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-bio-record.h"
10 #include "dm-cache-metadata.h"
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/jiffies.h>
15 #include <linux/init.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
21 #define DM_MSG_PREFIX "cache"
23 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle
,
24 "A percentage of time allocated for copying to and/or from cache");
26 /*----------------------------------------------------------------*/
28 #define IOT_RESOLUTION 4
34 * Sectors of in-flight IO.
39 * The time, in jiffies, when this device became idle (if it is
42 unsigned long idle_time
;
43 unsigned long last_update_time
;
46 static void iot_init(struct io_tracker
*iot
)
48 spin_lock_init(&iot
->lock
);
51 iot
->last_update_time
= jiffies
;
54 static bool __iot_idle_for(struct io_tracker
*iot
, unsigned long jifs
)
59 return time_after(jiffies
, iot
->idle_time
+ jifs
);
62 static bool iot_idle_for(struct io_tracker
*iot
, unsigned long jifs
)
67 spin_lock_irqsave(&iot
->lock
, flags
);
68 r
= __iot_idle_for(iot
, jifs
);
69 spin_unlock_irqrestore(&iot
->lock
, flags
);
74 static void iot_io_begin(struct io_tracker
*iot
, sector_t len
)
78 spin_lock_irqsave(&iot
->lock
, flags
);
79 iot
->in_flight
+= len
;
80 spin_unlock_irqrestore(&iot
->lock
, flags
);
83 static void __iot_io_end(struct io_tracker
*iot
, sector_t len
)
85 iot
->in_flight
-= len
;
87 iot
->idle_time
= jiffies
;
90 static void iot_io_end(struct io_tracker
*iot
, sector_t len
)
94 spin_lock_irqsave(&iot
->lock
, flags
);
95 __iot_io_end(iot
, len
);
96 spin_unlock_irqrestore(&iot
->lock
, flags
);
99 /*----------------------------------------------------------------*/
104 * oblock: index of an origin block
105 * cblock: index of a cache block
106 * promotion: movement of a block from origin to cache
107 * demotion: movement of a block from cache to origin
108 * migration: movement of a block between the origin and cache device,
112 /*----------------------------------------------------------------*/
115 * There are a couple of places where we let a bio run, but want to do some
116 * work before calling its endio function. We do this by temporarily
117 * changing the endio fn.
119 struct dm_hook_info
{
120 bio_end_io_t
*bi_end_io
;
123 static void dm_hook_bio(struct dm_hook_info
*h
, struct bio
*bio
,
124 bio_end_io_t
*bi_end_io
, void *bi_private
)
126 h
->bi_end_io
= bio
->bi_end_io
;
128 bio
->bi_end_io
= bi_end_io
;
129 bio
->bi_private
= bi_private
;
132 static void dm_unhook_bio(struct dm_hook_info
*h
, struct bio
*bio
)
134 bio
->bi_end_io
= h
->bi_end_io
;
137 /*----------------------------------------------------------------*/
139 #define MIGRATION_POOL_SIZE 128
140 #define COMMIT_PERIOD HZ
141 #define MIGRATION_COUNT_WINDOW 10
144 * The block size of the device holding cache data must be
145 * between 32KB and 1GB.
147 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT)
148 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
150 enum cache_metadata_mode
{
151 CM_WRITE
, /* metadata may be changed */
152 CM_READ_ONLY
, /* metadata may not be changed */
158 * Data is written to cached blocks only. These blocks are marked
159 * dirty. If you lose the cache device you will lose data.
160 * Potential performance increase for both reads and writes.
165 * Data is written to both cache and origin. Blocks are never
166 * dirty. Potential performance benfit for reads only.
171 * A degraded mode useful for various cache coherency situations
172 * (eg, rolling back snapshots). Reads and writes always go to the
173 * origin. If a write goes to a cached oblock, then the cache
174 * block is invalidated.
179 struct cache_features
{
180 enum cache_metadata_mode mode
;
181 enum cache_io_mode io_mode
;
191 atomic_t copies_avoided
;
192 atomic_t cache_cell_clash
;
193 atomic_t commit_count
;
194 atomic_t discard_count
;
198 * Defines a range of cblocks, begin to (end - 1) are in the range. end is
199 * the one-past-the-end value.
201 struct cblock_range
{
206 struct invalidation_request
{
207 struct list_head list
;
208 struct cblock_range
*cblocks
;
213 wait_queue_head_t result_wait
;
217 struct dm_target
*ti
;
218 struct dm_target_callbacks callbacks
;
220 struct dm_cache_metadata
*cmd
;
223 * Metadata is written to this device.
225 struct dm_dev
*metadata_dev
;
228 * The slower of the two data devices. Typically a spindle.
230 struct dm_dev
*origin_dev
;
233 * The faster of the two data devices. Typically an SSD.
235 struct dm_dev
*cache_dev
;
238 * Size of the origin device in _complete_ blocks and native sectors.
240 dm_oblock_t origin_blocks
;
241 sector_t origin_sectors
;
244 * Size of the cache device in blocks.
246 dm_cblock_t cache_size
;
249 * Fields for converting from sectors to blocks.
251 uint32_t sectors_per_block
;
252 int sectors_per_block_shift
;
255 struct list_head deferred_cells
;
256 struct bio_list deferred_bios
;
257 struct bio_list deferred_flush_bios
;
258 struct bio_list deferred_writethrough_bios
;
259 struct list_head quiesced_migrations
;
260 struct list_head completed_migrations
;
261 struct list_head need_commit_migrations
;
262 sector_t migration_threshold
;
263 wait_queue_head_t migration_wait
;
264 atomic_t nr_allocated_migrations
;
267 * The number of in flight migrations that are performing
268 * background io. eg, promotion, writeback.
270 atomic_t nr_io_migrations
;
272 wait_queue_head_t quiescing_wait
;
274 atomic_t quiescing_ack
;
277 * cache_size entries, dirty if set
280 unsigned long *dirty_bitset
;
283 * origin_blocks entries, discarded if set.
285 dm_dblock_t discard_nr_blocks
;
286 unsigned long *discard_bitset
;
287 uint32_t discard_block_size
; /* a power of 2 times sectors per block */
290 * Rather than reconstructing the table line for the status we just
291 * save it and regurgitate.
293 unsigned nr_ctr_args
;
294 const char **ctr_args
;
296 struct dm_kcopyd_client
*copier
;
297 struct workqueue_struct
*wq
;
298 struct work_struct worker
;
300 struct delayed_work waker
;
301 unsigned long last_commit_jiffies
;
303 struct dm_bio_prison
*prison
;
304 struct dm_deferred_set
*all_io_ds
;
306 mempool_t
*migration_pool
;
308 struct dm_cache_policy
*policy
;
309 unsigned policy_nr_args
;
311 bool need_tick_bio
:1;
314 bool commit_requested
:1;
315 bool loaded_mappings
:1;
316 bool loaded_discards
:1;
319 * Cache features such as write-through.
321 struct cache_features features
;
323 struct cache_stats stats
;
326 * Invalidation fields.
328 spinlock_t invalidation_lock
;
329 struct list_head invalidation_requests
;
331 struct io_tracker origin_tracker
;
334 struct per_bio_data
{
337 struct dm_deferred_entry
*all_io_entry
;
338 struct dm_hook_info hook_info
;
342 * writethrough fields. These MUST remain at the end of this
343 * structure and the 'cache' member must be the first as it
344 * is used to determine the offset of the writethrough fields.
348 struct dm_bio_details bio_details
;
351 struct dm_cache_migration
{
352 struct list_head list
;
355 unsigned long start_jiffies
;
356 dm_oblock_t old_oblock
;
357 dm_oblock_t new_oblock
;
365 bool requeue_holder
:1;
368 struct dm_bio_prison_cell
*old_ocell
;
369 struct dm_bio_prison_cell
*new_ocell
;
373 * Processing a bio in the worker thread may require these memory
374 * allocations. We prealloc to avoid deadlocks (the same worker thread
375 * frees them back to the mempool).
378 struct dm_cache_migration
*mg
;
379 struct dm_bio_prison_cell
*cell1
;
380 struct dm_bio_prison_cell
*cell2
;
383 static enum cache_metadata_mode
get_cache_mode(struct cache
*cache
);
385 static void wake_worker(struct cache
*cache
)
387 queue_work(cache
->wq
, &cache
->worker
);
390 /*----------------------------------------------------------------*/
392 static struct dm_bio_prison_cell
*alloc_prison_cell(struct cache
*cache
)
394 /* FIXME: change to use a local slab. */
395 return dm_bio_prison_alloc_cell(cache
->prison
, GFP_NOWAIT
);
398 static void free_prison_cell(struct cache
*cache
, struct dm_bio_prison_cell
*cell
)
400 dm_bio_prison_free_cell(cache
->prison
, cell
);
403 static struct dm_cache_migration
*alloc_migration(struct cache
*cache
)
405 struct dm_cache_migration
*mg
;
407 mg
= mempool_alloc(cache
->migration_pool
, GFP_NOWAIT
);
410 atomic_inc(&mg
->cache
->nr_allocated_migrations
);
416 static void free_migration(struct dm_cache_migration
*mg
)
418 struct cache
*cache
= mg
->cache
;
420 if (atomic_dec_and_test(&cache
->nr_allocated_migrations
))
421 wake_up(&cache
->migration_wait
);
423 mempool_free(mg
, cache
->migration_pool
);
426 static int prealloc_data_structs(struct cache
*cache
, struct prealloc
*p
)
429 p
->mg
= alloc_migration(cache
);
435 p
->cell1
= alloc_prison_cell(cache
);
441 p
->cell2
= alloc_prison_cell(cache
);
449 static void prealloc_free_structs(struct cache
*cache
, struct prealloc
*p
)
452 free_prison_cell(cache
, p
->cell2
);
455 free_prison_cell(cache
, p
->cell1
);
458 free_migration(p
->mg
);
461 static struct dm_cache_migration
*prealloc_get_migration(struct prealloc
*p
)
463 struct dm_cache_migration
*mg
= p
->mg
;
472 * You must have a cell within the prealloc struct to return. If not this
473 * function will BUG() rather than returning NULL.
475 static struct dm_bio_prison_cell
*prealloc_get_cell(struct prealloc
*p
)
477 struct dm_bio_prison_cell
*r
= NULL
;
483 } else if (p
->cell2
) {
493 * You can't have more than two cells in a prealloc struct. BUG() will be
494 * called if you try and overfill.
496 static void prealloc_put_cell(struct prealloc
*p
, struct dm_bio_prison_cell
*cell
)
508 /*----------------------------------------------------------------*/
510 static void build_key(dm_oblock_t begin
, dm_oblock_t end
, struct dm_cell_key
*key
)
514 key
->block_begin
= from_oblock(begin
);
515 key
->block_end
= from_oblock(end
);
519 * The caller hands in a preallocated cell, and a free function for it.
520 * The cell will be freed if there's an error, or if it wasn't used because
521 * a cell with that key already exists.
523 typedef void (*cell_free_fn
)(void *context
, struct dm_bio_prison_cell
*cell
);
525 static int bio_detain_range(struct cache
*cache
, dm_oblock_t oblock_begin
, dm_oblock_t oblock_end
,
526 struct bio
*bio
, struct dm_bio_prison_cell
*cell_prealloc
,
527 cell_free_fn free_fn
, void *free_context
,
528 struct dm_bio_prison_cell
**cell_result
)
531 struct dm_cell_key key
;
533 build_key(oblock_begin
, oblock_end
, &key
);
534 r
= dm_bio_detain(cache
->prison
, &key
, bio
, cell_prealloc
, cell_result
);
536 free_fn(free_context
, cell_prealloc
);
541 static int bio_detain(struct cache
*cache
, dm_oblock_t oblock
,
542 struct bio
*bio
, struct dm_bio_prison_cell
*cell_prealloc
,
543 cell_free_fn free_fn
, void *free_context
,
544 struct dm_bio_prison_cell
**cell_result
)
546 dm_oblock_t end
= to_oblock(from_oblock(oblock
) + 1ULL);
547 return bio_detain_range(cache
, oblock
, end
, bio
,
548 cell_prealloc
, free_fn
, free_context
, cell_result
);
551 static int get_cell(struct cache
*cache
,
553 struct prealloc
*structs
,
554 struct dm_bio_prison_cell
**cell_result
)
557 struct dm_cell_key key
;
558 struct dm_bio_prison_cell
*cell_prealloc
;
560 cell_prealloc
= prealloc_get_cell(structs
);
562 build_key(oblock
, to_oblock(from_oblock(oblock
) + 1ULL), &key
);
563 r
= dm_get_cell(cache
->prison
, &key
, cell_prealloc
, cell_result
);
565 prealloc_put_cell(structs
, cell_prealloc
);
570 /*----------------------------------------------------------------*/
572 static bool is_dirty(struct cache
*cache
, dm_cblock_t b
)
574 return test_bit(from_cblock(b
), cache
->dirty_bitset
);
577 static void set_dirty(struct cache
*cache
, dm_oblock_t oblock
, dm_cblock_t cblock
)
579 if (!test_and_set_bit(from_cblock(cblock
), cache
->dirty_bitset
)) {
580 atomic_inc(&cache
->nr_dirty
);
581 policy_set_dirty(cache
->policy
, oblock
);
585 static void clear_dirty(struct cache
*cache
, dm_oblock_t oblock
, dm_cblock_t cblock
)
587 if (test_and_clear_bit(from_cblock(cblock
), cache
->dirty_bitset
)) {
588 policy_clear_dirty(cache
->policy
, oblock
);
589 if (atomic_dec_return(&cache
->nr_dirty
) == 0)
590 dm_table_event(cache
->ti
->table
);
594 /*----------------------------------------------------------------*/
596 static bool block_size_is_power_of_two(struct cache
*cache
)
598 return cache
->sectors_per_block_shift
>= 0;
601 /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */
602 #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6
605 static dm_block_t
block_div(dm_block_t b
, uint32_t n
)
612 static dm_block_t
oblocks_per_dblock(struct cache
*cache
)
614 dm_block_t oblocks
= cache
->discard_block_size
;
616 if (block_size_is_power_of_two(cache
))
617 oblocks
>>= cache
->sectors_per_block_shift
;
619 oblocks
= block_div(oblocks
, cache
->sectors_per_block
);
624 static dm_dblock_t
oblock_to_dblock(struct cache
*cache
, dm_oblock_t oblock
)
626 return to_dblock(block_div(from_oblock(oblock
),
627 oblocks_per_dblock(cache
)));
630 static dm_oblock_t
dblock_to_oblock(struct cache
*cache
, dm_dblock_t dblock
)
632 return to_oblock(from_dblock(dblock
) * oblocks_per_dblock(cache
));
635 static void set_discard(struct cache
*cache
, dm_dblock_t b
)
639 BUG_ON(from_dblock(b
) >= from_dblock(cache
->discard_nr_blocks
));
640 atomic_inc(&cache
->stats
.discard_count
);
642 spin_lock_irqsave(&cache
->lock
, flags
);
643 set_bit(from_dblock(b
), cache
->discard_bitset
);
644 spin_unlock_irqrestore(&cache
->lock
, flags
);
647 static void clear_discard(struct cache
*cache
, dm_dblock_t b
)
651 spin_lock_irqsave(&cache
->lock
, flags
);
652 clear_bit(from_dblock(b
), cache
->discard_bitset
);
653 spin_unlock_irqrestore(&cache
->lock
, flags
);
656 static bool is_discarded(struct cache
*cache
, dm_dblock_t b
)
661 spin_lock_irqsave(&cache
->lock
, flags
);
662 r
= test_bit(from_dblock(b
), cache
->discard_bitset
);
663 spin_unlock_irqrestore(&cache
->lock
, flags
);
668 static bool is_discarded_oblock(struct cache
*cache
, dm_oblock_t b
)
673 spin_lock_irqsave(&cache
->lock
, flags
);
674 r
= test_bit(from_dblock(oblock_to_dblock(cache
, b
)),
675 cache
->discard_bitset
);
676 spin_unlock_irqrestore(&cache
->lock
, flags
);
681 /*----------------------------------------------------------------*/
683 static void load_stats(struct cache
*cache
)
685 struct dm_cache_statistics stats
;
687 dm_cache_metadata_get_stats(cache
->cmd
, &stats
);
688 atomic_set(&cache
->stats
.read_hit
, stats
.read_hits
);
689 atomic_set(&cache
->stats
.read_miss
, stats
.read_misses
);
690 atomic_set(&cache
->stats
.write_hit
, stats
.write_hits
);
691 atomic_set(&cache
->stats
.write_miss
, stats
.write_misses
);
694 static void save_stats(struct cache
*cache
)
696 struct dm_cache_statistics stats
;
698 if (get_cache_mode(cache
) >= CM_READ_ONLY
)
701 stats
.read_hits
= atomic_read(&cache
->stats
.read_hit
);
702 stats
.read_misses
= atomic_read(&cache
->stats
.read_miss
);
703 stats
.write_hits
= atomic_read(&cache
->stats
.write_hit
);
704 stats
.write_misses
= atomic_read(&cache
->stats
.write_miss
);
706 dm_cache_metadata_set_stats(cache
->cmd
, &stats
);
709 /*----------------------------------------------------------------
711 *--------------------------------------------------------------*/
714 * If using writeback, leave out struct per_bio_data's writethrough fields.
716 #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache))
717 #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data))
719 static bool writethrough_mode(struct cache_features
*f
)
721 return f
->io_mode
== CM_IO_WRITETHROUGH
;
724 static bool writeback_mode(struct cache_features
*f
)
726 return f
->io_mode
== CM_IO_WRITEBACK
;
729 static bool passthrough_mode(struct cache_features
*f
)
731 return f
->io_mode
== CM_IO_PASSTHROUGH
;
734 static size_t get_per_bio_data_size(struct cache
*cache
)
736 return writethrough_mode(&cache
->features
) ? PB_DATA_SIZE_WT
: PB_DATA_SIZE_WB
;
739 static struct per_bio_data
*get_per_bio_data(struct bio
*bio
, size_t data_size
)
741 struct per_bio_data
*pb
= dm_per_bio_data(bio
, data_size
);
746 static struct per_bio_data
*init_per_bio_data(struct bio
*bio
, size_t data_size
)
748 struct per_bio_data
*pb
= get_per_bio_data(bio
, data_size
);
751 pb
->req_nr
= dm_bio_get_target_bio_nr(bio
);
752 pb
->all_io_entry
= NULL
;
758 /*----------------------------------------------------------------
760 *--------------------------------------------------------------*/
761 static void remap_to_origin(struct cache
*cache
, struct bio
*bio
)
763 bio
->bi_bdev
= cache
->origin_dev
->bdev
;
766 static void remap_to_cache(struct cache
*cache
, struct bio
*bio
,
769 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
770 sector_t block
= from_cblock(cblock
);
772 bio
->bi_bdev
= cache
->cache_dev
->bdev
;
773 if (!block_size_is_power_of_two(cache
))
774 bio
->bi_iter
.bi_sector
=
775 (block
* cache
->sectors_per_block
) +
776 sector_div(bi_sector
, cache
->sectors_per_block
);
778 bio
->bi_iter
.bi_sector
=
779 (block
<< cache
->sectors_per_block_shift
) |
780 (bi_sector
& (cache
->sectors_per_block
- 1));
783 static void check_if_tick_bio_needed(struct cache
*cache
, struct bio
*bio
)
786 size_t pb_data_size
= get_per_bio_data_size(cache
);
787 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
789 spin_lock_irqsave(&cache
->lock
, flags
);
790 if (cache
->need_tick_bio
&& !op_is_flush(bio
->bi_opf
) &&
791 bio_op(bio
) != REQ_OP_DISCARD
) {
793 cache
->need_tick_bio
= false;
795 spin_unlock_irqrestore(&cache
->lock
, flags
);
798 static void remap_to_origin_clear_discard(struct cache
*cache
, struct bio
*bio
,
801 check_if_tick_bio_needed(cache
, bio
);
802 remap_to_origin(cache
, bio
);
803 if (bio_data_dir(bio
) == WRITE
)
804 clear_discard(cache
, oblock_to_dblock(cache
, oblock
));
807 static void remap_to_cache_dirty(struct cache
*cache
, struct bio
*bio
,
808 dm_oblock_t oblock
, dm_cblock_t cblock
)
810 check_if_tick_bio_needed(cache
, bio
);
811 remap_to_cache(cache
, bio
, cblock
);
812 if (bio_data_dir(bio
) == WRITE
) {
813 set_dirty(cache
, oblock
, cblock
);
814 clear_discard(cache
, oblock_to_dblock(cache
, oblock
));
818 static dm_oblock_t
get_bio_block(struct cache
*cache
, struct bio
*bio
)
820 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
822 if (!block_size_is_power_of_two(cache
))
823 (void) sector_div(block_nr
, cache
->sectors_per_block
);
825 block_nr
>>= cache
->sectors_per_block_shift
;
827 return to_oblock(block_nr
);
831 * You must increment the deferred set whilst the prison cell is held. To
832 * encourage this, we ask for 'cell' to be passed in.
834 static void inc_ds(struct cache
*cache
, struct bio
*bio
,
835 struct dm_bio_prison_cell
*cell
)
837 size_t pb_data_size
= get_per_bio_data_size(cache
);
838 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
841 BUG_ON(pb
->all_io_entry
);
843 pb
->all_io_entry
= dm_deferred_entry_inc(cache
->all_io_ds
);
846 static bool accountable_bio(struct cache
*cache
, struct bio
*bio
)
848 return ((bio
->bi_bdev
== cache
->origin_dev
->bdev
) &&
849 bio_op(bio
) != REQ_OP_DISCARD
);
852 static void accounted_begin(struct cache
*cache
, struct bio
*bio
)
854 size_t pb_data_size
= get_per_bio_data_size(cache
);
855 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
857 if (accountable_bio(cache
, bio
)) {
858 pb
->len
= bio_sectors(bio
);
859 iot_io_begin(&cache
->origin_tracker
, pb
->len
);
863 static void accounted_complete(struct cache
*cache
, struct bio
*bio
)
865 size_t pb_data_size
= get_per_bio_data_size(cache
);
866 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
868 iot_io_end(&cache
->origin_tracker
, pb
->len
);
871 static void accounted_request(struct cache
*cache
, struct bio
*bio
)
873 accounted_begin(cache
, bio
);
874 generic_make_request(bio
);
877 static void issue(struct cache
*cache
, struct bio
*bio
)
881 if (!op_is_flush(bio
->bi_opf
)) {
882 accounted_request(cache
, bio
);
887 * Batch together any bios that trigger commits and then issue a
888 * single commit for them in do_worker().
890 spin_lock_irqsave(&cache
->lock
, flags
);
891 cache
->commit_requested
= true;
892 bio_list_add(&cache
->deferred_flush_bios
, bio
);
893 spin_unlock_irqrestore(&cache
->lock
, flags
);
896 static void inc_and_issue(struct cache
*cache
, struct bio
*bio
, struct dm_bio_prison_cell
*cell
)
898 inc_ds(cache
, bio
, cell
);
902 static void defer_writethrough_bio(struct cache
*cache
, struct bio
*bio
)
906 spin_lock_irqsave(&cache
->lock
, flags
);
907 bio_list_add(&cache
->deferred_writethrough_bios
, bio
);
908 spin_unlock_irqrestore(&cache
->lock
, flags
);
913 static void writethrough_endio(struct bio
*bio
)
915 struct per_bio_data
*pb
= get_per_bio_data(bio
, PB_DATA_SIZE_WT
);
917 dm_unhook_bio(&pb
->hook_info
, bio
);
924 dm_bio_restore(&pb
->bio_details
, bio
);
925 remap_to_cache(pb
->cache
, bio
, pb
->cblock
);
928 * We can't issue this bio directly, since we're in interrupt
929 * context. So it gets put on a bio list for processing by the
932 defer_writethrough_bio(pb
->cache
, bio
);
936 * When running in writethrough mode we need to send writes to clean blocks
937 * to both the cache and origin devices. In future we'd like to clone the
938 * bio and send them in parallel, but for now we're doing them in
939 * series as this is easier.
941 static void remap_to_origin_then_cache(struct cache
*cache
, struct bio
*bio
,
942 dm_oblock_t oblock
, dm_cblock_t cblock
)
944 struct per_bio_data
*pb
= get_per_bio_data(bio
, PB_DATA_SIZE_WT
);
948 dm_hook_bio(&pb
->hook_info
, bio
, writethrough_endio
, NULL
);
949 dm_bio_record(&pb
->bio_details
, bio
);
951 remap_to_origin_clear_discard(pb
->cache
, bio
, oblock
);
954 /*----------------------------------------------------------------
956 *--------------------------------------------------------------*/
957 static enum cache_metadata_mode
get_cache_mode(struct cache
*cache
)
959 return cache
->features
.mode
;
962 static const char *cache_device_name(struct cache
*cache
)
964 return dm_device_name(dm_table_get_md(cache
->ti
->table
));
967 static void notify_mode_switch(struct cache
*cache
, enum cache_metadata_mode mode
)
969 const char *descs
[] = {
975 dm_table_event(cache
->ti
->table
);
976 DMINFO("%s: switching cache to %s mode",
977 cache_device_name(cache
), descs
[(int)mode
]);
980 static void set_cache_mode(struct cache
*cache
, enum cache_metadata_mode new_mode
)
983 enum cache_metadata_mode old_mode
= get_cache_mode(cache
);
985 if (dm_cache_metadata_needs_check(cache
->cmd
, &needs_check
)) {
986 DMERR("%s: unable to read needs_check flag, setting failure mode.",
987 cache_device_name(cache
));
991 if (new_mode
== CM_WRITE
&& needs_check
) {
992 DMERR("%s: unable to switch cache to write mode until repaired.",
993 cache_device_name(cache
));
994 if (old_mode
!= new_mode
)
997 new_mode
= CM_READ_ONLY
;
1000 /* Never move out of fail mode */
1001 if (old_mode
== CM_FAIL
)
1007 dm_cache_metadata_set_read_only(cache
->cmd
);
1011 dm_cache_metadata_set_read_write(cache
->cmd
);
1015 cache
->features
.mode
= new_mode
;
1017 if (new_mode
!= old_mode
)
1018 notify_mode_switch(cache
, new_mode
);
1021 static void abort_transaction(struct cache
*cache
)
1023 const char *dev_name
= cache_device_name(cache
);
1025 if (get_cache_mode(cache
) >= CM_READ_ONLY
)
1028 if (dm_cache_metadata_set_needs_check(cache
->cmd
)) {
1029 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name
);
1030 set_cache_mode(cache
, CM_FAIL
);
1033 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name
);
1034 if (dm_cache_metadata_abort(cache
->cmd
)) {
1035 DMERR("%s: failed to abort metadata transaction", dev_name
);
1036 set_cache_mode(cache
, CM_FAIL
);
1040 static void metadata_operation_failed(struct cache
*cache
, const char *op
, int r
)
1042 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1043 cache_device_name(cache
), op
, r
);
1044 abort_transaction(cache
);
1045 set_cache_mode(cache
, CM_READ_ONLY
);
1048 /*----------------------------------------------------------------
1049 * Migration processing
1051 * Migration covers moving data from the origin device to the cache, or
1053 *--------------------------------------------------------------*/
1054 static void inc_io_migrations(struct cache
*cache
)
1056 atomic_inc(&cache
->nr_io_migrations
);
1059 static void dec_io_migrations(struct cache
*cache
)
1061 atomic_dec(&cache
->nr_io_migrations
);
1064 static bool discard_or_flush(struct bio
*bio
)
1066 return bio_op(bio
) == REQ_OP_DISCARD
|| op_is_flush(bio
->bi_opf
);
1069 static void __cell_defer(struct cache
*cache
, struct dm_bio_prison_cell
*cell
)
1071 if (discard_or_flush(cell
->holder
)) {
1073 * We have to handle these bios individually.
1075 dm_cell_release(cache
->prison
, cell
, &cache
->deferred_bios
);
1076 free_prison_cell(cache
, cell
);
1078 list_add_tail(&cell
->user_list
, &cache
->deferred_cells
);
1081 static void cell_defer(struct cache
*cache
, struct dm_bio_prison_cell
*cell
, bool holder
)
1083 unsigned long flags
;
1085 if (!holder
&& dm_cell_promote_or_release(cache
->prison
, cell
)) {
1087 * There was no prisoner to promote to holder, the
1088 * cell has been released.
1090 free_prison_cell(cache
, cell
);
1094 spin_lock_irqsave(&cache
->lock
, flags
);
1095 __cell_defer(cache
, cell
);
1096 spin_unlock_irqrestore(&cache
->lock
, flags
);
1101 static void cell_error_with_code(struct cache
*cache
, struct dm_bio_prison_cell
*cell
, int err
)
1103 dm_cell_error(cache
->prison
, cell
, err
);
1104 free_prison_cell(cache
, cell
);
1107 static void cell_requeue(struct cache
*cache
, struct dm_bio_prison_cell
*cell
)
1109 cell_error_with_code(cache
, cell
, DM_ENDIO_REQUEUE
);
1112 static void free_io_migration(struct dm_cache_migration
*mg
)
1114 struct cache
*cache
= mg
->cache
;
1116 dec_io_migrations(cache
);
1121 static void migration_failure(struct dm_cache_migration
*mg
)
1123 struct cache
*cache
= mg
->cache
;
1124 const char *dev_name
= cache_device_name(cache
);
1126 if (mg
->writeback
) {
1127 DMERR_LIMIT("%s: writeback failed; couldn't copy block", dev_name
);
1128 set_dirty(cache
, mg
->old_oblock
, mg
->cblock
);
1129 cell_defer(cache
, mg
->old_ocell
, false);
1131 } else if (mg
->demote
) {
1132 DMERR_LIMIT("%s: demotion failed; couldn't copy block", dev_name
);
1133 policy_force_mapping(cache
->policy
, mg
->new_oblock
, mg
->old_oblock
);
1135 cell_defer(cache
, mg
->old_ocell
, mg
->promote
? false : true);
1137 cell_defer(cache
, mg
->new_ocell
, true);
1139 DMERR_LIMIT("%s: promotion failed; couldn't copy block", dev_name
);
1140 policy_remove_mapping(cache
->policy
, mg
->new_oblock
);
1141 cell_defer(cache
, mg
->new_ocell
, true);
1144 free_io_migration(mg
);
1147 static void migration_success_pre_commit(struct dm_cache_migration
*mg
)
1150 unsigned long flags
;
1151 struct cache
*cache
= mg
->cache
;
1153 if (mg
->writeback
) {
1154 clear_dirty(cache
, mg
->old_oblock
, mg
->cblock
);
1155 cell_defer(cache
, mg
->old_ocell
, false);
1156 free_io_migration(mg
);
1159 } else if (mg
->demote
) {
1160 r
= dm_cache_remove_mapping(cache
->cmd
, mg
->cblock
);
1162 DMERR_LIMIT("%s: demotion failed; couldn't update on disk metadata",
1163 cache_device_name(cache
));
1164 metadata_operation_failed(cache
, "dm_cache_remove_mapping", r
);
1165 policy_force_mapping(cache
->policy
, mg
->new_oblock
,
1168 cell_defer(cache
, mg
->new_ocell
, true);
1169 free_io_migration(mg
);
1173 r
= dm_cache_insert_mapping(cache
->cmd
, mg
->cblock
, mg
->new_oblock
);
1175 DMERR_LIMIT("%s: promotion failed; couldn't update on disk metadata",
1176 cache_device_name(cache
));
1177 metadata_operation_failed(cache
, "dm_cache_insert_mapping", r
);
1178 policy_remove_mapping(cache
->policy
, mg
->new_oblock
);
1179 free_io_migration(mg
);
1184 spin_lock_irqsave(&cache
->lock
, flags
);
1185 list_add_tail(&mg
->list
, &cache
->need_commit_migrations
);
1186 cache
->commit_requested
= true;
1187 spin_unlock_irqrestore(&cache
->lock
, flags
);
1190 static void migration_success_post_commit(struct dm_cache_migration
*mg
)
1192 unsigned long flags
;
1193 struct cache
*cache
= mg
->cache
;
1195 if (mg
->writeback
) {
1196 DMWARN_LIMIT("%s: writeback unexpectedly triggered commit",
1197 cache_device_name(cache
));
1200 } else if (mg
->demote
) {
1201 cell_defer(cache
, mg
->old_ocell
, mg
->promote
? false : true);
1206 spin_lock_irqsave(&cache
->lock
, flags
);
1207 list_add_tail(&mg
->list
, &cache
->quiesced_migrations
);
1208 spin_unlock_irqrestore(&cache
->lock
, flags
);
1212 policy_remove_mapping(cache
->policy
, mg
->old_oblock
);
1213 free_io_migration(mg
);
1217 if (mg
->requeue_holder
) {
1218 clear_dirty(cache
, mg
->new_oblock
, mg
->cblock
);
1219 cell_defer(cache
, mg
->new_ocell
, true);
1222 * The block was promoted via an overwrite, so it's dirty.
1224 set_dirty(cache
, mg
->new_oblock
, mg
->cblock
);
1225 bio_endio(mg
->new_ocell
->holder
);
1226 cell_defer(cache
, mg
->new_ocell
, false);
1228 free_io_migration(mg
);
1232 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
1234 unsigned long flags
;
1235 struct dm_cache_migration
*mg
= (struct dm_cache_migration
*) context
;
1236 struct cache
*cache
= mg
->cache
;
1238 if (read_err
|| write_err
)
1241 spin_lock_irqsave(&cache
->lock
, flags
);
1242 list_add_tail(&mg
->list
, &cache
->completed_migrations
);
1243 spin_unlock_irqrestore(&cache
->lock
, flags
);
1248 static void issue_copy(struct dm_cache_migration
*mg
)
1251 struct dm_io_region o_region
, c_region
;
1252 struct cache
*cache
= mg
->cache
;
1253 sector_t cblock
= from_cblock(mg
->cblock
);
1255 o_region
.bdev
= cache
->origin_dev
->bdev
;
1256 o_region
.count
= cache
->sectors_per_block
;
1258 c_region
.bdev
= cache
->cache_dev
->bdev
;
1259 c_region
.sector
= cblock
* cache
->sectors_per_block
;
1260 c_region
.count
= cache
->sectors_per_block
;
1262 if (mg
->writeback
|| mg
->demote
) {
1264 o_region
.sector
= from_oblock(mg
->old_oblock
) * cache
->sectors_per_block
;
1265 r
= dm_kcopyd_copy(cache
->copier
, &c_region
, 1, &o_region
, 0, copy_complete
, mg
);
1268 o_region
.sector
= from_oblock(mg
->new_oblock
) * cache
->sectors_per_block
;
1269 r
= dm_kcopyd_copy(cache
->copier
, &o_region
, 1, &c_region
, 0, copy_complete
, mg
);
1273 DMERR_LIMIT("%s: issuing migration failed", cache_device_name(cache
));
1274 migration_failure(mg
);
1278 static void overwrite_endio(struct bio
*bio
)
1280 struct dm_cache_migration
*mg
= bio
->bi_private
;
1281 struct cache
*cache
= mg
->cache
;
1282 size_t pb_data_size
= get_per_bio_data_size(cache
);
1283 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
1284 unsigned long flags
;
1286 dm_unhook_bio(&pb
->hook_info
, bio
);
1291 mg
->requeue_holder
= false;
1293 spin_lock_irqsave(&cache
->lock
, flags
);
1294 list_add_tail(&mg
->list
, &cache
->completed_migrations
);
1295 spin_unlock_irqrestore(&cache
->lock
, flags
);
1300 static void issue_overwrite(struct dm_cache_migration
*mg
, struct bio
*bio
)
1302 size_t pb_data_size
= get_per_bio_data_size(mg
->cache
);
1303 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
1305 dm_hook_bio(&pb
->hook_info
, bio
, overwrite_endio
, mg
);
1306 remap_to_cache_dirty(mg
->cache
, bio
, mg
->new_oblock
, mg
->cblock
);
1309 * No need to inc_ds() here, since the cell will be held for the
1310 * duration of the io.
1312 accounted_request(mg
->cache
, bio
);
1315 static bool bio_writes_complete_block(struct cache
*cache
, struct bio
*bio
)
1317 return (bio_data_dir(bio
) == WRITE
) &&
1318 (bio
->bi_iter
.bi_size
== (cache
->sectors_per_block
<< SECTOR_SHIFT
));
1321 static void avoid_copy(struct dm_cache_migration
*mg
)
1323 atomic_inc(&mg
->cache
->stats
.copies_avoided
);
1324 migration_success_pre_commit(mg
);
1327 static void calc_discard_block_range(struct cache
*cache
, struct bio
*bio
,
1328 dm_dblock_t
*b
, dm_dblock_t
*e
)
1330 sector_t sb
= bio
->bi_iter
.bi_sector
;
1331 sector_t se
= bio_end_sector(bio
);
1333 *b
= to_dblock(dm_sector_div_up(sb
, cache
->discard_block_size
));
1335 if (se
- sb
< cache
->discard_block_size
)
1338 *e
= to_dblock(block_div(se
, cache
->discard_block_size
));
1341 static void issue_discard(struct dm_cache_migration
*mg
)
1344 struct bio
*bio
= mg
->new_ocell
->holder
;
1345 struct cache
*cache
= mg
->cache
;
1347 calc_discard_block_range(cache
, bio
, &b
, &e
);
1349 set_discard(cache
, b
);
1350 b
= to_dblock(from_dblock(b
) + 1);
1354 cell_defer(cache
, mg
->new_ocell
, false);
1359 static void issue_copy_or_discard(struct dm_cache_migration
*mg
)
1362 struct cache
*cache
= mg
->cache
;
1369 if (mg
->writeback
|| mg
->demote
)
1370 avoid
= !is_dirty(cache
, mg
->cblock
) ||
1371 is_discarded_oblock(cache
, mg
->old_oblock
);
1373 struct bio
*bio
= mg
->new_ocell
->holder
;
1375 avoid
= is_discarded_oblock(cache
, mg
->new_oblock
);
1377 if (writeback_mode(&cache
->features
) &&
1378 !avoid
&& bio_writes_complete_block(cache
, bio
)) {
1379 issue_overwrite(mg
, bio
);
1384 avoid
? avoid_copy(mg
) : issue_copy(mg
);
1387 static void complete_migration(struct dm_cache_migration
*mg
)
1390 migration_failure(mg
);
1392 migration_success_pre_commit(mg
);
1395 static void process_migrations(struct cache
*cache
, struct list_head
*head
,
1396 void (*fn
)(struct dm_cache_migration
*))
1398 unsigned long flags
;
1399 struct list_head list
;
1400 struct dm_cache_migration
*mg
, *tmp
;
1402 INIT_LIST_HEAD(&list
);
1403 spin_lock_irqsave(&cache
->lock
, flags
);
1404 list_splice_init(head
, &list
);
1405 spin_unlock_irqrestore(&cache
->lock
, flags
);
1407 list_for_each_entry_safe(mg
, tmp
, &list
, list
)
1411 static void __queue_quiesced_migration(struct dm_cache_migration
*mg
)
1413 list_add_tail(&mg
->list
, &mg
->cache
->quiesced_migrations
);
1416 static void queue_quiesced_migration(struct dm_cache_migration
*mg
)
1418 unsigned long flags
;
1419 struct cache
*cache
= mg
->cache
;
1421 spin_lock_irqsave(&cache
->lock
, flags
);
1422 __queue_quiesced_migration(mg
);
1423 spin_unlock_irqrestore(&cache
->lock
, flags
);
1428 static void queue_quiesced_migrations(struct cache
*cache
, struct list_head
*work
)
1430 unsigned long flags
;
1431 struct dm_cache_migration
*mg
, *tmp
;
1433 spin_lock_irqsave(&cache
->lock
, flags
);
1434 list_for_each_entry_safe(mg
, tmp
, work
, list
)
1435 __queue_quiesced_migration(mg
);
1436 spin_unlock_irqrestore(&cache
->lock
, flags
);
1441 static void check_for_quiesced_migrations(struct cache
*cache
,
1442 struct per_bio_data
*pb
)
1444 struct list_head work
;
1446 if (!pb
->all_io_entry
)
1449 INIT_LIST_HEAD(&work
);
1450 dm_deferred_entry_dec(pb
->all_io_entry
, &work
);
1452 if (!list_empty(&work
))
1453 queue_quiesced_migrations(cache
, &work
);
1456 static void quiesce_migration(struct dm_cache_migration
*mg
)
1458 if (!dm_deferred_set_add_work(mg
->cache
->all_io_ds
, &mg
->list
))
1459 queue_quiesced_migration(mg
);
1462 static void promote(struct cache
*cache
, struct prealloc
*structs
,
1463 dm_oblock_t oblock
, dm_cblock_t cblock
,
1464 struct dm_bio_prison_cell
*cell
)
1466 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
1469 mg
->discard
= false;
1470 mg
->writeback
= false;
1473 mg
->requeue_holder
= true;
1474 mg
->invalidate
= false;
1476 mg
->new_oblock
= oblock
;
1477 mg
->cblock
= cblock
;
1478 mg
->old_ocell
= NULL
;
1479 mg
->new_ocell
= cell
;
1480 mg
->start_jiffies
= jiffies
;
1482 inc_io_migrations(cache
);
1483 quiesce_migration(mg
);
1486 static void writeback(struct cache
*cache
, struct prealloc
*structs
,
1487 dm_oblock_t oblock
, dm_cblock_t cblock
,
1488 struct dm_bio_prison_cell
*cell
)
1490 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
1493 mg
->discard
= false;
1494 mg
->writeback
= true;
1496 mg
->promote
= false;
1497 mg
->requeue_holder
= true;
1498 mg
->invalidate
= false;
1500 mg
->old_oblock
= oblock
;
1501 mg
->cblock
= cblock
;
1502 mg
->old_ocell
= cell
;
1503 mg
->new_ocell
= NULL
;
1504 mg
->start_jiffies
= jiffies
;
1506 inc_io_migrations(cache
);
1507 quiesce_migration(mg
);
1510 static void demote_then_promote(struct cache
*cache
, struct prealloc
*structs
,
1511 dm_oblock_t old_oblock
, dm_oblock_t new_oblock
,
1513 struct dm_bio_prison_cell
*old_ocell
,
1514 struct dm_bio_prison_cell
*new_ocell
)
1516 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
1519 mg
->discard
= false;
1520 mg
->writeback
= false;
1523 mg
->requeue_holder
= true;
1524 mg
->invalidate
= false;
1526 mg
->old_oblock
= old_oblock
;
1527 mg
->new_oblock
= new_oblock
;
1528 mg
->cblock
= cblock
;
1529 mg
->old_ocell
= old_ocell
;
1530 mg
->new_ocell
= new_ocell
;
1531 mg
->start_jiffies
= jiffies
;
1533 inc_io_migrations(cache
);
1534 quiesce_migration(mg
);
1538 * Invalidate a cache entry. No writeback occurs; any changes in the cache
1539 * block are thrown away.
1541 static void invalidate(struct cache
*cache
, struct prealloc
*structs
,
1542 dm_oblock_t oblock
, dm_cblock_t cblock
,
1543 struct dm_bio_prison_cell
*cell
)
1545 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
1548 mg
->discard
= false;
1549 mg
->writeback
= false;
1551 mg
->promote
= false;
1552 mg
->requeue_holder
= true;
1553 mg
->invalidate
= true;
1555 mg
->old_oblock
= oblock
;
1556 mg
->cblock
= cblock
;
1557 mg
->old_ocell
= cell
;
1558 mg
->new_ocell
= NULL
;
1559 mg
->start_jiffies
= jiffies
;
1561 inc_io_migrations(cache
);
1562 quiesce_migration(mg
);
1565 static void discard(struct cache
*cache
, struct prealloc
*structs
,
1566 struct dm_bio_prison_cell
*cell
)
1568 struct dm_cache_migration
*mg
= prealloc_get_migration(structs
);
1572 mg
->writeback
= false;
1574 mg
->promote
= false;
1575 mg
->requeue_holder
= false;
1576 mg
->invalidate
= false;
1578 mg
->old_ocell
= NULL
;
1579 mg
->new_ocell
= cell
;
1580 mg
->start_jiffies
= jiffies
;
1582 quiesce_migration(mg
);
1585 /*----------------------------------------------------------------
1587 *--------------------------------------------------------------*/
1588 static void defer_bio(struct cache
*cache
, struct bio
*bio
)
1590 unsigned long flags
;
1592 spin_lock_irqsave(&cache
->lock
, flags
);
1593 bio_list_add(&cache
->deferred_bios
, bio
);
1594 spin_unlock_irqrestore(&cache
->lock
, flags
);
1599 static void process_flush_bio(struct cache
*cache
, struct bio
*bio
)
1601 size_t pb_data_size
= get_per_bio_data_size(cache
);
1602 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
1604 BUG_ON(bio
->bi_iter
.bi_size
);
1606 remap_to_origin(cache
, bio
);
1608 remap_to_cache(cache
, bio
, 0);
1611 * REQ_PREFLUSH is not directed at any particular block so we don't
1612 * need to inc_ds(). REQ_FUA's are split into a write + REQ_PREFLUSH
1618 static void process_discard_bio(struct cache
*cache
, struct prealloc
*structs
,
1623 struct dm_bio_prison_cell
*cell_prealloc
, *new_ocell
;
1625 calc_discard_block_range(cache
, bio
, &b
, &e
);
1631 cell_prealloc
= prealloc_get_cell(structs
);
1632 r
= bio_detain_range(cache
, dblock_to_oblock(cache
, b
), dblock_to_oblock(cache
, e
), bio
, cell_prealloc
,
1633 (cell_free_fn
) prealloc_put_cell
,
1634 structs
, &new_ocell
);
1638 discard(cache
, structs
, new_ocell
);
1641 static bool spare_migration_bandwidth(struct cache
*cache
)
1643 sector_t current_volume
= (atomic_read(&cache
->nr_io_migrations
) + 1) *
1644 cache
->sectors_per_block
;
1645 return current_volume
< cache
->migration_threshold
;
1648 static void inc_hit_counter(struct cache
*cache
, struct bio
*bio
)
1650 atomic_inc(bio_data_dir(bio
) == READ
?
1651 &cache
->stats
.read_hit
: &cache
->stats
.write_hit
);
1654 static void inc_miss_counter(struct cache
*cache
, struct bio
*bio
)
1656 atomic_inc(bio_data_dir(bio
) == READ
?
1657 &cache
->stats
.read_miss
: &cache
->stats
.write_miss
);
1660 /*----------------------------------------------------------------*/
1663 struct cache
*cache
;
1664 struct bio_list bios_for_issue
;
1665 struct bio_list unhandled_bios
;
1669 static void inc_fn(void *context
, struct dm_bio_prison_cell
*cell
)
1672 struct inc_detail
*detail
= context
;
1673 struct cache
*cache
= detail
->cache
;
1675 inc_ds(cache
, cell
->holder
, cell
);
1676 if (bio_data_dir(cell
->holder
) == WRITE
)
1677 detail
->any_writes
= true;
1679 while ((bio
= bio_list_pop(&cell
->bios
))) {
1680 if (discard_or_flush(bio
)) {
1681 bio_list_add(&detail
->unhandled_bios
, bio
);
1685 if (bio_data_dir(bio
) == WRITE
)
1686 detail
->any_writes
= true;
1688 bio_list_add(&detail
->bios_for_issue
, bio
);
1689 inc_ds(cache
, bio
, cell
);
1693 // FIXME: refactor these two
1694 static void remap_cell_to_origin_clear_discard(struct cache
*cache
,
1695 struct dm_bio_prison_cell
*cell
,
1696 dm_oblock_t oblock
, bool issue_holder
)
1699 unsigned long flags
;
1700 struct inc_detail detail
;
1702 detail
.cache
= cache
;
1703 bio_list_init(&detail
.bios_for_issue
);
1704 bio_list_init(&detail
.unhandled_bios
);
1705 detail
.any_writes
= false;
1707 spin_lock_irqsave(&cache
->lock
, flags
);
1708 dm_cell_visit_release(cache
->prison
, inc_fn
, &detail
, cell
);
1709 bio_list_merge(&cache
->deferred_bios
, &detail
.unhandled_bios
);
1710 spin_unlock_irqrestore(&cache
->lock
, flags
);
1712 remap_to_origin(cache
, cell
->holder
);
1714 issue(cache
, cell
->holder
);
1716 accounted_begin(cache
, cell
->holder
);
1718 if (detail
.any_writes
)
1719 clear_discard(cache
, oblock_to_dblock(cache
, oblock
));
1721 while ((bio
= bio_list_pop(&detail
.bios_for_issue
))) {
1722 remap_to_origin(cache
, bio
);
1726 free_prison_cell(cache
, cell
);
1729 static void remap_cell_to_cache_dirty(struct cache
*cache
, struct dm_bio_prison_cell
*cell
,
1730 dm_oblock_t oblock
, dm_cblock_t cblock
, bool issue_holder
)
1733 unsigned long flags
;
1734 struct inc_detail detail
;
1736 detail
.cache
= cache
;
1737 bio_list_init(&detail
.bios_for_issue
);
1738 bio_list_init(&detail
.unhandled_bios
);
1739 detail
.any_writes
= false;
1741 spin_lock_irqsave(&cache
->lock
, flags
);
1742 dm_cell_visit_release(cache
->prison
, inc_fn
, &detail
, cell
);
1743 bio_list_merge(&cache
->deferred_bios
, &detail
.unhandled_bios
);
1744 spin_unlock_irqrestore(&cache
->lock
, flags
);
1746 remap_to_cache(cache
, cell
->holder
, cblock
);
1748 issue(cache
, cell
->holder
);
1750 accounted_begin(cache
, cell
->holder
);
1752 if (detail
.any_writes
) {
1753 set_dirty(cache
, oblock
, cblock
);
1754 clear_discard(cache
, oblock_to_dblock(cache
, oblock
));
1757 while ((bio
= bio_list_pop(&detail
.bios_for_issue
))) {
1758 remap_to_cache(cache
, bio
, cblock
);
1762 free_prison_cell(cache
, cell
);
1765 /*----------------------------------------------------------------*/
1767 struct old_oblock_lock
{
1768 struct policy_locker locker
;
1769 struct cache
*cache
;
1770 struct prealloc
*structs
;
1771 struct dm_bio_prison_cell
*cell
;
1774 static int null_locker(struct policy_locker
*locker
, dm_oblock_t b
)
1776 /* This should never be called */
1781 static int cell_locker(struct policy_locker
*locker
, dm_oblock_t b
)
1783 struct old_oblock_lock
*l
= container_of(locker
, struct old_oblock_lock
, locker
);
1784 struct dm_bio_prison_cell
*cell_prealloc
= prealloc_get_cell(l
->structs
);
1786 return bio_detain(l
->cache
, b
, NULL
, cell_prealloc
,
1787 (cell_free_fn
) prealloc_put_cell
,
1788 l
->structs
, &l
->cell
);
1791 static void process_cell(struct cache
*cache
, struct prealloc
*structs
,
1792 struct dm_bio_prison_cell
*new_ocell
)
1795 bool release_cell
= true;
1796 struct bio
*bio
= new_ocell
->holder
;
1797 dm_oblock_t block
= get_bio_block(cache
, bio
);
1798 struct policy_result lookup_result
;
1799 bool passthrough
= passthrough_mode(&cache
->features
);
1800 bool fast_promotion
, can_migrate
;
1801 struct old_oblock_lock ool
;
1803 fast_promotion
= is_discarded_oblock(cache
, block
) || bio_writes_complete_block(cache
, bio
);
1804 can_migrate
= !passthrough
&& (fast_promotion
|| spare_migration_bandwidth(cache
));
1806 ool
.locker
.fn
= cell_locker
;
1808 ool
.structs
= structs
;
1810 r
= policy_map(cache
->policy
, block
, true, can_migrate
, fast_promotion
,
1811 bio
, &ool
.locker
, &lookup_result
);
1813 if (r
== -EWOULDBLOCK
)
1814 /* migration has been denied */
1815 lookup_result
.op
= POLICY_MISS
;
1817 switch (lookup_result
.op
) {
1820 inc_miss_counter(cache
, bio
);
1823 * Passthrough always maps to the origin,
1824 * invalidating any cache blocks that are written
1828 if (bio_data_dir(bio
) == WRITE
) {
1829 atomic_inc(&cache
->stats
.demotion
);
1830 invalidate(cache
, structs
, block
, lookup_result
.cblock
, new_ocell
);
1831 release_cell
= false;
1834 /* FIXME: factor out issue_origin() */
1835 remap_to_origin_clear_discard(cache
, bio
, block
);
1836 inc_and_issue(cache
, bio
, new_ocell
);
1839 inc_hit_counter(cache
, bio
);
1841 if (bio_data_dir(bio
) == WRITE
&&
1842 writethrough_mode(&cache
->features
) &&
1843 !is_dirty(cache
, lookup_result
.cblock
)) {
1844 remap_to_origin_then_cache(cache
, bio
, block
, lookup_result
.cblock
);
1845 inc_and_issue(cache
, bio
, new_ocell
);
1848 remap_cell_to_cache_dirty(cache
, new_ocell
, block
, lookup_result
.cblock
, true);
1849 release_cell
= false;
1856 inc_miss_counter(cache
, bio
);
1857 remap_cell_to_origin_clear_discard(cache
, new_ocell
, block
, true);
1858 release_cell
= false;
1862 atomic_inc(&cache
->stats
.promotion
);
1863 promote(cache
, structs
, block
, lookup_result
.cblock
, new_ocell
);
1864 release_cell
= false;
1867 case POLICY_REPLACE
:
1868 atomic_inc(&cache
->stats
.demotion
);
1869 atomic_inc(&cache
->stats
.promotion
);
1870 demote_then_promote(cache
, structs
, lookup_result
.old_oblock
,
1871 block
, lookup_result
.cblock
,
1872 ool
.cell
, new_ocell
);
1873 release_cell
= false;
1877 DMERR_LIMIT("%s: %s: erroring bio, unknown policy op: %u",
1878 cache_device_name(cache
), __func__
,
1879 (unsigned) lookup_result
.op
);
1884 cell_defer(cache
, new_ocell
, false);
1887 static void process_bio(struct cache
*cache
, struct prealloc
*structs
,
1891 dm_oblock_t block
= get_bio_block(cache
, bio
);
1892 struct dm_bio_prison_cell
*cell_prealloc
, *new_ocell
;
1895 * Check to see if that block is currently migrating.
1897 cell_prealloc
= prealloc_get_cell(structs
);
1898 r
= bio_detain(cache
, block
, bio
, cell_prealloc
,
1899 (cell_free_fn
) prealloc_put_cell
,
1900 structs
, &new_ocell
);
1904 process_cell(cache
, structs
, new_ocell
);
1907 static int need_commit_due_to_time(struct cache
*cache
)
1909 return jiffies
< cache
->last_commit_jiffies
||
1910 jiffies
> cache
->last_commit_jiffies
+ COMMIT_PERIOD
;
1914 * A non-zero return indicates read_only or fail_io mode.
1916 static int commit(struct cache
*cache
, bool clean_shutdown
)
1920 if (get_cache_mode(cache
) >= CM_READ_ONLY
)
1923 atomic_inc(&cache
->stats
.commit_count
);
1924 r
= dm_cache_commit(cache
->cmd
, clean_shutdown
);
1926 metadata_operation_failed(cache
, "dm_cache_commit", r
);
1931 static int commit_if_needed(struct cache
*cache
)
1935 if ((cache
->commit_requested
|| need_commit_due_to_time(cache
)) &&
1936 dm_cache_changed_this_transaction(cache
->cmd
)) {
1937 r
= commit(cache
, false);
1938 cache
->commit_requested
= false;
1939 cache
->last_commit_jiffies
= jiffies
;
1945 static void process_deferred_bios(struct cache
*cache
)
1947 bool prealloc_used
= false;
1948 unsigned long flags
;
1949 struct bio_list bios
;
1951 struct prealloc structs
;
1953 memset(&structs
, 0, sizeof(structs
));
1954 bio_list_init(&bios
);
1956 spin_lock_irqsave(&cache
->lock
, flags
);
1957 bio_list_merge(&bios
, &cache
->deferred_bios
);
1958 bio_list_init(&cache
->deferred_bios
);
1959 spin_unlock_irqrestore(&cache
->lock
, flags
);
1961 while (!bio_list_empty(&bios
)) {
1963 * If we've got no free migration structs, and processing
1964 * this bio might require one, we pause until there are some
1965 * prepared mappings to process.
1967 prealloc_used
= true;
1968 if (prealloc_data_structs(cache
, &structs
)) {
1969 spin_lock_irqsave(&cache
->lock
, flags
);
1970 bio_list_merge(&cache
->deferred_bios
, &bios
);
1971 spin_unlock_irqrestore(&cache
->lock
, flags
);
1975 bio
= bio_list_pop(&bios
);
1977 if (bio
->bi_opf
& REQ_PREFLUSH
)
1978 process_flush_bio(cache
, bio
);
1979 else if (bio_op(bio
) == REQ_OP_DISCARD
)
1980 process_discard_bio(cache
, &structs
, bio
);
1982 process_bio(cache
, &structs
, bio
);
1986 prealloc_free_structs(cache
, &structs
);
1989 static void process_deferred_cells(struct cache
*cache
)
1991 bool prealloc_used
= false;
1992 unsigned long flags
;
1993 struct dm_bio_prison_cell
*cell
, *tmp
;
1994 struct list_head cells
;
1995 struct prealloc structs
;
1997 memset(&structs
, 0, sizeof(structs
));
1999 INIT_LIST_HEAD(&cells
);
2001 spin_lock_irqsave(&cache
->lock
, flags
);
2002 list_splice_init(&cache
->deferred_cells
, &cells
);
2003 spin_unlock_irqrestore(&cache
->lock
, flags
);
2005 list_for_each_entry_safe(cell
, tmp
, &cells
, user_list
) {
2007 * If we've got no free migration structs, and processing
2008 * this bio might require one, we pause until there are some
2009 * prepared mappings to process.
2011 prealloc_used
= true;
2012 if (prealloc_data_structs(cache
, &structs
)) {
2013 spin_lock_irqsave(&cache
->lock
, flags
);
2014 list_splice(&cells
, &cache
->deferred_cells
);
2015 spin_unlock_irqrestore(&cache
->lock
, flags
);
2019 process_cell(cache
, &structs
, cell
);
2023 prealloc_free_structs(cache
, &structs
);
2026 static void process_deferred_flush_bios(struct cache
*cache
, bool submit_bios
)
2028 unsigned long flags
;
2029 struct bio_list bios
;
2032 bio_list_init(&bios
);
2034 spin_lock_irqsave(&cache
->lock
, flags
);
2035 bio_list_merge(&bios
, &cache
->deferred_flush_bios
);
2036 bio_list_init(&cache
->deferred_flush_bios
);
2037 spin_unlock_irqrestore(&cache
->lock
, flags
);
2040 * These bios have already been through inc_ds()
2042 while ((bio
= bio_list_pop(&bios
)))
2043 submit_bios
? accounted_request(cache
, bio
) : bio_io_error(bio
);
2046 static void process_deferred_writethrough_bios(struct cache
*cache
)
2048 unsigned long flags
;
2049 struct bio_list bios
;
2052 bio_list_init(&bios
);
2054 spin_lock_irqsave(&cache
->lock
, flags
);
2055 bio_list_merge(&bios
, &cache
->deferred_writethrough_bios
);
2056 bio_list_init(&cache
->deferred_writethrough_bios
);
2057 spin_unlock_irqrestore(&cache
->lock
, flags
);
2060 * These bios have already been through inc_ds()
2062 while ((bio
= bio_list_pop(&bios
)))
2063 accounted_request(cache
, bio
);
2066 static void writeback_some_dirty_blocks(struct cache
*cache
)
2068 bool prealloc_used
= false;
2071 struct prealloc structs
;
2072 struct dm_bio_prison_cell
*old_ocell
;
2073 bool busy
= !iot_idle_for(&cache
->origin_tracker
, HZ
);
2075 memset(&structs
, 0, sizeof(structs
));
2077 while (spare_migration_bandwidth(cache
)) {
2078 if (policy_writeback_work(cache
->policy
, &oblock
, &cblock
, busy
))
2079 break; /* no work to do */
2081 prealloc_used
= true;
2082 if (prealloc_data_structs(cache
, &structs
) ||
2083 get_cell(cache
, oblock
, &structs
, &old_ocell
)) {
2084 policy_set_dirty(cache
->policy
, oblock
);
2088 writeback(cache
, &structs
, oblock
, cblock
, old_ocell
);
2092 prealloc_free_structs(cache
, &structs
);
2095 /*----------------------------------------------------------------
2097 * Dropping something from the cache *without* writing back.
2098 *--------------------------------------------------------------*/
2100 static void process_invalidation_request(struct cache
*cache
, struct invalidation_request
*req
)
2103 uint64_t begin
= from_cblock(req
->cblocks
->begin
);
2104 uint64_t end
= from_cblock(req
->cblocks
->end
);
2106 while (begin
!= end
) {
2107 r
= policy_remove_cblock(cache
->policy
, to_cblock(begin
));
2109 r
= dm_cache_remove_mapping(cache
->cmd
, to_cblock(begin
));
2111 metadata_operation_failed(cache
, "dm_cache_remove_mapping", r
);
2115 } else if (r
== -ENODATA
) {
2116 /* harmless, already unmapped */
2120 DMERR("%s: policy_remove_cblock failed", cache_device_name(cache
));
2127 cache
->commit_requested
= true;
2130 atomic_set(&req
->complete
, 1);
2132 wake_up(&req
->result_wait
);
2135 static void process_invalidation_requests(struct cache
*cache
)
2137 struct list_head list
;
2138 struct invalidation_request
*req
, *tmp
;
2140 INIT_LIST_HEAD(&list
);
2141 spin_lock(&cache
->invalidation_lock
);
2142 list_splice_init(&cache
->invalidation_requests
, &list
);
2143 spin_unlock(&cache
->invalidation_lock
);
2145 list_for_each_entry_safe (req
, tmp
, &list
, list
)
2146 process_invalidation_request(cache
, req
);
2149 /*----------------------------------------------------------------
2151 *--------------------------------------------------------------*/
2152 static bool is_quiescing(struct cache
*cache
)
2154 return atomic_read(&cache
->quiescing
);
2157 static void ack_quiescing(struct cache
*cache
)
2159 if (is_quiescing(cache
)) {
2160 atomic_inc(&cache
->quiescing_ack
);
2161 wake_up(&cache
->quiescing_wait
);
2165 static void wait_for_quiescing_ack(struct cache
*cache
)
2167 wait_event(cache
->quiescing_wait
, atomic_read(&cache
->quiescing_ack
));
2170 static void start_quiescing(struct cache
*cache
)
2172 atomic_inc(&cache
->quiescing
);
2173 wait_for_quiescing_ack(cache
);
2176 static void stop_quiescing(struct cache
*cache
)
2178 atomic_set(&cache
->quiescing
, 0);
2179 atomic_set(&cache
->quiescing_ack
, 0);
2182 static void wait_for_migrations(struct cache
*cache
)
2184 wait_event(cache
->migration_wait
, !atomic_read(&cache
->nr_allocated_migrations
));
2187 static void stop_worker(struct cache
*cache
)
2189 cancel_delayed_work(&cache
->waker
);
2190 flush_workqueue(cache
->wq
);
2193 static void requeue_deferred_cells(struct cache
*cache
)
2195 unsigned long flags
;
2196 struct list_head cells
;
2197 struct dm_bio_prison_cell
*cell
, *tmp
;
2199 INIT_LIST_HEAD(&cells
);
2200 spin_lock_irqsave(&cache
->lock
, flags
);
2201 list_splice_init(&cache
->deferred_cells
, &cells
);
2202 spin_unlock_irqrestore(&cache
->lock
, flags
);
2204 list_for_each_entry_safe(cell
, tmp
, &cells
, user_list
)
2205 cell_requeue(cache
, cell
);
2208 static void requeue_deferred_bios(struct cache
*cache
)
2211 struct bio_list bios
;
2213 bio_list_init(&bios
);
2214 bio_list_merge(&bios
, &cache
->deferred_bios
);
2215 bio_list_init(&cache
->deferred_bios
);
2217 while ((bio
= bio_list_pop(&bios
))) {
2218 bio
->bi_error
= DM_ENDIO_REQUEUE
;
2223 static int more_work(struct cache
*cache
)
2225 if (is_quiescing(cache
))
2226 return !list_empty(&cache
->quiesced_migrations
) ||
2227 !list_empty(&cache
->completed_migrations
) ||
2228 !list_empty(&cache
->need_commit_migrations
);
2230 return !bio_list_empty(&cache
->deferred_bios
) ||
2231 !list_empty(&cache
->deferred_cells
) ||
2232 !bio_list_empty(&cache
->deferred_flush_bios
) ||
2233 !bio_list_empty(&cache
->deferred_writethrough_bios
) ||
2234 !list_empty(&cache
->quiesced_migrations
) ||
2235 !list_empty(&cache
->completed_migrations
) ||
2236 !list_empty(&cache
->need_commit_migrations
) ||
2240 static void do_worker(struct work_struct
*ws
)
2242 struct cache
*cache
= container_of(ws
, struct cache
, worker
);
2245 if (!is_quiescing(cache
)) {
2246 writeback_some_dirty_blocks(cache
);
2247 process_deferred_writethrough_bios(cache
);
2248 process_deferred_bios(cache
);
2249 process_deferred_cells(cache
);
2250 process_invalidation_requests(cache
);
2253 process_migrations(cache
, &cache
->quiesced_migrations
, issue_copy_or_discard
);
2254 process_migrations(cache
, &cache
->completed_migrations
, complete_migration
);
2256 if (commit_if_needed(cache
)) {
2257 process_deferred_flush_bios(cache
, false);
2258 process_migrations(cache
, &cache
->need_commit_migrations
, migration_failure
);
2260 process_deferred_flush_bios(cache
, true);
2261 process_migrations(cache
, &cache
->need_commit_migrations
,
2262 migration_success_post_commit
);
2265 ack_quiescing(cache
);
2267 } while (more_work(cache
));
2271 * We want to commit periodically so that not too much
2272 * unwritten metadata builds up.
2274 static void do_waker(struct work_struct
*ws
)
2276 struct cache
*cache
= container_of(to_delayed_work(ws
), struct cache
, waker
);
2277 policy_tick(cache
->policy
, true);
2279 queue_delayed_work(cache
->wq
, &cache
->waker
, COMMIT_PERIOD
);
2282 /*----------------------------------------------------------------*/
2284 static int is_congested(struct dm_dev
*dev
, int bdi_bits
)
2286 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
2287 return bdi_congested(&q
->backing_dev_info
, bdi_bits
);
2290 static int cache_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
2292 struct cache
*cache
= container_of(cb
, struct cache
, callbacks
);
2294 return is_congested(cache
->origin_dev
, bdi_bits
) ||
2295 is_congested(cache
->cache_dev
, bdi_bits
);
2298 /*----------------------------------------------------------------
2300 *--------------------------------------------------------------*/
2303 * This function gets called on the error paths of the constructor, so we
2304 * have to cope with a partially initialised struct.
2306 static void destroy(struct cache
*cache
)
2310 mempool_destroy(cache
->migration_pool
);
2312 if (cache
->all_io_ds
)
2313 dm_deferred_set_destroy(cache
->all_io_ds
);
2316 dm_bio_prison_destroy(cache
->prison
);
2319 destroy_workqueue(cache
->wq
);
2321 if (cache
->dirty_bitset
)
2322 free_bitset(cache
->dirty_bitset
);
2324 if (cache
->discard_bitset
)
2325 free_bitset(cache
->discard_bitset
);
2328 dm_kcopyd_client_destroy(cache
->copier
);
2331 dm_cache_metadata_close(cache
->cmd
);
2333 if (cache
->metadata_dev
)
2334 dm_put_device(cache
->ti
, cache
->metadata_dev
);
2336 if (cache
->origin_dev
)
2337 dm_put_device(cache
->ti
, cache
->origin_dev
);
2339 if (cache
->cache_dev
)
2340 dm_put_device(cache
->ti
, cache
->cache_dev
);
2343 dm_cache_policy_destroy(cache
->policy
);
2345 for (i
= 0; i
< cache
->nr_ctr_args
; i
++)
2346 kfree(cache
->ctr_args
[i
]);
2347 kfree(cache
->ctr_args
);
2352 static void cache_dtr(struct dm_target
*ti
)
2354 struct cache
*cache
= ti
->private;
2359 static sector_t
get_dev_size(struct dm_dev
*dev
)
2361 return i_size_read(dev
->bdev
->bd_inode
) >> SECTOR_SHIFT
;
2364 /*----------------------------------------------------------------*/
2367 * Construct a cache device mapping.
2369 * cache <metadata dev> <cache dev> <origin dev> <block size>
2370 * <#feature args> [<feature arg>]*
2371 * <policy> <#policy args> [<policy arg>]*
2373 * metadata dev : fast device holding the persistent metadata
2374 * cache dev : fast device holding cached data blocks
2375 * origin dev : slow device holding original data blocks
2376 * block size : cache unit size in sectors
2378 * #feature args : number of feature arguments passed
2379 * feature args : writethrough. (The default is writeback.)
2381 * policy : the replacement policy to use
2382 * #policy args : an even number of policy arguments corresponding
2383 * to key/value pairs passed to the policy
2384 * policy args : key/value pairs passed to the policy
2385 * E.g. 'sequential_threshold 1024'
2386 * See cache-policies.txt for details.
2388 * Optional feature arguments are:
2389 * writethrough : write through caching that prohibits cache block
2390 * content from being different from origin block content.
2391 * Without this argument, the default behaviour is to write
2392 * back cache block contents later for performance reasons,
2393 * so they may differ from the corresponding origin blocks.
2396 struct dm_target
*ti
;
2398 struct dm_dev
*metadata_dev
;
2400 struct dm_dev
*cache_dev
;
2401 sector_t cache_sectors
;
2403 struct dm_dev
*origin_dev
;
2404 sector_t origin_sectors
;
2406 uint32_t block_size
;
2408 const char *policy_name
;
2410 const char **policy_argv
;
2412 struct cache_features features
;
2415 static void destroy_cache_args(struct cache_args
*ca
)
2417 if (ca
->metadata_dev
)
2418 dm_put_device(ca
->ti
, ca
->metadata_dev
);
2421 dm_put_device(ca
->ti
, ca
->cache_dev
);
2424 dm_put_device(ca
->ti
, ca
->origin_dev
);
2429 static bool at_least_one_arg(struct dm_arg_set
*as
, char **error
)
2432 *error
= "Insufficient args";
2439 static int parse_metadata_dev(struct cache_args
*ca
, struct dm_arg_set
*as
,
2443 sector_t metadata_dev_size
;
2444 char b
[BDEVNAME_SIZE
];
2446 if (!at_least_one_arg(as
, error
))
2449 r
= dm_get_device(ca
->ti
, dm_shift_arg(as
), FMODE_READ
| FMODE_WRITE
,
2452 *error
= "Error opening metadata device";
2456 metadata_dev_size
= get_dev_size(ca
->metadata_dev
);
2457 if (metadata_dev_size
> DM_CACHE_METADATA_MAX_SECTORS_WARNING
)
2458 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2459 bdevname(ca
->metadata_dev
->bdev
, b
), THIN_METADATA_MAX_SECTORS
);
2464 static int parse_cache_dev(struct cache_args
*ca
, struct dm_arg_set
*as
,
2469 if (!at_least_one_arg(as
, error
))
2472 r
= dm_get_device(ca
->ti
, dm_shift_arg(as
), FMODE_READ
| FMODE_WRITE
,
2475 *error
= "Error opening cache device";
2478 ca
->cache_sectors
= get_dev_size(ca
->cache_dev
);
2483 static int parse_origin_dev(struct cache_args
*ca
, struct dm_arg_set
*as
,
2488 if (!at_least_one_arg(as
, error
))
2491 r
= dm_get_device(ca
->ti
, dm_shift_arg(as
), FMODE_READ
| FMODE_WRITE
,
2494 *error
= "Error opening origin device";
2498 ca
->origin_sectors
= get_dev_size(ca
->origin_dev
);
2499 if (ca
->ti
->len
> ca
->origin_sectors
) {
2500 *error
= "Device size larger than cached device";
2507 static int parse_block_size(struct cache_args
*ca
, struct dm_arg_set
*as
,
2510 unsigned long block_size
;
2512 if (!at_least_one_arg(as
, error
))
2515 if (kstrtoul(dm_shift_arg(as
), 10, &block_size
) || !block_size
||
2516 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2517 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2518 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2519 *error
= "Invalid data block size";
2523 if (block_size
> ca
->cache_sectors
) {
2524 *error
= "Data block size is larger than the cache device";
2528 ca
->block_size
= block_size
;
2533 static void init_features(struct cache_features
*cf
)
2535 cf
->mode
= CM_WRITE
;
2536 cf
->io_mode
= CM_IO_WRITEBACK
;
2539 static int parse_features(struct cache_args
*ca
, struct dm_arg_set
*as
,
2542 static struct dm_arg _args
[] = {
2543 {0, 1, "Invalid number of cache feature arguments"},
2549 struct cache_features
*cf
= &ca
->features
;
2553 r
= dm_read_arg_group(_args
, as
, &argc
, error
);
2558 arg
= dm_shift_arg(as
);
2560 if (!strcasecmp(arg
, "writeback"))
2561 cf
->io_mode
= CM_IO_WRITEBACK
;
2563 else if (!strcasecmp(arg
, "writethrough"))
2564 cf
->io_mode
= CM_IO_WRITETHROUGH
;
2566 else if (!strcasecmp(arg
, "passthrough"))
2567 cf
->io_mode
= CM_IO_PASSTHROUGH
;
2570 *error
= "Unrecognised cache feature requested";
2578 static int parse_policy(struct cache_args
*ca
, struct dm_arg_set
*as
,
2581 static struct dm_arg _args
[] = {
2582 {0, 1024, "Invalid number of policy arguments"},
2587 if (!at_least_one_arg(as
, error
))
2590 ca
->policy_name
= dm_shift_arg(as
);
2592 r
= dm_read_arg_group(_args
, as
, &ca
->policy_argc
, error
);
2596 ca
->policy_argv
= (const char **)as
->argv
;
2597 dm_consume_args(as
, ca
->policy_argc
);
2602 static int parse_cache_args(struct cache_args
*ca
, int argc
, char **argv
,
2606 struct dm_arg_set as
;
2611 r
= parse_metadata_dev(ca
, &as
, error
);
2615 r
= parse_cache_dev(ca
, &as
, error
);
2619 r
= parse_origin_dev(ca
, &as
, error
);
2623 r
= parse_block_size(ca
, &as
, error
);
2627 r
= parse_features(ca
, &as
, error
);
2631 r
= parse_policy(ca
, &as
, error
);
2638 /*----------------------------------------------------------------*/
2640 static struct kmem_cache
*migration_cache
;
2642 #define NOT_CORE_OPTION 1
2644 static int process_config_option(struct cache
*cache
, const char *key
, const char *value
)
2648 if (!strcasecmp(key
, "migration_threshold")) {
2649 if (kstrtoul(value
, 10, &tmp
))
2652 cache
->migration_threshold
= tmp
;
2656 return NOT_CORE_OPTION
;
2659 static int set_config_value(struct cache
*cache
, const char *key
, const char *value
)
2661 int r
= process_config_option(cache
, key
, value
);
2663 if (r
== NOT_CORE_OPTION
)
2664 r
= policy_set_config_value(cache
->policy
, key
, value
);
2667 DMWARN("bad config value for %s: %s", key
, value
);
2672 static int set_config_values(struct cache
*cache
, int argc
, const char **argv
)
2677 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
2682 r
= set_config_value(cache
, argv
[0], argv
[1]);
2693 static int create_cache_policy(struct cache
*cache
, struct cache_args
*ca
,
2696 struct dm_cache_policy
*p
= dm_cache_policy_create(ca
->policy_name
,
2698 cache
->origin_sectors
,
2699 cache
->sectors_per_block
);
2701 *error
= "Error creating cache's policy";
2710 * We want the discard block size to be at least the size of the cache
2711 * block size and have no more than 2^14 discard blocks across the origin.
2713 #define MAX_DISCARD_BLOCKS (1 << 14)
2715 static bool too_many_discard_blocks(sector_t discard_block_size
,
2716 sector_t origin_size
)
2718 (void) sector_div(origin_size
, discard_block_size
);
2720 return origin_size
> MAX_DISCARD_BLOCKS
;
2723 static sector_t
calculate_discard_block_size(sector_t cache_block_size
,
2724 sector_t origin_size
)
2726 sector_t discard_block_size
= cache_block_size
;
2729 while (too_many_discard_blocks(discard_block_size
, origin_size
))
2730 discard_block_size
*= 2;
2732 return discard_block_size
;
2735 static void set_cache_size(struct cache
*cache
, dm_cblock_t size
)
2737 dm_block_t nr_blocks
= from_cblock(size
);
2739 if (nr_blocks
> (1 << 20) && cache
->cache_size
!= size
)
2740 DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n"
2741 "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n"
2742 "Please consider increasing the cache block size to reduce the overall cache block count.",
2743 (unsigned long long) nr_blocks
);
2745 cache
->cache_size
= size
;
2748 #define DEFAULT_MIGRATION_THRESHOLD 2048
2750 static int cache_create(struct cache_args
*ca
, struct cache
**result
)
2753 char **error
= &ca
->ti
->error
;
2754 struct cache
*cache
;
2755 struct dm_target
*ti
= ca
->ti
;
2756 dm_block_t origin_blocks
;
2757 struct dm_cache_metadata
*cmd
;
2758 bool may_format
= ca
->features
.mode
== CM_WRITE
;
2760 cache
= kzalloc(sizeof(*cache
), GFP_KERNEL
);
2765 ti
->private = cache
;
2766 ti
->num_flush_bios
= 2;
2767 ti
->flush_supported
= true;
2769 ti
->num_discard_bios
= 1;
2770 ti
->discards_supported
= true;
2771 ti
->discard_zeroes_data_unsupported
= true;
2772 ti
->split_discard_bios
= false;
2774 cache
->features
= ca
->features
;
2775 ti
->per_io_data_size
= get_per_bio_data_size(cache
);
2777 cache
->callbacks
.congested_fn
= cache_is_congested
;
2778 dm_table_add_target_callbacks(ti
->table
, &cache
->callbacks
);
2780 cache
->metadata_dev
= ca
->metadata_dev
;
2781 cache
->origin_dev
= ca
->origin_dev
;
2782 cache
->cache_dev
= ca
->cache_dev
;
2784 ca
->metadata_dev
= ca
->origin_dev
= ca
->cache_dev
= NULL
;
2786 /* FIXME: factor out this whole section */
2787 origin_blocks
= cache
->origin_sectors
= ca
->origin_sectors
;
2788 origin_blocks
= block_div(origin_blocks
, ca
->block_size
);
2789 cache
->origin_blocks
= to_oblock(origin_blocks
);
2791 cache
->sectors_per_block
= ca
->block_size
;
2792 if (dm_set_target_max_io_len(ti
, cache
->sectors_per_block
)) {
2797 if (ca
->block_size
& (ca
->block_size
- 1)) {
2798 dm_block_t cache_size
= ca
->cache_sectors
;
2800 cache
->sectors_per_block_shift
= -1;
2801 cache_size
= block_div(cache_size
, ca
->block_size
);
2802 set_cache_size(cache
, to_cblock(cache_size
));
2804 cache
->sectors_per_block_shift
= __ffs(ca
->block_size
);
2805 set_cache_size(cache
, to_cblock(ca
->cache_sectors
>> cache
->sectors_per_block_shift
));
2808 r
= create_cache_policy(cache
, ca
, error
);
2812 cache
->policy_nr_args
= ca
->policy_argc
;
2813 cache
->migration_threshold
= DEFAULT_MIGRATION_THRESHOLD
;
2815 r
= set_config_values(cache
, ca
->policy_argc
, ca
->policy_argv
);
2817 *error
= "Error setting cache policy's config values";
2821 cmd
= dm_cache_metadata_open(cache
->metadata_dev
->bdev
,
2822 ca
->block_size
, may_format
,
2823 dm_cache_policy_get_hint_size(cache
->policy
));
2825 *error
= "Error creating metadata object";
2830 set_cache_mode(cache
, CM_WRITE
);
2831 if (get_cache_mode(cache
) != CM_WRITE
) {
2832 *error
= "Unable to get write access to metadata, please check/repair metadata.";
2837 if (passthrough_mode(&cache
->features
)) {
2840 r
= dm_cache_metadata_all_clean(cache
->cmd
, &all_clean
);
2842 *error
= "dm_cache_metadata_all_clean() failed";
2847 *error
= "Cannot enter passthrough mode unless all blocks are clean";
2853 spin_lock_init(&cache
->lock
);
2854 INIT_LIST_HEAD(&cache
->deferred_cells
);
2855 bio_list_init(&cache
->deferred_bios
);
2856 bio_list_init(&cache
->deferred_flush_bios
);
2857 bio_list_init(&cache
->deferred_writethrough_bios
);
2858 INIT_LIST_HEAD(&cache
->quiesced_migrations
);
2859 INIT_LIST_HEAD(&cache
->completed_migrations
);
2860 INIT_LIST_HEAD(&cache
->need_commit_migrations
);
2861 atomic_set(&cache
->nr_allocated_migrations
, 0);
2862 atomic_set(&cache
->nr_io_migrations
, 0);
2863 init_waitqueue_head(&cache
->migration_wait
);
2865 init_waitqueue_head(&cache
->quiescing_wait
);
2866 atomic_set(&cache
->quiescing
, 0);
2867 atomic_set(&cache
->quiescing_ack
, 0);
2870 atomic_set(&cache
->nr_dirty
, 0);
2871 cache
->dirty_bitset
= alloc_bitset(from_cblock(cache
->cache_size
));
2872 if (!cache
->dirty_bitset
) {
2873 *error
= "could not allocate dirty bitset";
2876 clear_bitset(cache
->dirty_bitset
, from_cblock(cache
->cache_size
));
2878 cache
->discard_block_size
=
2879 calculate_discard_block_size(cache
->sectors_per_block
,
2880 cache
->origin_sectors
);
2881 cache
->discard_nr_blocks
= to_dblock(dm_sector_div_up(cache
->origin_sectors
,
2882 cache
->discard_block_size
));
2883 cache
->discard_bitset
= alloc_bitset(from_dblock(cache
->discard_nr_blocks
));
2884 if (!cache
->discard_bitset
) {
2885 *error
= "could not allocate discard bitset";
2888 clear_bitset(cache
->discard_bitset
, from_dblock(cache
->discard_nr_blocks
));
2890 cache
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
2891 if (IS_ERR(cache
->copier
)) {
2892 *error
= "could not create kcopyd client";
2893 r
= PTR_ERR(cache
->copier
);
2897 cache
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
2899 *error
= "could not create workqueue for metadata object";
2902 INIT_WORK(&cache
->worker
, do_worker
);
2903 INIT_DELAYED_WORK(&cache
->waker
, do_waker
);
2904 cache
->last_commit_jiffies
= jiffies
;
2906 cache
->prison
= dm_bio_prison_create();
2907 if (!cache
->prison
) {
2908 *error
= "could not create bio prison";
2912 cache
->all_io_ds
= dm_deferred_set_create();
2913 if (!cache
->all_io_ds
) {
2914 *error
= "could not create all_io deferred set";
2918 cache
->migration_pool
= mempool_create_slab_pool(MIGRATION_POOL_SIZE
,
2920 if (!cache
->migration_pool
) {
2921 *error
= "Error creating cache's migration mempool";
2925 cache
->need_tick_bio
= true;
2926 cache
->sized
= false;
2927 cache
->invalidate
= false;
2928 cache
->commit_requested
= false;
2929 cache
->loaded_mappings
= false;
2930 cache
->loaded_discards
= false;
2934 atomic_set(&cache
->stats
.demotion
, 0);
2935 atomic_set(&cache
->stats
.promotion
, 0);
2936 atomic_set(&cache
->stats
.copies_avoided
, 0);
2937 atomic_set(&cache
->stats
.cache_cell_clash
, 0);
2938 atomic_set(&cache
->stats
.commit_count
, 0);
2939 atomic_set(&cache
->stats
.discard_count
, 0);
2941 spin_lock_init(&cache
->invalidation_lock
);
2942 INIT_LIST_HEAD(&cache
->invalidation_requests
);
2944 iot_init(&cache
->origin_tracker
);
2954 static int copy_ctr_args(struct cache
*cache
, int argc
, const char **argv
)
2959 copy
= kcalloc(argc
, sizeof(*copy
), GFP_KERNEL
);
2962 for (i
= 0; i
< argc
; i
++) {
2963 copy
[i
] = kstrdup(argv
[i
], GFP_KERNEL
);
2972 cache
->nr_ctr_args
= argc
;
2973 cache
->ctr_args
= copy
;
2978 static int cache_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2981 struct cache_args
*ca
;
2982 struct cache
*cache
= NULL
;
2984 ca
= kzalloc(sizeof(*ca
), GFP_KERNEL
);
2986 ti
->error
= "Error allocating memory for cache";
2991 r
= parse_cache_args(ca
, argc
, argv
, &ti
->error
);
2995 r
= cache_create(ca
, &cache
);
2999 r
= copy_ctr_args(cache
, argc
- 3, (const char **)argv
+ 3);
3005 ti
->private = cache
;
3008 destroy_cache_args(ca
);
3012 /*----------------------------------------------------------------*/
3014 static int cache_map(struct dm_target
*ti
, struct bio
*bio
)
3016 struct cache
*cache
= ti
->private;
3019 struct dm_bio_prison_cell
*cell
= NULL
;
3020 dm_oblock_t block
= get_bio_block(cache
, bio
);
3021 size_t pb_data_size
= get_per_bio_data_size(cache
);
3022 bool can_migrate
= false;
3023 bool fast_promotion
;
3024 struct policy_result lookup_result
;
3025 struct per_bio_data
*pb
= init_per_bio_data(bio
, pb_data_size
);
3026 struct old_oblock_lock ool
;
3028 ool
.locker
.fn
= null_locker
;
3030 if (unlikely(from_oblock(block
) >= from_oblock(cache
->origin_blocks
))) {
3032 * This can only occur if the io goes to a partial block at
3033 * the end of the origin device. We don't cache these.
3034 * Just remap to the origin and carry on.
3036 remap_to_origin(cache
, bio
);
3037 accounted_begin(cache
, bio
);
3038 return DM_MAPIO_REMAPPED
;
3041 if (discard_or_flush(bio
)) {
3042 defer_bio(cache
, bio
);
3043 return DM_MAPIO_SUBMITTED
;
3047 * Check to see if that block is currently migrating.
3049 cell
= alloc_prison_cell(cache
);
3051 defer_bio(cache
, bio
);
3052 return DM_MAPIO_SUBMITTED
;
3055 r
= bio_detain(cache
, block
, bio
, cell
,
3056 (cell_free_fn
) free_prison_cell
,
3060 defer_bio(cache
, bio
);
3062 return DM_MAPIO_SUBMITTED
;
3065 fast_promotion
= is_discarded_oblock(cache
, block
) || bio_writes_complete_block(cache
, bio
);
3067 r
= policy_map(cache
->policy
, block
, false, can_migrate
, fast_promotion
,
3068 bio
, &ool
.locker
, &lookup_result
);
3069 if (r
== -EWOULDBLOCK
) {
3070 cell_defer(cache
, cell
, true);
3071 return DM_MAPIO_SUBMITTED
;
3074 DMERR_LIMIT("%s: Unexpected return from cache replacement policy: %d",
3075 cache_device_name(cache
), r
);
3076 cell_defer(cache
, cell
, false);
3078 return DM_MAPIO_SUBMITTED
;
3081 r
= DM_MAPIO_REMAPPED
;
3082 switch (lookup_result
.op
) {
3084 if (passthrough_mode(&cache
->features
)) {
3085 if (bio_data_dir(bio
) == WRITE
) {
3087 * We need to invalidate this block, so
3088 * defer for the worker thread.
3090 cell_defer(cache
, cell
, true);
3091 r
= DM_MAPIO_SUBMITTED
;
3094 inc_miss_counter(cache
, bio
);
3095 remap_to_origin_clear_discard(cache
, bio
, block
);
3096 accounted_begin(cache
, bio
);
3097 inc_ds(cache
, bio
, cell
);
3098 // FIXME: we want to remap hits or misses straight
3099 // away rather than passing over to the worker.
3100 cell_defer(cache
, cell
, false);
3104 inc_hit_counter(cache
, bio
);
3105 if (bio_data_dir(bio
) == WRITE
&& writethrough_mode(&cache
->features
) &&
3106 !is_dirty(cache
, lookup_result
.cblock
)) {
3107 remap_to_origin_then_cache(cache
, bio
, block
, lookup_result
.cblock
);
3108 accounted_begin(cache
, bio
);
3109 inc_ds(cache
, bio
, cell
);
3110 cell_defer(cache
, cell
, false);
3113 remap_cell_to_cache_dirty(cache
, cell
, block
, lookup_result
.cblock
, false);
3118 inc_miss_counter(cache
, bio
);
3119 if (pb
->req_nr
!= 0) {
3121 * This is a duplicate writethrough io that is no
3122 * longer needed because the block has been demoted.
3125 // FIXME: remap everything as a miss
3126 cell_defer(cache
, cell
, false);
3127 r
= DM_MAPIO_SUBMITTED
;
3130 remap_cell_to_origin_clear_discard(cache
, cell
, block
, false);
3134 DMERR_LIMIT("%s: %s: erroring bio: unknown policy op: %u",
3135 cache_device_name(cache
), __func__
,
3136 (unsigned) lookup_result
.op
);
3137 cell_defer(cache
, cell
, false);
3139 r
= DM_MAPIO_SUBMITTED
;
3145 static int cache_end_io(struct dm_target
*ti
, struct bio
*bio
, int error
)
3147 struct cache
*cache
= ti
->private;
3148 unsigned long flags
;
3149 size_t pb_data_size
= get_per_bio_data_size(cache
);
3150 struct per_bio_data
*pb
= get_per_bio_data(bio
, pb_data_size
);
3153 policy_tick(cache
->policy
, false);
3155 spin_lock_irqsave(&cache
->lock
, flags
);
3156 cache
->need_tick_bio
= true;
3157 spin_unlock_irqrestore(&cache
->lock
, flags
);
3160 check_for_quiesced_migrations(cache
, pb
);
3161 accounted_complete(cache
, bio
);
3166 static int write_dirty_bitset(struct cache
*cache
)
3170 if (get_cache_mode(cache
) >= CM_READ_ONLY
)
3173 for (i
= 0; i
< from_cblock(cache
->cache_size
); i
++) {
3174 r
= dm_cache_set_dirty(cache
->cmd
, to_cblock(i
),
3175 is_dirty(cache
, to_cblock(i
)));
3177 metadata_operation_failed(cache
, "dm_cache_set_dirty", r
);
3185 static int write_discard_bitset(struct cache
*cache
)
3189 if (get_cache_mode(cache
) >= CM_READ_ONLY
)
3192 r
= dm_cache_discard_bitset_resize(cache
->cmd
, cache
->discard_block_size
,
3193 cache
->discard_nr_blocks
);
3195 DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache
));
3196 metadata_operation_failed(cache
, "dm_cache_discard_bitset_resize", r
);
3200 for (i
= 0; i
< from_dblock(cache
->discard_nr_blocks
); i
++) {
3201 r
= dm_cache_set_discard(cache
->cmd
, to_dblock(i
),
3202 is_discarded(cache
, to_dblock(i
)));
3204 metadata_operation_failed(cache
, "dm_cache_set_discard", r
);
3212 static int write_hints(struct cache
*cache
)
3216 if (get_cache_mode(cache
) >= CM_READ_ONLY
)
3219 r
= dm_cache_write_hints(cache
->cmd
, cache
->policy
);
3221 metadata_operation_failed(cache
, "dm_cache_write_hints", r
);
3229 * returns true on success
3231 static bool sync_metadata(struct cache
*cache
)
3235 r1
= write_dirty_bitset(cache
);
3237 DMERR("%s: could not write dirty bitset", cache_device_name(cache
));
3239 r2
= write_discard_bitset(cache
);
3241 DMERR("%s: could not write discard bitset", cache_device_name(cache
));
3245 r3
= write_hints(cache
);
3247 DMERR("%s: could not write hints", cache_device_name(cache
));
3250 * If writing the above metadata failed, we still commit, but don't
3251 * set the clean shutdown flag. This will effectively force every
3252 * dirty bit to be set on reload.
3254 r4
= commit(cache
, !r1
&& !r2
&& !r3
);
3256 DMERR("%s: could not write cache metadata", cache_device_name(cache
));
3258 return !r1
&& !r2
&& !r3
&& !r4
;
3261 static void cache_postsuspend(struct dm_target
*ti
)
3263 struct cache
*cache
= ti
->private;
3265 start_quiescing(cache
);
3266 wait_for_migrations(cache
);
3268 requeue_deferred_bios(cache
);
3269 requeue_deferred_cells(cache
);
3270 stop_quiescing(cache
);
3272 if (get_cache_mode(cache
) == CM_WRITE
)
3273 (void) sync_metadata(cache
);
3276 static int load_mapping(void *context
, dm_oblock_t oblock
, dm_cblock_t cblock
,
3277 bool dirty
, uint32_t hint
, bool hint_valid
)
3280 struct cache
*cache
= context
;
3282 r
= policy_load_mapping(cache
->policy
, oblock
, cblock
, hint
, hint_valid
);
3287 set_dirty(cache
, oblock
, cblock
);
3289 clear_dirty(cache
, oblock
, cblock
);
3295 * The discard block size in the on disk metadata is not
3296 * neccessarily the same as we're currently using. So we have to
3297 * be careful to only set the discarded attribute if we know it
3298 * covers a complete block of the new size.
3300 struct discard_load_info
{
3301 struct cache
*cache
;
3304 * These blocks are sized using the on disk dblock size, rather
3305 * than the current one.
3307 dm_block_t block_size
;
3308 dm_block_t discard_begin
, discard_end
;
3311 static void discard_load_info_init(struct cache
*cache
,
3312 struct discard_load_info
*li
)
3315 li
->discard_begin
= li
->discard_end
= 0;
3318 static void set_discard_range(struct discard_load_info
*li
)
3322 if (li
->discard_begin
== li
->discard_end
)
3326 * Convert to sectors.
3328 b
= li
->discard_begin
* li
->block_size
;
3329 e
= li
->discard_end
* li
->block_size
;
3332 * Then convert back to the current dblock size.
3334 b
= dm_sector_div_up(b
, li
->cache
->discard_block_size
);
3335 sector_div(e
, li
->cache
->discard_block_size
);
3338 * The origin may have shrunk, so we need to check we're still in
3341 if (e
> from_dblock(li
->cache
->discard_nr_blocks
))
3342 e
= from_dblock(li
->cache
->discard_nr_blocks
);
3345 set_discard(li
->cache
, to_dblock(b
));
3348 static int load_discard(void *context
, sector_t discard_block_size
,
3349 dm_dblock_t dblock
, bool discard
)
3351 struct discard_load_info
*li
= context
;
3353 li
->block_size
= discard_block_size
;
3356 if (from_dblock(dblock
) == li
->discard_end
)
3358 * We're already in a discard range, just extend it.
3360 li
->discard_end
= li
->discard_end
+ 1ULL;
3364 * Emit the old range and start a new one.
3366 set_discard_range(li
);
3367 li
->discard_begin
= from_dblock(dblock
);
3368 li
->discard_end
= li
->discard_begin
+ 1ULL;
3371 set_discard_range(li
);
3372 li
->discard_begin
= li
->discard_end
= 0;
3378 static dm_cblock_t
get_cache_dev_size(struct cache
*cache
)
3380 sector_t size
= get_dev_size(cache
->cache_dev
);
3381 (void) sector_div(size
, cache
->sectors_per_block
);
3382 return to_cblock(size
);
3385 static bool can_resize(struct cache
*cache
, dm_cblock_t new_size
)
3387 if (from_cblock(new_size
) > from_cblock(cache
->cache_size
))
3391 * We can't drop a dirty block when shrinking the cache.
3393 while (from_cblock(new_size
) < from_cblock(cache
->cache_size
)) {
3394 new_size
= to_cblock(from_cblock(new_size
) + 1);
3395 if (is_dirty(cache
, new_size
)) {
3396 DMERR("%s: unable to shrink cache; cache block %llu is dirty",
3397 cache_device_name(cache
),
3398 (unsigned long long) from_cblock(new_size
));
3406 static int resize_cache_dev(struct cache
*cache
, dm_cblock_t new_size
)
3410 r
= dm_cache_resize(cache
->cmd
, new_size
);
3412 DMERR("%s: could not resize cache metadata", cache_device_name(cache
));
3413 metadata_operation_failed(cache
, "dm_cache_resize", r
);
3417 set_cache_size(cache
, new_size
);
3422 static int cache_preresume(struct dm_target
*ti
)
3425 struct cache
*cache
= ti
->private;
3426 dm_cblock_t csize
= get_cache_dev_size(cache
);
3429 * Check to see if the cache has resized.
3431 if (!cache
->sized
) {
3432 r
= resize_cache_dev(cache
, csize
);
3436 cache
->sized
= true;
3438 } else if (csize
!= cache
->cache_size
) {
3439 if (!can_resize(cache
, csize
))
3442 r
= resize_cache_dev(cache
, csize
);
3447 if (!cache
->loaded_mappings
) {
3448 r
= dm_cache_load_mappings(cache
->cmd
, cache
->policy
,
3449 load_mapping
, cache
);
3451 DMERR("%s: could not load cache mappings", cache_device_name(cache
));
3452 metadata_operation_failed(cache
, "dm_cache_load_mappings", r
);
3456 cache
->loaded_mappings
= true;
3459 if (!cache
->loaded_discards
) {
3460 struct discard_load_info li
;
3463 * The discard bitset could have been resized, or the
3464 * discard block size changed. To be safe we start by
3465 * setting every dblock to not discarded.
3467 clear_bitset(cache
->discard_bitset
, from_dblock(cache
->discard_nr_blocks
));
3469 discard_load_info_init(cache
, &li
);
3470 r
= dm_cache_load_discards(cache
->cmd
, load_discard
, &li
);
3472 DMERR("%s: could not load origin discards", cache_device_name(cache
));
3473 metadata_operation_failed(cache
, "dm_cache_load_discards", r
);
3476 set_discard_range(&li
);
3478 cache
->loaded_discards
= true;
3484 static void cache_resume(struct dm_target
*ti
)
3486 struct cache
*cache
= ti
->private;
3488 cache
->need_tick_bio
= true;
3489 do_waker(&cache
->waker
.work
);
3495 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
3496 * <cache block size> <#used cache blocks>/<#total cache blocks>
3497 * <#read hits> <#read misses> <#write hits> <#write misses>
3498 * <#demotions> <#promotions> <#dirty>
3499 * <#features> <features>*
3500 * <#core args> <core args>
3501 * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check>
3503 static void cache_status(struct dm_target
*ti
, status_type_t type
,
3504 unsigned status_flags
, char *result
, unsigned maxlen
)
3509 dm_block_t nr_free_blocks_metadata
= 0;
3510 dm_block_t nr_blocks_metadata
= 0;
3511 char buf
[BDEVNAME_SIZE
];
3512 struct cache
*cache
= ti
->private;
3513 dm_cblock_t residency
;
3517 case STATUSTYPE_INFO
:
3518 if (get_cache_mode(cache
) == CM_FAIL
) {
3523 /* Commit to ensure statistics aren't out-of-date */
3524 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
3525 (void) commit(cache
, false);
3527 r
= dm_cache_get_free_metadata_block_count(cache
->cmd
, &nr_free_blocks_metadata
);
3529 DMERR("%s: dm_cache_get_free_metadata_block_count returned %d",
3530 cache_device_name(cache
), r
);
3534 r
= dm_cache_get_metadata_dev_size(cache
->cmd
, &nr_blocks_metadata
);
3536 DMERR("%s: dm_cache_get_metadata_dev_size returned %d",
3537 cache_device_name(cache
), r
);
3541 residency
= policy_residency(cache
->policy
);
3543 DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ",
3544 (unsigned)DM_CACHE_METADATA_BLOCK_SIZE
,
3545 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
3546 (unsigned long long)nr_blocks_metadata
,
3547 cache
->sectors_per_block
,
3548 (unsigned long long) from_cblock(residency
),
3549 (unsigned long long) from_cblock(cache
->cache_size
),
3550 (unsigned) atomic_read(&cache
->stats
.read_hit
),
3551 (unsigned) atomic_read(&cache
->stats
.read_miss
),
3552 (unsigned) atomic_read(&cache
->stats
.write_hit
),
3553 (unsigned) atomic_read(&cache
->stats
.write_miss
),
3554 (unsigned) atomic_read(&cache
->stats
.demotion
),
3555 (unsigned) atomic_read(&cache
->stats
.promotion
),
3556 (unsigned long) atomic_read(&cache
->nr_dirty
));
3558 if (writethrough_mode(&cache
->features
))
3559 DMEMIT("1 writethrough ");
3561 else if (passthrough_mode(&cache
->features
))
3562 DMEMIT("1 passthrough ");
3564 else if (writeback_mode(&cache
->features
))
3565 DMEMIT("1 writeback ");
3568 DMERR("%s: internal error: unknown io mode: %d",
3569 cache_device_name(cache
), (int) cache
->features
.io_mode
);
3573 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache
->migration_threshold
);
3575 DMEMIT("%s ", dm_cache_policy_get_name(cache
->policy
));
3577 r
= policy_emit_config_values(cache
->policy
, result
, maxlen
, &sz
);
3579 DMERR("%s: policy_emit_config_values returned %d",
3580 cache_device_name(cache
), r
);
3583 if (get_cache_mode(cache
) == CM_READ_ONLY
)
3588 r
= dm_cache_metadata_needs_check(cache
->cmd
, &needs_check
);
3590 if (r
|| needs_check
)
3591 DMEMIT("needs_check ");
3597 case STATUSTYPE_TABLE
:
3598 format_dev_t(buf
, cache
->metadata_dev
->bdev
->bd_dev
);
3600 format_dev_t(buf
, cache
->cache_dev
->bdev
->bd_dev
);
3602 format_dev_t(buf
, cache
->origin_dev
->bdev
->bd_dev
);
3605 for (i
= 0; i
< cache
->nr_ctr_args
- 1; i
++)
3606 DMEMIT(" %s", cache
->ctr_args
[i
]);
3607 if (cache
->nr_ctr_args
)
3608 DMEMIT(" %s", cache
->ctr_args
[cache
->nr_ctr_args
- 1]);
3618 * A cache block range can take two forms:
3620 * i) A single cblock, eg. '3456'
3621 * ii) A begin and end cblock with dots between, eg. 123-234
3623 static int parse_cblock_range(struct cache
*cache
, const char *str
,
3624 struct cblock_range
*result
)
3631 * Try and parse form (ii) first.
3633 r
= sscanf(str
, "%llu-%llu%c", &b
, &e
, &dummy
);
3638 result
->begin
= to_cblock(b
);
3639 result
->end
= to_cblock(e
);
3644 * That didn't work, try form (i).
3646 r
= sscanf(str
, "%llu%c", &b
, &dummy
);
3651 result
->begin
= to_cblock(b
);
3652 result
->end
= to_cblock(from_cblock(result
->begin
) + 1u);
3656 DMERR("%s: invalid cblock range '%s'", cache_device_name(cache
), str
);
3660 static int validate_cblock_range(struct cache
*cache
, struct cblock_range
*range
)
3662 uint64_t b
= from_cblock(range
->begin
);
3663 uint64_t e
= from_cblock(range
->end
);
3664 uint64_t n
= from_cblock(cache
->cache_size
);
3667 DMERR("%s: begin cblock out of range: %llu >= %llu",
3668 cache_device_name(cache
), b
, n
);
3673 DMERR("%s: end cblock out of range: %llu > %llu",
3674 cache_device_name(cache
), e
, n
);
3679 DMERR("%s: invalid cblock range: %llu >= %llu",
3680 cache_device_name(cache
), b
, e
);
3687 static int request_invalidation(struct cache
*cache
, struct cblock_range
*range
)
3689 struct invalidation_request req
;
3691 INIT_LIST_HEAD(&req
.list
);
3692 req
.cblocks
= range
;
3693 atomic_set(&req
.complete
, 0);
3695 init_waitqueue_head(&req
.result_wait
);
3697 spin_lock(&cache
->invalidation_lock
);
3698 list_add(&req
.list
, &cache
->invalidation_requests
);
3699 spin_unlock(&cache
->invalidation_lock
);
3702 wait_event(req
.result_wait
, atomic_read(&req
.complete
));
3706 static int process_invalidate_cblocks_message(struct cache
*cache
, unsigned count
,
3707 const char **cblock_ranges
)
3711 struct cblock_range range
;
3713 if (!passthrough_mode(&cache
->features
)) {
3714 DMERR("%s: cache has to be in passthrough mode for invalidation",
3715 cache_device_name(cache
));
3719 for (i
= 0; i
< count
; i
++) {
3720 r
= parse_cblock_range(cache
, cblock_ranges
[i
], &range
);
3724 r
= validate_cblock_range(cache
, &range
);
3729 * Pass begin and end origin blocks to the worker and wake it.
3731 r
= request_invalidation(cache
, &range
);
3743 * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]*
3745 * The key migration_threshold is supported by the cache target core.
3747 static int cache_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
3749 struct cache
*cache
= ti
->private;
3754 if (get_cache_mode(cache
) >= CM_READ_ONLY
) {
3755 DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode",
3756 cache_device_name(cache
));
3760 if (!strcasecmp(argv
[0], "invalidate_cblocks"))
3761 return process_invalidate_cblocks_message(cache
, argc
- 1, (const char **) argv
+ 1);
3766 return set_config_value(cache
, argv
[0], argv
[1]);
3769 static int cache_iterate_devices(struct dm_target
*ti
,
3770 iterate_devices_callout_fn fn
, void *data
)
3773 struct cache
*cache
= ti
->private;
3775 r
= fn(ti
, cache
->cache_dev
, 0, get_dev_size(cache
->cache_dev
), data
);
3777 r
= fn(ti
, cache
->origin_dev
, 0, ti
->len
, data
);
3782 static void set_discard_limits(struct cache
*cache
, struct queue_limits
*limits
)
3785 * FIXME: these limits may be incompatible with the cache device
3787 limits
->max_discard_sectors
= min_t(sector_t
, cache
->discard_block_size
* 1024,
3788 cache
->origin_sectors
);
3789 limits
->discard_granularity
= cache
->discard_block_size
<< SECTOR_SHIFT
;
3792 static void cache_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3794 struct cache
*cache
= ti
->private;
3795 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
3798 * If the system-determined stacked limits are compatible with the
3799 * cache's blocksize (io_opt is a factor) do not override them.
3801 if (io_opt_sectors
< cache
->sectors_per_block
||
3802 do_div(io_opt_sectors
, cache
->sectors_per_block
)) {
3803 blk_limits_io_min(limits
, cache
->sectors_per_block
<< SECTOR_SHIFT
);
3804 blk_limits_io_opt(limits
, cache
->sectors_per_block
<< SECTOR_SHIFT
);
3806 set_discard_limits(cache
, limits
);
3809 /*----------------------------------------------------------------*/
3811 static struct target_type cache_target
= {
3813 .version
= {1, 9, 0},
3814 .module
= THIS_MODULE
,
3818 .end_io
= cache_end_io
,
3819 .postsuspend
= cache_postsuspend
,
3820 .preresume
= cache_preresume
,
3821 .resume
= cache_resume
,
3822 .status
= cache_status
,
3823 .message
= cache_message
,
3824 .iterate_devices
= cache_iterate_devices
,
3825 .io_hints
= cache_io_hints
,
3828 static int __init
dm_cache_init(void)
3832 r
= dm_register_target(&cache_target
);
3834 DMERR("cache target registration failed: %d", r
);
3838 migration_cache
= KMEM_CACHE(dm_cache_migration
, 0);
3839 if (!migration_cache
) {
3840 dm_unregister_target(&cache_target
);
3847 static void __exit
dm_cache_exit(void)
3849 dm_unregister_target(&cache_target
);
3850 kmem_cache_destroy(migration_cache
);
3853 module_init(dm_cache_init
);
3854 module_exit(dm_cache_exit
);
3856 MODULE_DESCRIPTION(DM_NAME
" cache target");
3857 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
3858 MODULE_LICENSE("GPL");