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, int err)
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 void __cell_release(struct cache *cache, struct dm_bio_prison_cell *cell,
1067 bool holder, struct bio_list *bios)
1069 (holder ? dm_cell_release : dm_cell_release_no_holder)
1070 (cache->prison, cell, bios);
1071 free_prison_cell(cache, cell);
1074 static bool discard_or_flush(struct bio *bio)
1076 return bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_DISCARD);
1079 static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell)
1081 if (discard_or_flush(cell->holder))
1083 * We have to handle these bios
1086 __cell_release(cache, cell, true, &cache->deferred_bios);
1089 list_add_tail(&cell->user_list, &cache->deferred_cells);
1092 static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, bool holder)
1094 unsigned long flags;
1096 if (!holder && dm_cell_promote_or_release(cache->prison, cell)) {
1098 * There was no prisoner to promote to holder, the
1099 * cell has been released.
1101 free_prison_cell(cache, cell);
1105 spin_lock_irqsave(&cache->lock, flags);
1106 __cell_defer(cache, cell);
1107 spin_unlock_irqrestore(&cache->lock, flags);
1112 static void cell_error_with_code(struct cache *cache, struct dm_bio_prison_cell *cell, int err)
1114 dm_cell_error(cache->prison, cell, err);
1115 dm_bio_prison_free_cell(cache->prison, cell);
1118 static void cell_requeue(struct cache *cache, struct dm_bio_prison_cell *cell)
1120 cell_error_with_code(cache, cell, DM_ENDIO_REQUEUE);
1123 static void free_io_migration(struct dm_cache_migration *mg)
1125 dec_io_migrations(mg->cache);
1127 wake_worker(mg->cache);
1130 static void migration_failure(struct dm_cache_migration *mg)
1132 struct cache *cache = mg->cache;
1133 const char *dev_name = cache_device_name(cache);
1135 if (mg->writeback) {
1136 DMERR_LIMIT("%s: writeback failed; couldn't copy block", dev_name);
1137 set_dirty(cache, mg->old_oblock, mg->cblock);
1138 cell_defer(cache, mg->old_ocell, false);
1140 } else if (mg->demote) {
1141 DMERR_LIMIT("%s: demotion failed; couldn't copy block", dev_name);
1142 policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock);
1144 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1146 cell_defer(cache, mg->new_ocell, true);
1148 DMERR_LIMIT("%s: promotion failed; couldn't copy block", dev_name);
1149 policy_remove_mapping(cache->policy, mg->new_oblock);
1150 cell_defer(cache, mg->new_ocell, true);
1153 free_io_migration(mg);
1156 static void migration_success_pre_commit(struct dm_cache_migration *mg)
1159 unsigned long flags;
1160 struct cache *cache = mg->cache;
1162 if (mg->writeback) {
1163 clear_dirty(cache, mg->old_oblock, mg->cblock);
1164 cell_defer(cache, mg->old_ocell, false);
1165 free_io_migration(mg);
1168 } else if (mg->demote) {
1169 r = dm_cache_remove_mapping(cache->cmd, mg->cblock);
1171 DMERR_LIMIT("%s: demotion failed; couldn't update on disk metadata",
1172 cache_device_name(cache));
1173 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
1174 policy_force_mapping(cache->policy, mg->new_oblock,
1177 cell_defer(cache, mg->new_ocell, true);
1178 free_io_migration(mg);
1182 r = dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock);
1184 DMERR_LIMIT("%s: promotion failed; couldn't update on disk metadata",
1185 cache_device_name(cache));
1186 metadata_operation_failed(cache, "dm_cache_insert_mapping", r);
1187 policy_remove_mapping(cache->policy, mg->new_oblock);
1188 free_io_migration(mg);
1193 spin_lock_irqsave(&cache->lock, flags);
1194 list_add_tail(&mg->list, &cache->need_commit_migrations);
1195 cache->commit_requested = true;
1196 spin_unlock_irqrestore(&cache->lock, flags);
1199 static void migration_success_post_commit(struct dm_cache_migration *mg)
1201 unsigned long flags;
1202 struct cache *cache = mg->cache;
1204 if (mg->writeback) {
1205 DMWARN_LIMIT("%s: writeback unexpectedly triggered commit",
1206 cache_device_name(cache));
1209 } else if (mg->demote) {
1210 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1215 spin_lock_irqsave(&cache->lock, flags);
1216 list_add_tail(&mg->list, &cache->quiesced_migrations);
1217 spin_unlock_irqrestore(&cache->lock, flags);
1221 policy_remove_mapping(cache->policy, mg->old_oblock);
1222 free_io_migration(mg);
1226 if (mg->requeue_holder) {
1227 clear_dirty(cache, mg->new_oblock, mg->cblock);
1228 cell_defer(cache, mg->new_ocell, true);
1231 * The block was promoted via an overwrite, so it's dirty.
1233 set_dirty(cache, mg->new_oblock, mg->cblock);
1234 bio_endio(mg->new_ocell->holder, 0);
1235 cell_defer(cache, mg->new_ocell, false);
1237 free_io_migration(mg);
1241 static void copy_complete(int read_err, unsigned long write_err, void *context)
1243 unsigned long flags;
1244 struct dm_cache_migration *mg = (struct dm_cache_migration *) context;
1245 struct cache *cache = mg->cache;
1247 if (read_err || write_err)
1250 spin_lock_irqsave(&cache->lock, flags);
1251 list_add_tail(&mg->list, &cache->completed_migrations);
1252 spin_unlock_irqrestore(&cache->lock, flags);
1257 static void issue_copy(struct dm_cache_migration *mg)
1260 struct dm_io_region o_region, c_region;
1261 struct cache *cache = mg->cache;
1262 sector_t cblock = from_cblock(mg->cblock);
1264 o_region.bdev = cache->origin_dev->bdev;
1265 o_region.count = cache->sectors_per_block;
1267 c_region.bdev = cache->cache_dev->bdev;
1268 c_region.sector = cblock * cache->sectors_per_block;
1269 c_region.count = cache->sectors_per_block;
1271 if (mg->writeback || mg->demote) {
1273 o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block;
1274 r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg);
1277 o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block;
1278 r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg);
1282 DMERR_LIMIT("%s: issuing migration failed", cache_device_name(cache));
1283 migration_failure(mg);
1287 static void overwrite_endio(struct bio *bio, int err)
1289 struct dm_cache_migration *mg = bio->bi_private;
1290 struct cache *cache = mg->cache;
1291 size_t pb_data_size = get_per_bio_data_size(cache);
1292 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1293 unsigned long flags;
1295 dm_unhook_bio(&pb->hook_info, bio);
1300 mg->requeue_holder = false;
1302 spin_lock_irqsave(&cache->lock, flags);
1303 list_add_tail(&mg->list, &cache->completed_migrations);
1304 spin_unlock_irqrestore(&cache->lock, flags);
1309 static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio)
1311 size_t pb_data_size = get_per_bio_data_size(mg->cache);
1312 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1314 dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg);
1315 remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock);
1318 * No need to inc_ds() here, since the cell will be held for the
1319 * duration of the io.
1321 accounted_request(mg->cache, bio);
1324 static bool bio_writes_complete_block(struct cache *cache, struct bio *bio)
1326 return (bio_data_dir(bio) == WRITE) &&
1327 (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT));
1330 static void avoid_copy(struct dm_cache_migration *mg)
1332 atomic_inc(&mg->cache->stats.copies_avoided);
1333 migration_success_pre_commit(mg);
1336 static void calc_discard_block_range(struct cache *cache, struct bio *bio,
1337 dm_dblock_t *b, dm_dblock_t *e)
1339 sector_t sb = bio->bi_iter.bi_sector;
1340 sector_t se = bio_end_sector(bio);
1342 *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size));
1344 if (se - sb < cache->discard_block_size)
1347 *e = to_dblock(block_div(se, cache->discard_block_size));
1350 static void issue_discard(struct dm_cache_migration *mg)
1353 struct bio *bio = mg->new_ocell->holder;
1355 calc_discard_block_range(mg->cache, bio, &b, &e);
1357 set_discard(mg->cache, b);
1358 b = to_dblock(from_dblock(b) + 1);
1362 cell_defer(mg->cache, mg->new_ocell, false);
1364 wake_worker(mg->cache);
1367 static void issue_copy_or_discard(struct dm_cache_migration *mg)
1370 struct cache *cache = mg->cache;
1377 if (mg->writeback || mg->demote)
1378 avoid = !is_dirty(cache, mg->cblock) ||
1379 is_discarded_oblock(cache, mg->old_oblock);
1381 struct bio *bio = mg->new_ocell->holder;
1383 avoid = is_discarded_oblock(cache, mg->new_oblock);
1385 if (writeback_mode(&cache->features) &&
1386 !avoid && bio_writes_complete_block(cache, bio)) {
1387 issue_overwrite(mg, bio);
1392 avoid ? avoid_copy(mg) : issue_copy(mg);
1395 static void complete_migration(struct dm_cache_migration *mg)
1398 migration_failure(mg);
1400 migration_success_pre_commit(mg);
1403 static void process_migrations(struct cache *cache, struct list_head *head,
1404 void (*fn)(struct dm_cache_migration *))
1406 unsigned long flags;
1407 struct list_head list;
1408 struct dm_cache_migration *mg, *tmp;
1410 INIT_LIST_HEAD(&list);
1411 spin_lock_irqsave(&cache->lock, flags);
1412 list_splice_init(head, &list);
1413 spin_unlock_irqrestore(&cache->lock, flags);
1415 list_for_each_entry_safe(mg, tmp, &list, list)
1419 static void __queue_quiesced_migration(struct dm_cache_migration *mg)
1421 list_add_tail(&mg->list, &mg->cache->quiesced_migrations);
1424 static void queue_quiesced_migration(struct dm_cache_migration *mg)
1426 unsigned long flags;
1427 struct cache *cache = mg->cache;
1429 spin_lock_irqsave(&cache->lock, flags);
1430 __queue_quiesced_migration(mg);
1431 spin_unlock_irqrestore(&cache->lock, flags);
1436 static void queue_quiesced_migrations(struct cache *cache, struct list_head *work)
1438 unsigned long flags;
1439 struct dm_cache_migration *mg, *tmp;
1441 spin_lock_irqsave(&cache->lock, flags);
1442 list_for_each_entry_safe(mg, tmp, work, list)
1443 __queue_quiesced_migration(mg);
1444 spin_unlock_irqrestore(&cache->lock, flags);
1449 static void check_for_quiesced_migrations(struct cache *cache,
1450 struct per_bio_data *pb)
1452 struct list_head work;
1454 if (!pb->all_io_entry)
1457 INIT_LIST_HEAD(&work);
1458 dm_deferred_entry_dec(pb->all_io_entry, &work);
1460 if (!list_empty(&work))
1461 queue_quiesced_migrations(cache, &work);
1464 static void quiesce_migration(struct dm_cache_migration *mg)
1466 if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list))
1467 queue_quiesced_migration(mg);
1470 static void promote(struct cache *cache, struct prealloc *structs,
1471 dm_oblock_t oblock, dm_cblock_t cblock,
1472 struct dm_bio_prison_cell *cell)
1474 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1477 mg->discard = false;
1478 mg->writeback = false;
1481 mg->requeue_holder = true;
1482 mg->invalidate = false;
1484 mg->new_oblock = oblock;
1485 mg->cblock = cblock;
1486 mg->old_ocell = NULL;
1487 mg->new_ocell = cell;
1488 mg->start_jiffies = jiffies;
1490 inc_io_migrations(cache);
1491 quiesce_migration(mg);
1494 static void writeback(struct cache *cache, struct prealloc *structs,
1495 dm_oblock_t oblock, dm_cblock_t cblock,
1496 struct dm_bio_prison_cell *cell)
1498 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1501 mg->discard = false;
1502 mg->writeback = true;
1504 mg->promote = false;
1505 mg->requeue_holder = true;
1506 mg->invalidate = false;
1508 mg->old_oblock = oblock;
1509 mg->cblock = cblock;
1510 mg->old_ocell = cell;
1511 mg->new_ocell = NULL;
1512 mg->start_jiffies = jiffies;
1514 inc_io_migrations(cache);
1515 quiesce_migration(mg);
1518 static void demote_then_promote(struct cache *cache, struct prealloc *structs,
1519 dm_oblock_t old_oblock, dm_oblock_t new_oblock,
1521 struct dm_bio_prison_cell *old_ocell,
1522 struct dm_bio_prison_cell *new_ocell)
1524 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1527 mg->discard = false;
1528 mg->writeback = false;
1531 mg->requeue_holder = true;
1532 mg->invalidate = false;
1534 mg->old_oblock = old_oblock;
1535 mg->new_oblock = new_oblock;
1536 mg->cblock = cblock;
1537 mg->old_ocell = old_ocell;
1538 mg->new_ocell = new_ocell;
1539 mg->start_jiffies = jiffies;
1541 inc_io_migrations(cache);
1542 quiesce_migration(mg);
1546 * Invalidate a cache entry. No writeback occurs; any changes in the cache
1547 * block are thrown away.
1549 static void invalidate(struct cache *cache, struct prealloc *structs,
1550 dm_oblock_t oblock, dm_cblock_t cblock,
1551 struct dm_bio_prison_cell *cell)
1553 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1556 mg->discard = false;
1557 mg->writeback = false;
1559 mg->promote = false;
1560 mg->requeue_holder = true;
1561 mg->invalidate = true;
1563 mg->old_oblock = oblock;
1564 mg->cblock = cblock;
1565 mg->old_ocell = cell;
1566 mg->new_ocell = NULL;
1567 mg->start_jiffies = jiffies;
1569 inc_io_migrations(cache);
1570 quiesce_migration(mg);
1573 static void discard(struct cache *cache, struct prealloc *structs,
1574 struct dm_bio_prison_cell *cell)
1576 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1580 mg->writeback = false;
1582 mg->promote = false;
1583 mg->requeue_holder = false;
1584 mg->invalidate = false;
1586 mg->old_ocell = NULL;
1587 mg->new_ocell = cell;
1588 mg->start_jiffies = jiffies;
1590 quiesce_migration(mg);
1593 /*----------------------------------------------------------------
1595 *--------------------------------------------------------------*/
1596 static void defer_bio(struct cache *cache, struct bio *bio)
1598 unsigned long flags;
1600 spin_lock_irqsave(&cache->lock, flags);
1601 bio_list_add(&cache->deferred_bios, bio);
1602 spin_unlock_irqrestore(&cache->lock, flags);
1607 static void process_flush_bio(struct cache *cache, struct bio *bio)
1609 size_t pb_data_size = get_per_bio_data_size(cache);
1610 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1612 BUG_ON(bio->bi_iter.bi_size);
1614 remap_to_origin(cache, bio);
1616 remap_to_cache(cache, bio, 0);
1619 * REQ_FLUSH is not directed at any particular block so we don't
1620 * need to inc_ds(). REQ_FUA's are split into a write + REQ_FLUSH
1626 static void process_discard_bio(struct cache *cache, struct prealloc *structs,
1631 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1633 calc_discard_block_range(cache, bio, &b, &e);
1639 cell_prealloc = prealloc_get_cell(structs);
1640 r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc,
1641 (cell_free_fn) prealloc_put_cell,
1642 structs, &new_ocell);
1646 discard(cache, structs, new_ocell);
1649 static bool spare_migration_bandwidth(struct cache *cache)
1651 sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) *
1652 cache->sectors_per_block;
1653 return current_volume < cache->migration_threshold;
1656 static void inc_hit_counter(struct cache *cache, struct bio *bio)
1658 atomic_inc(bio_data_dir(bio) == READ ?
1659 &cache->stats.read_hit : &cache->stats.write_hit);
1662 static void inc_miss_counter(struct cache *cache, struct bio *bio)
1664 atomic_inc(bio_data_dir(bio) == READ ?
1665 &cache->stats.read_miss : &cache->stats.write_miss);
1668 /*----------------------------------------------------------------*/
1671 struct cache *cache;
1672 struct bio_list bios_for_issue;
1673 struct bio_list unhandled_bios;
1677 static void inc_fn(void *context, struct dm_bio_prison_cell *cell)
1680 struct inc_detail *detail = context;
1681 struct cache *cache = detail->cache;
1683 inc_ds(cache, cell->holder, cell);
1684 if (bio_data_dir(cell->holder) == WRITE)
1685 detail->any_writes = true;
1687 while ((bio = bio_list_pop(&cell->bios))) {
1688 if (discard_or_flush(bio)) {
1689 bio_list_add(&detail->unhandled_bios, bio);
1693 if (bio_data_dir(bio) == WRITE)
1694 detail->any_writes = true;
1696 bio_list_add(&detail->bios_for_issue, bio);
1697 inc_ds(cache, bio, cell);
1701 // FIXME: refactor these two
1702 static void remap_cell_to_origin_clear_discard(struct cache *cache,
1703 struct dm_bio_prison_cell *cell,
1704 dm_oblock_t oblock, bool issue_holder)
1707 unsigned long flags;
1708 struct inc_detail detail;
1710 detail.cache = cache;
1711 bio_list_init(&detail.bios_for_issue);
1712 bio_list_init(&detail.unhandled_bios);
1713 detail.any_writes = false;
1715 spin_lock_irqsave(&cache->lock, flags);
1716 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1717 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1718 spin_unlock_irqrestore(&cache->lock, flags);
1720 remap_to_origin(cache, cell->holder);
1722 issue(cache, cell->holder);
1724 accounted_begin(cache, cell->holder);
1726 if (detail.any_writes)
1727 clear_discard(cache, oblock_to_dblock(cache, oblock));
1729 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1730 remap_to_origin(cache, bio);
1734 free_prison_cell(cache, cell);
1737 static void remap_cell_to_cache_dirty(struct cache *cache, struct dm_bio_prison_cell *cell,
1738 dm_oblock_t oblock, dm_cblock_t cblock, bool issue_holder)
1741 unsigned long flags;
1742 struct inc_detail detail;
1744 detail.cache = cache;
1745 bio_list_init(&detail.bios_for_issue);
1746 bio_list_init(&detail.unhandled_bios);
1747 detail.any_writes = false;
1749 spin_lock_irqsave(&cache->lock, flags);
1750 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1751 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1752 spin_unlock_irqrestore(&cache->lock, flags);
1754 remap_to_cache(cache, cell->holder, cblock);
1756 issue(cache, cell->holder);
1758 accounted_begin(cache, cell->holder);
1760 if (detail.any_writes) {
1761 set_dirty(cache, oblock, cblock);
1762 clear_discard(cache, oblock_to_dblock(cache, oblock));
1765 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1766 remap_to_cache(cache, bio, cblock);
1770 free_prison_cell(cache, cell);
1773 /*----------------------------------------------------------------*/
1775 struct old_oblock_lock {
1776 struct policy_locker locker;
1777 struct cache *cache;
1778 struct prealloc *structs;
1779 struct dm_bio_prison_cell *cell;
1782 static int null_locker(struct policy_locker *locker, dm_oblock_t b)
1784 /* This should never be called */
1789 static int cell_locker(struct policy_locker *locker, dm_oblock_t b)
1791 struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker);
1792 struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs);
1794 return bio_detain(l->cache, b, NULL, cell_prealloc,
1795 (cell_free_fn) prealloc_put_cell,
1796 l->structs, &l->cell);
1799 static void process_cell(struct cache *cache, struct prealloc *structs,
1800 struct dm_bio_prison_cell *new_ocell)
1803 bool release_cell = true;
1804 struct bio *bio = new_ocell->holder;
1805 dm_oblock_t block = get_bio_block(cache, bio);
1806 struct policy_result lookup_result;
1807 bool passthrough = passthrough_mode(&cache->features);
1808 bool fast_promotion, can_migrate;
1809 struct old_oblock_lock ool;
1811 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
1812 can_migrate = !passthrough && (fast_promotion || spare_migration_bandwidth(cache));
1814 ool.locker.fn = cell_locker;
1816 ool.structs = structs;
1818 r = policy_map(cache->policy, block, true, can_migrate, fast_promotion,
1819 bio, &ool.locker, &lookup_result);
1821 if (r == -EWOULDBLOCK)
1822 /* migration has been denied */
1823 lookup_result.op = POLICY_MISS;
1825 switch (lookup_result.op) {
1828 inc_miss_counter(cache, bio);
1831 * Passthrough always maps to the origin,
1832 * invalidating any cache blocks that are written
1836 if (bio_data_dir(bio) == WRITE) {
1837 atomic_inc(&cache->stats.demotion);
1838 invalidate(cache, structs, block, lookup_result.cblock, new_ocell);
1839 release_cell = false;
1842 /* FIXME: factor out issue_origin() */
1843 remap_to_origin_clear_discard(cache, bio, block);
1844 inc_and_issue(cache, bio, new_ocell);
1847 inc_hit_counter(cache, bio);
1849 if (bio_data_dir(bio) == WRITE &&
1850 writethrough_mode(&cache->features) &&
1851 !is_dirty(cache, lookup_result.cblock)) {
1852 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
1853 inc_and_issue(cache, bio, new_ocell);
1856 remap_cell_to_cache_dirty(cache, new_ocell, block, lookup_result.cblock, true);
1857 release_cell = false;
1864 inc_miss_counter(cache, bio);
1865 remap_cell_to_origin_clear_discard(cache, new_ocell, block, true);
1866 release_cell = false;
1870 atomic_inc(&cache->stats.promotion);
1871 promote(cache, structs, block, lookup_result.cblock, new_ocell);
1872 release_cell = false;
1875 case POLICY_REPLACE:
1876 atomic_inc(&cache->stats.demotion);
1877 atomic_inc(&cache->stats.promotion);
1878 demote_then_promote(cache, structs, lookup_result.old_oblock,
1879 block, lookup_result.cblock,
1880 ool.cell, new_ocell);
1881 release_cell = false;
1885 DMERR_LIMIT("%s: %s: erroring bio, unknown policy op: %u",
1886 cache_device_name(cache), __func__,
1887 (unsigned) lookup_result.op);
1892 cell_defer(cache, new_ocell, false);
1895 static void process_bio(struct cache *cache, struct prealloc *structs,
1899 dm_oblock_t block = get_bio_block(cache, bio);
1900 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1903 * Check to see if that block is currently migrating.
1905 cell_prealloc = prealloc_get_cell(structs);
1906 r = bio_detain(cache, block, bio, cell_prealloc,
1907 (cell_free_fn) prealloc_put_cell,
1908 structs, &new_ocell);
1912 process_cell(cache, structs, new_ocell);
1915 static int need_commit_due_to_time(struct cache *cache)
1917 return jiffies < cache->last_commit_jiffies ||
1918 jiffies > cache->last_commit_jiffies + COMMIT_PERIOD;
1922 * A non-zero return indicates read_only or fail_io mode.
1924 static int commit(struct cache *cache, bool clean_shutdown)
1928 if (get_cache_mode(cache) >= CM_READ_ONLY)
1931 atomic_inc(&cache->stats.commit_count);
1932 r = dm_cache_commit(cache->cmd, clean_shutdown);
1934 metadata_operation_failed(cache, "dm_cache_commit", r);
1939 static int commit_if_needed(struct cache *cache)
1943 if ((cache->commit_requested || need_commit_due_to_time(cache)) &&
1944 dm_cache_changed_this_transaction(cache->cmd)) {
1945 r = commit(cache, false);
1946 cache->commit_requested = false;
1947 cache->last_commit_jiffies = jiffies;
1953 static void process_deferred_bios(struct cache *cache)
1955 bool prealloc_used = false;
1956 unsigned long flags;
1957 struct bio_list bios;
1959 struct prealloc structs;
1961 memset(&structs, 0, sizeof(structs));
1962 bio_list_init(&bios);
1964 spin_lock_irqsave(&cache->lock, flags);
1965 bio_list_merge(&bios, &cache->deferred_bios);
1966 bio_list_init(&cache->deferred_bios);
1967 spin_unlock_irqrestore(&cache->lock, flags);
1969 while (!bio_list_empty(&bios)) {
1971 * If we've got no free migration structs, and processing
1972 * this bio might require one, we pause until there are some
1973 * prepared mappings to process.
1975 prealloc_used = true;
1976 if (prealloc_data_structs(cache, &structs)) {
1977 spin_lock_irqsave(&cache->lock, flags);
1978 bio_list_merge(&cache->deferred_bios, &bios);
1979 spin_unlock_irqrestore(&cache->lock, flags);
1983 bio = bio_list_pop(&bios);
1985 if (bio->bi_rw & REQ_FLUSH)
1986 process_flush_bio(cache, bio);
1987 else if (bio->bi_rw & REQ_DISCARD)
1988 process_discard_bio(cache, &structs, bio);
1990 process_bio(cache, &structs, bio);
1994 prealloc_free_structs(cache, &structs);
1997 static void process_deferred_cells(struct cache *cache)
1999 bool prealloc_used = false;
2000 unsigned long flags;
2001 struct dm_bio_prison_cell *cell, *tmp;
2002 struct list_head cells;
2003 struct prealloc structs;
2005 memset(&structs, 0, sizeof(structs));
2007 INIT_LIST_HEAD(&cells);
2009 spin_lock_irqsave(&cache->lock, flags);
2010 list_splice_init(&cache->deferred_cells, &cells);
2011 spin_unlock_irqrestore(&cache->lock, flags);
2013 list_for_each_entry_safe(cell, tmp, &cells, user_list) {
2015 * If we've got no free migration structs, and processing
2016 * this bio might require one, we pause until there are some
2017 * prepared mappings to process.
2019 prealloc_used = true;
2020 if (prealloc_data_structs(cache, &structs)) {
2021 spin_lock_irqsave(&cache->lock, flags);
2022 list_splice(&cells, &cache->deferred_cells);
2023 spin_unlock_irqrestore(&cache->lock, flags);
2027 process_cell(cache, &structs, cell);
2031 prealloc_free_structs(cache, &structs);
2034 static void process_deferred_flush_bios(struct cache *cache, bool submit_bios)
2036 unsigned long flags;
2037 struct bio_list bios;
2040 bio_list_init(&bios);
2042 spin_lock_irqsave(&cache->lock, flags);
2043 bio_list_merge(&bios, &cache->deferred_flush_bios);
2044 bio_list_init(&cache->deferred_flush_bios);
2045 spin_unlock_irqrestore(&cache->lock, flags);
2048 * These bios have already been through inc_ds()
2050 while ((bio = bio_list_pop(&bios)))
2051 submit_bios ? accounted_request(cache, bio) : bio_io_error(bio);
2054 static void process_deferred_writethrough_bios(struct cache *cache)
2056 unsigned long flags;
2057 struct bio_list bios;
2060 bio_list_init(&bios);
2062 spin_lock_irqsave(&cache->lock, flags);
2063 bio_list_merge(&bios, &cache->deferred_writethrough_bios);
2064 bio_list_init(&cache->deferred_writethrough_bios);
2065 spin_unlock_irqrestore(&cache->lock, flags);
2068 * These bios have already been through inc_ds()
2070 while ((bio = bio_list_pop(&bios)))
2071 accounted_request(cache, bio);
2074 static void writeback_some_dirty_blocks(struct cache *cache)
2076 bool prealloc_used = false;
2079 struct prealloc structs;
2080 struct dm_bio_prison_cell *old_ocell;
2081 bool busy = !iot_idle_for(&cache->origin_tracker, HZ);
2083 memset(&structs, 0, sizeof(structs));
2085 while (spare_migration_bandwidth(cache)) {
2086 if (policy_writeback_work(cache->policy, &oblock, &cblock, busy))
2087 break; /* no work to do */
2089 prealloc_used = true;
2090 if (prealloc_data_structs(cache, &structs) ||
2091 get_cell(cache, oblock, &structs, &old_ocell)) {
2092 policy_set_dirty(cache->policy, oblock);
2096 writeback(cache, &structs, oblock, cblock, old_ocell);
2100 prealloc_free_structs(cache, &structs);
2103 /*----------------------------------------------------------------
2105 * Dropping something from the cache *without* writing back.
2106 *--------------------------------------------------------------*/
2108 static void process_invalidation_request(struct cache *cache, struct invalidation_request *req)
2111 uint64_t begin = from_cblock(req->cblocks->begin);
2112 uint64_t end = from_cblock(req->cblocks->end);
2114 while (begin != end) {
2115 r = policy_remove_cblock(cache->policy, to_cblock(begin));
2117 r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin));
2119 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
2123 } else if (r == -ENODATA) {
2124 /* harmless, already unmapped */
2128 DMERR("%s: policy_remove_cblock failed", cache_device_name(cache));
2135 cache->commit_requested = true;
2138 atomic_set(&req->complete, 1);
2140 wake_up(&req->result_wait);
2143 static void process_invalidation_requests(struct cache *cache)
2145 struct list_head list;
2146 struct invalidation_request *req, *tmp;
2148 INIT_LIST_HEAD(&list);
2149 spin_lock(&cache->invalidation_lock);
2150 list_splice_init(&cache->invalidation_requests, &list);
2151 spin_unlock(&cache->invalidation_lock);
2153 list_for_each_entry_safe (req, tmp, &list, list)
2154 process_invalidation_request(cache, req);
2157 /*----------------------------------------------------------------
2159 *--------------------------------------------------------------*/
2160 static bool is_quiescing(struct cache *cache)
2162 return atomic_read(&cache->quiescing);
2165 static void ack_quiescing(struct cache *cache)
2167 if (is_quiescing(cache)) {
2168 atomic_inc(&cache->quiescing_ack);
2169 wake_up(&cache->quiescing_wait);
2173 static void wait_for_quiescing_ack(struct cache *cache)
2175 wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack));
2178 static void start_quiescing(struct cache *cache)
2180 atomic_inc(&cache->quiescing);
2181 wait_for_quiescing_ack(cache);
2184 static void stop_quiescing(struct cache *cache)
2186 atomic_set(&cache->quiescing, 0);
2187 atomic_set(&cache->quiescing_ack, 0);
2190 static void wait_for_migrations(struct cache *cache)
2192 wait_event(cache->migration_wait, !atomic_read(&cache->nr_allocated_migrations));
2195 static void stop_worker(struct cache *cache)
2197 cancel_delayed_work(&cache->waker);
2198 flush_workqueue(cache->wq);
2201 static void requeue_deferred_cells(struct cache *cache)
2203 unsigned long flags;
2204 struct list_head cells;
2205 struct dm_bio_prison_cell *cell, *tmp;
2207 INIT_LIST_HEAD(&cells);
2208 spin_lock_irqsave(&cache->lock, flags);
2209 list_splice_init(&cache->deferred_cells, &cells);
2210 spin_unlock_irqrestore(&cache->lock, flags);
2212 list_for_each_entry_safe(cell, tmp, &cells, user_list)
2213 cell_requeue(cache, cell);
2216 static void requeue_deferred_bios(struct cache *cache)
2219 struct bio_list bios;
2221 bio_list_init(&bios);
2222 bio_list_merge(&bios, &cache->deferred_bios);
2223 bio_list_init(&cache->deferred_bios);
2225 while ((bio = bio_list_pop(&bios)))
2226 bio_endio(bio, DM_ENDIO_REQUEUE);
2229 static int more_work(struct cache *cache)
2231 if (is_quiescing(cache))
2232 return !list_empty(&cache->quiesced_migrations) ||
2233 !list_empty(&cache->completed_migrations) ||
2234 !list_empty(&cache->need_commit_migrations);
2236 return !bio_list_empty(&cache->deferred_bios) ||
2237 !list_empty(&cache->deferred_cells) ||
2238 !bio_list_empty(&cache->deferred_flush_bios) ||
2239 !bio_list_empty(&cache->deferred_writethrough_bios) ||
2240 !list_empty(&cache->quiesced_migrations) ||
2241 !list_empty(&cache->completed_migrations) ||
2242 !list_empty(&cache->need_commit_migrations) ||
2246 static void do_worker(struct work_struct *ws)
2248 struct cache *cache = container_of(ws, struct cache, worker);
2251 if (!is_quiescing(cache)) {
2252 writeback_some_dirty_blocks(cache);
2253 process_deferred_writethrough_bios(cache);
2254 process_deferred_bios(cache);
2255 process_deferred_cells(cache);
2256 process_invalidation_requests(cache);
2259 process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard);
2260 process_migrations(cache, &cache->completed_migrations, complete_migration);
2262 if (commit_if_needed(cache)) {
2263 process_deferred_flush_bios(cache, false);
2264 process_migrations(cache, &cache->need_commit_migrations, migration_failure);
2266 process_deferred_flush_bios(cache, true);
2267 process_migrations(cache, &cache->need_commit_migrations,
2268 migration_success_post_commit);
2271 ack_quiescing(cache);
2273 } while (more_work(cache));
2277 * We want to commit periodically so that not too much
2278 * unwritten metadata builds up.
2280 static void do_waker(struct work_struct *ws)
2282 struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker);
2283 policy_tick(cache->policy, true);
2285 queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD);
2288 /*----------------------------------------------------------------*/
2290 static int is_congested(struct dm_dev *dev, int bdi_bits)
2292 struct request_queue *q = bdev_get_queue(dev->bdev);
2293 return bdi_congested(&q->backing_dev_info, bdi_bits);
2296 static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2298 struct cache *cache = container_of(cb, struct cache, callbacks);
2300 return is_congested(cache->origin_dev, bdi_bits) ||
2301 is_congested(cache->cache_dev, bdi_bits);
2304 /*----------------------------------------------------------------
2306 *--------------------------------------------------------------*/
2309 * This function gets called on the error paths of the constructor, so we
2310 * have to cope with a partially initialised struct.
2312 static void destroy(struct cache *cache)
2316 if (cache->migration_pool)
2317 mempool_destroy(cache->migration_pool);
2319 if (cache->all_io_ds)
2320 dm_deferred_set_destroy(cache->all_io_ds);
2323 dm_bio_prison_destroy(cache->prison);
2326 destroy_workqueue(cache->wq);
2328 if (cache->dirty_bitset)
2329 free_bitset(cache->dirty_bitset);
2331 if (cache->discard_bitset)
2332 free_bitset(cache->discard_bitset);
2335 dm_kcopyd_client_destroy(cache->copier);
2338 dm_cache_metadata_close(cache->cmd);
2340 if (cache->metadata_dev)
2341 dm_put_device(cache->ti, cache->metadata_dev);
2343 if (cache->origin_dev)
2344 dm_put_device(cache->ti, cache->origin_dev);
2346 if (cache->cache_dev)
2347 dm_put_device(cache->ti, cache->cache_dev);
2350 dm_cache_policy_destroy(cache->policy);
2352 for (i = 0; i < cache->nr_ctr_args ; i++)
2353 kfree(cache->ctr_args[i]);
2354 kfree(cache->ctr_args);
2359 static void cache_dtr(struct dm_target *ti)
2361 struct cache *cache = ti->private;
2366 static sector_t get_dev_size(struct dm_dev *dev)
2368 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
2371 /*----------------------------------------------------------------*/
2374 * Construct a cache device mapping.
2376 * cache <metadata dev> <cache dev> <origin dev> <block size>
2377 * <#feature args> [<feature arg>]*
2378 * <policy> <#policy args> [<policy arg>]*
2380 * metadata dev : fast device holding the persistent metadata
2381 * cache dev : fast device holding cached data blocks
2382 * origin dev : slow device holding original data blocks
2383 * block size : cache unit size in sectors
2385 * #feature args : number of feature arguments passed
2386 * feature args : writethrough. (The default is writeback.)
2388 * policy : the replacement policy to use
2389 * #policy args : an even number of policy arguments corresponding
2390 * to key/value pairs passed to the policy
2391 * policy args : key/value pairs passed to the policy
2392 * E.g. 'sequential_threshold 1024'
2393 * See cache-policies.txt for details.
2395 * Optional feature arguments are:
2396 * writethrough : write through caching that prohibits cache block
2397 * content from being different from origin block content.
2398 * Without this argument, the default behaviour is to write
2399 * back cache block contents later for performance reasons,
2400 * so they may differ from the corresponding origin blocks.
2403 struct dm_target *ti;
2405 struct dm_dev *metadata_dev;
2407 struct dm_dev *cache_dev;
2408 sector_t cache_sectors;
2410 struct dm_dev *origin_dev;
2411 sector_t origin_sectors;
2413 uint32_t block_size;
2415 const char *policy_name;
2417 const char **policy_argv;
2419 struct cache_features features;
2422 static void destroy_cache_args(struct cache_args *ca)
2424 if (ca->metadata_dev)
2425 dm_put_device(ca->ti, ca->metadata_dev);
2428 dm_put_device(ca->ti, ca->cache_dev);
2431 dm_put_device(ca->ti, ca->origin_dev);
2436 static bool at_least_one_arg(struct dm_arg_set *as, char **error)
2439 *error = "Insufficient args";
2446 static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as,
2450 sector_t metadata_dev_size;
2451 char b[BDEVNAME_SIZE];
2453 if (!at_least_one_arg(as, error))
2456 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2459 *error = "Error opening metadata device";
2463 metadata_dev_size = get_dev_size(ca->metadata_dev);
2464 if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING)
2465 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2466 bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
2471 static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as,
2476 if (!at_least_one_arg(as, error))
2479 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2482 *error = "Error opening cache device";
2485 ca->cache_sectors = get_dev_size(ca->cache_dev);
2490 static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as,
2495 if (!at_least_one_arg(as, error))
2498 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2501 *error = "Error opening origin device";
2505 ca->origin_sectors = get_dev_size(ca->origin_dev);
2506 if (ca->ti->len > ca->origin_sectors) {
2507 *error = "Device size larger than cached device";
2514 static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as,
2517 unsigned long block_size;
2519 if (!at_least_one_arg(as, error))
2522 if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size ||
2523 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2524 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2525 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2526 *error = "Invalid data block size";
2530 if (block_size > ca->cache_sectors) {
2531 *error = "Data block size is larger than the cache device";
2535 ca->block_size = block_size;
2540 static void init_features(struct cache_features *cf)
2542 cf->mode = CM_WRITE;
2543 cf->io_mode = CM_IO_WRITEBACK;
2546 static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
2549 static struct dm_arg _args[] = {
2550 {0, 1, "Invalid number of cache feature arguments"},
2556 struct cache_features *cf = &ca->features;
2560 r = dm_read_arg_group(_args, as, &argc, error);
2565 arg = dm_shift_arg(as);
2567 if (!strcasecmp(arg, "writeback"))
2568 cf->io_mode = CM_IO_WRITEBACK;
2570 else if (!strcasecmp(arg, "writethrough"))
2571 cf->io_mode = CM_IO_WRITETHROUGH;
2573 else if (!strcasecmp(arg, "passthrough"))
2574 cf->io_mode = CM_IO_PASSTHROUGH;
2577 *error = "Unrecognised cache feature requested";
2585 static int parse_policy(struct cache_args *ca, struct dm_arg_set *as,
2588 static struct dm_arg _args[] = {
2589 {0, 1024, "Invalid number of policy arguments"},
2594 if (!at_least_one_arg(as, error))
2597 ca->policy_name = dm_shift_arg(as);
2599 r = dm_read_arg_group(_args, as, &ca->policy_argc, error);
2603 ca->policy_argv = (const char **)as->argv;
2604 dm_consume_args(as, ca->policy_argc);
2609 static int parse_cache_args(struct cache_args *ca, int argc, char **argv,
2613 struct dm_arg_set as;
2618 r = parse_metadata_dev(ca, &as, error);
2622 r = parse_cache_dev(ca, &as, error);
2626 r = parse_origin_dev(ca, &as, error);
2630 r = parse_block_size(ca, &as, error);
2634 r = parse_features(ca, &as, error);
2638 r = parse_policy(ca, &as, error);
2645 /*----------------------------------------------------------------*/
2647 static struct kmem_cache *migration_cache;
2649 #define NOT_CORE_OPTION 1
2651 static int process_config_option(struct cache *cache, const char *key, const char *value)
2655 if (!strcasecmp(key, "migration_threshold")) {
2656 if (kstrtoul(value, 10, &tmp))
2659 cache->migration_threshold = tmp;
2663 return NOT_CORE_OPTION;
2666 static int set_config_value(struct cache *cache, const char *key, const char *value)
2668 int r = process_config_option(cache, key, value);
2670 if (r == NOT_CORE_OPTION)
2671 r = policy_set_config_value(cache->policy, key, value);
2674 DMWARN("bad config value for %s: %s", key, value);
2679 static int set_config_values(struct cache *cache, int argc, const char **argv)
2684 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
2689 r = set_config_value(cache, argv[0], argv[1]);
2700 static int create_cache_policy(struct cache *cache, struct cache_args *ca,
2703 struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name,
2705 cache->origin_sectors,
2706 cache->sectors_per_block);
2708 *error = "Error creating cache's policy";
2717 * We want the discard block size to be at least the size of the cache
2718 * block size and have no more than 2^14 discard blocks across the origin.
2720 #define MAX_DISCARD_BLOCKS (1 << 14)
2722 static bool too_many_discard_blocks(sector_t discard_block_size,
2723 sector_t origin_size)
2725 (void) sector_div(origin_size, discard_block_size);
2727 return origin_size > MAX_DISCARD_BLOCKS;
2730 static sector_t calculate_discard_block_size(sector_t cache_block_size,
2731 sector_t origin_size)
2733 sector_t discard_block_size = cache_block_size;
2736 while (too_many_discard_blocks(discard_block_size, origin_size))
2737 discard_block_size *= 2;
2739 return discard_block_size;
2742 static void set_cache_size(struct cache *cache, dm_cblock_t size)
2744 dm_block_t nr_blocks = from_cblock(size);
2746 if (nr_blocks > (1 << 20) && cache->cache_size != size)
2747 DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n"
2748 "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n"
2749 "Please consider increasing the cache block size to reduce the overall cache block count.",
2750 (unsigned long long) nr_blocks);
2752 cache->cache_size = size;
2755 #define DEFAULT_MIGRATION_THRESHOLD 2048
2757 static int cache_create(struct cache_args *ca, struct cache **result)
2760 char **error = &ca->ti->error;
2761 struct cache *cache;
2762 struct dm_target *ti = ca->ti;
2763 dm_block_t origin_blocks;
2764 struct dm_cache_metadata *cmd;
2765 bool may_format = ca->features.mode == CM_WRITE;
2767 cache = kzalloc(sizeof(*cache), GFP_KERNEL);
2772 ti->private = cache;
2773 ti->num_flush_bios = 2;
2774 ti->flush_supported = true;
2776 ti->num_discard_bios = 1;
2777 ti->discards_supported = true;
2778 ti->discard_zeroes_data_unsupported = true;
2779 ti->split_discard_bios = false;
2781 cache->features = ca->features;
2782 ti->per_bio_data_size = get_per_bio_data_size(cache);
2784 cache->callbacks.congested_fn = cache_is_congested;
2785 dm_table_add_target_callbacks(ti->table, &cache->callbacks);
2787 cache->metadata_dev = ca->metadata_dev;
2788 cache->origin_dev = ca->origin_dev;
2789 cache->cache_dev = ca->cache_dev;
2791 ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL;
2793 /* FIXME: factor out this whole section */
2794 origin_blocks = cache->origin_sectors = ca->origin_sectors;
2795 origin_blocks = block_div(origin_blocks, ca->block_size);
2796 cache->origin_blocks = to_oblock(origin_blocks);
2798 cache->sectors_per_block = ca->block_size;
2799 if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) {
2804 if (ca->block_size & (ca->block_size - 1)) {
2805 dm_block_t cache_size = ca->cache_sectors;
2807 cache->sectors_per_block_shift = -1;
2808 cache_size = block_div(cache_size, ca->block_size);
2809 set_cache_size(cache, to_cblock(cache_size));
2811 cache->sectors_per_block_shift = __ffs(ca->block_size);
2812 set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift));
2815 r = create_cache_policy(cache, ca, error);
2819 cache->policy_nr_args = ca->policy_argc;
2820 cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD;
2822 r = set_config_values(cache, ca->policy_argc, ca->policy_argv);
2824 *error = "Error setting cache policy's config values";
2828 cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
2829 ca->block_size, may_format,
2830 dm_cache_policy_get_hint_size(cache->policy));
2832 *error = "Error creating metadata object";
2837 set_cache_mode(cache, CM_WRITE);
2838 if (get_cache_mode(cache) != CM_WRITE) {
2839 *error = "Unable to get write access to metadata, please check/repair metadata.";
2844 if (passthrough_mode(&cache->features)) {
2847 r = dm_cache_metadata_all_clean(cache->cmd, &all_clean);
2849 *error = "dm_cache_metadata_all_clean() failed";
2854 *error = "Cannot enter passthrough mode unless all blocks are clean";
2860 spin_lock_init(&cache->lock);
2861 INIT_LIST_HEAD(&cache->deferred_cells);
2862 bio_list_init(&cache->deferred_bios);
2863 bio_list_init(&cache->deferred_flush_bios);
2864 bio_list_init(&cache->deferred_writethrough_bios);
2865 INIT_LIST_HEAD(&cache->quiesced_migrations);
2866 INIT_LIST_HEAD(&cache->completed_migrations);
2867 INIT_LIST_HEAD(&cache->need_commit_migrations);
2868 atomic_set(&cache->nr_allocated_migrations, 0);
2869 atomic_set(&cache->nr_io_migrations, 0);
2870 init_waitqueue_head(&cache->migration_wait);
2872 init_waitqueue_head(&cache->quiescing_wait);
2873 atomic_set(&cache->quiescing, 0);
2874 atomic_set(&cache->quiescing_ack, 0);
2877 atomic_set(&cache->nr_dirty, 0);
2878 cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size));
2879 if (!cache->dirty_bitset) {
2880 *error = "could not allocate dirty bitset";
2883 clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size));
2885 cache->discard_block_size =
2886 calculate_discard_block_size(cache->sectors_per_block,
2887 cache->origin_sectors);
2888 cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors,
2889 cache->discard_block_size));
2890 cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks));
2891 if (!cache->discard_bitset) {
2892 *error = "could not allocate discard bitset";
2895 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
2897 cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2898 if (IS_ERR(cache->copier)) {
2899 *error = "could not create kcopyd client";
2900 r = PTR_ERR(cache->copier);
2904 cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2906 *error = "could not create workqueue for metadata object";
2909 INIT_WORK(&cache->worker, do_worker);
2910 INIT_DELAYED_WORK(&cache->waker, do_waker);
2911 cache->last_commit_jiffies = jiffies;
2913 cache->prison = dm_bio_prison_create();
2914 if (!cache->prison) {
2915 *error = "could not create bio prison";
2919 cache->all_io_ds = dm_deferred_set_create();
2920 if (!cache->all_io_ds) {
2921 *error = "could not create all_io deferred set";
2925 cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE,
2927 if (!cache->migration_pool) {
2928 *error = "Error creating cache's migration mempool";
2932 cache->need_tick_bio = true;
2933 cache->sized = false;
2934 cache->invalidate = false;
2935 cache->commit_requested = false;
2936 cache->loaded_mappings = false;
2937 cache->loaded_discards = false;
2941 atomic_set(&cache->stats.demotion, 0);
2942 atomic_set(&cache->stats.promotion, 0);
2943 atomic_set(&cache->stats.copies_avoided, 0);
2944 atomic_set(&cache->stats.cache_cell_clash, 0);
2945 atomic_set(&cache->stats.commit_count, 0);
2946 atomic_set(&cache->stats.discard_count, 0);
2948 spin_lock_init(&cache->invalidation_lock);
2949 INIT_LIST_HEAD(&cache->invalidation_requests);
2951 iot_init(&cache->origin_tracker);
2961 static int copy_ctr_args(struct cache *cache, int argc, const char **argv)
2966 copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL);
2969 for (i = 0; i < argc; i++) {
2970 copy[i] = kstrdup(argv[i], GFP_KERNEL);
2979 cache->nr_ctr_args = argc;
2980 cache->ctr_args = copy;
2985 static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv)
2988 struct cache_args *ca;
2989 struct cache *cache = NULL;
2991 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2993 ti->error = "Error allocating memory for cache";
2998 r = parse_cache_args(ca, argc, argv, &ti->error);
3002 r = cache_create(ca, &cache);
3006 r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3);
3012 ti->private = cache;
3015 destroy_cache_args(ca);
3019 /*----------------------------------------------------------------*/
3021 static int cache_map(struct dm_target *ti, struct bio *bio)
3023 struct cache *cache = ti->private;
3026 struct dm_bio_prison_cell *cell = NULL;
3027 dm_oblock_t block = get_bio_block(cache, bio);
3028 size_t pb_data_size = get_per_bio_data_size(cache);
3029 bool can_migrate = false;
3030 bool fast_promotion;
3031 struct policy_result lookup_result;
3032 struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size);
3033 struct old_oblock_lock ool;
3035 ool.locker.fn = null_locker;
3037 if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) {
3039 * This can only occur if the io goes to a partial block at
3040 * the end of the origin device. We don't cache these.
3041 * Just remap to the origin and carry on.
3043 remap_to_origin(cache, bio);
3044 accounted_begin(cache, bio);
3045 return DM_MAPIO_REMAPPED;
3048 if (discard_or_flush(bio)) {
3049 defer_bio(cache, bio);
3050 return DM_MAPIO_SUBMITTED;
3054 * Check to see if that block is currently migrating.
3056 cell = alloc_prison_cell(cache);
3058 defer_bio(cache, bio);
3059 return DM_MAPIO_SUBMITTED;
3062 r = bio_detain(cache, block, bio, cell,
3063 (cell_free_fn) free_prison_cell,
3067 defer_bio(cache, bio);
3069 return DM_MAPIO_SUBMITTED;
3072 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
3074 r = policy_map(cache->policy, block, false, can_migrate, fast_promotion,
3075 bio, &ool.locker, &lookup_result);
3076 if (r == -EWOULDBLOCK) {
3077 cell_defer(cache, cell, true);
3078 return DM_MAPIO_SUBMITTED;
3081 DMERR_LIMIT("%s: Unexpected return from cache replacement policy: %d",
3082 cache_device_name(cache), r);
3083 cell_defer(cache, cell, false);
3085 return DM_MAPIO_SUBMITTED;
3088 r = DM_MAPIO_REMAPPED;
3089 switch (lookup_result.op) {
3091 if (passthrough_mode(&cache->features)) {
3092 if (bio_data_dir(bio) == WRITE) {
3094 * We need to invalidate this block, so
3095 * defer for the worker thread.
3097 cell_defer(cache, cell, true);
3098 r = DM_MAPIO_SUBMITTED;
3101 inc_miss_counter(cache, bio);
3102 remap_to_origin_clear_discard(cache, bio, block);
3103 accounted_begin(cache, bio);
3104 inc_ds(cache, bio, cell);
3105 // FIXME: we want to remap hits or misses straight
3106 // away rather than passing over to the worker.
3107 cell_defer(cache, cell, false);
3111 inc_hit_counter(cache, bio);
3112 if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) &&
3113 !is_dirty(cache, lookup_result.cblock)) {
3114 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
3115 accounted_begin(cache, bio);
3116 inc_ds(cache, bio, cell);
3117 cell_defer(cache, cell, false);
3120 remap_cell_to_cache_dirty(cache, cell, block, lookup_result.cblock, false);
3125 inc_miss_counter(cache, bio);
3126 if (pb->req_nr != 0) {
3128 * This is a duplicate writethrough io that is no
3129 * longer needed because the block has been demoted.
3132 // FIXME: remap everything as a miss
3133 cell_defer(cache, cell, false);
3134 r = DM_MAPIO_SUBMITTED;
3137 remap_cell_to_origin_clear_discard(cache, cell, block, false);
3141 DMERR_LIMIT("%s: %s: erroring bio: unknown policy op: %u",
3142 cache_device_name(cache), __func__,
3143 (unsigned) lookup_result.op);
3144 cell_defer(cache, cell, false);
3146 r = DM_MAPIO_SUBMITTED;
3152 static int cache_end_io(struct dm_target *ti, struct bio *bio, int error)
3154 struct cache *cache = ti->private;
3155 unsigned long flags;
3156 size_t pb_data_size = get_per_bio_data_size(cache);
3157 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
3160 policy_tick(cache->policy, false);
3162 spin_lock_irqsave(&cache->lock, flags);
3163 cache->need_tick_bio = true;
3164 spin_unlock_irqrestore(&cache->lock, flags);
3167 check_for_quiesced_migrations(cache, pb);
3168 accounted_complete(cache, bio);
3173 static int write_dirty_bitset(struct cache *cache)
3177 if (get_cache_mode(cache) >= CM_READ_ONLY)
3180 for (i = 0; i < from_cblock(cache->cache_size); i++) {
3181 r = dm_cache_set_dirty(cache->cmd, to_cblock(i),
3182 is_dirty(cache, to_cblock(i)));
3184 metadata_operation_failed(cache, "dm_cache_set_dirty", r);
3192 static int write_discard_bitset(struct cache *cache)
3196 if (get_cache_mode(cache) >= CM_READ_ONLY)
3199 r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size,
3200 cache->discard_nr_blocks);
3202 DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache));
3203 metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r);
3207 for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) {
3208 r = dm_cache_set_discard(cache->cmd, to_dblock(i),
3209 is_discarded(cache, to_dblock(i)));
3211 metadata_operation_failed(cache, "dm_cache_set_discard", r);
3219 static int write_hints(struct cache *cache)
3223 if (get_cache_mode(cache) >= CM_READ_ONLY)
3226 r = dm_cache_write_hints(cache->cmd, cache->policy);
3228 metadata_operation_failed(cache, "dm_cache_write_hints", r);
3236 * returns true on success
3238 static bool sync_metadata(struct cache *cache)
3242 r1 = write_dirty_bitset(cache);
3244 DMERR("%s: could not write dirty bitset", cache_device_name(cache));
3246 r2 = write_discard_bitset(cache);
3248 DMERR("%s: could not write discard bitset", cache_device_name(cache));
3252 r3 = write_hints(cache);
3254 DMERR("%s: could not write hints", cache_device_name(cache));
3257 * If writing the above metadata failed, we still commit, but don't
3258 * set the clean shutdown flag. This will effectively force every
3259 * dirty bit to be set on reload.
3261 r4 = commit(cache, !r1 && !r2 && !r3);
3263 DMERR("%s: could not write cache metadata", cache_device_name(cache));
3265 return !r1 && !r2 && !r3 && !r4;
3268 static void cache_postsuspend(struct dm_target *ti)
3270 struct cache *cache = ti->private;
3272 start_quiescing(cache);
3273 wait_for_migrations(cache);
3275 requeue_deferred_bios(cache);
3276 requeue_deferred_cells(cache);
3277 stop_quiescing(cache);
3279 if (get_cache_mode(cache) == CM_WRITE)
3280 (void) sync_metadata(cache);
3283 static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock,
3284 bool dirty, uint32_t hint, bool hint_valid)
3287 struct cache *cache = context;
3289 r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid);
3294 set_dirty(cache, oblock, cblock);
3296 clear_dirty(cache, oblock, cblock);
3302 * The discard block size in the on disk metadata is not
3303 * neccessarily the same as we're currently using. So we have to
3304 * be careful to only set the discarded attribute if we know it
3305 * covers a complete block of the new size.
3307 struct discard_load_info {
3308 struct cache *cache;
3311 * These blocks are sized using the on disk dblock size, rather
3312 * than the current one.
3314 dm_block_t block_size;
3315 dm_block_t discard_begin, discard_end;
3318 static void discard_load_info_init(struct cache *cache,
3319 struct discard_load_info *li)
3322 li->discard_begin = li->discard_end = 0;
3325 static void set_discard_range(struct discard_load_info *li)
3329 if (li->discard_begin == li->discard_end)
3333 * Convert to sectors.
3335 b = li->discard_begin * li->block_size;
3336 e = li->discard_end * li->block_size;
3339 * Then convert back to the current dblock size.
3341 b = dm_sector_div_up(b, li->cache->discard_block_size);
3342 sector_div(e, li->cache->discard_block_size);
3345 * The origin may have shrunk, so we need to check we're still in
3348 if (e > from_dblock(li->cache->discard_nr_blocks))
3349 e = from_dblock(li->cache->discard_nr_blocks);
3352 set_discard(li->cache, to_dblock(b));
3355 static int load_discard(void *context, sector_t discard_block_size,
3356 dm_dblock_t dblock, bool discard)
3358 struct discard_load_info *li = context;
3360 li->block_size = discard_block_size;
3363 if (from_dblock(dblock) == li->discard_end)
3365 * We're already in a discard range, just extend it.
3367 li->discard_end = li->discard_end + 1ULL;
3371 * Emit the old range and start a new one.
3373 set_discard_range(li);
3374 li->discard_begin = from_dblock(dblock);
3375 li->discard_end = li->discard_begin + 1ULL;
3378 set_discard_range(li);
3379 li->discard_begin = li->discard_end = 0;
3385 static dm_cblock_t get_cache_dev_size(struct cache *cache)
3387 sector_t size = get_dev_size(cache->cache_dev);
3388 (void) sector_div(size, cache->sectors_per_block);
3389 return to_cblock(size);
3392 static bool can_resize(struct cache *cache, dm_cblock_t new_size)
3394 if (from_cblock(new_size) > from_cblock(cache->cache_size))
3398 * We can't drop a dirty block when shrinking the cache.
3400 while (from_cblock(new_size) < from_cblock(cache->cache_size)) {
3401 new_size = to_cblock(from_cblock(new_size) + 1);
3402 if (is_dirty(cache, new_size)) {
3403 DMERR("%s: unable to shrink cache; cache block %llu is dirty",
3404 cache_device_name(cache),
3405 (unsigned long long) from_cblock(new_size));
3413 static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size)
3417 r = dm_cache_resize(cache->cmd, new_size);
3419 DMERR("%s: could not resize cache metadata", cache_device_name(cache));
3420 metadata_operation_failed(cache, "dm_cache_resize", r);
3424 set_cache_size(cache, new_size);
3429 static int cache_preresume(struct dm_target *ti)
3432 struct cache *cache = ti->private;
3433 dm_cblock_t csize = get_cache_dev_size(cache);
3436 * Check to see if the cache has resized.
3438 if (!cache->sized) {
3439 r = resize_cache_dev(cache, csize);
3443 cache->sized = true;
3445 } else if (csize != cache->cache_size) {
3446 if (!can_resize(cache, csize))
3449 r = resize_cache_dev(cache, csize);
3454 if (!cache->loaded_mappings) {
3455 r = dm_cache_load_mappings(cache->cmd, cache->policy,
3456 load_mapping, cache);
3458 DMERR("%s: could not load cache mappings", cache_device_name(cache));
3459 metadata_operation_failed(cache, "dm_cache_load_mappings", r);
3463 cache->loaded_mappings = true;
3466 if (!cache->loaded_discards) {
3467 struct discard_load_info li;
3470 * The discard bitset could have been resized, or the
3471 * discard block size changed. To be safe we start by
3472 * setting every dblock to not discarded.
3474 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
3476 discard_load_info_init(cache, &li);
3477 r = dm_cache_load_discards(cache->cmd, load_discard, &li);
3479 DMERR("%s: could not load origin discards", cache_device_name(cache));
3480 metadata_operation_failed(cache, "dm_cache_load_discards", r);
3483 set_discard_range(&li);
3485 cache->loaded_discards = true;
3491 static void cache_resume(struct dm_target *ti)
3493 struct cache *cache = ti->private;
3495 cache->need_tick_bio = true;
3496 do_waker(&cache->waker.work);
3502 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
3503 * <cache block size> <#used cache blocks>/<#total cache blocks>
3504 * <#read hits> <#read misses> <#write hits> <#write misses>
3505 * <#demotions> <#promotions> <#dirty>
3506 * <#features> <features>*
3507 * <#core args> <core args>
3508 * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check>
3510 static void cache_status(struct dm_target *ti, status_type_t type,
3511 unsigned status_flags, char *result, unsigned maxlen)
3516 dm_block_t nr_free_blocks_metadata = 0;
3517 dm_block_t nr_blocks_metadata = 0;
3518 char buf[BDEVNAME_SIZE];
3519 struct cache *cache = ti->private;
3520 dm_cblock_t residency;
3523 case STATUSTYPE_INFO:
3524 if (get_cache_mode(cache) == CM_FAIL) {
3529 /* Commit to ensure statistics aren't out-of-date */
3530 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3531 (void) commit(cache, false);
3533 r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata);
3535 DMERR("%s: dm_cache_get_free_metadata_block_count returned %d",
3536 cache_device_name(cache), r);
3540 r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata);
3542 DMERR("%s: dm_cache_get_metadata_dev_size returned %d",
3543 cache_device_name(cache), r);
3547 residency = policy_residency(cache->policy);
3549 DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ",
3550 (unsigned)DM_CACHE_METADATA_BLOCK_SIZE,
3551 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3552 (unsigned long long)nr_blocks_metadata,
3553 cache->sectors_per_block,
3554 (unsigned long long) from_cblock(residency),
3555 (unsigned long long) from_cblock(cache->cache_size),
3556 (unsigned) atomic_read(&cache->stats.read_hit),
3557 (unsigned) atomic_read(&cache->stats.read_miss),
3558 (unsigned) atomic_read(&cache->stats.write_hit),
3559 (unsigned) atomic_read(&cache->stats.write_miss),
3560 (unsigned) atomic_read(&cache->stats.demotion),
3561 (unsigned) atomic_read(&cache->stats.promotion),
3562 (unsigned long) atomic_read(&cache->nr_dirty));
3564 if (writethrough_mode(&cache->features))
3565 DMEMIT("1 writethrough ");
3567 else if (passthrough_mode(&cache->features))
3568 DMEMIT("1 passthrough ");
3570 else if (writeback_mode(&cache->features))
3571 DMEMIT("1 writeback ");
3574 DMERR("%s: internal error: unknown io mode: %d",
3575 cache_device_name(cache), (int) cache->features.io_mode);
3579 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
3581 DMEMIT("%s ", dm_cache_policy_get_name(cache->policy));
3583 r = policy_emit_config_values(cache->policy, result, maxlen, &sz);
3585 DMERR("%s: policy_emit_config_values returned %d",
3586 cache_device_name(cache), r);
3589 if (get_cache_mode(cache) == CM_READ_ONLY)
3594 if (dm_cache_metadata_needs_check(cache->cmd))
3595 DMEMIT("needs_check ");
3601 case STATUSTYPE_TABLE:
3602 format_dev_t(buf, cache->metadata_dev->bdev->bd_dev);
3604 format_dev_t(buf, cache->cache_dev->bdev->bd_dev);
3606 format_dev_t(buf, cache->origin_dev->bdev->bd_dev);
3609 for (i = 0; i < cache->nr_ctr_args - 1; i++)
3610 DMEMIT(" %s", cache->ctr_args[i]);
3611 if (cache->nr_ctr_args)
3612 DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]);
3622 * A cache block range can take two forms:
3624 * i) A single cblock, eg. '3456'
3625 * ii) A begin and end cblock with dots between, eg. 123-234
3627 static int parse_cblock_range(struct cache *cache, const char *str,
3628 struct cblock_range *result)
3635 * Try and parse form (ii) first.
3637 r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy);
3642 result->begin = to_cblock(b);
3643 result->end = to_cblock(e);
3648 * That didn't work, try form (i).
3650 r = sscanf(str, "%llu%c", &b, &dummy);
3655 result->begin = to_cblock(b);
3656 result->end = to_cblock(from_cblock(result->begin) + 1u);
3660 DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str);
3664 static int validate_cblock_range(struct cache *cache, struct cblock_range *range)
3666 uint64_t b = from_cblock(range->begin);
3667 uint64_t e = from_cblock(range->end);
3668 uint64_t n = from_cblock(cache->cache_size);
3671 DMERR("%s: begin cblock out of range: %llu >= %llu",
3672 cache_device_name(cache), b, n);
3677 DMERR("%s: end cblock out of range: %llu > %llu",
3678 cache_device_name(cache), e, n);
3683 DMERR("%s: invalid cblock range: %llu >= %llu",
3684 cache_device_name(cache), b, e);
3691 static int request_invalidation(struct cache *cache, struct cblock_range *range)
3693 struct invalidation_request req;
3695 INIT_LIST_HEAD(&req.list);
3696 req.cblocks = range;
3697 atomic_set(&req.complete, 0);
3699 init_waitqueue_head(&req.result_wait);
3701 spin_lock(&cache->invalidation_lock);
3702 list_add(&req.list, &cache->invalidation_requests);
3703 spin_unlock(&cache->invalidation_lock);
3706 wait_event(req.result_wait, atomic_read(&req.complete));
3710 static int process_invalidate_cblocks_message(struct cache *cache, unsigned count,
3711 const char **cblock_ranges)
3715 struct cblock_range range;
3717 if (!passthrough_mode(&cache->features)) {
3718 DMERR("%s: cache has to be in passthrough mode for invalidation",
3719 cache_device_name(cache));
3723 for (i = 0; i < count; i++) {
3724 r = parse_cblock_range(cache, cblock_ranges[i], &range);
3728 r = validate_cblock_range(cache, &range);
3733 * Pass begin and end origin blocks to the worker and wake it.
3735 r = request_invalidation(cache, &range);
3747 * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]*
3749 * The key migration_threshold is supported by the cache target core.
3751 static int cache_message(struct dm_target *ti, unsigned argc, char **argv)
3753 struct cache *cache = ti->private;
3758 if (get_cache_mode(cache) >= CM_READ_ONLY) {
3759 DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode",
3760 cache_device_name(cache));
3764 if (!strcasecmp(argv[0], "invalidate_cblocks"))
3765 return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1);
3770 return set_config_value(cache, argv[0], argv[1]);
3773 static int cache_iterate_devices(struct dm_target *ti,
3774 iterate_devices_callout_fn fn, void *data)
3777 struct cache *cache = ti->private;
3779 r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data);
3781 r = fn(ti, cache->origin_dev, 0, ti->len, data);
3787 * We assume I/O is going to the origin (which is the volume
3788 * more likely to have restrictions e.g. by being striped).
3789 * (Looking up the exact location of the data would be expensive
3790 * and could always be out of date by the time the bio is submitted.)
3792 static int cache_bvec_merge(struct dm_target *ti,
3793 struct bvec_merge_data *bvm,
3794 struct bio_vec *biovec, int max_size)
3796 struct cache *cache = ti->private;
3797 struct request_queue *q = bdev_get_queue(cache->origin_dev->bdev);
3799 if (!q->merge_bvec_fn)
3802 bvm->bi_bdev = cache->origin_dev->bdev;
3803 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
3806 static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
3809 * FIXME: these limits may be incompatible with the cache device
3811 limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024,
3812 cache->origin_sectors);
3813 limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
3816 static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
3818 struct cache *cache = ti->private;
3819 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
3822 * If the system-determined stacked limits are compatible with the
3823 * cache's blocksize (io_opt is a factor) do not override them.
3825 if (io_opt_sectors < cache->sectors_per_block ||
3826 do_div(io_opt_sectors, cache->sectors_per_block)) {
3827 blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT);
3828 blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
3830 set_discard_limits(cache, limits);
3833 /*----------------------------------------------------------------*/
3835 static struct target_type cache_target = {
3837 .version = {1, 8, 0},
3838 .module = THIS_MODULE,
3842 .end_io = cache_end_io,
3843 .postsuspend = cache_postsuspend,
3844 .preresume = cache_preresume,
3845 .resume = cache_resume,
3846 .status = cache_status,
3847 .message = cache_message,
3848 .iterate_devices = cache_iterate_devices,
3849 .merge = cache_bvec_merge,
3850 .io_hints = cache_io_hints,
3853 static int __init dm_cache_init(void)
3857 r = dm_register_target(&cache_target);
3859 DMERR("cache target registration failed: %d", r);
3863 migration_cache = KMEM_CACHE(dm_cache_migration, 0);
3864 if (!migration_cache) {
3865 dm_unregister_target(&cache_target);
3872 static void __exit dm_cache_exit(void)
3874 dm_unregister_target(&cache_target);
3875 kmem_cache_destroy(migration_cache);
3878 module_init(dm_cache_init);
3879 module_exit(dm_cache_exit);
3881 MODULE_DESCRIPTION(DM_NAME " cache target");
3882 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
3883 MODULE_LICENSE("GPL");