2 * Copyright (C) 2012 Red Hat. All rights reserved.
4 * This file is released under the GPL.
8 #include "dm-bio-prison.h"
9 #include "dm-cache-metadata.h"
11 #include <linux/dm-io.h>
12 #include <linux/dm-kcopyd.h>
13 #include <linux/init.h>
14 #include <linux/mempool.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/vmalloc.h>
19 #define DM_MSG_PREFIX "cache"
21 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle
,
22 "A percentage of time allocated for copying to and/or from cache");
24 /*----------------------------------------------------------------*/
29 * oblock: index of an origin block
30 * cblock: index of a cache block
31 * promotion: movement of a block from origin to cache
32 * demotion: movement of a block from cache to origin
33 * migration: movement of a block between the origin and cache device,
37 /*----------------------------------------------------------------*/
39 static size_t bitset_size_in_bytes(unsigned nr_entries
)
41 return sizeof(unsigned long) * dm_div_up(nr_entries
, BITS_PER_LONG
);
44 static unsigned long *alloc_bitset(unsigned nr_entries
)
46 size_t s
= bitset_size_in_bytes(nr_entries
);
50 static void clear_bitset(void *bitset
, unsigned nr_entries
)
52 size_t s
= bitset_size_in_bytes(nr_entries
);
56 static void free_bitset(unsigned long *bits
)
61 /*----------------------------------------------------------------*/
63 #define PRISON_CELLS 1024
64 #define MIGRATION_POOL_SIZE 128
65 #define COMMIT_PERIOD HZ
66 #define MIGRATION_COUNT_WINDOW 10
69 * The block size of the device holding cache data must be >= 32KB
71 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT)
74 * FIXME: the cache is read/write for the time being.
77 CM_WRITE
, /* metadata may be changed */
78 CM_READ_ONLY
, /* metadata may not be changed */
81 struct cache_features
{
93 atomic_t copies_avoided
;
94 atomic_t cache_cell_clash
;
95 atomic_t commit_count
;
96 atomic_t discard_count
;
100 struct dm_target
*ti
;
101 struct dm_target_callbacks callbacks
;
104 * Metadata is written to this device.
106 struct dm_dev
*metadata_dev
;
109 * The slower of the two data devices. Typically a spindle.
111 struct dm_dev
*origin_dev
;
114 * The faster of the two data devices. Typically an SSD.
116 struct dm_dev
*cache_dev
;
119 * Cache features such as write-through.
121 struct cache_features features
;
124 * Size of the origin device in _complete_ blocks and native sectors.
126 dm_oblock_t origin_blocks
;
127 sector_t origin_sectors
;
130 * Size of the cache device in blocks.
132 dm_cblock_t cache_size
;
135 * Fields for converting from sectors to blocks.
137 uint32_t sectors_per_block
;
138 int sectors_per_block_shift
;
140 struct dm_cache_metadata
*cmd
;
143 struct bio_list deferred_bios
;
144 struct bio_list deferred_flush_bios
;
145 struct bio_list deferred_writethrough_bios
;
146 struct list_head quiesced_migrations
;
147 struct list_head completed_migrations
;
148 struct list_head need_commit_migrations
;
149 sector_t migration_threshold
;
150 atomic_t nr_migrations
;
151 wait_queue_head_t migration_wait
;
154 * cache_size entries, dirty if set
156 dm_cblock_t nr_dirty
;
157 unsigned long *dirty_bitset
;
160 * origin_blocks entries, discarded if set.
162 uint32_t discard_block_size
; /* a power of 2 times sectors per block */
163 dm_dblock_t discard_nr_blocks
;
164 unsigned long *discard_bitset
;
166 struct dm_kcopyd_client
*copier
;
167 struct workqueue_struct
*wq
;
168 struct work_struct worker
;
170 struct delayed_work waker
;
171 unsigned long last_commit_jiffies
;
173 struct dm_bio_prison
*prison
;
174 struct dm_deferred_set
*all_io_ds
;
176 mempool_t
*migration_pool
;
177 struct dm_cache_migration
*next_migration
;
179 struct dm_cache_policy
*policy
;
180 unsigned policy_nr_args
;
182 bool need_tick_bio
:1;
185 bool commit_requested
:1;
186 bool loaded_mappings
:1;
187 bool loaded_discards
:1;
189 struct cache_stats stats
;
192 * Rather than reconstructing the table line for the status we just
193 * save it and regurgitate.
195 unsigned nr_ctr_args
;
196 const char **ctr_args
;
199 struct per_bio_data
{
202 struct dm_deferred_entry
*all_io_entry
;
204 /* writethrough fields */
207 bio_end_io_t
*saved_bi_end_io
;
210 struct dm_cache_migration
{
211 struct list_head list
;
214 unsigned long start_jiffies
;
215 dm_oblock_t old_oblock
;
216 dm_oblock_t new_oblock
;
224 struct dm_bio_prison_cell
*old_ocell
;
225 struct dm_bio_prison_cell
*new_ocell
;
229 * Processing a bio in the worker thread may require these memory
230 * allocations. We prealloc to avoid deadlocks (the same worker thread
231 * frees them back to the mempool).
234 struct dm_cache_migration
*mg
;
235 struct dm_bio_prison_cell
*cell1
;
236 struct dm_bio_prison_cell
*cell2
;
239 static void wake_worker(struct cache
*cache
)
241 queue_work(cache
->wq
, &cache
->worker
);
244 /*----------------------------------------------------------------*/
246 static struct dm_bio_prison_cell
*alloc_prison_cell(struct cache
*cache
)
248 /* FIXME: change to use a local slab. */
249 return dm_bio_prison_alloc_cell(cache
->prison
, GFP_NOWAIT
);
252 static void free_prison_cell(struct cache
*cache
, struct dm_bio_prison_cell
*cell
)
254 dm_bio_prison_free_cell(cache
->prison
, cell
);
257 static int prealloc_data_structs(struct cache
*cache
, struct prealloc
*p
)
260 p
->mg
= mempool_alloc(cache
->migration_pool
, GFP_NOWAIT
);
266 p
->cell1
= alloc_prison_cell(cache
);
272 p
->cell2
= alloc_prison_cell(cache
);
280 static void prealloc_free_structs(struct cache
*cache
, struct prealloc
*p
)
283 free_prison_cell(cache
, p
->cell2
);
286 free_prison_cell(cache
, p
->cell1
);
289 mempool_free(p
->mg
, cache
->migration_pool
);
292 static struct dm_cache_migration
*prealloc_get_migration(struct prealloc
*p
)
294 struct dm_cache_migration
*mg
= p
->mg
;
303 * You must have a cell within the prealloc struct to return. If not this
304 * function will BUG() rather than returning NULL.
306 static struct dm_bio_prison_cell
*prealloc_get_cell(struct prealloc
*p
)
308 struct dm_bio_prison_cell
*r
= NULL
;
314 } else if (p
->cell2
) {
324 * You can't have more than two cells in a prealloc struct. BUG() will be
325 * called if you try and overfill.
327 static void prealloc_put_cell(struct prealloc
*p
, struct dm_bio_prison_cell
*cell
)
339 /*----------------------------------------------------------------*/
341 static void build_key(dm_oblock_t oblock
, struct dm_cell_key
*key
)
345 key
->block
= from_oblock(oblock
);
349 * The caller hands in a preallocated cell, and a free function for it.
350 * The cell will be freed if there's an error, or if it wasn't used because
351 * a cell with that key already exists.
353 typedef void (*cell_free_fn
)(void *context
, struct dm_bio_prison_cell
*cell
);
355 static int bio_detain(struct cache
*cache
, dm_oblock_t oblock
,
356 struct bio
*bio
, struct dm_bio_prison_cell
*cell_prealloc
,
357 cell_free_fn free_fn
, void *free_context
,
358 struct dm_bio_prison_cell
**cell_result
)
361 struct dm_cell_key key
;
363 build_key(oblock
, &key
);
364 r
= dm_bio_detain(cache
->prison
, &key
, bio
, cell_prealloc
, cell_result
);
366 free_fn(free_context
, cell_prealloc
);
371 static int get_cell(struct cache
*cache
,
373 struct prealloc
*structs
,
374 struct dm_bio_prison_cell
**cell_result
)
377 struct dm_cell_key key
;
378 struct dm_bio_prison_cell
*cell_prealloc
;
380 cell_prealloc
= prealloc_get_cell(structs
);
382 build_key(oblock
, &key
);
383 r
= dm_get_cell(cache
->prison
, &key
, cell_prealloc
, cell_result
);
385 prealloc_put_cell(structs
, cell_prealloc
);
390 /*----------------------------------------------------------------*/
392 static bool is_dirty(struct cache
*cache
, dm_cblock_t b
)
394 return test_bit(from_cblock(b
), cache
->dirty_bitset
);
397 static void set_dirty(struct cache
*cache
, dm_oblock_t oblock
, dm_cblock_t cblock
)
399 if (!test_and_set_bit(from_cblock(cblock
), cache
->dirty_bitset
)) {
400 cache
->nr_dirty
= to_cblock(from_cblock(cache
->nr_dirty
) + 1);
401 policy_set_dirty(cache
->policy
, oblock
);
405 static void clear_dirty(struct cache
*cache
, dm_oblock_t oblock
, dm_cblock_t cblock
)
407 if (test_and_clear_bit(from_cblock(cblock
), cache
->dirty_bitset
)) {
408 policy_clear_dirty(cache
->policy
, oblock
);
409 cache
->nr_dirty
= to_cblock(from_cblock(cache
->nr_dirty
) - 1);
410 if (!from_cblock(cache
->nr_dirty
))
411 dm_table_event(cache
->ti
->table
);
415 /*----------------------------------------------------------------*/
416 static bool block_size_is_power_of_two(struct cache
*cache
)
418 return cache
->sectors_per_block_shift
>= 0;
421 static dm_block_t
block_div(dm_block_t b
, uint32_t n
)
428 static dm_dblock_t
oblock_to_dblock(struct cache
*cache
, dm_oblock_t oblock
)
430 uint32_t discard_blocks
= cache
->discard_block_size
;
431 dm_block_t b
= from_oblock(oblock
);
433 if (!block_size_is_power_of_two(cache
))
434 discard_blocks
= discard_blocks
/ cache
->sectors_per_block
;
436 discard_blocks
>>= cache
->sectors_per_block_shift
;
438 b
= block_div(b
, discard_blocks
);
443 static void set_discard(struct cache
*cache
, dm_dblock_t b
)
447 atomic_inc(&cache
->stats
.discard_count
);
449 spin_lock_irqsave(&cache
->lock
, flags
);
450 set_bit(from_dblock(b
), cache
->discard_bitset
);
451 spin_unlock_irqrestore(&cache
->lock
, flags
);
454 static void clear_discard(struct cache
*cache
, dm_dblock_t b
)
458 spin_lock_irqsave(&cache
->lock
, flags
);
459 clear_bit(from_dblock(b
), cache
->discard_bitset
);
460 spin_unlock_irqrestore(&cache
->lock
, flags
);
463 static bool is_discarded(struct cache
*cache
, dm_dblock_t b
)
468 spin_lock_irqsave(&cache
->lock
, flags
);
469 r
= test_bit(from_dblock(b
), cache
->discard_bitset
);
470 spin_unlock_irqrestore(&cache
->lock
, flags
);
475 static bool is_discarded_oblock(struct cache
*cache
, dm_oblock_t b
)
480 spin_lock_irqsave(&cache
->lock
, flags
);
481 r
= test_bit(from_dblock(oblock_to_dblock(cache
, b
)),
482 cache
->discard_bitset
);
483 spin_unlock_irqrestore(&cache
->lock
, flags
);
488 /*----------------------------------------------------------------*/
490 static void load_stats(struct cache
*cache
)
492 struct dm_cache_statistics stats
;
494 dm_cache_metadata_get_stats(cache
->cmd
, &stats
);
495 atomic_set(&cache
->stats
.read_hit
, stats
.read_hits
);
496 atomic_set(&cache
->stats
.read_miss
, stats
.read_misses
);
497 atomic_set(&cache
->stats
.write_hit
, stats
.write_hits
);
498 atomic_set(&cache
->stats
.write_miss
, stats
.write_misses
);
501 static void save_stats(struct cache
*cache
)
503 struct dm_cache_statistics stats
;
505 stats
.read_hits
= atomic_read(&cache
->stats
.read_hit
);
506 stats
.read_misses
= atomic_read(&cache
->stats
.read_miss
);
507 stats
.write_hits
= atomic_read(&cache
->stats
.write_hit
);
508 stats
.write_misses
= atomic_read(&cache
->stats
.write_miss
);
510 dm_cache_metadata_set_stats(cache
->cmd
, &stats
);
513 /*----------------------------------------------------------------
515 *--------------------------------------------------------------*/
516 static struct per_bio_data
*get_per_bio_data(struct bio
*bio
)
518 struct per_bio_data
*pb
= dm_per_bio_data(bio
, sizeof(struct per_bio_data
));
523 static struct per_bio_data
*init_per_bio_data(struct bio
*bio
)
525 struct per_bio_data
*pb
= get_per_bio_data(bio
);
528 pb
->req_nr
= dm_bio_get_target_bio_nr(bio
);
529 pb
->all_io_entry
= NULL
;
534 /*----------------------------------------------------------------
536 *--------------------------------------------------------------*/
537 static void remap_to_origin(struct cache
*cache
, struct bio
*bio
)
539 bio
->bi_bdev
= cache
->origin_dev
->bdev
;
542 static void remap_to_cache(struct cache
*cache
, struct bio
*bio
,
545 sector_t bi_sector
= bio
->bi_sector
;
547 bio
->bi_bdev
= cache
->cache_dev
->bdev
;
548 if (!block_size_is_power_of_two(cache
))
549 bio
->bi_sector
= (from_cblock(cblock
) * cache
->sectors_per_block
) +
550 sector_div(bi_sector
, cache
->sectors_per_block
);
552 bio
->bi_sector
= (from_cblock(cblock
) << cache
->sectors_per_block_shift
) |
553 (bi_sector
& (cache
->sectors_per_block
- 1));
556 static void check_if_tick_bio_needed(struct cache
*cache
, struct bio
*bio
)
559 struct per_bio_data
*pb
= get_per_bio_data(bio
);
561 spin_lock_irqsave(&cache
->lock
, flags
);
562 if (cache
->need_tick_bio
&&
563 !(bio
->bi_rw
& (REQ_FUA
| REQ_FLUSH
| REQ_DISCARD
))) {
565 cache
->need_tick_bio
= false;
567 spin_unlock_irqrestore(&cache
->lock
, flags
);
570 static void remap_to_origin_clear_discard(struct cache
*cache
, struct bio
*bio
,
573 check_if_tick_bio_needed(cache
, bio
);
574 remap_to_origin(cache
, bio
);
575 if (bio_data_dir(bio
) == WRITE
)
576 clear_discard(cache
, oblock_to_dblock(cache
, oblock
));
579 static void remap_to_cache_dirty(struct cache
*cache
, struct bio
*bio
,
580 dm_oblock_t oblock
, dm_cblock_t cblock
)
582 remap_to_cache(cache
, bio
, cblock
);
583 if (bio_data_dir(bio
) == WRITE
) {
584 set_dirty(cache
, oblock
, cblock
);
585 clear_discard(cache
, oblock_to_dblock(cache
, oblock
));
589 static dm_oblock_t
get_bio_block(struct cache
*cache
, struct bio
*bio
)
591 sector_t block_nr
= bio
->bi_sector
;
593 if (!block_size_is_power_of_two(cache
))
594 (void) sector_div(block_nr
, cache
->sectors_per_block
);
596 block_nr
>>= cache
->sectors_per_block_shift
;
598 return to_oblock(block_nr
);
601 static int bio_triggers_commit(struct cache
*cache
, struct bio
*bio
)
603 return bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
);
606 static void issue(struct cache
*cache
, struct bio
*bio
)
610 if (!bio_triggers_commit(cache
, bio
)) {
611 generic_make_request(bio
);
616 * Batch together any bios that trigger commits and then issue a
617 * single commit for them in do_worker().
619 spin_lock_irqsave(&cache
->lock
, flags
);
620 cache
->commit_requested
= true;
621 bio_list_add(&cache
->deferred_flush_bios
, bio
);
622 spin_unlock_irqrestore(&cache
->lock
, flags
);
625 static void defer_writethrough_bio(struct cache
*cache
, struct bio
*bio
)
629 spin_lock_irqsave(&cache
->lock
, flags
);
630 bio_list_add(&cache
->deferred_writethrough_bios
, bio
);
631 spin_unlock_irqrestore(&cache
->lock
, flags
);
636 static void writethrough_endio(struct bio
*bio
, int err
)
638 struct per_bio_data
*pb
= get_per_bio_data(bio
);
639 bio
->bi_end_io
= pb
->saved_bi_end_io
;
646 remap_to_cache(pb
->cache
, bio
, pb
->cblock
);
649 * We can't issue this bio directly, since we're in interrupt
650 * context. So it get's put on a bio list for processing by the
653 defer_writethrough_bio(pb
->cache
, bio
);
657 * When running in writethrough mode we need to send writes to clean blocks
658 * to both the cache and origin devices. In future we'd like to clone the
659 * bio and send them in parallel, but for now we're doing them in
660 * series as this is easier.
662 static void remap_to_origin_then_cache(struct cache
*cache
, struct bio
*bio
,
663 dm_oblock_t oblock
, dm_cblock_t cblock
)
665 struct per_bio_data
*pb
= get_per_bio_data(bio
);
669 pb
->saved_bi_end_io
= bio
->bi_end_io
;
670 bio
->bi_end_io
= writethrough_endio
;
672 remap_to_origin_clear_discard(pb
->cache
, bio
, oblock
);
675 /*----------------------------------------------------------------
676 * Migration processing
678 * Migration covers moving data from the origin device to the cache, or
680 *--------------------------------------------------------------*/
681 static void free_migration(struct dm_cache_migration
*mg
)
683 mempool_free(mg
, mg
->cache
->migration_pool
);
686 static void inc_nr_migrations(struct cache
*cache
)
688 atomic_inc(&cache
->nr_migrations
);
691 static void dec_nr_migrations(struct cache
*cache
)
693 atomic_dec(&cache
->nr_migrations
);
696 * Wake the worker in case we're suspending the target.
698 wake_up(&cache
->migration_wait
);
701 static void __cell_defer(struct cache
*cache
, struct dm_bio_prison_cell
*cell
,
704 (holder
? dm_cell_release
: dm_cell_release_no_holder
)
705 (cache
->prison
, cell
, &cache
->deferred_bios
);
706 free_prison_cell(cache
, cell
);
709 static void cell_defer(struct cache
*cache
, struct dm_bio_prison_cell
*cell
,
714 spin_lock_irqsave(&cache
->lock
, flags
);
715 __cell_defer(cache
, cell
, holder
);
716 spin_unlock_irqrestore(&cache
->lock
, flags
);
721 static void cleanup_migration(struct dm_cache_migration
*mg
)
723 dec_nr_migrations(mg
->cache
);
727 static void migration_failure(struct dm_cache_migration
*mg
)
729 struct cache
*cache
= mg
->cache
;
732 DMWARN_LIMIT("writeback failed; couldn't copy block");
733 set_dirty(cache
, mg
->old_oblock
, mg
->cblock
);
734 cell_defer(cache
, mg
->old_ocell
, false);
736 } else if (mg
->demote
) {
737 DMWARN_LIMIT("demotion failed; couldn't copy block");
738 policy_force_mapping(cache
->policy
, mg
->new_oblock
, mg
->old_oblock
);
740 cell_defer(cache
, mg
->old_ocell
, mg
->promote
? 0 : 1);
742 cell_defer(cache
, mg
->new_ocell
, 1);
744 DMWARN_LIMIT("promotion failed; couldn't copy block");
745 policy_remove_mapping(cache
->policy
, mg
->new_oblock
);
746 cell_defer(cache
, mg
->new_ocell
, 1);
749 cleanup_migration(mg
);
752 static void migration_success_pre_commit(struct dm_cache_migration
*mg
)
755 struct cache
*cache
= mg
->cache
;
758 cell_defer(cache
, mg
->old_ocell
, false);
759 clear_dirty(cache
, mg
->old_oblock
, mg
->cblock
);
760 cleanup_migration(mg
);
763 } else if (mg
->demote
) {
764 if (dm_cache_remove_mapping(cache
->cmd
, mg
->cblock
)) {
765 DMWARN_LIMIT("demotion failed; couldn't update on disk metadata");
766 policy_force_mapping(cache
->policy
, mg
->new_oblock
,
769 cell_defer(cache
, mg
->new_ocell
, true);
770 cleanup_migration(mg
);
774 if (dm_cache_insert_mapping(cache
->cmd
, mg
->cblock
, mg
->new_oblock
)) {
775 DMWARN_LIMIT("promotion failed; couldn't update on disk metadata");
776 policy_remove_mapping(cache
->policy
, mg
->new_oblock
);
777 cleanup_migration(mg
);
782 spin_lock_irqsave(&cache
->lock
, flags
);
783 list_add_tail(&mg
->list
, &cache
->need_commit_migrations
);
784 cache
->commit_requested
= true;
785 spin_unlock_irqrestore(&cache
->lock
, flags
);
788 static void migration_success_post_commit(struct dm_cache_migration
*mg
)
791 struct cache
*cache
= mg
->cache
;
794 DMWARN("writeback unexpectedly triggered commit");
797 } else if (mg
->demote
) {
798 cell_defer(cache
, mg
->old_ocell
, mg
->promote
? 0 : 1);
803 spin_lock_irqsave(&cache
->lock
, flags
);
804 list_add_tail(&mg
->list
, &cache
->quiesced_migrations
);
805 spin_unlock_irqrestore(&cache
->lock
, flags
);
808 cleanup_migration(mg
);
811 cell_defer(cache
, mg
->new_ocell
, true);
812 clear_dirty(cache
, mg
->new_oblock
, mg
->cblock
);
813 cleanup_migration(mg
);
817 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
820 struct dm_cache_migration
*mg
= (struct dm_cache_migration
*) context
;
821 struct cache
*cache
= mg
->cache
;
823 if (read_err
|| write_err
)
826 spin_lock_irqsave(&cache
->lock
, flags
);
827 list_add_tail(&mg
->list
, &cache
->completed_migrations
);
828 spin_unlock_irqrestore(&cache
->lock
, flags
);
833 static void issue_copy_real(struct dm_cache_migration
*mg
)
836 struct dm_io_region o_region
, c_region
;
837 struct cache
*cache
= mg
->cache
;
839 o_region
.bdev
= cache
->origin_dev
->bdev
;
840 o_region
.count
= cache
->sectors_per_block
;
842 c_region
.bdev
= cache
->cache_dev
->bdev
;
843 c_region
.sector
= from_cblock(mg
->cblock
) * cache
->sectors_per_block
;
844 c_region
.count
= cache
->sectors_per_block
;
846 if (mg
->writeback
|| mg
->demote
) {
848 o_region
.sector
= from_oblock(mg
->old_oblock
) * cache
->sectors_per_block
;
849 r
= dm_kcopyd_copy(cache
->copier
, &c_region
, 1, &o_region
, 0, copy_complete
, mg
);
852 o_region
.sector
= from_oblock(mg
->new_oblock
) * cache
->sectors_per_block
;
853 r
= dm_kcopyd_copy(cache
->copier
, &o_region
, 1, &c_region
, 0, copy_complete
, mg
);
857 migration_failure(mg
);
860 static void avoid_copy(struct dm_cache_migration
*mg
)
862 atomic_inc(&mg
->cache
->stats
.copies_avoided
);
863 migration_success_pre_commit(mg
);
866 static void issue_copy(struct dm_cache_migration
*mg
)
869 struct cache
*cache
= mg
->cache
;
871 if (mg
->writeback
|| mg
->demote
)
872 avoid
= !is_dirty(cache
, mg
->cblock
) ||
873 is_discarded_oblock(cache
, mg
->old_oblock
);
875 avoid
= is_discarded_oblock(cache
, mg
->new_oblock
);
877 avoid
? avoid_copy(mg
) : issue_copy_real(mg
);
880 static void complete_migration(struct dm_cache_migration
*mg
)
883 migration_failure(mg
);
885 migration_success_pre_commit(mg
);
888 static void process_migrations(struct cache
*cache
, struct list_head
*head
,
889 void (*fn
)(struct dm_cache_migration
*))
892 struct list_head list
;
893 struct dm_cache_migration
*mg
, *tmp
;
895 INIT_LIST_HEAD(&list
);
896 spin_lock_irqsave(&cache
->lock
, flags
);
897 list_splice_init(head
, &list
);
898 spin_unlock_irqrestore(&cache
->lock
, flags
);
900 list_for_each_entry_safe(mg
, tmp
, &list
, list
)
904 static void __queue_quiesced_migration(struct dm_cache_migration
*mg
)
906 list_add_tail(&mg
->list
, &mg
->cache
->quiesced_migrations
);
909 static void queue_quiesced_migration(struct dm_cache_migration
*mg
)
912 struct cache
*cache
= mg
->cache
;
914 spin_lock_irqsave(&cache
->lock
, flags
);
915 __queue_quiesced_migration(mg
);
916 spin_unlock_irqrestore(&cache
->lock
, flags
);
921 static void queue_quiesced_migrations(struct cache
*cache
, struct list_head
*work
)
924 struct dm_cache_migration
*mg
, *tmp
;
926 spin_lock_irqsave(&cache
->lock
, flags
);
927 list_for_each_entry_safe(mg
, tmp
, work
, list
)
928 __queue_quiesced_migration(mg
);
929 spin_unlock_irqrestore(&cache
->lock
, flags
);
934 static void check_for_quiesced_migrations(struct cache
*cache
,
935 struct per_bio_data
*pb
)
937 struct list_head work
;
939 if (!pb
->all_io_entry
)
942 INIT_LIST_HEAD(&work
);
943 if (pb
->all_io_entry
)
944 dm_deferred_entry_dec(pb
->all_io_entry
, &work
);
946 if (!list_empty(&work
))
947 queue_quiesced_migrations(cache
, &work
);
950 static void quiesce_migration(struct dm_cache_migration
*mg
)
952 if (!dm_deferred_set_add_work(mg
->cache
->all_io_ds
, &mg
->list
))
953 queue_quiesced_migration(mg
);
956 static void promote(struct cache
*cache
, struct prealloc
*structs
,
957 dm_oblock_t oblock
, dm_cblock_t cblock
,
958 struct dm_bio_prison_cell
*cell
)
960 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
963 mg
->writeback
= false;
967 mg
->new_oblock
= oblock
;
969 mg
->old_ocell
= NULL
;
970 mg
->new_ocell
= cell
;
971 mg
->start_jiffies
= jiffies
;
973 inc_nr_migrations(cache
);
974 quiesce_migration(mg
);
977 static void writeback(struct cache
*cache
, struct prealloc
*structs
,
978 dm_oblock_t oblock
, dm_cblock_t cblock
,
979 struct dm_bio_prison_cell
*cell
)
981 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
984 mg
->writeback
= true;
988 mg
->old_oblock
= oblock
;
990 mg
->old_ocell
= cell
;
991 mg
->new_ocell
= NULL
;
992 mg
->start_jiffies
= jiffies
;
994 inc_nr_migrations(cache
);
995 quiesce_migration(mg
);
998 static void demote_then_promote(struct cache
*cache
, struct prealloc
*structs
,
999 dm_oblock_t old_oblock
, dm_oblock_t new_oblock
,
1001 struct dm_bio_prison_cell
*old_ocell
,
1002 struct dm_bio_prison_cell
*new_ocell
)
1004 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
1007 mg
->writeback
= false;
1011 mg
->old_oblock
= old_oblock
;
1012 mg
->new_oblock
= new_oblock
;
1013 mg
->cblock
= cblock
;
1014 mg
->old_ocell
= old_ocell
;
1015 mg
->new_ocell
= new_ocell
;
1016 mg
->start_jiffies
= jiffies
;
1018 inc_nr_migrations(cache
);
1019 quiesce_migration(mg
);
1022 /*----------------------------------------------------------------
1024 *--------------------------------------------------------------*/
1025 static void defer_bio(struct cache
*cache
, struct bio
*bio
)
1027 unsigned long flags
;
1029 spin_lock_irqsave(&cache
->lock
, flags
);
1030 bio_list_add(&cache
->deferred_bios
, bio
);
1031 spin_unlock_irqrestore(&cache
->lock
, flags
);
1036 static void process_flush_bio(struct cache
*cache
, struct bio
*bio
)
1038 struct per_bio_data
*pb
= get_per_bio_data(bio
);
1040 BUG_ON(bio
->bi_size
);
1042 remap_to_origin(cache
, bio
);
1044 remap_to_cache(cache
, bio
, 0);
1050 * People generally discard large parts of a device, eg, the whole device
1051 * when formatting. Splitting these large discards up into cache block
1052 * sized ios and then quiescing (always neccessary for discard) takes too
1055 * We keep it simple, and allow any size of discard to come in, and just
1056 * mark off blocks on the discard bitset. No passdown occurs!
1058 * To implement passdown we need to change the bio_prison such that a cell
1059 * can have a key that spans many blocks.
1061 static void process_discard_bio(struct cache
*cache
, struct bio
*bio
)
1063 dm_block_t start_block
= dm_sector_div_up(bio
->bi_sector
,
1064 cache
->discard_block_size
);
1065 dm_block_t end_block
= bio
->bi_sector
+ bio_sectors(bio
);
1068 end_block
= block_div(end_block
, cache
->discard_block_size
);
1070 for (b
= start_block
; b
< end_block
; b
++)
1071 set_discard(cache
, to_dblock(b
));
1076 static bool spare_migration_bandwidth(struct cache
*cache
)
1078 sector_t current_volume
= (atomic_read(&cache
->nr_migrations
) + 1) *
1079 cache
->sectors_per_block
;
1080 return current_volume
< cache
->migration_threshold
;
1083 static bool is_writethrough_io(struct cache
*cache
, struct bio
*bio
,
1086 return bio_data_dir(bio
) == WRITE
&&
1087 cache
->features
.write_through
&& !is_dirty(cache
, cblock
);
1090 static void inc_hit_counter(struct cache
*cache
, struct bio
*bio
)
1092 atomic_inc(bio_data_dir(bio
) == READ
?
1093 &cache
->stats
.read_hit
: &cache
->stats
.write_hit
);
1096 static void inc_miss_counter(struct cache
*cache
, struct bio
*bio
)
1098 atomic_inc(bio_data_dir(bio
) == READ
?
1099 &cache
->stats
.read_miss
: &cache
->stats
.write_miss
);
1102 static void process_bio(struct cache
*cache
, struct prealloc
*structs
,
1106 bool release_cell
= true;
1107 dm_oblock_t block
= get_bio_block(cache
, bio
);
1108 struct dm_bio_prison_cell
*cell_prealloc
, *old_ocell
, *new_ocell
;
1109 struct policy_result lookup_result
;
1110 struct per_bio_data
*pb
= get_per_bio_data(bio
);
1111 bool discarded_block
= is_discarded_oblock(cache
, block
);
1112 bool can_migrate
= discarded_block
|| spare_migration_bandwidth(cache
);
1115 * Check to see if that block is currently migrating.
1117 cell_prealloc
= prealloc_get_cell(structs
);
1118 r
= bio_detain(cache
, block
, bio
, cell_prealloc
,
1119 (cell_free_fn
) prealloc_put_cell
,
1120 structs
, &new_ocell
);
1124 r
= policy_map(cache
->policy
, block
, true, can_migrate
, discarded_block
,
1125 bio
, &lookup_result
);
1127 if (r
== -EWOULDBLOCK
)
1128 /* migration has been denied */
1129 lookup_result
.op
= POLICY_MISS
;
1131 switch (lookup_result
.op
) {
1133 inc_hit_counter(cache
, bio
);
1134 pb
->all_io_entry
= dm_deferred_entry_inc(cache
->all_io_ds
);
1136 if (is_writethrough_io(cache
, bio
, lookup_result
.cblock
))
1137 remap_to_origin_then_cache(cache
, bio
, block
, lookup_result
.cblock
);
1139 remap_to_cache_dirty(cache
, bio
, block
, lookup_result
.cblock
);
1145 inc_miss_counter(cache
, bio
);
1146 pb
->all_io_entry
= dm_deferred_entry_inc(cache
->all_io_ds
);
1147 remap_to_origin_clear_discard(cache
, bio
, block
);
1152 atomic_inc(&cache
->stats
.promotion
);
1153 promote(cache
, structs
, block
, lookup_result
.cblock
, new_ocell
);
1154 release_cell
= false;
1157 case POLICY_REPLACE
:
1158 cell_prealloc
= prealloc_get_cell(structs
);
1159 r
= bio_detain(cache
, lookup_result
.old_oblock
, bio
, cell_prealloc
,
1160 (cell_free_fn
) prealloc_put_cell
,
1161 structs
, &old_ocell
);
1164 * We have to be careful to avoid lock inversion of
1165 * the cells. So we back off, and wait for the
1166 * old_ocell to become free.
1168 policy_force_mapping(cache
->policy
, block
,
1169 lookup_result
.old_oblock
);
1170 atomic_inc(&cache
->stats
.cache_cell_clash
);
1173 atomic_inc(&cache
->stats
.demotion
);
1174 atomic_inc(&cache
->stats
.promotion
);
1176 demote_then_promote(cache
, structs
, lookup_result
.old_oblock
,
1177 block
, lookup_result
.cblock
,
1178 old_ocell
, new_ocell
);
1179 release_cell
= false;
1183 DMERR_LIMIT("%s: erroring bio, unknown policy op: %u", __func__
,
1184 (unsigned) lookup_result
.op
);
1189 cell_defer(cache
, new_ocell
, false);
1192 static int need_commit_due_to_time(struct cache
*cache
)
1194 return jiffies
< cache
->last_commit_jiffies
||
1195 jiffies
> cache
->last_commit_jiffies
+ COMMIT_PERIOD
;
1198 static int commit_if_needed(struct cache
*cache
)
1200 if (dm_cache_changed_this_transaction(cache
->cmd
) &&
1201 (cache
->commit_requested
|| need_commit_due_to_time(cache
))) {
1202 atomic_inc(&cache
->stats
.commit_count
);
1203 cache
->last_commit_jiffies
= jiffies
;
1204 cache
->commit_requested
= false;
1205 return dm_cache_commit(cache
->cmd
, false);
1211 static void process_deferred_bios(struct cache
*cache
)
1213 unsigned long flags
;
1214 struct bio_list bios
;
1216 struct prealloc structs
;
1218 memset(&structs
, 0, sizeof(structs
));
1219 bio_list_init(&bios
);
1221 spin_lock_irqsave(&cache
->lock
, flags
);
1222 bio_list_merge(&bios
, &cache
->deferred_bios
);
1223 bio_list_init(&cache
->deferred_bios
);
1224 spin_unlock_irqrestore(&cache
->lock
, flags
);
1226 while (!bio_list_empty(&bios
)) {
1228 * If we've got no free migration structs, and processing
1229 * this bio might require one, we pause until there are some
1230 * prepared mappings to process.
1232 if (prealloc_data_structs(cache
, &structs
)) {
1233 spin_lock_irqsave(&cache
->lock
, flags
);
1234 bio_list_merge(&cache
->deferred_bios
, &bios
);
1235 spin_unlock_irqrestore(&cache
->lock
, flags
);
1239 bio
= bio_list_pop(&bios
);
1241 if (bio
->bi_rw
& REQ_FLUSH
)
1242 process_flush_bio(cache
, bio
);
1243 else if (bio
->bi_rw
& REQ_DISCARD
)
1244 process_discard_bio(cache
, bio
);
1246 process_bio(cache
, &structs
, bio
);
1249 prealloc_free_structs(cache
, &structs
);
1252 static void process_deferred_flush_bios(struct cache
*cache
, bool submit_bios
)
1254 unsigned long flags
;
1255 struct bio_list bios
;
1258 bio_list_init(&bios
);
1260 spin_lock_irqsave(&cache
->lock
, flags
);
1261 bio_list_merge(&bios
, &cache
->deferred_flush_bios
);
1262 bio_list_init(&cache
->deferred_flush_bios
);
1263 spin_unlock_irqrestore(&cache
->lock
, flags
);
1265 while ((bio
= bio_list_pop(&bios
)))
1266 submit_bios
? generic_make_request(bio
) : bio_io_error(bio
);
1269 static void process_deferred_writethrough_bios(struct cache
*cache
)
1271 unsigned long flags
;
1272 struct bio_list bios
;
1275 bio_list_init(&bios
);
1277 spin_lock_irqsave(&cache
->lock
, flags
);
1278 bio_list_merge(&bios
, &cache
->deferred_writethrough_bios
);
1279 bio_list_init(&cache
->deferred_writethrough_bios
);
1280 spin_unlock_irqrestore(&cache
->lock
, flags
);
1282 while ((bio
= bio_list_pop(&bios
)))
1283 generic_make_request(bio
);
1286 static void writeback_some_dirty_blocks(struct cache
*cache
)
1291 struct prealloc structs
;
1292 struct dm_bio_prison_cell
*old_ocell
;
1294 memset(&structs
, 0, sizeof(structs
));
1296 while (spare_migration_bandwidth(cache
)) {
1297 if (prealloc_data_structs(cache
, &structs
))
1300 r
= policy_writeback_work(cache
->policy
, &oblock
, &cblock
);
1304 r
= get_cell(cache
, oblock
, &structs
, &old_ocell
);
1306 policy_set_dirty(cache
->policy
, oblock
);
1310 writeback(cache
, &structs
, oblock
, cblock
, old_ocell
);
1313 prealloc_free_structs(cache
, &structs
);
1316 /*----------------------------------------------------------------
1318 *--------------------------------------------------------------*/
1319 static void start_quiescing(struct cache
*cache
)
1321 unsigned long flags
;
1323 spin_lock_irqsave(&cache
->lock
, flags
);
1324 cache
->quiescing
= 1;
1325 spin_unlock_irqrestore(&cache
->lock
, flags
);
1328 static void stop_quiescing(struct cache
*cache
)
1330 unsigned long flags
;
1332 spin_lock_irqsave(&cache
->lock
, flags
);
1333 cache
->quiescing
= 0;
1334 spin_unlock_irqrestore(&cache
->lock
, flags
);
1337 static bool is_quiescing(struct cache
*cache
)
1340 unsigned long flags
;
1342 spin_lock_irqsave(&cache
->lock
, flags
);
1343 r
= cache
->quiescing
;
1344 spin_unlock_irqrestore(&cache
->lock
, flags
);
1349 static void wait_for_migrations(struct cache
*cache
)
1351 wait_event(cache
->migration_wait
, !atomic_read(&cache
->nr_migrations
));
1354 static void stop_worker(struct cache
*cache
)
1356 cancel_delayed_work(&cache
->waker
);
1357 flush_workqueue(cache
->wq
);
1360 static void requeue_deferred_io(struct cache
*cache
)
1363 struct bio_list bios
;
1365 bio_list_init(&bios
);
1366 bio_list_merge(&bios
, &cache
->deferred_bios
);
1367 bio_list_init(&cache
->deferred_bios
);
1369 while ((bio
= bio_list_pop(&bios
)))
1370 bio_endio(bio
, DM_ENDIO_REQUEUE
);
1373 static int more_work(struct cache
*cache
)
1375 if (is_quiescing(cache
))
1376 return !list_empty(&cache
->quiesced_migrations
) ||
1377 !list_empty(&cache
->completed_migrations
) ||
1378 !list_empty(&cache
->need_commit_migrations
);
1380 return !bio_list_empty(&cache
->deferred_bios
) ||
1381 !bio_list_empty(&cache
->deferred_flush_bios
) ||
1382 !bio_list_empty(&cache
->deferred_writethrough_bios
) ||
1383 !list_empty(&cache
->quiesced_migrations
) ||
1384 !list_empty(&cache
->completed_migrations
) ||
1385 !list_empty(&cache
->need_commit_migrations
);
1388 static void do_worker(struct work_struct
*ws
)
1390 struct cache
*cache
= container_of(ws
, struct cache
, worker
);
1393 if (!is_quiescing(cache
))
1394 process_deferred_bios(cache
);
1396 process_migrations(cache
, &cache
->quiesced_migrations
, issue_copy
);
1397 process_migrations(cache
, &cache
->completed_migrations
, complete_migration
);
1399 writeback_some_dirty_blocks(cache
);
1401 process_deferred_writethrough_bios(cache
);
1403 if (commit_if_needed(cache
)) {
1404 process_deferred_flush_bios(cache
, false);
1407 * FIXME: rollback metadata or just go into a
1408 * failure mode and error everything
1411 process_deferred_flush_bios(cache
, true);
1412 process_migrations(cache
, &cache
->need_commit_migrations
,
1413 migration_success_post_commit
);
1415 } while (more_work(cache
));
1419 * We want to commit periodically so that not too much
1420 * unwritten metadata builds up.
1422 static void do_waker(struct work_struct
*ws
)
1424 struct cache
*cache
= container_of(to_delayed_work(ws
), struct cache
, waker
);
1426 queue_delayed_work(cache
->wq
, &cache
->waker
, COMMIT_PERIOD
);
1429 /*----------------------------------------------------------------*/
1431 static int is_congested(struct dm_dev
*dev
, int bdi_bits
)
1433 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1434 return bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1437 static int cache_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1439 struct cache
*cache
= container_of(cb
, struct cache
, callbacks
);
1441 return is_congested(cache
->origin_dev
, bdi_bits
) ||
1442 is_congested(cache
->cache_dev
, bdi_bits
);
1445 /*----------------------------------------------------------------
1447 *--------------------------------------------------------------*/
1450 * This function gets called on the error paths of the constructor, so we
1451 * have to cope with a partially initialised struct.
1453 static void destroy(struct cache
*cache
)
1457 if (cache
->next_migration
)
1458 mempool_free(cache
->next_migration
, cache
->migration_pool
);
1460 if (cache
->migration_pool
)
1461 mempool_destroy(cache
->migration_pool
);
1463 if (cache
->all_io_ds
)
1464 dm_deferred_set_destroy(cache
->all_io_ds
);
1467 dm_bio_prison_destroy(cache
->prison
);
1470 destroy_workqueue(cache
->wq
);
1472 if (cache
->dirty_bitset
)
1473 free_bitset(cache
->dirty_bitset
);
1475 if (cache
->discard_bitset
)
1476 free_bitset(cache
->discard_bitset
);
1479 dm_kcopyd_client_destroy(cache
->copier
);
1482 dm_cache_metadata_close(cache
->cmd
);
1484 if (cache
->metadata_dev
)
1485 dm_put_device(cache
->ti
, cache
->metadata_dev
);
1487 if (cache
->origin_dev
)
1488 dm_put_device(cache
->ti
, cache
->origin_dev
);
1490 if (cache
->cache_dev
)
1491 dm_put_device(cache
->ti
, cache
->cache_dev
);
1494 dm_cache_policy_destroy(cache
->policy
);
1496 for (i
= 0; i
< cache
->nr_ctr_args
; i
++)
1497 kfree(cache
->ctr_args
[i
]);
1498 kfree(cache
->ctr_args
);
1503 static void cache_dtr(struct dm_target
*ti
)
1505 struct cache
*cache
= ti
->private;
1510 static sector_t
get_dev_size(struct dm_dev
*dev
)
1512 return i_size_read(dev
->bdev
->bd_inode
) >> SECTOR_SHIFT
;
1515 /*----------------------------------------------------------------*/
1518 * Construct a cache device mapping.
1520 * cache <metadata dev> <cache dev> <origin dev> <block size>
1521 * <#feature args> [<feature arg>]*
1522 * <policy> <#policy args> [<policy arg>]*
1524 * metadata dev : fast device holding the persistent metadata
1525 * cache dev : fast device holding cached data blocks
1526 * origin dev : slow device holding original data blocks
1527 * block size : cache unit size in sectors
1529 * #feature args : number of feature arguments passed
1530 * feature args : writethrough. (The default is writeback.)
1532 * policy : the replacement policy to use
1533 * #policy args : an even number of policy arguments corresponding
1534 * to key/value pairs passed to the policy
1535 * policy args : key/value pairs passed to the policy
1536 * E.g. 'sequential_threshold 1024'
1537 * See cache-policies.txt for details.
1539 * Optional feature arguments are:
1540 * writethrough : write through caching that prohibits cache block
1541 * content from being different from origin block content.
1542 * Without this argument, the default behaviour is to write
1543 * back cache block contents later for performance reasons,
1544 * so they may differ from the corresponding origin blocks.
1547 struct dm_target
*ti
;
1549 struct dm_dev
*metadata_dev
;
1551 struct dm_dev
*cache_dev
;
1552 sector_t cache_sectors
;
1554 struct dm_dev
*origin_dev
;
1555 sector_t origin_sectors
;
1557 uint32_t block_size
;
1559 const char *policy_name
;
1561 const char **policy_argv
;
1563 struct cache_features features
;
1566 static void destroy_cache_args(struct cache_args
*ca
)
1568 if (ca
->metadata_dev
)
1569 dm_put_device(ca
->ti
, ca
->metadata_dev
);
1572 dm_put_device(ca
->ti
, ca
->cache_dev
);
1575 dm_put_device(ca
->ti
, ca
->origin_dev
);
1580 static bool at_least_one_arg(struct dm_arg_set
*as
, char **error
)
1583 *error
= "Insufficient args";
1590 static int parse_metadata_dev(struct cache_args
*ca
, struct dm_arg_set
*as
,
1594 sector_t metadata_dev_size
;
1595 char b
[BDEVNAME_SIZE
];
1597 if (!at_least_one_arg(as
, error
))
1600 r
= dm_get_device(ca
->ti
, dm_shift_arg(as
), FMODE_READ
| FMODE_WRITE
,
1603 *error
= "Error opening metadata device";
1607 metadata_dev_size
= get_dev_size(ca
->metadata_dev
);
1608 if (metadata_dev_size
> DM_CACHE_METADATA_MAX_SECTORS_WARNING
)
1609 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1610 bdevname(ca
->metadata_dev
->bdev
, b
), THIN_METADATA_MAX_SECTORS
);
1615 static int parse_cache_dev(struct cache_args
*ca
, struct dm_arg_set
*as
,
1620 if (!at_least_one_arg(as
, error
))
1623 r
= dm_get_device(ca
->ti
, dm_shift_arg(as
), FMODE_READ
| FMODE_WRITE
,
1626 *error
= "Error opening cache device";
1629 ca
->cache_sectors
= get_dev_size(ca
->cache_dev
);
1634 static int parse_origin_dev(struct cache_args
*ca
, struct dm_arg_set
*as
,
1639 if (!at_least_one_arg(as
, error
))
1642 r
= dm_get_device(ca
->ti
, dm_shift_arg(as
), FMODE_READ
| FMODE_WRITE
,
1645 *error
= "Error opening origin device";
1649 ca
->origin_sectors
= get_dev_size(ca
->origin_dev
);
1650 if (ca
->ti
->len
> ca
->origin_sectors
) {
1651 *error
= "Device size larger than cached device";
1658 static int parse_block_size(struct cache_args
*ca
, struct dm_arg_set
*as
,
1663 if (!at_least_one_arg(as
, error
))
1666 if (kstrtoul(dm_shift_arg(as
), 10, &tmp
) || !tmp
||
1667 tmp
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
1668 tmp
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
1669 *error
= "Invalid data block size";
1673 if (tmp
> ca
->cache_sectors
) {
1674 *error
= "Data block size is larger than the cache device";
1678 ca
->block_size
= tmp
;
1683 static void init_features(struct cache_features
*cf
)
1685 cf
->mode
= CM_WRITE
;
1686 cf
->write_through
= false;
1689 static int parse_features(struct cache_args
*ca
, struct dm_arg_set
*as
,
1692 static struct dm_arg _args
[] = {
1693 {0, 1, "Invalid number of cache feature arguments"},
1699 struct cache_features
*cf
= &ca
->features
;
1703 r
= dm_read_arg_group(_args
, as
, &argc
, error
);
1708 arg
= dm_shift_arg(as
);
1710 if (!strcasecmp(arg
, "writeback"))
1711 cf
->write_through
= false;
1713 else if (!strcasecmp(arg
, "writethrough"))
1714 cf
->write_through
= true;
1717 *error
= "Unrecognised cache feature requested";
1725 static int parse_policy(struct cache_args
*ca
, struct dm_arg_set
*as
,
1728 static struct dm_arg _args
[] = {
1729 {0, 1024, "Invalid number of policy arguments"},
1734 if (!at_least_one_arg(as
, error
))
1737 ca
->policy_name
= dm_shift_arg(as
);
1739 r
= dm_read_arg_group(_args
, as
, &ca
->policy_argc
, error
);
1743 ca
->policy_argv
= (const char **)as
->argv
;
1744 dm_consume_args(as
, ca
->policy_argc
);
1749 static int parse_cache_args(struct cache_args
*ca
, int argc
, char **argv
,
1753 struct dm_arg_set as
;
1758 r
= parse_metadata_dev(ca
, &as
, error
);
1762 r
= parse_cache_dev(ca
, &as
, error
);
1766 r
= parse_origin_dev(ca
, &as
, error
);
1770 r
= parse_block_size(ca
, &as
, error
);
1774 r
= parse_features(ca
, &as
, error
);
1778 r
= parse_policy(ca
, &as
, error
);
1785 /*----------------------------------------------------------------*/
1787 static struct kmem_cache
*migration_cache
;
1789 static int set_config_values(struct dm_cache_policy
*p
, int argc
, const char **argv
)
1794 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
1799 r
= policy_set_config_value(p
, argv
[0], argv
[1]);
1801 DMWARN("policy_set_config_value failed: key = '%s', value = '%s'",
1813 static int create_cache_policy(struct cache
*cache
, struct cache_args
*ca
,
1818 cache
->policy
= dm_cache_policy_create(ca
->policy_name
,
1820 cache
->origin_sectors
,
1821 cache
->sectors_per_block
);
1822 if (!cache
->policy
) {
1823 *error
= "Error creating cache's policy";
1827 r
= set_config_values(cache
->policy
, ca
->policy_argc
, ca
->policy_argv
);
1829 *error
= "Error setting cache policy's config values";
1830 dm_cache_policy_destroy(cache
->policy
);
1831 cache
->policy
= NULL
;
1838 * We want the discard block size to be a power of two, at least the size
1839 * of the cache block size, and have no more than 2^14 discard blocks
1840 * across the origin.
1842 #define MAX_DISCARD_BLOCKS (1 << 14)
1844 static bool too_many_discard_blocks(sector_t discard_block_size
,
1845 sector_t origin_size
)
1847 (void) sector_div(origin_size
, discard_block_size
);
1849 return origin_size
> MAX_DISCARD_BLOCKS
;
1852 static sector_t
calculate_discard_block_size(sector_t cache_block_size
,
1853 sector_t origin_size
)
1855 sector_t discard_block_size
;
1857 discard_block_size
= roundup_pow_of_two(cache_block_size
);
1860 while (too_many_discard_blocks(discard_block_size
, origin_size
))
1861 discard_block_size
*= 2;
1863 return discard_block_size
;
1866 #define DEFAULT_MIGRATION_THRESHOLD (2048 * 100)
1868 static int cache_create(struct cache_args
*ca
, struct cache
**result
)
1871 char **error
= &ca
->ti
->error
;
1872 struct cache
*cache
;
1873 struct dm_target
*ti
= ca
->ti
;
1874 dm_block_t origin_blocks
;
1875 struct dm_cache_metadata
*cmd
;
1876 bool may_format
= ca
->features
.mode
== CM_WRITE
;
1878 cache
= kzalloc(sizeof(*cache
), GFP_KERNEL
);
1883 ti
->private = cache
;
1884 ti
->per_bio_data_size
= sizeof(struct per_bio_data
);
1885 ti
->num_flush_bios
= 2;
1886 ti
->flush_supported
= true;
1888 ti
->num_discard_bios
= 1;
1889 ti
->discards_supported
= true;
1890 ti
->discard_zeroes_data_unsupported
= true;
1892 memcpy(&cache
->features
, &ca
->features
, sizeof(cache
->features
));
1894 cache
->callbacks
.congested_fn
= cache_is_congested
;
1895 dm_table_add_target_callbacks(ti
->table
, &cache
->callbacks
);
1897 cache
->metadata_dev
= ca
->metadata_dev
;
1898 cache
->origin_dev
= ca
->origin_dev
;
1899 cache
->cache_dev
= ca
->cache_dev
;
1901 ca
->metadata_dev
= ca
->origin_dev
= ca
->cache_dev
= NULL
;
1903 /* FIXME: factor out this whole section */
1904 origin_blocks
= cache
->origin_sectors
= ca
->origin_sectors
;
1905 origin_blocks
= block_div(origin_blocks
, ca
->block_size
);
1906 cache
->origin_blocks
= to_oblock(origin_blocks
);
1908 cache
->sectors_per_block
= ca
->block_size
;
1909 if (dm_set_target_max_io_len(ti
, cache
->sectors_per_block
)) {
1914 if (ca
->block_size
& (ca
->block_size
- 1)) {
1915 dm_block_t cache_size
= ca
->cache_sectors
;
1917 cache
->sectors_per_block_shift
= -1;
1918 cache_size
= block_div(cache_size
, ca
->block_size
);
1919 cache
->cache_size
= to_cblock(cache_size
);
1921 cache
->sectors_per_block_shift
= __ffs(ca
->block_size
);
1922 cache
->cache_size
= to_cblock(ca
->cache_sectors
>> cache
->sectors_per_block_shift
);
1925 r
= create_cache_policy(cache
, ca
, error
);
1928 cache
->policy_nr_args
= ca
->policy_argc
;
1930 cmd
= dm_cache_metadata_open(cache
->metadata_dev
->bdev
,
1931 ca
->block_size
, may_format
,
1932 dm_cache_policy_get_hint_size(cache
->policy
));
1934 *error
= "Error creating metadata object";
1940 spin_lock_init(&cache
->lock
);
1941 bio_list_init(&cache
->deferred_bios
);
1942 bio_list_init(&cache
->deferred_flush_bios
);
1943 bio_list_init(&cache
->deferred_writethrough_bios
);
1944 INIT_LIST_HEAD(&cache
->quiesced_migrations
);
1945 INIT_LIST_HEAD(&cache
->completed_migrations
);
1946 INIT_LIST_HEAD(&cache
->need_commit_migrations
);
1947 cache
->migration_threshold
= DEFAULT_MIGRATION_THRESHOLD
;
1948 atomic_set(&cache
->nr_migrations
, 0);
1949 init_waitqueue_head(&cache
->migration_wait
);
1951 cache
->nr_dirty
= 0;
1952 cache
->dirty_bitset
= alloc_bitset(from_cblock(cache
->cache_size
));
1953 if (!cache
->dirty_bitset
) {
1954 *error
= "could not allocate dirty bitset";
1957 clear_bitset(cache
->dirty_bitset
, from_cblock(cache
->cache_size
));
1959 cache
->discard_block_size
=
1960 calculate_discard_block_size(cache
->sectors_per_block
,
1961 cache
->origin_sectors
);
1962 cache
->discard_nr_blocks
= oblock_to_dblock(cache
, cache
->origin_blocks
);
1963 cache
->discard_bitset
= alloc_bitset(from_dblock(cache
->discard_nr_blocks
));
1964 if (!cache
->discard_bitset
) {
1965 *error
= "could not allocate discard bitset";
1968 clear_bitset(cache
->discard_bitset
, from_dblock(cache
->discard_nr_blocks
));
1970 cache
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1971 if (IS_ERR(cache
->copier
)) {
1972 *error
= "could not create kcopyd client";
1973 r
= PTR_ERR(cache
->copier
);
1977 cache
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1979 *error
= "could not create workqueue for metadata object";
1982 INIT_WORK(&cache
->worker
, do_worker
);
1983 INIT_DELAYED_WORK(&cache
->waker
, do_waker
);
1984 cache
->last_commit_jiffies
= jiffies
;
1986 cache
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1987 if (!cache
->prison
) {
1988 *error
= "could not create bio prison";
1992 cache
->all_io_ds
= dm_deferred_set_create();
1993 if (!cache
->all_io_ds
) {
1994 *error
= "could not create all_io deferred set";
1998 cache
->migration_pool
= mempool_create_slab_pool(MIGRATION_POOL_SIZE
,
2000 if (!cache
->migration_pool
) {
2001 *error
= "Error creating cache's migration mempool";
2005 cache
->next_migration
= NULL
;
2007 cache
->need_tick_bio
= true;
2008 cache
->sized
= false;
2009 cache
->quiescing
= false;
2010 cache
->commit_requested
= false;
2011 cache
->loaded_mappings
= false;
2012 cache
->loaded_discards
= false;
2016 atomic_set(&cache
->stats
.demotion
, 0);
2017 atomic_set(&cache
->stats
.promotion
, 0);
2018 atomic_set(&cache
->stats
.copies_avoided
, 0);
2019 atomic_set(&cache
->stats
.cache_cell_clash
, 0);
2020 atomic_set(&cache
->stats
.commit_count
, 0);
2021 atomic_set(&cache
->stats
.discard_count
, 0);
2031 static int copy_ctr_args(struct cache
*cache
, int argc
, const char **argv
)
2036 copy
= kcalloc(argc
, sizeof(*copy
), GFP_KERNEL
);
2039 for (i
= 0; i
< argc
; i
++) {
2040 copy
[i
] = kstrdup(argv
[i
], GFP_KERNEL
);
2049 cache
->nr_ctr_args
= argc
;
2050 cache
->ctr_args
= copy
;
2055 static int cache_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2058 struct cache_args
*ca
;
2059 struct cache
*cache
= NULL
;
2061 ca
= kzalloc(sizeof(*ca
), GFP_KERNEL
);
2063 ti
->error
= "Error allocating memory for cache";
2068 r
= parse_cache_args(ca
, argc
, argv
, &ti
->error
);
2072 r
= cache_create(ca
, &cache
);
2076 r
= copy_ctr_args(cache
, argc
- 3, (const char **)argv
+ 3);
2082 ti
->private = cache
;
2085 destroy_cache_args(ca
);
2089 static int cache_map(struct dm_target
*ti
, struct bio
*bio
)
2091 struct cache
*cache
= ti
->private;
2094 dm_oblock_t block
= get_bio_block(cache
, bio
);
2095 bool can_migrate
= false;
2096 bool discarded_block
;
2097 struct dm_bio_prison_cell
*cell
;
2098 struct policy_result lookup_result
;
2099 struct per_bio_data
*pb
;
2101 if (from_oblock(block
) > from_oblock(cache
->origin_blocks
)) {
2103 * This can only occur if the io goes to a partial block at
2104 * the end of the origin device. We don't cache these.
2105 * Just remap to the origin and carry on.
2107 remap_to_origin_clear_discard(cache
, bio
, block
);
2108 return DM_MAPIO_REMAPPED
;
2111 pb
= init_per_bio_data(bio
);
2113 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
| REQ_DISCARD
)) {
2114 defer_bio(cache
, bio
);
2115 return DM_MAPIO_SUBMITTED
;
2119 * Check to see if that block is currently migrating.
2121 cell
= alloc_prison_cell(cache
);
2123 defer_bio(cache
, bio
);
2124 return DM_MAPIO_SUBMITTED
;
2127 r
= bio_detain(cache
, block
, bio
, cell
,
2128 (cell_free_fn
) free_prison_cell
,
2132 defer_bio(cache
, bio
);
2134 return DM_MAPIO_SUBMITTED
;
2137 discarded_block
= is_discarded_oblock(cache
, block
);
2139 r
= policy_map(cache
->policy
, block
, false, can_migrate
, discarded_block
,
2140 bio
, &lookup_result
);
2141 if (r
== -EWOULDBLOCK
) {
2142 cell_defer(cache
, cell
, true);
2143 return DM_MAPIO_SUBMITTED
;
2146 DMERR_LIMIT("Unexpected return from cache replacement policy: %d", r
);
2148 return DM_MAPIO_SUBMITTED
;
2151 switch (lookup_result
.op
) {
2153 inc_hit_counter(cache
, bio
);
2154 pb
->all_io_entry
= dm_deferred_entry_inc(cache
->all_io_ds
);
2156 if (is_writethrough_io(cache
, bio
, lookup_result
.cblock
))
2157 remap_to_origin_then_cache(cache
, bio
, block
, lookup_result
.cblock
);
2159 remap_to_cache_dirty(cache
, bio
, block
, lookup_result
.cblock
);
2161 cell_defer(cache
, cell
, false);
2165 inc_miss_counter(cache
, bio
);
2166 pb
->all_io_entry
= dm_deferred_entry_inc(cache
->all_io_ds
);
2168 if (pb
->req_nr
!= 0) {
2170 * This is a duplicate writethrough io that is no
2171 * longer needed because the block has been demoted.
2174 cell_defer(cache
, cell
, false);
2175 return DM_MAPIO_SUBMITTED
;
2177 remap_to_origin_clear_discard(cache
, bio
, block
);
2178 cell_defer(cache
, cell
, false);
2183 DMERR_LIMIT("%s: erroring bio: unknown policy op: %u", __func__
,
2184 (unsigned) lookup_result
.op
);
2186 return DM_MAPIO_SUBMITTED
;
2189 return DM_MAPIO_REMAPPED
;
2192 static int cache_end_io(struct dm_target
*ti
, struct bio
*bio
, int error
)
2194 struct cache
*cache
= ti
->private;
2195 unsigned long flags
;
2196 struct per_bio_data
*pb
= get_per_bio_data(bio
);
2199 policy_tick(cache
->policy
);
2201 spin_lock_irqsave(&cache
->lock
, flags
);
2202 cache
->need_tick_bio
= true;
2203 spin_unlock_irqrestore(&cache
->lock
, flags
);
2206 check_for_quiesced_migrations(cache
, pb
);
2211 static int write_dirty_bitset(struct cache
*cache
)
2215 for (i
= 0; i
< from_cblock(cache
->cache_size
); i
++) {
2216 r
= dm_cache_set_dirty(cache
->cmd
, to_cblock(i
),
2217 is_dirty(cache
, to_cblock(i
)));
2225 static int write_discard_bitset(struct cache
*cache
)
2229 r
= dm_cache_discard_bitset_resize(cache
->cmd
, cache
->discard_block_size
,
2230 cache
->discard_nr_blocks
);
2232 DMERR("could not resize on-disk discard bitset");
2236 for (i
= 0; i
< from_dblock(cache
->discard_nr_blocks
); i
++) {
2237 r
= dm_cache_set_discard(cache
->cmd
, to_dblock(i
),
2238 is_discarded(cache
, to_dblock(i
)));
2246 static int save_hint(void *context
, dm_cblock_t cblock
, dm_oblock_t oblock
,
2249 struct cache
*cache
= context
;
2250 return dm_cache_save_hint(cache
->cmd
, cblock
, hint
);
2253 static int write_hints(struct cache
*cache
)
2257 r
= dm_cache_begin_hints(cache
->cmd
, cache
->policy
);
2259 DMERR("dm_cache_begin_hints failed");
2263 r
= policy_walk_mappings(cache
->policy
, save_hint
, cache
);
2265 DMERR("policy_walk_mappings failed");
2271 * returns true on success
2273 static bool sync_metadata(struct cache
*cache
)
2277 r1
= write_dirty_bitset(cache
);
2279 DMERR("could not write dirty bitset");
2281 r2
= write_discard_bitset(cache
);
2283 DMERR("could not write discard bitset");
2287 r3
= write_hints(cache
);
2289 DMERR("could not write hints");
2292 * If writing the above metadata failed, we still commit, but don't
2293 * set the clean shutdown flag. This will effectively force every
2294 * dirty bit to be set on reload.
2296 r4
= dm_cache_commit(cache
->cmd
, !r1
&& !r2
&& !r3
);
2298 DMERR("could not write cache metadata. Data loss may occur.");
2300 return !r1
&& !r2
&& !r3
&& !r4
;
2303 static void cache_postsuspend(struct dm_target
*ti
)
2305 struct cache
*cache
= ti
->private;
2307 start_quiescing(cache
);
2308 wait_for_migrations(cache
);
2310 requeue_deferred_io(cache
);
2311 stop_quiescing(cache
);
2313 (void) sync_metadata(cache
);
2316 static int load_mapping(void *context
, dm_oblock_t oblock
, dm_cblock_t cblock
,
2317 bool dirty
, uint32_t hint
, bool hint_valid
)
2320 struct cache
*cache
= context
;
2322 r
= policy_load_mapping(cache
->policy
, oblock
, cblock
, hint
, hint_valid
);
2327 set_dirty(cache
, oblock
, cblock
);
2329 clear_dirty(cache
, oblock
, cblock
);
2334 static int load_discard(void *context
, sector_t discard_block_size
,
2335 dm_dblock_t dblock
, bool discard
)
2337 struct cache
*cache
= context
;
2339 /* FIXME: handle mis-matched block size */
2342 set_discard(cache
, dblock
);
2344 clear_discard(cache
, dblock
);
2349 static int cache_preresume(struct dm_target
*ti
)
2352 struct cache
*cache
= ti
->private;
2353 sector_t actual_cache_size
= get_dev_size(cache
->cache_dev
);
2354 (void) sector_div(actual_cache_size
, cache
->sectors_per_block
);
2357 * Check to see if the cache has resized.
2359 if (from_cblock(cache
->cache_size
) != actual_cache_size
|| !cache
->sized
) {
2360 cache
->cache_size
= to_cblock(actual_cache_size
);
2362 r
= dm_cache_resize(cache
->cmd
, cache
->cache_size
);
2364 DMERR("could not resize cache metadata");
2368 cache
->sized
= true;
2371 if (!cache
->loaded_mappings
) {
2372 r
= dm_cache_load_mappings(cache
->cmd
, cache
->policy
,
2373 load_mapping
, cache
);
2375 DMERR("could not load cache mappings");
2379 cache
->loaded_mappings
= true;
2382 if (!cache
->loaded_discards
) {
2383 r
= dm_cache_load_discards(cache
->cmd
, load_discard
, cache
);
2385 DMERR("could not load origin discards");
2389 cache
->loaded_discards
= true;
2395 static void cache_resume(struct dm_target
*ti
)
2397 struct cache
*cache
= ti
->private;
2399 cache
->need_tick_bio
= true;
2400 do_waker(&cache
->waker
.work
);
2406 * <#used metadata blocks>/<#total metadata blocks>
2407 * <#read hits> <#read misses> <#write hits> <#write misses>
2408 * <#demotions> <#promotions> <#blocks in cache> <#dirty>
2409 * <#features> <features>*
2410 * <#core args> <core args>
2411 * <#policy args> <policy args>*
2413 static void cache_status(struct dm_target
*ti
, status_type_t type
,
2414 unsigned status_flags
, char *result
, unsigned maxlen
)
2419 dm_block_t nr_free_blocks_metadata
= 0;
2420 dm_block_t nr_blocks_metadata
= 0;
2421 char buf
[BDEVNAME_SIZE
];
2422 struct cache
*cache
= ti
->private;
2423 dm_cblock_t residency
;
2426 case STATUSTYPE_INFO
:
2427 /* Commit to ensure statistics aren't out-of-date */
2428 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
)) {
2429 r
= dm_cache_commit(cache
->cmd
, false);
2431 DMERR("could not commit metadata for accurate status");
2434 r
= dm_cache_get_free_metadata_block_count(cache
->cmd
,
2435 &nr_free_blocks_metadata
);
2437 DMERR("could not get metadata free block count");
2441 r
= dm_cache_get_metadata_dev_size(cache
->cmd
, &nr_blocks_metadata
);
2443 DMERR("could not get metadata device size");
2447 residency
= policy_residency(cache
->policy
);
2449 DMEMIT("%llu/%llu %u %u %u %u %u %u %llu %u ",
2450 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2451 (unsigned long long)nr_blocks_metadata
,
2452 (unsigned) atomic_read(&cache
->stats
.read_hit
),
2453 (unsigned) atomic_read(&cache
->stats
.read_miss
),
2454 (unsigned) atomic_read(&cache
->stats
.write_hit
),
2455 (unsigned) atomic_read(&cache
->stats
.write_miss
),
2456 (unsigned) atomic_read(&cache
->stats
.demotion
),
2457 (unsigned) atomic_read(&cache
->stats
.promotion
),
2458 (unsigned long long) from_cblock(residency
),
2461 if (cache
->features
.write_through
)
2462 DMEMIT("1 writethrough ");
2466 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache
->migration_threshold
);
2468 r
= policy_emit_config_values(cache
->policy
, result
+ sz
, maxlen
- sz
);
2470 DMERR("policy_emit_config_values returned %d", r
);
2475 case STATUSTYPE_TABLE
:
2476 format_dev_t(buf
, cache
->metadata_dev
->bdev
->bd_dev
);
2478 format_dev_t(buf
, cache
->cache_dev
->bdev
->bd_dev
);
2480 format_dev_t(buf
, cache
->origin_dev
->bdev
->bd_dev
);
2483 for (i
= 0; i
< cache
->nr_ctr_args
- 1; i
++)
2484 DMEMIT(" %s", cache
->ctr_args
[i
]);
2485 if (cache
->nr_ctr_args
)
2486 DMEMIT(" %s", cache
->ctr_args
[cache
->nr_ctr_args
- 1]);
2495 #define NOT_CORE_OPTION 1
2497 static int process_config_option(struct cache
*cache
, char **argv
)
2501 if (!strcasecmp(argv
[0], "migration_threshold")) {
2502 if (kstrtoul(argv
[1], 10, &tmp
))
2505 cache
->migration_threshold
= tmp
;
2509 return NOT_CORE_OPTION
;
2513 * Supports <key> <value>.
2515 * The key migration_threshold is supported by the cache target core.
2517 static int cache_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2520 struct cache
*cache
= ti
->private;
2525 r
= process_config_option(cache
, argv
);
2526 if (r
== NOT_CORE_OPTION
)
2527 return policy_set_config_value(cache
->policy
, argv
[0], argv
[1]);
2532 static int cache_iterate_devices(struct dm_target
*ti
,
2533 iterate_devices_callout_fn fn
, void *data
)
2536 struct cache
*cache
= ti
->private;
2538 r
= fn(ti
, cache
->cache_dev
, 0, get_dev_size(cache
->cache_dev
), data
);
2540 r
= fn(ti
, cache
->origin_dev
, 0, ti
->len
, data
);
2546 * We assume I/O is going to the origin (which is the volume
2547 * more likely to have restrictions e.g. by being striped).
2548 * (Looking up the exact location of the data would be expensive
2549 * and could always be out of date by the time the bio is submitted.)
2551 static int cache_bvec_merge(struct dm_target
*ti
,
2552 struct bvec_merge_data
*bvm
,
2553 struct bio_vec
*biovec
, int max_size
)
2555 struct cache
*cache
= ti
->private;
2556 struct request_queue
*q
= bdev_get_queue(cache
->origin_dev
->bdev
);
2558 if (!q
->merge_bvec_fn
)
2561 bvm
->bi_bdev
= cache
->origin_dev
->bdev
;
2562 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2565 static void set_discard_limits(struct cache
*cache
, struct queue_limits
*limits
)
2568 * FIXME: these limits may be incompatible with the cache device
2570 limits
->max_discard_sectors
= cache
->discard_block_size
* 1024;
2571 limits
->discard_granularity
= cache
->discard_block_size
<< SECTOR_SHIFT
;
2574 static void cache_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2576 struct cache
*cache
= ti
->private;
2578 blk_limits_io_min(limits
, 0);
2579 blk_limits_io_opt(limits
, cache
->sectors_per_block
<< SECTOR_SHIFT
);
2580 set_discard_limits(cache
, limits
);
2583 /*----------------------------------------------------------------*/
2585 static struct target_type cache_target
= {
2587 .version
= {1, 1, 0},
2588 .module
= THIS_MODULE
,
2592 .end_io
= cache_end_io
,
2593 .postsuspend
= cache_postsuspend
,
2594 .preresume
= cache_preresume
,
2595 .resume
= cache_resume
,
2596 .status
= cache_status
,
2597 .message
= cache_message
,
2598 .iterate_devices
= cache_iterate_devices
,
2599 .merge
= cache_bvec_merge
,
2600 .io_hints
= cache_io_hints
,
2603 static int __init
dm_cache_init(void)
2607 r
= dm_register_target(&cache_target
);
2609 DMERR("cache target registration failed: %d", r
);
2613 migration_cache
= KMEM_CACHE(dm_cache_migration
, 0);
2614 if (!migration_cache
) {
2615 dm_unregister_target(&cache_target
);
2622 static void __exit
dm_cache_exit(void)
2624 dm_unregister_target(&cache_target
);
2625 kmem_cache_destroy(migration_cache
);
2628 module_init(dm_cache_init
);
2629 module_exit(dm_cache_exit
);
2631 MODULE_DESCRIPTION(DM_NAME
" cache target");
2632 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
2633 MODULE_LICENSE("GPL");