 if (end
- start
>= range
->minlen
) {
10998 if (!block_group_cache_done(cache
)) {
10999 ret
= cache_block_group(cache
, 0);
11001 btrfs_put_block_group(cache
);
11004 ret
= wait_block_group_cache_done(cache
);
11006 btrfs_put_block_group(cache
);
11010 ret
= btrfs_trim_block_group(cache
,
11016 trimmed
+= group_trimmed
;
11018 btrfs_put_block_group(cache
);
11023 cache
= next_block_group(fs_info
, cache
);
11026 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
11027 devices
= &fs_info
->fs_devices
->alloc_list
;
11028 list_for_each_entry(device
, devices
, dev_alloc_list
) {
11029 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
11034 trimmed
+= group_trimmed
;
11036 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
11038 range
->len
= trimmed
;
11043 * btrfs_{start,end}_write_no_snapshotting() are similar to
11044 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11045 * data into the page cache through nocow before the subvolume is snapshoted,
11046 * but flush the data into disk after the snapshot creation, or to prevent
11047 * operations while snapshotting is ongoing and that cause the snapshot to be
11048 * inconsistent (writes followed by expanding truncates for example).
11050 void btrfs_end_write_no_snapshotting(struct btrfs_root
*root
)
11052 percpu_counter_dec(&root
->subv_writers
->counter
);
11053 cond_wake_up(&root
->subv_writers
->wait
);
11056 int btrfs_start_write_no_snapshotting(struct btrfs_root
*root
)
11058 if (atomic_read(&root
->will_be_snapshotted
))
11061 percpu_counter_inc(&root
->subv_writers
->counter
);
11063 * Make sure counter is updated before we check for snapshot creation.
11066 if (atomic_read(&root
->will_be_snapshotted
)) {
11067 btrfs_end_write_no_snapshotting(root
);
11073 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11078 ret
= btrfs_start_write_no_snapshotting(root
);
11081 wait_var_event(&root
->will_be_snapshotted
,
11082 !atomic_read(&root
->will_be_snapshotted
));