Lines Matching refs:cache
11 #include "dm-cache-metadata.h"
24 #define DM_MSG_PREFIX "cache"
27 "A percentage of time allocated for copying to and/or from cache");
35 * cblock: index of a cache block
36 * promotion: movement of a block from origin to cache
37 * demotion: movement of a block from cache to origin
38 * migration: movement of a block between the origin and cache device,
241 * The block size of the device holding cache data must be
256 * dirty. If you lose the cache device you will lose data.
262 * Data is written to both cache and origin. Blocks are never
268 * A degraded mode useful for various cache coherency situations
270 * origin. If a write goes to a cached oblock, then the cache
297 struct cache {
331 * Size of the cache device in blocks.
421 struct cache *cache;
433 static bool writethrough_mode(struct cache *cache)
435 return cache->features.io_mode == CM_IO_WRITETHROUGH;
438 static bool writeback_mode(struct cache *cache)
440 return cache->features.io_mode == CM_IO_WRITEBACK;
443 static inline bool passthrough_mode(struct cache *cache)
445 return unlikely(cache->features.io_mode == CM_IO_PASSTHROUGH);
450 static void wake_deferred_bio_worker(struct cache *cache)
452 queue_work(cache->wq, &cache->deferred_bio_worker);
455 static void wake_migration_worker(struct cache *cache)
457 if (passthrough_mode(cache))
460 queue_work(cache->wq, &cache->migration_worker);
465 static struct dm_bio_prison_cell_v2 *alloc_prison_cell(struct cache *cache)
467 return dm_bio_prison_alloc_cell_v2(cache->prison, GFP_NOIO);
470 static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell_v2 *cell)
472 dm_bio_prison_free_cell_v2(cache->prison, cell);
475 static struct dm_cache_migration *alloc_migration(struct cache *cache)
479 mg = mempool_alloc(&cache->migration_pool, GFP_NOIO);
483 mg->cache = cache;
484 atomic_inc(&cache->nr_allocated_migrations);
491 struct cache *cache = mg->cache;
493 if (atomic_dec_and_test(&cache->nr_allocated_migrations))
494 wake_up(&cache->migration_wait);
496 mempool_free(mg, &cache->migration_pool);
556 static void defer_bio(struct cache *cache, struct bio *bio)
558 spin_lock_irq(&cache->lock);
559 bio_list_add(&cache->deferred_bios, bio);
560 spin_unlock_irq(&cache->lock);
562 wake_deferred_bio_worker(cache);
565 static void defer_bios(struct cache *cache, struct bio_list *bios)
567 spin_lock_irq(&cache->lock);
568 bio_list_merge(&cache->deferred_bios, bios);
570 spin_unlock_irq(&cache->lock);
572 wake_deferred_bio_worker(cache);
577 static bool bio_detain_shared(struct cache *cache, dm_oblock_t oblock, struct bio *bio)
585 cell_prealloc = alloc_prison_cell(cache); /* FIXME: allow wait if calling from worker */
588 r = dm_cell_get_v2(cache->prison, &key, lock_level(bio), bio, cell_prealloc, &cell);
593 free_prison_cell(cache, cell_prealloc);
598 free_prison_cell(cache, cell_prealloc);
608 static bool is_dirty(struct cache *cache, dm_cblock_t b)
610 return test_bit(from_cblock(b), cache->dirty_bitset);
613 static void set_dirty(struct cache *cache, dm_cblock_t cblock)
615 if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) {
616 atomic_inc(&cache->nr_dirty);
617 policy_set_dirty(cache->policy, cblock);
625 static void force_set_dirty(struct cache *cache, dm_cblock_t cblock)
627 if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset))
628 atomic_inc(&cache->nr_dirty);
629 policy_set_dirty(cache->policy, cblock);
632 static void force_clear_dirty(struct cache *cache, dm_cblock_t cblock)
634 if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) {
635 if (atomic_dec_return(&cache->nr_dirty) == 0)
636 dm_table_event(cache->ti->table);
639 policy_clear_dirty(cache->policy, cblock);
644 static bool block_size_is_power_of_two(struct cache *cache)
646 return cache->sectors_per_block_shift >= 0;
656 static dm_block_t oblocks_per_dblock(struct cache *cache)
658 dm_block_t oblocks = cache->discard_block_size;
660 if (block_size_is_power_of_two(cache))
661 oblocks >>= cache->sectors_per_block_shift;
663 oblocks = block_div(oblocks, cache->sectors_per_block);
668 static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock)
671 oblocks_per_dblock(cache)));
674 static void set_discard(struct cache *cache, dm_dblock_t b)
676 BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks));
677 atomic_inc(&cache->stats.discard_count);
679 spin_lock_irq(&cache->lock);
680 set_bit(from_dblock(b), cache->discard_bitset);
681 spin_unlock_irq(&cache->lock);
684 static void clear_discard(struct cache *cache, dm_dblock_t b)
686 spin_lock_irq(&cache->lock);
687 clear_bit(from_dblock(b), cache->discard_bitset);
688 spin_unlock_irq(&cache->lock);
691 static bool is_discarded(struct cache *cache, dm_dblock_t b)
695 spin_lock_irq(&cache->lock);
696 r = test_bit(from_dblock(b), cache->discard_bitset);
697 spin_unlock_irq(&cache->lock);
702 static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b)
706 spin_lock_irq(&cache->lock);
707 r = test_bit(from_dblock(oblock_to_dblock(cache, b)),
708 cache->discard_bitset);
709 spin_unlock_irq(&cache->lock);
719 static void remap_to_origin(struct cache *cache, struct bio *bio)
721 bio_set_dev(bio, cache->origin_dev->bdev);
724 static void remap_to_cache(struct cache *cache, struct bio *bio,
730 bio_set_dev(bio, cache->cache_dev->bdev);
731 if (!block_size_is_power_of_two(cache))
733 (block * cache->sectors_per_block) +
734 sector_div(bi_sector, cache->sectors_per_block);
737 (block << cache->sectors_per_block_shift) |
738 (bi_sector & (cache->sectors_per_block - 1));
741 static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio)
745 spin_lock_irq(&cache->lock);
746 if (cache->need_tick_bio && !op_is_flush(bio->bi_opf) &&
750 cache->need_tick_bio = false;
752 spin_unlock_irq(&cache->lock);
755 static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio,
759 check_if_tick_bio_needed(cache, bio);
760 remap_to_origin(cache, bio);
762 clear_discard(cache, oblock_to_dblock(cache, oblock));
765 static void remap_to_cache_dirty(struct cache *cache, struct bio *bio,
768 check_if_tick_bio_needed(cache, bio);
769 remap_to_cache(cache, bio, cblock);
771 set_dirty(cache, cblock);
772 clear_discard(cache, oblock_to_dblock(cache, oblock));
776 static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio)
780 if (!block_size_is_power_of_two(cache))
781 (void) sector_div(block_nr, cache->sectors_per_block);
783 block_nr >>= cache->sectors_per_block_shift;
788 static bool accountable_bio(struct cache *cache, struct bio *bio)
793 static void accounted_begin(struct cache *cache, struct bio *bio)
797 if (accountable_bio(cache, bio)) {
800 dm_iot_io_begin(&cache->tracker, pb->len);
804 static void accounted_complete(struct cache *cache, struct bio *bio)
808 dm_iot_io_end(&cache->tracker, pb->len);
811 static void accounted_request(struct cache *cache, struct bio *bio)
813 accounted_begin(cache, bio);
819 struct cache *cache = context;
821 accounted_request(cache, bio);
826 * to both the cache and origin devices. Clone the bio and send them in parallel.
828 static void remap_to_origin_and_cache(struct cache *cache, struct bio *bio,
831 struct bio *origin_bio = bio_alloc_clone(cache->origin_dev->bdev, bio,
832 GFP_NOIO, &cache->bs);
839 clear_discard(cache, oblock_to_dblock(cache, oblock));
842 remap_to_cache(cache, bio, cblock);
850 static enum cache_metadata_mode get_cache_mode(struct cache *cache)
852 return cache->features.mode;
855 static const char *cache_device_name(struct cache *cache)
857 return dm_table_device_name(cache->ti->table);
860 static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode)
868 dm_table_event(cache->ti->table);
869 DMINFO("%s: switching cache to %s mode",
870 cache_device_name(cache), descs[(int)mode]);
873 static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode)
876 enum cache_metadata_mode old_mode = get_cache_mode(cache);
878 if (dm_cache_metadata_needs_check(cache->cmd, &needs_check)) {
880 cache_device_name(cache));
885 DMERR("%s: unable to switch cache to write mode until repaired.",
886 cache_device_name(cache));
900 dm_cache_metadata_set_read_only(cache->cmd);
904 dm_cache_metadata_set_read_write(cache->cmd);
908 cache->features.mode = new_mode;
911 notify_mode_switch(cache, new_mode);
914 static void abort_transaction(struct cache *cache)
916 const char *dev_name = cache_device_name(cache);
918 if (get_cache_mode(cache) >= CM_READ_ONLY)
922 if (dm_cache_metadata_abort(cache->cmd)) {
924 set_cache_mode(cache, CM_FAIL);
927 if (dm_cache_metadata_set_needs_check(cache->cmd)) {
929 set_cache_mode(cache, CM_FAIL);
933 static void metadata_operation_failed(struct cache *cache, const char *op, int r)
936 cache_device_name(cache), op, r);
937 abort_transaction(cache);
938 set_cache_mode(cache, CM_READ_ONLY);
943 static void load_stats(struct cache *cache)
947 dm_cache_metadata_get_stats(cache->cmd, &stats);
948 atomic_set(&cache->stats.read_hit, stats.read_hits);
949 atomic_set(&cache->stats.read_miss, stats.read_misses);
950 atomic_set(&cache->stats.write_hit, stats.write_hits);
951 atomic_set(&cache->stats.write_miss, stats.write_misses);
954 static void save_stats(struct cache *cache)
958 if (get_cache_mode(cache) >= CM_READ_ONLY)
961 stats.read_hits = atomic_read(&cache->stats.read_hit);
962 stats.read_misses = atomic_read(&cache->stats.read_miss);
963 stats.write_hits = atomic_read(&cache->stats.write_hit);
964 stats.write_misses = atomic_read(&cache->stats.write_miss);
966 dm_cache_metadata_set_stats(cache->cmd, &stats);
990 * Migration covers moving data from the origin device to the cache, or
994 static void inc_io_migrations(struct cache *cache)
996 atomic_inc(&cache->nr_io_migrations);
999 static void dec_io_migrations(struct cache *cache)
1001 atomic_dec(&cache->nr_io_migrations);
1009 static void calc_discard_block_range(struct cache *cache, struct bio *bio,
1015 *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size));
1017 if (se - sb < cache->discard_block_size)
1020 *e = to_dblock(block_div(se, cache->discard_block_size));
1025 static void prevent_background_work(struct cache *cache)
1028 down_write(&cache->background_work_lock);
1032 static void allow_background_work(struct cache *cache)
1035 up_write(&cache->background_work_lock);
1039 static bool background_work_begin(struct cache *cache)
1044 r = down_read_trylock(&cache->background_work_lock);
1050 static void background_work_end(struct cache *cache)
1053 up_read(&cache->background_work_lock);
1059 static bool bio_writes_complete_block(struct cache *cache, struct bio *bio)
1062 (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT));
1065 static bool optimisable_bio(struct cache *cache, struct bio *bio, dm_oblock_t block)
1067 return writeback_mode(cache) &&
1068 (is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio));
1075 dm_cell_quiesce_v2(mg->cache->prison, mg->cell, &mg->k.ws);
1092 queue_continuation(mg->cache->wq, &mg->k);
1098 struct cache *cache = mg->cache;
1100 o_region.bdev = cache->origin_dev->bdev;
1101 o_region.sector = from_oblock(mg->op->oblock) * cache->sectors_per_block;
1102 o_region.count = cache->sectors_per_block;
1104 c_region.bdev = cache->cache_dev->bdev;
1105 c_region.sector = from_cblock(mg->op->cblock) * cache->sectors_per_block;
1106 c_region.count = cache->sectors_per_block;
1109 dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, &mg->k);
1111 dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, &mg->k);
1114 static void bio_drop_shared_lock(struct cache *cache, struct bio *bio)
1118 if (pb->cell && dm_cell_put_v2(cache->prison, pb->cell))
1119 free_prison_cell(cache, pb->cell);
1126 struct cache *cache = mg->cache;
1134 queue_continuation(cache->wq, &mg->k);
1150 remap_to_cache(mg->cache, bio, mg->op->cblock);
1152 remap_to_origin(mg->cache, bio);
1155 accounted_request(mg->cache, bio);
1172 struct cache *cache = mg->cache;
1177 update_stats(&cache->stats, op->op);
1181 clear_discard(cache, oblock_to_dblock(cache, op->oblock));
1182 policy_complete_background_work(cache->policy, op, success);
1186 force_set_dirty(cache, cblock);
1194 force_clear_dirty(cache, cblock);
1195 dec_io_migrations(cache);
1204 force_clear_dirty(cache, cblock);
1205 policy_complete_background_work(cache->policy, op, success);
1206 dec_io_migrations(cache);
1211 force_clear_dirty(cache, cblock);
1212 policy_complete_background_work(cache->policy, op, success);
1213 dec_io_migrations(cache);
1219 if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios))
1220 free_prison_cell(cache, mg->cell);
1224 defer_bios(cache, &bios);
1225 wake_migration_worker(cache);
1227 background_work_end(cache);
1241 struct cache *cache = mg->cache;
1246 r = dm_cache_insert_mapping(cache->cmd, op->cblock, op->oblock);
1249 cache_device_name(cache));
1250 metadata_operation_failed(cache, "dm_cache_insert_mapping", r);
1259 r = dm_cache_remove_mapping(cache->cmd, op->cblock);
1262 cache_device_name(cache));
1263 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
1274 * - vblock x in a cache block
1276 * - cache block gets reallocated and over written
1279 * When we recover, because there was no commit the cache will
1280 * rollback to having the data for vblock x in the cache block.
1281 * But the cache block has since been overwritten, so it'll end
1289 continue_after_commit(&cache->committer, &mg->k);
1290 schedule_commit(&cache->committer);
1327 r = dm_cell_lock_promote_v2(mg->cache->prison, mg->cell,
1343 struct cache *cache = mg->cache;
1347 if ((!is_policy_promote && !is_dirty(cache, op->cblock)) ||
1348 is_discarded_oblock(cache, op->oblock)) {
1367 if (!optimisable_bio(mg->cache, mg->overwrite_bio, mg->op->oblock)) {
1371 bool rb = bio_detain_shared(mg->cache, mg->op->oblock, mg->overwrite_bio);
1375 inc_io_migrations(mg->cache);
1397 struct cache *cache = mg->cache;
1400 prealloc = alloc_prison_cell(cache);
1408 r = dm_cell_lock_v2(cache->prison, &key,
1412 free_prison_cell(cache, prealloc);
1418 free_prison_cell(cache, prealloc);
1428 static int mg_start(struct cache *cache, struct policy_work *op, struct bio *bio)
1432 if (!background_work_begin(cache)) {
1433 policy_complete_background_work(cache->policy, op, false);
1437 mg = alloc_migration(cache);
1443 inc_io_migrations(cache);
1457 struct cache *cache = mg->cache;
1460 if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios))
1461 free_prison_cell(cache, mg->cell);
1467 defer_bios(cache, &bios);
1469 background_work_end(cache);
1479 static int invalidate_cblock(struct cache *cache, dm_cblock_t cblock)
1483 r = policy_invalidate_mapping(cache->policy, cblock);
1485 r = dm_cache_remove_mapping(cache->cmd, cblock);
1488 cache_device_name(cache));
1489 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
1499 DMERR("%s: policy_invalidate_mapping failed", cache_device_name(cache));
1508 struct cache *cache = mg->cache;
1510 r = invalidate_cblock(cache, mg->invalidate_cblock);
1517 continue_after_commit(&cache->committer, &mg->k);
1518 remap_to_origin_clear_discard(cache, mg->overwrite_bio, mg->invalidate_oblock);
1520 schedule_commit(&cache->committer);
1527 struct cache *cache = mg->cache;
1530 prealloc = alloc_prison_cell(cache);
1533 r = dm_cell_lock_v2(cache->prison, &key,
1536 free_prison_cell(cache, prealloc);
1542 free_prison_cell(cache, prealloc);
1553 queue_work(cache->wq, &mg->k.ws);
1559 static int invalidate_start(struct cache *cache, dm_cblock_t cblock,
1564 if (!background_work_begin(cache))
1567 mg = alloc_migration(cache);
1587 static enum busy spare_migration_bandwidth(struct cache *cache)
1589 bool idle = dm_iot_idle_for(&cache->tracker, HZ);
1590 sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) *
1591 cache->sectors_per_block;
1593 if (idle && current_volume <= cache->migration_threshold)
1599 static void inc_hit_counter(struct cache *cache, struct bio *bio)
1602 &cache->stats.read_hit : &cache->stats.write_hit);
1605 static void inc_miss_counter(struct cache *cache, struct bio *bio)
1608 &cache->stats.read_miss : &cache->stats.write_miss);
1613 static int map_bio(struct cache *cache, struct bio *bio, dm_oblock_t block,
1622 rb = bio_detain_shared(cache, block, bio);
1636 if (optimisable_bio(cache, bio, block)) {
1639 r = policy_lookup_with_work(cache->policy, block, &cblock, data_dir, true, &op);
1642 cache_device_name(cache), r);
1648 bio_drop_shared_lock(cache, bio);
1650 mg_start(cache, op, bio);
1654 r = policy_lookup(cache->policy, block, &cblock, data_dir, false, &background_queued);
1657 cache_device_name(cache), r);
1663 wake_migration_worker(cache);
1672 inc_miss_counter(cache, bio);
1674 accounted_begin(cache, bio);
1675 remap_to_origin_clear_discard(cache, bio, block);
1688 inc_hit_counter(cache, bio);
1692 * cache blocks that are written to.
1694 if (passthrough_mode(cache)) {
1696 bio_drop_shared_lock(cache, bio);
1697 atomic_inc(&cache->stats.demotion);
1698 invalidate_start(cache, cblock, block, bio);
1700 remap_to_origin_clear_discard(cache, bio, block);
1702 if (bio_data_dir(bio) == WRITE && writethrough_mode(cache) &&
1703 !is_dirty(cache, cblock)) {
1704 remap_to_origin_and_cache(cache, bio, block, cblock);
1705 accounted_begin(cache, bio);
1707 remap_to_cache_dirty(cache, bio, block, cblock);
1719 accounted_complete(cache, bio);
1720 issue_after_commit(&cache->committer, bio);
1728 static bool process_bio(struct cache *cache, struct bio *bio)
1732 if (map_bio(cache, bio, get_bio_block(cache, bio), &commit_needed) == DM_MAPIO_REMAPPED)
1741 static int commit(struct cache *cache, bool clean_shutdown)
1745 if (get_cache_mode(cache) >= CM_READ_ONLY)
1748 atomic_inc(&cache->stats.commit_count);
1749 r = dm_cache_commit(cache->cmd, clean_shutdown);
1751 metadata_operation_failed(cache, "dm_cache_commit", r);
1761 struct cache *cache = context;
1763 if (dm_cache_changed_this_transaction(cache->cmd))
1764 return errno_to_blk_status(commit(cache, false));
1771 static bool process_flush_bio(struct cache *cache, struct bio *bio)
1776 remap_to_origin(cache, bio);
1778 remap_to_cache(cache, bio, 0);
1780 issue_after_commit(&cache->committer, bio);
1784 static bool process_discard_bio(struct cache *cache, struct bio *bio)
1793 calc_discard_block_range(cache, bio, &b, &e);
1795 set_discard(cache, b);
1799 if (cache->features.discard_passdown) {
1800 remap_to_origin(cache, bio);
1810 struct cache *cache = container_of(ws, struct cache, deferred_bio_worker);
1818 spin_lock_irq(&cache->lock);
1819 bio_list_merge(&bios, &cache->deferred_bios);
1820 bio_list_init(&cache->deferred_bios);
1821 spin_unlock_irq(&cache->lock);
1825 commit_needed = process_flush_bio(cache, bio) || commit_needed;
1828 commit_needed = process_discard_bio(cache, bio) || commit_needed;
1831 commit_needed = process_bio(cache, bio) || commit_needed;
1836 schedule_commit(&cache->committer);
1844 static void requeue_deferred_bios(struct cache *cache)
1850 bio_list_merge(&bios, &cache->deferred_bios);
1851 bio_list_init(&cache->deferred_bios);
1866 struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker);
1868 policy_tick(cache->policy, true);
1869 wake_migration_worker(cache);
1870 schedule_commit(&cache->committer);
1871 queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD);
1878 struct cache *cache = container_of(ws, struct cache, migration_worker);
1882 b = spare_migration_bandwidth(cache);
1884 r = policy_get_background_work(cache->policy, b == IDLE, &op);
1890 cache_device_name(cache));
1894 r = mg_start(cache, op, NULL);
1912 static void destroy(struct cache *cache)
1916 mempool_exit(&cache->migration_pool);
1918 if (cache->prison)
1919 dm_bio_prison_destroy_v2(cache->prison);
1921 cancel_delayed_work_sync(&cache->waker);
1922 if (cache->wq)
1923 destroy_workqueue(cache->wq);
1925 if (cache->dirty_bitset)
1926 free_bitset(cache->dirty_bitset);
1928 if (cache->discard_bitset)
1929 free_bitset(cache->discard_bitset);
1931 if (cache->copier)
1932 dm_kcopyd_client_destroy(cache->copier);
1934 if (cache->cmd)
1935 dm_cache_metadata_close(cache->cmd);
1937 if (cache->metadata_dev)
1938 dm_put_device(cache->ti, cache->metadata_dev);
1940 if (cache->origin_dev)
1941 dm_put_device(cache->ti, cache->origin_dev);
1943 if (cache->cache_dev)
1944 dm_put_device(cache->ti, cache->cache_dev);
1946 if (cache->policy)
1947 dm_cache_policy_destroy(cache->policy);
1949 for (i = 0; i < cache->nr_ctr_args ; i++)
1950 kfree(cache->ctr_args[i]);
1951 kfree(cache->ctr_args);
1953 bioset_exit(&cache->bs);
1955 kfree(cache);
1960 struct cache *cache = ti->private;
1962 destroy(cache);
1973 * Construct a cache device mapping.
1975 * cache <metadata dev> <cache dev> <origin dev> <block size>
1980 * cache dev : fast device holding cached data blocks
1982 * block size : cache unit size in sectors
1992 * See cache-policies.txt for details.
1995 * writethrough : write through caching that prohibits cache block
1998 * back cache block contents later for performance reasons,
2080 *error = "Error opening cache device";
2129 *error = "Data block size is larger than the cache device";
2150 {0, 3, "Invalid number of cache feature arguments"},
2189 *error = "Unrecognised cache feature requested";
2195 *error = "Duplicate cache io_mode features requested";
2268 static int process_config_option(struct cache *cache, const char *key, const char *value)
2276 cache->migration_threshold = tmp;
2283 static int set_config_value(struct cache *cache, const char *key, const char *value)
2285 int r = process_config_option(cache, key, value);
2288 r = policy_set_config_value(cache->policy, key, value);
2296 static int set_config_values(struct cache *cache, int argc, const char **argv)
2306 r = set_config_value(cache, argv[0], argv[1]);
2317 static int create_cache_policy(struct cache *cache, struct cache_args *ca,
2321 cache->cache_size,
2322 cache->origin_sectors,
2323 cache->sectors_per_block);
2325 *error = "Error creating cache's policy";
2328 cache->policy = p;
2329 BUG_ON(!cache->policy);
2335 * We want the discard block size to be at least the size of the cache
2360 static void set_cache_size(struct cache *cache, dm_cblock_t size)
2364 if (nr_blocks > (1 << 20) && cache->cache_size != size)
2365 DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n"
2367 "Please consider increasing the cache block size to reduce the overall cache block count.",
2370 cache->cache_size = size;
2375 static int cache_create(struct cache_args *ca, struct cache **result)
2379 struct cache *cache;
2385 cache = kzalloc(sizeof(*cache), GFP_KERNEL);
2386 if (!cache)
2389 cache->ti = ca->ti;
2390 ti->private = cache;
2400 cache->features = ca->features;
2401 if (writethrough_mode(cache)) {
2403 r = bioset_init(&cache->bs, BIO_POOL_SIZE, 0, 0);
2408 cache->metadata_dev = ca->metadata_dev;
2409 cache->origin_dev = ca->origin_dev;
2410 cache->cache_dev = ca->cache_dev;
2414 origin_blocks = cache->origin_sectors = ca->origin_sectors;
2416 cache->origin_blocks = to_oblock(origin_blocks);
2418 cache->sectors_per_block = ca->block_size;
2419 if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) {
2427 cache->sectors_per_block_shift = -1;
2429 set_cache_size(cache, to_cblock(cache_size));
2431 cache->sectors_per_block_shift = __ffs(ca->block_size);
2432 set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift));
2435 r = create_cache_policy(cache, ca, error);
2439 cache->policy_nr_args = ca->policy_argc;
2440 cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD;
2442 r = set_config_values(cache, ca->policy_argc, ca->policy_argv);
2444 *error = "Error setting cache policy's config values";
2448 cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
2450 dm_cache_policy_get_hint_size(cache->policy),
2457 cache->cmd = cmd;
2458 set_cache_mode(cache, CM_WRITE);
2459 if (get_cache_mode(cache) != CM_WRITE) {
2465 if (passthrough_mode(cache)) {
2468 r = dm_cache_metadata_all_clean(cache->cmd, &all_clean);
2480 policy_allow_migrations(cache->policy, false);
2483 spin_lock_init(&cache->lock);
2484 bio_list_init(&cache->deferred_bios);
2485 atomic_set(&cache->nr_allocated_migrations, 0);
2486 atomic_set(&cache->nr_io_migrations, 0);
2487 init_waitqueue_head(&cache->migration_wait);
2490 atomic_set(&cache->nr_dirty, 0);
2491 cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size));
2492 if (!cache->dirty_bitset) {
2496 clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size));
2498 cache->discard_block_size =
2499 calculate_discard_block_size(cache->sectors_per_block,
2500 cache->origin_sectors);
2501 cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors,
2502 cache->discard_block_size));
2503 cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks));
2504 if (!cache->discard_bitset) {
2508 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
2510 cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2511 if (IS_ERR(cache->copier)) {
2513 r = PTR_ERR(cache->copier);
2517 cache->wq = alloc_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM, 0);
2518 if (!cache->wq) {
2522 INIT_WORK(&cache->deferred_bio_worker, process_deferred_bios);
2523 INIT_WORK(&cache->migration_worker, check_migrations);
2524 INIT_DELAYED_WORK(&cache->waker, do_waker);
2526 cache->prison = dm_bio_prison_create_v2(cache->wq);
2527 if (!cache->prison) {
2532 r = mempool_init_slab_pool(&cache->migration_pool, MIGRATION_POOL_SIZE,
2535 *error = "Error creating cache's migration mempool";
2539 cache->need_tick_bio = true;
2540 cache->sized = false;
2541 cache->invalidate = false;
2542 cache->commit_requested = false;
2543 cache->loaded_mappings = false;
2544 cache->loaded_discards = false;
2546 load_stats(cache);
2548 atomic_set(&cache->stats.demotion, 0);
2549 atomic_set(&cache->stats.promotion, 0);
2550 atomic_set(&cache->stats.copies_avoided, 0);
2551 atomic_set(&cache->stats.cache_cell_clash, 0);
2552 atomic_set(&cache->stats.commit_count, 0);
2553 atomic_set(&cache->stats.discard_count, 0);
2555 spin_lock_init(&cache->invalidation_lock);
2556 INIT_LIST_HEAD(&cache->invalidation_requests);
2558 batcher_init(&cache->committer, commit_op, cache,
2559 issue_op, cache, cache->wq);
2560 dm_iot_init(&cache->tracker);
2562 init_rwsem(&cache->background_work_lock);
2563 prevent_background_work(cache);
2565 *result = cache;
2568 destroy(cache);
2572 static int copy_ctr_args(struct cache *cache, int argc, const char **argv)
2590 cache->nr_ctr_args = argc;
2591 cache->ctr_args = copy;
2600 struct cache *cache = NULL;
2604 ti->error = "Error allocating memory for cache";
2613 r = cache_create(ca, &cache);
2617 r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3);
2619 destroy(cache);
2623 ti->private = cache;
2633 struct cache *cache = ti->private;
2637 dm_oblock_t block = get_bio_block(cache, bio);
2640 if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) {
2643 * the end of the origin device. We don't cache these.
2646 remap_to_origin(cache, bio);
2647 accounted_begin(cache, bio);
2652 defer_bio(cache, bio);
2656 r = map_bio(cache, bio, block, &commit_needed);
2658 schedule_commit(&cache->committer);
2665 struct cache *cache = ti->private;
2670 policy_tick(cache->policy, false);
2672 spin_lock_irqsave(&cache->lock, flags);
2673 cache->need_tick_bio = true;
2674 spin_unlock_irqrestore(&cache->lock, flags);
2677 bio_drop_shared_lock(cache, bio);
2678 accounted_complete(cache, bio);
2683 static int write_dirty_bitset(struct cache *cache)
2687 if (get_cache_mode(cache) >= CM_READ_ONLY)
2690 r = dm_cache_set_dirty_bits(cache->cmd, from_cblock(cache->cache_size), cache->dirty_bitset);
2692 metadata_operation_failed(cache, "dm_cache_set_dirty_bits", r);
2697 static int write_discard_bitset(struct cache *cache)
2701 if (get_cache_mode(cache) >= CM_READ_ONLY)
2704 r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size,
2705 cache->discard_nr_blocks);
2707 DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache));
2708 metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r);
2712 for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) {
2713 r = dm_cache_set_discard(cache->cmd, to_dblock(i),
2714 is_discarded(cache, to_dblock(i)));
2716 metadata_operation_failed(cache, "dm_cache_set_discard", r);
2724 static int write_hints(struct cache *cache)
2728 if (get_cache_mode(cache) >= CM_READ_ONLY)
2731 r = dm_cache_write_hints(cache->cmd, cache->policy);
2733 metadata_operation_failed(cache, "dm_cache_write_hints", r);
2743 static bool sync_metadata(struct cache *cache)
2747 r1 = write_dirty_bitset(cache);
2749 DMERR("%s: could not write dirty bitset", cache_device_name(cache));
2751 r2 = write_discard_bitset(cache);
2753 DMERR("%s: could not write discard bitset", cache_device_name(cache));
2755 save_stats(cache);
2757 r3 = write_hints(cache);
2759 DMERR("%s: could not write hints", cache_device_name(cache));
2766 r4 = commit(cache, !r1 && !r2 && !r3);
2768 DMERR("%s: could not write cache metadata", cache_device_name(cache));
2775 struct cache *cache = ti->private;
2777 prevent_background_work(cache);
2778 BUG_ON(atomic_read(&cache->nr_io_migrations));
2780 cancel_delayed_work_sync(&cache->waker);
2781 drain_workqueue(cache->wq);
2782 WARN_ON(cache->tracker.in_flight);
2788 requeue_deferred_bios(cache);
2790 if (get_cache_mode(cache) == CM_WRITE)
2791 (void) sync_metadata(cache);
2797 struct cache *cache = context;
2800 set_bit(from_cblock(cblock), cache->dirty_bitset);
2801 atomic_inc(&cache->nr_dirty);
2803 clear_bit(from_cblock(cblock), cache->dirty_bitset);
2805 return policy_load_mapping(cache->policy, oblock, cblock, dirty, hint, hint_valid);
2815 struct cache *cache;
2825 static void discard_load_info_init(struct cache *cache,
2828 li->cache = cache;
2848 b = dm_sector_div_up(b, li->cache->discard_block_size);
2849 sector_div(e, li->cache->discard_block_size);
2855 if (e > from_dblock(li->cache->discard_nr_blocks))
2856 e = from_dblock(li->cache->discard_nr_blocks);
2859 set_discard(li->cache, to_dblock(b));
2892 static dm_cblock_t get_cache_dev_size(struct cache *cache)
2894 sector_t size = get_dev_size(cache->cache_dev);
2895 (void) sector_div(size, cache->sectors_per_block);
2899 static bool can_resize(struct cache *cache, dm_cblock_t new_size)
2901 if (from_cblock(new_size) > from_cblock(cache->cache_size)) {
2902 if (cache->sized) {
2903 DMERR("%s: unable to extend cache due to missing cache table reload",
2904 cache_device_name(cache));
2910 * We can't drop a dirty block when shrinking the cache.
2912 while (from_cblock(new_size) < from_cblock(cache->cache_size)) {
2914 if (is_dirty(cache, new_size)) {
2915 DMERR("%s: unable to shrink cache; cache block %llu is dirty",
2916 cache_device_name(cache),
2925 static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size)
2929 r = dm_cache_resize(cache->cmd, new_size);
2931 DMERR("%s: could not resize cache metadata", cache_device_name(cache));
2932 metadata_operation_failed(cache, "dm_cache_resize", r);
2936 set_cache_size(cache, new_size);
2944 struct cache *cache = ti->private;
2945 dm_cblock_t csize = get_cache_dev_size(cache);
2948 * Check to see if the cache has resized.
2950 if (!cache->sized) {
2951 r = resize_cache_dev(cache, csize);
2955 cache->sized = true;
2957 } else if (csize != cache->cache_size) {
2958 if (!can_resize(cache, csize))
2961 r = resize_cache_dev(cache, csize);
2966 if (!cache->loaded_mappings) {
2967 r = dm_cache_load_mappings(cache->cmd, cache->policy,
2968 load_mapping, cache);
2970 DMERR("%s: could not load cache mappings", cache_device_name(cache));
2971 metadata_operation_failed(cache, "dm_cache_load_mappings", r);
2975 cache->loaded_mappings = true;
2978 if (!cache->loaded_discards) {
2986 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
2988 discard_load_info_init(cache, &li);
2989 r = dm_cache_load_discards(cache->cmd, load_discard, &li);
2991 DMERR("%s: could not load origin discards", cache_device_name(cache));
2992 metadata_operation_failed(cache, "dm_cache_load_discards", r);
2997 cache->loaded_discards = true;
3005 struct cache *cache = ti->private;
3007 cache->need_tick_bio = true;
3008 allow_background_work(cache);
3009 do_waker(&cache->waker.work);
3012 static void emit_flags(struct cache *cache, char *result,
3016 struct cache_features *cf = &cache->features;
3024 if (writethrough_mode(cache))
3027 else if (passthrough_mode(cache))
3030 else if (writeback_mode(cache))
3036 cache_device_name(cache), (int) cf->io_mode);
3049 * <cache block size> <#used cache blocks>/<#total cache blocks>
3054 * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check>
3065 struct cache *cache = ti->private;
3071 if (get_cache_mode(cache) == CM_FAIL) {
3078 (void) commit(cache, false);
3080 r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata);
3083 cache_device_name(cache), r);
3087 r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata);
3090 cache_device_name(cache), r);
3094 residency = policy_residency(cache->policy);
3100 (unsigned long long)cache->sectors_per_block,
3102 (unsigned long long) from_cblock(cache->cache_size),
3103 (unsigned int) atomic_read(&cache->stats.read_hit),
3104 (unsigned int) atomic_read(&cache->stats.read_miss),
3105 (unsigned int) atomic_read(&cache->stats.write_hit),
3106 (unsigned int) atomic_read(&cache->stats.write_miss),
3107 (unsigned int) atomic_read(&cache->stats.demotion),
3108 (unsigned int) atomic_read(&cache->stats.promotion),
3109 (unsigned long) atomic_read(&cache->nr_dirty));
3111 emit_flags(cache, result, maxlen, &sz);
3113 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
3115 DMEMIT("%s ", dm_cache_policy_get_name(cache->policy));
3117 r = policy_emit_config_values(cache->policy, result, maxlen, &sz);
3120 cache_device_name(cache), r);
3123 if (get_cache_mode(cache) == CM_READ_ONLY)
3128 r = dm_cache_metadata_needs_check(cache->cmd, &needs_check);
3138 format_dev_t(buf, cache->metadata_dev->bdev->bd_dev);
3140 format_dev_t(buf, cache->cache_dev->bdev->bd_dev);
3142 format_dev_t(buf, cache->origin_dev->bdev->bd_dev);
3145 for (i = 0; i < cache->nr_ctr_args - 1; i++)
3146 DMEMIT(" %s", cache->ctr_args[i]);
3147 if (cache->nr_ctr_args)
3148 DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]);
3153 if (get_cache_mode(cache) == CM_FAIL)
3155 else if (get_cache_mode(cache) == CM_READ_ONLY)
3160 format_dev_t(buf, cache->metadata_dev->bdev->bd_dev);
3162 format_dev_t(buf, cache->cache_dev->bdev->bd_dev);
3164 format_dev_t(buf, cache->origin_dev->bdev->bd_dev);
3166 DMEMIT(",writethrough=%c", writethrough_mode(cache) ? 'y' : 'n');
3167 DMEMIT(",writeback=%c", writeback_mode(cache) ? 'y' : 'n');
3168 DMEMIT(",passthrough=%c", passthrough_mode(cache) ? 'y' : 'n');
3169 DMEMIT(",metadata2=%c", cache->features.metadata_version == 2 ? 'y' : 'n');
3170 DMEMIT(",no_discard_passdown=%c", cache->features.discard_passdown ? 'n' : 'y');
3191 * A cache block range can take two forms:
3196 static int parse_cblock_range(struct cache *cache, const char *str,
3229 DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str);
3233 static int validate_cblock_range(struct cache *cache, struct cblock_range *range)
3237 uint64_t n = from_cblock(cache->cache_size);
3241 cache_device_name(cache), b, n);
3247 cache_device_name(cache), e, n);
3253 cache_device_name(cache), b, e);
3265 static int request_invalidation(struct cache *cache, struct cblock_range *range)
3276 r = invalidate_cblock(cache, range->begin);
3283 cache->commit_requested = true;
3287 static int process_invalidate_cblocks_message(struct cache *cache, unsigned int count,
3294 if (!passthrough_mode(cache)) {
3295 DMERR("%s: cache has to be in passthrough mode for invalidation",
3296 cache_device_name(cache));
3301 r = parse_cblock_range(cache, cblock_ranges[i], &range);
3305 r = validate_cblock_range(cache, &range);
3312 r = request_invalidation(cache, &range);
3326 * The key migration_threshold is supported by the cache target core.
3331 struct cache *cache = ti->private;
3336 if (get_cache_mode(cache) >= CM_READ_ONLY) {
3337 DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode",
3338 cache_device_name(cache));
3343 return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1);
3348 return set_config_value(cache, argv[0], argv[1]);
3355 struct cache *cache = ti->private;
3357 r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data);
3359 r = fn(ti, cache->origin_dev, 0, ti->len, data);
3368 static void disable_passdown_if_not_supported(struct cache *cache)
3370 struct block_device *origin_bdev = cache->origin_dev->bdev;
3374 if (!cache->features.discard_passdown)
3380 else if (origin_limits->max_discard_sectors < cache->sectors_per_block)
3386 cache->features.discard_passdown = false;
3390 static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
3392 struct block_device *origin_bdev = cache->origin_dev->bdev;
3395 if (!cache->features.discard_passdown) {
3397 limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024,
3398 cache->origin_sectors);
3399 limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
3416 struct cache *cache = ti->private;
3421 * cache's blocksize (io_opt is a factor) do not override them.
3423 if (io_opt_sectors < cache->sectors_per_block ||
3424 do_div(io_opt_sectors, cache->sectors_per_block)) {
3425 blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT);
3426 blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
3429 disable_passdown_if_not_supported(cache);
3430 set_discard_limits(cache, limits);
3436 .name = "cache",
3478 MODULE_DESCRIPTION(DM_NAME " cache target");