2 * Copyright (C) 2012 Red Hat. All rights reserved.
4 * This file is released under the GPL.
8 #include "dm-bio-prison.h"
9 #include "dm-bio-record.h"
10 #include "dm-cache-metadata.h"
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/jiffies.h>
15 #include <linux/init.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
21 #define DM_MSG_PREFIX "cache"
23 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle,
24 "A percentage of time allocated for copying to and/or from cache");
26 /*----------------------------------------------------------------*/
28 #define IOT_RESOLUTION 4
34 * Sectors of in-flight IO.
39 * The time, in jiffies, when this device became idle (if it is
42 unsigned long idle_time;
43 unsigned long last_update_time;
46 static void iot_init(struct io_tracker *iot)
48 spin_lock_init(&iot->lock);
51 iot->last_update_time = jiffies;
54 static bool __iot_idle_for(struct io_tracker *iot, unsigned long jifs)
59 return time_after(jiffies, iot->idle_time + jifs);
62 static bool iot_idle_for(struct io_tracker *iot, unsigned long jifs)
67 spin_lock_irqsave(&iot->lock, flags);
68 r = __iot_idle_for(iot, jifs);
69 spin_unlock_irqrestore(&iot->lock, flags);
74 static void iot_io_begin(struct io_tracker *iot, sector_t len)
78 spin_lock_irqsave(&iot->lock, flags);
79 iot->in_flight += len;
80 spin_unlock_irqrestore(&iot->lock, flags);
83 static void __iot_io_end(struct io_tracker *iot, sector_t len)
85 iot->in_flight -= len;
87 iot->idle_time = jiffies;
90 static void iot_io_end(struct io_tracker *iot, sector_t len)
94 spin_lock_irqsave(&iot->lock, flags);
95 __iot_io_end(iot, len);
96 spin_unlock_irqrestore(&iot->lock, flags);
99 /*----------------------------------------------------------------*/
104 * oblock: index of an origin block
105 * cblock: index of a cache block
106 * promotion: movement of a block from origin to cache
107 * demotion: movement of a block from cache to origin
108 * migration: movement of a block between the origin and cache device,
112 /*----------------------------------------------------------------*/
115 * There are a couple of places where we let a bio run, but want to do some
116 * work before calling its endio function. We do this by temporarily
117 * changing the endio fn.
119 struct dm_hook_info {
120 bio_end_io_t *bi_end_io;
123 static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio,
124 bio_end_io_t *bi_end_io, void *bi_private)
126 h->bi_end_io = bio->bi_end_io;
128 bio->bi_end_io = bi_end_io;
129 bio->bi_private = bi_private;
132 static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio)
134 bio->bi_end_io = h->bi_end_io;
137 /*----------------------------------------------------------------*/
139 #define MIGRATION_POOL_SIZE 128
140 #define COMMIT_PERIOD HZ
141 #define MIGRATION_COUNT_WINDOW 10
144 * The block size of the device holding cache data must be
145 * between 32KB and 1GB.
147 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT)
148 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
150 enum cache_metadata_mode {
151 CM_WRITE, /* metadata may be changed */
152 CM_READ_ONLY, /* metadata may not be changed */
158 * Data is written to cached blocks only. These blocks are marked
159 * dirty. If you lose the cache device you will lose data.
160 * Potential performance increase for both reads and writes.
165 * Data is written to both cache and origin. Blocks are never
166 * dirty. Potential performance benfit for reads only.
171 * A degraded mode useful for various cache coherency situations
172 * (eg, rolling back snapshots). Reads and writes always go to the
173 * origin. If a write goes to a cached oblock, then the cache
174 * block is invalidated.
179 struct cache_features {
180 enum cache_metadata_mode mode;
181 enum cache_io_mode io_mode;
182 unsigned metadata_version;
192 atomic_t copies_avoided;
193 atomic_t cache_cell_clash;
194 atomic_t commit_count;
195 atomic_t discard_count;
199 * Defines a range of cblocks, begin to (end - 1) are in the range. end is
200 * the one-past-the-end value.
202 struct cblock_range {
207 struct invalidation_request {
208 struct list_head list;
209 struct cblock_range *cblocks;
214 wait_queue_head_t result_wait;
218 struct dm_target *ti;
219 struct dm_target_callbacks callbacks;
221 struct dm_cache_metadata *cmd;
224 * Metadata is written to this device.
226 struct dm_dev *metadata_dev;
229 * The slower of the two data devices. Typically a spindle.
231 struct dm_dev *origin_dev;
234 * The faster of the two data devices. Typically an SSD.
236 struct dm_dev *cache_dev;
239 * Size of the origin device in _complete_ blocks and native sectors.
241 dm_oblock_t origin_blocks;
242 sector_t origin_sectors;
245 * Size of the cache device in blocks.
247 dm_cblock_t cache_size;
250 * Fields for converting from sectors to blocks.
252 sector_t sectors_per_block;
253 int sectors_per_block_shift;
256 struct list_head deferred_cells;
257 struct bio_list deferred_bios;
258 struct bio_list deferred_flush_bios;
259 struct bio_list deferred_writethrough_bios;
260 struct list_head quiesced_migrations;
261 struct list_head completed_migrations;
262 struct list_head need_commit_migrations;
263 sector_t migration_threshold;
264 wait_queue_head_t migration_wait;
265 atomic_t nr_allocated_migrations;
268 * The number of in flight migrations that are performing
269 * background io. eg, promotion, writeback.
271 atomic_t nr_io_migrations;
273 wait_queue_head_t quiescing_wait;
275 atomic_t quiescing_ack;
278 * cache_size entries, dirty if set
281 unsigned long *dirty_bitset;
284 * origin_blocks entries, discarded if set.
286 dm_dblock_t discard_nr_blocks;
287 unsigned long *discard_bitset;
288 uint32_t discard_block_size; /* a power of 2 times sectors per block */
291 * Rather than reconstructing the table line for the status we just
292 * save it and regurgitate.
294 unsigned nr_ctr_args;
295 const char **ctr_args;
297 struct dm_kcopyd_client *copier;
298 struct workqueue_struct *wq;
299 struct work_struct worker;
301 struct delayed_work waker;
302 unsigned long last_commit_jiffies;
304 struct dm_bio_prison *prison;
305 struct dm_deferred_set *all_io_ds;
307 mempool_t *migration_pool;
309 struct dm_cache_policy *policy;
310 unsigned policy_nr_args;
312 bool need_tick_bio:1;
315 bool commit_requested:1;
316 bool loaded_mappings:1;
317 bool loaded_discards:1;
320 * Cache features such as write-through.
322 struct cache_features features;
324 struct cache_stats stats;
327 * Invalidation fields.
329 spinlock_t invalidation_lock;
330 struct list_head invalidation_requests;
332 struct io_tracker origin_tracker;
335 struct per_bio_data {
338 struct dm_deferred_entry *all_io_entry;
339 struct dm_hook_info hook_info;
343 * writethrough fields. These MUST remain at the end of this
344 * structure and the 'cache' member must be the first as it
345 * is used to determine the offset of the writethrough fields.
349 struct dm_bio_details bio_details;
352 struct dm_cache_migration {
353 struct list_head list;
356 unsigned long start_jiffies;
357 dm_oblock_t old_oblock;
358 dm_oblock_t new_oblock;
366 bool requeue_holder:1;
369 struct dm_bio_prison_cell *old_ocell;
370 struct dm_bio_prison_cell *new_ocell;
374 * Processing a bio in the worker thread may require these memory
375 * allocations. We prealloc to avoid deadlocks (the same worker thread
376 * frees them back to the mempool).
379 struct dm_cache_migration *mg;
380 struct dm_bio_prison_cell *cell1;
381 struct dm_bio_prison_cell *cell2;
384 static enum cache_metadata_mode get_cache_mode(struct cache *cache);
386 static void wake_worker(struct cache *cache)
388 queue_work(cache->wq, &cache->worker);
391 /*----------------------------------------------------------------*/
393 static struct dm_bio_prison_cell *alloc_prison_cell(struct cache *cache)
395 /* FIXME: change to use a local slab. */
396 return dm_bio_prison_alloc_cell(cache->prison, GFP_NOWAIT);
399 static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell *cell)
401 dm_bio_prison_free_cell(cache->prison, cell);
404 static struct dm_cache_migration *alloc_migration(struct cache *cache)
406 struct dm_cache_migration *mg;
408 mg = mempool_alloc(cache->migration_pool, GFP_NOWAIT);
411 atomic_inc(&mg->cache->nr_allocated_migrations);
417 static void free_migration(struct dm_cache_migration *mg)
419 struct cache *cache = mg->cache;
421 if (atomic_dec_and_test(&cache->nr_allocated_migrations))
422 wake_up(&cache->migration_wait);
424 mempool_free(mg, cache->migration_pool);
427 static int prealloc_data_structs(struct cache *cache, struct prealloc *p)
430 p->mg = alloc_migration(cache);
436 p->cell1 = alloc_prison_cell(cache);
442 p->cell2 = alloc_prison_cell(cache);
450 static void prealloc_free_structs(struct cache *cache, struct prealloc *p)
453 free_prison_cell(cache, p->cell2);
456 free_prison_cell(cache, p->cell1);
459 free_migration(p->mg);
462 static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p)
464 struct dm_cache_migration *mg = p->mg;
473 * You must have a cell within the prealloc struct to return. If not this
474 * function will BUG() rather than returning NULL.
476 static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p)
478 struct dm_bio_prison_cell *r = NULL;
484 } else if (p->cell2) {
494 * You can't have more than two cells in a prealloc struct. BUG() will be
495 * called if you try and overfill.
497 static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell)
509 /*----------------------------------------------------------------*/
511 static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key *key)
515 key->block_begin = from_oblock(begin);
516 key->block_end = from_oblock(end);
520 * The caller hands in a preallocated cell, and a free function for it.
521 * The cell will be freed if there's an error, or if it wasn't used because
522 * a cell with that key already exists.
524 typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell);
526 static int bio_detain_range(struct cache *cache, dm_oblock_t oblock_begin, dm_oblock_t oblock_end,
527 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
528 cell_free_fn free_fn, void *free_context,
529 struct dm_bio_prison_cell **cell_result)
532 struct dm_cell_key key;
534 build_key(oblock_begin, oblock_end, &key);
535 r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result);
537 free_fn(free_context, cell_prealloc);
542 static int bio_detain(struct cache *cache, dm_oblock_t oblock,
543 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
544 cell_free_fn free_fn, void *free_context,
545 struct dm_bio_prison_cell **cell_result)
547 dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL);
548 return bio_detain_range(cache, oblock, end, bio,
549 cell_prealloc, free_fn, free_context, cell_result);
552 static int get_cell(struct cache *cache,
554 struct prealloc *structs,
555 struct dm_bio_prison_cell **cell_result)
558 struct dm_cell_key key;
559 struct dm_bio_prison_cell *cell_prealloc;
561 cell_prealloc = prealloc_get_cell(structs);
563 build_key(oblock, to_oblock(from_oblock(oblock) + 1ULL), &key);
564 r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result);
566 prealloc_put_cell(structs, cell_prealloc);
571 /*----------------------------------------------------------------*/
573 static bool is_dirty(struct cache *cache, dm_cblock_t b)
575 return test_bit(from_cblock(b), cache->dirty_bitset);
578 static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
580 if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) {
581 atomic_inc(&cache->nr_dirty);
582 policy_set_dirty(cache->policy, oblock);
586 static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
588 if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) {
589 policy_clear_dirty(cache->policy, oblock);
590 if (atomic_dec_return(&cache->nr_dirty) == 0)
591 dm_table_event(cache->ti->table);
595 /*----------------------------------------------------------------*/
597 static bool block_size_is_power_of_two(struct cache *cache)
599 return cache->sectors_per_block_shift >= 0;
602 /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */
603 #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6
606 static dm_block_t block_div(dm_block_t b, uint32_t n)
613 static dm_block_t oblocks_per_dblock(struct cache *cache)
615 dm_block_t oblocks = cache->discard_block_size;
617 if (block_size_is_power_of_two(cache))
618 oblocks >>= cache->sectors_per_block_shift;
620 oblocks = block_div(oblocks, cache->sectors_per_block);
625 static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock)
627 return to_dblock(block_div(from_oblock(oblock),
628 oblocks_per_dblock(cache)));
631 static dm_oblock_t dblock_to_oblock(struct cache *cache, dm_dblock_t dblock)
633 return to_oblock(from_dblock(dblock) * oblocks_per_dblock(cache));
636 static void set_discard(struct cache *cache, dm_dblock_t b)
640 BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks));
641 atomic_inc(&cache->stats.discard_count);
643 spin_lock_irqsave(&cache->lock, flags);
644 set_bit(from_dblock(b), cache->discard_bitset);
645 spin_unlock_irqrestore(&cache->lock, flags);
648 static void clear_discard(struct cache *cache, dm_dblock_t b)
652 spin_lock_irqsave(&cache->lock, flags);
653 clear_bit(from_dblock(b), cache->discard_bitset);
654 spin_unlock_irqrestore(&cache->lock, flags);
657 static bool is_discarded(struct cache *cache, dm_dblock_t b)
662 spin_lock_irqsave(&cache->lock, flags);
663 r = test_bit(from_dblock(b), cache->discard_bitset);
664 spin_unlock_irqrestore(&cache->lock, flags);
669 static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b)
674 spin_lock_irqsave(&cache->lock, flags);
675 r = test_bit(from_dblock(oblock_to_dblock(cache, b)),
676 cache->discard_bitset);
677 spin_unlock_irqrestore(&cache->lock, flags);
682 /*----------------------------------------------------------------*/
684 static void load_stats(struct cache *cache)
686 struct dm_cache_statistics stats;
688 dm_cache_metadata_get_stats(cache->cmd, &stats);
689 atomic_set(&cache->stats.read_hit, stats.read_hits);
690 atomic_set(&cache->stats.read_miss, stats.read_misses);
691 atomic_set(&cache->stats.write_hit, stats.write_hits);
692 atomic_set(&cache->stats.write_miss, stats.write_misses);
695 static void save_stats(struct cache *cache)
697 struct dm_cache_statistics stats;
699 if (get_cache_mode(cache) >= CM_READ_ONLY)
702 stats.read_hits = atomic_read(&cache->stats.read_hit);
703 stats.read_misses = atomic_read(&cache->stats.read_miss);
704 stats.write_hits = atomic_read(&cache->stats.write_hit);
705 stats.write_misses = atomic_read(&cache->stats.write_miss);
707 dm_cache_metadata_set_stats(cache->cmd, &stats);
710 /*----------------------------------------------------------------
712 *--------------------------------------------------------------*/
715 * If using writeback, leave out struct per_bio_data's writethrough fields.
717 #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache))
718 #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data))
720 static bool writethrough_mode(struct cache_features *f)
722 return f->io_mode == CM_IO_WRITETHROUGH;
725 static bool writeback_mode(struct cache_features *f)
727 return f->io_mode == CM_IO_WRITEBACK;
730 static bool passthrough_mode(struct cache_features *f)
732 return f->io_mode == CM_IO_PASSTHROUGH;
735 static size_t get_per_bio_data_size(struct cache *cache)
737 return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB;
740 static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size)
742 struct per_bio_data *pb = dm_per_bio_data(bio, data_size);
747 static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size)
749 struct per_bio_data *pb = get_per_bio_data(bio, data_size);
752 pb->req_nr = dm_bio_get_target_bio_nr(bio);
753 pb->all_io_entry = NULL;
759 /*----------------------------------------------------------------
761 *--------------------------------------------------------------*/
762 static void remap_to_origin(struct cache *cache, struct bio *bio)
764 bio->bi_bdev = cache->origin_dev->bdev;
767 static void remap_to_cache(struct cache *cache, struct bio *bio,
770 sector_t bi_sector = bio->bi_iter.bi_sector;
771 sector_t block = from_cblock(cblock);
773 bio->bi_bdev = cache->cache_dev->bdev;
774 if (!block_size_is_power_of_two(cache))
775 bio->bi_iter.bi_sector =
776 (block * cache->sectors_per_block) +
777 sector_div(bi_sector, cache->sectors_per_block);
779 bio->bi_iter.bi_sector =
780 (block << cache->sectors_per_block_shift) |
781 (bi_sector & (cache->sectors_per_block - 1));
784 static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio)
787 size_t pb_data_size = get_per_bio_data_size(cache);
788 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
790 spin_lock_irqsave(&cache->lock, flags);
791 if (cache->need_tick_bio && !op_is_flush(bio->bi_opf) &&
792 bio_op(bio) != REQ_OP_DISCARD) {
794 cache->need_tick_bio = false;
796 spin_unlock_irqrestore(&cache->lock, flags);
799 static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio,
802 check_if_tick_bio_needed(cache, bio);
803 remap_to_origin(cache, bio);
804 if (bio_data_dir(bio) == WRITE)
805 clear_discard(cache, oblock_to_dblock(cache, oblock));
808 static void remap_to_cache_dirty(struct cache *cache, struct bio *bio,
809 dm_oblock_t oblock, dm_cblock_t cblock)
811 check_if_tick_bio_needed(cache, bio);
812 remap_to_cache(cache, bio, cblock);
813 if (bio_data_dir(bio) == WRITE) {
814 set_dirty(cache, oblock, cblock);
815 clear_discard(cache, oblock_to_dblock(cache, oblock));
819 static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio)
821 sector_t block_nr = bio->bi_iter.bi_sector;
823 if (!block_size_is_power_of_two(cache))
824 (void) sector_div(block_nr, cache->sectors_per_block);
826 block_nr >>= cache->sectors_per_block_shift;
828 return to_oblock(block_nr);
832 * You must increment the deferred set whilst the prison cell is held. To
833 * encourage this, we ask for 'cell' to be passed in.
835 static void inc_ds(struct cache *cache, struct bio *bio,
836 struct dm_bio_prison_cell *cell)
838 size_t pb_data_size = get_per_bio_data_size(cache);
839 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
842 BUG_ON(pb->all_io_entry);
844 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
847 static bool accountable_bio(struct cache *cache, struct bio *bio)
849 return ((bio->bi_bdev == cache->origin_dev->bdev) &&
850 bio_op(bio) != REQ_OP_DISCARD);
853 static void accounted_begin(struct cache *cache, struct bio *bio)
855 size_t pb_data_size = get_per_bio_data_size(cache);
856 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
858 if (accountable_bio(cache, bio)) {
859 pb->len = bio_sectors(bio);
860 iot_io_begin(&cache->origin_tracker, pb->len);
864 static void accounted_complete(struct cache *cache, struct bio *bio)
866 size_t pb_data_size = get_per_bio_data_size(cache);
867 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
869 iot_io_end(&cache->origin_tracker, pb->len);
872 static void accounted_request(struct cache *cache, struct bio *bio)
874 accounted_begin(cache, bio);
875 generic_make_request(bio);
878 static void issue(struct cache *cache, struct bio *bio)
882 if (!op_is_flush(bio->bi_opf)) {
883 accounted_request(cache, bio);
888 * Batch together any bios that trigger commits and then issue a
889 * single commit for them in do_worker().
891 spin_lock_irqsave(&cache->lock, flags);
892 cache->commit_requested = true;
893 bio_list_add(&cache->deferred_flush_bios, bio);
894 spin_unlock_irqrestore(&cache->lock, flags);
897 static void inc_and_issue(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell *cell)
899 inc_ds(cache, bio, cell);
903 static void defer_writethrough_bio(struct cache *cache, struct bio *bio)
907 spin_lock_irqsave(&cache->lock, flags);
908 bio_list_add(&cache->deferred_writethrough_bios, bio);
909 spin_unlock_irqrestore(&cache->lock, flags);
914 static void writethrough_endio(struct bio *bio)
916 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
918 dm_unhook_bio(&pb->hook_info, bio);
925 dm_bio_restore(&pb->bio_details, bio);
926 remap_to_cache(pb->cache, bio, pb->cblock);
929 * We can't issue this bio directly, since we're in interrupt
930 * context. So it gets put on a bio list for processing by the
933 defer_writethrough_bio(pb->cache, bio);
937 * When running in writethrough mode we need to send writes to clean blocks
938 * to both the cache and origin devices. In future we'd like to clone the
939 * bio and send them in parallel, but for now we're doing them in
940 * series as this is easier.
942 static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio,
943 dm_oblock_t oblock, dm_cblock_t cblock)
945 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
949 dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL);
950 dm_bio_record(&pb->bio_details, bio);
952 remap_to_origin_clear_discard(pb->cache, bio, oblock);
955 /*----------------------------------------------------------------
957 *--------------------------------------------------------------*/
958 static enum cache_metadata_mode get_cache_mode(struct cache *cache)
960 return cache->features.mode;
963 static const char *cache_device_name(struct cache *cache)
965 return dm_device_name(dm_table_get_md(cache->ti->table));
968 static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode)
970 const char *descs[] = {
976 dm_table_event(cache->ti->table);
977 DMINFO("%s: switching cache to %s mode",
978 cache_device_name(cache), descs[(int)mode]);
981 static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode)
984 enum cache_metadata_mode old_mode = get_cache_mode(cache);
986 if (dm_cache_metadata_needs_check(cache->cmd, &needs_check)) {
987 DMERR("%s: unable to read needs_check flag, setting failure mode.",
988 cache_device_name(cache));
992 if (new_mode == CM_WRITE && needs_check) {
993 DMERR("%s: unable to switch cache to write mode until repaired.",
994 cache_device_name(cache));
995 if (old_mode != new_mode)
998 new_mode = CM_READ_ONLY;
1001 /* Never move out of fail mode */
1002 if (old_mode == CM_FAIL)
1008 dm_cache_metadata_set_read_only(cache->cmd);
1012 dm_cache_metadata_set_read_write(cache->cmd);
1016 cache->features.mode = new_mode;
1018 if (new_mode != old_mode)
1019 notify_mode_switch(cache, new_mode);
1022 static void abort_transaction(struct cache *cache)
1024 const char *dev_name = cache_device_name(cache);
1026 if (get_cache_mode(cache) >= CM_READ_ONLY)
1029 if (dm_cache_metadata_set_needs_check(cache->cmd)) {
1030 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
1031 set_cache_mode(cache, CM_FAIL);
1034 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
1035 if (dm_cache_metadata_abort(cache->cmd)) {
1036 DMERR("%s: failed to abort metadata transaction", dev_name);
1037 set_cache_mode(cache, CM_FAIL);
1041 static void metadata_operation_failed(struct cache *cache, const char *op, int r)
1043 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1044 cache_device_name(cache), op, r);
1045 abort_transaction(cache);
1046 set_cache_mode(cache, CM_READ_ONLY);
1049 /*----------------------------------------------------------------
1050 * Migration processing
1052 * Migration covers moving data from the origin device to the cache, or
1054 *--------------------------------------------------------------*/
1055 static void inc_io_migrations(struct cache *cache)
1057 atomic_inc(&cache->nr_io_migrations);
1060 static void dec_io_migrations(struct cache *cache)
1062 atomic_dec(&cache->nr_io_migrations);
1065 static bool discard_or_flush(struct bio *bio)
1067 return bio_op(bio) == REQ_OP_DISCARD || op_is_flush(bio->bi_opf);
1070 static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell)
1072 if (discard_or_flush(cell->holder)) {
1074 * We have to handle these bios individually.
1076 dm_cell_release(cache->prison, cell, &cache->deferred_bios);
1077 free_prison_cell(cache, cell);
1079 list_add_tail(&cell->user_list, &cache->deferred_cells);
1082 static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, bool holder)
1084 unsigned long flags;
1086 if (!holder && dm_cell_promote_or_release(cache->prison, cell)) {
1088 * There was no prisoner to promote to holder, the
1089 * cell has been released.
1091 free_prison_cell(cache, cell);
1095 spin_lock_irqsave(&cache->lock, flags);
1096 __cell_defer(cache, cell);
1097 spin_unlock_irqrestore(&cache->lock, flags);
1102 static void cell_error_with_code(struct cache *cache, struct dm_bio_prison_cell *cell, int err)
1104 dm_cell_error(cache->prison, cell, err);
1105 free_prison_cell(cache, cell);
1108 static void cell_requeue(struct cache *cache, struct dm_bio_prison_cell *cell)
1110 cell_error_with_code(cache, cell, DM_ENDIO_REQUEUE);
1113 static void free_io_migration(struct dm_cache_migration *mg)
1115 struct cache *cache = mg->cache;
1117 dec_io_migrations(cache);
1122 static void migration_failure(struct dm_cache_migration *mg)
1124 struct cache *cache = mg->cache;
1125 const char *dev_name = cache_device_name(cache);
1127 if (mg->writeback) {
1128 DMERR_LIMIT("%s: writeback failed; couldn't copy block", dev_name);
1129 set_dirty(cache, mg->old_oblock, mg->cblock);
1130 cell_defer(cache, mg->old_ocell, false);
1132 } else if (mg->demote) {
1133 DMERR_LIMIT("%s: demotion failed; couldn't copy block", dev_name);
1134 policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock);
1136 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1138 cell_defer(cache, mg->new_ocell, true);
1140 DMERR_LIMIT("%s: promotion failed; couldn't copy block", dev_name);
1141 policy_remove_mapping(cache->policy, mg->new_oblock);
1142 cell_defer(cache, mg->new_ocell, true);
1145 free_io_migration(mg);
1148 static void migration_success_pre_commit(struct dm_cache_migration *mg)
1151 unsigned long flags;
1152 struct cache *cache = mg->cache;
1154 if (mg->writeback) {
1155 clear_dirty(cache, mg->old_oblock, mg->cblock);
1156 cell_defer(cache, mg->old_ocell, false);
1157 free_io_migration(mg);
1160 } else if (mg->demote) {
1161 r = dm_cache_remove_mapping(cache->cmd, mg->cblock);
1163 DMERR_LIMIT("%s: demotion failed; couldn't update on disk metadata",
1164 cache_device_name(cache));
1165 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
1166 policy_force_mapping(cache->policy, mg->new_oblock,
1169 cell_defer(cache, mg->new_ocell, true);
1170 free_io_migration(mg);
1174 r = dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock);
1176 DMERR_LIMIT("%s: promotion failed; couldn't update on disk metadata",
1177 cache_device_name(cache));
1178 metadata_operation_failed(cache, "dm_cache_insert_mapping", r);
1179 policy_remove_mapping(cache->policy, mg->new_oblock);
1180 free_io_migration(mg);
1185 spin_lock_irqsave(&cache->lock, flags);
1186 list_add_tail(&mg->list, &cache->need_commit_migrations);
1187 cache->commit_requested = true;
1188 spin_unlock_irqrestore(&cache->lock, flags);
1191 static void migration_success_post_commit(struct dm_cache_migration *mg)
1193 unsigned long flags;
1194 struct cache *cache = mg->cache;
1196 if (mg->writeback) {
1197 DMWARN_LIMIT("%s: writeback unexpectedly triggered commit",
1198 cache_device_name(cache));
1201 } else if (mg->demote) {
1202 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1207 spin_lock_irqsave(&cache->lock, flags);
1208 list_add_tail(&mg->list, &cache->quiesced_migrations);
1209 spin_unlock_irqrestore(&cache->lock, flags);
1213 policy_remove_mapping(cache->policy, mg->old_oblock);
1214 free_io_migration(mg);
1218 if (mg->requeue_holder) {
1219 clear_dirty(cache, mg->new_oblock, mg->cblock);
1220 cell_defer(cache, mg->new_ocell, true);
1223 * The block was promoted via an overwrite, so it's dirty.
1225 set_dirty(cache, mg->new_oblock, mg->cblock);
1226 bio_endio(mg->new_ocell->holder);
1227 cell_defer(cache, mg->new_ocell, false);
1229 free_io_migration(mg);
1233 static void copy_complete(int read_err, unsigned long write_err, void *context)
1235 unsigned long flags;
1236 struct dm_cache_migration *mg = (struct dm_cache_migration *) context;
1237 struct cache *cache = mg->cache;
1239 if (read_err || write_err)
1242 spin_lock_irqsave(&cache->lock, flags);
1243 list_add_tail(&mg->list, &cache->completed_migrations);
1244 spin_unlock_irqrestore(&cache->lock, flags);
1249 static void issue_copy(struct dm_cache_migration *mg)
1252 struct dm_io_region o_region, c_region;
1253 struct cache *cache = mg->cache;
1254 sector_t cblock = from_cblock(mg->cblock);
1256 o_region.bdev = cache->origin_dev->bdev;
1257 o_region.count = cache->sectors_per_block;
1259 c_region.bdev = cache->cache_dev->bdev;
1260 c_region.sector = cblock * cache->sectors_per_block;
1261 c_region.count = cache->sectors_per_block;
1263 if (mg->writeback || mg->demote) {
1265 o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block;
1266 r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg);
1269 o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block;
1270 r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg);
1274 DMERR_LIMIT("%s: issuing migration failed", cache_device_name(cache));
1275 migration_failure(mg);
1279 static void overwrite_endio(struct bio *bio)
1281 struct dm_cache_migration *mg = bio->bi_private;
1282 struct cache *cache = mg->cache;
1283 size_t pb_data_size = get_per_bio_data_size(cache);
1284 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1285 unsigned long flags;
1287 dm_unhook_bio(&pb->hook_info, bio);
1292 mg->requeue_holder = false;
1294 spin_lock_irqsave(&cache->lock, flags);
1295 list_add_tail(&mg->list, &cache->completed_migrations);
1296 spin_unlock_irqrestore(&cache->lock, flags);
1301 static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio)
1303 size_t pb_data_size = get_per_bio_data_size(mg->cache);
1304 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1306 dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg);
1307 remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock);
1310 * No need to inc_ds() here, since the cell will be held for the
1311 * duration of the io.
1313 accounted_request(mg->cache, bio);
1316 static bool bio_writes_complete_block(struct cache *cache, struct bio *bio)
1318 return (bio_data_dir(bio) == WRITE) &&
1319 (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT));
1322 static void avoid_copy(struct dm_cache_migration *mg)
1324 atomic_inc(&mg->cache->stats.copies_avoided);
1325 migration_success_pre_commit(mg);
1328 static void calc_discard_block_range(struct cache *cache, struct bio *bio,
1329 dm_dblock_t *b, dm_dblock_t *e)
1331 sector_t sb = bio->bi_iter.bi_sector;
1332 sector_t se = bio_end_sector(bio);
1334 *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size));
1336 if (se - sb < cache->discard_block_size)
1339 *e = to_dblock(block_div(se, cache->discard_block_size));
1342 static void issue_discard(struct dm_cache_migration *mg)
1345 struct bio *bio = mg->new_ocell->holder;
1346 struct cache *cache = mg->cache;
1348 calc_discard_block_range(cache, bio, &b, &e);
1350 set_discard(cache, b);
1351 b = to_dblock(from_dblock(b) + 1);
1355 cell_defer(cache, mg->new_ocell, false);
1360 static void issue_copy_or_discard(struct dm_cache_migration *mg)
1363 struct cache *cache = mg->cache;
1370 if (mg->writeback || mg->demote)
1371 avoid = !is_dirty(cache, mg->cblock) ||
1372 is_discarded_oblock(cache, mg->old_oblock);
1374 struct bio *bio = mg->new_ocell->holder;
1376 avoid = is_discarded_oblock(cache, mg->new_oblock);
1378 if (writeback_mode(&cache->features) &&
1379 !avoid && bio_writes_complete_block(cache, bio)) {
1380 issue_overwrite(mg, bio);
1385 avoid ? avoid_copy(mg) : issue_copy(mg);
1388 static void complete_migration(struct dm_cache_migration *mg)
1391 migration_failure(mg);
1393 migration_success_pre_commit(mg);
1396 static void process_migrations(struct cache *cache, struct list_head *head,
1397 void (*fn)(struct dm_cache_migration *))
1399 unsigned long flags;
1400 struct list_head list;
1401 struct dm_cache_migration *mg, *tmp;
1403 INIT_LIST_HEAD(&list);
1404 spin_lock_irqsave(&cache->lock, flags);
1405 list_splice_init(head, &list);
1406 spin_unlock_irqrestore(&cache->lock, flags);
1408 list_for_each_entry_safe(mg, tmp, &list, list)
1412 static void __queue_quiesced_migration(struct dm_cache_migration *mg)
1414 list_add_tail(&mg->list, &mg->cache->quiesced_migrations);
1417 static void queue_quiesced_migration(struct dm_cache_migration *mg)
1419 unsigned long flags;
1420 struct cache *cache = mg->cache;
1422 spin_lock_irqsave(&cache->lock, flags);
1423 __queue_quiesced_migration(mg);
1424 spin_unlock_irqrestore(&cache->lock, flags);
1429 static void queue_quiesced_migrations(struct cache *cache, struct list_head *work)
1431 unsigned long flags;
1432 struct dm_cache_migration *mg, *tmp;
1434 spin_lock_irqsave(&cache->lock, flags);
1435 list_for_each_entry_safe(mg, tmp, work, list)
1436 __queue_quiesced_migration(mg);
1437 spin_unlock_irqrestore(&cache->lock, flags);
1442 static void check_for_quiesced_migrations(struct cache *cache,
1443 struct per_bio_data *pb)
1445 struct list_head work;
1447 if (!pb->all_io_entry)
1450 INIT_LIST_HEAD(&work);
1451 dm_deferred_entry_dec(pb->all_io_entry, &work);
1453 if (!list_empty(&work))
1454 queue_quiesced_migrations(cache, &work);
1457 static void quiesce_migration(struct dm_cache_migration *mg)
1459 if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list))
1460 queue_quiesced_migration(mg);
1463 static void promote(struct cache *cache, struct prealloc *structs,
1464 dm_oblock_t oblock, dm_cblock_t cblock,
1465 struct dm_bio_prison_cell *cell)
1467 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1470 mg->discard = false;
1471 mg->writeback = false;
1474 mg->requeue_holder = true;
1475 mg->invalidate = false;
1477 mg->new_oblock = oblock;
1478 mg->cblock = cblock;
1479 mg->old_ocell = NULL;
1480 mg->new_ocell = cell;
1481 mg->start_jiffies = jiffies;
1483 inc_io_migrations(cache);
1484 quiesce_migration(mg);
1487 static void writeback(struct cache *cache, struct prealloc *structs,
1488 dm_oblock_t oblock, dm_cblock_t cblock,
1489 struct dm_bio_prison_cell *cell)
1491 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1494 mg->discard = false;
1495 mg->writeback = true;
1497 mg->promote = false;
1498 mg->requeue_holder = true;
1499 mg->invalidate = false;
1501 mg->old_oblock = oblock;
1502 mg->cblock = cblock;
1503 mg->old_ocell = cell;
1504 mg->new_ocell = NULL;
1505 mg->start_jiffies = jiffies;
1507 inc_io_migrations(cache);
1508 quiesce_migration(mg);
1511 static void demote_then_promote(struct cache *cache, struct prealloc *structs,
1512 dm_oblock_t old_oblock, dm_oblock_t new_oblock,
1514 struct dm_bio_prison_cell *old_ocell,
1515 struct dm_bio_prison_cell *new_ocell)
1517 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1520 mg->discard = false;
1521 mg->writeback = false;
1524 mg->requeue_holder = true;
1525 mg->invalidate = false;
1527 mg->old_oblock = old_oblock;
1528 mg->new_oblock = new_oblock;
1529 mg->cblock = cblock;
1530 mg->old_ocell = old_ocell;
1531 mg->new_ocell = new_ocell;
1532 mg->start_jiffies = jiffies;
1534 inc_io_migrations(cache);
1535 quiesce_migration(mg);
1539 * Invalidate a cache entry. No writeback occurs; any changes in the cache
1540 * block are thrown away.
1542 static void invalidate(struct cache *cache, struct prealloc *structs,
1543 dm_oblock_t oblock, dm_cblock_t cblock,
1544 struct dm_bio_prison_cell *cell)
1546 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1549 mg->discard = false;
1550 mg->writeback = false;
1552 mg->promote = false;
1553 mg->requeue_holder = true;
1554 mg->invalidate = true;
1556 mg->old_oblock = oblock;
1557 mg->cblock = cblock;
1558 mg->old_ocell = cell;
1559 mg->new_ocell = NULL;
1560 mg->start_jiffies = jiffies;
1562 inc_io_migrations(cache);
1563 quiesce_migration(mg);
1566 static void discard(struct cache *cache, struct prealloc *structs,
1567 struct dm_bio_prison_cell *cell)
1569 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1573 mg->writeback = false;
1575 mg->promote = false;
1576 mg->requeue_holder = false;
1577 mg->invalidate = false;
1579 mg->old_ocell = NULL;
1580 mg->new_ocell = cell;
1581 mg->start_jiffies = jiffies;
1583 quiesce_migration(mg);
1586 /*----------------------------------------------------------------
1588 *--------------------------------------------------------------*/
1589 static void defer_bio(struct cache *cache, struct bio *bio)
1591 unsigned long flags;
1593 spin_lock_irqsave(&cache->lock, flags);
1594 bio_list_add(&cache->deferred_bios, bio);
1595 spin_unlock_irqrestore(&cache->lock, flags);
1600 static void process_flush_bio(struct cache *cache, struct bio *bio)
1602 size_t pb_data_size = get_per_bio_data_size(cache);
1603 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1605 BUG_ON(bio->bi_iter.bi_size);
1607 remap_to_origin(cache, bio);
1609 remap_to_cache(cache, bio, 0);
1612 * REQ_PREFLUSH is not directed at any particular block so we don't
1613 * need to inc_ds(). REQ_FUA's are split into a write + REQ_PREFLUSH
1619 static void process_discard_bio(struct cache *cache, struct prealloc *structs,
1624 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1626 calc_discard_block_range(cache, bio, &b, &e);
1632 cell_prealloc = prealloc_get_cell(structs);
1633 r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc,
1634 (cell_free_fn) prealloc_put_cell,
1635 structs, &new_ocell);
1639 discard(cache, structs, new_ocell);
1642 static bool spare_migration_bandwidth(struct cache *cache)
1644 sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) *
1645 cache->sectors_per_block;
1646 return current_volume < cache->migration_threshold;
1649 static void inc_hit_counter(struct cache *cache, struct bio *bio)
1651 atomic_inc(bio_data_dir(bio) == READ ?
1652 &cache->stats.read_hit : &cache->stats.write_hit);
1655 static void inc_miss_counter(struct cache *cache, struct bio *bio)
1657 atomic_inc(bio_data_dir(bio) == READ ?
1658 &cache->stats.read_miss : &cache->stats.write_miss);
1661 /*----------------------------------------------------------------*/
1664 struct cache *cache;
1665 struct bio_list bios_for_issue;
1666 struct bio_list unhandled_bios;
1670 static void inc_fn(void *context, struct dm_bio_prison_cell *cell)
1673 struct inc_detail *detail = context;
1674 struct cache *cache = detail->cache;
1676 inc_ds(cache, cell->holder, cell);
1677 if (bio_data_dir(cell->holder) == WRITE)
1678 detail->any_writes = true;
1680 while ((bio = bio_list_pop(&cell->bios))) {
1681 if (discard_or_flush(bio)) {
1682 bio_list_add(&detail->unhandled_bios, bio);
1686 if (bio_data_dir(bio) == WRITE)
1687 detail->any_writes = true;
1689 bio_list_add(&detail->bios_for_issue, bio);
1690 inc_ds(cache, bio, cell);
1694 // FIXME: refactor these two
1695 static void remap_cell_to_origin_clear_discard(struct cache *cache,
1696 struct dm_bio_prison_cell *cell,
1697 dm_oblock_t oblock, bool issue_holder)
1700 unsigned long flags;
1701 struct inc_detail detail;
1703 detail.cache = cache;
1704 bio_list_init(&detail.bios_for_issue);
1705 bio_list_init(&detail.unhandled_bios);
1706 detail.any_writes = false;
1708 spin_lock_irqsave(&cache->lock, flags);
1709 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1710 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1711 spin_unlock_irqrestore(&cache->lock, flags);
1713 remap_to_origin(cache, cell->holder);
1715 issue(cache, cell->holder);
1717 accounted_begin(cache, cell->holder);
1719 if (detail.any_writes)
1720 clear_discard(cache, oblock_to_dblock(cache, oblock));
1722 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1723 remap_to_origin(cache, bio);
1727 free_prison_cell(cache, cell);
1730 static void remap_cell_to_cache_dirty(struct cache *cache, struct dm_bio_prison_cell *cell,
1731 dm_oblock_t oblock, dm_cblock_t cblock, bool issue_holder)
1734 unsigned long flags;
1735 struct inc_detail detail;
1737 detail.cache = cache;
1738 bio_list_init(&detail.bios_for_issue);
1739 bio_list_init(&detail.unhandled_bios);
1740 detail.any_writes = false;
1742 spin_lock_irqsave(&cache->lock, flags);
1743 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1744 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1745 spin_unlock_irqrestore(&cache->lock, flags);
1747 remap_to_cache(cache, cell->holder, cblock);
1749 issue(cache, cell->holder);
1751 accounted_begin(cache, cell->holder);
1753 if (detail.any_writes) {
1754 set_dirty(cache, oblock, cblock);
1755 clear_discard(cache, oblock_to_dblock(cache, oblock));
1758 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1759 remap_to_cache(cache, bio, cblock);
1763 free_prison_cell(cache, cell);
1766 /*----------------------------------------------------------------*/
1768 struct old_oblock_lock {
1769 struct policy_locker locker;
1770 struct cache *cache;
1771 struct prealloc *structs;
1772 struct dm_bio_prison_cell *cell;
1775 static int null_locker(struct policy_locker *locker, dm_oblock_t b)
1777 /* This should never be called */
1782 static int cell_locker(struct policy_locker *locker, dm_oblock_t b)
1784 struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker);
1785 struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs);
1787 return bio_detain(l->cache, b, NULL, cell_prealloc,
1788 (cell_free_fn) prealloc_put_cell,
1789 l->structs, &l->cell);
1792 static void process_cell(struct cache *cache, struct prealloc *structs,
1793 struct dm_bio_prison_cell *new_ocell)
1796 bool release_cell = true;
1797 struct bio *bio = new_ocell->holder;
1798 dm_oblock_t block = get_bio_block(cache, bio);
1799 struct policy_result lookup_result;
1800 bool passthrough = passthrough_mode(&cache->features);
1801 bool fast_promotion, can_migrate;
1802 struct old_oblock_lock ool;
1804 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
1805 can_migrate = !passthrough && (fast_promotion || spare_migration_bandwidth(cache));
1807 ool.locker.fn = cell_locker;
1809 ool.structs = structs;
1811 r = policy_map(cache->policy, block, true, can_migrate, fast_promotion,
1812 bio, &ool.locker, &lookup_result);
1814 if (r == -EWOULDBLOCK)
1815 /* migration has been denied */
1816 lookup_result.op = POLICY_MISS;
1818 switch (lookup_result.op) {
1821 inc_miss_counter(cache, bio);
1824 * Passthrough always maps to the origin,
1825 * invalidating any cache blocks that are written
1829 if (bio_data_dir(bio) == WRITE) {
1830 atomic_inc(&cache->stats.demotion);
1831 invalidate(cache, structs, block, lookup_result.cblock, new_ocell);
1832 release_cell = false;
1835 /* FIXME: factor out issue_origin() */
1836 remap_to_origin_clear_discard(cache, bio, block);
1837 inc_and_issue(cache, bio, new_ocell);
1840 inc_hit_counter(cache, bio);
1842 if (bio_data_dir(bio) == WRITE &&
1843 writethrough_mode(&cache->features) &&
1844 !is_dirty(cache, lookup_result.cblock)) {
1845 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
1846 inc_and_issue(cache, bio, new_ocell);
1849 remap_cell_to_cache_dirty(cache, new_ocell, block, lookup_result.cblock, true);
1850 release_cell = false;
1857 inc_miss_counter(cache, bio);
1858 remap_cell_to_origin_clear_discard(cache, new_ocell, block, true);
1859 release_cell = false;
1863 atomic_inc(&cache->stats.promotion);
1864 promote(cache, structs, block, lookup_result.cblock, new_ocell);
1865 release_cell = false;
1868 case POLICY_REPLACE:
1869 atomic_inc(&cache->stats.demotion);
1870 atomic_inc(&cache->stats.promotion);
1871 demote_then_promote(cache, structs, lookup_result.old_oblock,
1872 block, lookup_result.cblock,
1873 ool.cell, new_ocell);
1874 release_cell = false;
1878 DMERR_LIMIT("%s: %s: erroring bio, unknown policy op: %u",
1879 cache_device_name(cache), __func__,
1880 (unsigned) lookup_result.op);
1885 cell_defer(cache, new_ocell, false);
1888 static void process_bio(struct cache *cache, struct prealloc *structs,
1892 dm_oblock_t block = get_bio_block(cache, bio);
1893 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1896 * Check to see if that block is currently migrating.
1898 cell_prealloc = prealloc_get_cell(structs);
1899 r = bio_detain(cache, block, bio, cell_prealloc,
1900 (cell_free_fn) prealloc_put_cell,
1901 structs, &new_ocell);
1905 process_cell(cache, structs, new_ocell);
1908 static int need_commit_due_to_time(struct cache *cache)
1910 return jiffies < cache->last_commit_jiffies ||
1911 jiffies > cache->last_commit_jiffies + COMMIT_PERIOD;
1915 * A non-zero return indicates read_only or fail_io mode.
1917 static int commit(struct cache *cache, bool clean_shutdown)
1921 if (get_cache_mode(cache) >= CM_READ_ONLY)
1924 atomic_inc(&cache->stats.commit_count);
1925 r = dm_cache_commit(cache->cmd, clean_shutdown);
1927 metadata_operation_failed(cache, "dm_cache_commit", r);
1932 static int commit_if_needed(struct cache *cache)
1936 if ((cache->commit_requested || need_commit_due_to_time(cache)) &&
1937 dm_cache_changed_this_transaction(cache->cmd)) {
1938 r = commit(cache, false);
1939 cache->commit_requested = false;
1940 cache->last_commit_jiffies = jiffies;
1946 static void process_deferred_bios(struct cache *cache)
1948 bool prealloc_used = false;
1949 unsigned long flags;
1950 struct bio_list bios;
1952 struct prealloc structs;
1954 memset(&structs, 0, sizeof(structs));
1955 bio_list_init(&bios);
1957 spin_lock_irqsave(&cache->lock, flags);
1958 bio_list_merge(&bios, &cache->deferred_bios);
1959 bio_list_init(&cache->deferred_bios);
1960 spin_unlock_irqrestore(&cache->lock, flags);
1962 while (!bio_list_empty(&bios)) {
1964 * If we've got no free migration structs, and processing
1965 * this bio might require one, we pause until there are some
1966 * prepared mappings to process.
1968 prealloc_used = true;
1969 if (prealloc_data_structs(cache, &structs)) {
1970 spin_lock_irqsave(&cache->lock, flags);
1971 bio_list_merge(&cache->deferred_bios, &bios);
1972 spin_unlock_irqrestore(&cache->lock, flags);
1976 bio = bio_list_pop(&bios);
1978 if (bio->bi_opf & REQ_PREFLUSH)
1979 process_flush_bio(cache, bio);
1980 else if (bio_op(bio) == REQ_OP_DISCARD)
1981 process_discard_bio(cache, &structs, bio);
1983 process_bio(cache, &structs, bio);
1987 prealloc_free_structs(cache, &structs);
1990 static void process_deferred_cells(struct cache *cache)
1992 bool prealloc_used = false;
1993 unsigned long flags;
1994 struct dm_bio_prison_cell *cell, *tmp;
1995 struct list_head cells;
1996 struct prealloc structs;
1998 memset(&structs, 0, sizeof(structs));
2000 INIT_LIST_HEAD(&cells);
2002 spin_lock_irqsave(&cache->lock, flags);
2003 list_splice_init(&cache->deferred_cells, &cells);
2004 spin_unlock_irqrestore(&cache->lock, flags);
2006 list_for_each_entry_safe(cell, tmp, &cells, user_list) {
2008 * If we've got no free migration structs, and processing
2009 * this bio might require one, we pause until there are some
2010 * prepared mappings to process.
2012 prealloc_used = true;
2013 if (prealloc_data_structs(cache, &structs)) {
2014 spin_lock_irqsave(&cache->lock, flags);
2015 list_splice(&cells, &cache->deferred_cells);
2016 spin_unlock_irqrestore(&cache->lock, flags);
2020 process_cell(cache, &structs, cell);
2024 prealloc_free_structs(cache, &structs);
2027 static void process_deferred_flush_bios(struct cache *cache, bool submit_bios)
2029 unsigned long flags;
2030 struct bio_list bios;
2033 bio_list_init(&bios);
2035 spin_lock_irqsave(&cache->lock, flags);
2036 bio_list_merge(&bios, &cache->deferred_flush_bios);
2037 bio_list_init(&cache->deferred_flush_bios);
2038 spin_unlock_irqrestore(&cache->lock, flags);
2041 * These bios have already been through inc_ds()
2043 while ((bio = bio_list_pop(&bios)))
2044 submit_bios ? accounted_request(cache, bio) : bio_io_error(bio);
2047 static void process_deferred_writethrough_bios(struct cache *cache)
2049 unsigned long flags;
2050 struct bio_list bios;
2053 bio_list_init(&bios);
2055 spin_lock_irqsave(&cache->lock, flags);
2056 bio_list_merge(&bios, &cache->deferred_writethrough_bios);
2057 bio_list_init(&cache->deferred_writethrough_bios);
2058 spin_unlock_irqrestore(&cache->lock, flags);
2061 * These bios have already been through inc_ds()
2063 while ((bio = bio_list_pop(&bios)))
2064 accounted_request(cache, bio);
2067 static void writeback_some_dirty_blocks(struct cache *cache)
2069 bool prealloc_used = false;
2072 struct prealloc structs;
2073 struct dm_bio_prison_cell *old_ocell;
2074 bool busy = !iot_idle_for(&cache->origin_tracker, HZ);
2076 memset(&structs, 0, sizeof(structs));
2078 while (spare_migration_bandwidth(cache)) {
2079 if (policy_writeback_work(cache->policy, &oblock, &cblock, busy))
2080 break; /* no work to do */
2082 prealloc_used = true;
2083 if (prealloc_data_structs(cache, &structs) ||
2084 get_cell(cache, oblock, &structs, &old_ocell)) {
2085 policy_set_dirty(cache->policy, oblock);
2089 writeback(cache, &structs, oblock, cblock, old_ocell);
2093 prealloc_free_structs(cache, &structs);
2096 /*----------------------------------------------------------------
2098 * Dropping something from the cache *without* writing back.
2099 *--------------------------------------------------------------*/
2101 static void process_invalidation_request(struct cache *cache, struct invalidation_request *req)
2104 uint64_t begin = from_cblock(req->cblocks->begin);
2105 uint64_t end = from_cblock(req->cblocks->end);
2107 while (begin != end) {
2108 r = policy_remove_cblock(cache->policy, to_cblock(begin));
2110 r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin));
2112 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
2116 } else if (r == -ENODATA) {
2117 /* harmless, already unmapped */
2121 DMERR("%s: policy_remove_cblock failed", cache_device_name(cache));
2128 cache->commit_requested = true;
2131 atomic_set(&req->complete, 1);
2133 wake_up(&req->result_wait);
2136 static void process_invalidation_requests(struct cache *cache)
2138 struct list_head list;
2139 struct invalidation_request *req, *tmp;
2141 INIT_LIST_HEAD(&list);
2142 spin_lock(&cache->invalidation_lock);
2143 list_splice_init(&cache->invalidation_requests, &list);
2144 spin_unlock(&cache->invalidation_lock);
2146 list_for_each_entry_safe (req, tmp, &list, list)
2147 process_invalidation_request(cache, req);
2150 /*----------------------------------------------------------------
2152 *--------------------------------------------------------------*/
2153 static bool is_quiescing(struct cache *cache)
2155 return atomic_read(&cache->quiescing);
2158 static void ack_quiescing(struct cache *cache)
2160 if (is_quiescing(cache)) {
2161 atomic_inc(&cache->quiescing_ack);
2162 wake_up(&cache->quiescing_wait);
2166 static void wait_for_quiescing_ack(struct cache *cache)
2168 wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack));
2171 static void start_quiescing(struct cache *cache)
2173 atomic_inc(&cache->quiescing);
2174 wait_for_quiescing_ack(cache);
2177 static void stop_quiescing(struct cache *cache)
2179 atomic_set(&cache->quiescing, 0);
2180 atomic_set(&cache->quiescing_ack, 0);
2183 static void wait_for_migrations(struct cache *cache)
2185 wait_event(cache->migration_wait, !atomic_read(&cache->nr_allocated_migrations));
2188 static void stop_worker(struct cache *cache)
2190 cancel_delayed_work(&cache->waker);
2191 flush_workqueue(cache->wq);
2194 static void requeue_deferred_cells(struct cache *cache)
2196 unsigned long flags;
2197 struct list_head cells;
2198 struct dm_bio_prison_cell *cell, *tmp;
2200 INIT_LIST_HEAD(&cells);
2201 spin_lock_irqsave(&cache->lock, flags);
2202 list_splice_init(&cache->deferred_cells, &cells);
2203 spin_unlock_irqrestore(&cache->lock, flags);
2205 list_for_each_entry_safe(cell, tmp, &cells, user_list)
2206 cell_requeue(cache, cell);
2209 static void requeue_deferred_bios(struct cache *cache)
2212 struct bio_list bios;
2214 bio_list_init(&bios);
2215 bio_list_merge(&bios, &cache->deferred_bios);
2216 bio_list_init(&cache->deferred_bios);
2218 while ((bio = bio_list_pop(&bios))) {
2219 bio->bi_error = DM_ENDIO_REQUEUE;
2224 static int more_work(struct cache *cache)
2226 if (is_quiescing(cache))
2227 return !list_empty(&cache->quiesced_migrations) ||
2228 !list_empty(&cache->completed_migrations) ||
2229 !list_empty(&cache->need_commit_migrations);
2231 return !bio_list_empty(&cache->deferred_bios) ||
2232 !list_empty(&cache->deferred_cells) ||
2233 !bio_list_empty(&cache->deferred_flush_bios) ||
2234 !bio_list_empty(&cache->deferred_writethrough_bios) ||
2235 !list_empty(&cache->quiesced_migrations) ||
2236 !list_empty(&cache->completed_migrations) ||
2237 !list_empty(&cache->need_commit_migrations) ||
2241 static void do_worker(struct work_struct *ws)
2243 struct cache *cache = container_of(ws, struct cache, worker);
2246 if (!is_quiescing(cache)) {
2247 writeback_some_dirty_blocks(cache);
2248 process_deferred_writethrough_bios(cache);
2249 process_deferred_bios(cache);
2250 process_deferred_cells(cache);
2251 process_invalidation_requests(cache);
2254 process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard);
2255 process_migrations(cache, &cache->completed_migrations, complete_migration);
2257 if (commit_if_needed(cache)) {
2258 process_deferred_flush_bios(cache, false);
2259 process_migrations(cache, &cache->need_commit_migrations, migration_failure);
2261 process_deferred_flush_bios(cache, true);
2262 process_migrations(cache, &cache->need_commit_migrations,
2263 migration_success_post_commit);
2266 ack_quiescing(cache);
2268 } while (more_work(cache));
2272 * We want to commit periodically so that not too much
2273 * unwritten metadata builds up.
2275 static void do_waker(struct work_struct *ws)
2277 struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker);
2278 policy_tick(cache->policy, true);
2280 queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD);
2283 /*----------------------------------------------------------------*/
2285 static int is_congested(struct dm_dev *dev, int bdi_bits)
2287 struct request_queue *q = bdev_get_queue(dev->bdev);
2288 return bdi_congested(q->backing_dev_info, bdi_bits);
2291 static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2293 struct cache *cache = container_of(cb, struct cache, callbacks);
2295 return is_congested(cache->origin_dev, bdi_bits) ||
2296 is_congested(cache->cache_dev, bdi_bits);
2299 /*----------------------------------------------------------------
2301 *--------------------------------------------------------------*/
2304 * This function gets called on the error paths of the constructor, so we
2305 * have to cope with a partially initialised struct.
2307 static void destroy(struct cache *cache)
2311 mempool_destroy(cache->migration_pool);
2313 if (cache->all_io_ds)
2314 dm_deferred_set_destroy(cache->all_io_ds);
2317 dm_bio_prison_destroy(cache->prison);
2320 destroy_workqueue(cache->wq);
2322 if (cache->dirty_bitset)
2323 free_bitset(cache->dirty_bitset);
2325 if (cache->discard_bitset)
2326 free_bitset(cache->discard_bitset);
2329 dm_kcopyd_client_destroy(cache->copier);
2332 dm_cache_metadata_close(cache->cmd);
2334 if (cache->metadata_dev)
2335 dm_put_device(cache->ti, cache->metadata_dev);
2337 if (cache->origin_dev)
2338 dm_put_device(cache->ti, cache->origin_dev);
2340 if (cache->cache_dev)
2341 dm_put_device(cache->ti, cache->cache_dev);
2344 dm_cache_policy_destroy(cache->policy);
2346 for (i = 0; i < cache->nr_ctr_args ; i++)
2347 kfree(cache->ctr_args[i]);
2348 kfree(cache->ctr_args);
2353 static void cache_dtr(struct dm_target *ti)
2355 struct cache *cache = ti->private;
2360 static sector_t get_dev_size(struct dm_dev *dev)
2362 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
2365 /*----------------------------------------------------------------*/
2368 * Construct a cache device mapping.
2370 * cache <metadata dev> <cache dev> <origin dev> <block size>
2371 * <#feature args> [<feature arg>]*
2372 * <policy> <#policy args> [<policy arg>]*
2374 * metadata dev : fast device holding the persistent metadata
2375 * cache dev : fast device holding cached data blocks
2376 * origin dev : slow device holding original data blocks
2377 * block size : cache unit size in sectors
2379 * #feature args : number of feature arguments passed
2380 * feature args : writethrough. (The default is writeback.)
2382 * policy : the replacement policy to use
2383 * #policy args : an even number of policy arguments corresponding
2384 * to key/value pairs passed to the policy
2385 * policy args : key/value pairs passed to the policy
2386 * E.g. 'sequential_threshold 1024'
2387 * See cache-policies.txt for details.
2389 * Optional feature arguments are:
2390 * writethrough : write through caching that prohibits cache block
2391 * content from being different from origin block content.
2392 * Without this argument, the default behaviour is to write
2393 * back cache block contents later for performance reasons,
2394 * so they may differ from the corresponding origin blocks.
2397 struct dm_target *ti;
2399 struct dm_dev *metadata_dev;
2401 struct dm_dev *cache_dev;
2402 sector_t cache_sectors;
2404 struct dm_dev *origin_dev;
2405 sector_t origin_sectors;
2407 uint32_t block_size;
2409 const char *policy_name;
2411 const char **policy_argv;
2413 struct cache_features features;
2416 static void destroy_cache_args(struct cache_args *ca)
2418 if (ca->metadata_dev)
2419 dm_put_device(ca->ti, ca->metadata_dev);
2422 dm_put_device(ca->ti, ca->cache_dev);
2425 dm_put_device(ca->ti, ca->origin_dev);
2430 static bool at_least_one_arg(struct dm_arg_set *as, char **error)
2433 *error = "Insufficient args";
2440 static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as,
2444 sector_t metadata_dev_size;
2445 char b[BDEVNAME_SIZE];
2447 if (!at_least_one_arg(as, error))
2450 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2453 *error = "Error opening metadata device";
2457 metadata_dev_size = get_dev_size(ca->metadata_dev);
2458 if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING)
2459 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2460 bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
2465 static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as,
2470 if (!at_least_one_arg(as, error))
2473 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2476 *error = "Error opening cache device";
2479 ca->cache_sectors = get_dev_size(ca->cache_dev);
2484 static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as,
2489 if (!at_least_one_arg(as, error))
2492 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2495 *error = "Error opening origin device";
2499 ca->origin_sectors = get_dev_size(ca->origin_dev);
2500 if (ca->ti->len > ca->origin_sectors) {
2501 *error = "Device size larger than cached device";
2508 static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as,
2511 unsigned long block_size;
2513 if (!at_least_one_arg(as, error))
2516 if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size ||
2517 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2518 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2519 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2520 *error = "Invalid data block size";
2524 if (block_size > ca->cache_sectors) {
2525 *error = "Data block size is larger than the cache device";
2529 ca->block_size = block_size;
2534 static void init_features(struct cache_features *cf)
2536 cf->mode = CM_WRITE;
2537 cf->io_mode = CM_IO_WRITEBACK;
2538 cf->metadata_version = 1;
2541 static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
2544 static struct dm_arg _args[] = {
2545 {0, 2, "Invalid number of cache feature arguments"},
2551 struct cache_features *cf = &ca->features;
2555 r = dm_read_arg_group(_args, as, &argc, error);
2560 arg = dm_shift_arg(as);
2562 if (!strcasecmp(arg, "writeback"))
2563 cf->io_mode = CM_IO_WRITEBACK;
2565 else if (!strcasecmp(arg, "writethrough"))
2566 cf->io_mode = CM_IO_WRITETHROUGH;
2568 else if (!strcasecmp(arg, "passthrough"))
2569 cf->io_mode = CM_IO_PASSTHROUGH;
2571 else if (!strcasecmp(arg, "metadata2"))
2572 cf->metadata_version = 2;
2575 *error = "Unrecognised cache feature requested";
2583 static int parse_policy(struct cache_args *ca, struct dm_arg_set *as,
2586 static struct dm_arg _args[] = {
2587 {0, 1024, "Invalid number of policy arguments"},
2592 if (!at_least_one_arg(as, error))
2595 ca->policy_name = dm_shift_arg(as);
2597 r = dm_read_arg_group(_args, as, &ca->policy_argc, error);
2601 ca->policy_argv = (const char **)as->argv;
2602 dm_consume_args(as, ca->policy_argc);
2607 static int parse_cache_args(struct cache_args *ca, int argc, char **argv,
2611 struct dm_arg_set as;
2616 r = parse_metadata_dev(ca, &as, error);
2620 r = parse_cache_dev(ca, &as, error);
2624 r = parse_origin_dev(ca, &as, error);
2628 r = parse_block_size(ca, &as, error);
2632 r = parse_features(ca, &as, error);
2636 r = parse_policy(ca, &as, error);
2643 /*----------------------------------------------------------------*/
2645 static struct kmem_cache *migration_cache;
2647 #define NOT_CORE_OPTION 1
2649 static int process_config_option(struct cache *cache, const char *key, const char *value)
2653 if (!strcasecmp(key, "migration_threshold")) {
2654 if (kstrtoul(value, 10, &tmp))
2657 cache->migration_threshold = tmp;
2661 return NOT_CORE_OPTION;
2664 static int set_config_value(struct cache *cache, const char *key, const char *value)
2666 int r = process_config_option(cache, key, value);
2668 if (r == NOT_CORE_OPTION)
2669 r = policy_set_config_value(cache->policy, key, value);
2672 DMWARN("bad config value for %s: %s", key, value);
2677 static int set_config_values(struct cache *cache, int argc, const char **argv)
2682 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
2687 r = set_config_value(cache, argv[0], argv[1]);
2698 static int create_cache_policy(struct cache *cache, struct cache_args *ca,
2701 struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name,
2703 cache->origin_sectors,
2704 cache->sectors_per_block);
2706 *error = "Error creating cache's policy";
2715 * We want the discard block size to be at least the size of the cache
2716 * block size and have no more than 2^14 discard blocks across the origin.
2718 #define MAX_DISCARD_BLOCKS (1 << 14)
2720 static bool too_many_discard_blocks(sector_t discard_block_size,
2721 sector_t origin_size)
2723 (void) sector_div(origin_size, discard_block_size);
2725 return origin_size > MAX_DISCARD_BLOCKS;
2728 static sector_t calculate_discard_block_size(sector_t cache_block_size,
2729 sector_t origin_size)
2731 sector_t discard_block_size = cache_block_size;
2734 while (too_many_discard_blocks(discard_block_size, origin_size))
2735 discard_block_size *= 2;
2737 return discard_block_size;
2740 static void set_cache_size(struct cache *cache, dm_cblock_t size)
2742 dm_block_t nr_blocks = from_cblock(size);
2744 if (nr_blocks > (1 << 20) && cache->cache_size != size)
2745 DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n"
2746 "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n"
2747 "Please consider increasing the cache block size to reduce the overall cache block count.",
2748 (unsigned long long) nr_blocks);
2750 cache->cache_size = size;
2753 #define DEFAULT_MIGRATION_THRESHOLD 2048
2755 static int cache_create(struct cache_args *ca, struct cache **result)
2758 char **error = &ca->ti->error;
2759 struct cache *cache;
2760 struct dm_target *ti = ca->ti;
2761 dm_block_t origin_blocks;
2762 struct dm_cache_metadata *cmd;
2763 bool may_format = ca->features.mode == CM_WRITE;
2765 cache = kzalloc(sizeof(*cache), GFP_KERNEL);
2770 ti->private = cache;
2771 ti->num_flush_bios = 2;
2772 ti->flush_supported = true;
2774 ti->num_discard_bios = 1;
2775 ti->discards_supported = true;
2776 ti->split_discard_bios = false;
2778 cache->features = ca->features;
2779 ti->per_io_data_size = get_per_bio_data_size(cache);
2781 cache->callbacks.congested_fn = cache_is_congested;
2782 dm_table_add_target_callbacks(ti->table, &cache->callbacks);
2784 cache->metadata_dev = ca->metadata_dev;
2785 cache->origin_dev = ca->origin_dev;
2786 cache->cache_dev = ca->cache_dev;
2788 ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL;
2790 /* FIXME: factor out this whole section */
2791 origin_blocks = cache->origin_sectors = ca->origin_sectors;
2792 origin_blocks = block_div(origin_blocks, ca->block_size);
2793 cache->origin_blocks = to_oblock(origin_blocks);
2795 cache->sectors_per_block = ca->block_size;
2796 if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) {
2801 if (ca->block_size & (ca->block_size - 1)) {
2802 dm_block_t cache_size = ca->cache_sectors;
2804 cache->sectors_per_block_shift = -1;
2805 cache_size = block_div(cache_size, ca->block_size);
2806 set_cache_size(cache, to_cblock(cache_size));
2808 cache->sectors_per_block_shift = __ffs(ca->block_size);
2809 set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift));
2812 r = create_cache_policy(cache, ca, error);
2816 cache->policy_nr_args = ca->policy_argc;
2817 cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD;
2819 r = set_config_values(cache, ca->policy_argc, ca->policy_argv);
2821 *error = "Error setting cache policy's config values";
2825 cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
2826 ca->block_size, may_format,
2827 dm_cache_policy_get_hint_size(cache->policy),
2828 ca->features.metadata_version);
2830 *error = "Error creating metadata object";
2835 set_cache_mode(cache, CM_WRITE);
2836 if (get_cache_mode(cache) != CM_WRITE) {
2837 *error = "Unable to get write access to metadata, please check/repair metadata.";
2842 if (passthrough_mode(&cache->features)) {
2845 r = dm_cache_metadata_all_clean(cache->cmd, &all_clean);
2847 *error = "dm_cache_metadata_all_clean() failed";
2852 *error = "Cannot enter passthrough mode unless all blocks are clean";
2858 spin_lock_init(&cache->lock);
2859 INIT_LIST_HEAD(&cache->deferred_cells);
2860 bio_list_init(&cache->deferred_bios);
2861 bio_list_init(&cache->deferred_flush_bios);
2862 bio_list_init(&cache->deferred_writethrough_bios);
2863 INIT_LIST_HEAD(&cache->quiesced_migrations);
2864 INIT_LIST_HEAD(&cache->completed_migrations);
2865 INIT_LIST_HEAD(&cache->need_commit_migrations);
2866 atomic_set(&cache->nr_allocated_migrations, 0);
2867 atomic_set(&cache->nr_io_migrations, 0);
2868 init_waitqueue_head(&cache->migration_wait);
2870 init_waitqueue_head(&cache->quiescing_wait);
2871 atomic_set(&cache->quiescing, 0);
2872 atomic_set(&cache->quiescing_ack, 0);
2875 atomic_set(&cache->nr_dirty, 0);
2876 cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size));
2877 if (!cache->dirty_bitset) {
2878 *error = "could not allocate dirty bitset";
2881 clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size));
2883 cache->discard_block_size =
2884 calculate_discard_block_size(cache->sectors_per_block,
2885 cache->origin_sectors);
2886 cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors,
2887 cache->discard_block_size));
2888 cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks));
2889 if (!cache->discard_bitset) {
2890 *error = "could not allocate discard bitset";
2893 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
2895 cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2896 if (IS_ERR(cache->copier)) {
2897 *error = "could not create kcopyd client";
2898 r = PTR_ERR(cache->copier);
2902 cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2904 *error = "could not create workqueue for metadata object";
2907 INIT_WORK(&cache->worker, do_worker);
2908 INIT_DELAYED_WORK(&cache->waker, do_waker);
2909 cache->last_commit_jiffies = jiffies;
2911 cache->prison = dm_bio_prison_create();
2912 if (!cache->prison) {
2913 *error = "could not create bio prison";
2917 cache->all_io_ds = dm_deferred_set_create();
2918 if (!cache->all_io_ds) {
2919 *error = "could not create all_io deferred set";
2923 cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE,
2925 if (!cache->migration_pool) {
2926 *error = "Error creating cache's migration mempool";
2930 cache->need_tick_bio = true;
2931 cache->sized = false;
2932 cache->invalidate = false;
2933 cache->commit_requested = false;
2934 cache->loaded_mappings = false;
2935 cache->loaded_discards = false;
2939 atomic_set(&cache->stats.demotion, 0);
2940 atomic_set(&cache->stats.promotion, 0);
2941 atomic_set(&cache->stats.copies_avoided, 0);
2942 atomic_set(&cache->stats.cache_cell_clash, 0);
2943 atomic_set(&cache->stats.commit_count, 0);
2944 atomic_set(&cache->stats.discard_count, 0);
2946 spin_lock_init(&cache->invalidation_lock);
2947 INIT_LIST_HEAD(&cache->invalidation_requests);
2949 iot_init(&cache->origin_tracker);
2959 static int copy_ctr_args(struct cache *cache, int argc, const char **argv)
2964 copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL);
2967 for (i = 0; i < argc; i++) {
2968 copy[i] = kstrdup(argv[i], GFP_KERNEL);
2977 cache->nr_ctr_args = argc;
2978 cache->ctr_args = copy;
2983 static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv)
2986 struct cache_args *ca;
2987 struct cache *cache = NULL;
2989 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2991 ti->error = "Error allocating memory for cache";
2996 r = parse_cache_args(ca, argc, argv, &ti->error);
3000 r = cache_create(ca, &cache);
3004 r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3);
3010 ti->private = cache;
3013 destroy_cache_args(ca);
3017 /*----------------------------------------------------------------*/
3019 static int cache_map(struct dm_target *ti, struct bio *bio)
3021 struct cache *cache = ti->private;
3024 struct dm_bio_prison_cell *cell = NULL;
3025 dm_oblock_t block = get_bio_block(cache, bio);
3026 size_t pb_data_size = get_per_bio_data_size(cache);
3027 bool can_migrate = false;
3028 bool fast_promotion;
3029 struct policy_result lookup_result;
3030 struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size);
3031 struct old_oblock_lock ool;
3033 ool.locker.fn = null_locker;
3035 if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) {
3037 * This can only occur if the io goes to a partial block at
3038 * the end of the origin device. We don't cache these.
3039 * Just remap to the origin and carry on.
3041 remap_to_origin(cache, bio);
3042 accounted_begin(cache, bio);
3043 return DM_MAPIO_REMAPPED;
3046 if (discard_or_flush(bio)) {
3047 defer_bio(cache, bio);
3048 return DM_MAPIO_SUBMITTED;
3052 * Check to see if that block is currently migrating.
3054 cell = alloc_prison_cell(cache);
3056 defer_bio(cache, bio);
3057 return DM_MAPIO_SUBMITTED;
3060 r = bio_detain(cache, block, bio, cell,
3061 (cell_free_fn) free_prison_cell,
3065 defer_bio(cache, bio);
3067 return DM_MAPIO_SUBMITTED;
3070 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
3072 r = policy_map(cache->policy, block, false, can_migrate, fast_promotion,
3073 bio, &ool.locker, &lookup_result);
3074 if (r == -EWOULDBLOCK) {
3075 cell_defer(cache, cell, true);
3076 return DM_MAPIO_SUBMITTED;
3079 DMERR_LIMIT("%s: Unexpected return from cache replacement policy: %d",
3080 cache_device_name(cache), r);
3081 cell_defer(cache, cell, false);
3083 return DM_MAPIO_SUBMITTED;
3086 r = DM_MAPIO_REMAPPED;
3087 switch (lookup_result.op) {
3089 if (passthrough_mode(&cache->features)) {
3090 if (bio_data_dir(bio) == WRITE) {
3092 * We need to invalidate this block, so
3093 * defer for the worker thread.
3095 cell_defer(cache, cell, true);
3096 r = DM_MAPIO_SUBMITTED;
3099 inc_miss_counter(cache, bio);
3100 remap_to_origin_clear_discard(cache, bio, block);
3101 accounted_begin(cache, bio);
3102 inc_ds(cache, bio, cell);
3103 // FIXME: we want to remap hits or misses straight
3104 // away rather than passing over to the worker.
3105 cell_defer(cache, cell, false);
3109 inc_hit_counter(cache, bio);
3110 if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) &&
3111 !is_dirty(cache, lookup_result.cblock)) {
3112 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
3113 accounted_begin(cache, bio);
3114 inc_ds(cache, bio, cell);
3115 cell_defer(cache, cell, false);
3118 remap_cell_to_cache_dirty(cache, cell, block, lookup_result.cblock, false);
3123 inc_miss_counter(cache, bio);
3124 if (pb->req_nr != 0) {
3126 * This is a duplicate writethrough io that is no
3127 * longer needed because the block has been demoted.
3130 // FIXME: remap everything as a miss
3131 cell_defer(cache, cell, false);
3132 r = DM_MAPIO_SUBMITTED;
3135 remap_cell_to_origin_clear_discard(cache, cell, block, false);
3139 DMERR_LIMIT("%s: %s: erroring bio: unknown policy op: %u",
3140 cache_device_name(cache), __func__,
3141 (unsigned) lookup_result.op);
3142 cell_defer(cache, cell, false);
3144 r = DM_MAPIO_SUBMITTED;
3150 static int cache_end_io(struct dm_target *ti, struct bio *bio, int error)
3152 struct cache *cache = ti->private;
3153 unsigned long flags;
3154 size_t pb_data_size = get_per_bio_data_size(cache);
3155 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
3158 policy_tick(cache->policy, false);
3160 spin_lock_irqsave(&cache->lock, flags);
3161 cache->need_tick_bio = true;
3162 spin_unlock_irqrestore(&cache->lock, flags);
3165 check_for_quiesced_migrations(cache, pb);
3166 accounted_complete(cache, bio);
3171 static int write_dirty_bitset(struct cache *cache)
3175 if (get_cache_mode(cache) >= CM_READ_ONLY)
3178 r = dm_cache_set_dirty_bits(cache->cmd, from_cblock(cache->cache_size), cache->dirty_bitset);
3180 metadata_operation_failed(cache, "dm_cache_set_dirty_bits", r);
3185 static int write_discard_bitset(struct cache *cache)
3189 if (get_cache_mode(cache) >= CM_READ_ONLY)
3192 r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size,
3193 cache->discard_nr_blocks);
3195 DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache));
3196 metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r);
3200 for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) {
3201 r = dm_cache_set_discard(cache->cmd, to_dblock(i),
3202 is_discarded(cache, to_dblock(i)));
3204 metadata_operation_failed(cache, "dm_cache_set_discard", r);
3212 static int write_hints(struct cache *cache)
3216 if (get_cache_mode(cache) >= CM_READ_ONLY)
3219 r = dm_cache_write_hints(cache->cmd, cache->policy);
3221 metadata_operation_failed(cache, "dm_cache_write_hints", r);
3229 * returns true on success
3231 static bool sync_metadata(struct cache *cache)
3235 r1 = write_dirty_bitset(cache);
3237 DMERR("%s: could not write dirty bitset", cache_device_name(cache));
3239 r2 = write_discard_bitset(cache);
3241 DMERR("%s: could not write discard bitset", cache_device_name(cache));
3245 r3 = write_hints(cache);
3247 DMERR("%s: could not write hints", cache_device_name(cache));
3250 * If writing the above metadata failed, we still commit, but don't
3251 * set the clean shutdown flag. This will effectively force every
3252 * dirty bit to be set on reload.
3254 r4 = commit(cache, !r1 && !r2 && !r3);
3256 DMERR("%s: could not write cache metadata", cache_device_name(cache));
3258 return !r1 && !r2 && !r3 && !r4;
3261 static void cache_postsuspend(struct dm_target *ti)
3263 struct cache *cache = ti->private;
3265 start_quiescing(cache);
3266 wait_for_migrations(cache);
3268 requeue_deferred_bios(cache);
3269 requeue_deferred_cells(cache);
3270 stop_quiescing(cache);
3272 if (get_cache_mode(cache) == CM_WRITE)
3273 (void) sync_metadata(cache);
3276 static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock,
3277 bool dirty, uint32_t hint, bool hint_valid)
3280 struct cache *cache = context;
3282 r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid);
3287 set_dirty(cache, oblock, cblock);
3289 clear_dirty(cache, oblock, cblock);
3295 * The discard block size in the on disk metadata is not
3296 * neccessarily the same as we're currently using. So we have to
3297 * be careful to only set the discarded attribute if we know it
3298 * covers a complete block of the new size.
3300 struct discard_load_info {
3301 struct cache *cache;
3304 * These blocks are sized using the on disk dblock size, rather
3305 * than the current one.
3307 dm_block_t block_size;
3308 dm_block_t discard_begin, discard_end;
3311 static void discard_load_info_init(struct cache *cache,
3312 struct discard_load_info *li)
3315 li->discard_begin = li->discard_end = 0;
3318 static void set_discard_range(struct discard_load_info *li)
3322 if (li->discard_begin == li->discard_end)
3326 * Convert to sectors.
3328 b = li->discard_begin * li->block_size;
3329 e = li->discard_end * li->block_size;
3332 * Then convert back to the current dblock size.
3334 b = dm_sector_div_up(b, li->cache->discard_block_size);
3335 sector_div(e, li->cache->discard_block_size);
3338 * The origin may have shrunk, so we need to check we're still in
3341 if (e > from_dblock(li->cache->discard_nr_blocks))
3342 e = from_dblock(li->cache->discard_nr_blocks);
3345 set_discard(li->cache, to_dblock(b));
3348 static int load_discard(void *context, sector_t discard_block_size,
3349 dm_dblock_t dblock, bool discard)
3351 struct discard_load_info *li = context;
3353 li->block_size = discard_block_size;
3356 if (from_dblock(dblock) == li->discard_end)
3358 * We're already in a discard range, just extend it.
3360 li->discard_end = li->discard_end + 1ULL;
3364 * Emit the old range and start a new one.
3366 set_discard_range(li);
3367 li->discard_begin = from_dblock(dblock);
3368 li->discard_end = li->discard_begin + 1ULL;
3371 set_discard_range(li);
3372 li->discard_begin = li->discard_end = 0;
3378 static dm_cblock_t get_cache_dev_size(struct cache *cache)
3380 sector_t size = get_dev_size(cache->cache_dev);
3381 (void) sector_div(size, cache->sectors_per_block);
3382 return to_cblock(size);
3385 static bool can_resize(struct cache *cache, dm_cblock_t new_size)
3387 if (from_cblock(new_size) > from_cblock(cache->cache_size))
3391 * We can't drop a dirty block when shrinking the cache.
3393 while (from_cblock(new_size) < from_cblock(cache->cache_size)) {
3394 new_size = to_cblock(from_cblock(new_size) + 1);
3395 if (is_dirty(cache, new_size)) {
3396 DMERR("%s: unable to shrink cache; cache block %llu is dirty",
3397 cache_device_name(cache),
3398 (unsigned long long) from_cblock(new_size));
3406 static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size)
3410 r = dm_cache_resize(cache->cmd, new_size);
3412 DMERR("%s: could not resize cache metadata", cache_device_name(cache));
3413 metadata_operation_failed(cache, "dm_cache_resize", r);
3417 set_cache_size(cache, new_size);
3422 static int cache_preresume(struct dm_target *ti)
3425 struct cache *cache = ti->private;
3426 dm_cblock_t csize = get_cache_dev_size(cache);
3429 * Check to see if the cache has resized.
3431 if (!cache->sized) {
3432 r = resize_cache_dev(cache, csize);
3436 cache->sized = true;
3438 } else if (csize != cache->cache_size) {
3439 if (!can_resize(cache, csize))
3442 r = resize_cache_dev(cache, csize);
3447 if (!cache->loaded_mappings) {
3448 r = dm_cache_load_mappings(cache->cmd, cache->policy,
3449 load_mapping, cache);
3451 DMERR("%s: could not load cache mappings", cache_device_name(cache));
3452 metadata_operation_failed(cache, "dm_cache_load_mappings", r);
3456 cache->loaded_mappings = true;
3459 if (!cache->loaded_discards) {
3460 struct discard_load_info li;
3463 * The discard bitset could have been resized, or the
3464 * discard block size changed. To be safe we start by
3465 * setting every dblock to not discarded.
3467 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
3469 discard_load_info_init(cache, &li);
3470 r = dm_cache_load_discards(cache->cmd, load_discard, &li);
3472 DMERR("%s: could not load origin discards", cache_device_name(cache));
3473 metadata_operation_failed(cache, "dm_cache_load_discards", r);
3476 set_discard_range(&li);
3478 cache->loaded_discards = true;
3484 static void cache_resume(struct dm_target *ti)
3486 struct cache *cache = ti->private;
3488 cache->need_tick_bio = true;
3489 do_waker(&cache->waker.work);
3495 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
3496 * <cache block size> <#used cache blocks>/<#total cache blocks>
3497 * <#read hits> <#read misses> <#write hits> <#write misses>
3498 * <#demotions> <#promotions> <#dirty>
3499 * <#features> <features>*
3500 * <#core args> <core args>
3501 * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check>
3503 static void cache_status(struct dm_target *ti, status_type_t type,
3504 unsigned status_flags, char *result, unsigned maxlen)
3509 dm_block_t nr_free_blocks_metadata = 0;
3510 dm_block_t nr_blocks_metadata = 0;
3511 char buf[BDEVNAME_SIZE];
3512 struct cache *cache = ti->private;
3513 dm_cblock_t residency;
3517 case STATUSTYPE_INFO:
3518 if (get_cache_mode(cache) == CM_FAIL) {
3523 /* Commit to ensure statistics aren't out-of-date */
3524 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3525 (void) commit(cache, false);
3527 r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata);
3529 DMERR("%s: dm_cache_get_free_metadata_block_count returned %d",
3530 cache_device_name(cache), r);
3534 r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata);
3536 DMERR("%s: dm_cache_get_metadata_dev_size returned %d",
3537 cache_device_name(cache), r);
3541 residency = policy_residency(cache->policy);
3543 DMEMIT("%u %llu/%llu %llu %llu/%llu %u %u %u %u %u %u %lu ",
3544 (unsigned)DM_CACHE_METADATA_BLOCK_SIZE,
3545 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3546 (unsigned long long)nr_blocks_metadata,
3547 (unsigned long long)cache->sectors_per_block,
3548 (unsigned long long) from_cblock(residency),
3549 (unsigned long long) from_cblock(cache->cache_size),
3550 (unsigned) atomic_read(&cache->stats.read_hit),
3551 (unsigned) atomic_read(&cache->stats.read_miss),
3552 (unsigned) atomic_read(&cache->stats.write_hit),
3553 (unsigned) atomic_read(&cache->stats.write_miss),
3554 (unsigned) atomic_read(&cache->stats.demotion),
3555 (unsigned) atomic_read(&cache->stats.promotion),
3556 (unsigned long) atomic_read(&cache->nr_dirty));
3558 if (cache->features.metadata_version == 2)
3559 DMEMIT("2 metadata2 ");
3563 if (writethrough_mode(&cache->features))
3564 DMEMIT("writethrough ");
3566 else if (passthrough_mode(&cache->features))
3567 DMEMIT("passthrough ");
3569 else if (writeback_mode(&cache->features))
3570 DMEMIT("writeback ");
3573 DMERR("%s: internal error: unknown io mode: %d",
3574 cache_device_name(cache), (int) cache->features.io_mode);
3578 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
3580 DMEMIT("%s ", dm_cache_policy_get_name(cache->policy));
3582 r = policy_emit_config_values(cache->policy, result, maxlen, &sz);
3584 DMERR("%s: policy_emit_config_values returned %d",
3585 cache_device_name(cache), r);
3588 if (get_cache_mode(cache) == CM_READ_ONLY)
3593 r = dm_cache_metadata_needs_check(cache->cmd, &needs_check);
3595 if (r || needs_check)
3596 DMEMIT("needs_check ");
3602 case STATUSTYPE_TABLE:
3603 format_dev_t(buf, cache->metadata_dev->bdev->bd_dev);
3605 format_dev_t(buf, cache->cache_dev->bdev->bd_dev);
3607 format_dev_t(buf, cache->origin_dev->bdev->bd_dev);
3610 for (i = 0; i < cache->nr_ctr_args - 1; i++)
3611 DMEMIT(" %s", cache->ctr_args[i]);
3612 if (cache->nr_ctr_args)
3613 DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]);
3623 * A cache block range can take two forms:
3625 * i) A single cblock, eg. '3456'
3626 * ii) A begin and end cblock with dots between, eg. 123-234
3628 static int parse_cblock_range(struct cache *cache, const char *str,
3629 struct cblock_range *result)
3636 * Try and parse form (ii) first.
3638 r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy);
3643 result->begin = to_cblock(b);
3644 result->end = to_cblock(e);
3649 * That didn't work, try form (i).
3651 r = sscanf(str, "%llu%c", &b, &dummy);
3656 result->begin = to_cblock(b);
3657 result->end = to_cblock(from_cblock(result->begin) + 1u);
3661 DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str);
3665 static int validate_cblock_range(struct cache *cache, struct cblock_range *range)
3667 uint64_t b = from_cblock(range->begin);
3668 uint64_t e = from_cblock(range->end);
3669 uint64_t n = from_cblock(cache->cache_size);
3672 DMERR("%s: begin cblock out of range: %llu >= %llu",
3673 cache_device_name(cache), b, n);
3678 DMERR("%s: end cblock out of range: %llu > %llu",
3679 cache_device_name(cache), e, n);
3684 DMERR("%s: invalid cblock range: %llu >= %llu",
3685 cache_device_name(cache), b, e);
3692 static int request_invalidation(struct cache *cache, struct cblock_range *range)
3694 struct invalidation_request req;
3696 INIT_LIST_HEAD(&req.list);
3697 req.cblocks = range;
3698 atomic_set(&req.complete, 0);
3700 init_waitqueue_head(&req.result_wait);
3702 spin_lock(&cache->invalidation_lock);
3703 list_add(&req.list, &cache->invalidation_requests);
3704 spin_unlock(&cache->invalidation_lock);
3707 wait_event(req.result_wait, atomic_read(&req.complete));
3711 static int process_invalidate_cblocks_message(struct cache *cache, unsigned count,
3712 const char **cblock_ranges)
3716 struct cblock_range range;
3718 if (!passthrough_mode(&cache->features)) {
3719 DMERR("%s: cache has to be in passthrough mode for invalidation",
3720 cache_device_name(cache));
3724 for (i = 0; i < count; i++) {
3725 r = parse_cblock_range(cache, cblock_ranges[i], &range);
3729 r = validate_cblock_range(cache, &range);
3734 * Pass begin and end origin blocks to the worker and wake it.
3736 r = request_invalidation(cache, &range);
3748 * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]*
3750 * The key migration_threshold is supported by the cache target core.
3752 static int cache_message(struct dm_target *ti, unsigned argc, char **argv)
3754 struct cache *cache = ti->private;
3759 if (get_cache_mode(cache) >= CM_READ_ONLY) {
3760 DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode",
3761 cache_device_name(cache));
3765 if (!strcasecmp(argv[0], "invalidate_cblocks"))
3766 return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1);
3771 return set_config_value(cache, argv[0], argv[1]);
3774 static int cache_iterate_devices(struct dm_target *ti,
3775 iterate_devices_callout_fn fn, void *data)
3778 struct cache *cache = ti->private;
3780 r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data);
3782 r = fn(ti, cache->origin_dev, 0, ti->len, data);
3787 static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
3790 * FIXME: these limits may be incompatible with the cache device
3792 limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024,
3793 cache->origin_sectors);
3794 limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
3797 static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
3799 struct cache *cache = ti->private;
3800 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
3803 * If the system-determined stacked limits are compatible with the
3804 * cache's blocksize (io_opt is a factor) do not override them.
3806 if (io_opt_sectors < cache->sectors_per_block ||
3807 do_div(io_opt_sectors, cache->sectors_per_block)) {
3808 blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT);
3809 blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
3811 set_discard_limits(cache, limits);
3814 /*----------------------------------------------------------------*/
3816 static struct target_type cache_target = {
3818 .version = {1, 10, 0},
3819 .module = THIS_MODULE,
3823 .end_io = cache_end_io,
3824 .postsuspend = cache_postsuspend,
3825 .preresume = cache_preresume,
3826 .resume = cache_resume,
3827 .status = cache_status,
3828 .message = cache_message,
3829 .iterate_devices = cache_iterate_devices,
3830 .io_hints = cache_io_hints,
3833 static int __init dm_cache_init(void)
3837 r = dm_register_target(&cache_target);
3839 DMERR("cache target registration failed: %d", r);
3843 migration_cache = KMEM_CACHE(dm_cache_migration, 0);
3844 if (!migration_cache) {
3845 dm_unregister_target(&cache_target);
3852 static void __exit dm_cache_exit(void)
3854 dm_unregister_target(&cache_target);
3855 kmem_cache_destroy(migration_cache);
3858 module_init(dm_cache_init);
3859 module_exit(dm_cache_exit);
3861 MODULE_DESCRIPTION(DM_NAME " cache target");
3862 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
3863 MODULE_LICENSE("GPL");