blob: 1af70145fab9bcee990dd54c08224f9e16924851 [file] [log] [blame]
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001/*
2 * Copyright (C) 2012 Red Hat. All rights reserved.
3 *
4 * This file is released under the GPL.
5 */
6
7#include "dm.h"
8#include "dm-bio-prison.h"
Darrick J. Wongb844fe62013-04-05 15:36:32 +01009#include "dm-bio-record.h"
Joe Thornberc6b4fcb2013-03-01 22:45:51 +000010#include "dm-cache-metadata.h"
11
12#include <linux/dm-io.h>
13#include <linux/dm-kcopyd.h>
14#include <linux/init.h>
15#include <linux/mempool.h>
16#include <linux/module.h>
17#include <linux/slab.h>
18#include <linux/vmalloc.h>
19
20#define DM_MSG_PREFIX "cache"
21
22DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle,
23 "A percentage of time allocated for copying to and/or from cache");
24
25/*----------------------------------------------------------------*/
26
27/*
28 * Glossary:
29 *
30 * oblock: index of an origin block
31 * cblock: index of a cache block
32 * promotion: movement of a block from origin to cache
33 * demotion: movement of a block from cache to origin
34 * migration: movement of a block between the origin and cache device,
35 * either direction
36 */
37
38/*----------------------------------------------------------------*/
39
40static size_t bitset_size_in_bytes(unsigned nr_entries)
41{
42 return sizeof(unsigned long) * dm_div_up(nr_entries, BITS_PER_LONG);
43}
44
45static unsigned long *alloc_bitset(unsigned nr_entries)
46{
47 size_t s = bitset_size_in_bytes(nr_entries);
48 return vzalloc(s);
49}
50
51static void clear_bitset(void *bitset, unsigned nr_entries)
52{
53 size_t s = bitset_size_in_bytes(nr_entries);
54 memset(bitset, 0, s);
55}
56
57static void free_bitset(unsigned long *bits)
58{
59 vfree(bits);
60}
61
62/*----------------------------------------------------------------*/
63
Joe Thornberc9d28d52013-10-31 13:55:48 -040064/*
65 * There are a couple of places where we let a bio run, but want to do some
66 * work before calling its endio function. We do this by temporarily
67 * changing the endio fn.
68 */
69struct dm_hook_info {
70 bio_end_io_t *bi_end_io;
71 void *bi_private;
72};
73
74static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio,
75 bio_end_io_t *bi_end_io, void *bi_private)
76{
77 h->bi_end_io = bio->bi_end_io;
78 h->bi_private = bio->bi_private;
79
80 bio->bi_end_io = bi_end_io;
81 bio->bi_private = bi_private;
82}
83
84static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio)
85{
86 bio->bi_end_io = h->bi_end_io;
87 bio->bi_private = h->bi_private;
Mike Snitzer8d307262013-12-03 19:16:04 -070088
89 /*
90 * Must bump bi_remaining to allow bio to complete with
91 * restored bi_end_io.
92 */
93 atomic_inc(&bio->bi_remaining);
Joe Thornberc9d28d52013-10-31 13:55:48 -040094}
95
96/*----------------------------------------------------------------*/
97
Joe Thornberc6b4fcb2013-03-01 22:45:51 +000098#define PRISON_CELLS 1024
99#define MIGRATION_POOL_SIZE 128
100#define COMMIT_PERIOD HZ
101#define MIGRATION_COUNT_WINDOW 10
102
103/*
Mike Snitzer05473042013-08-16 10:54:19 -0400104 * The block size of the device holding cache data must be
105 * between 32KB and 1GB.
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000106 */
107#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT)
Mike Snitzer05473042013-08-16 10:54:19 -0400108#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000109
110/*
111 * FIXME: the cache is read/write for the time being.
112 */
Joe Thornber2ee57d52013-10-24 14:10:29 -0400113enum cache_metadata_mode {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000114 CM_WRITE, /* metadata may be changed */
115 CM_READ_ONLY, /* metadata may not be changed */
116};
117
Joe Thornber2ee57d52013-10-24 14:10:29 -0400118enum cache_io_mode {
119 /*
120 * Data is written to cached blocks only. These blocks are marked
121 * dirty. If you lose the cache device you will lose data.
122 * Potential performance increase for both reads and writes.
123 */
124 CM_IO_WRITEBACK,
125
126 /*
127 * Data is written to both cache and origin. Blocks are never
128 * dirty. Potential performance benfit for reads only.
129 */
130 CM_IO_WRITETHROUGH,
131
132 /*
133 * A degraded mode useful for various cache coherency situations
134 * (eg, rolling back snapshots). Reads and writes always go to the
135 * origin. If a write goes to a cached oblock, then the cache
136 * block is invalidated.
137 */
138 CM_IO_PASSTHROUGH
139};
140
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000141struct cache_features {
Joe Thornber2ee57d52013-10-24 14:10:29 -0400142 enum cache_metadata_mode mode;
143 enum cache_io_mode io_mode;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000144};
145
146struct cache_stats {
147 atomic_t read_hit;
148 atomic_t read_miss;
149 atomic_t write_hit;
150 atomic_t write_miss;
151 atomic_t demotion;
152 atomic_t promotion;
153 atomic_t copies_avoided;
154 atomic_t cache_cell_clash;
155 atomic_t commit_count;
156 atomic_t discard_count;
157};
158
Joe Thornber65790ff2013-11-08 16:39:50 +0000159/*
160 * Defines a range of cblocks, begin to (end - 1) are in the range. end is
161 * the one-past-the-end value.
162 */
163struct cblock_range {
164 dm_cblock_t begin;
165 dm_cblock_t end;
166};
167
168struct invalidation_request {
169 struct list_head list;
170 struct cblock_range *cblocks;
171
172 atomic_t complete;
173 int err;
174
175 wait_queue_head_t result_wait;
176};
177
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000178struct cache {
179 struct dm_target *ti;
180 struct dm_target_callbacks callbacks;
181
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400182 struct dm_cache_metadata *cmd;
183
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000184 /*
185 * Metadata is written to this device.
186 */
187 struct dm_dev *metadata_dev;
188
189 /*
190 * The slower of the two data devices. Typically a spindle.
191 */
192 struct dm_dev *origin_dev;
193
194 /*
195 * The faster of the two data devices. Typically an SSD.
196 */
197 struct dm_dev *cache_dev;
198
199 /*
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000200 * Size of the origin device in _complete_ blocks and native sectors.
201 */
202 dm_oblock_t origin_blocks;
203 sector_t origin_sectors;
204
205 /*
206 * Size of the cache device in blocks.
207 */
208 dm_cblock_t cache_size;
209
210 /*
211 * Fields for converting from sectors to blocks.
212 */
213 uint32_t sectors_per_block;
214 int sectors_per_block_shift;
215
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000216 spinlock_t lock;
217 struct bio_list deferred_bios;
218 struct bio_list deferred_flush_bios;
Joe Thornbere2e74d62013-03-20 17:21:27 +0000219 struct bio_list deferred_writethrough_bios;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000220 struct list_head quiesced_migrations;
221 struct list_head completed_migrations;
222 struct list_head need_commit_migrations;
223 sector_t migration_threshold;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000224 wait_queue_head_t migration_wait;
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400225 atomic_t nr_migrations;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000226
Joe Thornber66cb1912013-10-30 17:11:58 +0000227 wait_queue_head_t quiescing_wait;
Joe Thornber238f8362013-10-30 17:29:30 +0000228 atomic_t quiescing;
Joe Thornber66cb1912013-10-30 17:11:58 +0000229 atomic_t quiescing_ack;
230
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000231 /*
232 * cache_size entries, dirty if set
233 */
234 dm_cblock_t nr_dirty;
235 unsigned long *dirty_bitset;
236
237 /*
238 * origin_blocks entries, discarded if set.
239 */
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000240 dm_dblock_t discard_nr_blocks;
241 unsigned long *discard_bitset;
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400242 uint32_t discard_block_size; /* a power of 2 times sectors per block */
243
244 /*
245 * Rather than reconstructing the table line for the status we just
246 * save it and regurgitate.
247 */
248 unsigned nr_ctr_args;
249 const char **ctr_args;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000250
251 struct dm_kcopyd_client *copier;
252 struct workqueue_struct *wq;
253 struct work_struct worker;
254
255 struct delayed_work waker;
256 unsigned long last_commit_jiffies;
257
258 struct dm_bio_prison *prison;
259 struct dm_deferred_set *all_io_ds;
260
261 mempool_t *migration_pool;
262 struct dm_cache_migration *next_migration;
263
264 struct dm_cache_policy *policy;
265 unsigned policy_nr_args;
266
267 bool need_tick_bio:1;
268 bool sized:1;
Joe Thornber65790ff2013-11-08 16:39:50 +0000269 bool invalidate:1;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000270 bool commit_requested:1;
271 bool loaded_mappings:1;
272 bool loaded_discards:1;
273
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000274 /*
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400275 * Cache features such as write-through.
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000276 */
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400277 struct cache_features features;
278
279 struct cache_stats stats;
Joe Thornber65790ff2013-11-08 16:39:50 +0000280
281 /*
282 * Invalidation fields.
283 */
284 spinlock_t invalidation_lock;
285 struct list_head invalidation_requests;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000286};
287
288struct per_bio_data {
289 bool tick:1;
290 unsigned req_nr:2;
291 struct dm_deferred_entry *all_io_entry;
Mike Snitzerc6eda5e2014-01-31 14:11:54 -0500292 struct dm_hook_info hook_info;
Joe Thornbere2e74d62013-03-20 17:21:27 +0000293
Mike Snitzer19b00922013-04-05 15:36:34 +0100294 /*
295 * writethrough fields. These MUST remain at the end of this
296 * structure and the 'cache' member must be the first as it
Joe Thornberaeed1422013-05-10 14:37:18 +0100297 * is used to determine the offset of the writethrough fields.
Mike Snitzer19b00922013-04-05 15:36:34 +0100298 */
Joe Thornbere2e74d62013-03-20 17:21:27 +0000299 struct cache *cache;
300 dm_cblock_t cblock;
Darrick J. Wongb844fe62013-04-05 15:36:32 +0100301 struct dm_bio_details bio_details;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000302};
303
304struct dm_cache_migration {
305 struct list_head list;
306 struct cache *cache;
307
308 unsigned long start_jiffies;
309 dm_oblock_t old_oblock;
310 dm_oblock_t new_oblock;
311 dm_cblock_t cblock;
312
313 bool err:1;
314 bool writeback:1;
315 bool demote:1;
316 bool promote:1;
Joe Thornberc9d28d52013-10-31 13:55:48 -0400317 bool requeue_holder:1;
Joe Thornber65790ff2013-11-08 16:39:50 +0000318 bool invalidate:1;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000319
320 struct dm_bio_prison_cell *old_ocell;
321 struct dm_bio_prison_cell *new_ocell;
322};
323
324/*
325 * Processing a bio in the worker thread may require these memory
326 * allocations. We prealloc to avoid deadlocks (the same worker thread
327 * frees them back to the mempool).
328 */
329struct prealloc {
330 struct dm_cache_migration *mg;
331 struct dm_bio_prison_cell *cell1;
332 struct dm_bio_prison_cell *cell2;
333};
334
335static void wake_worker(struct cache *cache)
336{
337 queue_work(cache->wq, &cache->worker);
338}
339
340/*----------------------------------------------------------------*/
341
342static struct dm_bio_prison_cell *alloc_prison_cell(struct cache *cache)
343{
344 /* FIXME: change to use a local slab. */
345 return dm_bio_prison_alloc_cell(cache->prison, GFP_NOWAIT);
346}
347
348static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell *cell)
349{
350 dm_bio_prison_free_cell(cache->prison, cell);
351}
352
353static int prealloc_data_structs(struct cache *cache, struct prealloc *p)
354{
355 if (!p->mg) {
356 p->mg = mempool_alloc(cache->migration_pool, GFP_NOWAIT);
357 if (!p->mg)
358 return -ENOMEM;
359 }
360
361 if (!p->cell1) {
362 p->cell1 = alloc_prison_cell(cache);
363 if (!p->cell1)
364 return -ENOMEM;
365 }
366
367 if (!p->cell2) {
368 p->cell2 = alloc_prison_cell(cache);
369 if (!p->cell2)
370 return -ENOMEM;
371 }
372
373 return 0;
374}
375
376static void prealloc_free_structs(struct cache *cache, struct prealloc *p)
377{
378 if (p->cell2)
379 free_prison_cell(cache, p->cell2);
380
381 if (p->cell1)
382 free_prison_cell(cache, p->cell1);
383
384 if (p->mg)
385 mempool_free(p->mg, cache->migration_pool);
386}
387
388static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p)
389{
390 struct dm_cache_migration *mg = p->mg;
391
392 BUG_ON(!mg);
393 p->mg = NULL;
394
395 return mg;
396}
397
398/*
399 * You must have a cell within the prealloc struct to return. If not this
400 * function will BUG() rather than returning NULL.
401 */
402static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p)
403{
404 struct dm_bio_prison_cell *r = NULL;
405
406 if (p->cell1) {
407 r = p->cell1;
408 p->cell1 = NULL;
409
410 } else if (p->cell2) {
411 r = p->cell2;
412 p->cell2 = NULL;
413 } else
414 BUG();
415
416 return r;
417}
418
419/*
420 * You can't have more than two cells in a prealloc struct. BUG() will be
421 * called if you try and overfill.
422 */
423static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell)
424{
425 if (!p->cell2)
426 p->cell2 = cell;
427
428 else if (!p->cell1)
429 p->cell1 = cell;
430
431 else
432 BUG();
433}
434
435/*----------------------------------------------------------------*/
436
437static void build_key(dm_oblock_t oblock, struct dm_cell_key *key)
438{
439 key->virtual = 0;
440 key->dev = 0;
441 key->block = from_oblock(oblock);
442}
443
444/*
445 * The caller hands in a preallocated cell, and a free function for it.
446 * The cell will be freed if there's an error, or if it wasn't used because
447 * a cell with that key already exists.
448 */
449typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell);
450
451static int bio_detain(struct cache *cache, dm_oblock_t oblock,
452 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
453 cell_free_fn free_fn, void *free_context,
454 struct dm_bio_prison_cell **cell_result)
455{
456 int r;
457 struct dm_cell_key key;
458
459 build_key(oblock, &key);
460 r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result);
461 if (r)
462 free_fn(free_context, cell_prealloc);
463
464 return r;
465}
466
467static int get_cell(struct cache *cache,
468 dm_oblock_t oblock,
469 struct prealloc *structs,
470 struct dm_bio_prison_cell **cell_result)
471{
472 int r;
473 struct dm_cell_key key;
474 struct dm_bio_prison_cell *cell_prealloc;
475
476 cell_prealloc = prealloc_get_cell(structs);
477
478 build_key(oblock, &key);
479 r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result);
480 if (r)
481 prealloc_put_cell(structs, cell_prealloc);
482
483 return r;
484}
485
Joe Thornberaeed1422013-05-10 14:37:18 +0100486/*----------------------------------------------------------------*/
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000487
488static bool is_dirty(struct cache *cache, dm_cblock_t b)
489{
490 return test_bit(from_cblock(b), cache->dirty_bitset);
491}
492
493static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
494{
495 if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) {
496 cache->nr_dirty = to_cblock(from_cblock(cache->nr_dirty) + 1);
497 policy_set_dirty(cache->policy, oblock);
498 }
499}
500
501static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
502{
503 if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) {
504 policy_clear_dirty(cache->policy, oblock);
505 cache->nr_dirty = to_cblock(from_cblock(cache->nr_dirty) - 1);
506 if (!from_cblock(cache->nr_dirty))
507 dm_table_event(cache->ti->table);
508 }
509}
510
511/*----------------------------------------------------------------*/
Joe Thornberaeed1422013-05-10 14:37:18 +0100512
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000513static bool block_size_is_power_of_two(struct cache *cache)
514{
515 return cache->sectors_per_block_shift >= 0;
516}
517
Mikulas Patocka43aeaa22013-07-10 23:41:17 +0100518/* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */
519#if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6
520__always_inline
521#endif
Joe Thornber414dd672013-03-20 17:21:25 +0000522static dm_block_t block_div(dm_block_t b, uint32_t n)
523{
524 do_div(b, n);
525
526 return b;
527}
528
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000529static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock)
530{
Joe Thornber414dd672013-03-20 17:21:25 +0000531 uint32_t discard_blocks = cache->discard_block_size;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000532 dm_block_t b = from_oblock(oblock);
533
534 if (!block_size_is_power_of_two(cache))
Joe Thornber414dd672013-03-20 17:21:25 +0000535 discard_blocks = discard_blocks / cache->sectors_per_block;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000536 else
537 discard_blocks >>= cache->sectors_per_block_shift;
538
Joe Thornber414dd672013-03-20 17:21:25 +0000539 b = block_div(b, discard_blocks);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000540
541 return to_dblock(b);
542}
543
544static void set_discard(struct cache *cache, dm_dblock_t b)
545{
546 unsigned long flags;
547
548 atomic_inc(&cache->stats.discard_count);
549
550 spin_lock_irqsave(&cache->lock, flags);
551 set_bit(from_dblock(b), cache->discard_bitset);
552 spin_unlock_irqrestore(&cache->lock, flags);
553}
554
555static void clear_discard(struct cache *cache, dm_dblock_t b)
556{
557 unsigned long flags;
558
559 spin_lock_irqsave(&cache->lock, flags);
560 clear_bit(from_dblock(b), cache->discard_bitset);
561 spin_unlock_irqrestore(&cache->lock, flags);
562}
563
564static bool is_discarded(struct cache *cache, dm_dblock_t b)
565{
566 int r;
567 unsigned long flags;
568
569 spin_lock_irqsave(&cache->lock, flags);
570 r = test_bit(from_dblock(b), cache->discard_bitset);
571 spin_unlock_irqrestore(&cache->lock, flags);
572
573 return r;
574}
575
576static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b)
577{
578 int r;
579 unsigned long flags;
580
581 spin_lock_irqsave(&cache->lock, flags);
582 r = test_bit(from_dblock(oblock_to_dblock(cache, b)),
583 cache->discard_bitset);
584 spin_unlock_irqrestore(&cache->lock, flags);
585
586 return r;
587}
588
589/*----------------------------------------------------------------*/
590
591static void load_stats(struct cache *cache)
592{
593 struct dm_cache_statistics stats;
594
595 dm_cache_metadata_get_stats(cache->cmd, &stats);
596 atomic_set(&cache->stats.read_hit, stats.read_hits);
597 atomic_set(&cache->stats.read_miss, stats.read_misses);
598 atomic_set(&cache->stats.write_hit, stats.write_hits);
599 atomic_set(&cache->stats.write_miss, stats.write_misses);
600}
601
602static void save_stats(struct cache *cache)
603{
604 struct dm_cache_statistics stats;
605
606 stats.read_hits = atomic_read(&cache->stats.read_hit);
607 stats.read_misses = atomic_read(&cache->stats.read_miss);
608 stats.write_hits = atomic_read(&cache->stats.write_hit);
609 stats.write_misses = atomic_read(&cache->stats.write_miss);
610
611 dm_cache_metadata_set_stats(cache->cmd, &stats);
612}
613
614/*----------------------------------------------------------------
615 * Per bio data
616 *--------------------------------------------------------------*/
Mike Snitzer19b00922013-04-05 15:36:34 +0100617
618/*
619 * If using writeback, leave out struct per_bio_data's writethrough fields.
620 */
621#define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache))
622#define PB_DATA_SIZE_WT (sizeof(struct per_bio_data))
623
Joe Thornber2ee57d52013-10-24 14:10:29 -0400624static bool writethrough_mode(struct cache_features *f)
625{
626 return f->io_mode == CM_IO_WRITETHROUGH;
627}
628
629static bool writeback_mode(struct cache_features *f)
630{
631 return f->io_mode == CM_IO_WRITEBACK;
632}
633
634static bool passthrough_mode(struct cache_features *f)
635{
636 return f->io_mode == CM_IO_PASSTHROUGH;
637}
638
Mike Snitzer19b00922013-04-05 15:36:34 +0100639static size_t get_per_bio_data_size(struct cache *cache)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000640{
Joe Thornber2ee57d52013-10-24 14:10:29 -0400641 return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB;
Mike Snitzer19b00922013-04-05 15:36:34 +0100642}
643
644static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size)
645{
646 struct per_bio_data *pb = dm_per_bio_data(bio, data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000647 BUG_ON(!pb);
648 return pb;
649}
650
Mike Snitzer19b00922013-04-05 15:36:34 +0100651static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000652{
Mike Snitzer19b00922013-04-05 15:36:34 +0100653 struct per_bio_data *pb = get_per_bio_data(bio, data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000654
655 pb->tick = false;
656 pb->req_nr = dm_bio_get_target_bio_nr(bio);
657 pb->all_io_entry = NULL;
658
659 return pb;
660}
661
662/*----------------------------------------------------------------
663 * Remapping
664 *--------------------------------------------------------------*/
665static void remap_to_origin(struct cache *cache, struct bio *bio)
666{
667 bio->bi_bdev = cache->origin_dev->bdev;
668}
669
670static void remap_to_cache(struct cache *cache, struct bio *bio,
671 dm_cblock_t cblock)
672{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700673 sector_t bi_sector = bio->bi_iter.bi_sector;
Heinz Mauelshagene0d849f2014-02-27 22:46:48 +0100674 sector_t block = from_cblock(cblock);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000675
676 bio->bi_bdev = cache->cache_dev->bdev;
677 if (!block_size_is_power_of_two(cache))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700678 bio->bi_iter.bi_sector =
Heinz Mauelshagene0d849f2014-02-27 22:46:48 +0100679 (block * cache->sectors_per_block) +
Kent Overstreet4f024f32013-10-11 15:44:27 -0700680 sector_div(bi_sector, cache->sectors_per_block);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000681 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700682 bio->bi_iter.bi_sector =
Heinz Mauelshagene0d849f2014-02-27 22:46:48 +0100683 (block << cache->sectors_per_block_shift) |
Kent Overstreet4f024f32013-10-11 15:44:27 -0700684 (bi_sector & (cache->sectors_per_block - 1));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000685}
686
687static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio)
688{
689 unsigned long flags;
Mike Snitzer19b00922013-04-05 15:36:34 +0100690 size_t pb_data_size = get_per_bio_data_size(cache);
691 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000692
693 spin_lock_irqsave(&cache->lock, flags);
694 if (cache->need_tick_bio &&
695 !(bio->bi_rw & (REQ_FUA | REQ_FLUSH | REQ_DISCARD))) {
696 pb->tick = true;
697 cache->need_tick_bio = false;
698 }
699 spin_unlock_irqrestore(&cache->lock, flags);
700}
701
702static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio,
703 dm_oblock_t oblock)
704{
705 check_if_tick_bio_needed(cache, bio);
706 remap_to_origin(cache, bio);
707 if (bio_data_dir(bio) == WRITE)
708 clear_discard(cache, oblock_to_dblock(cache, oblock));
709}
710
711static void remap_to_cache_dirty(struct cache *cache, struct bio *bio,
712 dm_oblock_t oblock, dm_cblock_t cblock)
713{
Joe Thornberf8e5f012013-10-21 12:51:45 +0100714 check_if_tick_bio_needed(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000715 remap_to_cache(cache, bio, cblock);
716 if (bio_data_dir(bio) == WRITE) {
717 set_dirty(cache, oblock, cblock);
718 clear_discard(cache, oblock_to_dblock(cache, oblock));
719 }
720}
721
722static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio)
723{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700724 sector_t block_nr = bio->bi_iter.bi_sector;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000725
726 if (!block_size_is_power_of_two(cache))
727 (void) sector_div(block_nr, cache->sectors_per_block);
728 else
729 block_nr >>= cache->sectors_per_block_shift;
730
731 return to_oblock(block_nr);
732}
733
734static int bio_triggers_commit(struct cache *cache, struct bio *bio)
735{
736 return bio->bi_rw & (REQ_FLUSH | REQ_FUA);
737}
738
739static void issue(struct cache *cache, struct bio *bio)
740{
741 unsigned long flags;
742
743 if (!bio_triggers_commit(cache, bio)) {
744 generic_make_request(bio);
745 return;
746 }
747
748 /*
749 * Batch together any bios that trigger commits and then issue a
750 * single commit for them in do_worker().
751 */
752 spin_lock_irqsave(&cache->lock, flags);
753 cache->commit_requested = true;
754 bio_list_add(&cache->deferred_flush_bios, bio);
755 spin_unlock_irqrestore(&cache->lock, flags);
756}
757
Joe Thornbere2e74d62013-03-20 17:21:27 +0000758static void defer_writethrough_bio(struct cache *cache, struct bio *bio)
759{
760 unsigned long flags;
761
762 spin_lock_irqsave(&cache->lock, flags);
763 bio_list_add(&cache->deferred_writethrough_bios, bio);
764 spin_unlock_irqrestore(&cache->lock, flags);
765
766 wake_worker(cache);
767}
768
769static void writethrough_endio(struct bio *bio, int err)
770{
Mike Snitzer19b00922013-04-05 15:36:34 +0100771 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
Joe Thornberc9d28d52013-10-31 13:55:48 -0400772
773 dm_unhook_bio(&pb->hook_info, bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000774
775 if (err) {
776 bio_endio(bio, err);
777 return;
778 }
779
Darrick J. Wongb844fe62013-04-05 15:36:32 +0100780 dm_bio_restore(&pb->bio_details, bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000781 remap_to_cache(pb->cache, bio, pb->cblock);
782
783 /*
784 * We can't issue this bio directly, since we're in interrupt
Joe Thornberaeed1422013-05-10 14:37:18 +0100785 * context. So it gets put on a bio list for processing by the
Joe Thornbere2e74d62013-03-20 17:21:27 +0000786 * worker thread.
787 */
788 defer_writethrough_bio(pb->cache, bio);
789}
790
791/*
792 * When running in writethrough mode we need to send writes to clean blocks
793 * to both the cache and origin devices. In future we'd like to clone the
794 * bio and send them in parallel, but for now we're doing them in
795 * series as this is easier.
796 */
797static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio,
798 dm_oblock_t oblock, dm_cblock_t cblock)
799{
Mike Snitzer19b00922013-04-05 15:36:34 +0100800 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000801
802 pb->cache = cache;
803 pb->cblock = cblock;
Joe Thornberc9d28d52013-10-31 13:55:48 -0400804 dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL);
Darrick J. Wongb844fe62013-04-05 15:36:32 +0100805 dm_bio_record(&pb->bio_details, bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000806
807 remap_to_origin_clear_discard(pb->cache, bio, oblock);
808}
809
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000810/*----------------------------------------------------------------
811 * Migration processing
812 *
813 * Migration covers moving data from the origin device to the cache, or
814 * vice versa.
815 *--------------------------------------------------------------*/
816static void free_migration(struct dm_cache_migration *mg)
817{
818 mempool_free(mg, mg->cache->migration_pool);
819}
820
821static void inc_nr_migrations(struct cache *cache)
822{
823 atomic_inc(&cache->nr_migrations);
824}
825
826static void dec_nr_migrations(struct cache *cache)
827{
828 atomic_dec(&cache->nr_migrations);
829
830 /*
831 * Wake the worker in case we're suspending the target.
832 */
833 wake_up(&cache->migration_wait);
834}
835
836static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell,
837 bool holder)
838{
839 (holder ? dm_cell_release : dm_cell_release_no_holder)
840 (cache->prison, cell, &cache->deferred_bios);
841 free_prison_cell(cache, cell);
842}
843
844static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell,
845 bool holder)
846{
847 unsigned long flags;
848
849 spin_lock_irqsave(&cache->lock, flags);
850 __cell_defer(cache, cell, holder);
851 spin_unlock_irqrestore(&cache->lock, flags);
852
853 wake_worker(cache);
854}
855
856static void cleanup_migration(struct dm_cache_migration *mg)
857{
Joe Thornber66cb1912013-10-30 17:11:58 +0000858 struct cache *cache = mg->cache;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000859 free_migration(mg);
Joe Thornber66cb1912013-10-30 17:11:58 +0000860 dec_nr_migrations(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000861}
862
863static void migration_failure(struct dm_cache_migration *mg)
864{
865 struct cache *cache = mg->cache;
866
867 if (mg->writeback) {
868 DMWARN_LIMIT("writeback failed; couldn't copy block");
869 set_dirty(cache, mg->old_oblock, mg->cblock);
870 cell_defer(cache, mg->old_ocell, false);
871
872 } else if (mg->demote) {
873 DMWARN_LIMIT("demotion failed; couldn't copy block");
874 policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock);
875
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +0200876 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000877 if (mg->promote)
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +0200878 cell_defer(cache, mg->new_ocell, true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000879 } else {
880 DMWARN_LIMIT("promotion failed; couldn't copy block");
881 policy_remove_mapping(cache->policy, mg->new_oblock);
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +0200882 cell_defer(cache, mg->new_ocell, true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000883 }
884
885 cleanup_migration(mg);
886}
887
888static void migration_success_pre_commit(struct dm_cache_migration *mg)
889{
890 unsigned long flags;
891 struct cache *cache = mg->cache;
892
893 if (mg->writeback) {
894 cell_defer(cache, mg->old_ocell, false);
895 clear_dirty(cache, mg->old_oblock, mg->cblock);
896 cleanup_migration(mg);
897 return;
898
899 } else if (mg->demote) {
900 if (dm_cache_remove_mapping(cache->cmd, mg->cblock)) {
901 DMWARN_LIMIT("demotion failed; couldn't update on disk metadata");
902 policy_force_mapping(cache->policy, mg->new_oblock,
903 mg->old_oblock);
904 if (mg->promote)
905 cell_defer(cache, mg->new_ocell, true);
906 cleanup_migration(mg);
907 return;
908 }
909 } else {
910 if (dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock)) {
911 DMWARN_LIMIT("promotion failed; couldn't update on disk metadata");
912 policy_remove_mapping(cache->policy, mg->new_oblock);
913 cleanup_migration(mg);
914 return;
915 }
916 }
917
918 spin_lock_irqsave(&cache->lock, flags);
919 list_add_tail(&mg->list, &cache->need_commit_migrations);
920 cache->commit_requested = true;
921 spin_unlock_irqrestore(&cache->lock, flags);
922}
923
924static void migration_success_post_commit(struct dm_cache_migration *mg)
925{
926 unsigned long flags;
927 struct cache *cache = mg->cache;
928
929 if (mg->writeback) {
930 DMWARN("writeback unexpectedly triggered commit");
931 return;
932
933 } else if (mg->demote) {
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +0200934 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000935
936 if (mg->promote) {
937 mg->demote = false;
938
939 spin_lock_irqsave(&cache->lock, flags);
940 list_add_tail(&mg->list, &cache->quiesced_migrations);
941 spin_unlock_irqrestore(&cache->lock, flags);
942
Joe Thornber65790ff2013-11-08 16:39:50 +0000943 } else {
944 if (mg->invalidate)
945 policy_remove_mapping(cache->policy, mg->old_oblock);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000946 cleanup_migration(mg);
Joe Thornber65790ff2013-11-08 16:39:50 +0000947 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000948
949 } else {
Joe Thornberc9d28d52013-10-31 13:55:48 -0400950 if (mg->requeue_holder)
951 cell_defer(cache, mg->new_ocell, true);
952 else {
953 bio_endio(mg->new_ocell->holder, 0);
954 cell_defer(cache, mg->new_ocell, false);
955 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000956 clear_dirty(cache, mg->new_oblock, mg->cblock);
957 cleanup_migration(mg);
958 }
959}
960
961static void copy_complete(int read_err, unsigned long write_err, void *context)
962{
963 unsigned long flags;
964 struct dm_cache_migration *mg = (struct dm_cache_migration *) context;
965 struct cache *cache = mg->cache;
966
967 if (read_err || write_err)
968 mg->err = true;
969
970 spin_lock_irqsave(&cache->lock, flags);
971 list_add_tail(&mg->list, &cache->completed_migrations);
972 spin_unlock_irqrestore(&cache->lock, flags);
973
974 wake_worker(cache);
975}
976
977static void issue_copy_real(struct dm_cache_migration *mg)
978{
979 int r;
980 struct dm_io_region o_region, c_region;
981 struct cache *cache = mg->cache;
982
983 o_region.bdev = cache->origin_dev->bdev;
984 o_region.count = cache->sectors_per_block;
985
986 c_region.bdev = cache->cache_dev->bdev;
987 c_region.sector = from_cblock(mg->cblock) * cache->sectors_per_block;
988 c_region.count = cache->sectors_per_block;
989
990 if (mg->writeback || mg->demote) {
991 /* demote */
992 o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block;
993 r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg);
994 } else {
995 /* promote */
996 o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block;
997 r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg);
998 }
999
Heinz Mauelshagen2c2263c2013-10-14 17:14:45 +02001000 if (r < 0) {
1001 DMERR_LIMIT("issuing migration failed");
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001002 migration_failure(mg);
Heinz Mauelshagen2c2263c2013-10-14 17:14:45 +02001003 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001004}
1005
Joe Thornberc9d28d52013-10-31 13:55:48 -04001006static void overwrite_endio(struct bio *bio, int err)
1007{
1008 struct dm_cache_migration *mg = bio->bi_private;
1009 struct cache *cache = mg->cache;
1010 size_t pb_data_size = get_per_bio_data_size(cache);
1011 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1012 unsigned long flags;
1013
Mike Snitzer80ae49a2014-01-31 14:30:37 -05001014 dm_unhook_bio(&pb->hook_info, bio);
1015
Joe Thornberc9d28d52013-10-31 13:55:48 -04001016 if (err)
1017 mg->err = true;
1018
Mike Snitzer80ae49a2014-01-31 14:30:37 -05001019 mg->requeue_holder = false;
1020
Joe Thornberc9d28d52013-10-31 13:55:48 -04001021 spin_lock_irqsave(&cache->lock, flags);
1022 list_add_tail(&mg->list, &cache->completed_migrations);
Joe Thornberc9d28d52013-10-31 13:55:48 -04001023 spin_unlock_irqrestore(&cache->lock, flags);
1024
1025 wake_worker(cache);
1026}
1027
1028static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio)
1029{
1030 size_t pb_data_size = get_per_bio_data_size(mg->cache);
1031 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1032
1033 dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg);
1034 remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock);
1035 generic_make_request(bio);
1036}
1037
1038static bool bio_writes_complete_block(struct cache *cache, struct bio *bio)
1039{
1040 return (bio_data_dir(bio) == WRITE) &&
Kent Overstreet4f024f32013-10-11 15:44:27 -07001041 (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT));
Joe Thornberc9d28d52013-10-31 13:55:48 -04001042}
1043
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001044static void avoid_copy(struct dm_cache_migration *mg)
1045{
1046 atomic_inc(&mg->cache->stats.copies_avoided);
1047 migration_success_pre_commit(mg);
1048}
1049
1050static void issue_copy(struct dm_cache_migration *mg)
1051{
1052 bool avoid;
1053 struct cache *cache = mg->cache;
1054
1055 if (mg->writeback || mg->demote)
1056 avoid = !is_dirty(cache, mg->cblock) ||
1057 is_discarded_oblock(cache, mg->old_oblock);
Joe Thornberc9d28d52013-10-31 13:55:48 -04001058 else {
1059 struct bio *bio = mg->new_ocell->holder;
1060
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001061 avoid = is_discarded_oblock(cache, mg->new_oblock);
1062
Joe Thornberc9d28d52013-10-31 13:55:48 -04001063 if (!avoid && bio_writes_complete_block(cache, bio)) {
1064 issue_overwrite(mg, bio);
1065 return;
1066 }
1067 }
1068
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001069 avoid ? avoid_copy(mg) : issue_copy_real(mg);
1070}
1071
1072static void complete_migration(struct dm_cache_migration *mg)
1073{
1074 if (mg->err)
1075 migration_failure(mg);
1076 else
1077 migration_success_pre_commit(mg);
1078}
1079
1080static void process_migrations(struct cache *cache, struct list_head *head,
1081 void (*fn)(struct dm_cache_migration *))
1082{
1083 unsigned long flags;
1084 struct list_head list;
1085 struct dm_cache_migration *mg, *tmp;
1086
1087 INIT_LIST_HEAD(&list);
1088 spin_lock_irqsave(&cache->lock, flags);
1089 list_splice_init(head, &list);
1090 spin_unlock_irqrestore(&cache->lock, flags);
1091
1092 list_for_each_entry_safe(mg, tmp, &list, list)
1093 fn(mg);
1094}
1095
1096static void __queue_quiesced_migration(struct dm_cache_migration *mg)
1097{
1098 list_add_tail(&mg->list, &mg->cache->quiesced_migrations);
1099}
1100
1101static void queue_quiesced_migration(struct dm_cache_migration *mg)
1102{
1103 unsigned long flags;
1104 struct cache *cache = mg->cache;
1105
1106 spin_lock_irqsave(&cache->lock, flags);
1107 __queue_quiesced_migration(mg);
1108 spin_unlock_irqrestore(&cache->lock, flags);
1109
1110 wake_worker(cache);
1111}
1112
1113static void queue_quiesced_migrations(struct cache *cache, struct list_head *work)
1114{
1115 unsigned long flags;
1116 struct dm_cache_migration *mg, *tmp;
1117
1118 spin_lock_irqsave(&cache->lock, flags);
1119 list_for_each_entry_safe(mg, tmp, work, list)
1120 __queue_quiesced_migration(mg);
1121 spin_unlock_irqrestore(&cache->lock, flags);
1122
1123 wake_worker(cache);
1124}
1125
1126static void check_for_quiesced_migrations(struct cache *cache,
1127 struct per_bio_data *pb)
1128{
1129 struct list_head work;
1130
1131 if (!pb->all_io_entry)
1132 return;
1133
1134 INIT_LIST_HEAD(&work);
1135 if (pb->all_io_entry)
1136 dm_deferred_entry_dec(pb->all_io_entry, &work);
1137
1138 if (!list_empty(&work))
1139 queue_quiesced_migrations(cache, &work);
1140}
1141
1142static void quiesce_migration(struct dm_cache_migration *mg)
1143{
1144 if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list))
1145 queue_quiesced_migration(mg);
1146}
1147
1148static void promote(struct cache *cache, struct prealloc *structs,
1149 dm_oblock_t oblock, dm_cblock_t cblock,
1150 struct dm_bio_prison_cell *cell)
1151{
1152 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1153
1154 mg->err = false;
1155 mg->writeback = false;
1156 mg->demote = false;
1157 mg->promote = true;
Joe Thornberc9d28d52013-10-31 13:55:48 -04001158 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001159 mg->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001160 mg->cache = cache;
1161 mg->new_oblock = oblock;
1162 mg->cblock = cblock;
1163 mg->old_ocell = NULL;
1164 mg->new_ocell = cell;
1165 mg->start_jiffies = jiffies;
1166
1167 inc_nr_migrations(cache);
1168 quiesce_migration(mg);
1169}
1170
1171static void writeback(struct cache *cache, struct prealloc *structs,
1172 dm_oblock_t oblock, dm_cblock_t cblock,
1173 struct dm_bio_prison_cell *cell)
1174{
1175 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1176
1177 mg->err = false;
1178 mg->writeback = true;
1179 mg->demote = false;
1180 mg->promote = false;
Joe Thornberc9d28d52013-10-31 13:55:48 -04001181 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001182 mg->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001183 mg->cache = cache;
1184 mg->old_oblock = oblock;
1185 mg->cblock = cblock;
1186 mg->old_ocell = cell;
1187 mg->new_ocell = NULL;
1188 mg->start_jiffies = jiffies;
1189
1190 inc_nr_migrations(cache);
1191 quiesce_migration(mg);
1192}
1193
1194static void demote_then_promote(struct cache *cache, struct prealloc *structs,
1195 dm_oblock_t old_oblock, dm_oblock_t new_oblock,
1196 dm_cblock_t cblock,
1197 struct dm_bio_prison_cell *old_ocell,
1198 struct dm_bio_prison_cell *new_ocell)
1199{
1200 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1201
1202 mg->err = false;
1203 mg->writeback = false;
1204 mg->demote = true;
1205 mg->promote = true;
Joe Thornberc9d28d52013-10-31 13:55:48 -04001206 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001207 mg->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001208 mg->cache = cache;
1209 mg->old_oblock = old_oblock;
1210 mg->new_oblock = new_oblock;
1211 mg->cblock = cblock;
1212 mg->old_ocell = old_ocell;
1213 mg->new_ocell = new_ocell;
1214 mg->start_jiffies = jiffies;
1215
1216 inc_nr_migrations(cache);
1217 quiesce_migration(mg);
1218}
1219
Joe Thornber2ee57d52013-10-24 14:10:29 -04001220/*
1221 * Invalidate a cache entry. No writeback occurs; any changes in the cache
1222 * block are thrown away.
1223 */
1224static void invalidate(struct cache *cache, struct prealloc *structs,
1225 dm_oblock_t oblock, dm_cblock_t cblock,
1226 struct dm_bio_prison_cell *cell)
1227{
1228 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1229
1230 mg->err = false;
1231 mg->writeback = false;
1232 mg->demote = true;
1233 mg->promote = false;
1234 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001235 mg->invalidate = true;
Joe Thornber2ee57d52013-10-24 14:10:29 -04001236 mg->cache = cache;
1237 mg->old_oblock = oblock;
1238 mg->cblock = cblock;
1239 mg->old_ocell = cell;
1240 mg->new_ocell = NULL;
1241 mg->start_jiffies = jiffies;
1242
1243 inc_nr_migrations(cache);
1244 quiesce_migration(mg);
1245}
1246
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001247/*----------------------------------------------------------------
1248 * bio processing
1249 *--------------------------------------------------------------*/
1250static void defer_bio(struct cache *cache, struct bio *bio)
1251{
1252 unsigned long flags;
1253
1254 spin_lock_irqsave(&cache->lock, flags);
1255 bio_list_add(&cache->deferred_bios, bio);
1256 spin_unlock_irqrestore(&cache->lock, flags);
1257
1258 wake_worker(cache);
1259}
1260
1261static void process_flush_bio(struct cache *cache, struct bio *bio)
1262{
Mike Snitzer19b00922013-04-05 15:36:34 +01001263 size_t pb_data_size = get_per_bio_data_size(cache);
1264 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001265
Kent Overstreet4f024f32013-10-11 15:44:27 -07001266 BUG_ON(bio->bi_iter.bi_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001267 if (!pb->req_nr)
1268 remap_to_origin(cache, bio);
1269 else
1270 remap_to_cache(cache, bio, 0);
1271
1272 issue(cache, bio);
1273}
1274
1275/*
1276 * People generally discard large parts of a device, eg, the whole device
1277 * when formatting. Splitting these large discards up into cache block
1278 * sized ios and then quiescing (always neccessary for discard) takes too
1279 * long.
1280 *
1281 * We keep it simple, and allow any size of discard to come in, and just
1282 * mark off blocks on the discard bitset. No passdown occurs!
1283 *
1284 * To implement passdown we need to change the bio_prison such that a cell
1285 * can have a key that spans many blocks.
1286 */
1287static void process_discard_bio(struct cache *cache, struct bio *bio)
1288{
Kent Overstreet4f024f32013-10-11 15:44:27 -07001289 dm_block_t start_block = dm_sector_div_up(bio->bi_iter.bi_sector,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001290 cache->discard_block_size);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001291 dm_block_t end_block = bio_end_sector(bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001292 dm_block_t b;
1293
Joe Thornber414dd672013-03-20 17:21:25 +00001294 end_block = block_div(end_block, cache->discard_block_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001295
1296 for (b = start_block; b < end_block; b++)
1297 set_discard(cache, to_dblock(b));
1298
1299 bio_endio(bio, 0);
1300}
1301
1302static bool spare_migration_bandwidth(struct cache *cache)
1303{
1304 sector_t current_volume = (atomic_read(&cache->nr_migrations) + 1) *
1305 cache->sectors_per_block;
1306 return current_volume < cache->migration_threshold;
1307}
1308
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001309static void inc_hit_counter(struct cache *cache, struct bio *bio)
1310{
1311 atomic_inc(bio_data_dir(bio) == READ ?
1312 &cache->stats.read_hit : &cache->stats.write_hit);
1313}
1314
1315static void inc_miss_counter(struct cache *cache, struct bio *bio)
1316{
1317 atomic_inc(bio_data_dir(bio) == READ ?
1318 &cache->stats.read_miss : &cache->stats.write_miss);
1319}
1320
Joe Thornber2ee57d52013-10-24 14:10:29 -04001321static void issue_cache_bio(struct cache *cache, struct bio *bio,
1322 struct per_bio_data *pb,
1323 dm_oblock_t oblock, dm_cblock_t cblock)
1324{
1325 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
1326 remap_to_cache_dirty(cache, bio, oblock, cblock);
1327 issue(cache, bio);
1328}
1329
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001330static void process_bio(struct cache *cache, struct prealloc *structs,
1331 struct bio *bio)
1332{
1333 int r;
1334 bool release_cell = true;
1335 dm_oblock_t block = get_bio_block(cache, bio);
1336 struct dm_bio_prison_cell *cell_prealloc, *old_ocell, *new_ocell;
1337 struct policy_result lookup_result;
Mike Snitzer19b00922013-04-05 15:36:34 +01001338 size_t pb_data_size = get_per_bio_data_size(cache);
1339 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001340 bool discarded_block = is_discarded_oblock(cache, block);
Joe Thornber2ee57d52013-10-24 14:10:29 -04001341 bool passthrough = passthrough_mode(&cache->features);
1342 bool can_migrate = !passthrough && (discarded_block || spare_migration_bandwidth(cache));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001343
1344 /*
1345 * Check to see if that block is currently migrating.
1346 */
1347 cell_prealloc = prealloc_get_cell(structs);
1348 r = bio_detain(cache, block, bio, cell_prealloc,
1349 (cell_free_fn) prealloc_put_cell,
1350 structs, &new_ocell);
1351 if (r > 0)
1352 return;
1353
1354 r = policy_map(cache->policy, block, true, can_migrate, discarded_block,
1355 bio, &lookup_result);
1356
1357 if (r == -EWOULDBLOCK)
1358 /* migration has been denied */
1359 lookup_result.op = POLICY_MISS;
1360
1361 switch (lookup_result.op) {
1362 case POLICY_HIT:
Joe Thornber2ee57d52013-10-24 14:10:29 -04001363 if (passthrough) {
1364 inc_miss_counter(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001365
Joe Thornber2ee57d52013-10-24 14:10:29 -04001366 /*
1367 * Passthrough always maps to the origin,
1368 * invalidating any cache blocks that are written
1369 * to.
1370 */
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001371
Joe Thornber2ee57d52013-10-24 14:10:29 -04001372 if (bio_data_dir(bio) == WRITE) {
1373 atomic_inc(&cache->stats.demotion);
1374 invalidate(cache, structs, block, lookup_result.cblock, new_ocell);
1375 release_cell = false;
1376
1377 } else {
1378 /* FIXME: factor out issue_origin() */
1379 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
1380 remap_to_origin_clear_discard(cache, bio, block);
1381 issue(cache, bio);
1382 }
1383 } else {
1384 inc_hit_counter(cache, bio);
1385
1386 if (bio_data_dir(bio) == WRITE &&
1387 writethrough_mode(&cache->features) &&
1388 !is_dirty(cache, lookup_result.cblock)) {
1389 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
1390 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
1391 issue(cache, bio);
1392 } else
1393 issue_cache_bio(cache, bio, pb, block, lookup_result.cblock);
1394 }
1395
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001396 break;
1397
1398 case POLICY_MISS:
1399 inc_miss_counter(cache, bio);
1400 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
Joe Thornbere2e74d62013-03-20 17:21:27 +00001401 remap_to_origin_clear_discard(cache, bio, block);
1402 issue(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001403 break;
1404
1405 case POLICY_NEW:
1406 atomic_inc(&cache->stats.promotion);
1407 promote(cache, structs, block, lookup_result.cblock, new_ocell);
1408 release_cell = false;
1409 break;
1410
1411 case POLICY_REPLACE:
1412 cell_prealloc = prealloc_get_cell(structs);
1413 r = bio_detain(cache, lookup_result.old_oblock, bio, cell_prealloc,
1414 (cell_free_fn) prealloc_put_cell,
1415 structs, &old_ocell);
1416 if (r > 0) {
1417 /*
1418 * We have to be careful to avoid lock inversion of
1419 * the cells. So we back off, and wait for the
1420 * old_ocell to become free.
1421 */
1422 policy_force_mapping(cache->policy, block,
1423 lookup_result.old_oblock);
1424 atomic_inc(&cache->stats.cache_cell_clash);
1425 break;
1426 }
1427 atomic_inc(&cache->stats.demotion);
1428 atomic_inc(&cache->stats.promotion);
1429
1430 demote_then_promote(cache, structs, lookup_result.old_oblock,
1431 block, lookup_result.cblock,
1432 old_ocell, new_ocell);
1433 release_cell = false;
1434 break;
1435
1436 default:
1437 DMERR_LIMIT("%s: erroring bio, unknown policy op: %u", __func__,
1438 (unsigned) lookup_result.op);
1439 bio_io_error(bio);
1440 }
1441
1442 if (release_cell)
1443 cell_defer(cache, new_ocell, false);
1444}
1445
1446static int need_commit_due_to_time(struct cache *cache)
1447{
1448 return jiffies < cache->last_commit_jiffies ||
1449 jiffies > cache->last_commit_jiffies + COMMIT_PERIOD;
1450}
1451
1452static int commit_if_needed(struct cache *cache)
1453{
Heinz Mauelshagenffcbcb62013-10-14 17:24:43 +02001454 int r = 0;
1455
1456 if ((cache->commit_requested || need_commit_due_to_time(cache)) &&
1457 dm_cache_changed_this_transaction(cache->cmd)) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001458 atomic_inc(&cache->stats.commit_count);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001459 cache->commit_requested = false;
Heinz Mauelshagenffcbcb62013-10-14 17:24:43 +02001460 r = dm_cache_commit(cache->cmd, false);
1461 cache->last_commit_jiffies = jiffies;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001462 }
1463
Heinz Mauelshagenffcbcb62013-10-14 17:24:43 +02001464 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001465}
1466
1467static void process_deferred_bios(struct cache *cache)
1468{
1469 unsigned long flags;
1470 struct bio_list bios;
1471 struct bio *bio;
1472 struct prealloc structs;
1473
1474 memset(&structs, 0, sizeof(structs));
1475 bio_list_init(&bios);
1476
1477 spin_lock_irqsave(&cache->lock, flags);
1478 bio_list_merge(&bios, &cache->deferred_bios);
1479 bio_list_init(&cache->deferred_bios);
1480 spin_unlock_irqrestore(&cache->lock, flags);
1481
1482 while (!bio_list_empty(&bios)) {
1483 /*
1484 * If we've got no free migration structs, and processing
1485 * this bio might require one, we pause until there are some
1486 * prepared mappings to process.
1487 */
1488 if (prealloc_data_structs(cache, &structs)) {
1489 spin_lock_irqsave(&cache->lock, flags);
1490 bio_list_merge(&cache->deferred_bios, &bios);
1491 spin_unlock_irqrestore(&cache->lock, flags);
1492 break;
1493 }
1494
1495 bio = bio_list_pop(&bios);
1496
1497 if (bio->bi_rw & REQ_FLUSH)
1498 process_flush_bio(cache, bio);
1499 else if (bio->bi_rw & REQ_DISCARD)
1500 process_discard_bio(cache, bio);
1501 else
1502 process_bio(cache, &structs, bio);
1503 }
1504
1505 prealloc_free_structs(cache, &structs);
1506}
1507
1508static void process_deferred_flush_bios(struct cache *cache, bool submit_bios)
1509{
1510 unsigned long flags;
1511 struct bio_list bios;
1512 struct bio *bio;
1513
1514 bio_list_init(&bios);
1515
1516 spin_lock_irqsave(&cache->lock, flags);
1517 bio_list_merge(&bios, &cache->deferred_flush_bios);
1518 bio_list_init(&cache->deferred_flush_bios);
1519 spin_unlock_irqrestore(&cache->lock, flags);
1520
1521 while ((bio = bio_list_pop(&bios)))
1522 submit_bios ? generic_make_request(bio) : bio_io_error(bio);
1523}
1524
Joe Thornbere2e74d62013-03-20 17:21:27 +00001525static void process_deferred_writethrough_bios(struct cache *cache)
1526{
1527 unsigned long flags;
1528 struct bio_list bios;
1529 struct bio *bio;
1530
1531 bio_list_init(&bios);
1532
1533 spin_lock_irqsave(&cache->lock, flags);
1534 bio_list_merge(&bios, &cache->deferred_writethrough_bios);
1535 bio_list_init(&cache->deferred_writethrough_bios);
1536 spin_unlock_irqrestore(&cache->lock, flags);
1537
1538 while ((bio = bio_list_pop(&bios)))
1539 generic_make_request(bio);
1540}
1541
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001542static void writeback_some_dirty_blocks(struct cache *cache)
1543{
1544 int r = 0;
1545 dm_oblock_t oblock;
1546 dm_cblock_t cblock;
1547 struct prealloc structs;
1548 struct dm_bio_prison_cell *old_ocell;
1549
1550 memset(&structs, 0, sizeof(structs));
1551
1552 while (spare_migration_bandwidth(cache)) {
1553 if (prealloc_data_structs(cache, &structs))
1554 break;
1555
1556 r = policy_writeback_work(cache->policy, &oblock, &cblock);
1557 if (r)
1558 break;
1559
1560 r = get_cell(cache, oblock, &structs, &old_ocell);
1561 if (r) {
1562 policy_set_dirty(cache->policy, oblock);
1563 break;
1564 }
1565
1566 writeback(cache, &structs, oblock, cblock, old_ocell);
1567 }
1568
1569 prealloc_free_structs(cache, &structs);
1570}
1571
1572/*----------------------------------------------------------------
Joe Thornber65790ff2013-11-08 16:39:50 +00001573 * Invalidations.
1574 * Dropping something from the cache *without* writing back.
1575 *--------------------------------------------------------------*/
1576
1577static void process_invalidation_request(struct cache *cache, struct invalidation_request *req)
1578{
1579 int r = 0;
1580 uint64_t begin = from_cblock(req->cblocks->begin);
1581 uint64_t end = from_cblock(req->cblocks->end);
1582
1583 while (begin != end) {
1584 r = policy_remove_cblock(cache->policy, to_cblock(begin));
1585 if (!r) {
1586 r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin));
1587 if (r)
1588 break;
1589
1590 } else if (r == -ENODATA) {
1591 /* harmless, already unmapped */
1592 r = 0;
1593
1594 } else {
1595 DMERR("policy_remove_cblock failed");
1596 break;
1597 }
1598
1599 begin++;
1600 }
1601
1602 cache->commit_requested = true;
1603
1604 req->err = r;
1605 atomic_set(&req->complete, 1);
1606
1607 wake_up(&req->result_wait);
1608}
1609
1610static void process_invalidation_requests(struct cache *cache)
1611{
1612 struct list_head list;
1613 struct invalidation_request *req, *tmp;
1614
1615 INIT_LIST_HEAD(&list);
1616 spin_lock(&cache->invalidation_lock);
1617 list_splice_init(&cache->invalidation_requests, &list);
1618 spin_unlock(&cache->invalidation_lock);
1619
1620 list_for_each_entry_safe (req, tmp, &list, list)
1621 process_invalidation_request(cache, req);
1622}
1623
1624/*----------------------------------------------------------------
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001625 * Main worker loop
1626 *--------------------------------------------------------------*/
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001627static bool is_quiescing(struct cache *cache)
1628{
Joe Thornber238f8362013-10-30 17:29:30 +00001629 return atomic_read(&cache->quiescing);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001630}
1631
Joe Thornber66cb1912013-10-30 17:11:58 +00001632static void ack_quiescing(struct cache *cache)
1633{
1634 if (is_quiescing(cache)) {
1635 atomic_inc(&cache->quiescing_ack);
1636 wake_up(&cache->quiescing_wait);
1637 }
1638}
1639
1640static void wait_for_quiescing_ack(struct cache *cache)
1641{
1642 wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack));
1643}
1644
1645static void start_quiescing(struct cache *cache)
1646{
Joe Thornber238f8362013-10-30 17:29:30 +00001647 atomic_inc(&cache->quiescing);
Joe Thornber66cb1912013-10-30 17:11:58 +00001648 wait_for_quiescing_ack(cache);
1649}
1650
1651static void stop_quiescing(struct cache *cache)
1652{
Joe Thornber238f8362013-10-30 17:29:30 +00001653 atomic_set(&cache->quiescing, 0);
Joe Thornber66cb1912013-10-30 17:11:58 +00001654 atomic_set(&cache->quiescing_ack, 0);
1655}
1656
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001657static void wait_for_migrations(struct cache *cache)
1658{
1659 wait_event(cache->migration_wait, !atomic_read(&cache->nr_migrations));
1660}
1661
1662static void stop_worker(struct cache *cache)
1663{
1664 cancel_delayed_work(&cache->waker);
1665 flush_workqueue(cache->wq);
1666}
1667
1668static void requeue_deferred_io(struct cache *cache)
1669{
1670 struct bio *bio;
1671 struct bio_list bios;
1672
1673 bio_list_init(&bios);
1674 bio_list_merge(&bios, &cache->deferred_bios);
1675 bio_list_init(&cache->deferred_bios);
1676
1677 while ((bio = bio_list_pop(&bios)))
1678 bio_endio(bio, DM_ENDIO_REQUEUE);
1679}
1680
1681static int more_work(struct cache *cache)
1682{
1683 if (is_quiescing(cache))
1684 return !list_empty(&cache->quiesced_migrations) ||
1685 !list_empty(&cache->completed_migrations) ||
1686 !list_empty(&cache->need_commit_migrations);
1687 else
1688 return !bio_list_empty(&cache->deferred_bios) ||
1689 !bio_list_empty(&cache->deferred_flush_bios) ||
Joe Thornbere2e74d62013-03-20 17:21:27 +00001690 !bio_list_empty(&cache->deferred_writethrough_bios) ||
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001691 !list_empty(&cache->quiesced_migrations) ||
1692 !list_empty(&cache->completed_migrations) ||
Joe Thornber65790ff2013-11-08 16:39:50 +00001693 !list_empty(&cache->need_commit_migrations) ||
1694 cache->invalidate;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001695}
1696
1697static void do_worker(struct work_struct *ws)
1698{
1699 struct cache *cache = container_of(ws, struct cache, worker);
1700
1701 do {
Joe Thornber66cb1912013-10-30 17:11:58 +00001702 if (!is_quiescing(cache)) {
1703 writeback_some_dirty_blocks(cache);
1704 process_deferred_writethrough_bios(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001705 process_deferred_bios(cache);
Joe Thornber65790ff2013-11-08 16:39:50 +00001706 process_invalidation_requests(cache);
Joe Thornber66cb1912013-10-30 17:11:58 +00001707 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001708
1709 process_migrations(cache, &cache->quiesced_migrations, issue_copy);
1710 process_migrations(cache, &cache->completed_migrations, complete_migration);
1711
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001712 if (commit_if_needed(cache)) {
1713 process_deferred_flush_bios(cache, false);
1714
1715 /*
1716 * FIXME: rollback metadata or just go into a
1717 * failure mode and error everything
1718 */
1719 } else {
1720 process_deferred_flush_bios(cache, true);
1721 process_migrations(cache, &cache->need_commit_migrations,
1722 migration_success_post_commit);
1723 }
Joe Thornber66cb1912013-10-30 17:11:58 +00001724
1725 ack_quiescing(cache);
1726
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001727 } while (more_work(cache));
1728}
1729
1730/*
1731 * We want to commit periodically so that not too much
1732 * unwritten metadata builds up.
1733 */
1734static void do_waker(struct work_struct *ws)
1735{
1736 struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker);
Joe Thornberf8350da2013-05-10 14:37:16 +01001737 policy_tick(cache->policy);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001738 wake_worker(cache);
1739 queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD);
1740}
1741
1742/*----------------------------------------------------------------*/
1743
1744static int is_congested(struct dm_dev *dev, int bdi_bits)
1745{
1746 struct request_queue *q = bdev_get_queue(dev->bdev);
1747 return bdi_congested(&q->backing_dev_info, bdi_bits);
1748}
1749
1750static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1751{
1752 struct cache *cache = container_of(cb, struct cache, callbacks);
1753
1754 return is_congested(cache->origin_dev, bdi_bits) ||
1755 is_congested(cache->cache_dev, bdi_bits);
1756}
1757
1758/*----------------------------------------------------------------
1759 * Target methods
1760 *--------------------------------------------------------------*/
1761
1762/*
1763 * This function gets called on the error paths of the constructor, so we
1764 * have to cope with a partially initialised struct.
1765 */
1766static void destroy(struct cache *cache)
1767{
1768 unsigned i;
1769
1770 if (cache->next_migration)
1771 mempool_free(cache->next_migration, cache->migration_pool);
1772
1773 if (cache->migration_pool)
1774 mempool_destroy(cache->migration_pool);
1775
1776 if (cache->all_io_ds)
1777 dm_deferred_set_destroy(cache->all_io_ds);
1778
1779 if (cache->prison)
1780 dm_bio_prison_destroy(cache->prison);
1781
1782 if (cache->wq)
1783 destroy_workqueue(cache->wq);
1784
1785 if (cache->dirty_bitset)
1786 free_bitset(cache->dirty_bitset);
1787
1788 if (cache->discard_bitset)
1789 free_bitset(cache->discard_bitset);
1790
1791 if (cache->copier)
1792 dm_kcopyd_client_destroy(cache->copier);
1793
1794 if (cache->cmd)
1795 dm_cache_metadata_close(cache->cmd);
1796
1797 if (cache->metadata_dev)
1798 dm_put_device(cache->ti, cache->metadata_dev);
1799
1800 if (cache->origin_dev)
1801 dm_put_device(cache->ti, cache->origin_dev);
1802
1803 if (cache->cache_dev)
1804 dm_put_device(cache->ti, cache->cache_dev);
1805
1806 if (cache->policy)
1807 dm_cache_policy_destroy(cache->policy);
1808
1809 for (i = 0; i < cache->nr_ctr_args ; i++)
1810 kfree(cache->ctr_args[i]);
1811 kfree(cache->ctr_args);
1812
1813 kfree(cache);
1814}
1815
1816static void cache_dtr(struct dm_target *ti)
1817{
1818 struct cache *cache = ti->private;
1819
1820 destroy(cache);
1821}
1822
1823static sector_t get_dev_size(struct dm_dev *dev)
1824{
1825 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
1826}
1827
1828/*----------------------------------------------------------------*/
1829
1830/*
1831 * Construct a cache device mapping.
1832 *
1833 * cache <metadata dev> <cache dev> <origin dev> <block size>
1834 * <#feature args> [<feature arg>]*
1835 * <policy> <#policy args> [<policy arg>]*
1836 *
1837 * metadata dev : fast device holding the persistent metadata
1838 * cache dev : fast device holding cached data blocks
1839 * origin dev : slow device holding original data blocks
1840 * block size : cache unit size in sectors
1841 *
1842 * #feature args : number of feature arguments passed
1843 * feature args : writethrough. (The default is writeback.)
1844 *
1845 * policy : the replacement policy to use
1846 * #policy args : an even number of policy arguments corresponding
1847 * to key/value pairs passed to the policy
1848 * policy args : key/value pairs passed to the policy
1849 * E.g. 'sequential_threshold 1024'
1850 * See cache-policies.txt for details.
1851 *
1852 * Optional feature arguments are:
1853 * writethrough : write through caching that prohibits cache block
1854 * content from being different from origin block content.
1855 * Without this argument, the default behaviour is to write
1856 * back cache block contents later for performance reasons,
1857 * so they may differ from the corresponding origin blocks.
1858 */
1859struct cache_args {
1860 struct dm_target *ti;
1861
1862 struct dm_dev *metadata_dev;
1863
1864 struct dm_dev *cache_dev;
1865 sector_t cache_sectors;
1866
1867 struct dm_dev *origin_dev;
1868 sector_t origin_sectors;
1869
1870 uint32_t block_size;
1871
1872 const char *policy_name;
1873 int policy_argc;
1874 const char **policy_argv;
1875
1876 struct cache_features features;
1877};
1878
1879static void destroy_cache_args(struct cache_args *ca)
1880{
1881 if (ca->metadata_dev)
1882 dm_put_device(ca->ti, ca->metadata_dev);
1883
1884 if (ca->cache_dev)
1885 dm_put_device(ca->ti, ca->cache_dev);
1886
1887 if (ca->origin_dev)
1888 dm_put_device(ca->ti, ca->origin_dev);
1889
1890 kfree(ca);
1891}
1892
1893static bool at_least_one_arg(struct dm_arg_set *as, char **error)
1894{
1895 if (!as->argc) {
1896 *error = "Insufficient args";
1897 return false;
1898 }
1899
1900 return true;
1901}
1902
1903static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as,
1904 char **error)
1905{
1906 int r;
1907 sector_t metadata_dev_size;
1908 char b[BDEVNAME_SIZE];
1909
1910 if (!at_least_one_arg(as, error))
1911 return -EINVAL;
1912
1913 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
1914 &ca->metadata_dev);
1915 if (r) {
1916 *error = "Error opening metadata device";
1917 return r;
1918 }
1919
1920 metadata_dev_size = get_dev_size(ca->metadata_dev);
1921 if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING)
1922 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1923 bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
1924
1925 return 0;
1926}
1927
1928static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as,
1929 char **error)
1930{
1931 int r;
1932
1933 if (!at_least_one_arg(as, error))
1934 return -EINVAL;
1935
1936 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
1937 &ca->cache_dev);
1938 if (r) {
1939 *error = "Error opening cache device";
1940 return r;
1941 }
1942 ca->cache_sectors = get_dev_size(ca->cache_dev);
1943
1944 return 0;
1945}
1946
1947static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as,
1948 char **error)
1949{
1950 int r;
1951
1952 if (!at_least_one_arg(as, error))
1953 return -EINVAL;
1954
1955 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
1956 &ca->origin_dev);
1957 if (r) {
1958 *error = "Error opening origin device";
1959 return r;
1960 }
1961
1962 ca->origin_sectors = get_dev_size(ca->origin_dev);
1963 if (ca->ti->len > ca->origin_sectors) {
1964 *error = "Device size larger than cached device";
1965 return -EINVAL;
1966 }
1967
1968 return 0;
1969}
1970
1971static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as,
1972 char **error)
1973{
Mike Snitzer05473042013-08-16 10:54:19 -04001974 unsigned long block_size;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001975
1976 if (!at_least_one_arg(as, error))
1977 return -EINVAL;
1978
Mike Snitzer05473042013-08-16 10:54:19 -04001979 if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size ||
1980 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
1981 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
1982 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001983 *error = "Invalid data block size";
1984 return -EINVAL;
1985 }
1986
Mike Snitzer05473042013-08-16 10:54:19 -04001987 if (block_size > ca->cache_sectors) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001988 *error = "Data block size is larger than the cache device";
1989 return -EINVAL;
1990 }
1991
Mike Snitzer05473042013-08-16 10:54:19 -04001992 ca->block_size = block_size;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001993
1994 return 0;
1995}
1996
1997static void init_features(struct cache_features *cf)
1998{
1999 cf->mode = CM_WRITE;
Joe Thornber2ee57d52013-10-24 14:10:29 -04002000 cf->io_mode = CM_IO_WRITEBACK;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002001}
2002
2003static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
2004 char **error)
2005{
2006 static struct dm_arg _args[] = {
2007 {0, 1, "Invalid number of cache feature arguments"},
2008 };
2009
2010 int r;
2011 unsigned argc;
2012 const char *arg;
2013 struct cache_features *cf = &ca->features;
2014
2015 init_features(cf);
2016
2017 r = dm_read_arg_group(_args, as, &argc, error);
2018 if (r)
2019 return -EINVAL;
2020
2021 while (argc--) {
2022 arg = dm_shift_arg(as);
2023
2024 if (!strcasecmp(arg, "writeback"))
Joe Thornber2ee57d52013-10-24 14:10:29 -04002025 cf->io_mode = CM_IO_WRITEBACK;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002026
2027 else if (!strcasecmp(arg, "writethrough"))
Joe Thornber2ee57d52013-10-24 14:10:29 -04002028 cf->io_mode = CM_IO_WRITETHROUGH;
2029
2030 else if (!strcasecmp(arg, "passthrough"))
2031 cf->io_mode = CM_IO_PASSTHROUGH;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002032
2033 else {
2034 *error = "Unrecognised cache feature requested";
2035 return -EINVAL;
2036 }
2037 }
2038
2039 return 0;
2040}
2041
2042static int parse_policy(struct cache_args *ca, struct dm_arg_set *as,
2043 char **error)
2044{
2045 static struct dm_arg _args[] = {
2046 {0, 1024, "Invalid number of policy arguments"},
2047 };
2048
2049 int r;
2050
2051 if (!at_least_one_arg(as, error))
2052 return -EINVAL;
2053
2054 ca->policy_name = dm_shift_arg(as);
2055
2056 r = dm_read_arg_group(_args, as, &ca->policy_argc, error);
2057 if (r)
2058 return -EINVAL;
2059
2060 ca->policy_argv = (const char **)as->argv;
2061 dm_consume_args(as, ca->policy_argc);
2062
2063 return 0;
2064}
2065
2066static int parse_cache_args(struct cache_args *ca, int argc, char **argv,
2067 char **error)
2068{
2069 int r;
2070 struct dm_arg_set as;
2071
2072 as.argc = argc;
2073 as.argv = argv;
2074
2075 r = parse_metadata_dev(ca, &as, error);
2076 if (r)
2077 return r;
2078
2079 r = parse_cache_dev(ca, &as, error);
2080 if (r)
2081 return r;
2082
2083 r = parse_origin_dev(ca, &as, error);
2084 if (r)
2085 return r;
2086
2087 r = parse_block_size(ca, &as, error);
2088 if (r)
2089 return r;
2090
2091 r = parse_features(ca, &as, error);
2092 if (r)
2093 return r;
2094
2095 r = parse_policy(ca, &as, error);
2096 if (r)
2097 return r;
2098
2099 return 0;
2100}
2101
2102/*----------------------------------------------------------------*/
2103
2104static struct kmem_cache *migration_cache;
2105
Alasdair G Kergon2c73c472013-05-10 14:37:21 +01002106#define NOT_CORE_OPTION 1
2107
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002108static int process_config_option(struct cache *cache, const char *key, const char *value)
Alasdair G Kergon2c73c472013-05-10 14:37:21 +01002109{
2110 unsigned long tmp;
2111
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002112 if (!strcasecmp(key, "migration_threshold")) {
2113 if (kstrtoul(value, 10, &tmp))
Alasdair G Kergon2c73c472013-05-10 14:37:21 +01002114 return -EINVAL;
2115
2116 cache->migration_threshold = tmp;
2117 return 0;
2118 }
2119
2120 return NOT_CORE_OPTION;
2121}
2122
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002123static int set_config_value(struct cache *cache, const char *key, const char *value)
2124{
2125 int r = process_config_option(cache, key, value);
2126
2127 if (r == NOT_CORE_OPTION)
2128 r = policy_set_config_value(cache->policy, key, value);
2129
2130 if (r)
2131 DMWARN("bad config value for %s: %s", key, value);
2132
2133 return r;
2134}
2135
2136static int set_config_values(struct cache *cache, int argc, const char **argv)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002137{
2138 int r = 0;
2139
2140 if (argc & 1) {
2141 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
2142 return -EINVAL;
2143 }
2144
2145 while (argc) {
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002146 r = set_config_value(cache, argv[0], argv[1]);
2147 if (r)
2148 break;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002149
2150 argc -= 2;
2151 argv += 2;
2152 }
2153
2154 return r;
2155}
2156
2157static int create_cache_policy(struct cache *cache, struct cache_args *ca,
2158 char **error)
2159{
Mikulas Patocka4cb3e1d2013-10-01 18:35:39 -04002160 struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name,
2161 cache->cache_size,
2162 cache->origin_sectors,
2163 cache->sectors_per_block);
2164 if (IS_ERR(p)) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002165 *error = "Error creating cache's policy";
Mikulas Patocka4cb3e1d2013-10-01 18:35:39 -04002166 return PTR_ERR(p);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002167 }
Mikulas Patocka4cb3e1d2013-10-01 18:35:39 -04002168 cache->policy = p;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002169
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002170 return 0;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002171}
2172
2173/*
2174 * We want the discard block size to be a power of two, at least the size
2175 * of the cache block size, and have no more than 2^14 discard blocks
2176 * across the origin.
2177 */
2178#define MAX_DISCARD_BLOCKS (1 << 14)
2179
2180static bool too_many_discard_blocks(sector_t discard_block_size,
2181 sector_t origin_size)
2182{
2183 (void) sector_div(origin_size, discard_block_size);
2184
2185 return origin_size > MAX_DISCARD_BLOCKS;
2186}
2187
2188static sector_t calculate_discard_block_size(sector_t cache_block_size,
2189 sector_t origin_size)
2190{
2191 sector_t discard_block_size;
2192
2193 discard_block_size = roundup_pow_of_two(cache_block_size);
2194
2195 if (origin_size)
2196 while (too_many_discard_blocks(discard_block_size, origin_size))
2197 discard_block_size *= 2;
2198
2199 return discard_block_size;
2200}
2201
Joe Thornberf8350da2013-05-10 14:37:16 +01002202#define DEFAULT_MIGRATION_THRESHOLD 2048
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002203
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002204static int cache_create(struct cache_args *ca, struct cache **result)
2205{
2206 int r = 0;
2207 char **error = &ca->ti->error;
2208 struct cache *cache;
2209 struct dm_target *ti = ca->ti;
2210 dm_block_t origin_blocks;
2211 struct dm_cache_metadata *cmd;
2212 bool may_format = ca->features.mode == CM_WRITE;
2213
2214 cache = kzalloc(sizeof(*cache), GFP_KERNEL);
2215 if (!cache)
2216 return -ENOMEM;
2217
2218 cache->ti = ca->ti;
2219 ti->private = cache;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002220 ti->num_flush_bios = 2;
2221 ti->flush_supported = true;
2222
2223 ti->num_discard_bios = 1;
2224 ti->discards_supported = true;
2225 ti->discard_zeroes_data_unsupported = true;
2226
Joe Thornber8c5008f2013-05-10 14:37:18 +01002227 cache->features = ca->features;
Mike Snitzer19b00922013-04-05 15:36:34 +01002228 ti->per_bio_data_size = get_per_bio_data_size(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002229
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002230 cache->callbacks.congested_fn = cache_is_congested;
2231 dm_table_add_target_callbacks(ti->table, &cache->callbacks);
2232
2233 cache->metadata_dev = ca->metadata_dev;
2234 cache->origin_dev = ca->origin_dev;
2235 cache->cache_dev = ca->cache_dev;
2236
2237 ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL;
2238
2239 /* FIXME: factor out this whole section */
2240 origin_blocks = cache->origin_sectors = ca->origin_sectors;
Joe Thornber414dd672013-03-20 17:21:25 +00002241 origin_blocks = block_div(origin_blocks, ca->block_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002242 cache->origin_blocks = to_oblock(origin_blocks);
2243
2244 cache->sectors_per_block = ca->block_size;
2245 if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) {
2246 r = -EINVAL;
2247 goto bad;
2248 }
2249
2250 if (ca->block_size & (ca->block_size - 1)) {
2251 dm_block_t cache_size = ca->cache_sectors;
2252
2253 cache->sectors_per_block_shift = -1;
Joe Thornber414dd672013-03-20 17:21:25 +00002254 cache_size = block_div(cache_size, ca->block_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002255 cache->cache_size = to_cblock(cache_size);
2256 } else {
2257 cache->sectors_per_block_shift = __ffs(ca->block_size);
2258 cache->cache_size = to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift);
2259 }
2260
2261 r = create_cache_policy(cache, ca, error);
2262 if (r)
2263 goto bad;
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002264
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002265 cache->policy_nr_args = ca->policy_argc;
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002266 cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD;
2267
2268 r = set_config_values(cache, ca->policy_argc, ca->policy_argv);
2269 if (r) {
2270 *error = "Error setting cache policy's config values";
2271 goto bad;
2272 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002273
2274 cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
2275 ca->block_size, may_format,
2276 dm_cache_policy_get_hint_size(cache->policy));
2277 if (IS_ERR(cmd)) {
2278 *error = "Error creating metadata object";
2279 r = PTR_ERR(cmd);
2280 goto bad;
2281 }
2282 cache->cmd = cmd;
2283
Joe Thornber2ee57d52013-10-24 14:10:29 -04002284 if (passthrough_mode(&cache->features)) {
2285 bool all_clean;
2286
2287 r = dm_cache_metadata_all_clean(cache->cmd, &all_clean);
2288 if (r) {
2289 *error = "dm_cache_metadata_all_clean() failed";
2290 goto bad;
2291 }
2292
2293 if (!all_clean) {
2294 *error = "Cannot enter passthrough mode unless all blocks are clean";
2295 r = -EINVAL;
2296 goto bad;
2297 }
2298 }
2299
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002300 spin_lock_init(&cache->lock);
2301 bio_list_init(&cache->deferred_bios);
2302 bio_list_init(&cache->deferred_flush_bios);
Joe Thornbere2e74d62013-03-20 17:21:27 +00002303 bio_list_init(&cache->deferred_writethrough_bios);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002304 INIT_LIST_HEAD(&cache->quiesced_migrations);
2305 INIT_LIST_HEAD(&cache->completed_migrations);
2306 INIT_LIST_HEAD(&cache->need_commit_migrations);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002307 atomic_set(&cache->nr_migrations, 0);
2308 init_waitqueue_head(&cache->migration_wait);
2309
Joe Thornber66cb1912013-10-30 17:11:58 +00002310 init_waitqueue_head(&cache->quiescing_wait);
Joe Thornber238f8362013-10-30 17:29:30 +00002311 atomic_set(&cache->quiescing, 0);
Joe Thornber66cb1912013-10-30 17:11:58 +00002312 atomic_set(&cache->quiescing_ack, 0);
2313
Wei Yongjunfa4d6832013-05-10 14:37:14 +01002314 r = -ENOMEM;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002315 cache->nr_dirty = 0;
2316 cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size));
2317 if (!cache->dirty_bitset) {
2318 *error = "could not allocate dirty bitset";
2319 goto bad;
2320 }
2321 clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size));
2322
2323 cache->discard_block_size =
2324 calculate_discard_block_size(cache->sectors_per_block,
2325 cache->origin_sectors);
2326 cache->discard_nr_blocks = oblock_to_dblock(cache, cache->origin_blocks);
2327 cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks));
2328 if (!cache->discard_bitset) {
2329 *error = "could not allocate discard bitset";
2330 goto bad;
2331 }
2332 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
2333
2334 cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2335 if (IS_ERR(cache->copier)) {
2336 *error = "could not create kcopyd client";
2337 r = PTR_ERR(cache->copier);
2338 goto bad;
2339 }
2340
2341 cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2342 if (!cache->wq) {
2343 *error = "could not create workqueue for metadata object";
2344 goto bad;
2345 }
2346 INIT_WORK(&cache->worker, do_worker);
2347 INIT_DELAYED_WORK(&cache->waker, do_waker);
2348 cache->last_commit_jiffies = jiffies;
2349
2350 cache->prison = dm_bio_prison_create(PRISON_CELLS);
2351 if (!cache->prison) {
2352 *error = "could not create bio prison";
2353 goto bad;
2354 }
2355
2356 cache->all_io_ds = dm_deferred_set_create();
2357 if (!cache->all_io_ds) {
2358 *error = "could not create all_io deferred set";
2359 goto bad;
2360 }
2361
2362 cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE,
2363 migration_cache);
2364 if (!cache->migration_pool) {
2365 *error = "Error creating cache's migration mempool";
2366 goto bad;
2367 }
2368
2369 cache->next_migration = NULL;
2370
2371 cache->need_tick_bio = true;
2372 cache->sized = false;
Joe Thornber65790ff2013-11-08 16:39:50 +00002373 cache->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002374 cache->commit_requested = false;
2375 cache->loaded_mappings = false;
2376 cache->loaded_discards = false;
2377
2378 load_stats(cache);
2379
2380 atomic_set(&cache->stats.demotion, 0);
2381 atomic_set(&cache->stats.promotion, 0);
2382 atomic_set(&cache->stats.copies_avoided, 0);
2383 atomic_set(&cache->stats.cache_cell_clash, 0);
2384 atomic_set(&cache->stats.commit_count, 0);
2385 atomic_set(&cache->stats.discard_count, 0);
2386
Joe Thornber65790ff2013-11-08 16:39:50 +00002387 spin_lock_init(&cache->invalidation_lock);
2388 INIT_LIST_HEAD(&cache->invalidation_requests);
2389
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002390 *result = cache;
2391 return 0;
2392
2393bad:
2394 destroy(cache);
2395 return r;
2396}
2397
2398static int copy_ctr_args(struct cache *cache, int argc, const char **argv)
2399{
2400 unsigned i;
2401 const char **copy;
2402
2403 copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL);
2404 if (!copy)
2405 return -ENOMEM;
2406 for (i = 0; i < argc; i++) {
2407 copy[i] = kstrdup(argv[i], GFP_KERNEL);
2408 if (!copy[i]) {
2409 while (i--)
2410 kfree(copy[i]);
2411 kfree(copy);
2412 return -ENOMEM;
2413 }
2414 }
2415
2416 cache->nr_ctr_args = argc;
2417 cache->ctr_args = copy;
2418
2419 return 0;
2420}
2421
2422static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv)
2423{
2424 int r = -EINVAL;
2425 struct cache_args *ca;
2426 struct cache *cache = NULL;
2427
2428 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2429 if (!ca) {
2430 ti->error = "Error allocating memory for cache";
2431 return -ENOMEM;
2432 }
2433 ca->ti = ti;
2434
2435 r = parse_cache_args(ca, argc, argv, &ti->error);
2436 if (r)
2437 goto out;
2438
2439 r = cache_create(ca, &cache);
Heinz Mauelshagen617a0b82013-03-20 17:21:26 +00002440 if (r)
2441 goto out;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002442
2443 r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3);
2444 if (r) {
2445 destroy(cache);
2446 goto out;
2447 }
2448
2449 ti->private = cache;
2450
2451out:
2452 destroy_cache_args(ca);
2453 return r;
2454}
2455
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002456static int cache_map(struct dm_target *ti, struct bio *bio)
2457{
2458 struct cache *cache = ti->private;
2459
2460 int r;
2461 dm_oblock_t block = get_bio_block(cache, bio);
Mike Snitzer19b00922013-04-05 15:36:34 +01002462 size_t pb_data_size = get_per_bio_data_size(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002463 bool can_migrate = false;
2464 bool discarded_block;
2465 struct dm_bio_prison_cell *cell;
2466 struct policy_result lookup_result;
2467 struct per_bio_data *pb;
2468
2469 if (from_oblock(block) > from_oblock(cache->origin_blocks)) {
2470 /*
2471 * This can only occur if the io goes to a partial block at
2472 * the end of the origin device. We don't cache these.
2473 * Just remap to the origin and carry on.
2474 */
2475 remap_to_origin_clear_discard(cache, bio, block);
2476 return DM_MAPIO_REMAPPED;
2477 }
2478
Mike Snitzer19b00922013-04-05 15:36:34 +01002479 pb = init_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002480
2481 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_DISCARD)) {
2482 defer_bio(cache, bio);
2483 return DM_MAPIO_SUBMITTED;
2484 }
2485
2486 /*
2487 * Check to see if that block is currently migrating.
2488 */
2489 cell = alloc_prison_cell(cache);
2490 if (!cell) {
2491 defer_bio(cache, bio);
2492 return DM_MAPIO_SUBMITTED;
2493 }
2494
2495 r = bio_detain(cache, block, bio, cell,
2496 (cell_free_fn) free_prison_cell,
2497 cache, &cell);
2498 if (r) {
2499 if (r < 0)
2500 defer_bio(cache, bio);
2501
2502 return DM_MAPIO_SUBMITTED;
2503 }
2504
2505 discarded_block = is_discarded_oblock(cache, block);
2506
2507 r = policy_map(cache->policy, block, false, can_migrate, discarded_block,
2508 bio, &lookup_result);
2509 if (r == -EWOULDBLOCK) {
2510 cell_defer(cache, cell, true);
2511 return DM_MAPIO_SUBMITTED;
2512
2513 } else if (r) {
2514 DMERR_LIMIT("Unexpected return from cache replacement policy: %d", r);
2515 bio_io_error(bio);
2516 return DM_MAPIO_SUBMITTED;
2517 }
2518
Joe Thornber2ee57d52013-10-24 14:10:29 -04002519 r = DM_MAPIO_REMAPPED;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002520 switch (lookup_result.op) {
2521 case POLICY_HIT:
Joe Thornber2ee57d52013-10-24 14:10:29 -04002522 if (passthrough_mode(&cache->features)) {
2523 if (bio_data_dir(bio) == WRITE) {
2524 /*
2525 * We need to invalidate this block, so
2526 * defer for the worker thread.
2527 */
2528 cell_defer(cache, cell, true);
2529 r = DM_MAPIO_SUBMITTED;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002530
Joe Thornber2ee57d52013-10-24 14:10:29 -04002531 } else {
2532 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
2533 inc_miss_counter(cache, bio);
2534 remap_to_origin_clear_discard(cache, bio, block);
Joe Thornbere2e74d62013-03-20 17:21:27 +00002535
Joe Thornber2ee57d52013-10-24 14:10:29 -04002536 cell_defer(cache, cell, false);
2537 }
2538
2539 } else {
2540 inc_hit_counter(cache, bio);
2541
2542 if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) &&
2543 !is_dirty(cache, lookup_result.cblock))
2544 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
2545 else
2546 remap_to_cache_dirty(cache, bio, block, lookup_result.cblock);
2547
2548 cell_defer(cache, cell, false);
2549 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002550 break;
2551
2552 case POLICY_MISS:
2553 inc_miss_counter(cache, bio);
2554 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
2555
2556 if (pb->req_nr != 0) {
2557 /*
2558 * This is a duplicate writethrough io that is no
2559 * longer needed because the block has been demoted.
2560 */
2561 bio_endio(bio, 0);
2562 cell_defer(cache, cell, false);
2563 return DM_MAPIO_SUBMITTED;
2564 } else {
2565 remap_to_origin_clear_discard(cache, bio, block);
2566 cell_defer(cache, cell, false);
2567 }
2568 break;
2569
2570 default:
2571 DMERR_LIMIT("%s: erroring bio: unknown policy op: %u", __func__,
2572 (unsigned) lookup_result.op);
2573 bio_io_error(bio);
Joe Thornber2ee57d52013-10-24 14:10:29 -04002574 r = DM_MAPIO_SUBMITTED;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002575 }
2576
Joe Thornber2ee57d52013-10-24 14:10:29 -04002577 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002578}
2579
2580static int cache_end_io(struct dm_target *ti, struct bio *bio, int error)
2581{
2582 struct cache *cache = ti->private;
2583 unsigned long flags;
Mike Snitzer19b00922013-04-05 15:36:34 +01002584 size_t pb_data_size = get_per_bio_data_size(cache);
2585 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002586
2587 if (pb->tick) {
2588 policy_tick(cache->policy);
2589
2590 spin_lock_irqsave(&cache->lock, flags);
2591 cache->need_tick_bio = true;
2592 spin_unlock_irqrestore(&cache->lock, flags);
2593 }
2594
2595 check_for_quiesced_migrations(cache, pb);
2596
2597 return 0;
2598}
2599
2600static int write_dirty_bitset(struct cache *cache)
2601{
2602 unsigned i, r;
2603
2604 for (i = 0; i < from_cblock(cache->cache_size); i++) {
2605 r = dm_cache_set_dirty(cache->cmd, to_cblock(i),
2606 is_dirty(cache, to_cblock(i)));
2607 if (r)
2608 return r;
2609 }
2610
2611 return 0;
2612}
2613
2614static int write_discard_bitset(struct cache *cache)
2615{
2616 unsigned i, r;
2617
2618 r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size,
2619 cache->discard_nr_blocks);
2620 if (r) {
2621 DMERR("could not resize on-disk discard bitset");
2622 return r;
2623 }
2624
2625 for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) {
2626 r = dm_cache_set_discard(cache->cmd, to_dblock(i),
2627 is_discarded(cache, to_dblock(i)));
2628 if (r)
2629 return r;
2630 }
2631
2632 return 0;
2633}
2634
2635static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock,
2636 uint32_t hint)
2637{
2638 struct cache *cache = context;
2639 return dm_cache_save_hint(cache->cmd, cblock, hint);
2640}
2641
2642static int write_hints(struct cache *cache)
2643{
2644 int r;
2645
2646 r = dm_cache_begin_hints(cache->cmd, cache->policy);
2647 if (r) {
2648 DMERR("dm_cache_begin_hints failed");
2649 return r;
2650 }
2651
2652 r = policy_walk_mappings(cache->policy, save_hint, cache);
2653 if (r)
2654 DMERR("policy_walk_mappings failed");
2655
2656 return r;
2657}
2658
2659/*
2660 * returns true on success
2661 */
2662static bool sync_metadata(struct cache *cache)
2663{
2664 int r1, r2, r3, r4;
2665
2666 r1 = write_dirty_bitset(cache);
2667 if (r1)
2668 DMERR("could not write dirty bitset");
2669
2670 r2 = write_discard_bitset(cache);
2671 if (r2)
2672 DMERR("could not write discard bitset");
2673
2674 save_stats(cache);
2675
2676 r3 = write_hints(cache);
2677 if (r3)
2678 DMERR("could not write hints");
2679
2680 /*
2681 * If writing the above metadata failed, we still commit, but don't
2682 * set the clean shutdown flag. This will effectively force every
2683 * dirty bit to be set on reload.
2684 */
2685 r4 = dm_cache_commit(cache->cmd, !r1 && !r2 && !r3);
2686 if (r4)
2687 DMERR("could not write cache metadata. Data loss may occur.");
2688
2689 return !r1 && !r2 && !r3 && !r4;
2690}
2691
2692static void cache_postsuspend(struct dm_target *ti)
2693{
2694 struct cache *cache = ti->private;
2695
2696 start_quiescing(cache);
2697 wait_for_migrations(cache);
2698 stop_worker(cache);
2699 requeue_deferred_io(cache);
2700 stop_quiescing(cache);
2701
2702 (void) sync_metadata(cache);
2703}
2704
2705static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock,
2706 bool dirty, uint32_t hint, bool hint_valid)
2707{
2708 int r;
2709 struct cache *cache = context;
2710
2711 r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid);
2712 if (r)
2713 return r;
2714
2715 if (dirty)
2716 set_dirty(cache, oblock, cblock);
2717 else
2718 clear_dirty(cache, oblock, cblock);
2719
2720 return 0;
2721}
2722
2723static int load_discard(void *context, sector_t discard_block_size,
2724 dm_dblock_t dblock, bool discard)
2725{
2726 struct cache *cache = context;
2727
2728 /* FIXME: handle mis-matched block size */
2729
2730 if (discard)
2731 set_discard(cache, dblock);
2732 else
2733 clear_discard(cache, dblock);
2734
2735 return 0;
2736}
2737
Joe Thornberf494a9c2013-10-31 13:55:49 -04002738static dm_cblock_t get_cache_dev_size(struct cache *cache)
2739{
2740 sector_t size = get_dev_size(cache->cache_dev);
2741 (void) sector_div(size, cache->sectors_per_block);
2742 return to_cblock(size);
2743}
2744
2745static bool can_resize(struct cache *cache, dm_cblock_t new_size)
2746{
2747 if (from_cblock(new_size) > from_cblock(cache->cache_size))
2748 return true;
2749
2750 /*
2751 * We can't drop a dirty block when shrinking the cache.
2752 */
2753 while (from_cblock(new_size) < from_cblock(cache->cache_size)) {
2754 new_size = to_cblock(from_cblock(new_size) + 1);
2755 if (is_dirty(cache, new_size)) {
2756 DMERR("unable to shrink cache; cache block %llu is dirty",
2757 (unsigned long long) from_cblock(new_size));
2758 return false;
2759 }
2760 }
2761
2762 return true;
2763}
2764
2765static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size)
2766{
2767 int r;
2768
Vincent Pelletier08844802013-11-30 12:58:42 +01002769 r = dm_cache_resize(cache->cmd, new_size);
Joe Thornberf494a9c2013-10-31 13:55:49 -04002770 if (r) {
2771 DMERR("could not resize cache metadata");
2772 return r;
2773 }
2774
2775 cache->cache_size = new_size;
2776
2777 return 0;
2778}
2779
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002780static int cache_preresume(struct dm_target *ti)
2781{
2782 int r = 0;
2783 struct cache *cache = ti->private;
Joe Thornberf494a9c2013-10-31 13:55:49 -04002784 dm_cblock_t csize = get_cache_dev_size(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002785
2786 /*
2787 * Check to see if the cache has resized.
2788 */
Joe Thornberf494a9c2013-10-31 13:55:49 -04002789 if (!cache->sized) {
2790 r = resize_cache_dev(cache, csize);
2791 if (r)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002792 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002793
2794 cache->sized = true;
Joe Thornberf494a9c2013-10-31 13:55:49 -04002795
2796 } else if (csize != cache->cache_size) {
2797 if (!can_resize(cache, csize))
2798 return -EINVAL;
2799
2800 r = resize_cache_dev(cache, csize);
2801 if (r)
2802 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002803 }
2804
2805 if (!cache->loaded_mappings) {
Mike Snitzerea2dd8c2013-03-20 17:21:28 +00002806 r = dm_cache_load_mappings(cache->cmd, cache->policy,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002807 load_mapping, cache);
2808 if (r) {
2809 DMERR("could not load cache mappings");
2810 return r;
2811 }
2812
2813 cache->loaded_mappings = true;
2814 }
2815
2816 if (!cache->loaded_discards) {
2817 r = dm_cache_load_discards(cache->cmd, load_discard, cache);
2818 if (r) {
2819 DMERR("could not load origin discards");
2820 return r;
2821 }
2822
2823 cache->loaded_discards = true;
2824 }
2825
2826 return r;
2827}
2828
2829static void cache_resume(struct dm_target *ti)
2830{
2831 struct cache *cache = ti->private;
2832
2833 cache->need_tick_bio = true;
2834 do_waker(&cache->waker.work);
2835}
2836
2837/*
2838 * Status format:
2839 *
Mike Snitzer6a388612014-01-09 16:04:12 -05002840 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
2841 * <cache block size> <#used cache blocks>/<#total cache blocks>
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002842 * <#read hits> <#read misses> <#write hits> <#write misses>
Mike Snitzer6a388612014-01-09 16:04:12 -05002843 * <#demotions> <#promotions> <#dirty>
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002844 * <#features> <features>*
2845 * <#core args> <core args>
Mike Snitzer2e68c4e2014-01-15 21:06:55 -05002846 * <policy name> <#policy args> <policy args>*
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002847 */
2848static void cache_status(struct dm_target *ti, status_type_t type,
2849 unsigned status_flags, char *result, unsigned maxlen)
2850{
2851 int r = 0;
2852 unsigned i;
2853 ssize_t sz = 0;
2854 dm_block_t nr_free_blocks_metadata = 0;
2855 dm_block_t nr_blocks_metadata = 0;
2856 char buf[BDEVNAME_SIZE];
2857 struct cache *cache = ti->private;
2858 dm_cblock_t residency;
2859
2860 switch (type) {
2861 case STATUSTYPE_INFO:
2862 /* Commit to ensure statistics aren't out-of-date */
2863 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) {
2864 r = dm_cache_commit(cache->cmd, false);
2865 if (r)
2866 DMERR("could not commit metadata for accurate status");
2867 }
2868
2869 r = dm_cache_get_free_metadata_block_count(cache->cmd,
2870 &nr_free_blocks_metadata);
2871 if (r) {
2872 DMERR("could not get metadata free block count");
2873 goto err;
2874 }
2875
2876 r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata);
2877 if (r) {
2878 DMERR("could not get metadata device size");
2879 goto err;
2880 }
2881
2882 residency = policy_residency(cache->policy);
2883
Mike Snitzer6a388612014-01-09 16:04:12 -05002884 DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %llu ",
2885 (unsigned)(DM_CACHE_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002886 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2887 (unsigned long long)nr_blocks_metadata,
Mike Snitzer6a388612014-01-09 16:04:12 -05002888 cache->sectors_per_block,
2889 (unsigned long long) from_cblock(residency),
2890 (unsigned long long) from_cblock(cache->cache_size),
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002891 (unsigned) atomic_read(&cache->stats.read_hit),
2892 (unsigned) atomic_read(&cache->stats.read_miss),
2893 (unsigned) atomic_read(&cache->stats.write_hit),
2894 (unsigned) atomic_read(&cache->stats.write_miss),
2895 (unsigned) atomic_read(&cache->stats.demotion),
2896 (unsigned) atomic_read(&cache->stats.promotion),
Mike Snitzer6a388612014-01-09 16:04:12 -05002897 (unsigned long long) from_cblock(cache->nr_dirty));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002898
Joe Thornber2ee57d52013-10-24 14:10:29 -04002899 if (writethrough_mode(&cache->features))
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002900 DMEMIT("1 writethrough ");
Joe Thornber2ee57d52013-10-24 14:10:29 -04002901
2902 else if (passthrough_mode(&cache->features))
2903 DMEMIT("1 passthrough ");
2904
2905 else if (writeback_mode(&cache->features))
2906 DMEMIT("1 writeback ");
2907
2908 else {
2909 DMERR("internal error: unknown io mode: %d", (int) cache->features.io_mode);
2910 goto err;
2911 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002912
2913 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
Mike Snitzer2e68c4e2014-01-15 21:06:55 -05002914
2915 DMEMIT("%s ", dm_cache_policy_get_name(cache->policy));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002916 if (sz < maxlen) {
2917 r = policy_emit_config_values(cache->policy, result + sz, maxlen - sz);
2918 if (r)
2919 DMERR("policy_emit_config_values returned %d", r);
2920 }
2921
2922 break;
2923
2924 case STATUSTYPE_TABLE:
2925 format_dev_t(buf, cache->metadata_dev->bdev->bd_dev);
2926 DMEMIT("%s ", buf);
2927 format_dev_t(buf, cache->cache_dev->bdev->bd_dev);
2928 DMEMIT("%s ", buf);
2929 format_dev_t(buf, cache->origin_dev->bdev->bd_dev);
2930 DMEMIT("%s", buf);
2931
2932 for (i = 0; i < cache->nr_ctr_args - 1; i++)
2933 DMEMIT(" %s", cache->ctr_args[i]);
2934 if (cache->nr_ctr_args)
2935 DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]);
2936 }
2937
2938 return;
2939
2940err:
2941 DMEMIT("Error");
2942}
2943
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002944/*
Joe Thornber65790ff2013-11-08 16:39:50 +00002945 * A cache block range can take two forms:
2946 *
2947 * i) A single cblock, eg. '3456'
2948 * ii) A begin and end cblock with dots between, eg. 123-234
2949 */
2950static int parse_cblock_range(struct cache *cache, const char *str,
2951 struct cblock_range *result)
2952{
2953 char dummy;
2954 uint64_t b, e;
2955 int r;
2956
2957 /*
2958 * Try and parse form (ii) first.
2959 */
2960 r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy);
2961 if (r < 0)
2962 return r;
2963
2964 if (r == 2) {
2965 result->begin = to_cblock(b);
2966 result->end = to_cblock(e);
2967 return 0;
2968 }
2969
2970 /*
2971 * That didn't work, try form (i).
2972 */
2973 r = sscanf(str, "%llu%c", &b, &dummy);
2974 if (r < 0)
2975 return r;
2976
2977 if (r == 1) {
2978 result->begin = to_cblock(b);
2979 result->end = to_cblock(from_cblock(result->begin) + 1u);
2980 return 0;
2981 }
2982
2983 DMERR("invalid cblock range '%s'", str);
2984 return -EINVAL;
2985}
2986
2987static int validate_cblock_range(struct cache *cache, struct cblock_range *range)
2988{
2989 uint64_t b = from_cblock(range->begin);
2990 uint64_t e = from_cblock(range->end);
2991 uint64_t n = from_cblock(cache->cache_size);
2992
2993 if (b >= n) {
2994 DMERR("begin cblock out of range: %llu >= %llu", b, n);
2995 return -EINVAL;
2996 }
2997
2998 if (e > n) {
2999 DMERR("end cblock out of range: %llu > %llu", e, n);
3000 return -EINVAL;
3001 }
3002
3003 if (b >= e) {
3004 DMERR("invalid cblock range: %llu >= %llu", b, e);
3005 return -EINVAL;
3006 }
3007
3008 return 0;
3009}
3010
3011static int request_invalidation(struct cache *cache, struct cblock_range *range)
3012{
3013 struct invalidation_request req;
3014
3015 INIT_LIST_HEAD(&req.list);
3016 req.cblocks = range;
3017 atomic_set(&req.complete, 0);
3018 req.err = 0;
3019 init_waitqueue_head(&req.result_wait);
3020
3021 spin_lock(&cache->invalidation_lock);
3022 list_add(&req.list, &cache->invalidation_requests);
3023 spin_unlock(&cache->invalidation_lock);
3024 wake_worker(cache);
3025
3026 wait_event(req.result_wait, atomic_read(&req.complete));
3027 return req.err;
3028}
3029
3030static int process_invalidate_cblocks_message(struct cache *cache, unsigned count,
3031 const char **cblock_ranges)
3032{
3033 int r = 0;
3034 unsigned i;
3035 struct cblock_range range;
3036
3037 if (!passthrough_mode(&cache->features)) {
3038 DMERR("cache has to be in passthrough mode for invalidation");
3039 return -EPERM;
3040 }
3041
3042 for (i = 0; i < count; i++) {
3043 r = parse_cblock_range(cache, cblock_ranges[i], &range);
3044 if (r)
3045 break;
3046
3047 r = validate_cblock_range(cache, &range);
3048 if (r)
3049 break;
3050
3051 /*
3052 * Pass begin and end origin blocks to the worker and wake it.
3053 */
3054 r = request_invalidation(cache, &range);
3055 if (r)
3056 break;
3057 }
3058
3059 return r;
3060}
3061
3062/*
3063 * Supports
3064 * "<key> <value>"
3065 * and
3066 * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]*
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003067 *
3068 * The key migration_threshold is supported by the cache target core.
3069 */
3070static int cache_message(struct dm_target *ti, unsigned argc, char **argv)
3071{
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003072 struct cache *cache = ti->private;
3073
Joe Thornber65790ff2013-11-08 16:39:50 +00003074 if (!argc)
3075 return -EINVAL;
3076
Mike Snitzer7b6b2bc2013-11-12 12:17:43 -05003077 if (!strcasecmp(argv[0], "invalidate_cblocks"))
Joe Thornber65790ff2013-11-08 16:39:50 +00003078 return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1);
3079
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003080 if (argc != 2)
3081 return -EINVAL;
3082
Joe Thornber2f14f4b2013-05-10 14:37:21 +01003083 return set_config_value(cache, argv[0], argv[1]);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003084}
3085
3086static int cache_iterate_devices(struct dm_target *ti,
3087 iterate_devices_callout_fn fn, void *data)
3088{
3089 int r = 0;
3090 struct cache *cache = ti->private;
3091
3092 r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data);
3093 if (!r)
3094 r = fn(ti, cache->origin_dev, 0, ti->len, data);
3095
3096 return r;
3097}
3098
3099/*
3100 * We assume I/O is going to the origin (which is the volume
3101 * more likely to have restrictions e.g. by being striped).
3102 * (Looking up the exact location of the data would be expensive
3103 * and could always be out of date by the time the bio is submitted.)
3104 */
3105static int cache_bvec_merge(struct dm_target *ti,
3106 struct bvec_merge_data *bvm,
3107 struct bio_vec *biovec, int max_size)
3108{
3109 struct cache *cache = ti->private;
3110 struct request_queue *q = bdev_get_queue(cache->origin_dev->bdev);
3111
3112 if (!q->merge_bvec_fn)
3113 return max_size;
3114
3115 bvm->bi_bdev = cache->origin_dev->bdev;
3116 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
3117}
3118
3119static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
3120{
3121 /*
3122 * FIXME: these limits may be incompatible with the cache device
3123 */
3124 limits->max_discard_sectors = cache->discard_block_size * 1024;
3125 limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
3126}
3127
3128static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
3129{
3130 struct cache *cache = ti->private;
Mike Snitzerf6109372013-08-20 15:02:41 -04003131 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003132
Mike Snitzerf6109372013-08-20 15:02:41 -04003133 /*
3134 * If the system-determined stacked limits are compatible with the
3135 * cache's blocksize (io_opt is a factor) do not override them.
3136 */
3137 if (io_opt_sectors < cache->sectors_per_block ||
3138 do_div(io_opt_sectors, cache->sectors_per_block)) {
3139 blk_limits_io_min(limits, 0);
3140 blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
3141 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003142 set_discard_limits(cache, limits);
3143}
3144
3145/*----------------------------------------------------------------*/
3146
3147static struct target_type cache_target = {
3148 .name = "cache",
Mike Snitzer6a388612014-01-09 16:04:12 -05003149 .version = {1, 3, 0},
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003150 .module = THIS_MODULE,
3151 .ctr = cache_ctr,
3152 .dtr = cache_dtr,
3153 .map = cache_map,
3154 .end_io = cache_end_io,
3155 .postsuspend = cache_postsuspend,
3156 .preresume = cache_preresume,
3157 .resume = cache_resume,
3158 .status = cache_status,
3159 .message = cache_message,
3160 .iterate_devices = cache_iterate_devices,
3161 .merge = cache_bvec_merge,
3162 .io_hints = cache_io_hints,
3163};
3164
3165static int __init dm_cache_init(void)
3166{
3167 int r;
3168
3169 r = dm_register_target(&cache_target);
3170 if (r) {
3171 DMERR("cache target registration failed: %d", r);
3172 return r;
3173 }
3174
3175 migration_cache = KMEM_CACHE(dm_cache_migration, 0);
3176 if (!migration_cache) {
3177 dm_unregister_target(&cache_target);
3178 return -ENOMEM;
3179 }
3180
3181 return 0;
3182}
3183
3184static void __exit dm_cache_exit(void)
3185{
3186 dm_unregister_target(&cache_target);
3187 kmem_cache_destroy(migration_cache);
3188}
3189
3190module_init(dm_cache_init);
3191module_exit(dm_cache_exit);
3192
3193MODULE_DESCRIPTION(DM_NAME " cache target");
3194MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
3195MODULE_LICENSE("GPL");