1 // SPDX-License-Identifier: GPL-2.0
5 #include "block-group.h"
6 #include "space-info.h"
8 #include "free-space-cache.h"
9 #include "free-space-tree.h"
11 #include "transaction.h"
12 #include "ref-verify.h"
15 #include "delalloc-space.h"
20 * Return target flags in extended format or 0 if restripe for this chunk_type
23 * Should be called with balance_lock held
25 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
27 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
33 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
34 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
35 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
36 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
37 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
38 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
39 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
40 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
41 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
48 * @flags: available profiles in extended format (see ctree.h)
50 * Return reduced profile in chunk format. If profile changing is in progress
51 * (either running or paused) picks the target profile (if it's already
52 * available), otherwise falls back to plain reducing.
54 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
56 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
62 * See if restripe for this chunk_type is in progress, if so try to
63 * reduce to the target profile
65 spin_lock(&fs_info
->balance_lock
);
66 target
= get_restripe_target(fs_info
, flags
);
68 spin_unlock(&fs_info
->balance_lock
);
69 return extended_to_chunk(target
);
71 spin_unlock(&fs_info
->balance_lock
);
73 /* First, mask out the RAID levels which aren't possible */
74 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
75 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
76 allowed
|= btrfs_raid_array
[raid_type
].bg_flag
;
80 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
81 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
82 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
83 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
84 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
85 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
86 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
87 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
88 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
89 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
91 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
93 return extended_to_chunk(flags
| allowed
);
96 u64
btrfs_get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
103 seq
= read_seqbegin(&fs_info
->profiles_lock
);
105 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
106 flags
|= fs_info
->avail_data_alloc_bits
;
107 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
108 flags
|= fs_info
->avail_system_alloc_bits
;
109 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
110 flags
|= fs_info
->avail_metadata_alloc_bits
;
111 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
113 return btrfs_reduce_alloc_profile(fs_info
, flags
);
116 void btrfs_get_block_group(struct btrfs_block_group
*cache
)
118 refcount_inc(&cache
->refs
);
121 void btrfs_put_block_group(struct btrfs_block_group
*cache
)
123 if (refcount_dec_and_test(&cache
->refs
)) {
124 WARN_ON(cache
->pinned
> 0);
125 WARN_ON(cache
->reserved
> 0);
128 * A block_group shouldn't be on the discard_list anymore.
129 * Remove the block_group from the discard_list to prevent us
130 * from causing a panic due to NULL pointer dereference.
132 if (WARN_ON(!list_empty(&cache
->discard_list
)))
133 btrfs_discard_cancel_work(&cache
->fs_info
->discard_ctl
,
137 * If not empty, someone is still holding mutex of
138 * full_stripe_lock, which can only be released by caller.
139 * And it will definitely cause use-after-free when caller
140 * tries to release full stripe lock.
142 * No better way to resolve, but only to warn.
144 WARN_ON(!RB_EMPTY_ROOT(&cache
->full_stripe_locks_root
.root
));
145 kfree(cache
->free_space_ctl
);
151 * This adds the block group to the fs_info rb tree for the block group cache
153 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
154 struct btrfs_block_group
*block_group
)
157 struct rb_node
*parent
= NULL
;
158 struct btrfs_block_group
*cache
;
160 ASSERT(block_group
->length
!= 0);
162 spin_lock(&info
->block_group_cache_lock
);
163 p
= &info
->block_group_cache_tree
.rb_node
;
167 cache
= rb_entry(parent
, struct btrfs_block_group
, cache_node
);
168 if (block_group
->start
< cache
->start
) {
170 } else if (block_group
->start
> cache
->start
) {
173 spin_unlock(&info
->block_group_cache_lock
);
178 rb_link_node(&block_group
->cache_node
, parent
, p
);
179 rb_insert_color(&block_group
->cache_node
,
180 &info
->block_group_cache_tree
);
182 if (info
->first_logical_byte
> block_group
->start
)
183 info
->first_logical_byte
= block_group
->start
;
185 spin_unlock(&info
->block_group_cache_lock
);
191 * This will return the block group at or after bytenr if contains is 0, else
192 * it will return the block group that contains the bytenr
194 static struct btrfs_block_group
*block_group_cache_tree_search(
195 struct btrfs_fs_info
*info
, u64 bytenr
, int contains
)
197 struct btrfs_block_group
*cache
, *ret
= NULL
;
201 spin_lock(&info
->block_group_cache_lock
);
202 n
= info
->block_group_cache_tree
.rb_node
;
205 cache
= rb_entry(n
, struct btrfs_block_group
, cache_node
);
206 end
= cache
->start
+ cache
->length
- 1;
207 start
= cache
->start
;
209 if (bytenr
< start
) {
210 if (!contains
&& (!ret
|| start
< ret
->start
))
213 } else if (bytenr
> start
) {
214 if (contains
&& bytenr
<= end
) {
225 btrfs_get_block_group(ret
);
226 if (bytenr
== 0 && info
->first_logical_byte
> ret
->start
)
227 info
->first_logical_byte
= ret
->start
;
229 spin_unlock(&info
->block_group_cache_lock
);
235 * Return the block group that starts at or after bytenr
237 struct btrfs_block_group
*btrfs_lookup_first_block_group(
238 struct btrfs_fs_info
*info
, u64 bytenr
)
240 return block_group_cache_tree_search(info
, bytenr
, 0);
244 * Return the block group that contains the given bytenr
246 struct btrfs_block_group
*btrfs_lookup_block_group(
247 struct btrfs_fs_info
*info
, u64 bytenr
)
249 return block_group_cache_tree_search(info
, bytenr
, 1);
252 struct btrfs_block_group
*btrfs_next_block_group(
253 struct btrfs_block_group
*cache
)
255 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
256 struct rb_node
*node
;
258 spin_lock(&fs_info
->block_group_cache_lock
);
260 /* If our block group was removed, we need a full search. */
261 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
262 const u64 next_bytenr
= cache
->start
+ cache
->length
;
264 spin_unlock(&fs_info
->block_group_cache_lock
);
265 btrfs_put_block_group(cache
);
266 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
268 node
= rb_next(&cache
->cache_node
);
269 btrfs_put_block_group(cache
);
271 cache
= rb_entry(node
, struct btrfs_block_group
, cache_node
);
272 btrfs_get_block_group(cache
);
275 spin_unlock(&fs_info
->block_group_cache_lock
);
279 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
281 struct btrfs_block_group
*bg
;
284 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
288 spin_lock(&bg
->lock
);
292 atomic_inc(&bg
->nocow_writers
);
293 spin_unlock(&bg
->lock
);
295 /* No put on block group, done by btrfs_dec_nocow_writers */
297 btrfs_put_block_group(bg
);
302 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
304 struct btrfs_block_group
*bg
;
306 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
308 if (atomic_dec_and_test(&bg
->nocow_writers
))
309 wake_up_var(&bg
->nocow_writers
);
311 * Once for our lookup and once for the lookup done by a previous call
312 * to btrfs_inc_nocow_writers()
314 btrfs_put_block_group(bg
);
315 btrfs_put_block_group(bg
);
318 void btrfs_wait_nocow_writers(struct btrfs_block_group
*bg
)
320 wait_var_event(&bg
->nocow_writers
, !atomic_read(&bg
->nocow_writers
));
323 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
326 struct btrfs_block_group
*bg
;
328 bg
= btrfs_lookup_block_group(fs_info
, start
);
330 if (atomic_dec_and_test(&bg
->reservations
))
331 wake_up_var(&bg
->reservations
);
332 btrfs_put_block_group(bg
);
335 void btrfs_wait_block_group_reservations(struct btrfs_block_group
*bg
)
337 struct btrfs_space_info
*space_info
= bg
->space_info
;
341 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
345 * Our block group is read only but before we set it to read only,
346 * some task might have had allocated an extent from it already, but it
347 * has not yet created a respective ordered extent (and added it to a
348 * root's list of ordered extents).
349 * Therefore wait for any task currently allocating extents, since the
350 * block group's reservations counter is incremented while a read lock
351 * on the groups' semaphore is held and decremented after releasing
352 * the read access on that semaphore and creating the ordered extent.
354 down_write(&space_info
->groups_sem
);
355 up_write(&space_info
->groups_sem
);
357 wait_var_event(&bg
->reservations
, !atomic_read(&bg
->reservations
));
360 struct btrfs_caching_control
*btrfs_get_caching_control(
361 struct btrfs_block_group
*cache
)
363 struct btrfs_caching_control
*ctl
;
365 spin_lock(&cache
->lock
);
366 if (!cache
->caching_ctl
) {
367 spin_unlock(&cache
->lock
);
371 ctl
= cache
->caching_ctl
;
372 refcount_inc(&ctl
->count
);
373 spin_unlock(&cache
->lock
);
377 void btrfs_put_caching_control(struct btrfs_caching_control
*ctl
)
379 if (refcount_dec_and_test(&ctl
->count
))
384 * When we wait for progress in the block group caching, its because our
385 * allocation attempt failed at least once. So, we must sleep and let some
386 * progress happen before we try again.
388 * This function will sleep at least once waiting for new free space to show
389 * up, and then it will check the block group free space numbers for our min
390 * num_bytes. Another option is to have it go ahead and look in the rbtree for
391 * a free extent of a given size, but this is a good start.
393 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
394 * any of the information in this block group.
396 void btrfs_wait_block_group_cache_progress(struct btrfs_block_group
*cache
,
399 struct btrfs_caching_control
*caching_ctl
;
401 caching_ctl
= btrfs_get_caching_control(cache
);
405 wait_event(caching_ctl
->wait
, btrfs_block_group_done(cache
) ||
406 (cache
->free_space_ctl
->free_space
>= num_bytes
));
408 btrfs_put_caching_control(caching_ctl
);
411 int btrfs_wait_block_group_cache_done(struct btrfs_block_group
*cache
)
413 struct btrfs_caching_control
*caching_ctl
;
416 caching_ctl
= btrfs_get_caching_control(cache
);
418 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
420 wait_event(caching_ctl
->wait
, btrfs_block_group_done(cache
));
421 if (cache
->cached
== BTRFS_CACHE_ERROR
)
423 btrfs_put_caching_control(caching_ctl
);
427 #ifdef CONFIG_BTRFS_DEBUG
428 static void fragment_free_space(struct btrfs_block_group
*block_group
)
430 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
431 u64 start
= block_group
->start
;
432 u64 len
= block_group
->length
;
433 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
434 fs_info
->nodesize
: fs_info
->sectorsize
;
435 u64 step
= chunk
<< 1;
437 while (len
> chunk
) {
438 btrfs_remove_free_space(block_group
, start
, chunk
);
449 * This is only called by btrfs_cache_block_group, since we could have freed
450 * extents we need to check the pinned_extents for any extents that can't be
451 * used yet since their free space will be released as soon as the transaction
454 u64
add_new_free_space(struct btrfs_block_group
*block_group
, u64 start
, u64 end
)
456 struct btrfs_fs_info
*info
= block_group
->fs_info
;
457 u64 extent_start
, extent_end
, size
, total_added
= 0;
460 while (start
< end
) {
461 ret
= find_first_extent_bit(&info
->excluded_extents
, start
,
462 &extent_start
, &extent_end
,
463 EXTENT_DIRTY
| EXTENT_UPTODATE
,
468 if (extent_start
<= start
) {
469 start
= extent_end
+ 1;
470 } else if (extent_start
> start
&& extent_start
< end
) {
471 size
= extent_start
- start
;
473 ret
= btrfs_add_free_space_async_trimmed(block_group
,
475 BUG_ON(ret
); /* -ENOMEM or logic error */
476 start
= extent_end
+ 1;
485 ret
= btrfs_add_free_space_async_trimmed(block_group
, start
,
487 BUG_ON(ret
); /* -ENOMEM or logic error */
493 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
495 struct btrfs_block_group
*block_group
= caching_ctl
->block_group
;
496 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
497 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
498 struct btrfs_path
*path
;
499 struct extent_buffer
*leaf
;
500 struct btrfs_key key
;
507 path
= btrfs_alloc_path();
511 last
= max_t(u64
, block_group
->start
, BTRFS_SUPER_INFO_OFFSET
);
513 #ifdef CONFIG_BTRFS_DEBUG
515 * If we're fragmenting we don't want to make anybody think we can
516 * allocate from this block group until we've had a chance to fragment
519 if (btrfs_should_fragment_free_space(block_group
))
523 * We don't want to deadlock with somebody trying to allocate a new
524 * extent for the extent root while also trying to search the extent
525 * root to add free space. So we skip locking and search the commit
526 * root, since its read-only
528 path
->skip_locking
= 1;
529 path
->search_commit_root
= 1;
530 path
->reada
= READA_FORWARD
;
534 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
537 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
541 leaf
= path
->nodes
[0];
542 nritems
= btrfs_header_nritems(leaf
);
545 if (btrfs_fs_closing(fs_info
) > 1) {
550 if (path
->slots
[0] < nritems
) {
551 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
553 ret
= btrfs_find_next_key(extent_root
, path
, &key
, 0, 0);
557 if (need_resched() ||
558 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
560 caching_ctl
->progress
= last
;
561 btrfs_release_path(path
);
562 up_read(&fs_info
->commit_root_sem
);
563 mutex_unlock(&caching_ctl
->mutex
);
565 mutex_lock(&caching_ctl
->mutex
);
566 down_read(&fs_info
->commit_root_sem
);
570 ret
= btrfs_next_leaf(extent_root
, path
);
575 leaf
= path
->nodes
[0];
576 nritems
= btrfs_header_nritems(leaf
);
580 if (key
.objectid
< last
) {
583 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
586 caching_ctl
->progress
= last
;
587 btrfs_release_path(path
);
591 if (key
.objectid
< block_group
->start
) {
596 if (key
.objectid
>= block_group
->start
+ block_group
->length
)
599 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
600 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
601 total_found
+= add_new_free_space(block_group
, last
,
603 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
604 last
= key
.objectid
+
607 last
= key
.objectid
+ key
.offset
;
609 if (total_found
> CACHING_CTL_WAKE_UP
) {
612 wake_up(&caching_ctl
->wait
);
619 total_found
+= add_new_free_space(block_group
, last
,
620 block_group
->start
+ block_group
->length
);
621 caching_ctl
->progress
= (u64
)-1;
624 btrfs_free_path(path
);
628 static noinline
void caching_thread(struct btrfs_work
*work
)
630 struct btrfs_block_group
*block_group
;
631 struct btrfs_fs_info
*fs_info
;
632 struct btrfs_caching_control
*caching_ctl
;
635 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
636 block_group
= caching_ctl
->block_group
;
637 fs_info
= block_group
->fs_info
;
639 mutex_lock(&caching_ctl
->mutex
);
640 down_read(&fs_info
->commit_root_sem
);
642 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
643 ret
= load_free_space_tree(caching_ctl
);
645 ret
= load_extent_tree_free(caching_ctl
);
647 spin_lock(&block_group
->lock
);
648 block_group
->caching_ctl
= NULL
;
649 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
650 spin_unlock(&block_group
->lock
);
652 #ifdef CONFIG_BTRFS_DEBUG
653 if (btrfs_should_fragment_free_space(block_group
)) {
656 spin_lock(&block_group
->space_info
->lock
);
657 spin_lock(&block_group
->lock
);
658 bytes_used
= block_group
->length
- block_group
->used
;
659 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
660 spin_unlock(&block_group
->lock
);
661 spin_unlock(&block_group
->space_info
->lock
);
662 fragment_free_space(block_group
);
666 caching_ctl
->progress
= (u64
)-1;
668 up_read(&fs_info
->commit_root_sem
);
669 btrfs_free_excluded_extents(block_group
);
670 mutex_unlock(&caching_ctl
->mutex
);
672 wake_up(&caching_ctl
->wait
);
674 btrfs_put_caching_control(caching_ctl
);
675 btrfs_put_block_group(block_group
);
678 int btrfs_cache_block_group(struct btrfs_block_group
*cache
, int load_cache_only
)
681 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
682 struct btrfs_caching_control
*caching_ctl
;
685 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
689 INIT_LIST_HEAD(&caching_ctl
->list
);
690 mutex_init(&caching_ctl
->mutex
);
691 init_waitqueue_head(&caching_ctl
->wait
);
692 caching_ctl
->block_group
= cache
;
693 caching_ctl
->progress
= cache
->start
;
694 refcount_set(&caching_ctl
->count
, 1);
695 btrfs_init_work(&caching_ctl
->work
, caching_thread
, NULL
, NULL
);
697 spin_lock(&cache
->lock
);
699 * This should be a rare occasion, but this could happen I think in the
700 * case where one thread starts to load the space cache info, and then
701 * some other thread starts a transaction commit which tries to do an
702 * allocation while the other thread is still loading the space cache
703 * info. The previous loop should have kept us from choosing this block
704 * group, but if we've moved to the state where we will wait on caching
705 * block groups we need to first check if we're doing a fast load here,
706 * so we can wait for it to finish, otherwise we could end up allocating
707 * from a block group who's cache gets evicted for one reason or
710 while (cache
->cached
== BTRFS_CACHE_FAST
) {
711 struct btrfs_caching_control
*ctl
;
713 ctl
= cache
->caching_ctl
;
714 refcount_inc(&ctl
->count
);
715 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
716 spin_unlock(&cache
->lock
);
720 finish_wait(&ctl
->wait
, &wait
);
721 btrfs_put_caching_control(ctl
);
722 spin_lock(&cache
->lock
);
725 if (cache
->cached
!= BTRFS_CACHE_NO
) {
726 spin_unlock(&cache
->lock
);
730 WARN_ON(cache
->caching_ctl
);
731 cache
->caching_ctl
= caching_ctl
;
732 cache
->cached
= BTRFS_CACHE_FAST
;
733 spin_unlock(&cache
->lock
);
735 if (btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
736 mutex_lock(&caching_ctl
->mutex
);
737 ret
= load_free_space_cache(cache
);
739 spin_lock(&cache
->lock
);
741 cache
->caching_ctl
= NULL
;
742 cache
->cached
= BTRFS_CACHE_FINISHED
;
743 cache
->last_byte_to_unpin
= (u64
)-1;
744 caching_ctl
->progress
= (u64
)-1;
746 if (load_cache_only
) {
747 cache
->caching_ctl
= NULL
;
748 cache
->cached
= BTRFS_CACHE_NO
;
750 cache
->cached
= BTRFS_CACHE_STARTED
;
751 cache
->has_caching_ctl
= 1;
754 spin_unlock(&cache
->lock
);
755 #ifdef CONFIG_BTRFS_DEBUG
757 btrfs_should_fragment_free_space(cache
)) {
760 spin_lock(&cache
->space_info
->lock
);
761 spin_lock(&cache
->lock
);
762 bytes_used
= cache
->length
- cache
->used
;
763 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
764 spin_unlock(&cache
->lock
);
765 spin_unlock(&cache
->space_info
->lock
);
766 fragment_free_space(cache
);
769 mutex_unlock(&caching_ctl
->mutex
);
771 wake_up(&caching_ctl
->wait
);
773 btrfs_put_caching_control(caching_ctl
);
774 btrfs_free_excluded_extents(cache
);
779 * We're either using the free space tree or no caching at all.
780 * Set cached to the appropriate value and wakeup any waiters.
782 spin_lock(&cache
->lock
);
783 if (load_cache_only
) {
784 cache
->caching_ctl
= NULL
;
785 cache
->cached
= BTRFS_CACHE_NO
;
787 cache
->cached
= BTRFS_CACHE_STARTED
;
788 cache
->has_caching_ctl
= 1;
790 spin_unlock(&cache
->lock
);
791 wake_up(&caching_ctl
->wait
);
794 if (load_cache_only
) {
795 btrfs_put_caching_control(caching_ctl
);
799 down_write(&fs_info
->commit_root_sem
);
800 refcount_inc(&caching_ctl
->count
);
801 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
802 up_write(&fs_info
->commit_root_sem
);
804 btrfs_get_block_group(cache
);
806 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
811 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
813 u64 extra_flags
= chunk_to_extended(flags
) &
814 BTRFS_EXTENDED_PROFILE_MASK
;
816 write_seqlock(&fs_info
->profiles_lock
);
817 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
818 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
819 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
820 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
821 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
822 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
823 write_sequnlock(&fs_info
->profiles_lock
);
827 * Clear incompat bits for the following feature(s):
829 * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group
830 * in the whole filesystem
832 * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups
834 static void clear_incompat_bg_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
836 bool found_raid56
= false;
837 bool found_raid1c34
= false;
839 if ((flags
& BTRFS_BLOCK_GROUP_RAID56_MASK
) ||
840 (flags
& BTRFS_BLOCK_GROUP_RAID1C3
) ||
841 (flags
& BTRFS_BLOCK_GROUP_RAID1C4
)) {
842 struct list_head
*head
= &fs_info
->space_info
;
843 struct btrfs_space_info
*sinfo
;
845 list_for_each_entry_rcu(sinfo
, head
, list
) {
846 down_read(&sinfo
->groups_sem
);
847 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID5
]))
849 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID6
]))
851 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID1C3
]))
852 found_raid1c34
= true;
853 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID1C4
]))
854 found_raid1c34
= true;
855 up_read(&sinfo
->groups_sem
);
858 btrfs_clear_fs_incompat(fs_info
, RAID56
);
860 btrfs_clear_fs_incompat(fs_info
, RAID1C34
);
864 static int remove_block_group_item(struct btrfs_trans_handle
*trans
,
865 struct btrfs_path
*path
,
866 struct btrfs_block_group
*block_group
)
868 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
869 struct btrfs_root
*root
;
870 struct btrfs_key key
;
873 root
= fs_info
->extent_root
;
874 key
.objectid
= block_group
->start
;
875 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
876 key
.offset
= block_group
->length
;
878 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
884 ret
= btrfs_del_item(trans
, root
, path
);
888 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
889 u64 group_start
, struct extent_map
*em
)
891 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
892 struct btrfs_path
*path
;
893 struct btrfs_block_group
*block_group
;
894 struct btrfs_free_cluster
*cluster
;
895 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
896 struct btrfs_key key
;
898 struct kobject
*kobj
= NULL
;
902 struct btrfs_caching_control
*caching_ctl
= NULL
;
904 bool remove_rsv
= false;
906 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
907 BUG_ON(!block_group
);
908 BUG_ON(!block_group
->ro
);
910 trace_btrfs_remove_block_group(block_group
);
912 * Free the reserved super bytes from this block group before
915 btrfs_free_excluded_extents(block_group
);
916 btrfs_free_ref_tree_range(fs_info
, block_group
->start
,
917 block_group
->length
);
919 index
= btrfs_bg_flags_to_raid_index(block_group
->flags
);
920 factor
= btrfs_bg_type_to_factor(block_group
->flags
);
922 /* make sure this block group isn't part of an allocation cluster */
923 cluster
= &fs_info
->data_alloc_cluster
;
924 spin_lock(&cluster
->refill_lock
);
925 btrfs_return_cluster_to_free_space(block_group
, cluster
);
926 spin_unlock(&cluster
->refill_lock
);
929 * make sure this block group isn't part of a metadata
932 cluster
= &fs_info
->meta_alloc_cluster
;
933 spin_lock(&cluster
->refill_lock
);
934 btrfs_return_cluster_to_free_space(block_group
, cluster
);
935 spin_unlock(&cluster
->refill_lock
);
937 path
= btrfs_alloc_path();
944 * get the inode first so any iput calls done for the io_list
945 * aren't the final iput (no unlinks allowed now)
947 inode
= lookup_free_space_inode(block_group
, path
);
949 mutex_lock(&trans
->transaction
->cache_write_mutex
);
951 * Make sure our free space cache IO is done before removing the
954 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
955 if (!list_empty(&block_group
->io_list
)) {
956 list_del_init(&block_group
->io_list
);
958 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
960 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
961 btrfs_wait_cache_io(trans
, block_group
, path
);
962 btrfs_put_block_group(block_group
);
963 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
966 if (!list_empty(&block_group
->dirty_list
)) {
967 list_del_init(&block_group
->dirty_list
);
969 btrfs_put_block_group(block_group
);
971 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
972 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
974 if (!IS_ERR(inode
)) {
975 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
977 btrfs_add_delayed_iput(inode
);
981 /* One for the block groups ref */
982 spin_lock(&block_group
->lock
);
983 if (block_group
->iref
) {
984 block_group
->iref
= 0;
985 block_group
->inode
= NULL
;
986 spin_unlock(&block_group
->lock
);
989 spin_unlock(&block_group
->lock
);
991 /* One for our lookup ref */
992 btrfs_add_delayed_iput(inode
);
995 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
997 key
.offset
= block_group
->start
;
999 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
1003 btrfs_release_path(path
);
1005 ret
= btrfs_del_item(trans
, tree_root
, path
);
1008 btrfs_release_path(path
);
1011 spin_lock(&fs_info
->block_group_cache_lock
);
1012 rb_erase(&block_group
->cache_node
,
1013 &fs_info
->block_group_cache_tree
);
1014 RB_CLEAR_NODE(&block_group
->cache_node
);
1016 /* Once for the block groups rbtree */
1017 btrfs_put_block_group(block_group
);
1019 if (fs_info
->first_logical_byte
== block_group
->start
)
1020 fs_info
->first_logical_byte
= (u64
)-1;
1021 spin_unlock(&fs_info
->block_group_cache_lock
);
1023 down_write(&block_group
->space_info
->groups_sem
);
1025 * we must use list_del_init so people can check to see if they
1026 * are still on the list after taking the semaphore
1028 list_del_init(&block_group
->list
);
1029 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
1030 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
1031 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
1032 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
1034 up_write(&block_group
->space_info
->groups_sem
);
1035 clear_incompat_bg_bits(fs_info
, block_group
->flags
);
1041 if (block_group
->has_caching_ctl
)
1042 caching_ctl
= btrfs_get_caching_control(block_group
);
1043 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
1044 btrfs_wait_block_group_cache_done(block_group
);
1045 if (block_group
->has_caching_ctl
) {
1046 down_write(&fs_info
->commit_root_sem
);
1048 struct btrfs_caching_control
*ctl
;
1050 list_for_each_entry(ctl
,
1051 &fs_info
->caching_block_groups
, list
)
1052 if (ctl
->block_group
== block_group
) {
1054 refcount_inc(&caching_ctl
->count
);
1059 list_del_init(&caching_ctl
->list
);
1060 up_write(&fs_info
->commit_root_sem
);
1062 /* Once for the caching bgs list and once for us. */
1063 btrfs_put_caching_control(caching_ctl
);
1064 btrfs_put_caching_control(caching_ctl
);
1068 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
1069 WARN_ON(!list_empty(&block_group
->dirty_list
));
1070 WARN_ON(!list_empty(&block_group
->io_list
));
1071 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
1073 btrfs_remove_free_space_cache(block_group
);
1075 spin_lock(&block_group
->space_info
->lock
);
1076 list_del_init(&block_group
->ro_list
);
1078 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
1079 WARN_ON(block_group
->space_info
->total_bytes
1080 < block_group
->length
);
1081 WARN_ON(block_group
->space_info
->bytes_readonly
1082 < block_group
->length
);
1083 WARN_ON(block_group
->space_info
->disk_total
1084 < block_group
->length
* factor
);
1086 block_group
->space_info
->total_bytes
-= block_group
->length
;
1087 block_group
->space_info
->bytes_readonly
-= block_group
->length
;
1088 block_group
->space_info
->disk_total
-= block_group
->length
* factor
;
1090 spin_unlock(&block_group
->space_info
->lock
);
1093 * Remove the free space for the block group from the free space tree
1094 * and the block group's item from the extent tree before marking the
1095 * block group as removed. This is to prevent races with tasks that
1096 * freeze and unfreeze a block group, this task and another task
1097 * allocating a new block group - the unfreeze task ends up removing
1098 * the block group's extent map before the task calling this function
1099 * deletes the block group item from the extent tree, allowing for
1100 * another task to attempt to create another block group with the same
1101 * item key (and failing with -EEXIST and a transaction abort).
1103 ret
= remove_block_group_free_space(trans
, block_group
);
1107 ret
= remove_block_group_item(trans
, path
, block_group
);
1111 spin_lock(&block_group
->lock
);
1112 block_group
->removed
= 1;
1114 * At this point trimming or scrub can't start on this block group,
1115 * because we removed the block group from the rbtree
1116 * fs_info->block_group_cache_tree so no one can't find it anymore and
1117 * even if someone already got this block group before we removed it
1118 * from the rbtree, they have already incremented block_group->frozen -
1119 * if they didn't, for the trimming case they won't find any free space
1120 * entries because we already removed them all when we called
1121 * btrfs_remove_free_space_cache().
1123 * And we must not remove the extent map from the fs_info->mapping_tree
1124 * to prevent the same logical address range and physical device space
1125 * ranges from being reused for a new block group. This is needed to
1126 * avoid races with trimming and scrub.
1128 * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
1129 * completely transactionless, so while it is trimming a range the
1130 * currently running transaction might finish and a new one start,
1131 * allowing for new block groups to be created that can reuse the same
1132 * physical device locations unless we take this special care.
1134 * There may also be an implicit trim operation if the file system
1135 * is mounted with -odiscard. The same protections must remain
1136 * in place until the extents have been discarded completely when
1137 * the transaction commit has completed.
1139 remove_em
= (atomic_read(&block_group
->frozen
) == 0);
1140 spin_unlock(&block_group
->lock
);
1143 struct extent_map_tree
*em_tree
;
1145 em_tree
= &fs_info
->mapping_tree
;
1146 write_lock(&em_tree
->lock
);
1147 remove_extent_mapping(em_tree
, em
);
1148 write_unlock(&em_tree
->lock
);
1149 /* once for the tree */
1150 free_extent_map(em
);
1154 /* Once for the lookup reference */
1155 btrfs_put_block_group(block_group
);
1157 btrfs_delayed_refs_rsv_release(fs_info
, 1);
1158 btrfs_free_path(path
);
1162 struct btrfs_trans_handle
*btrfs_start_trans_remove_block_group(
1163 struct btrfs_fs_info
*fs_info
, const u64 chunk_offset
)
1165 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
;
1166 struct extent_map
*em
;
1167 struct map_lookup
*map
;
1168 unsigned int num_items
;
1170 read_lock(&em_tree
->lock
);
1171 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
1172 read_unlock(&em_tree
->lock
);
1173 ASSERT(em
&& em
->start
== chunk_offset
);
1176 * We need to reserve 3 + N units from the metadata space info in order
1177 * to remove a block group (done at btrfs_remove_chunk() and at
1178 * btrfs_remove_block_group()), which are used for:
1180 * 1 unit for adding the free space inode's orphan (located in the tree
1182 * 1 unit for deleting the block group item (located in the extent
1184 * 1 unit for deleting the free space item (located in tree of tree
1186 * N units for deleting N device extent items corresponding to each
1187 * stripe (located in the device tree).
1189 * In order to remove a block group we also need to reserve units in the
1190 * system space info in order to update the chunk tree (update one or
1191 * more device items and remove one chunk item), but this is done at
1192 * btrfs_remove_chunk() through a call to check_system_chunk().
1194 map
= em
->map_lookup
;
1195 num_items
= 3 + map
->num_stripes
;
1196 free_extent_map(em
);
1198 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
1203 * Mark block group @cache read-only, so later write won't happen to block
1206 * If @force is not set, this function will only mark the block group readonly
1207 * if we have enough free space (1M) in other metadata/system block groups.
1208 * If @force is not set, this function will mark the block group readonly
1209 * without checking free space.
1211 * NOTE: This function doesn't care if other block groups can contain all the
1212 * data in this block group. That check should be done by relocation routine,
1213 * not this function.
1215 static int inc_block_group_ro(struct btrfs_block_group
*cache
, int force
)
1217 struct btrfs_space_info
*sinfo
= cache
->space_info
;
1221 spin_lock(&sinfo
->lock
);
1222 spin_lock(&cache
->lock
);
1230 num_bytes
= cache
->length
- cache
->reserved
- cache
->pinned
-
1231 cache
->bytes_super
- cache
->used
;
1234 * Data never overcommits, even in mixed mode, so do just the straight
1235 * check of left over space in how much we have allocated.
1239 } else if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
) {
1240 u64 sinfo_used
= btrfs_space_info_used(sinfo
, true);
1243 * Here we make sure if we mark this bg RO, we still have enough
1244 * free space as buffer.
1246 if (sinfo_used
+ num_bytes
<= sinfo
->total_bytes
)
1250 * We overcommit metadata, so we need to do the
1251 * btrfs_can_overcommit check here, and we need to pass in
1252 * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of
1253 * leeway to allow us to mark this block group as read only.
1255 if (btrfs_can_overcommit(cache
->fs_info
, sinfo
, num_bytes
,
1256 BTRFS_RESERVE_NO_FLUSH
))
1261 sinfo
->bytes_readonly
+= num_bytes
;
1263 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
1266 spin_unlock(&cache
->lock
);
1267 spin_unlock(&sinfo
->lock
);
1268 if (ret
== -ENOSPC
&& btrfs_test_opt(cache
->fs_info
, ENOSPC_DEBUG
)) {
1269 btrfs_info(cache
->fs_info
,
1270 "unable to make block group %llu ro", cache
->start
);
1271 btrfs_dump_space_info(cache
->fs_info
, cache
->space_info
, 0, 0);
1276 static bool clean_pinned_extents(struct btrfs_trans_handle
*trans
,
1277 struct btrfs_block_group
*bg
)
1279 struct btrfs_fs_info
*fs_info
= bg
->fs_info
;
1280 struct btrfs_transaction
*prev_trans
= NULL
;
1281 const u64 start
= bg
->start
;
1282 const u64 end
= start
+ bg
->length
- 1;
1285 spin_lock(&fs_info
->trans_lock
);
1286 if (trans
->transaction
->list
.prev
!= &fs_info
->trans_list
) {
1287 prev_trans
= list_last_entry(&trans
->transaction
->list
,
1288 struct btrfs_transaction
, list
);
1289 refcount_inc(&prev_trans
->use_count
);
1291 spin_unlock(&fs_info
->trans_lock
);
1294 * Hold the unused_bg_unpin_mutex lock to avoid racing with
1295 * btrfs_finish_extent_commit(). If we are at transaction N, another
1296 * task might be running finish_extent_commit() for the previous
1297 * transaction N - 1, and have seen a range belonging to the block
1298 * group in pinned_extents before we were able to clear the whole block
1299 * group range from pinned_extents. This means that task can lookup for
1300 * the block group after we unpinned it from pinned_extents and removed
1301 * it, leading to a BUG_ON() at unpin_extent_range().
1303 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
1305 ret
= clear_extent_bits(&prev_trans
->pinned_extents
, start
, end
,
1311 ret
= clear_extent_bits(&trans
->transaction
->pinned_extents
, start
, end
,
1314 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
1316 btrfs_put_transaction(prev_trans
);
1322 * Process the unused_bgs list and remove any that don't have any allocated
1323 * space inside of them.
1325 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
1327 struct btrfs_block_group
*block_group
;
1328 struct btrfs_space_info
*space_info
;
1329 struct btrfs_trans_handle
*trans
;
1330 const bool async_trim_enabled
= btrfs_test_opt(fs_info
, DISCARD_ASYNC
);
1333 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1336 spin_lock(&fs_info
->unused_bgs_lock
);
1337 while (!list_empty(&fs_info
->unused_bgs
)) {
1340 block_group
= list_first_entry(&fs_info
->unused_bgs
,
1341 struct btrfs_block_group
,
1343 list_del_init(&block_group
->bg_list
);
1345 space_info
= block_group
->space_info
;
1347 if (ret
|| btrfs_mixed_space_info(space_info
)) {
1348 btrfs_put_block_group(block_group
);
1351 spin_unlock(&fs_info
->unused_bgs_lock
);
1353 btrfs_discard_cancel_work(&fs_info
->discard_ctl
, block_group
);
1355 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
1357 /* Don't want to race with allocators so take the groups_sem */
1358 down_write(&space_info
->groups_sem
);
1361 * Async discard moves the final block group discard to be prior
1362 * to the unused_bgs code path. Therefore, if it's not fully
1363 * trimmed, punt it back to the async discard lists.
1365 if (btrfs_test_opt(fs_info
, DISCARD_ASYNC
) &&
1366 !btrfs_is_free_space_trimmed(block_group
)) {
1367 trace_btrfs_skip_unused_block_group(block_group
);
1368 up_write(&space_info
->groups_sem
);
1369 /* Requeue if we failed because of async discard */
1370 btrfs_discard_queue_work(&fs_info
->discard_ctl
,
1375 spin_lock(&block_group
->lock
);
1376 if (block_group
->reserved
|| block_group
->pinned
||
1377 block_group
->used
|| block_group
->ro
||
1378 list_is_singular(&block_group
->list
)) {
1380 * We want to bail if we made new allocations or have
1381 * outstanding allocations in this block group. We do
1382 * the ro check in case balance is currently acting on
1385 trace_btrfs_skip_unused_block_group(block_group
);
1386 spin_unlock(&block_group
->lock
);
1387 up_write(&space_info
->groups_sem
);
1390 spin_unlock(&block_group
->lock
);
1392 /* We don't want to force the issue, only flip if it's ok. */
1393 ret
= inc_block_group_ro(block_group
, 0);
1394 up_write(&space_info
->groups_sem
);
1401 * Want to do this before we do anything else so we can recover
1402 * properly if we fail to join the transaction.
1404 trans
= btrfs_start_trans_remove_block_group(fs_info
,
1405 block_group
->start
);
1406 if (IS_ERR(trans
)) {
1407 btrfs_dec_block_group_ro(block_group
);
1408 ret
= PTR_ERR(trans
);
1413 * We could have pending pinned extents for this block group,
1414 * just delete them, we don't care about them anymore.
1416 if (!clean_pinned_extents(trans
, block_group
)) {
1417 btrfs_dec_block_group_ro(block_group
);
1422 * At this point, the block_group is read only and should fail
1423 * new allocations. However, btrfs_finish_extent_commit() can
1424 * cause this block_group to be placed back on the discard
1425 * lists because now the block_group isn't fully discarded.
1426 * Bail here and try again later after discarding everything.
1428 spin_lock(&fs_info
->discard_ctl
.lock
);
1429 if (!list_empty(&block_group
->discard_list
)) {
1430 spin_unlock(&fs_info
->discard_ctl
.lock
);
1431 btrfs_dec_block_group_ro(block_group
);
1432 btrfs_discard_queue_work(&fs_info
->discard_ctl
,
1436 spin_unlock(&fs_info
->discard_ctl
.lock
);
1438 /* Reset pinned so btrfs_put_block_group doesn't complain */
1439 spin_lock(&space_info
->lock
);
1440 spin_lock(&block_group
->lock
);
1442 btrfs_space_info_update_bytes_pinned(fs_info
, space_info
,
1443 -block_group
->pinned
);
1444 space_info
->bytes_readonly
+= block_group
->pinned
;
1445 percpu_counter_add_batch(&space_info
->total_bytes_pinned
,
1446 -block_group
->pinned
,
1447 BTRFS_TOTAL_BYTES_PINNED_BATCH
);
1448 block_group
->pinned
= 0;
1450 spin_unlock(&block_group
->lock
);
1451 spin_unlock(&space_info
->lock
);
1454 * The normal path here is an unused block group is passed here,
1455 * then trimming is handled in the transaction commit path.
1456 * Async discard interposes before this to do the trimming
1457 * before coming down the unused block group path as trimming
1458 * will no longer be done later in the transaction commit path.
1460 if (!async_trim_enabled
&& btrfs_test_opt(fs_info
, DISCARD_ASYNC
))
1463 /* DISCARD can flip during remount */
1464 trimming
= btrfs_test_opt(fs_info
, DISCARD_SYNC
);
1466 /* Implicit trim during transaction commit. */
1468 btrfs_freeze_block_group(block_group
);
1471 * Btrfs_remove_chunk will abort the transaction if things go
1474 ret
= btrfs_remove_chunk(trans
, block_group
->start
);
1478 btrfs_unfreeze_block_group(block_group
);
1483 * If we're not mounted with -odiscard, we can just forget
1484 * about this block group. Otherwise we'll need to wait
1485 * until transaction commit to do the actual discard.
1488 spin_lock(&fs_info
->unused_bgs_lock
);
1490 * A concurrent scrub might have added us to the list
1491 * fs_info->unused_bgs, so use a list_move operation
1492 * to add the block group to the deleted_bgs list.
1494 list_move(&block_group
->bg_list
,
1495 &trans
->transaction
->deleted_bgs
);
1496 spin_unlock(&fs_info
->unused_bgs_lock
);
1497 btrfs_get_block_group(block_group
);
1500 btrfs_end_transaction(trans
);
1502 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
1503 btrfs_put_block_group(block_group
);
1504 spin_lock(&fs_info
->unused_bgs_lock
);
1506 spin_unlock(&fs_info
->unused_bgs_lock
);
1510 btrfs_end_transaction(trans
);
1511 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
1512 btrfs_put_block_group(block_group
);
1513 btrfs_discard_punt_unused_bgs_list(fs_info
);
1516 void btrfs_mark_bg_unused(struct btrfs_block_group
*bg
)
1518 struct btrfs_fs_info
*fs_info
= bg
->fs_info
;
1520 spin_lock(&fs_info
->unused_bgs_lock
);
1521 if (list_empty(&bg
->bg_list
)) {
1522 btrfs_get_block_group(bg
);
1523 trace_btrfs_add_unused_block_group(bg
);
1524 list_add_tail(&bg
->bg_list
, &fs_info
->unused_bgs
);
1526 spin_unlock(&fs_info
->unused_bgs_lock
);
1529 static int read_bg_from_eb(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
1530 struct btrfs_path
*path
)
1532 struct extent_map_tree
*em_tree
;
1533 struct extent_map
*em
;
1534 struct btrfs_block_group_item bg
;
1535 struct extent_buffer
*leaf
;
1540 slot
= path
->slots
[0];
1541 leaf
= path
->nodes
[0];
1543 em_tree
= &fs_info
->mapping_tree
;
1544 read_lock(&em_tree
->lock
);
1545 em
= lookup_extent_mapping(em_tree
, key
->objectid
, key
->offset
);
1546 read_unlock(&em_tree
->lock
);
1549 "logical %llu len %llu found bg but no related chunk",
1550 key
->objectid
, key
->offset
);
1554 if (em
->start
!= key
->objectid
|| em
->len
!= key
->offset
) {
1556 "block group %llu len %llu mismatch with chunk %llu len %llu",
1557 key
->objectid
, key
->offset
, em
->start
, em
->len
);
1562 read_extent_buffer(leaf
, &bg
, btrfs_item_ptr_offset(leaf
, slot
),
1564 flags
= btrfs_stack_block_group_flags(&bg
) &
1565 BTRFS_BLOCK_GROUP_TYPE_MASK
;
1567 if (flags
!= (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
1569 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
1570 key
->objectid
, key
->offset
, flags
,
1571 (BTRFS_BLOCK_GROUP_TYPE_MASK
& em
->map_lookup
->type
));
1576 free_extent_map(em
);
1580 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
1581 struct btrfs_path
*path
,
1582 struct btrfs_key
*key
)
1584 struct btrfs_root
*root
= fs_info
->extent_root
;
1586 struct btrfs_key found_key
;
1587 struct extent_buffer
*leaf
;
1590 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
1595 slot
= path
->slots
[0];
1596 leaf
= path
->nodes
[0];
1597 if (slot
>= btrfs_header_nritems(leaf
)) {
1598 ret
= btrfs_next_leaf(root
, path
);
1605 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
1607 if (found_key
.objectid
>= key
->objectid
&&
1608 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1609 ret
= read_bg_from_eb(fs_info
, &found_key
, path
);
1619 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1621 u64 extra_flags
= chunk_to_extended(flags
) &
1622 BTRFS_EXTENDED_PROFILE_MASK
;
1624 write_seqlock(&fs_info
->profiles_lock
);
1625 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1626 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1627 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1628 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1629 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1630 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1631 write_sequnlock(&fs_info
->profiles_lock
);
1635 * btrfs_rmap_block - Map a physical disk address to a list of logical addresses
1636 * @chunk_start: logical address of block group
1637 * @physical: physical address to map to logical addresses
1638 * @logical: return array of logical addresses which map to @physical
1639 * @naddrs: length of @logical
1640 * @stripe_len: size of IO stripe for the given block group
1642 * Maps a particular @physical disk address to a list of @logical addresses.
1643 * Used primarily to exclude those portions of a block group that contain super
1647 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
, u64 chunk_start
,
1648 u64 physical
, u64
**logical
, int *naddrs
, int *stripe_len
)
1650 struct extent_map
*em
;
1651 struct map_lookup
*map
;
1654 u64 data_stripe_length
;
1659 em
= btrfs_get_chunk_map(fs_info
, chunk_start
, 1);
1663 map
= em
->map_lookup
;
1664 data_stripe_length
= em
->orig_block_len
;
1665 io_stripe_size
= map
->stripe_len
;
1667 /* For RAID5/6 adjust to a full IO stripe length */
1668 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
1669 io_stripe_size
= map
->stripe_len
* nr_data_stripes(map
);
1671 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
1677 for (i
= 0; i
< map
->num_stripes
; i
++) {
1678 bool already_inserted
= false;
1682 if (!in_range(physical
, map
->stripes
[i
].physical
,
1683 data_stripe_length
))
1686 stripe_nr
= physical
- map
->stripes
[i
].physical
;
1687 stripe_nr
= div64_u64(stripe_nr
, map
->stripe_len
);
1689 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
1690 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
1691 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
1692 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
1693 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
1696 * The remaining case would be for RAID56, multiply by
1697 * nr_data_stripes(). Alternatively, just use rmap_len below
1698 * instead of map->stripe_len
1701 bytenr
= chunk_start
+ stripe_nr
* io_stripe_size
;
1703 /* Ensure we don't add duplicate addresses */
1704 for (j
= 0; j
< nr
; j
++) {
1705 if (buf
[j
] == bytenr
) {
1706 already_inserted
= true;
1711 if (!already_inserted
)
1717 *stripe_len
= io_stripe_size
;
1719 free_extent_map(em
);
1723 static int exclude_super_stripes(struct btrfs_block_group
*cache
)
1725 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
1731 if (cache
->start
< BTRFS_SUPER_INFO_OFFSET
) {
1732 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->start
;
1733 cache
->bytes_super
+= stripe_len
;
1734 ret
= btrfs_add_excluded_extent(fs_info
, cache
->start
,
1740 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
1741 bytenr
= btrfs_sb_offset(i
);
1742 ret
= btrfs_rmap_block(fs_info
, cache
->start
,
1743 bytenr
, &logical
, &nr
, &stripe_len
);
1748 u64 len
= min_t(u64
, stripe_len
,
1749 cache
->start
+ cache
->length
- logical
[nr
]);
1751 cache
->bytes_super
+= len
;
1752 ret
= btrfs_add_excluded_extent(fs_info
, logical
[nr
],
1765 static void link_block_group(struct btrfs_block_group
*cache
)
1767 struct btrfs_space_info
*space_info
= cache
->space_info
;
1768 int index
= btrfs_bg_flags_to_raid_index(cache
->flags
);
1771 down_write(&space_info
->groups_sem
);
1772 if (list_empty(&space_info
->block_groups
[index
]))
1774 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
1775 up_write(&space_info
->groups_sem
);
1778 btrfs_sysfs_add_block_group_type(cache
);
1781 static struct btrfs_block_group
*btrfs_create_block_group_cache(
1782 struct btrfs_fs_info
*fs_info
, u64 start
)
1784 struct btrfs_block_group
*cache
;
1786 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
1790 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
1792 if (!cache
->free_space_ctl
) {
1797 cache
->start
= start
;
1799 cache
->fs_info
= fs_info
;
1800 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
, start
);
1802 cache
->discard_index
= BTRFS_DISCARD_INDEX_UNUSED
;
1804 refcount_set(&cache
->refs
, 1);
1805 spin_lock_init(&cache
->lock
);
1806 init_rwsem(&cache
->data_rwsem
);
1807 INIT_LIST_HEAD(&cache
->list
);
1808 INIT_LIST_HEAD(&cache
->cluster_list
);
1809 INIT_LIST_HEAD(&cache
->bg_list
);
1810 INIT_LIST_HEAD(&cache
->ro_list
);
1811 INIT_LIST_HEAD(&cache
->discard_list
);
1812 INIT_LIST_HEAD(&cache
->dirty_list
);
1813 INIT_LIST_HEAD(&cache
->io_list
);
1814 btrfs_init_free_space_ctl(cache
);
1815 atomic_set(&cache
->frozen
, 0);
1816 mutex_init(&cache
->free_space_lock
);
1817 btrfs_init_full_stripe_locks_tree(&cache
->full_stripe_locks_root
);
1823 * Iterate all chunks and verify that each of them has the corresponding block
1826 static int check_chunk_block_group_mappings(struct btrfs_fs_info
*fs_info
)
1828 struct extent_map_tree
*map_tree
= &fs_info
->mapping_tree
;
1829 struct extent_map
*em
;
1830 struct btrfs_block_group
*bg
;
1835 read_lock(&map_tree
->lock
);
1837 * lookup_extent_mapping will return the first extent map
1838 * intersecting the range, so setting @len to 1 is enough to
1839 * get the first chunk.
1841 em
= lookup_extent_mapping(map_tree
, start
, 1);
1842 read_unlock(&map_tree
->lock
);
1846 bg
= btrfs_lookup_block_group(fs_info
, em
->start
);
1849 "chunk start=%llu len=%llu doesn't have corresponding block group",
1850 em
->start
, em
->len
);
1852 free_extent_map(em
);
1855 if (bg
->start
!= em
->start
|| bg
->length
!= em
->len
||
1856 (bg
->flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
) !=
1857 (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
1859 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
1861 em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
,
1862 bg
->start
, bg
->length
,
1863 bg
->flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
);
1865 free_extent_map(em
);
1866 btrfs_put_block_group(bg
);
1869 start
= em
->start
+ em
->len
;
1870 free_extent_map(em
);
1871 btrfs_put_block_group(bg
);
1876 static int read_block_group_item(struct btrfs_block_group
*cache
,
1877 struct btrfs_path
*path
,
1878 const struct btrfs_key
*key
)
1880 struct extent_buffer
*leaf
= path
->nodes
[0];
1881 struct btrfs_block_group_item bgi
;
1882 int slot
= path
->slots
[0];
1884 cache
->length
= key
->offset
;
1886 read_extent_buffer(leaf
, &bgi
, btrfs_item_ptr_offset(leaf
, slot
),
1888 cache
->used
= btrfs_stack_block_group_used(&bgi
);
1889 cache
->flags
= btrfs_stack_block_group_flags(&bgi
);
1894 static int read_one_block_group(struct btrfs_fs_info
*info
,
1895 struct btrfs_path
*path
,
1896 const struct btrfs_key
*key
,
1899 struct btrfs_block_group
*cache
;
1900 struct btrfs_space_info
*space_info
;
1901 const bool mixed
= btrfs_fs_incompat(info
, MIXED_GROUPS
);
1904 ASSERT(key
->type
== BTRFS_BLOCK_GROUP_ITEM_KEY
);
1906 cache
= btrfs_create_block_group_cache(info
, key
->objectid
);
1910 ret
= read_block_group_item(cache
, path
, key
);
1914 set_free_space_tree_thresholds(cache
);
1918 * When we mount with old space cache, we need to
1919 * set BTRFS_DC_CLEAR and set dirty flag.
1921 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
1922 * truncate the old free space cache inode and
1924 * b) Setting 'dirty flag' makes sure that we flush
1925 * the new space cache info onto disk.
1927 if (btrfs_test_opt(info
, SPACE_CACHE
))
1928 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
1930 if (!mixed
&& ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
1931 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
1933 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
1940 * We need to exclude the super stripes now so that the space info has
1941 * super bytes accounted for, otherwise we'll think we have more space
1942 * than we actually do.
1944 ret
= exclude_super_stripes(cache
);
1946 /* We may have excluded something, so call this just in case. */
1947 btrfs_free_excluded_extents(cache
);
1952 * Check for two cases, either we are full, and therefore don't need
1953 * to bother with the caching work since we won't find any space, or we
1954 * are empty, and we can just add all the space in and be done with it.
1955 * This saves us _a_lot_ of time, particularly in the full case.
1957 if (cache
->length
== cache
->used
) {
1958 cache
->last_byte_to_unpin
= (u64
)-1;
1959 cache
->cached
= BTRFS_CACHE_FINISHED
;
1960 btrfs_free_excluded_extents(cache
);
1961 } else if (cache
->used
== 0) {
1962 cache
->last_byte_to_unpin
= (u64
)-1;
1963 cache
->cached
= BTRFS_CACHE_FINISHED
;
1964 add_new_free_space(cache
, cache
->start
,
1965 cache
->start
+ cache
->length
);
1966 btrfs_free_excluded_extents(cache
);
1969 ret
= btrfs_add_block_group_cache(info
, cache
);
1971 btrfs_remove_free_space_cache(cache
);
1974 trace_btrfs_add_block_group(info
, cache
, 0);
1975 btrfs_update_space_info(info
, cache
->flags
, cache
->length
,
1976 cache
->used
, cache
->bytes_super
, &space_info
);
1978 cache
->space_info
= space_info
;
1980 link_block_group(cache
);
1982 set_avail_alloc_bits(info
, cache
->flags
);
1983 if (btrfs_chunk_readonly(info
, cache
->start
)) {
1984 inc_block_group_ro(cache
, 1);
1985 } else if (cache
->used
== 0) {
1986 ASSERT(list_empty(&cache
->bg_list
));
1987 if (btrfs_test_opt(info
, DISCARD_ASYNC
))
1988 btrfs_discard_queue_work(&info
->discard_ctl
, cache
);
1990 btrfs_mark_bg_unused(cache
);
1994 btrfs_put_block_group(cache
);
1998 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
2000 struct btrfs_path
*path
;
2002 struct btrfs_block_group
*cache
;
2003 struct btrfs_space_info
*space_info
;
2004 struct btrfs_key key
;
2010 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
2011 path
= btrfs_alloc_path();
2015 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
2016 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
2017 btrfs_super_generation(info
->super_copy
) != cache_gen
)
2019 if (btrfs_test_opt(info
, CLEAR_CACHE
))
2023 ret
= find_first_block_group(info
, path
, &key
);
2029 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
2030 ret
= read_one_block_group(info
, path
, &key
, need_clear
);
2033 key
.objectid
+= key
.offset
;
2035 btrfs_release_path(path
);
2039 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
2040 if (!(btrfs_get_alloc_profile(info
, space_info
->flags
) &
2041 (BTRFS_BLOCK_GROUP_RAID10
|
2042 BTRFS_BLOCK_GROUP_RAID1_MASK
|
2043 BTRFS_BLOCK_GROUP_RAID56_MASK
|
2044 BTRFS_BLOCK_GROUP_DUP
)))
2047 * Avoid allocating from un-mirrored block group if there are
2048 * mirrored block groups.
2050 list_for_each_entry(cache
,
2051 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
2053 inc_block_group_ro(cache
, 1);
2054 list_for_each_entry(cache
,
2055 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
2057 inc_block_group_ro(cache
, 1);
2061 btrfs_init_global_block_rsv(info
);
2062 ret
= check_chunk_block_group_mappings(info
);
2064 btrfs_free_path(path
);
2068 static int insert_block_group_item(struct btrfs_trans_handle
*trans
,
2069 struct btrfs_block_group
*block_group
)
2071 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2072 struct btrfs_block_group_item bgi
;
2073 struct btrfs_root
*root
;
2074 struct btrfs_key key
;
2076 spin_lock(&block_group
->lock
);
2077 btrfs_set_stack_block_group_used(&bgi
, block_group
->used
);
2078 btrfs_set_stack_block_group_chunk_objectid(&bgi
,
2079 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
2080 btrfs_set_stack_block_group_flags(&bgi
, block_group
->flags
);
2081 key
.objectid
= block_group
->start
;
2082 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
2083 key
.offset
= block_group
->length
;
2084 spin_unlock(&block_group
->lock
);
2086 root
= fs_info
->extent_root
;
2087 return btrfs_insert_item(trans
, root
, &key
, &bgi
, sizeof(bgi
));
2090 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
)
2092 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2093 struct btrfs_block_group
*block_group
;
2096 if (!trans
->can_flush_pending_bgs
)
2099 while (!list_empty(&trans
->new_bgs
)) {
2100 block_group
= list_first_entry(&trans
->new_bgs
,
2101 struct btrfs_block_group
,
2106 ret
= insert_block_group_item(trans
, block_group
);
2108 btrfs_abort_transaction(trans
, ret
);
2109 ret
= btrfs_finish_chunk_alloc(trans
, block_group
->start
,
2110 block_group
->length
);
2112 btrfs_abort_transaction(trans
, ret
);
2113 add_block_group_free_space(trans
, block_group
);
2114 /* Already aborted the transaction if it failed. */
2116 btrfs_delayed_refs_rsv_release(fs_info
, 1);
2117 list_del_init(&block_group
->bg_list
);
2119 btrfs_trans_release_chunk_metadata(trans
);
2122 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
, u64 bytes_used
,
2123 u64 type
, u64 chunk_offset
, u64 size
)
2125 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2126 struct btrfs_block_group
*cache
;
2129 btrfs_set_log_full_commit(trans
);
2131 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
);
2135 cache
->length
= size
;
2136 set_free_space_tree_thresholds(cache
);
2137 cache
->used
= bytes_used
;
2138 cache
->flags
= type
;
2139 cache
->last_byte_to_unpin
= (u64
)-1;
2140 cache
->cached
= BTRFS_CACHE_FINISHED
;
2141 cache
->needs_free_space
= 1;
2142 ret
= exclude_super_stripes(cache
);
2144 /* We may have excluded something, so call this just in case */
2145 btrfs_free_excluded_extents(cache
);
2146 btrfs_put_block_group(cache
);
2150 add_new_free_space(cache
, chunk_offset
, chunk_offset
+ size
);
2152 btrfs_free_excluded_extents(cache
);
2154 #ifdef CONFIG_BTRFS_DEBUG
2155 if (btrfs_should_fragment_free_space(cache
)) {
2156 u64 new_bytes_used
= size
- bytes_used
;
2158 bytes_used
+= new_bytes_used
>> 1;
2159 fragment_free_space(cache
);
2163 * Ensure the corresponding space_info object is created and
2164 * assigned to our block group. We want our bg to be added to the rbtree
2165 * with its ->space_info set.
2167 cache
->space_info
= btrfs_find_space_info(fs_info
, cache
->flags
);
2168 ASSERT(cache
->space_info
);
2170 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
2172 btrfs_remove_free_space_cache(cache
);
2173 btrfs_put_block_group(cache
);
2178 * Now that our block group has its ->space_info set and is inserted in
2179 * the rbtree, update the space info's counters.
2181 trace_btrfs_add_block_group(fs_info
, cache
, 1);
2182 btrfs_update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
2183 cache
->bytes_super
, &cache
->space_info
);
2184 btrfs_update_global_block_rsv(fs_info
);
2186 link_block_group(cache
);
2188 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
2189 trans
->delayed_ref_updates
++;
2190 btrfs_update_delayed_refs_rsv(trans
);
2192 set_avail_alloc_bits(fs_info
, type
);
2197 * Mark one block group RO, can be called several times for the same block
2200 * @cache: the destination block group
2201 * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to
2202 * ensure we still have some free space after marking this
2205 int btrfs_inc_block_group_ro(struct btrfs_block_group
*cache
,
2206 bool do_chunk_alloc
)
2208 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
2209 struct btrfs_trans_handle
*trans
;
2214 trans
= btrfs_join_transaction(fs_info
->extent_root
);
2216 return PTR_ERR(trans
);
2219 * we're not allowed to set block groups readonly after the dirty
2220 * block groups cache has started writing. If it already started,
2221 * back off and let this transaction commit
2223 mutex_lock(&fs_info
->ro_block_group_mutex
);
2224 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
2225 u64 transid
= trans
->transid
;
2227 mutex_unlock(&fs_info
->ro_block_group_mutex
);
2228 btrfs_end_transaction(trans
);
2230 ret
= btrfs_wait_for_commit(fs_info
, transid
);
2236 if (do_chunk_alloc
) {
2238 * If we are changing raid levels, try to allocate a
2239 * corresponding block group with the new raid level.
2241 alloc_flags
= btrfs_get_alloc_profile(fs_info
, cache
->flags
);
2242 if (alloc_flags
!= cache
->flags
) {
2243 ret
= btrfs_chunk_alloc(trans
, alloc_flags
,
2246 * ENOSPC is allowed here, we may have enough space
2247 * already allocated at the new raid level to carry on
2256 ret
= inc_block_group_ro(cache
, 0);
2257 if (!do_chunk_alloc
)
2261 alloc_flags
= btrfs_get_alloc_profile(fs_info
, cache
->space_info
->flags
);
2262 ret
= btrfs_chunk_alloc(trans
, alloc_flags
, CHUNK_ALLOC_FORCE
);
2265 ret
= inc_block_group_ro(cache
, 0);
2267 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2268 alloc_flags
= btrfs_get_alloc_profile(fs_info
, cache
->flags
);
2269 mutex_lock(&fs_info
->chunk_mutex
);
2270 check_system_chunk(trans
, alloc_flags
);
2271 mutex_unlock(&fs_info
->chunk_mutex
);
2274 mutex_unlock(&fs_info
->ro_block_group_mutex
);
2276 btrfs_end_transaction(trans
);
2280 void btrfs_dec_block_group_ro(struct btrfs_block_group
*cache
)
2282 struct btrfs_space_info
*sinfo
= cache
->space_info
;
2287 spin_lock(&sinfo
->lock
);
2288 spin_lock(&cache
->lock
);
2290 num_bytes
= cache
->length
- cache
->reserved
-
2291 cache
->pinned
- cache
->bytes_super
- cache
->used
;
2292 sinfo
->bytes_readonly
-= num_bytes
;
2293 list_del_init(&cache
->ro_list
);
2295 spin_unlock(&cache
->lock
);
2296 spin_unlock(&sinfo
->lock
);
2299 static int update_block_group_item(struct btrfs_trans_handle
*trans
,
2300 struct btrfs_path
*path
,
2301 struct btrfs_block_group
*cache
)
2303 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2305 struct btrfs_root
*root
= fs_info
->extent_root
;
2307 struct extent_buffer
*leaf
;
2308 struct btrfs_block_group_item bgi
;
2309 struct btrfs_key key
;
2311 key
.objectid
= cache
->start
;
2312 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
2313 key
.offset
= cache
->length
;
2315 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2322 leaf
= path
->nodes
[0];
2323 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2324 btrfs_set_stack_block_group_used(&bgi
, cache
->used
);
2325 btrfs_set_stack_block_group_chunk_objectid(&bgi
,
2326 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
2327 btrfs_set_stack_block_group_flags(&bgi
, cache
->flags
);
2328 write_extent_buffer(leaf
, &bgi
, bi
, sizeof(bgi
));
2329 btrfs_mark_buffer_dirty(leaf
);
2331 btrfs_release_path(path
);
2336 static int cache_save_setup(struct btrfs_block_group
*block_group
,
2337 struct btrfs_trans_handle
*trans
,
2338 struct btrfs_path
*path
)
2340 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
2341 struct btrfs_root
*root
= fs_info
->tree_root
;
2342 struct inode
*inode
= NULL
;
2343 struct extent_changeset
*data_reserved
= NULL
;
2345 int dcs
= BTRFS_DC_ERROR
;
2351 * If this block group is smaller than 100 megs don't bother caching the
2354 if (block_group
->length
< (100 * SZ_1M
)) {
2355 spin_lock(&block_group
->lock
);
2356 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2357 spin_unlock(&block_group
->lock
);
2361 if (TRANS_ABORTED(trans
))
2364 inode
= lookup_free_space_inode(block_group
, path
);
2365 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2366 ret
= PTR_ERR(inode
);
2367 btrfs_release_path(path
);
2371 if (IS_ERR(inode
)) {
2375 if (block_group
->ro
)
2378 ret
= create_free_space_inode(trans
, block_group
, path
);
2385 * We want to set the generation to 0, that way if anything goes wrong
2386 * from here on out we know not to trust this cache when we load up next
2389 BTRFS_I(inode
)->generation
= 0;
2390 ret
= btrfs_update_inode(trans
, root
, inode
);
2393 * So theoretically we could recover from this, simply set the
2394 * super cache generation to 0 so we know to invalidate the
2395 * cache, but then we'd have to keep track of the block groups
2396 * that fail this way so we know we _have_ to reset this cache
2397 * before the next commit or risk reading stale cache. So to
2398 * limit our exposure to horrible edge cases lets just abort the
2399 * transaction, this only happens in really bad situations
2402 btrfs_abort_transaction(trans
, ret
);
2407 /* We've already setup this transaction, go ahead and exit */
2408 if (block_group
->cache_generation
== trans
->transid
&&
2409 i_size_read(inode
)) {
2410 dcs
= BTRFS_DC_SETUP
;
2414 if (i_size_read(inode
) > 0) {
2415 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
2416 &fs_info
->global_block_rsv
);
2420 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
2425 spin_lock(&block_group
->lock
);
2426 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
2427 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
2429 * don't bother trying to write stuff out _if_
2430 * a) we're not cached,
2431 * b) we're with nospace_cache mount option,
2432 * c) we're with v2 space_cache (FREE_SPACE_TREE).
2434 dcs
= BTRFS_DC_WRITTEN
;
2435 spin_unlock(&block_group
->lock
);
2438 spin_unlock(&block_group
->lock
);
2441 * We hit an ENOSPC when setting up the cache in this transaction, just
2442 * skip doing the setup, we've already cleared the cache so we're safe.
2444 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
2450 * Try to preallocate enough space based on how big the block group is.
2451 * Keep in mind this has to include any pinned space which could end up
2452 * taking up quite a bit since it's not folded into the other space
2455 num_pages
= div_u64(block_group
->length
, SZ_256M
);
2460 num_pages
*= PAGE_SIZE
;
2462 ret
= btrfs_check_data_free_space(BTRFS_I(inode
), &data_reserved
, 0,
2467 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2468 num_pages
, num_pages
,
2471 * Our cache requires contiguous chunks so that we don't modify a bunch
2472 * of metadata or split extents when writing the cache out, which means
2473 * we can enospc if we are heavily fragmented in addition to just normal
2474 * out of space conditions. So if we hit this just skip setting up any
2475 * other block groups for this transaction, maybe we'll unpin enough
2476 * space the next time around.
2479 dcs
= BTRFS_DC_SETUP
;
2480 else if (ret
== -ENOSPC
)
2481 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
2486 btrfs_release_path(path
);
2488 spin_lock(&block_group
->lock
);
2489 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
2490 block_group
->cache_generation
= trans
->transid
;
2491 block_group
->disk_cache_state
= dcs
;
2492 spin_unlock(&block_group
->lock
);
2494 extent_changeset_free(data_reserved
);
2498 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
)
2500 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2501 struct btrfs_block_group
*cache
, *tmp
;
2502 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
2503 struct btrfs_path
*path
;
2505 if (list_empty(&cur_trans
->dirty_bgs
) ||
2506 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
2509 path
= btrfs_alloc_path();
2513 /* Could add new block groups, use _safe just in case */
2514 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
2516 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2517 cache_save_setup(cache
, trans
, path
);
2520 btrfs_free_path(path
);
2525 * Transaction commit does final block group cache writeback during a critical
2526 * section where nothing is allowed to change the FS. This is required in
2527 * order for the cache to actually match the block group, but can introduce a
2528 * lot of latency into the commit.
2530 * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO.
2531 * There's a chance we'll have to redo some of it if the block group changes
2532 * again during the commit, but it greatly reduces the commit latency by
2533 * getting rid of the easy block groups while we're still allowing others to
2536 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
)
2538 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2539 struct btrfs_block_group
*cache
;
2540 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
2543 struct btrfs_path
*path
= NULL
;
2545 struct list_head
*io
= &cur_trans
->io_bgs
;
2546 int num_started
= 0;
2549 spin_lock(&cur_trans
->dirty_bgs_lock
);
2550 if (list_empty(&cur_trans
->dirty_bgs
)) {
2551 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2554 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
2555 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2558 /* Make sure all the block groups on our dirty list actually exist */
2559 btrfs_create_pending_block_groups(trans
);
2562 path
= btrfs_alloc_path();
2568 * cache_write_mutex is here only to save us from balance or automatic
2569 * removal of empty block groups deleting this block group while we are
2570 * writing out the cache
2572 mutex_lock(&trans
->transaction
->cache_write_mutex
);
2573 while (!list_empty(&dirty
)) {
2574 bool drop_reserve
= true;
2576 cache
= list_first_entry(&dirty
, struct btrfs_block_group
,
2579 * This can happen if something re-dirties a block group that
2580 * is already under IO. Just wait for it to finish and then do
2583 if (!list_empty(&cache
->io_list
)) {
2584 list_del_init(&cache
->io_list
);
2585 btrfs_wait_cache_io(trans
, cache
, path
);
2586 btrfs_put_block_group(cache
);
2591 * btrfs_wait_cache_io uses the cache->dirty_list to decide if
2592 * it should update the cache_state. Don't delete until after
2595 * Since we're not running in the commit critical section
2596 * we need the dirty_bgs_lock to protect from update_block_group
2598 spin_lock(&cur_trans
->dirty_bgs_lock
);
2599 list_del_init(&cache
->dirty_list
);
2600 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2604 cache_save_setup(cache
, trans
, path
);
2606 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
2607 cache
->io_ctl
.inode
= NULL
;
2608 ret
= btrfs_write_out_cache(trans
, cache
, path
);
2609 if (ret
== 0 && cache
->io_ctl
.inode
) {
2614 * The cache_write_mutex is protecting the
2615 * io_list, also refer to the definition of
2616 * btrfs_transaction::io_bgs for more details
2618 list_add_tail(&cache
->io_list
, io
);
2621 * If we failed to write the cache, the
2622 * generation will be bad and life goes on
2628 ret
= update_block_group_item(trans
, path
, cache
);
2630 * Our block group might still be attached to the list
2631 * of new block groups in the transaction handle of some
2632 * other task (struct btrfs_trans_handle->new_bgs). This
2633 * means its block group item isn't yet in the extent
2634 * tree. If this happens ignore the error, as we will
2635 * try again later in the critical section of the
2636 * transaction commit.
2638 if (ret
== -ENOENT
) {
2640 spin_lock(&cur_trans
->dirty_bgs_lock
);
2641 if (list_empty(&cache
->dirty_list
)) {
2642 list_add_tail(&cache
->dirty_list
,
2643 &cur_trans
->dirty_bgs
);
2644 btrfs_get_block_group(cache
);
2645 drop_reserve
= false;
2647 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2649 btrfs_abort_transaction(trans
, ret
);
2653 /* If it's not on the io list, we need to put the block group */
2655 btrfs_put_block_group(cache
);
2657 btrfs_delayed_refs_rsv_release(fs_info
, 1);
2663 * Avoid blocking other tasks for too long. It might even save
2664 * us from writing caches for block groups that are going to be
2667 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
2668 mutex_lock(&trans
->transaction
->cache_write_mutex
);
2670 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
2673 * Go through delayed refs for all the stuff we've just kicked off
2674 * and then loop back (just once)
2676 ret
= btrfs_run_delayed_refs(trans
, 0);
2677 if (!ret
&& loops
== 0) {
2679 spin_lock(&cur_trans
->dirty_bgs_lock
);
2680 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
2682 * dirty_bgs_lock protects us from concurrent block group
2683 * deletes too (not just cache_write_mutex).
2685 if (!list_empty(&dirty
)) {
2686 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2689 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2690 } else if (ret
< 0) {
2691 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
2694 btrfs_free_path(path
);
2698 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
)
2700 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2701 struct btrfs_block_group
*cache
;
2702 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
2705 struct btrfs_path
*path
;
2706 struct list_head
*io
= &cur_trans
->io_bgs
;
2707 int num_started
= 0;
2709 path
= btrfs_alloc_path();
2714 * Even though we are in the critical section of the transaction commit,
2715 * we can still have concurrent tasks adding elements to this
2716 * transaction's list of dirty block groups. These tasks correspond to
2717 * endio free space workers started when writeback finishes for a
2718 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
2719 * allocate new block groups as a result of COWing nodes of the root
2720 * tree when updating the free space inode. The writeback for the space
2721 * caches is triggered by an earlier call to
2722 * btrfs_start_dirty_block_groups() and iterations of the following
2724 * Also we want to do the cache_save_setup first and then run the
2725 * delayed refs to make sure we have the best chance at doing this all
2728 spin_lock(&cur_trans
->dirty_bgs_lock
);
2729 while (!list_empty(&cur_trans
->dirty_bgs
)) {
2730 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
2731 struct btrfs_block_group
,
2735 * This can happen if cache_save_setup re-dirties a block group
2736 * that is already under IO. Just wait for it to finish and
2737 * then do it all again
2739 if (!list_empty(&cache
->io_list
)) {
2740 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2741 list_del_init(&cache
->io_list
);
2742 btrfs_wait_cache_io(trans
, cache
, path
);
2743 btrfs_put_block_group(cache
);
2744 spin_lock(&cur_trans
->dirty_bgs_lock
);
2748 * Don't remove from the dirty list until after we've waited on
2751 list_del_init(&cache
->dirty_list
);
2752 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2755 cache_save_setup(cache
, trans
, path
);
2758 ret
= btrfs_run_delayed_refs(trans
,
2759 (unsigned long) -1);
2761 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
2762 cache
->io_ctl
.inode
= NULL
;
2763 ret
= btrfs_write_out_cache(trans
, cache
, path
);
2764 if (ret
== 0 && cache
->io_ctl
.inode
) {
2767 list_add_tail(&cache
->io_list
, io
);
2770 * If we failed to write the cache, the
2771 * generation will be bad and life goes on
2777 ret
= update_block_group_item(trans
, path
, cache
);
2779 * One of the free space endio workers might have
2780 * created a new block group while updating a free space
2781 * cache's inode (at inode.c:btrfs_finish_ordered_io())
2782 * and hasn't released its transaction handle yet, in
2783 * which case the new block group is still attached to
2784 * its transaction handle and its creation has not
2785 * finished yet (no block group item in the extent tree
2786 * yet, etc). If this is the case, wait for all free
2787 * space endio workers to finish and retry. This is a
2788 * a very rare case so no need for a more efficient and
2791 if (ret
== -ENOENT
) {
2792 wait_event(cur_trans
->writer_wait
,
2793 atomic_read(&cur_trans
->num_writers
) == 1);
2794 ret
= update_block_group_item(trans
, path
, cache
);
2797 btrfs_abort_transaction(trans
, ret
);
2800 /* If its not on the io list, we need to put the block group */
2802 btrfs_put_block_group(cache
);
2803 btrfs_delayed_refs_rsv_release(fs_info
, 1);
2804 spin_lock(&cur_trans
->dirty_bgs_lock
);
2806 spin_unlock(&cur_trans
->dirty_bgs_lock
);
2809 * Refer to the definition of io_bgs member for details why it's safe
2810 * to use it without any locking
2812 while (!list_empty(io
)) {
2813 cache
= list_first_entry(io
, struct btrfs_block_group
,
2815 list_del_init(&cache
->io_list
);
2816 btrfs_wait_cache_io(trans
, cache
, path
);
2817 btrfs_put_block_group(cache
);
2820 btrfs_free_path(path
);
2824 int btrfs_update_block_group(struct btrfs_trans_handle
*trans
,
2825 u64 bytenr
, u64 num_bytes
, int alloc
)
2827 struct btrfs_fs_info
*info
= trans
->fs_info
;
2828 struct btrfs_block_group
*cache
= NULL
;
2829 u64 total
= num_bytes
;
2835 /* Block accounting for super block */
2836 spin_lock(&info
->delalloc_root_lock
);
2837 old_val
= btrfs_super_bytes_used(info
->super_copy
);
2839 old_val
+= num_bytes
;
2841 old_val
-= num_bytes
;
2842 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
2843 spin_unlock(&info
->delalloc_root_lock
);
2846 cache
= btrfs_lookup_block_group(info
, bytenr
);
2851 factor
= btrfs_bg_type_to_factor(cache
->flags
);
2854 * If this block group has free space cache written out, we
2855 * need to make sure to load it if we are removing space. This
2856 * is because we need the unpinning stage to actually add the
2857 * space back to the block group, otherwise we will leak space.
2859 if (!alloc
&& !btrfs_block_group_done(cache
))
2860 btrfs_cache_block_group(cache
, 1);
2862 byte_in_group
= bytenr
- cache
->start
;
2863 WARN_ON(byte_in_group
> cache
->length
);
2865 spin_lock(&cache
->space_info
->lock
);
2866 spin_lock(&cache
->lock
);
2868 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
2869 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
2870 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
2872 old_val
= cache
->used
;
2873 num_bytes
= min(total
, cache
->length
- byte_in_group
);
2875 old_val
+= num_bytes
;
2876 cache
->used
= old_val
;
2877 cache
->reserved
-= num_bytes
;
2878 cache
->space_info
->bytes_reserved
-= num_bytes
;
2879 cache
->space_info
->bytes_used
+= num_bytes
;
2880 cache
->space_info
->disk_used
+= num_bytes
* factor
;
2881 spin_unlock(&cache
->lock
);
2882 spin_unlock(&cache
->space_info
->lock
);
2884 old_val
-= num_bytes
;
2885 cache
->used
= old_val
;
2886 cache
->pinned
+= num_bytes
;
2887 btrfs_space_info_update_bytes_pinned(info
,
2888 cache
->space_info
, num_bytes
);
2889 cache
->space_info
->bytes_used
-= num_bytes
;
2890 cache
->space_info
->disk_used
-= num_bytes
* factor
;
2891 spin_unlock(&cache
->lock
);
2892 spin_unlock(&cache
->space_info
->lock
);
2894 percpu_counter_add_batch(
2895 &cache
->space_info
->total_bytes_pinned
,
2897 BTRFS_TOTAL_BYTES_PINNED_BATCH
);
2898 set_extent_dirty(&trans
->transaction
->pinned_extents
,
2899 bytenr
, bytenr
+ num_bytes
- 1,
2900 GFP_NOFS
| __GFP_NOFAIL
);
2903 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
2904 if (list_empty(&cache
->dirty_list
)) {
2905 list_add_tail(&cache
->dirty_list
,
2906 &trans
->transaction
->dirty_bgs
);
2907 trans
->delayed_ref_updates
++;
2908 btrfs_get_block_group(cache
);
2910 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
2913 * No longer have used bytes in this block group, queue it for
2914 * deletion. We do this after adding the block group to the
2915 * dirty list to avoid races between cleaner kthread and space
2918 if (!alloc
&& old_val
== 0) {
2919 if (!btrfs_test_opt(info
, DISCARD_ASYNC
))
2920 btrfs_mark_bg_unused(cache
);
2923 btrfs_put_block_group(cache
);
2925 bytenr
+= num_bytes
;
2928 /* Modified block groups are accounted for in the delayed_refs_rsv. */
2929 btrfs_update_delayed_refs_rsv(trans
);
2934 * btrfs_add_reserved_bytes - update the block_group and space info counters
2935 * @cache: The cache we are manipulating
2936 * @ram_bytes: The number of bytes of file content, and will be same to
2937 * @num_bytes except for the compress path.
2938 * @num_bytes: The number of bytes in question
2939 * @delalloc: The blocks are allocated for the delalloc write
2941 * This is called by the allocator when it reserves space. If this is a
2942 * reservation and the block group has become read only we cannot make the
2943 * reservation and return -EAGAIN, otherwise this function always succeeds.
2945 int btrfs_add_reserved_bytes(struct btrfs_block_group
*cache
,
2946 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
2948 struct btrfs_space_info
*space_info
= cache
->space_info
;
2951 spin_lock(&space_info
->lock
);
2952 spin_lock(&cache
->lock
);
2956 cache
->reserved
+= num_bytes
;
2957 space_info
->bytes_reserved
+= num_bytes
;
2958 trace_btrfs_space_reservation(cache
->fs_info
, "space_info",
2959 space_info
->flags
, num_bytes
, 1);
2960 btrfs_space_info_update_bytes_may_use(cache
->fs_info
,
2961 space_info
, -ram_bytes
);
2963 cache
->delalloc_bytes
+= num_bytes
;
2965 spin_unlock(&cache
->lock
);
2966 spin_unlock(&space_info
->lock
);
2971 * btrfs_free_reserved_bytes - update the block_group and space info counters
2972 * @cache: The cache we are manipulating
2973 * @num_bytes: The number of bytes in question
2974 * @delalloc: The blocks are allocated for the delalloc write
2976 * This is called by somebody who is freeing space that was never actually used
2977 * on disk. For example if you reserve some space for a new leaf in transaction
2978 * A and before transaction A commits you free that leaf, you call this with
2979 * reserve set to 0 in order to clear the reservation.
2981 void btrfs_free_reserved_bytes(struct btrfs_block_group
*cache
,
2982 u64 num_bytes
, int delalloc
)
2984 struct btrfs_space_info
*space_info
= cache
->space_info
;
2986 spin_lock(&space_info
->lock
);
2987 spin_lock(&cache
->lock
);
2989 space_info
->bytes_readonly
+= num_bytes
;
2990 cache
->reserved
-= num_bytes
;
2991 space_info
->bytes_reserved
-= num_bytes
;
2992 space_info
->max_extent_size
= 0;
2995 cache
->delalloc_bytes
-= num_bytes
;
2996 spin_unlock(&cache
->lock
);
2997 spin_unlock(&space_info
->lock
);
3000 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3002 struct list_head
*head
= &info
->space_info
;
3003 struct btrfs_space_info
*found
;
3006 list_for_each_entry_rcu(found
, head
, list
) {
3007 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3008 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3013 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
3014 struct btrfs_space_info
*sinfo
, int force
)
3016 u64 bytes_used
= btrfs_space_info_used(sinfo
, false);
3019 if (force
== CHUNK_ALLOC_FORCE
)
3023 * in limited mode, we want to have some free space up to
3024 * about 1% of the FS size.
3026 if (force
== CHUNK_ALLOC_LIMITED
) {
3027 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
3028 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
3030 if (sinfo
->total_bytes
- bytes_used
< thresh
)
3034 if (bytes_used
+ SZ_2M
< div_factor(sinfo
->total_bytes
, 8))
3039 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 type
)
3041 u64 alloc_flags
= btrfs_get_alloc_profile(trans
->fs_info
, type
);
3043 return btrfs_chunk_alloc(trans
, alloc_flags
, CHUNK_ALLOC_FORCE
);
3047 * If force is CHUNK_ALLOC_FORCE:
3048 * - return 1 if it successfully allocates a chunk,
3049 * - return errors including -ENOSPC otherwise.
3050 * If force is NOT CHUNK_ALLOC_FORCE:
3051 * - return 0 if it doesn't need to allocate a new chunk,
3052 * - return 1 if it successfully allocates a chunk,
3053 * - return errors including -ENOSPC otherwise.
3055 int btrfs_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 flags
,
3056 enum btrfs_chunk_alloc_enum force
)
3058 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3059 struct btrfs_space_info
*space_info
;
3060 bool wait_for_alloc
= false;
3061 bool should_alloc
= false;
3064 /* Don't re-enter if we're already allocating a chunk */
3065 if (trans
->allocating_chunk
)
3068 space_info
= btrfs_find_space_info(fs_info
, flags
);
3072 spin_lock(&space_info
->lock
);
3073 if (force
< space_info
->force_alloc
)
3074 force
= space_info
->force_alloc
;
3075 should_alloc
= should_alloc_chunk(fs_info
, space_info
, force
);
3076 if (space_info
->full
) {
3077 /* No more free physical space */
3082 spin_unlock(&space_info
->lock
);
3084 } else if (!should_alloc
) {
3085 spin_unlock(&space_info
->lock
);
3087 } else if (space_info
->chunk_alloc
) {
3089 * Someone is already allocating, so we need to block
3090 * until this someone is finished and then loop to
3091 * recheck if we should continue with our allocation
3094 wait_for_alloc
= true;
3095 spin_unlock(&space_info
->lock
);
3096 mutex_lock(&fs_info
->chunk_mutex
);
3097 mutex_unlock(&fs_info
->chunk_mutex
);
3099 /* Proceed with allocation */
3100 space_info
->chunk_alloc
= 1;
3101 wait_for_alloc
= false;
3102 spin_unlock(&space_info
->lock
);
3106 } while (wait_for_alloc
);
3108 mutex_lock(&fs_info
->chunk_mutex
);
3109 trans
->allocating_chunk
= true;
3112 * If we have mixed data/metadata chunks we want to make sure we keep
3113 * allocating mixed chunks instead of individual chunks.
3115 if (btrfs_mixed_space_info(space_info
))
3116 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3119 * if we're doing a data chunk, go ahead and make sure that
3120 * we keep a reasonable number of metadata chunks allocated in the
3123 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3124 fs_info
->data_chunk_allocations
++;
3125 if (!(fs_info
->data_chunk_allocations
%
3126 fs_info
->metadata_ratio
))
3127 force_metadata_allocation(fs_info
);
3131 * Check if we have enough space in SYSTEM chunk because we may need
3132 * to update devices.
3134 check_system_chunk(trans
, flags
);
3136 ret
= btrfs_alloc_chunk(trans
, flags
);
3137 trans
->allocating_chunk
= false;
3139 spin_lock(&space_info
->lock
);
3142 space_info
->full
= 1;
3147 space_info
->max_extent_size
= 0;
3150 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3152 space_info
->chunk_alloc
= 0;
3153 spin_unlock(&space_info
->lock
);
3154 mutex_unlock(&fs_info
->chunk_mutex
);
3156 * When we allocate a new chunk we reserve space in the chunk block
3157 * reserve to make sure we can COW nodes/leafs in the chunk tree or
3158 * add new nodes/leafs to it if we end up needing to do it when
3159 * inserting the chunk item and updating device items as part of the
3160 * second phase of chunk allocation, performed by
3161 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
3162 * large number of new block groups to create in our transaction
3163 * handle's new_bgs list to avoid exhausting the chunk block reserve
3164 * in extreme cases - like having a single transaction create many new
3165 * block groups when starting to write out the free space caches of all
3166 * the block groups that were made dirty during the lifetime of the
3169 if (trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
)
3170 btrfs_create_pending_block_groups(trans
);
3175 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
3179 num_dev
= btrfs_raid_array
[btrfs_bg_flags_to_raid_index(type
)].devs_max
;
3181 num_dev
= fs_info
->fs_devices
->rw_devices
;
3187 * Reserve space in the system space for allocating or removing a chunk
3189 void check_system_chunk(struct btrfs_trans_handle
*trans
, u64 type
)
3191 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3192 struct btrfs_space_info
*info
;
3199 * Needed because we can end up allocating a system chunk and for an
3200 * atomic and race free space reservation in the chunk block reserve.
3202 lockdep_assert_held(&fs_info
->chunk_mutex
);
3204 info
= btrfs_find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3205 spin_lock(&info
->lock
);
3206 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
3207 spin_unlock(&info
->lock
);
3209 num_devs
= get_profile_num_devs(fs_info
, type
);
3211 /* num_devs device items to update and 1 chunk item to add or remove */
3212 thresh
= btrfs_calc_metadata_size(fs_info
, num_devs
) +
3213 btrfs_calc_insert_metadata_size(fs_info
, 1);
3215 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
3216 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
3217 left
, thresh
, type
);
3218 btrfs_dump_space_info(fs_info
, info
, 0, 0);
3221 if (left
< thresh
) {
3222 u64 flags
= btrfs_system_alloc_profile(fs_info
);
3225 * Ignore failure to create system chunk. We might end up not
3226 * needing it, as we might not need to COW all nodes/leafs from
3227 * the paths we visit in the chunk tree (they were already COWed
3228 * or created in the current transaction for example).
3230 ret
= btrfs_alloc_chunk(trans
, flags
);
3234 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
3235 &fs_info
->chunk_block_rsv
,
3236 thresh
, BTRFS_RESERVE_NO_FLUSH
);
3238 trans
->chunk_bytes_reserved
+= thresh
;
3242 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
3244 struct btrfs_block_group
*block_group
;
3248 struct inode
*inode
;
3250 block_group
= btrfs_lookup_first_block_group(info
, last
);
3251 while (block_group
) {
3252 btrfs_wait_block_group_cache_done(block_group
);
3253 spin_lock(&block_group
->lock
);
3254 if (block_group
->iref
)
3256 spin_unlock(&block_group
->lock
);
3257 block_group
= btrfs_next_block_group(block_group
);
3266 inode
= block_group
->inode
;
3267 block_group
->iref
= 0;
3268 block_group
->inode
= NULL
;
3269 spin_unlock(&block_group
->lock
);
3270 ASSERT(block_group
->io_ctl
.inode
== NULL
);
3272 last
= block_group
->start
+ block_group
->length
;
3273 btrfs_put_block_group(block_group
);
3278 * Must be called only after stopping all workers, since we could have block
3279 * group caching kthreads running, and therefore they could race with us if we
3280 * freed the block groups before stopping them.
3282 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
3284 struct btrfs_block_group
*block_group
;
3285 struct btrfs_space_info
*space_info
;
3286 struct btrfs_caching_control
*caching_ctl
;
3289 down_write(&info
->commit_root_sem
);
3290 while (!list_empty(&info
->caching_block_groups
)) {
3291 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
3292 struct btrfs_caching_control
, list
);
3293 list_del(&caching_ctl
->list
);
3294 btrfs_put_caching_control(caching_ctl
);
3296 up_write(&info
->commit_root_sem
);
3298 spin_lock(&info
->unused_bgs_lock
);
3299 while (!list_empty(&info
->unused_bgs
)) {
3300 block_group
= list_first_entry(&info
->unused_bgs
,
3301 struct btrfs_block_group
,
3303 list_del_init(&block_group
->bg_list
);
3304 btrfs_put_block_group(block_group
);
3306 spin_unlock(&info
->unused_bgs_lock
);
3308 spin_lock(&info
->block_group_cache_lock
);
3309 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
3310 block_group
= rb_entry(n
, struct btrfs_block_group
,
3312 rb_erase(&block_group
->cache_node
,
3313 &info
->block_group_cache_tree
);
3314 RB_CLEAR_NODE(&block_group
->cache_node
);
3315 spin_unlock(&info
->block_group_cache_lock
);
3317 down_write(&block_group
->space_info
->groups_sem
);
3318 list_del(&block_group
->list
);
3319 up_write(&block_group
->space_info
->groups_sem
);
3322 * We haven't cached this block group, which means we could
3323 * possibly have excluded extents on this block group.
3325 if (block_group
->cached
== BTRFS_CACHE_NO
||
3326 block_group
->cached
== BTRFS_CACHE_ERROR
)
3327 btrfs_free_excluded_extents(block_group
);
3329 btrfs_remove_free_space_cache(block_group
);
3330 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
3331 ASSERT(list_empty(&block_group
->dirty_list
));
3332 ASSERT(list_empty(&block_group
->io_list
));
3333 ASSERT(list_empty(&block_group
->bg_list
));
3334 ASSERT(refcount_read(&block_group
->refs
) == 1);
3335 btrfs_put_block_group(block_group
);
3337 spin_lock(&info
->block_group_cache_lock
);
3339 spin_unlock(&info
->block_group_cache_lock
);
3342 * Now that all the block groups are freed, go through and free all the
3343 * space_info structs. This is only called during the final stages of
3344 * unmount, and so we know nobody is using them. We call
3345 * synchronize_rcu() once before we start, just to be on the safe side.
3349 btrfs_release_global_block_rsv(info
);
3351 while (!list_empty(&info
->space_info
)) {
3352 space_info
= list_entry(info
->space_info
.next
,
3353 struct btrfs_space_info
,
3357 * Do not hide this behind enospc_debug, this is actually
3358 * important and indicates a real bug if this happens.
3360 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
3361 space_info
->bytes_reserved
> 0 ||
3362 space_info
->bytes_may_use
> 0))
3363 btrfs_dump_space_info(info
, space_info
, 0, 0);
3364 WARN_ON(space_info
->reclaim_size
> 0);
3365 list_del(&space_info
->list
);
3366 btrfs_sysfs_remove_space_info(space_info
);
3371 void btrfs_freeze_block_group(struct btrfs_block_group
*cache
)
3373 atomic_inc(&cache
->frozen
);
3376 void btrfs_unfreeze_block_group(struct btrfs_block_group
*block_group
)
3378 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3379 struct extent_map_tree
*em_tree
;
3380 struct extent_map
*em
;
3383 spin_lock(&block_group
->lock
);
3384 cleanup
= (atomic_dec_and_test(&block_group
->frozen
) &&
3385 block_group
->removed
);
3386 spin_unlock(&block_group
->lock
);
3389 em_tree
= &fs_info
->mapping_tree
;
3390 write_lock(&em_tree
->lock
);
3391 em
= lookup_extent_mapping(em_tree
, block_group
->start
,
3393 BUG_ON(!em
); /* logic error, can't happen */
3394 remove_extent_mapping(em_tree
, em
);
3395 write_unlock(&em_tree
->lock
);
3397 /* once for us and once for the tree */
3398 free_extent_map(em
);
3399 free_extent_map(em
);
3402 * We may have left one free space entry and other possible
3403 * tasks trimming this block group have left 1 entry each one.
3406 __btrfs_remove_free_space_cache(block_group
->free_space_ctl
);