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;
124 static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio,
125 bio_end_io_t *bi_end_io, void *bi_private)
127 h->bi_end_io = bio->bi_end_io;
128 h->bi_private = bio->bi_private;
130 bio->bi_end_io = bi_end_io;
131 bio->bi_private = bi_private;
134 static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio)
136 bio->bi_end_io = h->bi_end_io;
137 bio->bi_private = h->bi_private;
140 /*----------------------------------------------------------------*/
142 #define MIGRATION_POOL_SIZE 128
143 #define COMMIT_PERIOD HZ
144 #define MIGRATION_COUNT_WINDOW 10
147 * The block size of the device holding cache data must be
148 * between 32KB and 1GB.
150 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT)
151 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
153 enum cache_metadata_mode {
154 CM_WRITE, /* metadata may be changed */
155 CM_READ_ONLY, /* metadata may not be changed */
161 * Data is written to cached blocks only. These blocks are marked
162 * dirty. If you lose the cache device you will lose data.
163 * Potential performance increase for both reads and writes.
168 * Data is written to both cache and origin. Blocks are never
169 * dirty. Potential performance benfit for reads only.
174 * A degraded mode useful for various cache coherency situations
175 * (eg, rolling back snapshots). Reads and writes always go to the
176 * origin. If a write goes to a cached oblock, then the cache
177 * block is invalidated.
182 struct cache_features {
183 enum cache_metadata_mode mode;
184 enum cache_io_mode io_mode;
194 atomic_t copies_avoided;
195 atomic_t cache_cell_clash;
196 atomic_t commit_count;
197 atomic_t discard_count;
201 * Defines a range of cblocks, begin to (end - 1) are in the range. end is
202 * the one-past-the-end value.
204 struct cblock_range {
209 struct invalidation_request {
210 struct list_head list;
211 struct cblock_range *cblocks;
216 wait_queue_head_t result_wait;
220 struct dm_target *ti;
221 struct dm_target_callbacks callbacks;
223 struct dm_cache_metadata *cmd;
226 * Metadata is written to this device.
228 struct dm_dev *metadata_dev;
231 * The slower of the two data devices. Typically a spindle.
233 struct dm_dev *origin_dev;
236 * The faster of the two data devices. Typically an SSD.
238 struct dm_dev *cache_dev;
241 * Size of the origin device in _complete_ blocks and native sectors.
243 dm_oblock_t origin_blocks;
244 sector_t origin_sectors;
247 * Size of the cache device in blocks.
249 dm_cblock_t cache_size;
252 * Fields for converting from sectors to blocks.
254 uint32_t sectors_per_block;
255 int sectors_per_block_shift;
258 struct list_head deferred_cells;
259 struct bio_list deferred_bios;
260 struct bio_list deferred_flush_bios;
261 struct bio_list deferred_writethrough_bios;
262 struct list_head quiesced_migrations;
263 struct list_head completed_migrations;
264 struct list_head need_commit_migrations;
265 sector_t migration_threshold;
266 wait_queue_head_t migration_wait;
267 atomic_t nr_allocated_migrations;
270 * The number of in flight migrations that are performing
271 * background io. eg, promotion, writeback.
273 atomic_t nr_io_migrations;
275 wait_queue_head_t quiescing_wait;
277 atomic_t quiescing_ack;
280 * cache_size entries, dirty if set
283 unsigned long *dirty_bitset;
286 * origin_blocks entries, discarded if set.
288 dm_dblock_t discard_nr_blocks;
289 unsigned long *discard_bitset;
290 uint32_t discard_block_size; /* a power of 2 times sectors per block */
293 * Rather than reconstructing the table line for the status we just
294 * save it and regurgitate.
296 unsigned nr_ctr_args;
297 const char **ctr_args;
299 struct dm_kcopyd_client *copier;
300 struct workqueue_struct *wq;
301 struct work_struct worker;
303 struct delayed_work waker;
304 unsigned long last_commit_jiffies;
306 struct dm_bio_prison *prison;
307 struct dm_deferred_set *all_io_ds;
309 mempool_t *migration_pool;
311 struct dm_cache_policy *policy;
312 unsigned policy_nr_args;
314 bool need_tick_bio:1;
317 bool commit_requested:1;
318 bool loaded_mappings:1;
319 bool loaded_discards:1;
322 * Cache features such as write-through.
324 struct cache_features features;
326 struct cache_stats stats;
329 * Invalidation fields.
331 spinlock_t invalidation_lock;
332 struct list_head invalidation_requests;
334 struct io_tracker origin_tracker;
337 struct per_bio_data {
340 struct dm_deferred_entry *all_io_entry;
341 struct dm_hook_info hook_info;
345 * writethrough fields. These MUST remain at the end of this
346 * structure and the 'cache' member must be the first as it
347 * is used to determine the offset of the writethrough fields.
351 struct dm_bio_details bio_details;
354 struct dm_cache_migration {
355 struct list_head list;
358 unsigned long start_jiffies;
359 dm_oblock_t old_oblock;
360 dm_oblock_t new_oblock;
368 bool requeue_holder:1;
371 struct dm_bio_prison_cell *old_ocell;
372 struct dm_bio_prison_cell *new_ocell;
376 * Processing a bio in the worker thread may require these memory
377 * allocations. We prealloc to avoid deadlocks (the same worker thread
378 * frees them back to the mempool).
381 struct dm_cache_migration *mg;
382 struct dm_bio_prison_cell *cell1;
383 struct dm_bio_prison_cell *cell2;
386 static enum cache_metadata_mode get_cache_mode(struct cache *cache);
388 static void wake_worker(struct cache *cache)
390 queue_work(cache->wq, &cache->worker);
393 /*----------------------------------------------------------------*/
395 static struct dm_bio_prison_cell *alloc_prison_cell(struct cache *cache)
397 /* FIXME: change to use a local slab. */
398 return dm_bio_prison_alloc_cell(cache->prison, GFP_NOWAIT);
401 static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell *cell)
403 dm_bio_prison_free_cell(cache->prison, cell);
406 static struct dm_cache_migration *alloc_migration(struct cache *cache)
408 struct dm_cache_migration *mg;
410 mg = mempool_alloc(cache->migration_pool, GFP_NOWAIT);
413 atomic_inc(&mg->cache->nr_allocated_migrations);
419 static void free_migration(struct dm_cache_migration *mg)
421 struct cache *cache = mg->cache;
423 if (atomic_dec_and_test(&cache->nr_allocated_migrations))
424 wake_up(&cache->migration_wait);
426 mempool_free(mg, cache->migration_pool);
429 static int prealloc_data_structs(struct cache *cache, struct prealloc *p)
432 p->mg = alloc_migration(cache);
438 p->cell1 = alloc_prison_cell(cache);
444 p->cell2 = alloc_prison_cell(cache);
452 static void prealloc_free_structs(struct cache *cache, struct prealloc *p)
455 free_prison_cell(cache, p->cell2);
458 free_prison_cell(cache, p->cell1);
461 free_migration(p->mg);
464 static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p)
466 struct dm_cache_migration *mg = p->mg;
475 * You must have a cell within the prealloc struct to return. If not this
476 * function will BUG() rather than returning NULL.
478 static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p)
480 struct dm_bio_prison_cell *r = NULL;
486 } else if (p->cell2) {
496 * You can't have more than two cells in a prealloc struct. BUG() will be
497 * called if you try and overfill.
499 static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell)
511 /*----------------------------------------------------------------*/
513 static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key *key)
517 key->block_begin = from_oblock(begin);
518 key->block_end = from_oblock(end);
522 * The caller hands in a preallocated cell, and a free function for it.
523 * The cell will be freed if there's an error, or if it wasn't used because
524 * a cell with that key already exists.
526 typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell);
528 static int bio_detain_range(struct cache *cache, dm_oblock_t oblock_begin, dm_oblock_t oblock_end,
529 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
530 cell_free_fn free_fn, void *free_context,
531 struct dm_bio_prison_cell **cell_result)
534 struct dm_cell_key key;
536 build_key(oblock_begin, oblock_end, &key);
537 r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result);
539 free_fn(free_context, cell_prealloc);
544 static int bio_detain(struct cache *cache, dm_oblock_t oblock,
545 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
546 cell_free_fn free_fn, void *free_context,
547 struct dm_bio_prison_cell **cell_result)
549 dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL);
550 return bio_detain_range(cache, oblock, end, bio,
551 cell_prealloc, free_fn, free_context, cell_result);
554 static int get_cell(struct cache *cache,
556 struct prealloc *structs,
557 struct dm_bio_prison_cell **cell_result)
560 struct dm_cell_key key;
561 struct dm_bio_prison_cell *cell_prealloc;
563 cell_prealloc = prealloc_get_cell(structs);
565 build_key(oblock, to_oblock(from_oblock(oblock) + 1ULL), &key);
566 r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result);
568 prealloc_put_cell(structs, cell_prealloc);
573 /*----------------------------------------------------------------*/
575 static bool is_dirty(struct cache *cache, dm_cblock_t b)
577 return test_bit(from_cblock(b), cache->dirty_bitset);
580 static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
582 if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) {
583 atomic_inc(&cache->nr_dirty);
584 policy_set_dirty(cache->policy, oblock);
588 static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
590 if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) {
591 policy_clear_dirty(cache->policy, oblock);
592 if (atomic_dec_return(&cache->nr_dirty) == 0)
593 dm_table_event(cache->ti->table);
597 /*----------------------------------------------------------------*/
599 static bool block_size_is_power_of_two(struct cache *cache)
601 return cache->sectors_per_block_shift >= 0;
604 /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */
605 #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6
608 static dm_block_t block_div(dm_block_t b, uint32_t n)
615 static dm_block_t oblocks_per_dblock(struct cache *cache)
617 dm_block_t oblocks = cache->discard_block_size;
619 if (block_size_is_power_of_two(cache))
620 oblocks >>= cache->sectors_per_block_shift;
622 oblocks = block_div(oblocks, cache->sectors_per_block);
627 static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock)
629 return to_dblock(block_div(from_oblock(oblock),
630 oblocks_per_dblock(cache)));
633 static dm_oblock_t dblock_to_oblock(struct cache *cache, dm_dblock_t dblock)
635 return to_oblock(from_dblock(dblock) * oblocks_per_dblock(cache));
638 static void set_discard(struct cache *cache, dm_dblock_t b)
642 BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks));
643 atomic_inc(&cache->stats.discard_count);
645 spin_lock_irqsave(&cache->lock, flags);
646 set_bit(from_dblock(b), cache->discard_bitset);
647 spin_unlock_irqrestore(&cache->lock, flags);
650 static void clear_discard(struct cache *cache, dm_dblock_t b)
654 spin_lock_irqsave(&cache->lock, flags);
655 clear_bit(from_dblock(b), cache->discard_bitset);
656 spin_unlock_irqrestore(&cache->lock, flags);
659 static bool is_discarded(struct cache *cache, dm_dblock_t b)
664 spin_lock_irqsave(&cache->lock, flags);
665 r = test_bit(from_dblock(b), cache->discard_bitset);
666 spin_unlock_irqrestore(&cache->lock, flags);
671 static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b)
676 spin_lock_irqsave(&cache->lock, flags);
677 r = test_bit(from_dblock(oblock_to_dblock(cache, b)),
678 cache->discard_bitset);
679 spin_unlock_irqrestore(&cache->lock, flags);
684 /*----------------------------------------------------------------*/
686 static void load_stats(struct cache *cache)
688 struct dm_cache_statistics stats;
690 dm_cache_metadata_get_stats(cache->cmd, &stats);
691 atomic_set(&cache->stats.read_hit, stats.read_hits);
692 atomic_set(&cache->stats.read_miss, stats.read_misses);
693 atomic_set(&cache->stats.write_hit, stats.write_hits);
694 atomic_set(&cache->stats.write_miss, stats.write_misses);
697 static void save_stats(struct cache *cache)
699 struct dm_cache_statistics stats;
701 if (get_cache_mode(cache) >= CM_READ_ONLY)
704 stats.read_hits = atomic_read(&cache->stats.read_hit);
705 stats.read_misses = atomic_read(&cache->stats.read_miss);
706 stats.write_hits = atomic_read(&cache->stats.write_hit);
707 stats.write_misses = atomic_read(&cache->stats.write_miss);
709 dm_cache_metadata_set_stats(cache->cmd, &stats);
712 /*----------------------------------------------------------------
714 *--------------------------------------------------------------*/
717 * If using writeback, leave out struct per_bio_data's writethrough fields.
719 #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache))
720 #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data))
722 static bool writethrough_mode(struct cache_features *f)
724 return f->io_mode == CM_IO_WRITETHROUGH;
727 static bool writeback_mode(struct cache_features *f)
729 return f->io_mode == CM_IO_WRITEBACK;
732 static bool passthrough_mode(struct cache_features *f)
734 return f->io_mode == CM_IO_PASSTHROUGH;
737 static size_t get_per_bio_data_size(struct cache *cache)
739 return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB;
742 static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size)
744 struct per_bio_data *pb = dm_per_bio_data(bio, data_size);
749 static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size)
751 struct per_bio_data *pb = get_per_bio_data(bio, data_size);
754 pb->req_nr = dm_bio_get_target_bio_nr(bio);
755 pb->all_io_entry = NULL;
761 /*----------------------------------------------------------------
763 *--------------------------------------------------------------*/
764 static void remap_to_origin(struct cache *cache, struct bio *bio)
766 bio->bi_bdev = cache->origin_dev->bdev;
769 static void remap_to_cache(struct cache *cache, struct bio *bio,
772 sector_t bi_sector = bio->bi_iter.bi_sector;
773 sector_t block = from_cblock(cblock);
775 bio->bi_bdev = cache->cache_dev->bdev;
776 if (!block_size_is_power_of_two(cache))
777 bio->bi_iter.bi_sector =
778 (block * cache->sectors_per_block) +
779 sector_div(bi_sector, cache->sectors_per_block);
781 bio->bi_iter.bi_sector =
782 (block << cache->sectors_per_block_shift) |
783 (bi_sector & (cache->sectors_per_block - 1));
786 static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio)
789 size_t pb_data_size = get_per_bio_data_size(cache);
790 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
792 spin_lock_irqsave(&cache->lock, flags);
793 if (cache->need_tick_bio &&
794 !(bio->bi_rw & (REQ_FUA | REQ_FLUSH | REQ_DISCARD))) {
796 cache->need_tick_bio = false;
798 spin_unlock_irqrestore(&cache->lock, flags);
801 static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio,
804 check_if_tick_bio_needed(cache, bio);
805 remap_to_origin(cache, bio);
806 if (bio_data_dir(bio) == WRITE)
807 clear_discard(cache, oblock_to_dblock(cache, oblock));
810 static void remap_to_cache_dirty(struct cache *cache, struct bio *bio,
811 dm_oblock_t oblock, dm_cblock_t cblock)
813 check_if_tick_bio_needed(cache, bio);
814 remap_to_cache(cache, bio, cblock);
815 if (bio_data_dir(bio) == WRITE) {
816 set_dirty(cache, oblock, cblock);
817 clear_discard(cache, oblock_to_dblock(cache, oblock));
821 static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio)
823 sector_t block_nr = bio->bi_iter.bi_sector;
825 if (!block_size_is_power_of_two(cache))
826 (void) sector_div(block_nr, cache->sectors_per_block);
828 block_nr >>= cache->sectors_per_block_shift;
830 return to_oblock(block_nr);
833 static int bio_triggers_commit(struct cache *cache, struct bio *bio)
835 return bio->bi_rw & (REQ_FLUSH | REQ_FUA);
839 * You must increment the deferred set whilst the prison cell is held. To
840 * encourage this, we ask for 'cell' to be passed in.
842 static void inc_ds(struct cache *cache, struct bio *bio,
843 struct dm_bio_prison_cell *cell)
845 size_t pb_data_size = get_per_bio_data_size(cache);
846 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
849 BUG_ON(pb->all_io_entry);
851 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
854 static bool accountable_bio(struct cache *cache, struct bio *bio)
856 return ((bio->bi_bdev == cache->origin_dev->bdev) &&
857 !(bio->bi_rw & REQ_DISCARD));
860 static void accounted_begin(struct cache *cache, struct bio *bio)
862 size_t pb_data_size = get_per_bio_data_size(cache);
863 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
865 if (accountable_bio(cache, bio)) {
866 pb->len = bio_sectors(bio);
867 iot_io_begin(&cache->origin_tracker, pb->len);
871 static void accounted_complete(struct cache *cache, struct bio *bio)
873 size_t pb_data_size = get_per_bio_data_size(cache);
874 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
876 iot_io_end(&cache->origin_tracker, pb->len);
879 static void accounted_request(struct cache *cache, struct bio *bio)
881 accounted_begin(cache, bio);
882 generic_make_request(bio);
885 static void issue(struct cache *cache, struct bio *bio)
889 if (!bio_triggers_commit(cache, bio)) {
890 accounted_request(cache, bio);
895 * Batch together any bios that trigger commits and then issue a
896 * single commit for them in do_worker().
898 spin_lock_irqsave(&cache->lock, flags);
899 cache->commit_requested = true;
900 bio_list_add(&cache->deferred_flush_bios, bio);
901 spin_unlock_irqrestore(&cache->lock, flags);
904 static void inc_and_issue(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell *cell)
906 inc_ds(cache, bio, cell);
910 static void defer_writethrough_bio(struct cache *cache, struct bio *bio)
914 spin_lock_irqsave(&cache->lock, flags);
915 bio_list_add(&cache->deferred_writethrough_bios, bio);
916 spin_unlock_irqrestore(&cache->lock, flags);
921 static void writethrough_endio(struct bio *bio)
923 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
925 dm_unhook_bio(&pb->hook_info, bio);
932 dm_bio_restore(&pb->bio_details, bio);
933 remap_to_cache(pb->cache, bio, pb->cblock);
936 * We can't issue this bio directly, since we're in interrupt
937 * context. So it gets put on a bio list for processing by the
940 defer_writethrough_bio(pb->cache, bio);
944 * When running in writethrough mode we need to send writes to clean blocks
945 * to both the cache and origin devices. In future we'd like to clone the
946 * bio and send them in parallel, but for now we're doing them in
947 * series as this is easier.
949 static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio,
950 dm_oblock_t oblock, dm_cblock_t cblock)
952 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
956 dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL);
957 dm_bio_record(&pb->bio_details, bio);
959 remap_to_origin_clear_discard(pb->cache, bio, oblock);
962 /*----------------------------------------------------------------
964 *--------------------------------------------------------------*/
965 static enum cache_metadata_mode get_cache_mode(struct cache *cache)
967 return cache->features.mode;
970 static const char *cache_device_name(struct cache *cache)
972 return dm_device_name(dm_table_get_md(cache->ti->table));
975 static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode)
977 const char *descs[] = {
983 dm_table_event(cache->ti->table);
984 DMINFO("%s: switching cache to %s mode",
985 cache_device_name(cache), descs[(int)mode]);
988 static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode)
990 bool needs_check = dm_cache_metadata_needs_check(cache->cmd);
991 enum cache_metadata_mode old_mode = get_cache_mode(cache);
993 if (new_mode == CM_WRITE && needs_check) {
994 DMERR("%s: unable to switch cache to write mode until repaired.",
995 cache_device_name(cache));
996 if (old_mode != new_mode)
999 new_mode = CM_READ_ONLY;
1002 /* Never move out of fail mode */
1003 if (old_mode == CM_FAIL)
1009 dm_cache_metadata_set_read_only(cache->cmd);
1013 dm_cache_metadata_set_read_write(cache->cmd);
1017 cache->features.mode = new_mode;
1019 if (new_mode != old_mode)
1020 notify_mode_switch(cache, new_mode);
1023 static void abort_transaction(struct cache *cache)
1025 const char *dev_name = cache_device_name(cache);
1027 if (get_cache_mode(cache) >= CM_READ_ONLY)
1030 if (dm_cache_metadata_set_needs_check(cache->cmd)) {
1031 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
1032 set_cache_mode(cache, CM_FAIL);
1035 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
1036 if (dm_cache_metadata_abort(cache->cmd)) {
1037 DMERR("%s: failed to abort metadata transaction", dev_name);
1038 set_cache_mode(cache, CM_FAIL);
1042 static void metadata_operation_failed(struct cache *cache, const char *op, int r)
1044 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1045 cache_device_name(cache), op, r);
1046 abort_transaction(cache);
1047 set_cache_mode(cache, CM_READ_ONLY);
1050 /*----------------------------------------------------------------
1051 * Migration processing
1053 * Migration covers moving data from the origin device to the cache, or
1055 *--------------------------------------------------------------*/
1056 static void inc_io_migrations(struct cache *cache)
1058 atomic_inc(&cache->nr_io_migrations);
1061 static void dec_io_migrations(struct cache *cache)
1063 atomic_dec(&cache->nr_io_migrations);
1066 static bool discard_or_flush(struct bio *bio)
1068 return bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_DISCARD);
1071 static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell)
1073 if (discard_or_flush(cell->holder)) {
1075 * We have to handle these bios individually.
1077 dm_cell_release(cache->prison, cell, &cache->deferred_bios);
1078 free_prison_cell(cache, cell);
1080 list_add_tail(&cell->user_list, &cache->deferred_cells);
1083 static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, bool holder)
1085 unsigned long flags;
1087 if (!holder && dm_cell_promote_or_release(cache->prison, cell)) {
1089 * There was no prisoner to promote to holder, the
1090 * cell has been released.
1092 free_prison_cell(cache, cell);
1096 spin_lock_irqsave(&cache->lock, flags);
1097 __cell_defer(cache, cell);
1098 spin_unlock_irqrestore(&cache->lock, flags);
1103 static void cell_error_with_code(struct cache *cache, struct dm_bio_prison_cell *cell, int err)
1105 dm_cell_error(cache->prison, cell, err);
1106 free_prison_cell(cache, cell);
1109 static void cell_requeue(struct cache *cache, struct dm_bio_prison_cell *cell)
1111 cell_error_with_code(cache, cell, DM_ENDIO_REQUEUE);
1114 static void free_io_migration(struct dm_cache_migration *mg)
1116 struct cache *cache = mg->cache;
1118 dec_io_migrations(cache);
1123 static void migration_failure(struct dm_cache_migration *mg)
1125 struct cache *cache = mg->cache;
1126 const char *dev_name = cache_device_name(cache);
1128 if (mg->writeback) {
1129 DMERR_LIMIT("%s: writeback failed; couldn't copy block", dev_name);
1130 set_dirty(cache, mg->old_oblock, mg->cblock);
1131 cell_defer(cache, mg->old_ocell, false);
1133 } else if (mg->demote) {
1134 DMERR_LIMIT("%s: demotion failed; couldn't copy block", dev_name);
1135 policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock);
1137 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1139 cell_defer(cache, mg->new_ocell, true);
1141 DMERR_LIMIT("%s: promotion failed; couldn't copy block", dev_name);
1142 policy_remove_mapping(cache->policy, mg->new_oblock);
1143 cell_defer(cache, mg->new_ocell, true);
1146 free_io_migration(mg);
1149 static void migration_success_pre_commit(struct dm_cache_migration *mg)
1152 unsigned long flags;
1153 struct cache *cache = mg->cache;
1155 if (mg->writeback) {
1156 clear_dirty(cache, mg->old_oblock, mg->cblock);
1157 cell_defer(cache, mg->old_ocell, false);
1158 free_io_migration(mg);
1161 } else if (mg->demote) {
1162 r = dm_cache_remove_mapping(cache->cmd, mg->cblock);
1164 DMERR_LIMIT("%s: demotion failed; couldn't update on disk metadata",
1165 cache_device_name(cache));
1166 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
1167 policy_force_mapping(cache->policy, mg->new_oblock,
1170 cell_defer(cache, mg->new_ocell, true);
1171 free_io_migration(mg);
1175 r = dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock);
1177 DMERR_LIMIT("%s: promotion failed; couldn't update on disk metadata",
1178 cache_device_name(cache));
1179 metadata_operation_failed(cache, "dm_cache_insert_mapping", r);
1180 policy_remove_mapping(cache->policy, mg->new_oblock);
1181 free_io_migration(mg);
1186 spin_lock_irqsave(&cache->lock, flags);
1187 list_add_tail(&mg->list, &cache->need_commit_migrations);
1188 cache->commit_requested = true;
1189 spin_unlock_irqrestore(&cache->lock, flags);
1192 static void migration_success_post_commit(struct dm_cache_migration *mg)
1194 unsigned long flags;
1195 struct cache *cache = mg->cache;
1197 if (mg->writeback) {
1198 DMWARN_LIMIT("%s: writeback unexpectedly triggered commit",
1199 cache_device_name(cache));
1202 } else if (mg->demote) {
1203 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1208 spin_lock_irqsave(&cache->lock, flags);
1209 list_add_tail(&mg->list, &cache->quiesced_migrations);
1210 spin_unlock_irqrestore(&cache->lock, flags);
1214 policy_remove_mapping(cache->policy, mg->old_oblock);
1215 free_io_migration(mg);
1219 if (mg->requeue_holder) {
1220 clear_dirty(cache, mg->new_oblock, mg->cblock);
1221 cell_defer(cache, mg->new_ocell, true);
1224 * The block was promoted via an overwrite, so it's dirty.
1226 set_dirty(cache, mg->new_oblock, mg->cblock);
1227 bio_endio(mg->new_ocell->holder);
1228 cell_defer(cache, mg->new_ocell, false);
1230 free_io_migration(mg);
1234 static void copy_complete(int read_err, unsigned long write_err, void *context)
1236 unsigned long flags;
1237 struct dm_cache_migration *mg = (struct dm_cache_migration *) context;
1238 struct cache *cache = mg->cache;
1240 if (read_err || write_err)
1243 spin_lock_irqsave(&cache->lock, flags);
1244 list_add_tail(&mg->list, &cache->completed_migrations);
1245 spin_unlock_irqrestore(&cache->lock, flags);
1250 static void issue_copy(struct dm_cache_migration *mg)
1253 struct dm_io_region o_region, c_region;
1254 struct cache *cache = mg->cache;
1255 sector_t cblock = from_cblock(mg->cblock);
1257 o_region.bdev = cache->origin_dev->bdev;
1258 o_region.count = cache->sectors_per_block;
1260 c_region.bdev = cache->cache_dev->bdev;
1261 c_region.sector = cblock * cache->sectors_per_block;
1262 c_region.count = cache->sectors_per_block;
1264 if (mg->writeback || mg->demote) {
1266 o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block;
1267 r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg);
1270 o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block;
1271 r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg);
1275 DMERR_LIMIT("%s: issuing migration failed", cache_device_name(cache));
1276 migration_failure(mg);
1280 static void overwrite_endio(struct bio *bio)
1282 struct dm_cache_migration *mg = bio->bi_private;
1283 struct cache *cache = mg->cache;
1284 size_t pb_data_size = get_per_bio_data_size(cache);
1285 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1286 unsigned long flags;
1288 dm_unhook_bio(&pb->hook_info, bio);
1293 mg->requeue_holder = false;
1295 spin_lock_irqsave(&cache->lock, flags);
1296 list_add_tail(&mg->list, &cache->completed_migrations);
1297 spin_unlock_irqrestore(&cache->lock, flags);
1302 static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio)
1304 size_t pb_data_size = get_per_bio_data_size(mg->cache);
1305 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1307 dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg);
1308 remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock);
1311 * No need to inc_ds() here, since the cell will be held for the
1312 * duration of the io.
1314 accounted_request(mg->cache, bio);
1317 static bool bio_writes_complete_block(struct cache *cache, struct bio *bio)
1319 return (bio_data_dir(bio) == WRITE) &&
1320 (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT));
1323 static void avoid_copy(struct dm_cache_migration *mg)
1325 atomic_inc(&mg->cache->stats.copies_avoided);
1326 migration_success_pre_commit(mg);
1329 static void calc_discard_block_range(struct cache *cache, struct bio *bio,
1330 dm_dblock_t *b, dm_dblock_t *e)
1332 sector_t sb = bio->bi_iter.bi_sector;
1333 sector_t se = bio_end_sector(bio);
1335 *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size));
1337 if (se - sb < cache->discard_block_size)
1340 *e = to_dblock(block_div(se, cache->discard_block_size));
1343 static void issue_discard(struct dm_cache_migration *mg)
1346 struct bio *bio = mg->new_ocell->holder;
1347 struct cache *cache = mg->cache;
1349 calc_discard_block_range(cache, bio, &b, &e);
1351 set_discard(cache, b);
1352 b = to_dblock(from_dblock(b) + 1);
1356 cell_defer(cache, mg->new_ocell, false);
1361 static void issue_copy_or_discard(struct dm_cache_migration *mg)
1364 struct cache *cache = mg->cache;
1371 if (mg->writeback || mg->demote)
1372 avoid = !is_dirty(cache, mg->cblock) ||
1373 is_discarded_oblock(cache, mg->old_oblock);
1375 struct bio *bio = mg->new_ocell->holder;
1377 avoid = is_discarded_oblock(cache, mg->new_oblock);
1379 if (writeback_mode(&cache->features) &&
1380 !avoid && bio_writes_complete_block(cache, bio)) {
1381 issue_overwrite(mg, bio);
1386 avoid ? avoid_copy(mg) : issue_copy(mg);
1389 static void complete_migration(struct dm_cache_migration *mg)
1392 migration_failure(mg);
1394 migration_success_pre_commit(mg);
1397 static void process_migrations(struct cache *cache, struct list_head *head,
1398 void (*fn)(struct dm_cache_migration *))
1400 unsigned long flags;
1401 struct list_head list;
1402 struct dm_cache_migration *mg, *tmp;
1404 INIT_LIST_HEAD(&list);
1405 spin_lock_irqsave(&cache->lock, flags);
1406 list_splice_init(head, &list);
1407 spin_unlock_irqrestore(&cache->lock, flags);
1409 list_for_each_entry_safe(mg, tmp, &list, list)
1413 static void __queue_quiesced_migration(struct dm_cache_migration *mg)
1415 list_add_tail(&mg->list, &mg->cache->quiesced_migrations);
1418 static void queue_quiesced_migration(struct dm_cache_migration *mg)
1420 unsigned long flags;
1421 struct cache *cache = mg->cache;
1423 spin_lock_irqsave(&cache->lock, flags);
1424 __queue_quiesced_migration(mg);
1425 spin_unlock_irqrestore(&cache->lock, flags);
1430 static void queue_quiesced_migrations(struct cache *cache, struct list_head *work)
1432 unsigned long flags;
1433 struct dm_cache_migration *mg, *tmp;
1435 spin_lock_irqsave(&cache->lock, flags);
1436 list_for_each_entry_safe(mg, tmp, work, list)
1437 __queue_quiesced_migration(mg);
1438 spin_unlock_irqrestore(&cache->lock, flags);
1443 static void check_for_quiesced_migrations(struct cache *cache,
1444 struct per_bio_data *pb)
1446 struct list_head work;
1448 if (!pb->all_io_entry)
1451 INIT_LIST_HEAD(&work);
1452 dm_deferred_entry_dec(pb->all_io_entry, &work);
1454 if (!list_empty(&work))
1455 queue_quiesced_migrations(cache, &work);
1458 static void quiesce_migration(struct dm_cache_migration *mg)
1460 if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list))
1461 queue_quiesced_migration(mg);
1464 static void promote(struct cache *cache, struct prealloc *structs,
1465 dm_oblock_t oblock, dm_cblock_t cblock,
1466 struct dm_bio_prison_cell *cell)
1468 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1471 mg->discard = false;
1472 mg->writeback = false;
1475 mg->requeue_holder = true;
1476 mg->invalidate = false;
1478 mg->new_oblock = oblock;
1479 mg->cblock = cblock;
1480 mg->old_ocell = NULL;
1481 mg->new_ocell = cell;
1482 mg->start_jiffies = jiffies;
1484 inc_io_migrations(cache);
1485 quiesce_migration(mg);
1488 static void writeback(struct cache *cache, struct prealloc *structs,
1489 dm_oblock_t oblock, dm_cblock_t cblock,
1490 struct dm_bio_prison_cell *cell)
1492 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1495 mg->discard = false;
1496 mg->writeback = true;
1498 mg->promote = false;
1499 mg->requeue_holder = true;
1500 mg->invalidate = false;
1502 mg->old_oblock = oblock;
1503 mg->cblock = cblock;
1504 mg->old_ocell = cell;
1505 mg->new_ocell = NULL;
1506 mg->start_jiffies = jiffies;
1508 inc_io_migrations(cache);
1509 quiesce_migration(mg);
1512 static void demote_then_promote(struct cache *cache, struct prealloc *structs,
1513 dm_oblock_t old_oblock, dm_oblock_t new_oblock,
1515 struct dm_bio_prison_cell *old_ocell,
1516 struct dm_bio_prison_cell *new_ocell)
1518 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1521 mg->discard = false;
1522 mg->writeback = false;
1525 mg->requeue_holder = true;
1526 mg->invalidate = false;
1528 mg->old_oblock = old_oblock;
1529 mg->new_oblock = new_oblock;
1530 mg->cblock = cblock;
1531 mg->old_ocell = old_ocell;
1532 mg->new_ocell = new_ocell;
1533 mg->start_jiffies = jiffies;
1535 inc_io_migrations(cache);
1536 quiesce_migration(mg);
1540 * Invalidate a cache entry. No writeback occurs; any changes in the cache
1541 * block are thrown away.
1543 static void invalidate(struct cache *cache, struct prealloc *structs,
1544 dm_oblock_t oblock, dm_cblock_t cblock,
1545 struct dm_bio_prison_cell *cell)
1547 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1550 mg->discard = false;
1551 mg->writeback = false;
1553 mg->promote = false;
1554 mg->requeue_holder = true;
1555 mg->invalidate = true;
1557 mg->old_oblock = oblock;
1558 mg->cblock = cblock;
1559 mg->old_ocell = cell;
1560 mg->new_ocell = NULL;
1561 mg->start_jiffies = jiffies;
1563 inc_io_migrations(cache);
1564 quiesce_migration(mg);
1567 static void discard(struct cache *cache, struct prealloc *structs,
1568 struct dm_bio_prison_cell *cell)
1570 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1574 mg->writeback = false;
1576 mg->promote = false;
1577 mg->requeue_holder = false;
1578 mg->invalidate = false;
1580 mg->old_ocell = NULL;
1581 mg->new_ocell = cell;
1582 mg->start_jiffies = jiffies;
1584 quiesce_migration(mg);
1587 /*----------------------------------------------------------------
1589 *--------------------------------------------------------------*/
1590 static void defer_bio(struct cache *cache, struct bio *bio)
1592 unsigned long flags;
1594 spin_lock_irqsave(&cache->lock, flags);
1595 bio_list_add(&cache->deferred_bios, bio);
1596 spin_unlock_irqrestore(&cache->lock, flags);
1601 static void process_flush_bio(struct cache *cache, struct bio *bio)
1603 size_t pb_data_size = get_per_bio_data_size(cache);
1604 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1606 BUG_ON(bio->bi_iter.bi_size);
1608 remap_to_origin(cache, bio);
1610 remap_to_cache(cache, bio, 0);
1613 * REQ_FLUSH is not directed at any particular block so we don't
1614 * need to inc_ds(). REQ_FUA's are split into a write + REQ_FLUSH
1620 static void process_discard_bio(struct cache *cache, struct prealloc *structs,
1625 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1627 calc_discard_block_range(cache, bio, &b, &e);
1633 cell_prealloc = prealloc_get_cell(structs);
1634 r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc,
1635 (cell_free_fn) prealloc_put_cell,
1636 structs, &new_ocell);
1640 discard(cache, structs, new_ocell);
1643 static bool spare_migration_bandwidth(struct cache *cache)
1645 sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) *
1646 cache->sectors_per_block;
1647 return current_volume < cache->migration_threshold;
1650 static void inc_hit_counter(struct cache *cache, struct bio *bio)
1652 atomic_inc(bio_data_dir(bio) == READ ?
1653 &cache->stats.read_hit : &cache->stats.write_hit);
1656 static void inc_miss_counter(struct cache *cache, struct bio *bio)
1658 atomic_inc(bio_data_dir(bio) == READ ?
1659 &cache->stats.read_miss : &cache->stats.write_miss);
1662 /*----------------------------------------------------------------*/
1665 struct cache *cache;
1666 struct bio_list bios_for_issue;
1667 struct bio_list unhandled_bios;
1671 static void inc_fn(void *context, struct dm_bio_prison_cell *cell)
1674 struct inc_detail *detail = context;
1675 struct cache *cache = detail->cache;
1677 inc_ds(cache, cell->holder, cell);
1678 if (bio_data_dir(cell->holder) == WRITE)
1679 detail->any_writes = true;
1681 while ((bio = bio_list_pop(&cell->bios))) {
1682 if (discard_or_flush(bio)) {
1683 bio_list_add(&detail->unhandled_bios, bio);
1687 if (bio_data_dir(bio) == WRITE)
1688 detail->any_writes = true;
1690 bio_list_add(&detail->bios_for_issue, bio);
1691 inc_ds(cache, bio, cell);
1695 // FIXME: refactor these two
1696 static void remap_cell_to_origin_clear_discard(struct cache *cache,
1697 struct dm_bio_prison_cell *cell,
1698 dm_oblock_t oblock, bool issue_holder)
1701 unsigned long flags;
1702 struct inc_detail detail;
1704 detail.cache = cache;
1705 bio_list_init(&detail.bios_for_issue);
1706 bio_list_init(&detail.unhandled_bios);
1707 detail.any_writes = false;
1709 spin_lock_irqsave(&cache->lock, flags);
1710 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1711 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1712 spin_unlock_irqrestore(&cache->lock, flags);
1714 remap_to_origin(cache, cell->holder);
1716 issue(cache, cell->holder);
1718 accounted_begin(cache, cell->holder);
1720 if (detail.any_writes)
1721 clear_discard(cache, oblock_to_dblock(cache, oblock));
1723 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1724 remap_to_origin(cache, bio);
1728 free_prison_cell(cache, cell);
1731 static void remap_cell_to_cache_dirty(struct cache *cache, struct dm_bio_prison_cell *cell,
1732 dm_oblock_t oblock, dm_cblock_t cblock, bool issue_holder)
1735 unsigned long flags;
1736 struct inc_detail detail;
1738 detail.cache = cache;
1739 bio_list_init(&detail.bios_for_issue);
1740 bio_list_init(&detail.unhandled_bios);
1741 detail.any_writes = false;
1743 spin_lock_irqsave(&cache->lock, flags);
1744 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1745 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1746 spin_unlock_irqrestore(&cache->lock, flags);
1748 remap_to_cache(cache, cell->holder, cblock);
1750 issue(cache, cell->holder);
1752 accounted_begin(cache, cell->holder);
1754 if (detail.any_writes) {
1755 set_dirty(cache, oblock, cblock);
1756 clear_discard(cache, oblock_to_dblock(cache, oblock));
1759 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1760 remap_to_cache(cache, bio, cblock);
1764 free_prison_cell(cache, cell);
1767 /*----------------------------------------------------------------*/
1769 struct old_oblock_lock {
1770 struct policy_locker locker;
1771 struct cache *cache;
1772 struct prealloc *structs;
1773 struct dm_bio_prison_cell *cell;
1776 static int null_locker(struct policy_locker *locker, dm_oblock_t b)
1778 /* This should never be called */
1783 static int cell_locker(struct policy_locker *locker, dm_oblock_t b)
1785 struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker);
1786 struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs);
1788 return bio_detain(l->cache, b, NULL, cell_prealloc,
1789 (cell_free_fn) prealloc_put_cell,
1790 l->structs, &l->cell);
1793 static void process_cell(struct cache *cache, struct prealloc *structs,
1794 struct dm_bio_prison_cell *new_ocell)
1797 bool release_cell = true;
1798 struct bio *bio = new_ocell->holder;
1799 dm_oblock_t block = get_bio_block(cache, bio);
1800 struct policy_result lookup_result;
1801 bool passthrough = passthrough_mode(&cache->features);
1802 bool fast_promotion, can_migrate;
1803 struct old_oblock_lock ool;
1805 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
1806 can_migrate = !passthrough && (fast_promotion || spare_migration_bandwidth(cache));
1808 ool.locker.fn = cell_locker;
1810 ool.structs = structs;
1812 r = policy_map(cache->policy, block, true, can_migrate, fast_promotion,
1813 bio, &ool.locker, &lookup_result);
1815 if (r == -EWOULDBLOCK)
1816 /* migration has been denied */
1817 lookup_result.op = POLICY_MISS;
1819 switch (lookup_result.op) {
1822 inc_miss_counter(cache, bio);
1825 * Passthrough always maps to the origin,
1826 * invalidating any cache blocks that are written
1830 if (bio_data_dir(bio) == WRITE) {
1831 atomic_inc(&cache->stats.demotion);
1832 invalidate(cache, structs, block, lookup_result.cblock, new_ocell);
1833 release_cell = false;
1836 /* FIXME: factor out issue_origin() */
1837 remap_to_origin_clear_discard(cache, bio, block);
1838 inc_and_issue(cache, bio, new_ocell);
1841 inc_hit_counter(cache, bio);
1843 if (bio_data_dir(bio) == WRITE &&
1844 writethrough_mode(&cache->features) &&
1845 !is_dirty(cache, lookup_result.cblock)) {
1846 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
1847 inc_and_issue(cache, bio, new_ocell);
1850 remap_cell_to_cache_dirty(cache, new_ocell, block, lookup_result.cblock, true);
1851 release_cell = false;
1858 inc_miss_counter(cache, bio);
1859 remap_cell_to_origin_clear_discard(cache, new_ocell, block, true);
1860 release_cell = false;
1864 atomic_inc(&cache->stats.promotion);
1865 promote(cache, structs, block, lookup_result.cblock, new_ocell);
1866 release_cell = false;
1869 case POLICY_REPLACE:
1870 atomic_inc(&cache->stats.demotion);
1871 atomic_inc(&cache->stats.promotion);
1872 demote_then_promote(cache, structs, lookup_result.old_oblock,
1873 block, lookup_result.cblock,
1874 ool.cell, new_ocell);
1875 release_cell = false;
1879 DMERR_LIMIT("%s: %s: erroring bio, unknown policy op: %u",
1880 cache_device_name(cache), __func__,
1881 (unsigned) lookup_result.op);
1886 cell_defer(cache, new_ocell, false);
1889 static void process_bio(struct cache *cache, struct prealloc *structs,
1893 dm_oblock_t block = get_bio_block(cache, bio);
1894 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1897 * Check to see if that block is currently migrating.
1899 cell_prealloc = prealloc_get_cell(structs);
1900 r = bio_detain(cache, block, bio, cell_prealloc,
1901 (cell_free_fn) prealloc_put_cell,
1902 structs, &new_ocell);
1906 process_cell(cache, structs, new_ocell);
1909 static int need_commit_due_to_time(struct cache *cache)
1911 return jiffies < cache->last_commit_jiffies ||
1912 jiffies > cache->last_commit_jiffies + COMMIT_PERIOD;
1916 * A non-zero return indicates read_only or fail_io mode.
1918 static int commit(struct cache *cache, bool clean_shutdown)
1922 if (get_cache_mode(cache) >= CM_READ_ONLY)
1925 atomic_inc(&cache->stats.commit_count);
1926 r = dm_cache_commit(cache->cmd, clean_shutdown);
1928 metadata_operation_failed(cache, "dm_cache_commit", r);
1933 static int commit_if_needed(struct cache *cache)
1937 if ((cache->commit_requested || need_commit_due_to_time(cache)) &&
1938 dm_cache_changed_this_transaction(cache->cmd)) {
1939 r = commit(cache, false);
1940 cache->commit_requested = false;
1941 cache->last_commit_jiffies = jiffies;
1947 static void process_deferred_bios(struct cache *cache)
1949 bool prealloc_used = false;
1950 unsigned long flags;
1951 struct bio_list bios;
1953 struct prealloc structs;
1955 memset(&structs, 0, sizeof(structs));
1956 bio_list_init(&bios);
1958 spin_lock_irqsave(&cache->lock, flags);
1959 bio_list_merge(&bios, &cache->deferred_bios);
1960 bio_list_init(&cache->deferred_bios);
1961 spin_unlock_irqrestore(&cache->lock, flags);
1963 while (!bio_list_empty(&bios)) {
1965 * If we've got no free migration structs, and processing
1966 * this bio might require one, we pause until there are some
1967 * prepared mappings to process.
1969 prealloc_used = true;
1970 if (prealloc_data_structs(cache, &structs)) {
1971 spin_lock_irqsave(&cache->lock, flags);
1972 bio_list_merge(&cache->deferred_bios, &bios);
1973 spin_unlock_irqrestore(&cache->lock, flags);
1977 bio = bio_list_pop(&bios);
1979 if (bio->bi_rw & REQ_FLUSH)
1980 process_flush_bio(cache, bio);
1981 else if (bio->bi_rw & REQ_DISCARD)
1982 process_discard_bio(cache, &structs, bio);
1984 process_bio(cache, &structs, bio);
1988 prealloc_free_structs(cache, &structs);
1991 static void process_deferred_cells(struct cache *cache)
1993 bool prealloc_used = false;
1994 unsigned long flags;
1995 struct dm_bio_prison_cell *cell, *tmp;
1996 struct list_head cells;
1997 struct prealloc structs;
1999 memset(&structs, 0, sizeof(structs));
2001 INIT_LIST_HEAD(&cells);
2003 spin_lock_irqsave(&cache->lock, flags);
2004 list_splice_init(&cache->deferred_cells, &cells);
2005 spin_unlock_irqrestore(&cache->lock, flags);
2007 list_for_each_entry_safe(cell, tmp, &cells, user_list) {
2009 * If we've got no free migration structs, and processing
2010 * this bio might require one, we pause until there are some
2011 * prepared mappings to process.
2013 prealloc_used = true;
2014 if (prealloc_data_structs(cache, &structs)) {
2015 spin_lock_irqsave(&cache->lock, flags);
2016 list_splice(&cells, &cache->deferred_cells);
2017 spin_unlock_irqrestore(&cache->lock, flags);
2021 process_cell(cache, &structs, cell);
2025 prealloc_free_structs(cache, &structs);
2028 static void process_deferred_flush_bios(struct cache *cache, bool submit_bios)
2030 unsigned long flags;
2031 struct bio_list bios;
2034 bio_list_init(&bios);
2036 spin_lock_irqsave(&cache->lock, flags);
2037 bio_list_merge(&bios, &cache->deferred_flush_bios);
2038 bio_list_init(&cache->deferred_flush_bios);
2039 spin_unlock_irqrestore(&cache->lock, flags);
2042 * These bios have already been through inc_ds()
2044 while ((bio = bio_list_pop(&bios)))
2045 submit_bios ? accounted_request(cache, bio) : bio_io_error(bio);
2048 static void process_deferred_writethrough_bios(struct cache *cache)
2050 unsigned long flags;
2051 struct bio_list bios;
2054 bio_list_init(&bios);
2056 spin_lock_irqsave(&cache->lock, flags);
2057 bio_list_merge(&bios, &cache->deferred_writethrough_bios);
2058 bio_list_init(&cache->deferred_writethrough_bios);
2059 spin_unlock_irqrestore(&cache->lock, flags);
2062 * These bios have already been through inc_ds()
2064 while ((bio = bio_list_pop(&bios)))
2065 accounted_request(cache, bio);
2068 static void writeback_some_dirty_blocks(struct cache *cache)
2070 bool prealloc_used = false;
2073 struct prealloc structs;
2074 struct dm_bio_prison_cell *old_ocell;
2075 bool busy = !iot_idle_for(&cache->origin_tracker, HZ);
2077 memset(&structs, 0, sizeof(structs));
2079 while (spare_migration_bandwidth(cache)) {
2080 if (policy_writeback_work(cache->policy, &oblock, &cblock, busy))
2081 break; /* no work to do */
2083 prealloc_used = true;
2084 if (prealloc_data_structs(cache, &structs) ||
2085 get_cell(cache, oblock, &structs, &old_ocell)) {
2086 policy_set_dirty(cache->policy, oblock);
2090 writeback(cache, &structs, oblock, cblock, old_ocell);
2094 prealloc_free_structs(cache, &structs);
2097 /*----------------------------------------------------------------
2099 * Dropping something from the cache *without* writing back.
2100 *--------------------------------------------------------------*/
2102 static void process_invalidation_request(struct cache *cache, struct invalidation_request *req)
2105 uint64_t begin = from_cblock(req->cblocks->begin);
2106 uint64_t end = from_cblock(req->cblocks->end);
2108 while (begin != end) {
2109 r = policy_remove_cblock(cache->policy, to_cblock(begin));
2111 r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin));
2113 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
2117 } else if (r == -ENODATA) {
2118 /* harmless, already unmapped */
2122 DMERR("%s: policy_remove_cblock failed", cache_device_name(cache));
2129 cache->commit_requested = true;
2132 atomic_set(&req->complete, 1);
2134 wake_up(&req->result_wait);
2137 static void process_invalidation_requests(struct cache *cache)
2139 struct list_head list;
2140 struct invalidation_request *req, *tmp;
2142 INIT_LIST_HEAD(&list);
2143 spin_lock(&cache->invalidation_lock);
2144 list_splice_init(&cache->invalidation_requests, &list);
2145 spin_unlock(&cache->invalidation_lock);
2147 list_for_each_entry_safe (req, tmp, &list, list)
2148 process_invalidation_request(cache, req);
2151 /*----------------------------------------------------------------
2153 *--------------------------------------------------------------*/
2154 static bool is_quiescing(struct cache *cache)
2156 return atomic_read(&cache->quiescing);
2159 static void ack_quiescing(struct cache *cache)
2161 if (is_quiescing(cache)) {
2162 atomic_inc(&cache->quiescing_ack);
2163 wake_up(&cache->quiescing_wait);
2167 static void wait_for_quiescing_ack(struct cache *cache)
2169 wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack));
2172 static void start_quiescing(struct cache *cache)
2174 atomic_inc(&cache->quiescing);
2175 wait_for_quiescing_ack(cache);
2178 static void stop_quiescing(struct cache *cache)
2180 atomic_set(&cache->quiescing, 0);
2181 atomic_set(&cache->quiescing_ack, 0);
2184 static void wait_for_migrations(struct cache *cache)
2186 wait_event(cache->migration_wait, !atomic_read(&cache->nr_allocated_migrations));
2189 static void stop_worker(struct cache *cache)
2191 cancel_delayed_work(&cache->waker);
2192 flush_workqueue(cache->wq);
2195 static void requeue_deferred_cells(struct cache *cache)
2197 unsigned long flags;
2198 struct list_head cells;
2199 struct dm_bio_prison_cell *cell, *tmp;
2201 INIT_LIST_HEAD(&cells);
2202 spin_lock_irqsave(&cache->lock, flags);
2203 list_splice_init(&cache->deferred_cells, &cells);
2204 spin_unlock_irqrestore(&cache->lock, flags);
2206 list_for_each_entry_safe(cell, tmp, &cells, user_list)
2207 cell_requeue(cache, cell);
2210 static void requeue_deferred_bios(struct cache *cache)
2213 struct bio_list bios;
2215 bio_list_init(&bios);
2216 bio_list_merge(&bios, &cache->deferred_bios);
2217 bio_list_init(&cache->deferred_bios);
2219 while ((bio = bio_list_pop(&bios))) {
2220 bio->bi_error = DM_ENDIO_REQUEUE;
2225 static int more_work(struct cache *cache)
2227 if (is_quiescing(cache))
2228 return !list_empty(&cache->quiesced_migrations) ||
2229 !list_empty(&cache->completed_migrations) ||
2230 !list_empty(&cache->need_commit_migrations);
2232 return !bio_list_empty(&cache->deferred_bios) ||
2233 !list_empty(&cache->deferred_cells) ||
2234 !bio_list_empty(&cache->deferred_flush_bios) ||
2235 !bio_list_empty(&cache->deferred_writethrough_bios) ||
2236 !list_empty(&cache->quiesced_migrations) ||
2237 !list_empty(&cache->completed_migrations) ||
2238 !list_empty(&cache->need_commit_migrations) ||
2242 static void do_worker(struct work_struct *ws)
2244 struct cache *cache = container_of(ws, struct cache, worker);
2247 if (!is_quiescing(cache)) {
2248 writeback_some_dirty_blocks(cache);
2249 process_deferred_writethrough_bios(cache);
2250 process_deferred_bios(cache);
2251 process_deferred_cells(cache);
2252 process_invalidation_requests(cache);
2255 process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard);
2256 process_migrations(cache, &cache->completed_migrations, complete_migration);
2258 if (commit_if_needed(cache)) {
2259 process_deferred_flush_bios(cache, false);
2260 process_migrations(cache, &cache->need_commit_migrations, migration_failure);
2262 process_deferred_flush_bios(cache, true);
2263 process_migrations(cache, &cache->need_commit_migrations,
2264 migration_success_post_commit);
2267 ack_quiescing(cache);
2269 } while (more_work(cache));
2273 * We want to commit periodically so that not too much
2274 * unwritten metadata builds up.
2276 static void do_waker(struct work_struct *ws)
2278 struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker);
2279 policy_tick(cache->policy, true);
2281 queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD);
2284 /*----------------------------------------------------------------*/
2286 static int is_congested(struct dm_dev *dev, int bdi_bits)
2288 struct request_queue *q = bdev_get_queue(dev->bdev);
2289 return bdi_congested(&q->backing_dev_info, bdi_bits);
2292 static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2294 struct cache *cache = container_of(cb, struct cache, callbacks);
2296 return is_congested(cache->origin_dev, bdi_bits) ||
2297 is_congested(cache->cache_dev, bdi_bits);
2300 /*----------------------------------------------------------------
2302 *--------------------------------------------------------------*/
2305 * This function gets called on the error paths of the constructor, so we
2306 * have to cope with a partially initialised struct.
2308 static void destroy(struct cache *cache)
2312 if (cache->migration_pool)
2313 mempool_destroy(cache->migration_pool);
2315 if (cache->all_io_ds)
2316 dm_deferred_set_destroy(cache->all_io_ds);
2319 dm_bio_prison_destroy(cache->prison);
2322 destroy_workqueue(cache->wq);
2324 if (cache->dirty_bitset)
2325 free_bitset(cache->dirty_bitset);
2327 if (cache->discard_bitset)
2328 free_bitset(cache->discard_bitset);
2331 dm_kcopyd_client_destroy(cache->copier);
2334 dm_cache_metadata_close(cache->cmd);
2336 if (cache->metadata_dev)
2337 dm_put_device(cache->ti, cache->metadata_dev);
2339 if (cache->origin_dev)
2340 dm_put_device(cache->ti, cache->origin_dev);
2342 if (cache->cache_dev)
2343 dm_put_device(cache->ti, cache->cache_dev);
2346 dm_cache_policy_destroy(cache->policy);
2348 for (i = 0; i < cache->nr_ctr_args ; i++)
2349 kfree(cache->ctr_args[i]);
2350 kfree(cache->ctr_args);
2355 static void cache_dtr(struct dm_target *ti)
2357 struct cache *cache = ti->private;
2362 static sector_t get_dev_size(struct dm_dev *dev)
2364 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
2367 /*----------------------------------------------------------------*/
2370 * Construct a cache device mapping.
2372 * cache <metadata dev> <cache dev> <origin dev> <block size>
2373 * <#feature args> [<feature arg>]*
2374 * <policy> <#policy args> [<policy arg>]*
2376 * metadata dev : fast device holding the persistent metadata
2377 * cache dev : fast device holding cached data blocks
2378 * origin dev : slow device holding original data blocks
2379 * block size : cache unit size in sectors
2381 * #feature args : number of feature arguments passed
2382 * feature args : writethrough. (The default is writeback.)
2384 * policy : the replacement policy to use
2385 * #policy args : an even number of policy arguments corresponding
2386 * to key/value pairs passed to the policy
2387 * policy args : key/value pairs passed to the policy
2388 * E.g. 'sequential_threshold 1024'
2389 * See cache-policies.txt for details.
2391 * Optional feature arguments are:
2392 * writethrough : write through caching that prohibits cache block
2393 * content from being different from origin block content.
2394 * Without this argument, the default behaviour is to write
2395 * back cache block contents later for performance reasons,
2396 * so they may differ from the corresponding origin blocks.
2399 struct dm_target *ti;
2401 struct dm_dev *metadata_dev;
2403 struct dm_dev *cache_dev;
2404 sector_t cache_sectors;
2406 struct dm_dev *origin_dev;
2407 sector_t origin_sectors;
2409 uint32_t block_size;
2411 const char *policy_name;
2413 const char **policy_argv;
2415 struct cache_features features;
2418 static void destroy_cache_args(struct cache_args *ca)
2420 if (ca->metadata_dev)
2421 dm_put_device(ca->ti, ca->metadata_dev);
2424 dm_put_device(ca->ti, ca->cache_dev);
2427 dm_put_device(ca->ti, ca->origin_dev);
2432 static bool at_least_one_arg(struct dm_arg_set *as, char **error)
2435 *error = "Insufficient args";
2442 static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as,
2446 sector_t metadata_dev_size;
2447 char b[BDEVNAME_SIZE];
2449 if (!at_least_one_arg(as, error))
2452 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2455 *error = "Error opening metadata device";
2459 metadata_dev_size = get_dev_size(ca->metadata_dev);
2460 if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING)
2461 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2462 bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
2467 static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as,
2472 if (!at_least_one_arg(as, error))
2475 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2478 *error = "Error opening cache device";
2481 ca->cache_sectors = get_dev_size(ca->cache_dev);
2486 static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as,
2491 if (!at_least_one_arg(as, error))
2494 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2497 *error = "Error opening origin device";
2501 ca->origin_sectors = get_dev_size(ca->origin_dev);
2502 if (ca->ti->len > ca->origin_sectors) {
2503 *error = "Device size larger than cached device";
2510 static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as,
2513 unsigned long block_size;
2515 if (!at_least_one_arg(as, error))
2518 if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size ||
2519 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2520 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2521 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2522 *error = "Invalid data block size";
2526 if (block_size > ca->cache_sectors) {
2527 *error = "Data block size is larger than the cache device";
2531 ca->block_size = block_size;
2536 static void init_features(struct cache_features *cf)
2538 cf->mode = CM_WRITE;
2539 cf->io_mode = CM_IO_WRITEBACK;
2542 static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
2545 static struct dm_arg _args[] = {
2546 {0, 1, "Invalid number of cache feature arguments"},
2552 struct cache_features *cf = &ca->features;
2556 r = dm_read_arg_group(_args, as, &argc, error);
2561 arg = dm_shift_arg(as);
2563 if (!strcasecmp(arg, "writeback"))
2564 cf->io_mode = CM_IO_WRITEBACK;
2566 else if (!strcasecmp(arg, "writethrough"))
2567 cf->io_mode = CM_IO_WRITETHROUGH;
2569 else if (!strcasecmp(arg, "passthrough"))
2570 cf->io_mode = CM_IO_PASSTHROUGH;
2573 *error = "Unrecognised cache feature requested";
2581 static int parse_policy(struct cache_args *ca, struct dm_arg_set *as,
2584 static struct dm_arg _args[] = {
2585 {0, 1024, "Invalid number of policy arguments"},
2590 if (!at_least_one_arg(as, error))
2593 ca->policy_name = dm_shift_arg(as);
2595 r = dm_read_arg_group(_args, as, &ca->policy_argc, error);
2599 ca->policy_argv = (const char **)as->argv;
2600 dm_consume_args(as, ca->policy_argc);
2605 static int parse_cache_args(struct cache_args *ca, int argc, char **argv,
2609 struct dm_arg_set as;
2614 r = parse_metadata_dev(ca, &as, error);
2618 r = parse_cache_dev(ca, &as, error);
2622 r = parse_origin_dev(ca, &as, error);
2626 r = parse_block_size(ca, &as, error);
2630 r = parse_features(ca, &as, error);
2634 r = parse_policy(ca, &as, error);
2641 /*----------------------------------------------------------------*/
2643 static struct kmem_cache *migration_cache;
2645 #define NOT_CORE_OPTION 1
2647 static int process_config_option(struct cache *cache, const char *key, const char *value)
2651 if (!strcasecmp(key, "migration_threshold")) {
2652 if (kstrtoul(value, 10, &tmp))
2655 cache->migration_threshold = tmp;
2659 return NOT_CORE_OPTION;
2662 static int set_config_value(struct cache *cache, const char *key, const char *value)
2664 int r = process_config_option(cache, key, value);
2666 if (r == NOT_CORE_OPTION)
2667 r = policy_set_config_value(cache->policy, key, value);
2670 DMWARN("bad config value for %s: %s", key, value);
2675 static int set_config_values(struct cache *cache, int argc, const char **argv)
2680 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
2685 r = set_config_value(cache, argv[0], argv[1]);
2696 static int create_cache_policy(struct cache *cache, struct cache_args *ca,
2699 struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name,
2701 cache->origin_sectors,
2702 cache->sectors_per_block);
2704 *error = "Error creating cache's policy";
2713 * We want the discard block size to be at least the size of the cache
2714 * block size and have no more than 2^14 discard blocks across the origin.
2716 #define MAX_DISCARD_BLOCKS (1 << 14)
2718 static bool too_many_discard_blocks(sector_t discard_block_size,
2719 sector_t origin_size)
2721 (void) sector_div(origin_size, discard_block_size);
2723 return origin_size > MAX_DISCARD_BLOCKS;
2726 static sector_t calculate_discard_block_size(sector_t cache_block_size,
2727 sector_t origin_size)
2729 sector_t discard_block_size = cache_block_size;
2732 while (too_many_discard_blocks(discard_block_size, origin_size))
2733 discard_block_size *= 2;
2735 return discard_block_size;
2738 static void set_cache_size(struct cache *cache, dm_cblock_t size)
2740 dm_block_t nr_blocks = from_cblock(size);
2742 if (nr_blocks > (1 << 20) && cache->cache_size != size)
2743 DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n"
2744 "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n"
2745 "Please consider increasing the cache block size to reduce the overall cache block count.",
2746 (unsigned long long) nr_blocks);
2748 cache->cache_size = size;
2751 #define DEFAULT_MIGRATION_THRESHOLD 2048
2753 static int cache_create(struct cache_args *ca, struct cache **result)
2756 char **error = &ca->ti->error;
2757 struct cache *cache;
2758 struct dm_target *ti = ca->ti;
2759 dm_block_t origin_blocks;
2760 struct dm_cache_metadata *cmd;
2761 bool may_format = ca->features.mode == CM_WRITE;
2763 cache = kzalloc(sizeof(*cache), GFP_KERNEL);
2768 ti->private = cache;
2769 ti->num_flush_bios = 2;
2770 ti->flush_supported = true;
2772 ti->num_discard_bios = 1;
2773 ti->discards_supported = true;
2774 ti->discard_zeroes_data_unsupported = true;
2775 ti->split_discard_bios = false;
2777 cache->features = ca->features;
2778 ti->per_bio_data_size = get_per_bio_data_size(cache);
2780 cache->callbacks.congested_fn = cache_is_congested;
2781 dm_table_add_target_callbacks(ti->table, &cache->callbacks);
2783 cache->metadata_dev = ca->metadata_dev;
2784 cache->origin_dev = ca->origin_dev;
2785 cache->cache_dev = ca->cache_dev;
2787 ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL;
2789 /* FIXME: factor out this whole section */
2790 origin_blocks = cache->origin_sectors = ca->origin_sectors;
2791 origin_blocks = block_div(origin_blocks, ca->block_size);
2792 cache->origin_blocks = to_oblock(origin_blocks);
2794 cache->sectors_per_block = ca->block_size;
2795 if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) {
2800 if (ca->block_size & (ca->block_size - 1)) {
2801 dm_block_t cache_size = ca->cache_sectors;
2803 cache->sectors_per_block_shift = -1;
2804 cache_size = block_div(cache_size, ca->block_size);
2805 set_cache_size(cache, to_cblock(cache_size));
2807 cache->sectors_per_block_shift = __ffs(ca->block_size);
2808 set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift));
2811 r = create_cache_policy(cache, ca, error);
2815 cache->policy_nr_args = ca->policy_argc;
2816 cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD;
2818 r = set_config_values(cache, ca->policy_argc, ca->policy_argv);
2820 *error = "Error setting cache policy's config values";
2824 cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
2825 ca->block_size, may_format,
2826 dm_cache_policy_get_hint_size(cache->policy));
2828 *error = "Error creating metadata object";
2833 set_cache_mode(cache, CM_WRITE);
2834 if (get_cache_mode(cache) != CM_WRITE) {
2835 *error = "Unable to get write access to metadata, please check/repair metadata.";
2840 if (passthrough_mode(&cache->features)) {
2843 r = dm_cache_metadata_all_clean(cache->cmd, &all_clean);
2845 *error = "dm_cache_metadata_all_clean() failed";
2850 *error = "Cannot enter passthrough mode unless all blocks are clean";
2856 spin_lock_init(&cache->lock);
2857 INIT_LIST_HEAD(&cache->deferred_cells);
2858 bio_list_init(&cache->deferred_bios);
2859 bio_list_init(&cache->deferred_flush_bios);
2860 bio_list_init(&cache->deferred_writethrough_bios);
2861 INIT_LIST_HEAD(&cache->quiesced_migrations);
2862 INIT_LIST_HEAD(&cache->completed_migrations);
2863 INIT_LIST_HEAD(&cache->need_commit_migrations);
2864 atomic_set(&cache->nr_allocated_migrations, 0);
2865 atomic_set(&cache->nr_io_migrations, 0);
2866 init_waitqueue_head(&cache->migration_wait);
2868 init_waitqueue_head(&cache->quiescing_wait);
2869 atomic_set(&cache->quiescing, 0);
2870 atomic_set(&cache->quiescing_ack, 0);
2873 atomic_set(&cache->nr_dirty, 0);
2874 cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size));
2875 if (!cache->dirty_bitset) {
2876 *error = "could not allocate dirty bitset";
2879 clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size));
2881 cache->discard_block_size =
2882 calculate_discard_block_size(cache->sectors_per_block,
2883 cache->origin_sectors);
2884 cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors,
2885 cache->discard_block_size));
2886 cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks));
2887 if (!cache->discard_bitset) {
2888 *error = "could not allocate discard bitset";
2891 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
2893 cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2894 if (IS_ERR(cache->copier)) {
2895 *error = "could not create kcopyd client";
2896 r = PTR_ERR(cache->copier);
2900 cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2902 *error = "could not create workqueue for metadata object";
2905 INIT_WORK(&cache->worker, do_worker);
2906 INIT_DELAYED_WORK(&cache->waker, do_waker);
2907 cache->last_commit_jiffies = jiffies;
2909 cache->prison = dm_bio_prison_create();
2910 if (!cache->prison) {
2911 *error = "could not create bio prison";
2915 cache->all_io_ds = dm_deferred_set_create();
2916 if (!cache->all_io_ds) {
2917 *error = "could not create all_io deferred set";
2921 cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE,
2923 if (!cache->migration_pool) {
2924 *error = "Error creating cache's migration mempool";
2928 cache->need_tick_bio = true;
2929 cache->sized = false;
2930 cache->invalidate = false;
2931 cache->commit_requested = false;
2932 cache->loaded_mappings = false;
2933 cache->loaded_discards = false;
2937 atomic_set(&cache->stats.demotion, 0);
2938 atomic_set(&cache->stats.promotion, 0);
2939 atomic_set(&cache->stats.copies_avoided, 0);
2940 atomic_set(&cache->stats.cache_cell_clash, 0);
2941 atomic_set(&cache->stats.commit_count, 0);
2942 atomic_set(&cache->stats.discard_count, 0);
2944 spin_lock_init(&cache->invalidation_lock);
2945 INIT_LIST_HEAD(&cache->invalidation_requests);
2947 iot_init(&cache->origin_tracker);
2957 static int copy_ctr_args(struct cache *cache, int argc, const char **argv)
2962 copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL);
2965 for (i = 0; i < argc; i++) {
2966 copy[i] = kstrdup(argv[i], GFP_KERNEL);
2975 cache->nr_ctr_args = argc;
2976 cache->ctr_args = copy;
2981 static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv)
2984 struct cache_args *ca;
2985 struct cache *cache = NULL;
2987 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2989 ti->error = "Error allocating memory for cache";
2994 r = parse_cache_args(ca, argc, argv, &ti->error);
2998 r = cache_create(ca, &cache);
3002 r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3);
3008 ti->private = cache;
3011 destroy_cache_args(ca);
3015 /*----------------------------------------------------------------*/
3017 static int cache_map(struct dm_target *ti, struct bio *bio)
3019 struct cache *cache = ti->private;
3022 struct dm_bio_prison_cell *cell = NULL;
3023 dm_oblock_t block = get_bio_block(cache, bio);
3024 size_t pb_data_size = get_per_bio_data_size(cache);
3025 bool can_migrate = false;
3026 bool fast_promotion;
3027 struct policy_result lookup_result;
3028 struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size);
3029 struct old_oblock_lock ool;
3031 ool.locker.fn = null_locker;
3033 if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) {
3035 * This can only occur if the io goes to a partial block at
3036 * the end of the origin device. We don't cache these.
3037 * Just remap to the origin and carry on.
3039 remap_to_origin(cache, bio);
3040 accounted_begin(cache, bio);
3041 return DM_MAPIO_REMAPPED;
3044 if (discard_or_flush(bio)) {
3045 defer_bio(cache, bio);
3046 return DM_MAPIO_SUBMITTED;
3050 * Check to see if that block is currently migrating.
3052 cell = alloc_prison_cell(cache);
3054 defer_bio(cache, bio);
3055 return DM_MAPIO_SUBMITTED;
3058 r = bio_detain(cache, block, bio, cell,
3059 (cell_free_fn) free_prison_cell,
3063 defer_bio(cache, bio);
3065 return DM_MAPIO_SUBMITTED;
3068 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
3070 r = policy_map(cache->policy, block, false, can_migrate, fast_promotion,
3071 bio, &ool.locker, &lookup_result);
3072 if (r == -EWOULDBLOCK) {
3073 cell_defer(cache, cell, true);
3074 return DM_MAPIO_SUBMITTED;
3077 DMERR_LIMIT("%s: Unexpected return from cache replacement policy: %d",
3078 cache_device_name(cache), r);
3079 cell_defer(cache, cell, false);
3081 return DM_MAPIO_SUBMITTED;
3084 r = DM_MAPIO_REMAPPED;
3085 switch (lookup_result.op) {
3087 if (passthrough_mode(&cache->features)) {
3088 if (bio_data_dir(bio) == WRITE) {
3090 * We need to invalidate this block, so
3091 * defer for the worker thread.
3093 cell_defer(cache, cell, true);
3094 r = DM_MAPIO_SUBMITTED;
3097 inc_miss_counter(cache, bio);
3098 remap_to_origin_clear_discard(cache, bio, block);
3099 accounted_begin(cache, bio);
3100 inc_ds(cache, bio, cell);
3101 // FIXME: we want to remap hits or misses straight
3102 // away rather than passing over to the worker.
3103 cell_defer(cache, cell, false);
3107 inc_hit_counter(cache, bio);
3108 if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) &&
3109 !is_dirty(cache, lookup_result.cblock)) {
3110 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
3111 accounted_begin(cache, bio);
3112 inc_ds(cache, bio, cell);
3113 cell_defer(cache, cell, false);
3116 remap_cell_to_cache_dirty(cache, cell, block, lookup_result.cblock, false);
3121 inc_miss_counter(cache, bio);
3122 if (pb->req_nr != 0) {
3124 * This is a duplicate writethrough io that is no
3125 * longer needed because the block has been demoted.
3128 // FIXME: remap everything as a miss
3129 cell_defer(cache, cell, false);
3130 r = DM_MAPIO_SUBMITTED;
3133 remap_cell_to_origin_clear_discard(cache, cell, block, false);
3137 DMERR_LIMIT("%s: %s: erroring bio: unknown policy op: %u",
3138 cache_device_name(cache), __func__,
3139 (unsigned) lookup_result.op);
3140 cell_defer(cache, cell, false);
3142 r = DM_MAPIO_SUBMITTED;
3148 static int cache_end_io(struct dm_target *ti, struct bio *bio, int error)
3150 struct cache *cache = ti->private;
3151 unsigned long flags;
3152 size_t pb_data_size = get_per_bio_data_size(cache);
3153 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
3156 policy_tick(cache->policy, false);
3158 spin_lock_irqsave(&cache->lock, flags);
3159 cache->need_tick_bio = true;
3160 spin_unlock_irqrestore(&cache->lock, flags);
3163 check_for_quiesced_migrations(cache, pb);
3164 accounted_complete(cache, bio);
3169 static int write_dirty_bitset(struct cache *cache)
3173 if (get_cache_mode(cache) >= CM_READ_ONLY)
3176 for (i = 0; i < from_cblock(cache->cache_size); i++) {
3177 r = dm_cache_set_dirty(cache->cmd, to_cblock(i),
3178 is_dirty(cache, to_cblock(i)));
3180 metadata_operation_failed(cache, "dm_cache_set_dirty", r);
3188 static int write_discard_bitset(struct cache *cache)
3192 if (get_cache_mode(cache) >= CM_READ_ONLY)
3195 r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size,
3196 cache->discard_nr_blocks);
3198 DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache));
3199 metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r);
3203 for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) {
3204 r = dm_cache_set_discard(cache->cmd, to_dblock(i),
3205 is_discarded(cache, to_dblock(i)));
3207 metadata_operation_failed(cache, "dm_cache_set_discard", r);
3215 static int write_hints(struct cache *cache)
3219 if (get_cache_mode(cache) >= CM_READ_ONLY)
3222 r = dm_cache_write_hints(cache->cmd, cache->policy);
3224 metadata_operation_failed(cache, "dm_cache_write_hints", r);
3232 * returns true on success
3234 static bool sync_metadata(struct cache *cache)
3238 r1 = write_dirty_bitset(cache);
3240 DMERR("%s: could not write dirty bitset", cache_device_name(cache));
3242 r2 = write_discard_bitset(cache);
3244 DMERR("%s: could not write discard bitset", cache_device_name(cache));
3248 r3 = write_hints(cache);
3250 DMERR("%s: could not write hints", cache_device_name(cache));
3253 * If writing the above metadata failed, we still commit, but don't
3254 * set the clean shutdown flag. This will effectively force every
3255 * dirty bit to be set on reload.
3257 r4 = commit(cache, !r1 && !r2 && !r3);
3259 DMERR("%s: could not write cache metadata", cache_device_name(cache));
3261 return !r1 && !r2 && !r3 && !r4;
3264 static void cache_postsuspend(struct dm_target *ti)
3266 struct cache *cache = ti->private;
3268 start_quiescing(cache);
3269 wait_for_migrations(cache);
3271 requeue_deferred_bios(cache);
3272 requeue_deferred_cells(cache);
3273 stop_quiescing(cache);
3275 if (get_cache_mode(cache) == CM_WRITE)
3276 (void) sync_metadata(cache);
3279 static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock,
3280 bool dirty, uint32_t hint, bool hint_valid)
3283 struct cache *cache = context;
3285 r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid);
3290 set_dirty(cache, oblock, cblock);
3292 clear_dirty(cache, oblock, cblock);
3298 * The discard block size in the on disk metadata is not
3299 * neccessarily the same as we're currently using. So we have to
3300 * be careful to only set the discarded attribute if we know it
3301 * covers a complete block of the new size.
3303 struct discard_load_info {
3304 struct cache *cache;
3307 * These blocks are sized using the on disk dblock size, rather
3308 * than the current one.
3310 dm_block_t block_size;
3311 dm_block_t discard_begin, discard_end;
3314 static void discard_load_info_init(struct cache *cache,
3315 struct discard_load_info *li)
3318 li->discard_begin = li->discard_end = 0;
3321 static void set_discard_range(struct discard_load_info *li)
3325 if (li->discard_begin == li->discard_end)
3329 * Convert to sectors.
3331 b = li->discard_begin * li->block_size;
3332 e = li->discard_end * li->block_size;
3335 * Then convert back to the current dblock size.
3337 b = dm_sector_div_up(b, li->cache->discard_block_size);
3338 sector_div(e, li->cache->discard_block_size);
3341 * The origin may have shrunk, so we need to check we're still in
3344 if (e > from_dblock(li->cache->discard_nr_blocks))
3345 e = from_dblock(li->cache->discard_nr_blocks);
3348 set_discard(li->cache, to_dblock(b));
3351 static int load_discard(void *context, sector_t discard_block_size,
3352 dm_dblock_t dblock, bool discard)
3354 struct discard_load_info *li = context;
3356 li->block_size = discard_block_size;
3359 if (from_dblock(dblock) == li->discard_end)
3361 * We're already in a discard range, just extend it.
3363 li->discard_end = li->discard_end + 1ULL;
3367 * Emit the old range and start a new one.
3369 set_discard_range(li);
3370 li->discard_begin = from_dblock(dblock);
3371 li->discard_end = li->discard_begin + 1ULL;
3374 set_discard_range(li);
3375 li->discard_begin = li->discard_end = 0;
3381 static dm_cblock_t get_cache_dev_size(struct cache *cache)
3383 sector_t size = get_dev_size(cache->cache_dev);
3384 (void) sector_div(size, cache->sectors_per_block);
3385 return to_cblock(size);
3388 static bool can_resize(struct cache *cache, dm_cblock_t new_size)
3390 if (from_cblock(new_size) > from_cblock(cache->cache_size))
3394 * We can't drop a dirty block when shrinking the cache.
3396 while (from_cblock(new_size) < from_cblock(cache->cache_size)) {
3397 new_size = to_cblock(from_cblock(new_size) + 1);
3398 if (is_dirty(cache, new_size)) {
3399 DMERR("%s: unable to shrink cache; cache block %llu is dirty",
3400 cache_device_name(cache),
3401 (unsigned long long) from_cblock(new_size));
3409 static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size)
3413 r = dm_cache_resize(cache->cmd, new_size);
3415 DMERR("%s: could not resize cache metadata", cache_device_name(cache));
3416 metadata_operation_failed(cache, "dm_cache_resize", r);
3420 set_cache_size(cache, new_size);
3425 static int cache_preresume(struct dm_target *ti)
3428 struct cache *cache = ti->private;
3429 dm_cblock_t csize = get_cache_dev_size(cache);
3432 * Check to see if the cache has resized.
3434 if (!cache->sized) {
3435 r = resize_cache_dev(cache, csize);
3439 cache->sized = true;
3441 } else if (csize != cache->cache_size) {
3442 if (!can_resize(cache, csize))
3445 r = resize_cache_dev(cache, csize);
3450 if (!cache->loaded_mappings) {
3451 r = dm_cache_load_mappings(cache->cmd, cache->policy,
3452 load_mapping, cache);
3454 DMERR("%s: could not load cache mappings", cache_device_name(cache));
3455 metadata_operation_failed(cache, "dm_cache_load_mappings", r);
3459 cache->loaded_mappings = true;
3462 if (!cache->loaded_discards) {
3463 struct discard_load_info li;
3466 * The discard bitset could have been resized, or the
3467 * discard block size changed. To be safe we start by
3468 * setting every dblock to not discarded.
3470 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
3472 discard_load_info_init(cache, &li);
3473 r = dm_cache_load_discards(cache->cmd, load_discard, &li);
3475 DMERR("%s: could not load origin discards", cache_device_name(cache));
3476 metadata_operation_failed(cache, "dm_cache_load_discards", r);
3479 set_discard_range(&li);
3481 cache->loaded_discards = true;
3487 static void cache_resume(struct dm_target *ti)
3489 struct cache *cache = ti->private;
3491 cache->need_tick_bio = true;
3492 do_waker(&cache->waker.work);
3498 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
3499 * <cache block size> <#used cache blocks>/<#total cache blocks>
3500 * <#read hits> <#read misses> <#write hits> <#write misses>
3501 * <#demotions> <#promotions> <#dirty>
3502 * <#features> <features>*
3503 * <#core args> <core args>
3504 * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check>
3506 static void cache_status(struct dm_target *ti, status_type_t type,
3507 unsigned status_flags, char *result, unsigned maxlen)
3512 dm_block_t nr_free_blocks_metadata = 0;
3513 dm_block_t nr_blocks_metadata = 0;
3514 char buf[BDEVNAME_SIZE];
3515 struct cache *cache = ti->private;
3516 dm_cblock_t residency;
3519 case STATUSTYPE_INFO:
3520 if (get_cache_mode(cache) == CM_FAIL) {
3525 /* Commit to ensure statistics aren't out-of-date */
3526 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3527 (void) commit(cache, false);
3529 r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata);
3531 DMERR("%s: dm_cache_get_free_metadata_block_count returned %d",
3532 cache_device_name(cache), r);
3536 r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata);
3538 DMERR("%s: dm_cache_get_metadata_dev_size returned %d",
3539 cache_device_name(cache), r);
3543 residency = policy_residency(cache->policy);
3545 DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ",
3546 (unsigned)DM_CACHE_METADATA_BLOCK_SIZE,
3547 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3548 (unsigned long long)nr_blocks_metadata,
3549 cache->sectors_per_block,
3550 (unsigned long long) from_cblock(residency),
3551 (unsigned long long) from_cblock(cache->cache_size),
3552 (unsigned) atomic_read(&cache->stats.read_hit),
3553 (unsigned) atomic_read(&cache->stats.read_miss),
3554 (unsigned) atomic_read(&cache->stats.write_hit),
3555 (unsigned) atomic_read(&cache->stats.write_miss),
3556 (unsigned) atomic_read(&cache->stats.demotion),
3557 (unsigned) atomic_read(&cache->stats.promotion),
3558 (unsigned long) atomic_read(&cache->nr_dirty));
3560 if (writethrough_mode(&cache->features))
3561 DMEMIT("1 writethrough ");
3563 else if (passthrough_mode(&cache->features))
3564 DMEMIT("1 passthrough ");
3566 else if (writeback_mode(&cache->features))
3567 DMEMIT("1 writeback ");
3570 DMERR("%s: internal error: unknown io mode: %d",
3571 cache_device_name(cache), (int) cache->features.io_mode);
3575 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
3577 DMEMIT("%s ", dm_cache_policy_get_name(cache->policy));
3579 r = policy_emit_config_values(cache->policy, result, maxlen, &sz);
3581 DMERR("%s: policy_emit_config_values returned %d",
3582 cache_device_name(cache), r);
3585 if (get_cache_mode(cache) == CM_READ_ONLY)
3590 if (dm_cache_metadata_needs_check(cache->cmd))
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, 8, 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");