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);
430 static int prealloc_data_structs(struct cache *cache, struct prealloc *p)
433 p->mg = alloc_migration(cache);
439 p->cell1 = alloc_prison_cell(cache);
445 p->cell2 = alloc_prison_cell(cache);
453 static void prealloc_free_structs(struct cache *cache, struct prealloc *p)
456 free_prison_cell(cache, p->cell2);
459 free_prison_cell(cache, p->cell1);
462 free_migration(p->mg);
465 static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p)
467 struct dm_cache_migration *mg = p->mg;
476 * You must have a cell within the prealloc struct to return. If not this
477 * function will BUG() rather than returning NULL.
479 static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p)
481 struct dm_bio_prison_cell *r = NULL;
487 } else if (p->cell2) {
497 * You can't have more than two cells in a prealloc struct. BUG() will be
498 * called if you try and overfill.
500 static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell)
512 /*----------------------------------------------------------------*/
514 static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key *key)
518 key->block_begin = from_oblock(begin);
519 key->block_end = from_oblock(end);
523 * The caller hands in a preallocated cell, and a free function for it.
524 * The cell will be freed if there's an error, or if it wasn't used because
525 * a cell with that key already exists.
527 typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell);
529 static int bio_detain_range(struct cache *cache, dm_oblock_t oblock_begin, dm_oblock_t oblock_end,
530 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
531 cell_free_fn free_fn, void *free_context,
532 struct dm_bio_prison_cell **cell_result)
535 struct dm_cell_key key;
537 build_key(oblock_begin, oblock_end, &key);
538 r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result);
540 free_fn(free_context, cell_prealloc);
545 static int bio_detain(struct cache *cache, dm_oblock_t oblock,
546 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
547 cell_free_fn free_fn, void *free_context,
548 struct dm_bio_prison_cell **cell_result)
550 dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL);
551 return bio_detain_range(cache, oblock, end, bio,
552 cell_prealloc, free_fn, free_context, cell_result);
555 static int get_cell(struct cache *cache,
557 struct prealloc *structs,
558 struct dm_bio_prison_cell **cell_result)
561 struct dm_cell_key key;
562 struct dm_bio_prison_cell *cell_prealloc;
564 cell_prealloc = prealloc_get_cell(structs);
566 build_key(oblock, to_oblock(from_oblock(oblock) + 1ULL), &key);
567 r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result);
569 prealloc_put_cell(structs, cell_prealloc);
574 /*----------------------------------------------------------------*/
576 static bool is_dirty(struct cache *cache, dm_cblock_t b)
578 return test_bit(from_cblock(b), cache->dirty_bitset);
581 static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
583 if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) {
584 atomic_inc(&cache->nr_dirty);
585 policy_set_dirty(cache->policy, oblock);
589 static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
591 if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) {
592 policy_clear_dirty(cache->policy, oblock);
593 if (atomic_dec_return(&cache->nr_dirty) == 0)
594 dm_table_event(cache->ti->table);
598 /*----------------------------------------------------------------*/
600 static bool block_size_is_power_of_two(struct cache *cache)
602 return cache->sectors_per_block_shift >= 0;
605 /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */
606 #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6
609 static dm_block_t block_div(dm_block_t b, uint32_t n)
616 static dm_block_t oblocks_per_dblock(struct cache *cache)
618 dm_block_t oblocks = cache->discard_block_size;
620 if (block_size_is_power_of_two(cache))
621 oblocks >>= cache->sectors_per_block_shift;
623 oblocks = block_div(oblocks, cache->sectors_per_block);
628 static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock)
630 return to_dblock(block_div(from_oblock(oblock),
631 oblocks_per_dblock(cache)));
634 static dm_oblock_t dblock_to_oblock(struct cache *cache, dm_dblock_t dblock)
636 return to_oblock(from_dblock(dblock) * oblocks_per_dblock(cache));
639 static void set_discard(struct cache *cache, dm_dblock_t b)
643 BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks));
644 atomic_inc(&cache->stats.discard_count);
646 spin_lock_irqsave(&cache->lock, flags);
647 set_bit(from_dblock(b), cache->discard_bitset);
648 spin_unlock_irqrestore(&cache->lock, flags);
651 static void clear_discard(struct cache *cache, dm_dblock_t b)
655 spin_lock_irqsave(&cache->lock, flags);
656 clear_bit(from_dblock(b), cache->discard_bitset);
657 spin_unlock_irqrestore(&cache->lock, flags);
660 static bool is_discarded(struct cache *cache, dm_dblock_t b)
665 spin_lock_irqsave(&cache->lock, flags);
666 r = test_bit(from_dblock(b), cache->discard_bitset);
667 spin_unlock_irqrestore(&cache->lock, flags);
672 static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b)
677 spin_lock_irqsave(&cache->lock, flags);
678 r = test_bit(from_dblock(oblock_to_dblock(cache, b)),
679 cache->discard_bitset);
680 spin_unlock_irqrestore(&cache->lock, flags);
685 /*----------------------------------------------------------------*/
687 static void load_stats(struct cache *cache)
689 struct dm_cache_statistics stats;
691 dm_cache_metadata_get_stats(cache->cmd, &stats);
692 atomic_set(&cache->stats.read_hit, stats.read_hits);
693 atomic_set(&cache->stats.read_miss, stats.read_misses);
694 atomic_set(&cache->stats.write_hit, stats.write_hits);
695 atomic_set(&cache->stats.write_miss, stats.write_misses);
698 static void save_stats(struct cache *cache)
700 struct dm_cache_statistics stats;
702 if (get_cache_mode(cache) >= CM_READ_ONLY)
705 stats.read_hits = atomic_read(&cache->stats.read_hit);
706 stats.read_misses = atomic_read(&cache->stats.read_miss);
707 stats.write_hits = atomic_read(&cache->stats.write_hit);
708 stats.write_misses = atomic_read(&cache->stats.write_miss);
710 dm_cache_metadata_set_stats(cache->cmd, &stats);
713 /*----------------------------------------------------------------
715 *--------------------------------------------------------------*/
718 * If using writeback, leave out struct per_bio_data's writethrough fields.
720 #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache))
721 #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data))
723 static bool writethrough_mode(struct cache_features *f)
725 return f->io_mode == CM_IO_WRITETHROUGH;
728 static bool writeback_mode(struct cache_features *f)
730 return f->io_mode == CM_IO_WRITEBACK;
733 static bool passthrough_mode(struct cache_features *f)
735 return f->io_mode == CM_IO_PASSTHROUGH;
738 static size_t get_per_bio_data_size(struct cache *cache)
740 return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB;
743 static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size)
745 struct per_bio_data *pb = dm_per_bio_data(bio, data_size);
750 static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size)
752 struct per_bio_data *pb = get_per_bio_data(bio, data_size);
755 pb->req_nr = dm_bio_get_target_bio_nr(bio);
756 pb->all_io_entry = NULL;
762 /*----------------------------------------------------------------
764 *--------------------------------------------------------------*/
765 static void remap_to_origin(struct cache *cache, struct bio *bio)
767 bio->bi_bdev = cache->origin_dev->bdev;
770 static void remap_to_cache(struct cache *cache, struct bio *bio,
773 sector_t bi_sector = bio->bi_iter.bi_sector;
774 sector_t block = from_cblock(cblock);
776 bio->bi_bdev = cache->cache_dev->bdev;
777 if (!block_size_is_power_of_two(cache))
778 bio->bi_iter.bi_sector =
779 (block * cache->sectors_per_block) +
780 sector_div(bi_sector, cache->sectors_per_block);
782 bio->bi_iter.bi_sector =
783 (block << cache->sectors_per_block_shift) |
784 (bi_sector & (cache->sectors_per_block - 1));
787 static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio)
790 size_t pb_data_size = get_per_bio_data_size(cache);
791 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
793 spin_lock_irqsave(&cache->lock, flags);
794 if (cache->need_tick_bio &&
795 !(bio->bi_rw & (REQ_FUA | REQ_FLUSH | REQ_DISCARD))) {
797 cache->need_tick_bio = false;
799 spin_unlock_irqrestore(&cache->lock, flags);
802 static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio,
805 check_if_tick_bio_needed(cache, bio);
806 remap_to_origin(cache, bio);
807 if (bio_data_dir(bio) == WRITE)
808 clear_discard(cache, oblock_to_dblock(cache, oblock));
811 static void remap_to_cache_dirty(struct cache *cache, struct bio *bio,
812 dm_oblock_t oblock, dm_cblock_t cblock)
814 check_if_tick_bio_needed(cache, bio);
815 remap_to_cache(cache, bio, cblock);
816 if (bio_data_dir(bio) == WRITE) {
817 set_dirty(cache, oblock, cblock);
818 clear_discard(cache, oblock_to_dblock(cache, oblock));
822 static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio)
824 sector_t block_nr = bio->bi_iter.bi_sector;
826 if (!block_size_is_power_of_two(cache))
827 (void) sector_div(block_nr, cache->sectors_per_block);
829 block_nr >>= cache->sectors_per_block_shift;
831 return to_oblock(block_nr);
834 static int bio_triggers_commit(struct cache *cache, struct bio *bio)
836 return bio->bi_rw & (REQ_FLUSH | REQ_FUA);
840 * You must increment the deferred set whilst the prison cell is held. To
841 * encourage this, we ask for 'cell' to be passed in.
843 static void inc_ds(struct cache *cache, struct bio *bio,
844 struct dm_bio_prison_cell *cell)
846 size_t pb_data_size = get_per_bio_data_size(cache);
847 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
850 BUG_ON(pb->all_io_entry);
852 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
855 static bool accountable_bio(struct cache *cache, struct bio *bio)
857 return ((bio->bi_bdev == cache->origin_dev->bdev) &&
858 !(bio->bi_rw & REQ_DISCARD));
861 static void accounted_begin(struct cache *cache, struct bio *bio)
863 size_t pb_data_size = get_per_bio_data_size(cache);
864 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
866 if (accountable_bio(cache, bio)) {
867 pb->len = bio_sectors(bio);
868 iot_io_begin(&cache->origin_tracker, pb->len);
872 static void accounted_complete(struct cache *cache, struct bio *bio)
874 size_t pb_data_size = get_per_bio_data_size(cache);
875 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
877 iot_io_end(&cache->origin_tracker, pb->len);
880 static void accounted_request(struct cache *cache, struct bio *bio)
882 accounted_begin(cache, bio);
883 generic_make_request(bio);
886 static void issue(struct cache *cache, struct bio *bio)
890 if (!bio_triggers_commit(cache, bio)) {
891 accounted_request(cache, bio);
896 * Batch together any bios that trigger commits and then issue a
897 * single commit for them in do_worker().
899 spin_lock_irqsave(&cache->lock, flags);
900 cache->commit_requested = true;
901 bio_list_add(&cache->deferred_flush_bios, bio);
902 spin_unlock_irqrestore(&cache->lock, flags);
905 static void inc_and_issue(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell *cell)
907 inc_ds(cache, bio, cell);
911 static void defer_writethrough_bio(struct cache *cache, struct bio *bio)
915 spin_lock_irqsave(&cache->lock, flags);
916 bio_list_add(&cache->deferred_writethrough_bios, bio);
917 spin_unlock_irqrestore(&cache->lock, flags);
922 static void writethrough_endio(struct bio *bio, int err)
924 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
926 dm_unhook_bio(&pb->hook_info, bio);
933 dm_bio_restore(&pb->bio_details, bio);
934 remap_to_cache(pb->cache, bio, pb->cblock);
937 * We can't issue this bio directly, since we're in interrupt
938 * context. So it gets put on a bio list for processing by the
941 defer_writethrough_bio(pb->cache, bio);
945 * When running in writethrough mode we need to send writes to clean blocks
946 * to both the cache and origin devices. In future we'd like to clone the
947 * bio and send them in parallel, but for now we're doing them in
948 * series as this is easier.
950 static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio,
951 dm_oblock_t oblock, dm_cblock_t cblock)
953 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
957 dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL);
958 dm_bio_record(&pb->bio_details, bio);
960 remap_to_origin_clear_discard(pb->cache, bio, oblock);
963 /*----------------------------------------------------------------
965 *--------------------------------------------------------------*/
966 static enum cache_metadata_mode get_cache_mode(struct cache *cache)
968 return cache->features.mode;
971 static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode)
973 const char *descs[] = {
979 dm_table_event(cache->ti->table);
980 DMINFO("switching cache to %s mode", descs[(int)mode]);
983 static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode)
985 bool needs_check = dm_cache_metadata_needs_check(cache->cmd);
986 enum cache_metadata_mode old_mode = get_cache_mode(cache);
988 if (new_mode == CM_WRITE && needs_check) {
989 DMERR("unable to switch cache to write mode until repaired.");
990 if (old_mode != new_mode)
993 new_mode = CM_READ_ONLY;
996 /* Never move out of fail mode */
997 if (old_mode == CM_FAIL)
1003 dm_cache_metadata_set_read_only(cache->cmd);
1007 dm_cache_metadata_set_read_write(cache->cmd);
1011 cache->features.mode = new_mode;
1013 if (new_mode != old_mode)
1014 notify_mode_switch(cache, new_mode);
1017 static void abort_transaction(struct cache *cache)
1019 if (get_cache_mode(cache) >= CM_READ_ONLY)
1022 if (dm_cache_metadata_set_needs_check(cache->cmd)) {
1023 DMERR("failed to set 'needs_check' flag in metadata");
1024 set_cache_mode(cache, CM_FAIL);
1027 DMERR_LIMIT("aborting current metadata transaction");
1028 if (dm_cache_metadata_abort(cache->cmd)) {
1029 DMERR("failed to abort metadata transaction");
1030 set_cache_mode(cache, CM_FAIL);
1034 static void metadata_operation_failed(struct cache *cache, const char *op, int r)
1036 DMERR_LIMIT("metadata operation '%s' failed: error = %d", op, r);
1037 abort_transaction(cache);
1038 set_cache_mode(cache, CM_READ_ONLY);
1041 /*----------------------------------------------------------------
1042 * Migration processing
1044 * Migration covers moving data from the origin device to the cache, or
1046 *--------------------------------------------------------------*/
1047 static void inc_io_migrations(struct cache *cache)
1049 atomic_inc(&cache->nr_io_migrations);
1052 static void dec_io_migrations(struct cache *cache)
1054 atomic_dec(&cache->nr_io_migrations);
1057 static void __cell_release(struct cache *cache, struct dm_bio_prison_cell *cell,
1058 bool holder, struct bio_list *bios)
1060 (holder ? dm_cell_release : dm_cell_release_no_holder)
1061 (cache->prison, cell, bios);
1062 free_prison_cell(cache, cell);
1065 static bool discard_or_flush(struct bio *bio)
1067 return bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_DISCARD);
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
1077 __cell_release(cache, cell, true, &cache->deferred_bios);
1080 list_add_tail(&cell->user_list, &cache->deferred_cells);
1083 static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, bool holder)
1085 unsigned long flags;
1087 if (!holder && dm_cell_promote_or_release(cache->prison, cell)) {
1089 * There was no prisoner to promote to holder, the
1090 * cell has been released.
1092 free_prison_cell(cache, cell);
1096 spin_lock_irqsave(&cache->lock, flags);
1097 __cell_defer(cache, cell);
1098 spin_unlock_irqrestore(&cache->lock, flags);
1103 static void cell_error_with_code(struct cache *cache, struct dm_bio_prison_cell *cell, int err)
1105 dm_cell_error(cache->prison, cell, err);
1106 dm_bio_prison_free_cell(cache->prison, cell);
1109 static void cell_requeue(struct cache *cache, struct dm_bio_prison_cell *cell)
1111 cell_error_with_code(cache, cell, DM_ENDIO_REQUEUE);
1114 static void free_io_migration(struct dm_cache_migration *mg)
1116 dec_io_migrations(mg->cache);
1120 static void migration_failure(struct dm_cache_migration *mg)
1122 struct cache *cache = mg->cache;
1124 if (mg->writeback) {
1125 DMWARN_LIMIT("writeback failed; couldn't copy block");
1126 set_dirty(cache, mg->old_oblock, mg->cblock);
1127 cell_defer(cache, mg->old_ocell, false);
1129 } else if (mg->demote) {
1130 DMWARN_LIMIT("demotion failed; couldn't copy block");
1131 policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock);
1133 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1135 cell_defer(cache, mg->new_ocell, true);
1137 DMWARN_LIMIT("promotion failed; couldn't copy block");
1138 policy_remove_mapping(cache->policy, mg->new_oblock);
1139 cell_defer(cache, mg->new_ocell, true);
1142 free_io_migration(mg);
1145 static void migration_success_pre_commit(struct dm_cache_migration *mg)
1148 unsigned long flags;
1149 struct cache *cache = mg->cache;
1151 if (mg->writeback) {
1152 clear_dirty(cache, mg->old_oblock, mg->cblock);
1153 cell_defer(cache, mg->old_ocell, false);
1154 free_io_migration(mg);
1157 } else if (mg->demote) {
1158 r = dm_cache_remove_mapping(cache->cmd, mg->cblock);
1160 DMWARN_LIMIT("demotion failed; couldn't update on disk metadata");
1161 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
1162 policy_force_mapping(cache->policy, mg->new_oblock,
1165 cell_defer(cache, mg->new_ocell, true);
1166 free_io_migration(mg);
1170 r = dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock);
1172 DMWARN_LIMIT("promotion failed; couldn't update on disk metadata");
1173 metadata_operation_failed(cache, "dm_cache_insert_mapping", r);
1174 policy_remove_mapping(cache->policy, mg->new_oblock);
1175 free_io_migration(mg);
1180 spin_lock_irqsave(&cache->lock, flags);
1181 list_add_tail(&mg->list, &cache->need_commit_migrations);
1182 cache->commit_requested = true;
1183 spin_unlock_irqrestore(&cache->lock, flags);
1186 static void migration_success_post_commit(struct dm_cache_migration *mg)
1188 unsigned long flags;
1189 struct cache *cache = mg->cache;
1191 if (mg->writeback) {
1192 DMWARN("writeback unexpectedly triggered commit");
1195 } else if (mg->demote) {
1196 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
1201 spin_lock_irqsave(&cache->lock, flags);
1202 list_add_tail(&mg->list, &cache->quiesced_migrations);
1203 spin_unlock_irqrestore(&cache->lock, flags);
1207 policy_remove_mapping(cache->policy, mg->old_oblock);
1208 free_io_migration(mg);
1212 if (mg->requeue_holder) {
1213 clear_dirty(cache, mg->new_oblock, mg->cblock);
1214 cell_defer(cache, mg->new_ocell, true);
1217 * The block was promoted via an overwrite, so it's dirty.
1219 set_dirty(cache, mg->new_oblock, mg->cblock);
1220 bio_endio(mg->new_ocell->holder, 0);
1221 cell_defer(cache, mg->new_ocell, false);
1223 free_io_migration(mg);
1227 static void copy_complete(int read_err, unsigned long write_err, void *context)
1229 unsigned long flags;
1230 struct dm_cache_migration *mg = (struct dm_cache_migration *) context;
1231 struct cache *cache = mg->cache;
1233 if (read_err || write_err)
1236 spin_lock_irqsave(&cache->lock, flags);
1237 list_add_tail(&mg->list, &cache->completed_migrations);
1238 spin_unlock_irqrestore(&cache->lock, flags);
1243 static void issue_copy(struct dm_cache_migration *mg)
1246 struct dm_io_region o_region, c_region;
1247 struct cache *cache = mg->cache;
1248 sector_t cblock = from_cblock(mg->cblock);
1250 o_region.bdev = cache->origin_dev->bdev;
1251 o_region.count = cache->sectors_per_block;
1253 c_region.bdev = cache->cache_dev->bdev;
1254 c_region.sector = cblock * cache->sectors_per_block;
1255 c_region.count = cache->sectors_per_block;
1257 if (mg->writeback || mg->demote) {
1259 o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block;
1260 r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg);
1263 o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block;
1264 r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg);
1268 DMERR_LIMIT("issuing migration failed");
1269 migration_failure(mg);
1273 static void overwrite_endio(struct bio *bio, int err)
1275 struct dm_cache_migration *mg = bio->bi_private;
1276 struct cache *cache = mg->cache;
1277 size_t pb_data_size = get_per_bio_data_size(cache);
1278 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1279 unsigned long flags;
1281 dm_unhook_bio(&pb->hook_info, bio);
1286 mg->requeue_holder = false;
1288 spin_lock_irqsave(&cache->lock, flags);
1289 list_add_tail(&mg->list, &cache->completed_migrations);
1290 spin_unlock_irqrestore(&cache->lock, flags);
1295 static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio)
1297 size_t pb_data_size = get_per_bio_data_size(mg->cache);
1298 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1300 dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg);
1301 remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock);
1304 * No need to inc_ds() here, since the cell will be held for the
1305 * duration of the io.
1307 accounted_request(mg->cache, bio);
1310 static bool bio_writes_complete_block(struct cache *cache, struct bio *bio)
1312 return (bio_data_dir(bio) == WRITE) &&
1313 (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT));
1316 static void avoid_copy(struct dm_cache_migration *mg)
1318 atomic_inc(&mg->cache->stats.copies_avoided);
1319 migration_success_pre_commit(mg);
1322 static void calc_discard_block_range(struct cache *cache, struct bio *bio,
1323 dm_dblock_t *b, dm_dblock_t *e)
1325 sector_t sb = bio->bi_iter.bi_sector;
1326 sector_t se = bio_end_sector(bio);
1328 *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size));
1330 if (se - sb < cache->discard_block_size)
1333 *e = to_dblock(block_div(se, cache->discard_block_size));
1336 static void issue_discard(struct dm_cache_migration *mg)
1339 struct bio *bio = mg->new_ocell->holder;
1341 calc_discard_block_range(mg->cache, bio, &b, &e);
1343 set_discard(mg->cache, b);
1344 b = to_dblock(from_dblock(b) + 1);
1348 cell_defer(mg->cache, mg->new_ocell, false);
1352 static void issue_copy_or_discard(struct dm_cache_migration *mg)
1355 struct cache *cache = mg->cache;
1362 if (mg->writeback || mg->demote)
1363 avoid = !is_dirty(cache, mg->cblock) ||
1364 is_discarded_oblock(cache, mg->old_oblock);
1366 struct bio *bio = mg->new_ocell->holder;
1368 avoid = is_discarded_oblock(cache, mg->new_oblock);
1370 if (writeback_mode(&cache->features) &&
1371 !avoid && bio_writes_complete_block(cache, bio)) {
1372 issue_overwrite(mg, bio);
1377 avoid ? avoid_copy(mg) : issue_copy(mg);
1380 static void complete_migration(struct dm_cache_migration *mg)
1383 migration_failure(mg);
1385 migration_success_pre_commit(mg);
1388 static void process_migrations(struct cache *cache, struct list_head *head,
1389 void (*fn)(struct dm_cache_migration *))
1391 unsigned long flags;
1392 struct list_head list;
1393 struct dm_cache_migration *mg, *tmp;
1395 INIT_LIST_HEAD(&list);
1396 spin_lock_irqsave(&cache->lock, flags);
1397 list_splice_init(head, &list);
1398 spin_unlock_irqrestore(&cache->lock, flags);
1400 list_for_each_entry_safe(mg, tmp, &list, list)
1404 static void __queue_quiesced_migration(struct dm_cache_migration *mg)
1406 list_add_tail(&mg->list, &mg->cache->quiesced_migrations);
1409 static void queue_quiesced_migration(struct dm_cache_migration *mg)
1411 unsigned long flags;
1412 struct cache *cache = mg->cache;
1414 spin_lock_irqsave(&cache->lock, flags);
1415 __queue_quiesced_migration(mg);
1416 spin_unlock_irqrestore(&cache->lock, flags);
1421 static void queue_quiesced_migrations(struct cache *cache, struct list_head *work)
1423 unsigned long flags;
1424 struct dm_cache_migration *mg, *tmp;
1426 spin_lock_irqsave(&cache->lock, flags);
1427 list_for_each_entry_safe(mg, tmp, work, list)
1428 __queue_quiesced_migration(mg);
1429 spin_unlock_irqrestore(&cache->lock, flags);
1434 static void check_for_quiesced_migrations(struct cache *cache,
1435 struct per_bio_data *pb)
1437 struct list_head work;
1439 if (!pb->all_io_entry)
1442 INIT_LIST_HEAD(&work);
1443 dm_deferred_entry_dec(pb->all_io_entry, &work);
1445 if (!list_empty(&work))
1446 queue_quiesced_migrations(cache, &work);
1449 static void quiesce_migration(struct dm_cache_migration *mg)
1451 if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list))
1452 queue_quiesced_migration(mg);
1455 static void promote(struct cache *cache, struct prealloc *structs,
1456 dm_oblock_t oblock, dm_cblock_t cblock,
1457 struct dm_bio_prison_cell *cell)
1459 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1462 mg->discard = false;
1463 mg->writeback = false;
1466 mg->requeue_holder = true;
1467 mg->invalidate = false;
1469 mg->new_oblock = oblock;
1470 mg->cblock = cblock;
1471 mg->old_ocell = NULL;
1472 mg->new_ocell = cell;
1473 mg->start_jiffies = jiffies;
1475 inc_io_migrations(cache);
1476 quiesce_migration(mg);
1479 static void writeback(struct cache *cache, struct prealloc *structs,
1480 dm_oblock_t oblock, dm_cblock_t cblock,
1481 struct dm_bio_prison_cell *cell)
1483 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1486 mg->discard = false;
1487 mg->writeback = true;
1489 mg->promote = false;
1490 mg->requeue_holder = true;
1491 mg->invalidate = false;
1493 mg->old_oblock = oblock;
1494 mg->cblock = cblock;
1495 mg->old_ocell = cell;
1496 mg->new_ocell = NULL;
1497 mg->start_jiffies = jiffies;
1499 inc_io_migrations(cache);
1500 quiesce_migration(mg);
1503 static void demote_then_promote(struct cache *cache, struct prealloc *structs,
1504 dm_oblock_t old_oblock, dm_oblock_t new_oblock,
1506 struct dm_bio_prison_cell *old_ocell,
1507 struct dm_bio_prison_cell *new_ocell)
1509 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1512 mg->discard = false;
1513 mg->writeback = false;
1516 mg->requeue_holder = true;
1517 mg->invalidate = false;
1519 mg->old_oblock = old_oblock;
1520 mg->new_oblock = new_oblock;
1521 mg->cblock = cblock;
1522 mg->old_ocell = old_ocell;
1523 mg->new_ocell = new_ocell;
1524 mg->start_jiffies = jiffies;
1526 inc_io_migrations(cache);
1527 quiesce_migration(mg);
1531 * Invalidate a cache entry. No writeback occurs; any changes in the cache
1532 * block are thrown away.
1534 static void invalidate(struct cache *cache, struct prealloc *structs,
1535 dm_oblock_t oblock, dm_cblock_t cblock,
1536 struct dm_bio_prison_cell *cell)
1538 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1541 mg->discard = false;
1542 mg->writeback = false;
1544 mg->promote = false;
1545 mg->requeue_holder = true;
1546 mg->invalidate = true;
1548 mg->old_oblock = oblock;
1549 mg->cblock = cblock;
1550 mg->old_ocell = cell;
1551 mg->new_ocell = NULL;
1552 mg->start_jiffies = jiffies;
1554 inc_io_migrations(cache);
1555 quiesce_migration(mg);
1558 static void discard(struct cache *cache, struct prealloc *structs,
1559 struct dm_bio_prison_cell *cell)
1561 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1565 mg->writeback = false;
1567 mg->promote = false;
1568 mg->requeue_holder = false;
1569 mg->invalidate = false;
1571 mg->old_ocell = NULL;
1572 mg->new_ocell = cell;
1573 mg->start_jiffies = jiffies;
1575 quiesce_migration(mg);
1578 /*----------------------------------------------------------------
1580 *--------------------------------------------------------------*/
1581 static void defer_bio(struct cache *cache, struct bio *bio)
1583 unsigned long flags;
1585 spin_lock_irqsave(&cache->lock, flags);
1586 bio_list_add(&cache->deferred_bios, bio);
1587 spin_unlock_irqrestore(&cache->lock, flags);
1592 static void process_flush_bio(struct cache *cache, struct bio *bio)
1594 size_t pb_data_size = get_per_bio_data_size(cache);
1595 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1597 BUG_ON(bio->bi_iter.bi_size);
1599 remap_to_origin(cache, bio);
1601 remap_to_cache(cache, bio, 0);
1604 * REQ_FLUSH is not directed at any particular block so we don't
1605 * need to inc_ds(). REQ_FUA's are split into a write + REQ_FLUSH
1611 static void process_discard_bio(struct cache *cache, struct prealloc *structs,
1616 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1618 calc_discard_block_range(cache, bio, &b, &e);
1624 cell_prealloc = prealloc_get_cell(structs);
1625 r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc,
1626 (cell_free_fn) prealloc_put_cell,
1627 structs, &new_ocell);
1631 discard(cache, structs, new_ocell);
1634 static bool spare_migration_bandwidth(struct cache *cache)
1636 sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) *
1637 cache->sectors_per_block;
1638 return current_volume < cache->migration_threshold;
1641 static void inc_hit_counter(struct cache *cache, struct bio *bio)
1643 atomic_inc(bio_data_dir(bio) == READ ?
1644 &cache->stats.read_hit : &cache->stats.write_hit);
1647 static void inc_miss_counter(struct cache *cache, struct bio *bio)
1649 atomic_inc(bio_data_dir(bio) == READ ?
1650 &cache->stats.read_miss : &cache->stats.write_miss);
1653 /*----------------------------------------------------------------*/
1656 struct cache *cache;
1657 struct bio_list bios_for_issue;
1658 struct bio_list unhandled_bios;
1662 static void inc_fn(void *context, struct dm_bio_prison_cell *cell)
1665 struct inc_detail *detail = context;
1666 struct cache *cache = detail->cache;
1668 inc_ds(cache, cell->holder, cell);
1669 if (bio_data_dir(cell->holder) == WRITE)
1670 detail->any_writes = true;
1672 while ((bio = bio_list_pop(&cell->bios))) {
1673 if (discard_or_flush(bio)) {
1674 bio_list_add(&detail->unhandled_bios, bio);
1678 if (bio_data_dir(bio) == WRITE)
1679 detail->any_writes = true;
1681 bio_list_add(&detail->bios_for_issue, bio);
1682 inc_ds(cache, bio, cell);
1686 // FIXME: refactor these two
1687 static void remap_cell_to_origin_clear_discard(struct cache *cache,
1688 struct dm_bio_prison_cell *cell,
1689 dm_oblock_t oblock, bool issue_holder)
1692 unsigned long flags;
1693 struct inc_detail detail;
1695 detail.cache = cache;
1696 bio_list_init(&detail.bios_for_issue);
1697 bio_list_init(&detail.unhandled_bios);
1698 detail.any_writes = false;
1700 spin_lock_irqsave(&cache->lock, flags);
1701 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1702 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1703 spin_unlock_irqrestore(&cache->lock, flags);
1705 remap_to_origin(cache, cell->holder);
1707 issue(cache, cell->holder);
1709 accounted_begin(cache, cell->holder);
1711 if (detail.any_writes)
1712 clear_discard(cache, oblock_to_dblock(cache, oblock));
1714 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1715 remap_to_origin(cache, bio);
1720 static void remap_cell_to_cache_dirty(struct cache *cache, struct dm_bio_prison_cell *cell,
1721 dm_oblock_t oblock, dm_cblock_t cblock, bool issue_holder)
1724 unsigned long flags;
1725 struct inc_detail detail;
1727 detail.cache = cache;
1728 bio_list_init(&detail.bios_for_issue);
1729 bio_list_init(&detail.unhandled_bios);
1730 detail.any_writes = false;
1732 spin_lock_irqsave(&cache->lock, flags);
1733 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1734 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1735 spin_unlock_irqrestore(&cache->lock, flags);
1737 remap_to_cache(cache, cell->holder, cblock);
1739 issue(cache, cell->holder);
1741 accounted_begin(cache, cell->holder);
1743 if (detail.any_writes) {
1744 set_dirty(cache, oblock, cblock);
1745 clear_discard(cache, oblock_to_dblock(cache, oblock));
1748 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1749 remap_to_cache(cache, bio, cblock);
1754 /*----------------------------------------------------------------*/
1756 struct old_oblock_lock {
1757 struct policy_locker locker;
1758 struct cache *cache;
1759 struct prealloc *structs;
1760 struct dm_bio_prison_cell *cell;
1763 static int null_locker(struct policy_locker *locker, dm_oblock_t b)
1765 /* This should never be called */
1770 static int cell_locker(struct policy_locker *locker, dm_oblock_t b)
1772 struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker);
1773 struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs);
1775 return bio_detain(l->cache, b, NULL, cell_prealloc,
1776 (cell_free_fn) prealloc_put_cell,
1777 l->structs, &l->cell);
1780 static void process_cell(struct cache *cache, struct prealloc *structs,
1781 struct dm_bio_prison_cell *new_ocell)
1784 bool release_cell = true;
1785 struct bio *bio = new_ocell->holder;
1786 dm_oblock_t block = get_bio_block(cache, bio);
1787 struct policy_result lookup_result;
1788 bool passthrough = passthrough_mode(&cache->features);
1789 bool fast_promotion, can_migrate;
1790 struct old_oblock_lock ool;
1792 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
1793 can_migrate = !passthrough && (fast_promotion || spare_migration_bandwidth(cache));
1795 ool.locker.fn = cell_locker;
1797 ool.structs = structs;
1799 r = policy_map(cache->policy, block, true, can_migrate, fast_promotion,
1800 bio, &ool.locker, &lookup_result);
1802 if (r == -EWOULDBLOCK)
1803 /* migration has been denied */
1804 lookup_result.op = POLICY_MISS;
1806 switch (lookup_result.op) {
1809 inc_miss_counter(cache, bio);
1812 * Passthrough always maps to the origin,
1813 * invalidating any cache blocks that are written
1817 if (bio_data_dir(bio) == WRITE) {
1818 atomic_inc(&cache->stats.demotion);
1819 invalidate(cache, structs, block, lookup_result.cblock, new_ocell);
1820 release_cell = false;
1823 /* FIXME: factor out issue_origin() */
1824 remap_to_origin_clear_discard(cache, bio, block);
1825 inc_and_issue(cache, bio, new_ocell);
1828 inc_hit_counter(cache, bio);
1830 if (bio_data_dir(bio) == WRITE &&
1831 writethrough_mode(&cache->features) &&
1832 !is_dirty(cache, lookup_result.cblock)) {
1833 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
1834 inc_and_issue(cache, bio, new_ocell);
1837 remap_cell_to_cache_dirty(cache, new_ocell, block, lookup_result.cblock, true);
1838 release_cell = false;
1845 inc_miss_counter(cache, bio);
1846 remap_cell_to_origin_clear_discard(cache, new_ocell, block, true);
1847 release_cell = false;
1851 atomic_inc(&cache->stats.promotion);
1852 promote(cache, structs, block, lookup_result.cblock, new_ocell);
1853 release_cell = false;
1856 case POLICY_REPLACE:
1857 atomic_inc(&cache->stats.demotion);
1858 atomic_inc(&cache->stats.promotion);
1859 demote_then_promote(cache, structs, lookup_result.old_oblock,
1860 block, lookup_result.cblock,
1861 ool.cell, new_ocell);
1862 release_cell = false;
1866 DMERR_LIMIT("%s: erroring bio, unknown policy op: %u", __func__,
1867 (unsigned) lookup_result.op);
1872 cell_defer(cache, new_ocell, false);
1875 static void process_bio(struct cache *cache, struct prealloc *structs,
1879 dm_oblock_t block = get_bio_block(cache, bio);
1880 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1883 * Check to see if that block is currently migrating.
1885 cell_prealloc = prealloc_get_cell(structs);
1886 r = bio_detain(cache, block, bio, cell_prealloc,
1887 (cell_free_fn) prealloc_put_cell,
1888 structs, &new_ocell);
1892 process_cell(cache, structs, new_ocell);
1895 static int need_commit_due_to_time(struct cache *cache)
1897 return jiffies < cache->last_commit_jiffies ||
1898 jiffies > cache->last_commit_jiffies + COMMIT_PERIOD;
1902 * A non-zero return indicates read_only or fail_io mode.
1904 static int commit(struct cache *cache, bool clean_shutdown)
1908 if (get_cache_mode(cache) >= CM_READ_ONLY)
1911 atomic_inc(&cache->stats.commit_count);
1912 r = dm_cache_commit(cache->cmd, clean_shutdown);
1914 metadata_operation_failed(cache, "dm_cache_commit", r);
1919 static int commit_if_needed(struct cache *cache)
1923 if ((cache->commit_requested || need_commit_due_to_time(cache)) &&
1924 dm_cache_changed_this_transaction(cache->cmd)) {
1925 r = commit(cache, false);
1926 cache->commit_requested = false;
1927 cache->last_commit_jiffies = jiffies;
1933 static void process_deferred_bios(struct cache *cache)
1935 unsigned long flags;
1936 struct bio_list bios;
1938 struct prealloc structs;
1940 memset(&structs, 0, sizeof(structs));
1941 bio_list_init(&bios);
1943 spin_lock_irqsave(&cache->lock, flags);
1944 bio_list_merge(&bios, &cache->deferred_bios);
1945 bio_list_init(&cache->deferred_bios);
1946 spin_unlock_irqrestore(&cache->lock, flags);
1948 while (!bio_list_empty(&bios)) {
1950 * If we've got no free migration structs, and processing
1951 * this bio might require one, we pause until there are some
1952 * prepared mappings to process.
1954 if (prealloc_data_structs(cache, &structs)) {
1955 spin_lock_irqsave(&cache->lock, flags);
1956 bio_list_merge(&cache->deferred_bios, &bios);
1957 spin_unlock_irqrestore(&cache->lock, flags);
1961 bio = bio_list_pop(&bios);
1963 if (bio->bi_rw & REQ_FLUSH)
1964 process_flush_bio(cache, bio);
1965 else if (bio->bi_rw & REQ_DISCARD)
1966 process_discard_bio(cache, &structs, bio);
1968 process_bio(cache, &structs, bio);
1971 prealloc_free_structs(cache, &structs);
1974 static void process_deferred_cells(struct cache *cache)
1976 unsigned long flags;
1977 struct dm_bio_prison_cell *cell, *tmp;
1978 struct list_head cells;
1979 struct prealloc structs;
1981 memset(&structs, 0, sizeof(structs));
1983 INIT_LIST_HEAD(&cells);
1985 spin_lock_irqsave(&cache->lock, flags);
1986 list_splice_init(&cache->deferred_cells, &cells);
1987 spin_unlock_irqrestore(&cache->lock, flags);
1989 list_for_each_entry_safe(cell, tmp, &cells, user_list) {
1991 * If we've got no free migration structs, and processing
1992 * this bio might require one, we pause until there are some
1993 * prepared mappings to process.
1995 if (prealloc_data_structs(cache, &structs)) {
1996 spin_lock_irqsave(&cache->lock, flags);
1997 list_splice(&cells, &cache->deferred_cells);
1998 spin_unlock_irqrestore(&cache->lock, flags);
2002 process_cell(cache, &structs, cell);
2005 prealloc_free_structs(cache, &structs);
2008 static void process_deferred_flush_bios(struct cache *cache, bool submit_bios)
2010 unsigned long flags;
2011 struct bio_list bios;
2014 bio_list_init(&bios);
2016 spin_lock_irqsave(&cache->lock, flags);
2017 bio_list_merge(&bios, &cache->deferred_flush_bios);
2018 bio_list_init(&cache->deferred_flush_bios);
2019 spin_unlock_irqrestore(&cache->lock, flags);
2022 * These bios have already been through inc_ds()
2024 while ((bio = bio_list_pop(&bios)))
2025 submit_bios ? accounted_request(cache, bio) : bio_io_error(bio);
2028 static void process_deferred_writethrough_bios(struct cache *cache)
2030 unsigned long flags;
2031 struct bio_list bios;
2034 bio_list_init(&bios);
2036 spin_lock_irqsave(&cache->lock, flags);
2037 bio_list_merge(&bios, &cache->deferred_writethrough_bios);
2038 bio_list_init(&cache->deferred_writethrough_bios);
2039 spin_unlock_irqrestore(&cache->lock, flags);
2042 * These bios have already been through inc_ds()
2044 while ((bio = bio_list_pop(&bios)))
2045 accounted_request(cache, bio);
2048 static void writeback_some_dirty_blocks(struct cache *cache)
2053 struct prealloc structs;
2054 struct dm_bio_prison_cell *old_ocell;
2055 bool busy = !iot_idle_for(&cache->origin_tracker, HZ);
2057 memset(&structs, 0, sizeof(structs));
2059 while (spare_migration_bandwidth(cache)) {
2060 if (prealloc_data_structs(cache, &structs))
2063 r = policy_writeback_work(cache->policy, &oblock, &cblock, busy);
2067 r = get_cell(cache, oblock, &structs, &old_ocell);
2069 policy_set_dirty(cache->policy, oblock);
2073 writeback(cache, &structs, oblock, cblock, old_ocell);
2076 prealloc_free_structs(cache, &structs);
2079 /*----------------------------------------------------------------
2081 * Dropping something from the cache *without* writing back.
2082 *--------------------------------------------------------------*/
2084 static void process_invalidation_request(struct cache *cache, struct invalidation_request *req)
2087 uint64_t begin = from_cblock(req->cblocks->begin);
2088 uint64_t end = from_cblock(req->cblocks->end);
2090 while (begin != end) {
2091 r = policy_remove_cblock(cache->policy, to_cblock(begin));
2093 r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin));
2095 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
2099 } else if (r == -ENODATA) {
2100 /* harmless, already unmapped */
2104 DMERR("policy_remove_cblock failed");
2111 cache->commit_requested = true;
2114 atomic_set(&req->complete, 1);
2116 wake_up(&req->result_wait);
2119 static void process_invalidation_requests(struct cache *cache)
2121 struct list_head list;
2122 struct invalidation_request *req, *tmp;
2124 INIT_LIST_HEAD(&list);
2125 spin_lock(&cache->invalidation_lock);
2126 list_splice_init(&cache->invalidation_requests, &list);
2127 spin_unlock(&cache->invalidation_lock);
2129 list_for_each_entry_safe (req, tmp, &list, list)
2130 process_invalidation_request(cache, req);
2133 /*----------------------------------------------------------------
2135 *--------------------------------------------------------------*/
2136 static bool is_quiescing(struct cache *cache)
2138 return atomic_read(&cache->quiescing);
2141 static void ack_quiescing(struct cache *cache)
2143 if (is_quiescing(cache)) {
2144 atomic_inc(&cache->quiescing_ack);
2145 wake_up(&cache->quiescing_wait);
2149 static void wait_for_quiescing_ack(struct cache *cache)
2151 wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack));
2154 static void start_quiescing(struct cache *cache)
2156 atomic_inc(&cache->quiescing);
2157 wait_for_quiescing_ack(cache);
2160 static void stop_quiescing(struct cache *cache)
2162 atomic_set(&cache->quiescing, 0);
2163 atomic_set(&cache->quiescing_ack, 0);
2166 static void wait_for_migrations(struct cache *cache)
2168 wait_event(cache->migration_wait, !atomic_read(&cache->nr_allocated_migrations));
2171 static void stop_worker(struct cache *cache)
2173 cancel_delayed_work(&cache->waker);
2174 flush_workqueue(cache->wq);
2177 static void requeue_deferred_cells(struct cache *cache)
2179 unsigned long flags;
2180 struct list_head cells;
2181 struct dm_bio_prison_cell *cell, *tmp;
2183 INIT_LIST_HEAD(&cells);
2184 spin_lock_irqsave(&cache->lock, flags);
2185 list_splice_init(&cache->deferred_cells, &cells);
2186 spin_unlock_irqrestore(&cache->lock, flags);
2188 list_for_each_entry_safe(cell, tmp, &cells, user_list)
2189 cell_requeue(cache, cell);
2192 static void requeue_deferred_bios(struct cache *cache)
2195 struct bio_list bios;
2197 bio_list_init(&bios);
2198 bio_list_merge(&bios, &cache->deferred_bios);
2199 bio_list_init(&cache->deferred_bios);
2201 while ((bio = bio_list_pop(&bios)))
2202 bio_endio(bio, DM_ENDIO_REQUEUE);
2205 static int more_work(struct cache *cache)
2207 if (is_quiescing(cache))
2208 return !list_empty(&cache->quiesced_migrations) ||
2209 !list_empty(&cache->completed_migrations) ||
2210 !list_empty(&cache->need_commit_migrations);
2212 return !bio_list_empty(&cache->deferred_bios) ||
2213 !list_empty(&cache->deferred_cells) ||
2214 !bio_list_empty(&cache->deferred_flush_bios) ||
2215 !bio_list_empty(&cache->deferred_writethrough_bios) ||
2216 !list_empty(&cache->quiesced_migrations) ||
2217 !list_empty(&cache->completed_migrations) ||
2218 !list_empty(&cache->need_commit_migrations) ||
2222 static void do_worker(struct work_struct *ws)
2224 struct cache *cache = container_of(ws, struct cache, worker);
2227 if (!is_quiescing(cache)) {
2228 writeback_some_dirty_blocks(cache);
2229 process_deferred_writethrough_bios(cache);
2230 process_deferred_bios(cache);
2231 process_deferred_cells(cache);
2232 process_invalidation_requests(cache);
2235 process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard);
2236 process_migrations(cache, &cache->completed_migrations, complete_migration);
2238 if (commit_if_needed(cache)) {
2239 process_deferred_flush_bios(cache, false);
2240 process_migrations(cache, &cache->need_commit_migrations, migration_failure);
2242 process_deferred_flush_bios(cache, true);
2243 process_migrations(cache, &cache->need_commit_migrations,
2244 migration_success_post_commit);
2247 ack_quiescing(cache);
2249 } while (more_work(cache));
2253 * We want to commit periodically so that not too much
2254 * unwritten metadata builds up.
2256 static void do_waker(struct work_struct *ws)
2258 struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker);
2259 policy_tick(cache->policy);
2261 queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD);
2264 /*----------------------------------------------------------------*/
2266 static int is_congested(struct dm_dev *dev, int bdi_bits)
2268 struct request_queue *q = bdev_get_queue(dev->bdev);
2269 return bdi_congested(&q->backing_dev_info, bdi_bits);
2272 static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2274 struct cache *cache = container_of(cb, struct cache, callbacks);
2276 return is_congested(cache->origin_dev, bdi_bits) ||
2277 is_congested(cache->cache_dev, bdi_bits);
2280 /*----------------------------------------------------------------
2282 *--------------------------------------------------------------*/
2285 * This function gets called on the error paths of the constructor, so we
2286 * have to cope with a partially initialised struct.
2288 static void destroy(struct cache *cache)
2292 if (cache->migration_pool)
2293 mempool_destroy(cache->migration_pool);
2295 if (cache->all_io_ds)
2296 dm_deferred_set_destroy(cache->all_io_ds);
2299 dm_bio_prison_destroy(cache->prison);
2302 destroy_workqueue(cache->wq);
2304 if (cache->dirty_bitset)
2305 free_bitset(cache->dirty_bitset);
2307 if (cache->discard_bitset)
2308 free_bitset(cache->discard_bitset);
2311 dm_kcopyd_client_destroy(cache->copier);
2314 dm_cache_metadata_close(cache->cmd);
2316 if (cache->metadata_dev)
2317 dm_put_device(cache->ti, cache->metadata_dev);
2319 if (cache->origin_dev)
2320 dm_put_device(cache->ti, cache->origin_dev);
2322 if (cache->cache_dev)
2323 dm_put_device(cache->ti, cache->cache_dev);
2326 dm_cache_policy_destroy(cache->policy);
2328 for (i = 0; i < cache->nr_ctr_args ; i++)
2329 kfree(cache->ctr_args[i]);
2330 kfree(cache->ctr_args);
2335 static void cache_dtr(struct dm_target *ti)
2337 struct cache *cache = ti->private;
2342 static sector_t get_dev_size(struct dm_dev *dev)
2344 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
2347 /*----------------------------------------------------------------*/
2350 * Construct a cache device mapping.
2352 * cache <metadata dev> <cache dev> <origin dev> <block size>
2353 * <#feature args> [<feature arg>]*
2354 * <policy> <#policy args> [<policy arg>]*
2356 * metadata dev : fast device holding the persistent metadata
2357 * cache dev : fast device holding cached data blocks
2358 * origin dev : slow device holding original data blocks
2359 * block size : cache unit size in sectors
2361 * #feature args : number of feature arguments passed
2362 * feature args : writethrough. (The default is writeback.)
2364 * policy : the replacement policy to use
2365 * #policy args : an even number of policy arguments corresponding
2366 * to key/value pairs passed to the policy
2367 * policy args : key/value pairs passed to the policy
2368 * E.g. 'sequential_threshold 1024'
2369 * See cache-policies.txt for details.
2371 * Optional feature arguments are:
2372 * writethrough : write through caching that prohibits cache block
2373 * content from being different from origin block content.
2374 * Without this argument, the default behaviour is to write
2375 * back cache block contents later for performance reasons,
2376 * so they may differ from the corresponding origin blocks.
2379 struct dm_target *ti;
2381 struct dm_dev *metadata_dev;
2383 struct dm_dev *cache_dev;
2384 sector_t cache_sectors;
2386 struct dm_dev *origin_dev;
2387 sector_t origin_sectors;
2389 uint32_t block_size;
2391 const char *policy_name;
2393 const char **policy_argv;
2395 struct cache_features features;
2398 static void destroy_cache_args(struct cache_args *ca)
2400 if (ca->metadata_dev)
2401 dm_put_device(ca->ti, ca->metadata_dev);
2404 dm_put_device(ca->ti, ca->cache_dev);
2407 dm_put_device(ca->ti, ca->origin_dev);
2412 static bool at_least_one_arg(struct dm_arg_set *as, char **error)
2415 *error = "Insufficient args";
2422 static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as,
2426 sector_t metadata_dev_size;
2427 char b[BDEVNAME_SIZE];
2429 if (!at_least_one_arg(as, error))
2432 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2435 *error = "Error opening metadata device";
2439 metadata_dev_size = get_dev_size(ca->metadata_dev);
2440 if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING)
2441 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2442 bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
2447 static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as,
2452 if (!at_least_one_arg(as, error))
2455 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2458 *error = "Error opening cache device";
2461 ca->cache_sectors = get_dev_size(ca->cache_dev);
2466 static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as,
2471 if (!at_least_one_arg(as, error))
2474 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2477 *error = "Error opening origin device";
2481 ca->origin_sectors = get_dev_size(ca->origin_dev);
2482 if (ca->ti->len > ca->origin_sectors) {
2483 *error = "Device size larger than cached device";
2490 static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as,
2493 unsigned long block_size;
2495 if (!at_least_one_arg(as, error))
2498 if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size ||
2499 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2500 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2501 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2502 *error = "Invalid data block size";
2506 if (block_size > ca->cache_sectors) {
2507 *error = "Data block size is larger than the cache device";
2511 ca->block_size = block_size;
2516 static void init_features(struct cache_features *cf)
2518 cf->mode = CM_WRITE;
2519 cf->io_mode = CM_IO_WRITEBACK;
2522 static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
2525 static struct dm_arg _args[] = {
2526 {0, 1, "Invalid number of cache feature arguments"},
2532 struct cache_features *cf = &ca->features;
2536 r = dm_read_arg_group(_args, as, &argc, error);
2541 arg = dm_shift_arg(as);
2543 if (!strcasecmp(arg, "writeback"))
2544 cf->io_mode = CM_IO_WRITEBACK;
2546 else if (!strcasecmp(arg, "writethrough"))
2547 cf->io_mode = CM_IO_WRITETHROUGH;
2549 else if (!strcasecmp(arg, "passthrough"))
2550 cf->io_mode = CM_IO_PASSTHROUGH;
2553 *error = "Unrecognised cache feature requested";
2561 static int parse_policy(struct cache_args *ca, struct dm_arg_set *as,
2564 static struct dm_arg _args[] = {
2565 {0, 1024, "Invalid number of policy arguments"},
2570 if (!at_least_one_arg(as, error))
2573 ca->policy_name = dm_shift_arg(as);
2575 r = dm_read_arg_group(_args, as, &ca->policy_argc, error);
2579 ca->policy_argv = (const char **)as->argv;
2580 dm_consume_args(as, ca->policy_argc);
2585 static int parse_cache_args(struct cache_args *ca, int argc, char **argv,
2589 struct dm_arg_set as;
2594 r = parse_metadata_dev(ca, &as, error);
2598 r = parse_cache_dev(ca, &as, error);
2602 r = parse_origin_dev(ca, &as, error);
2606 r = parse_block_size(ca, &as, error);
2610 r = parse_features(ca, &as, error);
2614 r = parse_policy(ca, &as, error);
2621 /*----------------------------------------------------------------*/
2623 static struct kmem_cache *migration_cache;
2625 #define NOT_CORE_OPTION 1
2627 static int process_config_option(struct cache *cache, const char *key, const char *value)
2631 if (!strcasecmp(key, "migration_threshold")) {
2632 if (kstrtoul(value, 10, &tmp))
2635 cache->migration_threshold = tmp;
2639 return NOT_CORE_OPTION;
2642 static int set_config_value(struct cache *cache, const char *key, const char *value)
2644 int r = process_config_option(cache, key, value);
2646 if (r == NOT_CORE_OPTION)
2647 r = policy_set_config_value(cache->policy, key, value);
2650 DMWARN("bad config value for %s: %s", key, value);
2655 static int set_config_values(struct cache *cache, int argc, const char **argv)
2660 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
2665 r = set_config_value(cache, argv[0], argv[1]);
2676 static int create_cache_policy(struct cache *cache, struct cache_args *ca,
2679 struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name,
2681 cache->origin_sectors,
2682 cache->sectors_per_block);
2684 *error = "Error creating cache's policy";
2693 * We want the discard block size to be at least the size of the cache
2694 * block size and have no more than 2^14 discard blocks across the origin.
2696 #define MAX_DISCARD_BLOCKS (1 << 14)
2698 static bool too_many_discard_blocks(sector_t discard_block_size,
2699 sector_t origin_size)
2701 (void) sector_div(origin_size, discard_block_size);
2703 return origin_size > MAX_DISCARD_BLOCKS;
2706 static sector_t calculate_discard_block_size(sector_t cache_block_size,
2707 sector_t origin_size)
2709 sector_t discard_block_size = cache_block_size;
2712 while (too_many_discard_blocks(discard_block_size, origin_size))
2713 discard_block_size *= 2;
2715 return discard_block_size;
2718 static void set_cache_size(struct cache *cache, dm_cblock_t size)
2720 dm_block_t nr_blocks = from_cblock(size);
2722 if (nr_blocks > (1 << 20) && cache->cache_size != size)
2723 DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n"
2724 "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n"
2725 "Please consider increasing the cache block size to reduce the overall cache block count.",
2726 (unsigned long long) nr_blocks);
2728 cache->cache_size = size;
2731 #define DEFAULT_MIGRATION_THRESHOLD 2048
2733 static int cache_create(struct cache_args *ca, struct cache **result)
2736 char **error = &ca->ti->error;
2737 struct cache *cache;
2738 struct dm_target *ti = ca->ti;
2739 dm_block_t origin_blocks;
2740 struct dm_cache_metadata *cmd;
2741 bool may_format = ca->features.mode == CM_WRITE;
2743 cache = kzalloc(sizeof(*cache), GFP_KERNEL);
2748 ti->private = cache;
2749 ti->num_flush_bios = 2;
2750 ti->flush_supported = true;
2752 ti->num_discard_bios = 1;
2753 ti->discards_supported = true;
2754 ti->discard_zeroes_data_unsupported = true;
2755 ti->split_discard_bios = false;
2757 cache->features = ca->features;
2758 ti->per_bio_data_size = get_per_bio_data_size(cache);
2760 cache->callbacks.congested_fn = cache_is_congested;
2761 dm_table_add_target_callbacks(ti->table, &cache->callbacks);
2763 cache->metadata_dev = ca->metadata_dev;
2764 cache->origin_dev = ca->origin_dev;
2765 cache->cache_dev = ca->cache_dev;
2767 ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL;
2769 /* FIXME: factor out this whole section */
2770 origin_blocks = cache->origin_sectors = ca->origin_sectors;
2771 origin_blocks = block_div(origin_blocks, ca->block_size);
2772 cache->origin_blocks = to_oblock(origin_blocks);
2774 cache->sectors_per_block = ca->block_size;
2775 if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) {
2780 if (ca->block_size & (ca->block_size - 1)) {
2781 dm_block_t cache_size = ca->cache_sectors;
2783 cache->sectors_per_block_shift = -1;
2784 cache_size = block_div(cache_size, ca->block_size);
2785 set_cache_size(cache, to_cblock(cache_size));
2787 cache->sectors_per_block_shift = __ffs(ca->block_size);
2788 set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift));
2791 r = create_cache_policy(cache, ca, error);
2795 cache->policy_nr_args = ca->policy_argc;
2796 cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD;
2798 r = set_config_values(cache, ca->policy_argc, ca->policy_argv);
2800 *error = "Error setting cache policy's config values";
2804 cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
2805 ca->block_size, may_format,
2806 dm_cache_policy_get_hint_size(cache->policy));
2808 *error = "Error creating metadata object";
2813 set_cache_mode(cache, CM_WRITE);
2814 if (get_cache_mode(cache) != CM_WRITE) {
2815 *error = "Unable to get write access to metadata, please check/repair metadata.";
2820 if (passthrough_mode(&cache->features)) {
2823 r = dm_cache_metadata_all_clean(cache->cmd, &all_clean);
2825 *error = "dm_cache_metadata_all_clean() failed";
2830 *error = "Cannot enter passthrough mode unless all blocks are clean";
2836 spin_lock_init(&cache->lock);
2837 INIT_LIST_HEAD(&cache->deferred_cells);
2838 bio_list_init(&cache->deferred_bios);
2839 bio_list_init(&cache->deferred_flush_bios);
2840 bio_list_init(&cache->deferred_writethrough_bios);
2841 INIT_LIST_HEAD(&cache->quiesced_migrations);
2842 INIT_LIST_HEAD(&cache->completed_migrations);
2843 INIT_LIST_HEAD(&cache->need_commit_migrations);
2844 atomic_set(&cache->nr_allocated_migrations, 0);
2845 atomic_set(&cache->nr_io_migrations, 0);
2846 init_waitqueue_head(&cache->migration_wait);
2848 init_waitqueue_head(&cache->quiescing_wait);
2849 atomic_set(&cache->quiescing, 0);
2850 atomic_set(&cache->quiescing_ack, 0);
2853 atomic_set(&cache->nr_dirty, 0);
2854 cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size));
2855 if (!cache->dirty_bitset) {
2856 *error = "could not allocate dirty bitset";
2859 clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size));
2861 cache->discard_block_size =
2862 calculate_discard_block_size(cache->sectors_per_block,
2863 cache->origin_sectors);
2864 cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors,
2865 cache->discard_block_size));
2866 cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks));
2867 if (!cache->discard_bitset) {
2868 *error = "could not allocate discard bitset";
2871 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
2873 cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2874 if (IS_ERR(cache->copier)) {
2875 *error = "could not create kcopyd client";
2876 r = PTR_ERR(cache->copier);
2880 cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2882 *error = "could not create workqueue for metadata object";
2885 INIT_WORK(&cache->worker, do_worker);
2886 INIT_DELAYED_WORK(&cache->waker, do_waker);
2887 cache->last_commit_jiffies = jiffies;
2889 cache->prison = dm_bio_prison_create();
2890 if (!cache->prison) {
2891 *error = "could not create bio prison";
2895 cache->all_io_ds = dm_deferred_set_create();
2896 if (!cache->all_io_ds) {
2897 *error = "could not create all_io deferred set";
2901 cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE,
2903 if (!cache->migration_pool) {
2904 *error = "Error creating cache's migration mempool";
2908 cache->need_tick_bio = true;
2909 cache->sized = false;
2910 cache->invalidate = false;
2911 cache->commit_requested = false;
2912 cache->loaded_mappings = false;
2913 cache->loaded_discards = false;
2917 atomic_set(&cache->stats.demotion, 0);
2918 atomic_set(&cache->stats.promotion, 0);
2919 atomic_set(&cache->stats.copies_avoided, 0);
2920 atomic_set(&cache->stats.cache_cell_clash, 0);
2921 atomic_set(&cache->stats.commit_count, 0);
2922 atomic_set(&cache->stats.discard_count, 0);
2924 spin_lock_init(&cache->invalidation_lock);
2925 INIT_LIST_HEAD(&cache->invalidation_requests);
2927 iot_init(&cache->origin_tracker);
2937 static int copy_ctr_args(struct cache *cache, int argc, const char **argv)
2942 copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL);
2945 for (i = 0; i < argc; i++) {
2946 copy[i] = kstrdup(argv[i], GFP_KERNEL);
2955 cache->nr_ctr_args = argc;
2956 cache->ctr_args = copy;
2961 static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv)
2964 struct cache_args *ca;
2965 struct cache *cache = NULL;
2967 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2969 ti->error = "Error allocating memory for cache";
2974 r = parse_cache_args(ca, argc, argv, &ti->error);
2978 r = cache_create(ca, &cache);
2982 r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3);
2988 ti->private = cache;
2991 destroy_cache_args(ca);
2995 /*----------------------------------------------------------------*/
2997 static int cache_map(struct dm_target *ti, struct bio *bio)
2999 struct cache *cache = ti->private;
3002 struct dm_bio_prison_cell *cell = NULL;
3003 dm_oblock_t block = get_bio_block(cache, bio);
3004 size_t pb_data_size = get_per_bio_data_size(cache);
3005 bool can_migrate = false;
3006 bool fast_promotion;
3007 struct policy_result lookup_result;
3008 struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size);
3009 struct old_oblock_lock ool;
3011 ool.locker.fn = null_locker;
3013 if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) {
3015 * This can only occur if the io goes to a partial block at
3016 * the end of the origin device. We don't cache these.
3017 * Just remap to the origin and carry on.
3019 remap_to_origin(cache, bio);
3020 accounted_begin(cache, bio);
3021 return DM_MAPIO_REMAPPED;
3024 if (discard_or_flush(bio)) {
3025 defer_bio(cache, bio);
3026 return DM_MAPIO_SUBMITTED;
3030 * Check to see if that block is currently migrating.
3032 cell = alloc_prison_cell(cache);
3034 defer_bio(cache, bio);
3035 return DM_MAPIO_SUBMITTED;
3038 r = bio_detain(cache, block, bio, cell,
3039 (cell_free_fn) free_prison_cell,
3043 defer_bio(cache, bio);
3045 return DM_MAPIO_SUBMITTED;
3048 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
3050 r = policy_map(cache->policy, block, false, can_migrate, fast_promotion,
3051 bio, &ool.locker, &lookup_result);
3052 if (r == -EWOULDBLOCK) {
3053 cell_defer(cache, cell, true);
3054 return DM_MAPIO_SUBMITTED;
3057 DMERR_LIMIT("Unexpected return from cache replacement policy: %d", r);
3058 cell_defer(cache, cell, false);
3060 return DM_MAPIO_SUBMITTED;
3063 r = DM_MAPIO_REMAPPED;
3064 switch (lookup_result.op) {
3066 if (passthrough_mode(&cache->features)) {
3067 if (bio_data_dir(bio) == WRITE) {
3069 * We need to invalidate this block, so
3070 * defer for the worker thread.
3072 cell_defer(cache, cell, true);
3073 r = DM_MAPIO_SUBMITTED;
3076 inc_miss_counter(cache, bio);
3077 remap_to_origin_clear_discard(cache, bio, block);
3078 accounted_begin(cache, bio);
3079 inc_ds(cache, bio, cell);
3080 // FIXME: we want to remap hits or misses straight
3081 // away rather than passing over to the worker.
3082 cell_defer(cache, cell, false);
3086 inc_hit_counter(cache, bio);
3087 if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) &&
3088 !is_dirty(cache, lookup_result.cblock)) {
3089 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
3090 accounted_begin(cache, bio);
3091 inc_ds(cache, bio, cell);
3092 cell_defer(cache, cell, false);
3095 remap_cell_to_cache_dirty(cache, cell, block, lookup_result.cblock, false);
3100 inc_miss_counter(cache, bio);
3101 if (pb->req_nr != 0) {
3103 * This is a duplicate writethrough io that is no
3104 * longer needed because the block has been demoted.
3107 // FIXME: remap everything as a miss
3108 cell_defer(cache, cell, false);
3109 r = DM_MAPIO_SUBMITTED;
3112 remap_cell_to_origin_clear_discard(cache, cell, block, false);
3116 DMERR_LIMIT("%s: erroring bio: unknown policy op: %u", __func__,
3117 (unsigned) lookup_result.op);
3118 cell_defer(cache, cell, false);
3120 r = DM_MAPIO_SUBMITTED;
3126 static int cache_end_io(struct dm_target *ti, struct bio *bio, int error)
3128 struct cache *cache = ti->private;
3129 unsigned long flags;
3130 size_t pb_data_size = get_per_bio_data_size(cache);
3131 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
3134 policy_tick(cache->policy);
3136 spin_lock_irqsave(&cache->lock, flags);
3137 cache->need_tick_bio = true;
3138 spin_unlock_irqrestore(&cache->lock, flags);
3141 check_for_quiesced_migrations(cache, pb);
3142 accounted_complete(cache, bio);
3147 static int write_dirty_bitset(struct cache *cache)
3151 if (get_cache_mode(cache) >= CM_READ_ONLY)
3154 for (i = 0; i < from_cblock(cache->cache_size); i++) {
3155 r = dm_cache_set_dirty(cache->cmd, to_cblock(i),
3156 is_dirty(cache, to_cblock(i)));
3158 metadata_operation_failed(cache, "dm_cache_set_dirty", r);
3166 static int write_discard_bitset(struct cache *cache)
3170 if (get_cache_mode(cache) >= CM_READ_ONLY)
3173 r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size,
3174 cache->discard_nr_blocks);
3176 DMERR("could not resize on-disk discard bitset");
3177 metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r);
3181 for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) {
3182 r = dm_cache_set_discard(cache->cmd, to_dblock(i),
3183 is_discarded(cache, to_dblock(i)));
3185 metadata_operation_failed(cache, "dm_cache_set_discard", r);
3193 static int write_hints(struct cache *cache)
3197 if (get_cache_mode(cache) >= CM_READ_ONLY)
3200 r = dm_cache_write_hints(cache->cmd, cache->policy);
3202 metadata_operation_failed(cache, "dm_cache_write_hints", r);
3210 * returns true on success
3212 static bool sync_metadata(struct cache *cache)
3216 r1 = write_dirty_bitset(cache);
3218 DMERR("could not write dirty bitset");
3220 r2 = write_discard_bitset(cache);
3222 DMERR("could not write discard bitset");
3226 r3 = write_hints(cache);
3228 DMERR("could not write hints");
3231 * If writing the above metadata failed, we still commit, but don't
3232 * set the clean shutdown flag. This will effectively force every
3233 * dirty bit to be set on reload.
3235 r4 = commit(cache, !r1 && !r2 && !r3);
3237 DMERR("could not write cache metadata.");
3239 return !r1 && !r2 && !r3 && !r4;
3242 static void cache_postsuspend(struct dm_target *ti)
3244 struct cache *cache = ti->private;
3246 start_quiescing(cache);
3247 wait_for_migrations(cache);
3249 requeue_deferred_bios(cache);
3250 requeue_deferred_cells(cache);
3251 stop_quiescing(cache);
3253 if (get_cache_mode(cache) == CM_WRITE)
3254 (void) sync_metadata(cache);
3257 static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock,
3258 bool dirty, uint32_t hint, bool hint_valid)
3261 struct cache *cache = context;
3263 r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid);
3268 set_dirty(cache, oblock, cblock);
3270 clear_dirty(cache, oblock, cblock);
3276 * The discard block size in the on disk metadata is not
3277 * neccessarily the same as we're currently using. So we have to
3278 * be careful to only set the discarded attribute if we know it
3279 * covers a complete block of the new size.
3281 struct discard_load_info {
3282 struct cache *cache;
3285 * These blocks are sized using the on disk dblock size, rather
3286 * than the current one.
3288 dm_block_t block_size;
3289 dm_block_t discard_begin, discard_end;
3292 static void discard_load_info_init(struct cache *cache,
3293 struct discard_load_info *li)
3296 li->discard_begin = li->discard_end = 0;
3299 static void set_discard_range(struct discard_load_info *li)
3303 if (li->discard_begin == li->discard_end)
3307 * Convert to sectors.
3309 b = li->discard_begin * li->block_size;
3310 e = li->discard_end * li->block_size;
3313 * Then convert back to the current dblock size.
3315 b = dm_sector_div_up(b, li->cache->discard_block_size);
3316 sector_div(e, li->cache->discard_block_size);
3319 * The origin may have shrunk, so we need to check we're still in
3322 if (e > from_dblock(li->cache->discard_nr_blocks))
3323 e = from_dblock(li->cache->discard_nr_blocks);
3326 set_discard(li->cache, to_dblock(b));
3329 static int load_discard(void *context, sector_t discard_block_size,
3330 dm_dblock_t dblock, bool discard)
3332 struct discard_load_info *li = context;
3334 li->block_size = discard_block_size;
3337 if (from_dblock(dblock) == li->discard_end)
3339 * We're already in a discard range, just extend it.
3341 li->discard_end = li->discard_end + 1ULL;
3345 * Emit the old range and start a new one.
3347 set_discard_range(li);
3348 li->discard_begin = from_dblock(dblock);
3349 li->discard_end = li->discard_begin + 1ULL;
3352 set_discard_range(li);
3353 li->discard_begin = li->discard_end = 0;
3359 static dm_cblock_t get_cache_dev_size(struct cache *cache)
3361 sector_t size = get_dev_size(cache->cache_dev);
3362 (void) sector_div(size, cache->sectors_per_block);
3363 return to_cblock(size);
3366 static bool can_resize(struct cache *cache, dm_cblock_t new_size)
3368 if (from_cblock(new_size) > from_cblock(cache->cache_size))
3372 * We can't drop a dirty block when shrinking the cache.
3374 while (from_cblock(new_size) < from_cblock(cache->cache_size)) {
3375 new_size = to_cblock(from_cblock(new_size) + 1);
3376 if (is_dirty(cache, new_size)) {
3377 DMERR("unable to shrink cache; cache block %llu is dirty",
3378 (unsigned long long) from_cblock(new_size));
3386 static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size)
3390 r = dm_cache_resize(cache->cmd, new_size);
3392 DMERR("could not resize cache metadata");
3393 metadata_operation_failed(cache, "dm_cache_resize", r);
3397 set_cache_size(cache, new_size);
3402 static int cache_preresume(struct dm_target *ti)
3405 struct cache *cache = ti->private;
3406 dm_cblock_t csize = get_cache_dev_size(cache);
3409 * Check to see if the cache has resized.
3411 if (!cache->sized) {
3412 r = resize_cache_dev(cache, csize);
3416 cache->sized = true;
3418 } else if (csize != cache->cache_size) {
3419 if (!can_resize(cache, csize))
3422 r = resize_cache_dev(cache, csize);
3427 if (!cache->loaded_mappings) {
3428 r = dm_cache_load_mappings(cache->cmd, cache->policy,
3429 load_mapping, cache);
3431 DMERR("could not load cache mappings");
3432 metadata_operation_failed(cache, "dm_cache_load_mappings", r);
3436 cache->loaded_mappings = true;
3439 if (!cache->loaded_discards) {
3440 struct discard_load_info li;
3443 * The discard bitset could have been resized, or the
3444 * discard block size changed. To be safe we start by
3445 * setting every dblock to not discarded.
3447 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
3449 discard_load_info_init(cache, &li);
3450 r = dm_cache_load_discards(cache->cmd, load_discard, &li);
3452 DMERR("could not load origin discards");
3453 metadata_operation_failed(cache, "dm_cache_load_discards", r);
3456 set_discard_range(&li);
3458 cache->loaded_discards = true;
3464 static void cache_resume(struct dm_target *ti)
3466 struct cache *cache = ti->private;
3468 cache->need_tick_bio = true;
3469 do_waker(&cache->waker.work);
3475 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
3476 * <cache block size> <#used cache blocks>/<#total cache blocks>
3477 * <#read hits> <#read misses> <#write hits> <#write misses>
3478 * <#demotions> <#promotions> <#dirty>
3479 * <#features> <features>*
3480 * <#core args> <core args>
3481 * <policy name> <#policy args> <policy args>* <cache metadata mode>
3483 static void cache_status(struct dm_target *ti, status_type_t type,
3484 unsigned status_flags, char *result, unsigned maxlen)
3489 dm_block_t nr_free_blocks_metadata = 0;
3490 dm_block_t nr_blocks_metadata = 0;
3491 char buf[BDEVNAME_SIZE];
3492 struct cache *cache = ti->private;
3493 dm_cblock_t residency;
3496 case STATUSTYPE_INFO:
3497 if (get_cache_mode(cache) == CM_FAIL) {
3502 /* Commit to ensure statistics aren't out-of-date */
3503 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3504 (void) commit(cache, false);
3506 r = dm_cache_get_free_metadata_block_count(cache->cmd,
3507 &nr_free_blocks_metadata);
3509 DMERR("could not get metadata free block count");
3513 r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata);
3515 DMERR("could not get metadata device size");
3519 residency = policy_residency(cache->policy);
3521 DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ",
3522 (unsigned)DM_CACHE_METADATA_BLOCK_SIZE,
3523 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3524 (unsigned long long)nr_blocks_metadata,
3525 cache->sectors_per_block,
3526 (unsigned long long) from_cblock(residency),
3527 (unsigned long long) from_cblock(cache->cache_size),
3528 (unsigned) atomic_read(&cache->stats.read_hit),
3529 (unsigned) atomic_read(&cache->stats.read_miss),
3530 (unsigned) atomic_read(&cache->stats.write_hit),
3531 (unsigned) atomic_read(&cache->stats.write_miss),
3532 (unsigned) atomic_read(&cache->stats.demotion),
3533 (unsigned) atomic_read(&cache->stats.promotion),
3534 (unsigned long) atomic_read(&cache->nr_dirty));
3536 if (writethrough_mode(&cache->features))
3537 DMEMIT("1 writethrough ");
3539 else if (passthrough_mode(&cache->features))
3540 DMEMIT("1 passthrough ");
3542 else if (writeback_mode(&cache->features))
3543 DMEMIT("1 writeback ");
3546 DMERR("internal error: unknown io mode: %d", (int) cache->features.io_mode);
3550 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
3552 DMEMIT("%s ", dm_cache_policy_get_name(cache->policy));
3554 r = policy_emit_config_values(cache->policy, result, maxlen, &sz);
3556 DMERR("policy_emit_config_values returned %d", r);
3559 if (get_cache_mode(cache) == CM_READ_ONLY)
3566 case STATUSTYPE_TABLE:
3567 format_dev_t(buf, cache->metadata_dev->bdev->bd_dev);
3569 format_dev_t(buf, cache->cache_dev->bdev->bd_dev);
3571 format_dev_t(buf, cache->origin_dev->bdev->bd_dev);
3574 for (i = 0; i < cache->nr_ctr_args - 1; i++)
3575 DMEMIT(" %s", cache->ctr_args[i]);
3576 if (cache->nr_ctr_args)
3577 DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]);
3587 * A cache block range can take two forms:
3589 * i) A single cblock, eg. '3456'
3590 * ii) A begin and end cblock with dots between, eg. 123-234
3592 static int parse_cblock_range(struct cache *cache, const char *str,
3593 struct cblock_range *result)
3600 * Try and parse form (ii) first.
3602 r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy);
3607 result->begin = to_cblock(b);
3608 result->end = to_cblock(e);
3613 * That didn't work, try form (i).
3615 r = sscanf(str, "%llu%c", &b, &dummy);
3620 result->begin = to_cblock(b);
3621 result->end = to_cblock(from_cblock(result->begin) + 1u);
3625 DMERR("invalid cblock range '%s'", str);
3629 static int validate_cblock_range(struct cache *cache, struct cblock_range *range)
3631 uint64_t b = from_cblock(range->begin);
3632 uint64_t e = from_cblock(range->end);
3633 uint64_t n = from_cblock(cache->cache_size);
3636 DMERR("begin cblock out of range: %llu >= %llu", b, n);
3641 DMERR("end cblock out of range: %llu > %llu", e, n);
3646 DMERR("invalid cblock range: %llu >= %llu", b, e);
3653 static int request_invalidation(struct cache *cache, struct cblock_range *range)
3655 struct invalidation_request req;
3657 INIT_LIST_HEAD(&req.list);
3658 req.cblocks = range;
3659 atomic_set(&req.complete, 0);
3661 init_waitqueue_head(&req.result_wait);
3663 spin_lock(&cache->invalidation_lock);
3664 list_add(&req.list, &cache->invalidation_requests);
3665 spin_unlock(&cache->invalidation_lock);
3668 wait_event(req.result_wait, atomic_read(&req.complete));
3672 static int process_invalidate_cblocks_message(struct cache *cache, unsigned count,
3673 const char **cblock_ranges)
3677 struct cblock_range range;
3679 if (!passthrough_mode(&cache->features)) {
3680 DMERR("cache has to be in passthrough mode for invalidation");
3684 for (i = 0; i < count; i++) {
3685 r = parse_cblock_range(cache, cblock_ranges[i], &range);
3689 r = validate_cblock_range(cache, &range);
3694 * Pass begin and end origin blocks to the worker and wake it.
3696 r = request_invalidation(cache, &range);
3708 * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]*
3710 * The key migration_threshold is supported by the cache target core.
3712 static int cache_message(struct dm_target *ti, unsigned argc, char **argv)
3714 struct cache *cache = ti->private;
3719 if (get_cache_mode(cache) >= CM_READ_ONLY) {
3720 DMERR("unable to service cache target messages in READ_ONLY or FAIL mode");
3724 if (!strcasecmp(argv[0], "invalidate_cblocks"))
3725 return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1);
3730 return set_config_value(cache, argv[0], argv[1]);
3733 static int cache_iterate_devices(struct dm_target *ti,
3734 iterate_devices_callout_fn fn, void *data)
3737 struct cache *cache = ti->private;
3739 r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data);
3741 r = fn(ti, cache->origin_dev, 0, ti->len, data);
3747 * We assume I/O is going to the origin (which is the volume
3748 * more likely to have restrictions e.g. by being striped).
3749 * (Looking up the exact location of the data would be expensive
3750 * and could always be out of date by the time the bio is submitted.)
3752 static int cache_bvec_merge(struct dm_target *ti,
3753 struct bvec_merge_data *bvm,
3754 struct bio_vec *biovec, int max_size)
3756 struct cache *cache = ti->private;
3757 struct request_queue *q = bdev_get_queue(cache->origin_dev->bdev);
3759 if (!q->merge_bvec_fn)
3762 bvm->bi_bdev = cache->origin_dev->bdev;
3763 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
3766 static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
3769 * FIXME: these limits may be incompatible with the cache device
3771 limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024,
3772 cache->origin_sectors);
3773 limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
3776 static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
3778 struct cache *cache = ti->private;
3779 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
3782 * If the system-determined stacked limits are compatible with the
3783 * cache's blocksize (io_opt is a factor) do not override them.
3785 if (io_opt_sectors < cache->sectors_per_block ||
3786 do_div(io_opt_sectors, cache->sectors_per_block)) {
3787 blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT);
3788 blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
3790 set_discard_limits(cache, limits);
3793 /*----------------------------------------------------------------*/
3795 static struct target_type cache_target = {
3797 .version = {1, 7, 0},
3798 .module = THIS_MODULE,
3802 .end_io = cache_end_io,
3803 .postsuspend = cache_postsuspend,
3804 .preresume = cache_preresume,
3805 .resume = cache_resume,
3806 .status = cache_status,
3807 .message = cache_message,
3808 .iterate_devices = cache_iterate_devices,
3809 .merge = cache_bvec_merge,
3810 .io_hints = cache_io_hints,
3813 static int __init dm_cache_init(void)
3817 r = dm_register_target(&cache_target);
3819 DMERR("cache target registration failed: %d", r);
3823 migration_cache = KMEM_CACHE(dm_cache_migration, 0);
3824 if (!migration_cache) {
3825 dm_unregister_target(&cache_target);
3832 static void __exit dm_cache_exit(void)
3834 dm_unregister_target(&cache_target);
3835 kmem_cache_destroy(migration_cache);
3838 module_init(dm_cache_init);
3839 module_exit(dm_cache_exit);
3841 MODULE_DESCRIPTION(DM_NAME " cache target");
3842 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
3843 MODULE_LICENSE("GPL");