blob: 59b2c50562e4bc8e2a1ecc3ddf7f1ffaa0bce650 [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>
Manuel Schölling0f30af92014-05-22 22:42:37 +020014#include <linux/jiffies.h>
Joe Thornberc6b4fcb2013-03-01 22:45:51 +000015#include <linux/init.h>
16#include <linux/mempool.h>
17#include <linux/module.h>
18#include <linux/slab.h>
19#include <linux/vmalloc.h>
20
21#define DM_MSG_PREFIX "cache"
22
23DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle,
24 "A percentage of time allocated for copying to and/or from cache");
25
26/*----------------------------------------------------------------*/
27
Joe Thornber77289d32015-05-15 13:45:30 +010028#define IOT_RESOLUTION 4
29
30struct io_tracker {
31 spinlock_t lock;
32
33 /*
34 * Sectors of in-flight IO.
35 */
36 sector_t in_flight;
37
38 /*
39 * The time, in jiffies, when this device became idle (if it is
40 * indeed idle).
41 */
42 unsigned long idle_time;
43 unsigned long last_update_time;
44};
45
46static void iot_init(struct io_tracker *iot)
47{
48 spin_lock_init(&iot->lock);
49 iot->in_flight = 0ul;
50 iot->idle_time = 0ul;
51 iot->last_update_time = jiffies;
52}
53
54static bool __iot_idle_for(struct io_tracker *iot, unsigned long jifs)
55{
56 if (iot->in_flight)
57 return false;
58
59 return time_after(jiffies, iot->idle_time + jifs);
60}
61
62static bool iot_idle_for(struct io_tracker *iot, unsigned long jifs)
63{
64 bool r;
65 unsigned long flags;
66
67 spin_lock_irqsave(&iot->lock, flags);
68 r = __iot_idle_for(iot, jifs);
69 spin_unlock_irqrestore(&iot->lock, flags);
70
71 return r;
72}
73
74static void iot_io_begin(struct io_tracker *iot, sector_t len)
75{
76 unsigned long flags;
77
78 spin_lock_irqsave(&iot->lock, flags);
79 iot->in_flight += len;
80 spin_unlock_irqrestore(&iot->lock, flags);
81}
82
83static void __iot_io_end(struct io_tracker *iot, sector_t len)
84{
85 iot->in_flight -= len;
86 if (!iot->in_flight)
87 iot->idle_time = jiffies;
88}
89
90static void iot_io_end(struct io_tracker *iot, sector_t len)
91{
92 unsigned long flags;
93
94 spin_lock_irqsave(&iot->lock, flags);
95 __iot_io_end(iot, len);
96 spin_unlock_irqrestore(&iot->lock, flags);
97}
98
99/*----------------------------------------------------------------*/
100
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000101/*
102 * Glossary:
103 *
104 * oblock: index of an origin block
105 * cblock: index of a cache block
106 * promotion: movement of a block from origin to cache
107 * demotion: movement of a block from cache to origin
108 * migration: movement of a block between the origin and cache device,
109 * either direction
110 */
111
112/*----------------------------------------------------------------*/
113
Joe Thornberc9d28d52013-10-31 13:55:48 -0400114/*
115 * There are a couple of places where we let a bio run, but want to do some
116 * work before calling its endio function. We do this by temporarily
117 * changing the endio fn.
118 */
119struct dm_hook_info {
120 bio_end_io_t *bi_end_io;
Joe Thornberc9d28d52013-10-31 13:55:48 -0400121};
122
123static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio,
124 bio_end_io_t *bi_end_io, void *bi_private)
125{
126 h->bi_end_io = bio->bi_end_io;
Joe Thornberc9d28d52013-10-31 13:55:48 -0400127
128 bio->bi_end_io = bi_end_io;
129 bio->bi_private = bi_private;
130}
131
132static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio)
133{
134 bio->bi_end_io = h->bi_end_io;
Joe Thornberc9d28d52013-10-31 13:55:48 -0400135}
136
137/*----------------------------------------------------------------*/
138
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000139#define MIGRATION_POOL_SIZE 128
140#define COMMIT_PERIOD HZ
141#define MIGRATION_COUNT_WINDOW 10
142
143/*
Mike Snitzer05473042013-08-16 10:54:19 -0400144 * The block size of the device holding cache data must be
145 * between 32KB and 1GB.
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000146 */
147#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT)
Mike Snitzer05473042013-08-16 10:54:19 -0400148#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000149
Joe Thornber2ee57d52013-10-24 14:10:29 -0400150enum cache_metadata_mode {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000151 CM_WRITE, /* metadata may be changed */
152 CM_READ_ONLY, /* metadata may not be changed */
Joe Thornber028ae9f2015-04-22 16:42:35 -0400153 CM_FAIL
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000154};
155
Joe Thornber2ee57d52013-10-24 14:10:29 -0400156enum cache_io_mode {
157 /*
158 * Data is written to cached blocks only. These blocks are marked
159 * dirty. If you lose the cache device you will lose data.
160 * Potential performance increase for both reads and writes.
161 */
162 CM_IO_WRITEBACK,
163
164 /*
165 * Data is written to both cache and origin. Blocks are never
166 * dirty. Potential performance benfit for reads only.
167 */
168 CM_IO_WRITETHROUGH,
169
170 /*
171 * A degraded mode useful for various cache coherency situations
172 * (eg, rolling back snapshots). Reads and writes always go to the
173 * origin. If a write goes to a cached oblock, then the cache
174 * block is invalidated.
175 */
176 CM_IO_PASSTHROUGH
177};
178
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000179struct cache_features {
Joe Thornber2ee57d52013-10-24 14:10:29 -0400180 enum cache_metadata_mode mode;
181 enum cache_io_mode io_mode;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000182};
183
184struct cache_stats {
185 atomic_t read_hit;
186 atomic_t read_miss;
187 atomic_t write_hit;
188 atomic_t write_miss;
189 atomic_t demotion;
190 atomic_t promotion;
191 atomic_t copies_avoided;
192 atomic_t cache_cell_clash;
193 atomic_t commit_count;
194 atomic_t discard_count;
195};
196
Joe Thornber65790ff2013-11-08 16:39:50 +0000197/*
198 * Defines a range of cblocks, begin to (end - 1) are in the range. end is
199 * the one-past-the-end value.
200 */
201struct cblock_range {
202 dm_cblock_t begin;
203 dm_cblock_t end;
204};
205
206struct invalidation_request {
207 struct list_head list;
208 struct cblock_range *cblocks;
209
210 atomic_t complete;
211 int err;
212
213 wait_queue_head_t result_wait;
214};
215
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000216struct cache {
217 struct dm_target *ti;
218 struct dm_target_callbacks callbacks;
219
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400220 struct dm_cache_metadata *cmd;
221
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000222 /*
223 * Metadata is written to this device.
224 */
225 struct dm_dev *metadata_dev;
226
227 /*
228 * The slower of the two data devices. Typically a spindle.
229 */
230 struct dm_dev *origin_dev;
231
232 /*
233 * The faster of the two data devices. Typically an SSD.
234 */
235 struct dm_dev *cache_dev;
236
237 /*
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000238 * Size of the origin device in _complete_ blocks and native sectors.
239 */
240 dm_oblock_t origin_blocks;
241 sector_t origin_sectors;
242
243 /*
244 * Size of the cache device in blocks.
245 */
246 dm_cblock_t cache_size;
247
248 /*
249 * Fields for converting from sectors to blocks.
250 */
251 uint32_t sectors_per_block;
252 int sectors_per_block_shift;
253
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000254 spinlock_t lock;
Joe Thornber651f5fa2015-05-15 15:26:08 +0100255 struct list_head deferred_cells;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000256 struct bio_list deferred_bios;
257 struct bio_list deferred_flush_bios;
Joe Thornbere2e74d62013-03-20 17:21:27 +0000258 struct bio_list deferred_writethrough_bios;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000259 struct list_head quiesced_migrations;
260 struct list_head completed_migrations;
261 struct list_head need_commit_migrations;
262 sector_t migration_threshold;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000263 wait_queue_head_t migration_wait;
Joe Thornbera59db672015-01-23 10:16:16 +0000264 atomic_t nr_allocated_migrations;
265
266 /*
267 * The number of in flight migrations that are performing
268 * background io. eg, promotion, writeback.
269 */
270 atomic_t nr_io_migrations;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000271
Joe Thornber66cb1912013-10-30 17:11:58 +0000272 wait_queue_head_t quiescing_wait;
Joe Thornber238f8362013-10-30 17:29:30 +0000273 atomic_t quiescing;
Joe Thornber66cb1912013-10-30 17:11:58 +0000274 atomic_t quiescing_ack;
275
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000276 /*
277 * cache_size entries, dirty if set
278 */
Anssi Hannula44fa8162014-08-01 11:55:47 -0400279 atomic_t nr_dirty;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000280 unsigned long *dirty_bitset;
281
282 /*
283 * origin_blocks entries, discarded if set.
284 */
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000285 dm_dblock_t discard_nr_blocks;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000286 unsigned long *discard_bitset;
Joe Thornber08b18452014-11-06 14:38:01 +0000287 uint32_t discard_block_size; /* a power of 2 times sectors per block */
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400288
289 /*
290 * Rather than reconstructing the table line for the status we just
291 * save it and regurgitate.
292 */
293 unsigned nr_ctr_args;
294 const char **ctr_args;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000295
296 struct dm_kcopyd_client *copier;
297 struct workqueue_struct *wq;
298 struct work_struct worker;
299
300 struct delayed_work waker;
301 unsigned long last_commit_jiffies;
302
303 struct dm_bio_prison *prison;
304 struct dm_deferred_set *all_io_ds;
305
306 mempool_t *migration_pool;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000307
308 struct dm_cache_policy *policy;
309 unsigned policy_nr_args;
310
311 bool need_tick_bio:1;
312 bool sized:1;
Joe Thornber65790ff2013-11-08 16:39:50 +0000313 bool invalidate:1;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000314 bool commit_requested:1;
315 bool loaded_mappings:1;
316 bool loaded_discards:1;
317
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000318 /*
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400319 * Cache features such as write-through.
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000320 */
Mike Snitzerc9ec5d72013-08-16 10:54:21 -0400321 struct cache_features features;
322
323 struct cache_stats stats;
Joe Thornber65790ff2013-11-08 16:39:50 +0000324
325 /*
326 * Invalidation fields.
327 */
328 spinlock_t invalidation_lock;
329 struct list_head invalidation_requests;
Joe Thornber066dbaa2015-05-15 15:18:01 +0100330
331 struct io_tracker origin_tracker;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000332};
333
334struct per_bio_data {
335 bool tick:1;
336 unsigned req_nr:2;
337 struct dm_deferred_entry *all_io_entry;
Mike Snitzerc6eda5e2014-01-31 14:11:54 -0500338 struct dm_hook_info hook_info;
Joe Thornber066dbaa2015-05-15 15:18:01 +0100339 sector_t len;
Joe Thornbere2e74d62013-03-20 17:21:27 +0000340
Mike Snitzer19b00922013-04-05 15:36:34 +0100341 /*
342 * writethrough fields. These MUST remain at the end of this
343 * structure and the 'cache' member must be the first as it
Joe Thornberaeed1422013-05-10 14:37:18 +0100344 * is used to determine the offset of the writethrough fields.
Mike Snitzer19b00922013-04-05 15:36:34 +0100345 */
Joe Thornbere2e74d62013-03-20 17:21:27 +0000346 struct cache *cache;
347 dm_cblock_t cblock;
Darrick J. Wongb844fe62013-04-05 15:36:32 +0100348 struct dm_bio_details bio_details;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000349};
350
351struct dm_cache_migration {
352 struct list_head list;
353 struct cache *cache;
354
355 unsigned long start_jiffies;
356 dm_oblock_t old_oblock;
357 dm_oblock_t new_oblock;
358 dm_cblock_t cblock;
359
360 bool err:1;
Joe Thornber7ae34e72014-11-06 10:18:04 +0000361 bool discard:1;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000362 bool writeback:1;
363 bool demote:1;
364 bool promote:1;
Joe Thornberc9d28d52013-10-31 13:55:48 -0400365 bool requeue_holder:1;
Joe Thornber65790ff2013-11-08 16:39:50 +0000366 bool invalidate:1;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000367
368 struct dm_bio_prison_cell *old_ocell;
369 struct dm_bio_prison_cell *new_ocell;
370};
371
372/*
373 * Processing a bio in the worker thread may require these memory
374 * allocations. We prealloc to avoid deadlocks (the same worker thread
375 * frees them back to the mempool).
376 */
377struct prealloc {
378 struct dm_cache_migration *mg;
379 struct dm_bio_prison_cell *cell1;
380 struct dm_bio_prison_cell *cell2;
381};
382
Joe Thornber028ae9f2015-04-22 16:42:35 -0400383static enum cache_metadata_mode get_cache_mode(struct cache *cache);
384
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000385static void wake_worker(struct cache *cache)
386{
387 queue_work(cache->wq, &cache->worker);
388}
389
390/*----------------------------------------------------------------*/
391
392static struct dm_bio_prison_cell *alloc_prison_cell(struct cache *cache)
393{
394 /* FIXME: change to use a local slab. */
395 return dm_bio_prison_alloc_cell(cache->prison, GFP_NOWAIT);
396}
397
398static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell *cell)
399{
400 dm_bio_prison_free_cell(cache->prison, cell);
401}
402
Joe Thornbera59db672015-01-23 10:16:16 +0000403static struct dm_cache_migration *alloc_migration(struct cache *cache)
404{
405 struct dm_cache_migration *mg;
406
407 mg = mempool_alloc(cache->migration_pool, GFP_NOWAIT);
408 if (mg) {
409 mg->cache = cache;
410 atomic_inc(&mg->cache->nr_allocated_migrations);
411 }
412
413 return mg;
414}
415
416static void free_migration(struct dm_cache_migration *mg)
417{
Joe Thornber88bf51842015-05-27 15:39:45 +0100418 struct cache *cache = mg->cache;
Joe Thornbera59db672015-01-23 10:16:16 +0000419
Joe Thornber88bf51842015-05-27 15:39:45 +0100420 if (atomic_dec_and_test(&cache->nr_allocated_migrations))
421 wake_up(&cache->migration_wait);
422
423 mempool_free(mg, cache->migration_pool);
Joe Thornbera59db672015-01-23 10:16:16 +0000424}
425
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000426static int prealloc_data_structs(struct cache *cache, struct prealloc *p)
427{
428 if (!p->mg) {
Joe Thornbera59db672015-01-23 10:16:16 +0000429 p->mg = alloc_migration(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000430 if (!p->mg)
431 return -ENOMEM;
432 }
433
434 if (!p->cell1) {
435 p->cell1 = alloc_prison_cell(cache);
436 if (!p->cell1)
437 return -ENOMEM;
438 }
439
440 if (!p->cell2) {
441 p->cell2 = alloc_prison_cell(cache);
442 if (!p->cell2)
443 return -ENOMEM;
444 }
445
446 return 0;
447}
448
449static void prealloc_free_structs(struct cache *cache, struct prealloc *p)
450{
451 if (p->cell2)
452 free_prison_cell(cache, p->cell2);
453
454 if (p->cell1)
455 free_prison_cell(cache, p->cell1);
456
457 if (p->mg)
Joe Thornbera59db672015-01-23 10:16:16 +0000458 free_migration(p->mg);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000459}
460
461static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p)
462{
463 struct dm_cache_migration *mg = p->mg;
464
465 BUG_ON(!mg);
466 p->mg = NULL;
467
468 return mg;
469}
470
471/*
472 * You must have a cell within the prealloc struct to return. If not this
473 * function will BUG() rather than returning NULL.
474 */
475static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p)
476{
477 struct dm_bio_prison_cell *r = NULL;
478
479 if (p->cell1) {
480 r = p->cell1;
481 p->cell1 = NULL;
482
483 } else if (p->cell2) {
484 r = p->cell2;
485 p->cell2 = NULL;
486 } else
487 BUG();
488
489 return r;
490}
491
492/*
493 * You can't have more than two cells in a prealloc struct. BUG() will be
494 * called if you try and overfill.
495 */
496static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell)
497{
498 if (!p->cell2)
499 p->cell2 = cell;
500
501 else if (!p->cell1)
502 p->cell1 = cell;
503
504 else
505 BUG();
506}
507
508/*----------------------------------------------------------------*/
509
Joe Thornber7ae34e72014-11-06 10:18:04 +0000510static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key *key)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000511{
512 key->virtual = 0;
513 key->dev = 0;
Joe Thornber7ae34e72014-11-06 10:18:04 +0000514 key->block_begin = from_oblock(begin);
515 key->block_end = from_oblock(end);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000516}
517
518/*
519 * The caller hands in a preallocated cell, and a free function for it.
520 * The cell will be freed if there's an error, or if it wasn't used because
521 * a cell with that key already exists.
522 */
523typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell);
524
Joe Thornber7ae34e72014-11-06 10:18:04 +0000525static int bio_detain_range(struct cache *cache, dm_oblock_t oblock_begin, dm_oblock_t oblock_end,
526 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
527 cell_free_fn free_fn, void *free_context,
528 struct dm_bio_prison_cell **cell_result)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000529{
530 int r;
531 struct dm_cell_key key;
532
Joe Thornber7ae34e72014-11-06 10:18:04 +0000533 build_key(oblock_begin, oblock_end, &key);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000534 r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result);
535 if (r)
536 free_fn(free_context, cell_prealloc);
537
538 return r;
539}
540
Joe Thornber7ae34e72014-11-06 10:18:04 +0000541static int bio_detain(struct cache *cache, dm_oblock_t oblock,
542 struct bio *bio, struct dm_bio_prison_cell *cell_prealloc,
543 cell_free_fn free_fn, void *free_context,
544 struct dm_bio_prison_cell **cell_result)
545{
546 dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL);
547 return bio_detain_range(cache, oblock, end, bio,
548 cell_prealloc, free_fn, free_context, cell_result);
549}
550
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000551static int get_cell(struct cache *cache,
552 dm_oblock_t oblock,
553 struct prealloc *structs,
554 struct dm_bio_prison_cell **cell_result)
555{
556 int r;
557 struct dm_cell_key key;
558 struct dm_bio_prison_cell *cell_prealloc;
559
560 cell_prealloc = prealloc_get_cell(structs);
561
Joe Thornber7ae34e72014-11-06 10:18:04 +0000562 build_key(oblock, to_oblock(from_oblock(oblock) + 1ULL), &key);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000563 r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result);
564 if (r)
565 prealloc_put_cell(structs, cell_prealloc);
566
567 return r;
568}
569
Joe Thornberaeed1422013-05-10 14:37:18 +0100570/*----------------------------------------------------------------*/
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000571
572static bool is_dirty(struct cache *cache, dm_cblock_t b)
573{
574 return test_bit(from_cblock(b), cache->dirty_bitset);
575}
576
577static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
578{
579 if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) {
Anssi Hannula44fa8162014-08-01 11:55:47 -0400580 atomic_inc(&cache->nr_dirty);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000581 policy_set_dirty(cache->policy, oblock);
582 }
583}
584
585static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock)
586{
587 if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) {
588 policy_clear_dirty(cache->policy, oblock);
Anssi Hannula44fa8162014-08-01 11:55:47 -0400589 if (atomic_dec_return(&cache->nr_dirty) == 0)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000590 dm_table_event(cache->ti->table);
591 }
592}
593
594/*----------------------------------------------------------------*/
Joe Thornberaeed1422013-05-10 14:37:18 +0100595
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000596static bool block_size_is_power_of_two(struct cache *cache)
597{
598 return cache->sectors_per_block_shift >= 0;
599}
600
Mikulas Patocka43aeaa22013-07-10 23:41:17 +0100601/* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */
602#if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6
603__always_inline
604#endif
Joe Thornber414dd672013-03-20 17:21:25 +0000605static dm_block_t block_div(dm_block_t b, uint32_t n)
606{
607 do_div(b, n);
608
609 return b;
610}
611
Joe Thornber7ae34e72014-11-06 10:18:04 +0000612static dm_block_t oblocks_per_dblock(struct cache *cache)
613{
614 dm_block_t oblocks = cache->discard_block_size;
615
616 if (block_size_is_power_of_two(cache))
617 oblocks >>= cache->sectors_per_block_shift;
618 else
619 oblocks = block_div(oblocks, cache->sectors_per_block);
620
621 return oblocks;
622}
623
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000624static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock)
625{
Joe Thornber7ae34e72014-11-06 10:18:04 +0000626 return to_dblock(block_div(from_oblock(oblock),
627 oblocks_per_dblock(cache)));
628}
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000629
Joe Thornber7ae34e72014-11-06 10:18:04 +0000630static dm_oblock_t dblock_to_oblock(struct cache *cache, dm_dblock_t dblock)
631{
632 return to_oblock(from_dblock(dblock) * oblocks_per_dblock(cache));
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000633}
634
635static void set_discard(struct cache *cache, dm_dblock_t b)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000636{
637 unsigned long flags;
638
Joe Thornber7ae34e72014-11-06 10:18:04 +0000639 BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000640 atomic_inc(&cache->stats.discard_count);
641
642 spin_lock_irqsave(&cache->lock, flags);
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000643 set_bit(from_dblock(b), cache->discard_bitset);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000644 spin_unlock_irqrestore(&cache->lock, flags);
645}
646
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000647static void clear_discard(struct cache *cache, dm_dblock_t b)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000648{
649 unsigned long flags;
650
651 spin_lock_irqsave(&cache->lock, flags);
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000652 clear_bit(from_dblock(b), cache->discard_bitset);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000653 spin_unlock_irqrestore(&cache->lock, flags);
654}
655
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000656static bool is_discarded(struct cache *cache, dm_dblock_t b)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000657{
658 int r;
659 unsigned long flags;
660
661 spin_lock_irqsave(&cache->lock, flags);
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000662 r = test_bit(from_dblock(b), cache->discard_bitset);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000663 spin_unlock_irqrestore(&cache->lock, flags);
664
665 return r;
666}
667
668static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b)
669{
670 int r;
671 unsigned long flags;
672
673 spin_lock_irqsave(&cache->lock, flags);
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000674 r = test_bit(from_dblock(oblock_to_dblock(cache, b)),
675 cache->discard_bitset);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000676 spin_unlock_irqrestore(&cache->lock, flags);
677
678 return r;
679}
680
681/*----------------------------------------------------------------*/
682
683static void load_stats(struct cache *cache)
684{
685 struct dm_cache_statistics stats;
686
687 dm_cache_metadata_get_stats(cache->cmd, &stats);
688 atomic_set(&cache->stats.read_hit, stats.read_hits);
689 atomic_set(&cache->stats.read_miss, stats.read_misses);
690 atomic_set(&cache->stats.write_hit, stats.write_hits);
691 atomic_set(&cache->stats.write_miss, stats.write_misses);
692}
693
694static void save_stats(struct cache *cache)
695{
696 struct dm_cache_statistics stats;
697
Joe Thornber028ae9f2015-04-22 16:42:35 -0400698 if (get_cache_mode(cache) >= CM_READ_ONLY)
699 return;
700
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000701 stats.read_hits = atomic_read(&cache->stats.read_hit);
702 stats.read_misses = atomic_read(&cache->stats.read_miss);
703 stats.write_hits = atomic_read(&cache->stats.write_hit);
704 stats.write_misses = atomic_read(&cache->stats.write_miss);
705
706 dm_cache_metadata_set_stats(cache->cmd, &stats);
707}
708
709/*----------------------------------------------------------------
710 * Per bio data
711 *--------------------------------------------------------------*/
Mike Snitzer19b00922013-04-05 15:36:34 +0100712
713/*
714 * If using writeback, leave out struct per_bio_data's writethrough fields.
715 */
716#define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache))
717#define PB_DATA_SIZE_WT (sizeof(struct per_bio_data))
718
Joe Thornber2ee57d52013-10-24 14:10:29 -0400719static bool writethrough_mode(struct cache_features *f)
720{
721 return f->io_mode == CM_IO_WRITETHROUGH;
722}
723
724static bool writeback_mode(struct cache_features *f)
725{
726 return f->io_mode == CM_IO_WRITEBACK;
727}
728
729static bool passthrough_mode(struct cache_features *f)
730{
731 return f->io_mode == CM_IO_PASSTHROUGH;
732}
733
Mike Snitzer19b00922013-04-05 15:36:34 +0100734static size_t get_per_bio_data_size(struct cache *cache)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000735{
Joe Thornber2ee57d52013-10-24 14:10:29 -0400736 return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB;
Mike Snitzer19b00922013-04-05 15:36:34 +0100737}
738
739static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size)
740{
741 struct per_bio_data *pb = dm_per_bio_data(bio, data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000742 BUG_ON(!pb);
743 return pb;
744}
745
Mike Snitzer19b00922013-04-05 15:36:34 +0100746static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000747{
Mike Snitzer19b00922013-04-05 15:36:34 +0100748 struct per_bio_data *pb = get_per_bio_data(bio, data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000749
750 pb->tick = false;
751 pb->req_nr = dm_bio_get_target_bio_nr(bio);
752 pb->all_io_entry = NULL;
Joe Thornber066dbaa2015-05-15 15:18:01 +0100753 pb->len = 0;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000754
755 return pb;
756}
757
758/*----------------------------------------------------------------
759 * Remapping
760 *--------------------------------------------------------------*/
761static void remap_to_origin(struct cache *cache, struct bio *bio)
762{
763 bio->bi_bdev = cache->origin_dev->bdev;
764}
765
766static void remap_to_cache(struct cache *cache, struct bio *bio,
767 dm_cblock_t cblock)
768{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700769 sector_t bi_sector = bio->bi_iter.bi_sector;
Heinz Mauelshagene0d849f2014-02-27 22:46:48 +0100770 sector_t block = from_cblock(cblock);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000771
772 bio->bi_bdev = cache->cache_dev->bdev;
773 if (!block_size_is_power_of_two(cache))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700774 bio->bi_iter.bi_sector =
Heinz Mauelshagene0d849f2014-02-27 22:46:48 +0100775 (block * cache->sectors_per_block) +
Kent Overstreet4f024f32013-10-11 15:44:27 -0700776 sector_div(bi_sector, cache->sectors_per_block);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000777 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700778 bio->bi_iter.bi_sector =
Heinz Mauelshagene0d849f2014-02-27 22:46:48 +0100779 (block << cache->sectors_per_block_shift) |
Kent Overstreet4f024f32013-10-11 15:44:27 -0700780 (bi_sector & (cache->sectors_per_block - 1));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000781}
782
783static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio)
784{
785 unsigned long flags;
Mike Snitzer19b00922013-04-05 15:36:34 +0100786 size_t pb_data_size = get_per_bio_data_size(cache);
787 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000788
789 spin_lock_irqsave(&cache->lock, flags);
790 if (cache->need_tick_bio &&
Jens Axboe1eff9d32016-08-05 15:35:16 -0600791 !(bio->bi_opf & (REQ_FUA | REQ_PREFLUSH)) &&
Mike Christiee6047142016-06-05 14:32:04 -0500792 bio_op(bio) != REQ_OP_DISCARD) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000793 pb->tick = true;
794 cache->need_tick_bio = false;
795 }
796 spin_unlock_irqrestore(&cache->lock, flags);
797}
798
799static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio,
800 dm_oblock_t oblock)
801{
802 check_if_tick_bio_needed(cache, bio);
803 remap_to_origin(cache, bio);
804 if (bio_data_dir(bio) == WRITE)
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000805 clear_discard(cache, oblock_to_dblock(cache, oblock));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000806}
807
808static void remap_to_cache_dirty(struct cache *cache, struct bio *bio,
809 dm_oblock_t oblock, dm_cblock_t cblock)
810{
Joe Thornberf8e5f012013-10-21 12:51:45 +0100811 check_if_tick_bio_needed(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000812 remap_to_cache(cache, bio, cblock);
813 if (bio_data_dir(bio) == WRITE) {
814 set_dirty(cache, oblock, cblock);
Joe Thornber1bad9bc2014-11-07 14:47:07 +0000815 clear_discard(cache, oblock_to_dblock(cache, oblock));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000816 }
817}
818
819static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio)
820{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700821 sector_t block_nr = bio->bi_iter.bi_sector;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000822
823 if (!block_size_is_power_of_two(cache))
824 (void) sector_div(block_nr, cache->sectors_per_block);
825 else
826 block_nr >>= cache->sectors_per_block_shift;
827
828 return to_oblock(block_nr);
829}
830
831static int bio_triggers_commit(struct cache *cache, struct bio *bio)
832{
Jens Axboe1eff9d32016-08-05 15:35:16 -0600833 return bio->bi_opf & (REQ_PREFLUSH | REQ_FUA);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000834}
835
Joe Thornber8c081b52014-05-13 16:18:38 +0100836/*
837 * You must increment the deferred set whilst the prison cell is held. To
838 * encourage this, we ask for 'cell' to be passed in.
839 */
840static void inc_ds(struct cache *cache, struct bio *bio,
841 struct dm_bio_prison_cell *cell)
842{
843 size_t pb_data_size = get_per_bio_data_size(cache);
844 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
845
846 BUG_ON(!cell);
847 BUG_ON(pb->all_io_entry);
848
849 pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds);
850}
851
Joe Thornber066dbaa2015-05-15 15:18:01 +0100852static bool accountable_bio(struct cache *cache, struct bio *bio)
853{
854 return ((bio->bi_bdev == cache->origin_dev->bdev) &&
Mike Christiee6047142016-06-05 14:32:04 -0500855 bio_op(bio) != REQ_OP_DISCARD);
Joe Thornber066dbaa2015-05-15 15:18:01 +0100856}
857
858static void accounted_begin(struct cache *cache, struct bio *bio)
859{
860 size_t pb_data_size = get_per_bio_data_size(cache);
861 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
862
863 if (accountable_bio(cache, bio)) {
864 pb->len = bio_sectors(bio);
865 iot_io_begin(&cache->origin_tracker, pb->len);
866 }
867}
868
869static void accounted_complete(struct cache *cache, struct bio *bio)
870{
871 size_t pb_data_size = get_per_bio_data_size(cache);
872 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
873
874 iot_io_end(&cache->origin_tracker, pb->len);
875}
876
877static void accounted_request(struct cache *cache, struct bio *bio)
878{
879 accounted_begin(cache, bio);
880 generic_make_request(bio);
881}
882
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000883static void issue(struct cache *cache, struct bio *bio)
884{
885 unsigned long flags;
886
887 if (!bio_triggers_commit(cache, bio)) {
Joe Thornber066dbaa2015-05-15 15:18:01 +0100888 accounted_request(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000889 return;
890 }
891
892 /*
893 * Batch together any bios that trigger commits and then issue a
894 * single commit for them in do_worker().
895 */
896 spin_lock_irqsave(&cache->lock, flags);
897 cache->commit_requested = true;
898 bio_list_add(&cache->deferred_flush_bios, bio);
899 spin_unlock_irqrestore(&cache->lock, flags);
900}
901
Joe Thornber8c081b52014-05-13 16:18:38 +0100902static void inc_and_issue(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell *cell)
903{
904 inc_ds(cache, bio, cell);
905 issue(cache, bio);
906}
907
Joe Thornbere2e74d62013-03-20 17:21:27 +0000908static void defer_writethrough_bio(struct cache *cache, struct bio *bio)
909{
910 unsigned long flags;
911
912 spin_lock_irqsave(&cache->lock, flags);
913 bio_list_add(&cache->deferred_writethrough_bios, bio);
914 spin_unlock_irqrestore(&cache->lock, flags);
915
916 wake_worker(cache);
917}
918
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200919static void writethrough_endio(struct bio *bio)
Joe Thornbere2e74d62013-03-20 17:21:27 +0000920{
Mike Snitzer19b00922013-04-05 15:36:34 +0100921 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
Joe Thornberc9d28d52013-10-31 13:55:48 -0400922
923 dm_unhook_bio(&pb->hook_info, bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000924
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200925 if (bio->bi_error) {
926 bio_endio(bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000927 return;
928 }
929
Darrick J. Wongb844fe62013-04-05 15:36:32 +0100930 dm_bio_restore(&pb->bio_details, bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000931 remap_to_cache(pb->cache, bio, pb->cblock);
932
933 /*
934 * We can't issue this bio directly, since we're in interrupt
Joe Thornberaeed1422013-05-10 14:37:18 +0100935 * context. So it gets put on a bio list for processing by the
Joe Thornbere2e74d62013-03-20 17:21:27 +0000936 * worker thread.
937 */
938 defer_writethrough_bio(pb->cache, bio);
939}
940
941/*
942 * When running in writethrough mode we need to send writes to clean blocks
943 * to both the cache and origin devices. In future we'd like to clone the
944 * bio and send them in parallel, but for now we're doing them in
945 * series as this is easier.
946 */
947static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio,
948 dm_oblock_t oblock, dm_cblock_t cblock)
949{
Mike Snitzer19b00922013-04-05 15:36:34 +0100950 struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000951
952 pb->cache = cache;
953 pb->cblock = cblock;
Joe Thornberc9d28d52013-10-31 13:55:48 -0400954 dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL);
Darrick J. Wongb844fe62013-04-05 15:36:32 +0100955 dm_bio_record(&pb->bio_details, bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +0000956
957 remap_to_origin_clear_discard(pb->cache, bio, oblock);
958}
959
Joe Thornberc6b4fcb2013-03-01 22:45:51 +0000960/*----------------------------------------------------------------
Joe Thornber028ae9f2015-04-22 16:42:35 -0400961 * Failure modes
962 *--------------------------------------------------------------*/
963static enum cache_metadata_mode get_cache_mode(struct cache *cache)
964{
965 return cache->features.mode;
966}
967
Mike Snitzerb61d9502015-04-22 17:25:56 -0400968static const char *cache_device_name(struct cache *cache)
969{
970 return dm_device_name(dm_table_get_md(cache->ti->table));
971}
972
Joe Thornber028ae9f2015-04-22 16:42:35 -0400973static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode)
974{
975 const char *descs[] = {
976 "write",
977 "read-only",
978 "fail"
979 };
980
981 dm_table_event(cache->ti->table);
Mike Snitzerb61d9502015-04-22 17:25:56 -0400982 DMINFO("%s: switching cache to %s mode",
983 cache_device_name(cache), descs[(int)mode]);
Joe Thornber028ae9f2015-04-22 16:42:35 -0400984}
985
986static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode)
987{
Joe Thornberd14fcf32016-03-10 16:20:58 +0000988 bool needs_check;
Joe Thornber028ae9f2015-04-22 16:42:35 -0400989 enum cache_metadata_mode old_mode = get_cache_mode(cache);
990
Joe Thornberd14fcf32016-03-10 16:20:58 +0000991 if (dm_cache_metadata_needs_check(cache->cmd, &needs_check)) {
992 DMERR("unable to read needs_check flag, setting failure mode");
993 new_mode = CM_FAIL;
994 }
995
Joe Thornber028ae9f2015-04-22 16:42:35 -0400996 if (new_mode == CM_WRITE && needs_check) {
Mike Snitzerb61d9502015-04-22 17:25:56 -0400997 DMERR("%s: unable to switch cache to write mode until repaired.",
998 cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -0400999 if (old_mode != new_mode)
1000 new_mode = old_mode;
1001 else
1002 new_mode = CM_READ_ONLY;
1003 }
1004
1005 /* Never move out of fail mode */
1006 if (old_mode == CM_FAIL)
1007 new_mode = CM_FAIL;
1008
1009 switch (new_mode) {
1010 case CM_FAIL:
1011 case CM_READ_ONLY:
1012 dm_cache_metadata_set_read_only(cache->cmd);
1013 break;
1014
1015 case CM_WRITE:
1016 dm_cache_metadata_set_read_write(cache->cmd);
1017 break;
1018 }
1019
1020 cache->features.mode = new_mode;
1021
1022 if (new_mode != old_mode)
1023 notify_mode_switch(cache, new_mode);
1024}
1025
1026static void abort_transaction(struct cache *cache)
1027{
Mike Snitzerb61d9502015-04-22 17:25:56 -04001028 const char *dev_name = cache_device_name(cache);
1029
Joe Thornber028ae9f2015-04-22 16:42:35 -04001030 if (get_cache_mode(cache) >= CM_READ_ONLY)
1031 return;
1032
1033 if (dm_cache_metadata_set_needs_check(cache->cmd)) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001034 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
Joe Thornber028ae9f2015-04-22 16:42:35 -04001035 set_cache_mode(cache, CM_FAIL);
1036 }
1037
Mike Snitzerb61d9502015-04-22 17:25:56 -04001038 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
Joe Thornber028ae9f2015-04-22 16:42:35 -04001039 if (dm_cache_metadata_abort(cache->cmd)) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001040 DMERR("%s: failed to abort metadata transaction", dev_name);
Joe Thornber028ae9f2015-04-22 16:42:35 -04001041 set_cache_mode(cache, CM_FAIL);
1042 }
1043}
1044
1045static void metadata_operation_failed(struct cache *cache, const char *op, int r)
1046{
Mike Snitzerb61d9502015-04-22 17:25:56 -04001047 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1048 cache_device_name(cache), op, r);
Joe Thornber028ae9f2015-04-22 16:42:35 -04001049 abort_transaction(cache);
1050 set_cache_mode(cache, CM_READ_ONLY);
1051}
1052
1053/*----------------------------------------------------------------
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001054 * Migration processing
1055 *
1056 * Migration covers moving data from the origin device to the cache, or
1057 * vice versa.
1058 *--------------------------------------------------------------*/
Joe Thornbera59db672015-01-23 10:16:16 +00001059static void inc_io_migrations(struct cache *cache)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001060{
Joe Thornbera59db672015-01-23 10:16:16 +00001061 atomic_inc(&cache->nr_io_migrations);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001062}
1063
Joe Thornbera59db672015-01-23 10:16:16 +00001064static void dec_io_migrations(struct cache *cache)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001065{
Joe Thornbera59db672015-01-23 10:16:16 +00001066 atomic_dec(&cache->nr_io_migrations);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001067}
1068
Joe Thornber651f5fa2015-05-15 15:26:08 +01001069static bool discard_or_flush(struct bio *bio)
1070{
Mike Christiee6047142016-06-05 14:32:04 -05001071 return bio_op(bio) == REQ_OP_DISCARD ||
Jens Axboe1eff9d32016-08-05 15:35:16 -06001072 bio->bi_opf & (REQ_PREFLUSH | REQ_FUA);
Joe Thornber651f5fa2015-05-15 15:26:08 +01001073}
1074
1075static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell)
1076{
Mike Snitzerdc9cee52015-08-31 15:41:34 -04001077 if (discard_or_flush(cell->holder)) {
Joe Thornber651f5fa2015-05-15 15:26:08 +01001078 /*
Mike Snitzerdc9cee52015-08-31 15:41:34 -04001079 * We have to handle these bios individually.
Joe Thornber651f5fa2015-05-15 15:26:08 +01001080 */
Mike Snitzerdc9cee52015-08-31 15:41:34 -04001081 dm_cell_release(cache->prison, cell, &cache->deferred_bios);
1082 free_prison_cell(cache, cell);
1083 } else
Joe Thornber651f5fa2015-05-15 15:26:08 +01001084 list_add_tail(&cell->user_list, &cache->deferred_cells);
1085}
1086
1087static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, bool holder)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001088{
1089 unsigned long flags;
1090
Joe Thornber651f5fa2015-05-15 15:26:08 +01001091 if (!holder && dm_cell_promote_or_release(cache->prison, cell)) {
1092 /*
1093 * There was no prisoner to promote to holder, the
1094 * cell has been released.
1095 */
1096 free_prison_cell(cache, cell);
1097 return;
1098 }
1099
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001100 spin_lock_irqsave(&cache->lock, flags);
Joe Thornber651f5fa2015-05-15 15:26:08 +01001101 __cell_defer(cache, cell);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001102 spin_unlock_irqrestore(&cache->lock, flags);
1103
1104 wake_worker(cache);
1105}
1106
Joe Thornber651f5fa2015-05-15 15:26:08 +01001107static void cell_error_with_code(struct cache *cache, struct dm_bio_prison_cell *cell, int err)
1108{
1109 dm_cell_error(cache->prison, cell, err);
Mike Snitzerdc9cee52015-08-31 15:41:34 -04001110 free_prison_cell(cache, cell);
Joe Thornber651f5fa2015-05-15 15:26:08 +01001111}
1112
1113static void cell_requeue(struct cache *cache, struct dm_bio_prison_cell *cell)
1114{
1115 cell_error_with_code(cache, cell, DM_ENDIO_REQUEUE);
1116}
1117
Joe Thornbera59db672015-01-23 10:16:16 +00001118static void free_io_migration(struct dm_cache_migration *mg)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001119{
Joe Thornbercc7da0b2015-09-01 11:38:19 +01001120 struct cache *cache = mg->cache;
1121
1122 dec_io_migrations(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001123 free_migration(mg);
Joe Thornbercc7da0b2015-09-01 11:38:19 +01001124 wake_worker(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001125}
1126
1127static void migration_failure(struct dm_cache_migration *mg)
1128{
1129 struct cache *cache = mg->cache;
Mike Snitzerb61d9502015-04-22 17:25:56 -04001130 const char *dev_name = cache_device_name(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001131
1132 if (mg->writeback) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001133 DMERR_LIMIT("%s: writeback failed; couldn't copy block", dev_name);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001134 set_dirty(cache, mg->old_oblock, mg->cblock);
1135 cell_defer(cache, mg->old_ocell, false);
1136
1137 } else if (mg->demote) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001138 DMERR_LIMIT("%s: demotion failed; couldn't copy block", dev_name);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001139 policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock);
1140
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +02001141 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001142 if (mg->promote)
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +02001143 cell_defer(cache, mg->new_ocell, true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001144 } else {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001145 DMERR_LIMIT("%s: promotion failed; couldn't copy block", dev_name);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001146 policy_remove_mapping(cache->policy, mg->new_oblock);
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +02001147 cell_defer(cache, mg->new_ocell, true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001148 }
1149
Joe Thornbera59db672015-01-23 10:16:16 +00001150 free_io_migration(mg);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001151}
1152
1153static void migration_success_pre_commit(struct dm_cache_migration *mg)
1154{
Joe Thornber028ae9f2015-04-22 16:42:35 -04001155 int r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001156 unsigned long flags;
1157 struct cache *cache = mg->cache;
1158
1159 if (mg->writeback) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001160 clear_dirty(cache, mg->old_oblock, mg->cblock);
Anssi Hannula40aa9782014-09-05 03:11:28 +03001161 cell_defer(cache, mg->old_ocell, false);
Joe Thornbera59db672015-01-23 10:16:16 +00001162 free_io_migration(mg);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001163 return;
1164
1165 } else if (mg->demote) {
Joe Thornber028ae9f2015-04-22 16:42:35 -04001166 r = dm_cache_remove_mapping(cache->cmd, mg->cblock);
1167 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001168 DMERR_LIMIT("%s: demotion failed; couldn't update on disk metadata",
1169 cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -04001170 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001171 policy_force_mapping(cache->policy, mg->new_oblock,
1172 mg->old_oblock);
1173 if (mg->promote)
1174 cell_defer(cache, mg->new_ocell, true);
Joe Thornbera59db672015-01-23 10:16:16 +00001175 free_io_migration(mg);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001176 return;
1177 }
1178 } else {
Joe Thornber028ae9f2015-04-22 16:42:35 -04001179 r = dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock);
1180 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001181 DMERR_LIMIT("%s: promotion failed; couldn't update on disk metadata",
1182 cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -04001183 metadata_operation_failed(cache, "dm_cache_insert_mapping", r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001184 policy_remove_mapping(cache->policy, mg->new_oblock);
Joe Thornbera59db672015-01-23 10:16:16 +00001185 free_io_migration(mg);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001186 return;
1187 }
1188 }
1189
1190 spin_lock_irqsave(&cache->lock, flags);
1191 list_add_tail(&mg->list, &cache->need_commit_migrations);
1192 cache->commit_requested = true;
1193 spin_unlock_irqrestore(&cache->lock, flags);
1194}
1195
1196static void migration_success_post_commit(struct dm_cache_migration *mg)
1197{
1198 unsigned long flags;
1199 struct cache *cache = mg->cache;
1200
1201 if (mg->writeback) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001202 DMWARN_LIMIT("%s: writeback unexpectedly triggered commit",
1203 cache_device_name(cache));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001204 return;
1205
1206 } else if (mg->demote) {
Heinz Mauelshagen80f659f2013-10-14 17:10:47 +02001207 cell_defer(cache, mg->old_ocell, mg->promote ? false : true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001208
1209 if (mg->promote) {
1210 mg->demote = false;
1211
1212 spin_lock_irqsave(&cache->lock, flags);
1213 list_add_tail(&mg->list, &cache->quiesced_migrations);
1214 spin_unlock_irqrestore(&cache->lock, flags);
1215
Joe Thornber65790ff2013-11-08 16:39:50 +00001216 } else {
1217 if (mg->invalidate)
1218 policy_remove_mapping(cache->policy, mg->old_oblock);
Joe Thornbera59db672015-01-23 10:16:16 +00001219 free_io_migration(mg);
Joe Thornber65790ff2013-11-08 16:39:50 +00001220 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001221
1222 } else {
Joe Thornber1e321342014-11-27 12:26:46 +00001223 if (mg->requeue_holder) {
1224 clear_dirty(cache, mg->new_oblock, mg->cblock);
Joe Thornberc9d28d52013-10-31 13:55:48 -04001225 cell_defer(cache, mg->new_ocell, true);
Joe Thornber1e321342014-11-27 12:26:46 +00001226 } else {
1227 /*
1228 * The block was promoted via an overwrite, so it's dirty.
1229 */
1230 set_dirty(cache, mg->new_oblock, mg->cblock);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001231 bio_endio(mg->new_ocell->holder);
Joe Thornberc9d28d52013-10-31 13:55:48 -04001232 cell_defer(cache, mg->new_ocell, false);
1233 }
Joe Thornbera59db672015-01-23 10:16:16 +00001234 free_io_migration(mg);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001235 }
1236}
1237
1238static void copy_complete(int read_err, unsigned long write_err, void *context)
1239{
1240 unsigned long flags;
1241 struct dm_cache_migration *mg = (struct dm_cache_migration *) context;
1242 struct cache *cache = mg->cache;
1243
1244 if (read_err || write_err)
1245 mg->err = true;
1246
1247 spin_lock_irqsave(&cache->lock, flags);
1248 list_add_tail(&mg->list, &cache->completed_migrations);
1249 spin_unlock_irqrestore(&cache->lock, flags);
1250
1251 wake_worker(cache);
1252}
1253
Joe Thornber7ae34e72014-11-06 10:18:04 +00001254static void issue_copy(struct dm_cache_migration *mg)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001255{
1256 int r;
1257 struct dm_io_region o_region, c_region;
1258 struct cache *cache = mg->cache;
Heinz Mauelshagen8b9d9662014-03-12 00:40:05 +01001259 sector_t cblock = from_cblock(mg->cblock);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001260
1261 o_region.bdev = cache->origin_dev->bdev;
1262 o_region.count = cache->sectors_per_block;
1263
1264 c_region.bdev = cache->cache_dev->bdev;
Heinz Mauelshagen8b9d9662014-03-12 00:40:05 +01001265 c_region.sector = cblock * cache->sectors_per_block;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001266 c_region.count = cache->sectors_per_block;
1267
1268 if (mg->writeback || mg->demote) {
1269 /* demote */
1270 o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block;
1271 r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg);
1272 } else {
1273 /* promote */
1274 o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block;
1275 r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg);
1276 }
1277
Heinz Mauelshagen2c2263c2013-10-14 17:14:45 +02001278 if (r < 0) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04001279 DMERR_LIMIT("%s: issuing migration failed", cache_device_name(cache));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001280 migration_failure(mg);
Heinz Mauelshagen2c2263c2013-10-14 17:14:45 +02001281 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001282}
1283
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001284static void overwrite_endio(struct bio *bio)
Joe Thornberc9d28d52013-10-31 13:55:48 -04001285{
1286 struct dm_cache_migration *mg = bio->bi_private;
1287 struct cache *cache = mg->cache;
1288 size_t pb_data_size = get_per_bio_data_size(cache);
1289 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1290 unsigned long flags;
1291
Mike Snitzer80ae49a2014-01-31 14:30:37 -05001292 dm_unhook_bio(&pb->hook_info, bio);
1293
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001294 if (bio->bi_error)
Joe Thornberc9d28d52013-10-31 13:55:48 -04001295 mg->err = true;
1296
Mike Snitzer80ae49a2014-01-31 14:30:37 -05001297 mg->requeue_holder = false;
1298
Joe Thornberc9d28d52013-10-31 13:55:48 -04001299 spin_lock_irqsave(&cache->lock, flags);
1300 list_add_tail(&mg->list, &cache->completed_migrations);
Joe Thornberc9d28d52013-10-31 13:55:48 -04001301 spin_unlock_irqrestore(&cache->lock, flags);
1302
1303 wake_worker(cache);
1304}
1305
1306static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio)
1307{
1308 size_t pb_data_size = get_per_bio_data_size(mg->cache);
1309 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
1310
1311 dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg);
1312 remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock);
Joe Thornber8c081b52014-05-13 16:18:38 +01001313
1314 /*
1315 * No need to inc_ds() here, since the cell will be held for the
1316 * duration of the io.
1317 */
Joe Thornber066dbaa2015-05-15 15:18:01 +01001318 accounted_request(mg->cache, bio);
Joe Thornberc9d28d52013-10-31 13:55:48 -04001319}
1320
1321static bool bio_writes_complete_block(struct cache *cache, struct bio *bio)
1322{
1323 return (bio_data_dir(bio) == WRITE) &&
Kent Overstreet4f024f32013-10-11 15:44:27 -07001324 (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT));
Joe Thornberc9d28d52013-10-31 13:55:48 -04001325}
1326
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001327static void avoid_copy(struct dm_cache_migration *mg)
1328{
1329 atomic_inc(&mg->cache->stats.copies_avoided);
1330 migration_success_pre_commit(mg);
1331}
1332
Joe Thornber7ae34e72014-11-06 10:18:04 +00001333static void calc_discard_block_range(struct cache *cache, struct bio *bio,
1334 dm_dblock_t *b, dm_dblock_t *e)
1335{
1336 sector_t sb = bio->bi_iter.bi_sector;
1337 sector_t se = bio_end_sector(bio);
1338
1339 *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size));
1340
1341 if (se - sb < cache->discard_block_size)
1342 *e = *b;
1343 else
1344 *e = to_dblock(block_div(se, cache->discard_block_size));
1345}
1346
1347static void issue_discard(struct dm_cache_migration *mg)
1348{
1349 dm_dblock_t b, e;
1350 struct bio *bio = mg->new_ocell->holder;
Joe Thornbercc7da0b2015-09-01 11:38:19 +01001351 struct cache *cache = mg->cache;
Joe Thornber7ae34e72014-11-06 10:18:04 +00001352
Joe Thornbercc7da0b2015-09-01 11:38:19 +01001353 calc_discard_block_range(cache, bio, &b, &e);
Joe Thornber7ae34e72014-11-06 10:18:04 +00001354 while (b != e) {
Joe Thornbercc7da0b2015-09-01 11:38:19 +01001355 set_discard(cache, b);
Joe Thornber7ae34e72014-11-06 10:18:04 +00001356 b = to_dblock(from_dblock(b) + 1);
1357 }
1358
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001359 bio_endio(bio);
Joe Thornbercc7da0b2015-09-01 11:38:19 +01001360 cell_defer(cache, mg->new_ocell, false);
Joe Thornber7ae34e72014-11-06 10:18:04 +00001361 free_migration(mg);
Joe Thornbercc7da0b2015-09-01 11:38:19 +01001362 wake_worker(cache);
Joe Thornber7ae34e72014-11-06 10:18:04 +00001363}
1364
1365static void issue_copy_or_discard(struct dm_cache_migration *mg)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001366{
1367 bool avoid;
1368 struct cache *cache = mg->cache;
1369
Joe Thornber7ae34e72014-11-06 10:18:04 +00001370 if (mg->discard) {
1371 issue_discard(mg);
1372 return;
1373 }
1374
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001375 if (mg->writeback || mg->demote)
1376 avoid = !is_dirty(cache, mg->cblock) ||
1377 is_discarded_oblock(cache, mg->old_oblock);
Joe Thornberc9d28d52013-10-31 13:55:48 -04001378 else {
1379 struct bio *bio = mg->new_ocell->holder;
1380
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001381 avoid = is_discarded_oblock(cache, mg->new_oblock);
1382
Joe Thornberf29a3142014-11-27 12:21:08 +00001383 if (writeback_mode(&cache->features) &&
1384 !avoid && bio_writes_complete_block(cache, bio)) {
Joe Thornberc9d28d52013-10-31 13:55:48 -04001385 issue_overwrite(mg, bio);
1386 return;
1387 }
1388 }
1389
Joe Thornber7ae34e72014-11-06 10:18:04 +00001390 avoid ? avoid_copy(mg) : issue_copy(mg);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001391}
1392
1393static void complete_migration(struct dm_cache_migration *mg)
1394{
1395 if (mg->err)
1396 migration_failure(mg);
1397 else
1398 migration_success_pre_commit(mg);
1399}
1400
1401static void process_migrations(struct cache *cache, struct list_head *head,
1402 void (*fn)(struct dm_cache_migration *))
1403{
1404 unsigned long flags;
1405 struct list_head list;
1406 struct dm_cache_migration *mg, *tmp;
1407
1408 INIT_LIST_HEAD(&list);
1409 spin_lock_irqsave(&cache->lock, flags);
1410 list_splice_init(head, &list);
1411 spin_unlock_irqrestore(&cache->lock, flags);
1412
1413 list_for_each_entry_safe(mg, tmp, &list, list)
1414 fn(mg);
1415}
1416
1417static void __queue_quiesced_migration(struct dm_cache_migration *mg)
1418{
1419 list_add_tail(&mg->list, &mg->cache->quiesced_migrations);
1420}
1421
1422static void queue_quiesced_migration(struct dm_cache_migration *mg)
1423{
1424 unsigned long flags;
1425 struct cache *cache = mg->cache;
1426
1427 spin_lock_irqsave(&cache->lock, flags);
1428 __queue_quiesced_migration(mg);
1429 spin_unlock_irqrestore(&cache->lock, flags);
1430
1431 wake_worker(cache);
1432}
1433
1434static void queue_quiesced_migrations(struct cache *cache, struct list_head *work)
1435{
1436 unsigned long flags;
1437 struct dm_cache_migration *mg, *tmp;
1438
1439 spin_lock_irqsave(&cache->lock, flags);
1440 list_for_each_entry_safe(mg, tmp, work, list)
1441 __queue_quiesced_migration(mg);
1442 spin_unlock_irqrestore(&cache->lock, flags);
1443
1444 wake_worker(cache);
1445}
1446
1447static void check_for_quiesced_migrations(struct cache *cache,
1448 struct per_bio_data *pb)
1449{
1450 struct list_head work;
1451
1452 if (!pb->all_io_entry)
1453 return;
1454
1455 INIT_LIST_HEAD(&work);
Joe Thornber8c081b52014-05-13 16:18:38 +01001456 dm_deferred_entry_dec(pb->all_io_entry, &work);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001457
1458 if (!list_empty(&work))
1459 queue_quiesced_migrations(cache, &work);
1460}
1461
1462static void quiesce_migration(struct dm_cache_migration *mg)
1463{
1464 if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list))
1465 queue_quiesced_migration(mg);
1466}
1467
1468static void promote(struct cache *cache, struct prealloc *structs,
1469 dm_oblock_t oblock, dm_cblock_t cblock,
1470 struct dm_bio_prison_cell *cell)
1471{
1472 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1473
1474 mg->err = false;
Joe Thornber7ae34e72014-11-06 10:18:04 +00001475 mg->discard = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001476 mg->writeback = false;
1477 mg->demote = false;
1478 mg->promote = true;
Joe Thornberc9d28d52013-10-31 13:55:48 -04001479 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001480 mg->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001481 mg->cache = cache;
1482 mg->new_oblock = oblock;
1483 mg->cblock = cblock;
1484 mg->old_ocell = NULL;
1485 mg->new_ocell = cell;
1486 mg->start_jiffies = jiffies;
1487
Joe Thornbera59db672015-01-23 10:16:16 +00001488 inc_io_migrations(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001489 quiesce_migration(mg);
1490}
1491
1492static void writeback(struct cache *cache, struct prealloc *structs,
1493 dm_oblock_t oblock, dm_cblock_t cblock,
1494 struct dm_bio_prison_cell *cell)
1495{
1496 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1497
1498 mg->err = false;
Joe Thornber7ae34e72014-11-06 10:18:04 +00001499 mg->discard = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001500 mg->writeback = true;
1501 mg->demote = false;
1502 mg->promote = false;
Joe Thornberc9d28d52013-10-31 13:55:48 -04001503 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001504 mg->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001505 mg->cache = cache;
1506 mg->old_oblock = oblock;
1507 mg->cblock = cblock;
1508 mg->old_ocell = cell;
1509 mg->new_ocell = NULL;
1510 mg->start_jiffies = jiffies;
1511
Joe Thornbera59db672015-01-23 10:16:16 +00001512 inc_io_migrations(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001513 quiesce_migration(mg);
1514}
1515
1516static void demote_then_promote(struct cache *cache, struct prealloc *structs,
1517 dm_oblock_t old_oblock, dm_oblock_t new_oblock,
1518 dm_cblock_t cblock,
1519 struct dm_bio_prison_cell *old_ocell,
1520 struct dm_bio_prison_cell *new_ocell)
1521{
1522 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1523
1524 mg->err = false;
Joe Thornber7ae34e72014-11-06 10:18:04 +00001525 mg->discard = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001526 mg->writeback = false;
1527 mg->demote = true;
1528 mg->promote = true;
Joe Thornberc9d28d52013-10-31 13:55:48 -04001529 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001530 mg->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001531 mg->cache = cache;
1532 mg->old_oblock = old_oblock;
1533 mg->new_oblock = new_oblock;
1534 mg->cblock = cblock;
1535 mg->old_ocell = old_ocell;
1536 mg->new_ocell = new_ocell;
1537 mg->start_jiffies = jiffies;
1538
Joe Thornbera59db672015-01-23 10:16:16 +00001539 inc_io_migrations(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001540 quiesce_migration(mg);
1541}
1542
Joe Thornber2ee57d52013-10-24 14:10:29 -04001543/*
1544 * Invalidate a cache entry. No writeback occurs; any changes in the cache
1545 * block are thrown away.
1546 */
1547static void invalidate(struct cache *cache, struct prealloc *structs,
1548 dm_oblock_t oblock, dm_cblock_t cblock,
1549 struct dm_bio_prison_cell *cell)
1550{
1551 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1552
1553 mg->err = false;
Joe Thornber7ae34e72014-11-06 10:18:04 +00001554 mg->discard = false;
Joe Thornber2ee57d52013-10-24 14:10:29 -04001555 mg->writeback = false;
1556 mg->demote = true;
1557 mg->promote = false;
1558 mg->requeue_holder = true;
Joe Thornber65790ff2013-11-08 16:39:50 +00001559 mg->invalidate = true;
Joe Thornber2ee57d52013-10-24 14:10:29 -04001560 mg->cache = cache;
1561 mg->old_oblock = oblock;
1562 mg->cblock = cblock;
1563 mg->old_ocell = cell;
1564 mg->new_ocell = NULL;
1565 mg->start_jiffies = jiffies;
1566
Joe Thornbera59db672015-01-23 10:16:16 +00001567 inc_io_migrations(cache);
Joe Thornber2ee57d52013-10-24 14:10:29 -04001568 quiesce_migration(mg);
1569}
1570
Joe Thornber7ae34e72014-11-06 10:18:04 +00001571static void discard(struct cache *cache, struct prealloc *structs,
1572 struct dm_bio_prison_cell *cell)
1573{
1574 struct dm_cache_migration *mg = prealloc_get_migration(structs);
1575
1576 mg->err = false;
1577 mg->discard = true;
1578 mg->writeback = false;
1579 mg->demote = false;
1580 mg->promote = false;
1581 mg->requeue_holder = false;
1582 mg->invalidate = false;
1583 mg->cache = cache;
1584 mg->old_ocell = NULL;
1585 mg->new_ocell = cell;
1586 mg->start_jiffies = jiffies;
1587
1588 quiesce_migration(mg);
1589}
1590
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001591/*----------------------------------------------------------------
1592 * bio processing
1593 *--------------------------------------------------------------*/
1594static void defer_bio(struct cache *cache, struct bio *bio)
1595{
1596 unsigned long flags;
1597
1598 spin_lock_irqsave(&cache->lock, flags);
1599 bio_list_add(&cache->deferred_bios, bio);
1600 spin_unlock_irqrestore(&cache->lock, flags);
1601
1602 wake_worker(cache);
1603}
1604
1605static void process_flush_bio(struct cache *cache, struct bio *bio)
1606{
Mike Snitzer19b00922013-04-05 15:36:34 +01001607 size_t pb_data_size = get_per_bio_data_size(cache);
1608 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001609
Kent Overstreet4f024f32013-10-11 15:44:27 -07001610 BUG_ON(bio->bi_iter.bi_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001611 if (!pb->req_nr)
1612 remap_to_origin(cache, bio);
1613 else
1614 remap_to_cache(cache, bio, 0);
1615
Joe Thornber8c081b52014-05-13 16:18:38 +01001616 /*
Mike Christie28a8f0d2016-06-05 14:32:25 -05001617 * REQ_PREFLUSH is not directed at any particular block so we don't
1618 * need to inc_ds(). REQ_FUA's are split into a write + REQ_PREFLUSH
Joe Thornber8c081b52014-05-13 16:18:38 +01001619 * by dm-core.
1620 */
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001621 issue(cache, bio);
1622}
1623
Joe Thornber7ae34e72014-11-06 10:18:04 +00001624static void process_discard_bio(struct cache *cache, struct prealloc *structs,
1625 struct bio *bio)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001626{
Joe Thornber7ae34e72014-11-06 10:18:04 +00001627 int r;
1628 dm_dblock_t b, e;
1629 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001630
Joe Thornber7ae34e72014-11-06 10:18:04 +00001631 calc_discard_block_range(cache, bio, &b, &e);
1632 if (b == e) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001633 bio_endio(bio);
Joe Thornber7ae34e72014-11-06 10:18:04 +00001634 return;
1635 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001636
Joe Thornber7ae34e72014-11-06 10:18:04 +00001637 cell_prealloc = prealloc_get_cell(structs);
1638 r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc,
1639 (cell_free_fn) prealloc_put_cell,
1640 structs, &new_ocell);
1641 if (r > 0)
1642 return;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001643
Joe Thornber7ae34e72014-11-06 10:18:04 +00001644 discard(cache, structs, new_ocell);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001645}
1646
1647static bool spare_migration_bandwidth(struct cache *cache)
1648{
Joe Thornbera59db672015-01-23 10:16:16 +00001649 sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) *
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001650 cache->sectors_per_block;
1651 return current_volume < cache->migration_threshold;
1652}
1653
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001654static void inc_hit_counter(struct cache *cache, struct bio *bio)
1655{
1656 atomic_inc(bio_data_dir(bio) == READ ?
1657 &cache->stats.read_hit : &cache->stats.write_hit);
1658}
1659
1660static void inc_miss_counter(struct cache *cache, struct bio *bio)
1661{
1662 atomic_inc(bio_data_dir(bio) == READ ?
1663 &cache->stats.read_miss : &cache->stats.write_miss);
1664}
1665
Joe Thornberfb4100a2015-05-20 10:30:32 +01001666/*----------------------------------------------------------------*/
1667
Joe Thornber651f5fa2015-05-15 15:26:08 +01001668struct inc_detail {
1669 struct cache *cache;
1670 struct bio_list bios_for_issue;
1671 struct bio_list unhandled_bios;
1672 bool any_writes;
1673};
1674
1675static void inc_fn(void *context, struct dm_bio_prison_cell *cell)
1676{
1677 struct bio *bio;
1678 struct inc_detail *detail = context;
1679 struct cache *cache = detail->cache;
1680
1681 inc_ds(cache, cell->holder, cell);
1682 if (bio_data_dir(cell->holder) == WRITE)
1683 detail->any_writes = true;
1684
1685 while ((bio = bio_list_pop(&cell->bios))) {
1686 if (discard_or_flush(bio)) {
1687 bio_list_add(&detail->unhandled_bios, bio);
1688 continue;
1689 }
1690
1691 if (bio_data_dir(bio) == WRITE)
1692 detail->any_writes = true;
1693
1694 bio_list_add(&detail->bios_for_issue, bio);
1695 inc_ds(cache, bio, cell);
1696 }
1697}
1698
1699// FIXME: refactor these two
1700static void remap_cell_to_origin_clear_discard(struct cache *cache,
1701 struct dm_bio_prison_cell *cell,
1702 dm_oblock_t oblock, bool issue_holder)
1703{
1704 struct bio *bio;
1705 unsigned long flags;
1706 struct inc_detail detail;
1707
1708 detail.cache = cache;
1709 bio_list_init(&detail.bios_for_issue);
1710 bio_list_init(&detail.unhandled_bios);
1711 detail.any_writes = false;
1712
1713 spin_lock_irqsave(&cache->lock, flags);
1714 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1715 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1716 spin_unlock_irqrestore(&cache->lock, flags);
1717
1718 remap_to_origin(cache, cell->holder);
1719 if (issue_holder)
1720 issue(cache, cell->holder);
1721 else
1722 accounted_begin(cache, cell->holder);
1723
1724 if (detail.any_writes)
1725 clear_discard(cache, oblock_to_dblock(cache, oblock));
1726
1727 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1728 remap_to_origin(cache, bio);
1729 issue(cache, bio);
1730 }
Joe Thornber9153df72015-08-31 18:20:08 +01001731
1732 free_prison_cell(cache, cell);
Joe Thornber651f5fa2015-05-15 15:26:08 +01001733}
1734
1735static void remap_cell_to_cache_dirty(struct cache *cache, struct dm_bio_prison_cell *cell,
1736 dm_oblock_t oblock, dm_cblock_t cblock, bool issue_holder)
1737{
1738 struct bio *bio;
1739 unsigned long flags;
1740 struct inc_detail detail;
1741
1742 detail.cache = cache;
1743 bio_list_init(&detail.bios_for_issue);
1744 bio_list_init(&detail.unhandled_bios);
1745 detail.any_writes = false;
1746
1747 spin_lock_irqsave(&cache->lock, flags);
1748 dm_cell_visit_release(cache->prison, inc_fn, &detail, cell);
1749 bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios);
1750 spin_unlock_irqrestore(&cache->lock, flags);
1751
1752 remap_to_cache(cache, cell->holder, cblock);
1753 if (issue_holder)
1754 issue(cache, cell->holder);
1755 else
1756 accounted_begin(cache, cell->holder);
1757
1758 if (detail.any_writes) {
1759 set_dirty(cache, oblock, cblock);
1760 clear_discard(cache, oblock_to_dblock(cache, oblock));
1761 }
1762
1763 while ((bio = bio_list_pop(&detail.bios_for_issue))) {
1764 remap_to_cache(cache, bio, cblock);
1765 issue(cache, bio);
1766 }
Joe Thornber9153df72015-08-31 18:20:08 +01001767
1768 free_prison_cell(cache, cell);
Joe Thornber651f5fa2015-05-15 15:26:08 +01001769}
1770
1771/*----------------------------------------------------------------*/
1772
Joe Thornberfb4100a2015-05-20 10:30:32 +01001773struct old_oblock_lock {
1774 struct policy_locker locker;
1775 struct cache *cache;
1776 struct prealloc *structs;
1777 struct dm_bio_prison_cell *cell;
1778};
1779
1780static int null_locker(struct policy_locker *locker, dm_oblock_t b)
1781{
1782 /* This should never be called */
1783 BUG();
1784 return 0;
1785}
1786
1787static int cell_locker(struct policy_locker *locker, dm_oblock_t b)
1788{
1789 struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker);
1790 struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs);
1791
1792 return bio_detain(l->cache, b, NULL, cell_prealloc,
1793 (cell_free_fn) prealloc_put_cell,
1794 l->structs, &l->cell);
1795}
1796
Joe Thornber651f5fa2015-05-15 15:26:08 +01001797static void process_cell(struct cache *cache, struct prealloc *structs,
1798 struct dm_bio_prison_cell *new_ocell)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001799{
1800 int r;
1801 bool release_cell = true;
Joe Thornber651f5fa2015-05-15 15:26:08 +01001802 struct bio *bio = new_ocell->holder;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001803 dm_oblock_t block = get_bio_block(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001804 struct policy_result lookup_result;
Joe Thornber2ee57d52013-10-24 14:10:29 -04001805 bool passthrough = passthrough_mode(&cache->features);
Joe Thornber40775252015-05-15 15:29:58 +01001806 bool fast_promotion, can_migrate;
Joe Thornberfb4100a2015-05-20 10:30:32 +01001807 struct old_oblock_lock ool;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001808
Joe Thornber40775252015-05-15 15:29:58 +01001809 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
1810 can_migrate = !passthrough && (fast_promotion || spare_migration_bandwidth(cache));
Joe Thornber43c32bf2014-11-25 13:14:57 +00001811
Joe Thornberfb4100a2015-05-20 10:30:32 +01001812 ool.locker.fn = cell_locker;
1813 ool.cache = cache;
1814 ool.structs = structs;
1815 ool.cell = NULL;
Joe Thornber40775252015-05-15 15:29:58 +01001816 r = policy_map(cache->policy, block, true, can_migrate, fast_promotion,
Joe Thornberfb4100a2015-05-20 10:30:32 +01001817 bio, &ool.locker, &lookup_result);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001818
1819 if (r == -EWOULDBLOCK)
1820 /* migration has been denied */
1821 lookup_result.op = POLICY_MISS;
1822
1823 switch (lookup_result.op) {
1824 case POLICY_HIT:
Joe Thornber2ee57d52013-10-24 14:10:29 -04001825 if (passthrough) {
1826 inc_miss_counter(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001827
Joe Thornber2ee57d52013-10-24 14:10:29 -04001828 /*
1829 * Passthrough always maps to the origin,
1830 * invalidating any cache blocks that are written
1831 * to.
1832 */
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001833
Joe Thornber2ee57d52013-10-24 14:10:29 -04001834 if (bio_data_dir(bio) == WRITE) {
1835 atomic_inc(&cache->stats.demotion);
1836 invalidate(cache, structs, block, lookup_result.cblock, new_ocell);
1837 release_cell = false;
1838
1839 } else {
1840 /* FIXME: factor out issue_origin() */
Joe Thornber2ee57d52013-10-24 14:10:29 -04001841 remap_to_origin_clear_discard(cache, bio, block);
Joe Thornber8c081b52014-05-13 16:18:38 +01001842 inc_and_issue(cache, bio, new_ocell);
Joe Thornber2ee57d52013-10-24 14:10:29 -04001843 }
1844 } else {
1845 inc_hit_counter(cache, bio);
1846
1847 if (bio_data_dir(bio) == WRITE &&
1848 writethrough_mode(&cache->features) &&
1849 !is_dirty(cache, lookup_result.cblock)) {
Joe Thornber2ee57d52013-10-24 14:10:29 -04001850 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
Joe Thornber8c081b52014-05-13 16:18:38 +01001851 inc_and_issue(cache, bio, new_ocell);
1852
Joe Thornber651f5fa2015-05-15 15:26:08 +01001853 } else {
1854 remap_cell_to_cache_dirty(cache, new_ocell, block, lookup_result.cblock, true);
1855 release_cell = false;
Joe Thornber8c081b52014-05-13 16:18:38 +01001856 }
Joe Thornber2ee57d52013-10-24 14:10:29 -04001857 }
1858
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001859 break;
1860
1861 case POLICY_MISS:
1862 inc_miss_counter(cache, bio);
Joe Thornber651f5fa2015-05-15 15:26:08 +01001863 remap_cell_to_origin_clear_discard(cache, new_ocell, block, true);
1864 release_cell = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001865 break;
1866
1867 case POLICY_NEW:
1868 atomic_inc(&cache->stats.promotion);
1869 promote(cache, structs, block, lookup_result.cblock, new_ocell);
1870 release_cell = false;
1871 break;
1872
1873 case POLICY_REPLACE:
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001874 atomic_inc(&cache->stats.demotion);
1875 atomic_inc(&cache->stats.promotion);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001876 demote_then_promote(cache, structs, lookup_result.old_oblock,
1877 block, lookup_result.cblock,
Joe Thornberfb4100a2015-05-20 10:30:32 +01001878 ool.cell, new_ocell);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001879 release_cell = false;
1880 break;
1881
1882 default:
Mike Snitzerb61d9502015-04-22 17:25:56 -04001883 DMERR_LIMIT("%s: %s: erroring bio, unknown policy op: %u",
1884 cache_device_name(cache), __func__,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001885 (unsigned) lookup_result.op);
1886 bio_io_error(bio);
1887 }
1888
1889 if (release_cell)
1890 cell_defer(cache, new_ocell, false);
1891}
1892
Joe Thornber651f5fa2015-05-15 15:26:08 +01001893static void process_bio(struct cache *cache, struct prealloc *structs,
1894 struct bio *bio)
1895{
1896 int r;
1897 dm_oblock_t block = get_bio_block(cache, bio);
1898 struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
1899
1900 /*
1901 * Check to see if that block is currently migrating.
1902 */
1903 cell_prealloc = prealloc_get_cell(structs);
1904 r = bio_detain(cache, block, bio, cell_prealloc,
1905 (cell_free_fn) prealloc_put_cell,
1906 structs, &new_ocell);
1907 if (r > 0)
1908 return;
1909
1910 process_cell(cache, structs, new_ocell);
1911}
1912
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001913static int need_commit_due_to_time(struct cache *cache)
1914{
Joe Thornber651f5fa2015-05-15 15:26:08 +01001915 return jiffies < cache->last_commit_jiffies ||
1916 jiffies > cache->last_commit_jiffies + COMMIT_PERIOD;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001917}
1918
Joe Thornber028ae9f2015-04-22 16:42:35 -04001919/*
1920 * A non-zero return indicates read_only or fail_io mode.
1921 */
1922static int commit(struct cache *cache, bool clean_shutdown)
1923{
1924 int r;
1925
1926 if (get_cache_mode(cache) >= CM_READ_ONLY)
1927 return -EINVAL;
1928
1929 atomic_inc(&cache->stats.commit_count);
1930 r = dm_cache_commit(cache->cmd, clean_shutdown);
1931 if (r)
1932 metadata_operation_failed(cache, "dm_cache_commit", r);
1933
1934 return r;
1935}
1936
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001937static int commit_if_needed(struct cache *cache)
1938{
Heinz Mauelshagenffcbcb62013-10-14 17:24:43 +02001939 int r = 0;
1940
1941 if ((cache->commit_requested || need_commit_due_to_time(cache)) &&
1942 dm_cache_changed_this_transaction(cache->cmd)) {
Joe Thornber028ae9f2015-04-22 16:42:35 -04001943 r = commit(cache, false);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001944 cache->commit_requested = false;
Heinz Mauelshagenffcbcb62013-10-14 17:24:43 +02001945 cache->last_commit_jiffies = jiffies;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001946 }
1947
Heinz Mauelshagenffcbcb62013-10-14 17:24:43 +02001948 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001949}
1950
1951static void process_deferred_bios(struct cache *cache)
1952{
Mike Snitzer665022d2015-07-16 21:48:55 -04001953 bool prealloc_used = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001954 unsigned long flags;
1955 struct bio_list bios;
1956 struct bio *bio;
1957 struct prealloc structs;
1958
1959 memset(&structs, 0, sizeof(structs));
1960 bio_list_init(&bios);
1961
1962 spin_lock_irqsave(&cache->lock, flags);
1963 bio_list_merge(&bios, &cache->deferred_bios);
1964 bio_list_init(&cache->deferred_bios);
1965 spin_unlock_irqrestore(&cache->lock, flags);
1966
1967 while (!bio_list_empty(&bios)) {
1968 /*
1969 * If we've got no free migration structs, and processing
1970 * this bio might require one, we pause until there are some
1971 * prepared mappings to process.
1972 */
Mike Snitzer795e6332015-07-29 13:48:23 -04001973 prealloc_used = true;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001974 if (prealloc_data_structs(cache, &structs)) {
1975 spin_lock_irqsave(&cache->lock, flags);
1976 bio_list_merge(&cache->deferred_bios, &bios);
1977 spin_unlock_irqrestore(&cache->lock, flags);
1978 break;
1979 }
1980
1981 bio = bio_list_pop(&bios);
1982
Jens Axboe1eff9d32016-08-05 15:35:16 -06001983 if (bio->bi_opf & REQ_PREFLUSH)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001984 process_flush_bio(cache, bio);
Mike Christiee6047142016-06-05 14:32:04 -05001985 else if (bio_op(bio) == REQ_OP_DISCARD)
Joe Thornber7ae34e72014-11-06 10:18:04 +00001986 process_discard_bio(cache, &structs, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001987 else
1988 process_bio(cache, &structs, bio);
1989 }
1990
Mike Snitzer665022d2015-07-16 21:48:55 -04001991 if (prealloc_used)
1992 prealloc_free_structs(cache, &structs);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00001993}
1994
Joe Thornber651f5fa2015-05-15 15:26:08 +01001995static void process_deferred_cells(struct cache *cache)
1996{
Mike Snitzer665022d2015-07-16 21:48:55 -04001997 bool prealloc_used = false;
Joe Thornber651f5fa2015-05-15 15:26:08 +01001998 unsigned long flags;
1999 struct dm_bio_prison_cell *cell, *tmp;
2000 struct list_head cells;
2001 struct prealloc structs;
2002
2003 memset(&structs, 0, sizeof(structs));
2004
2005 INIT_LIST_HEAD(&cells);
2006
2007 spin_lock_irqsave(&cache->lock, flags);
2008 list_splice_init(&cache->deferred_cells, &cells);
2009 spin_unlock_irqrestore(&cache->lock, flags);
2010
2011 list_for_each_entry_safe(cell, tmp, &cells, user_list) {
2012 /*
2013 * If we've got no free migration structs, and processing
2014 * this bio might require one, we pause until there are some
2015 * prepared mappings to process.
2016 */
Mike Snitzer795e6332015-07-29 13:48:23 -04002017 prealloc_used = true;
Joe Thornber651f5fa2015-05-15 15:26:08 +01002018 if (prealloc_data_structs(cache, &structs)) {
2019 spin_lock_irqsave(&cache->lock, flags);
2020 list_splice(&cells, &cache->deferred_cells);
2021 spin_unlock_irqrestore(&cache->lock, flags);
2022 break;
2023 }
2024
2025 process_cell(cache, &structs, cell);
2026 }
2027
Mike Snitzer665022d2015-07-16 21:48:55 -04002028 if (prealloc_used)
2029 prealloc_free_structs(cache, &structs);
Joe Thornber651f5fa2015-05-15 15:26:08 +01002030}
2031
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002032static void process_deferred_flush_bios(struct cache *cache, bool submit_bios)
2033{
2034 unsigned long flags;
2035 struct bio_list bios;
2036 struct bio *bio;
2037
2038 bio_list_init(&bios);
2039
2040 spin_lock_irqsave(&cache->lock, flags);
2041 bio_list_merge(&bios, &cache->deferred_flush_bios);
2042 bio_list_init(&cache->deferred_flush_bios);
2043 spin_unlock_irqrestore(&cache->lock, flags);
2044
Joe Thornber8c081b52014-05-13 16:18:38 +01002045 /*
2046 * These bios have already been through inc_ds()
2047 */
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002048 while ((bio = bio_list_pop(&bios)))
Joe Thornber066dbaa2015-05-15 15:18:01 +01002049 submit_bios ? accounted_request(cache, bio) : bio_io_error(bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002050}
2051
Joe Thornbere2e74d62013-03-20 17:21:27 +00002052static void process_deferred_writethrough_bios(struct cache *cache)
2053{
2054 unsigned long flags;
2055 struct bio_list bios;
2056 struct bio *bio;
2057
2058 bio_list_init(&bios);
2059
2060 spin_lock_irqsave(&cache->lock, flags);
2061 bio_list_merge(&bios, &cache->deferred_writethrough_bios);
2062 bio_list_init(&cache->deferred_writethrough_bios);
2063 spin_unlock_irqrestore(&cache->lock, flags);
2064
Joe Thornber8c081b52014-05-13 16:18:38 +01002065 /*
2066 * These bios have already been through inc_ds()
2067 */
Joe Thornbere2e74d62013-03-20 17:21:27 +00002068 while ((bio = bio_list_pop(&bios)))
Joe Thornber066dbaa2015-05-15 15:18:01 +01002069 accounted_request(cache, bio);
Joe Thornbere2e74d62013-03-20 17:21:27 +00002070}
2071
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002072static void writeback_some_dirty_blocks(struct cache *cache)
2073{
Mike Snitzer665022d2015-07-16 21:48:55 -04002074 bool prealloc_used = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002075 dm_oblock_t oblock;
2076 dm_cblock_t cblock;
2077 struct prealloc structs;
2078 struct dm_bio_prison_cell *old_ocell;
Joe Thornber20f68142015-05-15 15:20:09 +01002079 bool busy = !iot_idle_for(&cache->origin_tracker, HZ);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002080
2081 memset(&structs, 0, sizeof(structs));
2082
2083 while (spare_migration_bandwidth(cache)) {
Mike Snitzere782eff2015-07-16 21:26:10 -04002084 if (policy_writeback_work(cache->policy, &oblock, &cblock, busy))
2085 break; /* no work to do */
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002086
Mike Snitzer795e6332015-07-29 13:48:23 -04002087 prealloc_used = true;
Mike Snitzere782eff2015-07-16 21:26:10 -04002088 if (prealloc_data_structs(cache, &structs) ||
2089 get_cell(cache, oblock, &structs, &old_ocell)) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002090 policy_set_dirty(cache->policy, oblock);
2091 break;
2092 }
2093
2094 writeback(cache, &structs, oblock, cblock, old_ocell);
2095 }
2096
Mike Snitzer665022d2015-07-16 21:48:55 -04002097 if (prealloc_used)
2098 prealloc_free_structs(cache, &structs);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002099}
2100
2101/*----------------------------------------------------------------
Joe Thornber65790ff2013-11-08 16:39:50 +00002102 * Invalidations.
2103 * Dropping something from the cache *without* writing back.
2104 *--------------------------------------------------------------*/
2105
2106static void process_invalidation_request(struct cache *cache, struct invalidation_request *req)
2107{
2108 int r = 0;
2109 uint64_t begin = from_cblock(req->cblocks->begin);
2110 uint64_t end = from_cblock(req->cblocks->end);
2111
2112 while (begin != end) {
2113 r = policy_remove_cblock(cache->policy, to_cblock(begin));
2114 if (!r) {
2115 r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin));
Joe Thornber028ae9f2015-04-22 16:42:35 -04002116 if (r) {
2117 metadata_operation_failed(cache, "dm_cache_remove_mapping", r);
Joe Thornber65790ff2013-11-08 16:39:50 +00002118 break;
Joe Thornber028ae9f2015-04-22 16:42:35 -04002119 }
Joe Thornber65790ff2013-11-08 16:39:50 +00002120
2121 } else if (r == -ENODATA) {
2122 /* harmless, already unmapped */
2123 r = 0;
2124
2125 } else {
Mike Snitzerb61d9502015-04-22 17:25:56 -04002126 DMERR("%s: policy_remove_cblock failed", cache_device_name(cache));
Joe Thornber65790ff2013-11-08 16:39:50 +00002127 break;
2128 }
2129
2130 begin++;
2131 }
2132
2133 cache->commit_requested = true;
2134
2135 req->err = r;
2136 atomic_set(&req->complete, 1);
2137
2138 wake_up(&req->result_wait);
2139}
2140
2141static void process_invalidation_requests(struct cache *cache)
2142{
2143 struct list_head list;
2144 struct invalidation_request *req, *tmp;
2145
2146 INIT_LIST_HEAD(&list);
2147 spin_lock(&cache->invalidation_lock);
2148 list_splice_init(&cache->invalidation_requests, &list);
2149 spin_unlock(&cache->invalidation_lock);
2150
2151 list_for_each_entry_safe (req, tmp, &list, list)
2152 process_invalidation_request(cache, req);
2153}
2154
2155/*----------------------------------------------------------------
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002156 * Main worker loop
2157 *--------------------------------------------------------------*/
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002158static bool is_quiescing(struct cache *cache)
2159{
Joe Thornber238f8362013-10-30 17:29:30 +00002160 return atomic_read(&cache->quiescing);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002161}
2162
Joe Thornber66cb1912013-10-30 17:11:58 +00002163static void ack_quiescing(struct cache *cache)
2164{
2165 if (is_quiescing(cache)) {
2166 atomic_inc(&cache->quiescing_ack);
2167 wake_up(&cache->quiescing_wait);
2168 }
2169}
2170
2171static void wait_for_quiescing_ack(struct cache *cache)
2172{
2173 wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack));
2174}
2175
2176static void start_quiescing(struct cache *cache)
2177{
Joe Thornber238f8362013-10-30 17:29:30 +00002178 atomic_inc(&cache->quiescing);
Joe Thornber66cb1912013-10-30 17:11:58 +00002179 wait_for_quiescing_ack(cache);
2180}
2181
2182static void stop_quiescing(struct cache *cache)
2183{
Joe Thornber238f8362013-10-30 17:29:30 +00002184 atomic_set(&cache->quiescing, 0);
Joe Thornber66cb1912013-10-30 17:11:58 +00002185 atomic_set(&cache->quiescing_ack, 0);
2186}
2187
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002188static void wait_for_migrations(struct cache *cache)
2189{
Joe Thornbera59db672015-01-23 10:16:16 +00002190 wait_event(cache->migration_wait, !atomic_read(&cache->nr_allocated_migrations));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002191}
2192
2193static void stop_worker(struct cache *cache)
2194{
2195 cancel_delayed_work(&cache->waker);
2196 flush_workqueue(cache->wq);
2197}
2198
Joe Thornber651f5fa2015-05-15 15:26:08 +01002199static void requeue_deferred_cells(struct cache *cache)
2200{
2201 unsigned long flags;
2202 struct list_head cells;
2203 struct dm_bio_prison_cell *cell, *tmp;
2204
2205 INIT_LIST_HEAD(&cells);
2206 spin_lock_irqsave(&cache->lock, flags);
2207 list_splice_init(&cache->deferred_cells, &cells);
2208 spin_unlock_irqrestore(&cache->lock, flags);
2209
2210 list_for_each_entry_safe(cell, tmp, &cells, user_list)
2211 cell_requeue(cache, cell);
2212}
2213
2214static void requeue_deferred_bios(struct cache *cache)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002215{
2216 struct bio *bio;
2217 struct bio_list bios;
2218
2219 bio_list_init(&bios);
2220 bio_list_merge(&bios, &cache->deferred_bios);
2221 bio_list_init(&cache->deferred_bios);
2222
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002223 while ((bio = bio_list_pop(&bios))) {
2224 bio->bi_error = DM_ENDIO_REQUEUE;
2225 bio_endio(bio);
2226 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002227}
2228
2229static int more_work(struct cache *cache)
2230{
2231 if (is_quiescing(cache))
2232 return !list_empty(&cache->quiesced_migrations) ||
2233 !list_empty(&cache->completed_migrations) ||
2234 !list_empty(&cache->need_commit_migrations);
2235 else
2236 return !bio_list_empty(&cache->deferred_bios) ||
Joe Thornber651f5fa2015-05-15 15:26:08 +01002237 !list_empty(&cache->deferred_cells) ||
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002238 !bio_list_empty(&cache->deferred_flush_bios) ||
Joe Thornbere2e74d62013-03-20 17:21:27 +00002239 !bio_list_empty(&cache->deferred_writethrough_bios) ||
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002240 !list_empty(&cache->quiesced_migrations) ||
2241 !list_empty(&cache->completed_migrations) ||
Joe Thornber65790ff2013-11-08 16:39:50 +00002242 !list_empty(&cache->need_commit_migrations) ||
2243 cache->invalidate;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002244}
2245
2246static void do_worker(struct work_struct *ws)
2247{
2248 struct cache *cache = container_of(ws, struct cache, worker);
2249
2250 do {
Joe Thornber66cb1912013-10-30 17:11:58 +00002251 if (!is_quiescing(cache)) {
2252 writeback_some_dirty_blocks(cache);
2253 process_deferred_writethrough_bios(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002254 process_deferred_bios(cache);
Joe Thornber651f5fa2015-05-15 15:26:08 +01002255 process_deferred_cells(cache);
Joe Thornber65790ff2013-11-08 16:39:50 +00002256 process_invalidation_requests(cache);
Joe Thornber66cb1912013-10-30 17:11:58 +00002257 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002258
Joe Thornber7ae34e72014-11-06 10:18:04 +00002259 process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002260 process_migrations(cache, &cache->completed_migrations, complete_migration);
2261
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002262 if (commit_if_needed(cache)) {
2263 process_deferred_flush_bios(cache, false);
Joe Thornber304affa2014-06-24 15:36:58 -04002264 process_migrations(cache, &cache->need_commit_migrations, migration_failure);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002265 } else {
2266 process_deferred_flush_bios(cache, true);
2267 process_migrations(cache, &cache->need_commit_migrations,
2268 migration_success_post_commit);
2269 }
Joe Thornber66cb1912013-10-30 17:11:58 +00002270
2271 ack_quiescing(cache);
2272
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002273 } while (more_work(cache));
2274}
2275
2276/*
2277 * We want to commit periodically so that not too much
2278 * unwritten metadata builds up.
2279 */
2280static void do_waker(struct work_struct *ws)
2281{
2282 struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker);
Joe Thornberfba10102015-05-29 10:20:56 +01002283 policy_tick(cache->policy, true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002284 wake_worker(cache);
2285 queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD);
2286}
2287
2288/*----------------------------------------------------------------*/
2289
2290static int is_congested(struct dm_dev *dev, int bdi_bits)
2291{
2292 struct request_queue *q = bdev_get_queue(dev->bdev);
2293 return bdi_congested(&q->backing_dev_info, bdi_bits);
2294}
2295
2296static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2297{
2298 struct cache *cache = container_of(cb, struct cache, callbacks);
2299
2300 return is_congested(cache->origin_dev, bdi_bits) ||
2301 is_congested(cache->cache_dev, bdi_bits);
2302}
2303
2304/*----------------------------------------------------------------
2305 * Target methods
2306 *--------------------------------------------------------------*/
2307
2308/*
2309 * This function gets called on the error paths of the constructor, so we
2310 * have to cope with a partially initialised struct.
2311 */
2312static void destroy(struct cache *cache)
2313{
2314 unsigned i;
2315
Julia Lawall6f659852015-09-13 14:15:05 +02002316 mempool_destroy(cache->migration_pool);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002317
2318 if (cache->all_io_ds)
2319 dm_deferred_set_destroy(cache->all_io_ds);
2320
2321 if (cache->prison)
2322 dm_bio_prison_destroy(cache->prison);
2323
2324 if (cache->wq)
2325 destroy_workqueue(cache->wq);
2326
2327 if (cache->dirty_bitset)
2328 free_bitset(cache->dirty_bitset);
2329
2330 if (cache->discard_bitset)
2331 free_bitset(cache->discard_bitset);
2332
2333 if (cache->copier)
2334 dm_kcopyd_client_destroy(cache->copier);
2335
2336 if (cache->cmd)
2337 dm_cache_metadata_close(cache->cmd);
2338
2339 if (cache->metadata_dev)
2340 dm_put_device(cache->ti, cache->metadata_dev);
2341
2342 if (cache->origin_dev)
2343 dm_put_device(cache->ti, cache->origin_dev);
2344
2345 if (cache->cache_dev)
2346 dm_put_device(cache->ti, cache->cache_dev);
2347
2348 if (cache->policy)
2349 dm_cache_policy_destroy(cache->policy);
2350
2351 for (i = 0; i < cache->nr_ctr_args ; i++)
2352 kfree(cache->ctr_args[i]);
2353 kfree(cache->ctr_args);
2354
2355 kfree(cache);
2356}
2357
2358static void cache_dtr(struct dm_target *ti)
2359{
2360 struct cache *cache = ti->private;
2361
2362 destroy(cache);
2363}
2364
2365static sector_t get_dev_size(struct dm_dev *dev)
2366{
2367 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
2368}
2369
2370/*----------------------------------------------------------------*/
2371
2372/*
2373 * Construct a cache device mapping.
2374 *
2375 * cache <metadata dev> <cache dev> <origin dev> <block size>
2376 * <#feature args> [<feature arg>]*
2377 * <policy> <#policy args> [<policy arg>]*
2378 *
2379 * metadata dev : fast device holding the persistent metadata
2380 * cache dev : fast device holding cached data blocks
2381 * origin dev : slow device holding original data blocks
2382 * block size : cache unit size in sectors
2383 *
2384 * #feature args : number of feature arguments passed
2385 * feature args : writethrough. (The default is writeback.)
2386 *
2387 * policy : the replacement policy to use
2388 * #policy args : an even number of policy arguments corresponding
2389 * to key/value pairs passed to the policy
2390 * policy args : key/value pairs passed to the policy
2391 * E.g. 'sequential_threshold 1024'
2392 * See cache-policies.txt for details.
2393 *
2394 * Optional feature arguments are:
2395 * writethrough : write through caching that prohibits cache block
2396 * content from being different from origin block content.
2397 * Without this argument, the default behaviour is to write
2398 * back cache block contents later for performance reasons,
2399 * so they may differ from the corresponding origin blocks.
2400 */
2401struct cache_args {
2402 struct dm_target *ti;
2403
2404 struct dm_dev *metadata_dev;
2405
2406 struct dm_dev *cache_dev;
2407 sector_t cache_sectors;
2408
2409 struct dm_dev *origin_dev;
2410 sector_t origin_sectors;
2411
2412 uint32_t block_size;
2413
2414 const char *policy_name;
2415 int policy_argc;
2416 const char **policy_argv;
2417
2418 struct cache_features features;
2419};
2420
2421static void destroy_cache_args(struct cache_args *ca)
2422{
2423 if (ca->metadata_dev)
2424 dm_put_device(ca->ti, ca->metadata_dev);
2425
2426 if (ca->cache_dev)
2427 dm_put_device(ca->ti, ca->cache_dev);
2428
2429 if (ca->origin_dev)
2430 dm_put_device(ca->ti, ca->origin_dev);
2431
2432 kfree(ca);
2433}
2434
2435static bool at_least_one_arg(struct dm_arg_set *as, char **error)
2436{
2437 if (!as->argc) {
2438 *error = "Insufficient args";
2439 return false;
2440 }
2441
2442 return true;
2443}
2444
2445static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as,
2446 char **error)
2447{
2448 int r;
2449 sector_t metadata_dev_size;
2450 char b[BDEVNAME_SIZE];
2451
2452 if (!at_least_one_arg(as, error))
2453 return -EINVAL;
2454
2455 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2456 &ca->metadata_dev);
2457 if (r) {
2458 *error = "Error opening metadata device";
2459 return r;
2460 }
2461
2462 metadata_dev_size = get_dev_size(ca->metadata_dev);
2463 if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING)
2464 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2465 bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
2466
2467 return 0;
2468}
2469
2470static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as,
2471 char **error)
2472{
2473 int r;
2474
2475 if (!at_least_one_arg(as, error))
2476 return -EINVAL;
2477
2478 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2479 &ca->cache_dev);
2480 if (r) {
2481 *error = "Error opening cache device";
2482 return r;
2483 }
2484 ca->cache_sectors = get_dev_size(ca->cache_dev);
2485
2486 return 0;
2487}
2488
2489static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as,
2490 char **error)
2491{
2492 int r;
2493
2494 if (!at_least_one_arg(as, error))
2495 return -EINVAL;
2496
2497 r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE,
2498 &ca->origin_dev);
2499 if (r) {
2500 *error = "Error opening origin device";
2501 return r;
2502 }
2503
2504 ca->origin_sectors = get_dev_size(ca->origin_dev);
2505 if (ca->ti->len > ca->origin_sectors) {
2506 *error = "Device size larger than cached device";
2507 return -EINVAL;
2508 }
2509
2510 return 0;
2511}
2512
2513static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as,
2514 char **error)
2515{
Mike Snitzer05473042013-08-16 10:54:19 -04002516 unsigned long block_size;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002517
2518 if (!at_least_one_arg(as, error))
2519 return -EINVAL;
2520
Mike Snitzer05473042013-08-16 10:54:19 -04002521 if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size ||
2522 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2523 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2524 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002525 *error = "Invalid data block size";
2526 return -EINVAL;
2527 }
2528
Mike Snitzer05473042013-08-16 10:54:19 -04002529 if (block_size > ca->cache_sectors) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002530 *error = "Data block size is larger than the cache device";
2531 return -EINVAL;
2532 }
2533
Mike Snitzer05473042013-08-16 10:54:19 -04002534 ca->block_size = block_size;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002535
2536 return 0;
2537}
2538
2539static void init_features(struct cache_features *cf)
2540{
2541 cf->mode = CM_WRITE;
Joe Thornber2ee57d52013-10-24 14:10:29 -04002542 cf->io_mode = CM_IO_WRITEBACK;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002543}
2544
2545static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
2546 char **error)
2547{
2548 static struct dm_arg _args[] = {
2549 {0, 1, "Invalid number of cache feature arguments"},
2550 };
2551
2552 int r;
2553 unsigned argc;
2554 const char *arg;
2555 struct cache_features *cf = &ca->features;
2556
2557 init_features(cf);
2558
2559 r = dm_read_arg_group(_args, as, &argc, error);
2560 if (r)
2561 return -EINVAL;
2562
2563 while (argc--) {
2564 arg = dm_shift_arg(as);
2565
2566 if (!strcasecmp(arg, "writeback"))
Joe Thornber2ee57d52013-10-24 14:10:29 -04002567 cf->io_mode = CM_IO_WRITEBACK;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002568
2569 else if (!strcasecmp(arg, "writethrough"))
Joe Thornber2ee57d52013-10-24 14:10:29 -04002570 cf->io_mode = CM_IO_WRITETHROUGH;
2571
2572 else if (!strcasecmp(arg, "passthrough"))
2573 cf->io_mode = CM_IO_PASSTHROUGH;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002574
2575 else {
2576 *error = "Unrecognised cache feature requested";
2577 return -EINVAL;
2578 }
2579 }
2580
2581 return 0;
2582}
2583
2584static int parse_policy(struct cache_args *ca, struct dm_arg_set *as,
2585 char **error)
2586{
2587 static struct dm_arg _args[] = {
2588 {0, 1024, "Invalid number of policy arguments"},
2589 };
2590
2591 int r;
2592
2593 if (!at_least_one_arg(as, error))
2594 return -EINVAL;
2595
2596 ca->policy_name = dm_shift_arg(as);
2597
2598 r = dm_read_arg_group(_args, as, &ca->policy_argc, error);
2599 if (r)
2600 return -EINVAL;
2601
2602 ca->policy_argv = (const char **)as->argv;
2603 dm_consume_args(as, ca->policy_argc);
2604
2605 return 0;
2606}
2607
2608static int parse_cache_args(struct cache_args *ca, int argc, char **argv,
2609 char **error)
2610{
2611 int r;
2612 struct dm_arg_set as;
2613
2614 as.argc = argc;
2615 as.argv = argv;
2616
2617 r = parse_metadata_dev(ca, &as, error);
2618 if (r)
2619 return r;
2620
2621 r = parse_cache_dev(ca, &as, error);
2622 if (r)
2623 return r;
2624
2625 r = parse_origin_dev(ca, &as, error);
2626 if (r)
2627 return r;
2628
2629 r = parse_block_size(ca, &as, error);
2630 if (r)
2631 return r;
2632
2633 r = parse_features(ca, &as, error);
2634 if (r)
2635 return r;
2636
2637 r = parse_policy(ca, &as, error);
2638 if (r)
2639 return r;
2640
2641 return 0;
2642}
2643
2644/*----------------------------------------------------------------*/
2645
2646static struct kmem_cache *migration_cache;
2647
Alasdair G Kergon2c73c472013-05-10 14:37:21 +01002648#define NOT_CORE_OPTION 1
2649
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002650static int process_config_option(struct cache *cache, const char *key, const char *value)
Alasdair G Kergon2c73c472013-05-10 14:37:21 +01002651{
2652 unsigned long tmp;
2653
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002654 if (!strcasecmp(key, "migration_threshold")) {
2655 if (kstrtoul(value, 10, &tmp))
Alasdair G Kergon2c73c472013-05-10 14:37:21 +01002656 return -EINVAL;
2657
2658 cache->migration_threshold = tmp;
2659 return 0;
2660 }
2661
2662 return NOT_CORE_OPTION;
2663}
2664
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002665static int set_config_value(struct cache *cache, const char *key, const char *value)
2666{
2667 int r = process_config_option(cache, key, value);
2668
2669 if (r == NOT_CORE_OPTION)
2670 r = policy_set_config_value(cache->policy, key, value);
2671
2672 if (r)
2673 DMWARN("bad config value for %s: %s", key, value);
2674
2675 return r;
2676}
2677
2678static int set_config_values(struct cache *cache, int argc, const char **argv)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002679{
2680 int r = 0;
2681
2682 if (argc & 1) {
2683 DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs.");
2684 return -EINVAL;
2685 }
2686
2687 while (argc) {
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002688 r = set_config_value(cache, argv[0], argv[1]);
2689 if (r)
2690 break;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002691
2692 argc -= 2;
2693 argv += 2;
2694 }
2695
2696 return r;
2697}
2698
2699static int create_cache_policy(struct cache *cache, struct cache_args *ca,
2700 char **error)
2701{
Mikulas Patocka4cb3e1d2013-10-01 18:35:39 -04002702 struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name,
2703 cache->cache_size,
2704 cache->origin_sectors,
2705 cache->sectors_per_block);
2706 if (IS_ERR(p)) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002707 *error = "Error creating cache's policy";
Mikulas Patocka4cb3e1d2013-10-01 18:35:39 -04002708 return PTR_ERR(p);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002709 }
Mikulas Patocka4cb3e1d2013-10-01 18:35:39 -04002710 cache->policy = p;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002711
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002712 return 0;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002713}
2714
Joe Thornber08b18452014-11-06 14:38:01 +00002715/*
Joe Thornber2bb812d2014-11-26 16:07:50 +00002716 * We want the discard block size to be at least the size of the cache
2717 * block size and have no more than 2^14 discard blocks across the origin.
Joe Thornber08b18452014-11-06 14:38:01 +00002718 */
2719#define MAX_DISCARD_BLOCKS (1 << 14)
2720
2721static bool too_many_discard_blocks(sector_t discard_block_size,
2722 sector_t origin_size)
2723{
2724 (void) sector_div(origin_size, discard_block_size);
2725
2726 return origin_size > MAX_DISCARD_BLOCKS;
2727}
2728
2729static sector_t calculate_discard_block_size(sector_t cache_block_size,
2730 sector_t origin_size)
2731{
Joe Thornber2bb812d2014-11-26 16:07:50 +00002732 sector_t discard_block_size = cache_block_size;
Joe Thornber08b18452014-11-06 14:38:01 +00002733
2734 if (origin_size)
2735 while (too_many_discard_blocks(discard_block_size, origin_size))
2736 discard_block_size *= 2;
2737
2738 return discard_block_size;
2739}
2740
Joe Thornberd1d92202014-11-11 11:58:32 +00002741static void set_cache_size(struct cache *cache, dm_cblock_t size)
2742{
2743 dm_block_t nr_blocks = from_cblock(size);
2744
2745 if (nr_blocks > (1 << 20) && cache->cache_size != size)
2746 DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n"
2747 "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n"
2748 "Please consider increasing the cache block size to reduce the overall cache block count.",
2749 (unsigned long long) nr_blocks);
2750
2751 cache->cache_size = size;
2752}
2753
Joe Thornberf8350da2013-05-10 14:37:16 +01002754#define DEFAULT_MIGRATION_THRESHOLD 2048
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002755
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002756static int cache_create(struct cache_args *ca, struct cache **result)
2757{
2758 int r = 0;
2759 char **error = &ca->ti->error;
2760 struct cache *cache;
2761 struct dm_target *ti = ca->ti;
2762 dm_block_t origin_blocks;
2763 struct dm_cache_metadata *cmd;
2764 bool may_format = ca->features.mode == CM_WRITE;
2765
2766 cache = kzalloc(sizeof(*cache), GFP_KERNEL);
2767 if (!cache)
2768 return -ENOMEM;
2769
2770 cache->ti = ca->ti;
2771 ti->private = cache;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002772 ti->num_flush_bios = 2;
2773 ti->flush_supported = true;
2774
2775 ti->num_discard_bios = 1;
2776 ti->discards_supported = true;
2777 ti->discard_zeroes_data_unsupported = true;
Joe Thornber25726292014-11-24 14:05:16 +00002778 ti->split_discard_bios = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002779
Joe Thornber8c5008f2013-05-10 14:37:18 +01002780 cache->features = ca->features;
Mike Snitzer30187e12016-01-31 13:28:26 -05002781 ti->per_io_data_size = get_per_bio_data_size(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002782
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002783 cache->callbacks.congested_fn = cache_is_congested;
2784 dm_table_add_target_callbacks(ti->table, &cache->callbacks);
2785
2786 cache->metadata_dev = ca->metadata_dev;
2787 cache->origin_dev = ca->origin_dev;
2788 cache->cache_dev = ca->cache_dev;
2789
2790 ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL;
2791
2792 /* FIXME: factor out this whole section */
2793 origin_blocks = cache->origin_sectors = ca->origin_sectors;
Joe Thornber414dd672013-03-20 17:21:25 +00002794 origin_blocks = block_div(origin_blocks, ca->block_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002795 cache->origin_blocks = to_oblock(origin_blocks);
2796
2797 cache->sectors_per_block = ca->block_size;
2798 if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) {
2799 r = -EINVAL;
2800 goto bad;
2801 }
2802
2803 if (ca->block_size & (ca->block_size - 1)) {
2804 dm_block_t cache_size = ca->cache_sectors;
2805
2806 cache->sectors_per_block_shift = -1;
Joe Thornber414dd672013-03-20 17:21:25 +00002807 cache_size = block_div(cache_size, ca->block_size);
Joe Thornberd1d92202014-11-11 11:58:32 +00002808 set_cache_size(cache, to_cblock(cache_size));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002809 } else {
2810 cache->sectors_per_block_shift = __ffs(ca->block_size);
Joe Thornberd1d92202014-11-11 11:58:32 +00002811 set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002812 }
2813
2814 r = create_cache_policy(cache, ca, error);
2815 if (r)
2816 goto bad;
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002817
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002818 cache->policy_nr_args = ca->policy_argc;
Joe Thornber2f14f4b2013-05-10 14:37:21 +01002819 cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD;
2820
2821 r = set_config_values(cache, ca->policy_argc, ca->policy_argv);
2822 if (r) {
2823 *error = "Error setting cache policy's config values";
2824 goto bad;
2825 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002826
2827 cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
2828 ca->block_size, may_format,
2829 dm_cache_policy_get_hint_size(cache->policy));
2830 if (IS_ERR(cmd)) {
2831 *error = "Error creating metadata object";
2832 r = PTR_ERR(cmd);
2833 goto bad;
2834 }
2835 cache->cmd = cmd;
Joe Thornber028ae9f2015-04-22 16:42:35 -04002836 set_cache_mode(cache, CM_WRITE);
2837 if (get_cache_mode(cache) != CM_WRITE) {
2838 *error = "Unable to get write access to metadata, please check/repair metadata.";
2839 r = -EINVAL;
2840 goto bad;
2841 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002842
Joe Thornber2ee57d52013-10-24 14:10:29 -04002843 if (passthrough_mode(&cache->features)) {
2844 bool all_clean;
2845
2846 r = dm_cache_metadata_all_clean(cache->cmd, &all_clean);
2847 if (r) {
2848 *error = "dm_cache_metadata_all_clean() failed";
2849 goto bad;
2850 }
2851
2852 if (!all_clean) {
2853 *error = "Cannot enter passthrough mode unless all blocks are clean";
2854 r = -EINVAL;
2855 goto bad;
2856 }
2857 }
2858
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002859 spin_lock_init(&cache->lock);
Joe Thornber651f5fa2015-05-15 15:26:08 +01002860 INIT_LIST_HEAD(&cache->deferred_cells);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002861 bio_list_init(&cache->deferred_bios);
2862 bio_list_init(&cache->deferred_flush_bios);
Joe Thornbere2e74d62013-03-20 17:21:27 +00002863 bio_list_init(&cache->deferred_writethrough_bios);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002864 INIT_LIST_HEAD(&cache->quiesced_migrations);
2865 INIT_LIST_HEAD(&cache->completed_migrations);
2866 INIT_LIST_HEAD(&cache->need_commit_migrations);
Joe Thornbera59db672015-01-23 10:16:16 +00002867 atomic_set(&cache->nr_allocated_migrations, 0);
2868 atomic_set(&cache->nr_io_migrations, 0);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002869 init_waitqueue_head(&cache->migration_wait);
2870
Joe Thornber66cb1912013-10-30 17:11:58 +00002871 init_waitqueue_head(&cache->quiescing_wait);
Joe Thornber238f8362013-10-30 17:29:30 +00002872 atomic_set(&cache->quiescing, 0);
Joe Thornber66cb1912013-10-30 17:11:58 +00002873 atomic_set(&cache->quiescing_ack, 0);
2874
Wei Yongjunfa4d6832013-05-10 14:37:14 +01002875 r = -ENOMEM;
Anssi Hannula44fa8162014-08-01 11:55:47 -04002876 atomic_set(&cache->nr_dirty, 0);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002877 cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size));
2878 if (!cache->dirty_bitset) {
2879 *error = "could not allocate dirty bitset";
2880 goto bad;
2881 }
2882 clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size));
2883
Joe Thornber08b18452014-11-06 14:38:01 +00002884 cache->discard_block_size =
2885 calculate_discard_block_size(cache->sectors_per_block,
2886 cache->origin_sectors);
Joe Thornber25726292014-11-24 14:05:16 +00002887 cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors,
2888 cache->discard_block_size));
Joe Thornber1bad9bc2014-11-07 14:47:07 +00002889 cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002890 if (!cache->discard_bitset) {
2891 *error = "could not allocate discard bitset";
2892 goto bad;
2893 }
Joe Thornber1bad9bc2014-11-07 14:47:07 +00002894 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002895
2896 cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2897 if (IS_ERR(cache->copier)) {
2898 *error = "could not create kcopyd client";
2899 r = PTR_ERR(cache->copier);
2900 goto bad;
2901 }
2902
2903 cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2904 if (!cache->wq) {
2905 *error = "could not create workqueue for metadata object";
2906 goto bad;
2907 }
2908 INIT_WORK(&cache->worker, do_worker);
2909 INIT_DELAYED_WORK(&cache->waker, do_waker);
2910 cache->last_commit_jiffies = jiffies;
2911
Joe Thornbera195db22014-10-06 16:30:06 -04002912 cache->prison = dm_bio_prison_create();
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002913 if (!cache->prison) {
2914 *error = "could not create bio prison";
2915 goto bad;
2916 }
2917
2918 cache->all_io_ds = dm_deferred_set_create();
2919 if (!cache->all_io_ds) {
2920 *error = "could not create all_io deferred set";
2921 goto bad;
2922 }
2923
2924 cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE,
2925 migration_cache);
2926 if (!cache->migration_pool) {
2927 *error = "Error creating cache's migration mempool";
2928 goto bad;
2929 }
2930
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002931 cache->need_tick_bio = true;
2932 cache->sized = false;
Joe Thornber65790ff2013-11-08 16:39:50 +00002933 cache->invalidate = false;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002934 cache->commit_requested = false;
2935 cache->loaded_mappings = false;
2936 cache->loaded_discards = false;
2937
2938 load_stats(cache);
2939
2940 atomic_set(&cache->stats.demotion, 0);
2941 atomic_set(&cache->stats.promotion, 0);
2942 atomic_set(&cache->stats.copies_avoided, 0);
2943 atomic_set(&cache->stats.cache_cell_clash, 0);
2944 atomic_set(&cache->stats.commit_count, 0);
2945 atomic_set(&cache->stats.discard_count, 0);
2946
Joe Thornber65790ff2013-11-08 16:39:50 +00002947 spin_lock_init(&cache->invalidation_lock);
2948 INIT_LIST_HEAD(&cache->invalidation_requests);
2949
Joe Thornber066dbaa2015-05-15 15:18:01 +01002950 iot_init(&cache->origin_tracker);
2951
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00002952 *result = cache;
2953 return 0;
2954
2955bad:
2956 destroy(cache);
2957 return r;
2958}
2959
2960static int copy_ctr_args(struct cache *cache, int argc, const char **argv)
2961{
2962 unsigned i;
2963 const char **copy;
2964
2965 copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL);
2966 if (!copy)
2967 return -ENOMEM;
2968 for (i = 0; i < argc; i++) {
2969 copy[i] = kstrdup(argv[i], GFP_KERNEL);
2970 if (!copy[i]) {
2971 while (i--)
2972 kfree(copy[i]);
2973 kfree(copy);
2974 return -ENOMEM;
2975 }
2976 }
2977
2978 cache->nr_ctr_args = argc;
2979 cache->ctr_args = copy;
2980
2981 return 0;
2982}
2983
2984static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv)
2985{
2986 int r = -EINVAL;
2987 struct cache_args *ca;
2988 struct cache *cache = NULL;
2989
2990 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2991 if (!ca) {
2992 ti->error = "Error allocating memory for cache";
2993 return -ENOMEM;
2994 }
2995 ca->ti = ti;
2996
2997 r = parse_cache_args(ca, argc, argv, &ti->error);
2998 if (r)
2999 goto out;
3000
3001 r = cache_create(ca, &cache);
Heinz Mauelshagen617a0b82013-03-20 17:21:26 +00003002 if (r)
3003 goto out;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003004
3005 r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3);
3006 if (r) {
3007 destroy(cache);
3008 goto out;
3009 }
3010
3011 ti->private = cache;
3012
3013out:
3014 destroy_cache_args(ca);
3015 return r;
3016}
3017
Joe Thornber651f5fa2015-05-15 15:26:08 +01003018/*----------------------------------------------------------------*/
3019
3020static int cache_map(struct dm_target *ti, struct bio *bio)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003021{
Joe Thornber651f5fa2015-05-15 15:26:08 +01003022 struct cache *cache = ti->private;
3023
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003024 int r;
Joe Thornber651f5fa2015-05-15 15:26:08 +01003025 struct dm_bio_prison_cell *cell = NULL;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003026 dm_oblock_t block = get_bio_block(cache, bio);
Mike Snitzer19b00922013-04-05 15:36:34 +01003027 size_t pb_data_size = get_per_bio_data_size(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003028 bool can_migrate = false;
Joe Thornber40775252015-05-15 15:29:58 +01003029 bool fast_promotion;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003030 struct policy_result lookup_result;
Heinz Mauelshagene893fba2014-03-12 16:13:39 +01003031 struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size);
Joe Thornberfb4100a2015-05-20 10:30:32 +01003032 struct old_oblock_lock ool;
3033
3034 ool.locker.fn = null_locker;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003035
Heinz Mauelshagene893fba2014-03-12 16:13:39 +01003036 if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003037 /*
3038 * This can only occur if the io goes to a partial block at
3039 * the end of the origin device. We don't cache these.
3040 * Just remap to the origin and carry on.
3041 */
Heinz Mauelshagene893fba2014-03-12 16:13:39 +01003042 remap_to_origin(cache, bio);
Joe Thornber651f5fa2015-05-15 15:26:08 +01003043 accounted_begin(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003044 return DM_MAPIO_REMAPPED;
3045 }
3046
Joe Thornber651f5fa2015-05-15 15:26:08 +01003047 if (discard_or_flush(bio)) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003048 defer_bio(cache, bio);
3049 return DM_MAPIO_SUBMITTED;
3050 }
3051
3052 /*
3053 * Check to see if that block is currently migrating.
3054 */
Joe Thornber651f5fa2015-05-15 15:26:08 +01003055 cell = alloc_prison_cell(cache);
3056 if (!cell) {
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003057 defer_bio(cache, bio);
3058 return DM_MAPIO_SUBMITTED;
3059 }
3060
Joe Thornber651f5fa2015-05-15 15:26:08 +01003061 r = bio_detain(cache, block, bio, cell,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003062 (cell_free_fn) free_prison_cell,
Joe Thornber651f5fa2015-05-15 15:26:08 +01003063 cache, &cell);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003064 if (r) {
3065 if (r < 0)
3066 defer_bio(cache, bio);
3067
3068 return DM_MAPIO_SUBMITTED;
3069 }
3070
Joe Thornber40775252015-05-15 15:29:58 +01003071 fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003072
Joe Thornber40775252015-05-15 15:29:58 +01003073 r = policy_map(cache->policy, block, false, can_migrate, fast_promotion,
Joe Thornberfb4100a2015-05-20 10:30:32 +01003074 bio, &ool.locker, &lookup_result);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003075 if (r == -EWOULDBLOCK) {
Joe Thornber651f5fa2015-05-15 15:26:08 +01003076 cell_defer(cache, cell, true);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003077 return DM_MAPIO_SUBMITTED;
3078
3079 } else if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003080 DMERR_LIMIT("%s: Unexpected return from cache replacement policy: %d",
3081 cache_device_name(cache), r);
Joe Thornber651f5fa2015-05-15 15:26:08 +01003082 cell_defer(cache, cell, false);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003083 bio_io_error(bio);
3084 return DM_MAPIO_SUBMITTED;
3085 }
3086
Joe Thornber2ee57d52013-10-24 14:10:29 -04003087 r = DM_MAPIO_REMAPPED;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003088 switch (lookup_result.op) {
3089 case POLICY_HIT:
Joe Thornber2ee57d52013-10-24 14:10:29 -04003090 if (passthrough_mode(&cache->features)) {
3091 if (bio_data_dir(bio) == WRITE) {
3092 /*
3093 * We need to invalidate this block, so
3094 * defer for the worker thread.
3095 */
Joe Thornber651f5fa2015-05-15 15:26:08 +01003096 cell_defer(cache, cell, true);
Joe Thornber2ee57d52013-10-24 14:10:29 -04003097 r = DM_MAPIO_SUBMITTED;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003098
Joe Thornber2ee57d52013-10-24 14:10:29 -04003099 } else {
Joe Thornber2ee57d52013-10-24 14:10:29 -04003100 inc_miss_counter(cache, bio);
3101 remap_to_origin_clear_discard(cache, bio, block);
Joe Thornber651f5fa2015-05-15 15:26:08 +01003102 accounted_begin(cache, bio);
3103 inc_ds(cache, bio, cell);
3104 // FIXME: we want to remap hits or misses straight
3105 // away rather than passing over to the worker.
3106 cell_defer(cache, cell, false);
Joe Thornber2ee57d52013-10-24 14:10:29 -04003107 }
3108
3109 } else {
3110 inc_hit_counter(cache, bio);
Joe Thornber2ee57d52013-10-24 14:10:29 -04003111 if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) &&
Joe Thornber651f5fa2015-05-15 15:26:08 +01003112 !is_dirty(cache, lookup_result.cblock)) {
Joe Thornber2ee57d52013-10-24 14:10:29 -04003113 remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock);
Joe Thornber651f5fa2015-05-15 15:26:08 +01003114 accounted_begin(cache, bio);
3115 inc_ds(cache, bio, cell);
3116 cell_defer(cache, cell, false);
3117
3118 } else
3119 remap_cell_to_cache_dirty(cache, cell, block, lookup_result.cblock, false);
Joe Thornber2ee57d52013-10-24 14:10:29 -04003120 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003121 break;
3122
3123 case POLICY_MISS:
3124 inc_miss_counter(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003125 if (pb->req_nr != 0) {
3126 /*
3127 * This is a duplicate writethrough io that is no
3128 * longer needed because the block has been demoted.
3129 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003130 bio_endio(bio);
Joe Thornber651f5fa2015-05-15 15:26:08 +01003131 // FIXME: remap everything as a miss
3132 cell_defer(cache, cell, false);
Joe Thornber8c081b52014-05-13 16:18:38 +01003133 r = DM_MAPIO_SUBMITTED;
3134
3135 } else
Joe Thornber651f5fa2015-05-15 15:26:08 +01003136 remap_cell_to_origin_clear_discard(cache, cell, block, false);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003137 break;
3138
3139 default:
Mike Snitzerb61d9502015-04-22 17:25:56 -04003140 DMERR_LIMIT("%s: %s: erroring bio: unknown policy op: %u",
3141 cache_device_name(cache), __func__,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003142 (unsigned) lookup_result.op);
Joe Thornber651f5fa2015-05-15 15:26:08 +01003143 cell_defer(cache, cell, false);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003144 bio_io_error(bio);
Joe Thornber2ee57d52013-10-24 14:10:29 -04003145 r = DM_MAPIO_SUBMITTED;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003146 }
3147
Joe Thornber2ee57d52013-10-24 14:10:29 -04003148 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003149}
3150
3151static int cache_end_io(struct dm_target *ti, struct bio *bio, int error)
3152{
3153 struct cache *cache = ti->private;
3154 unsigned long flags;
Mike Snitzer19b00922013-04-05 15:36:34 +01003155 size_t pb_data_size = get_per_bio_data_size(cache);
3156 struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003157
3158 if (pb->tick) {
Joe Thornberfba10102015-05-29 10:20:56 +01003159 policy_tick(cache->policy, false);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003160
3161 spin_lock_irqsave(&cache->lock, flags);
3162 cache->need_tick_bio = true;
3163 spin_unlock_irqrestore(&cache->lock, flags);
3164 }
3165
3166 check_for_quiesced_migrations(cache, pb);
Joe Thornber066dbaa2015-05-15 15:18:01 +01003167 accounted_complete(cache, bio);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003168
3169 return 0;
3170}
3171
3172static int write_dirty_bitset(struct cache *cache)
3173{
3174 unsigned i, r;
3175
Joe Thornber028ae9f2015-04-22 16:42:35 -04003176 if (get_cache_mode(cache) >= CM_READ_ONLY)
3177 return -EINVAL;
3178
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003179 for (i = 0; i < from_cblock(cache->cache_size); i++) {
3180 r = dm_cache_set_dirty(cache->cmd, to_cblock(i),
3181 is_dirty(cache, to_cblock(i)));
Joe Thornber028ae9f2015-04-22 16:42:35 -04003182 if (r) {
3183 metadata_operation_failed(cache, "dm_cache_set_dirty", r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003184 return r;
Joe Thornber028ae9f2015-04-22 16:42:35 -04003185 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003186 }
3187
3188 return 0;
3189}
3190
3191static int write_discard_bitset(struct cache *cache)
3192{
3193 unsigned i, r;
3194
Joe Thornber028ae9f2015-04-22 16:42:35 -04003195 if (get_cache_mode(cache) >= CM_READ_ONLY)
3196 return -EINVAL;
3197
Joe Thornber1bad9bc2014-11-07 14:47:07 +00003198 r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size,
3199 cache->discard_nr_blocks);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003200 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003201 DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -04003202 metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003203 return r;
3204 }
3205
Joe Thornber1bad9bc2014-11-07 14:47:07 +00003206 for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) {
3207 r = dm_cache_set_discard(cache->cmd, to_dblock(i),
3208 is_discarded(cache, to_dblock(i)));
Joe Thornber028ae9f2015-04-22 16:42:35 -04003209 if (r) {
3210 metadata_operation_failed(cache, "dm_cache_set_discard", r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003211 return r;
Joe Thornber028ae9f2015-04-22 16:42:35 -04003212 }
3213 }
3214
3215 return 0;
3216}
3217
3218static int write_hints(struct cache *cache)
3219{
3220 int r;
3221
3222 if (get_cache_mode(cache) >= CM_READ_ONLY)
3223 return -EINVAL;
3224
3225 r = dm_cache_write_hints(cache->cmd, cache->policy);
3226 if (r) {
3227 metadata_operation_failed(cache, "dm_cache_write_hints", r);
3228 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003229 }
3230
3231 return 0;
3232}
3233
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003234/*
3235 * returns true on success
3236 */
3237static bool sync_metadata(struct cache *cache)
3238{
3239 int r1, r2, r3, r4;
3240
3241 r1 = write_dirty_bitset(cache);
3242 if (r1)
Mike Snitzerb61d9502015-04-22 17:25:56 -04003243 DMERR("%s: could not write dirty bitset", cache_device_name(cache));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003244
3245 r2 = write_discard_bitset(cache);
3246 if (r2)
Mike Snitzerb61d9502015-04-22 17:25:56 -04003247 DMERR("%s: could not write discard bitset", cache_device_name(cache));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003248
3249 save_stats(cache);
3250
Joe Thornber028ae9f2015-04-22 16:42:35 -04003251 r3 = write_hints(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003252 if (r3)
Mike Snitzerb61d9502015-04-22 17:25:56 -04003253 DMERR("%s: could not write hints", cache_device_name(cache));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003254
3255 /*
3256 * If writing the above metadata failed, we still commit, but don't
3257 * set the clean shutdown flag. This will effectively force every
3258 * dirty bit to be set on reload.
3259 */
Joe Thornber028ae9f2015-04-22 16:42:35 -04003260 r4 = commit(cache, !r1 && !r2 && !r3);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003261 if (r4)
Mike Snitzerb61d9502015-04-22 17:25:56 -04003262 DMERR("%s: could not write cache metadata", cache_device_name(cache));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003263
3264 return !r1 && !r2 && !r3 && !r4;
3265}
3266
3267static void cache_postsuspend(struct dm_target *ti)
3268{
3269 struct cache *cache = ti->private;
3270
3271 start_quiescing(cache);
3272 wait_for_migrations(cache);
3273 stop_worker(cache);
Joe Thornber651f5fa2015-05-15 15:26:08 +01003274 requeue_deferred_bios(cache);
3275 requeue_deferred_cells(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003276 stop_quiescing(cache);
3277
Joe Thornber028ae9f2015-04-22 16:42:35 -04003278 if (get_cache_mode(cache) == CM_WRITE)
3279 (void) sync_metadata(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003280}
3281
3282static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock,
3283 bool dirty, uint32_t hint, bool hint_valid)
3284{
3285 int r;
3286 struct cache *cache = context;
3287
3288 r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid);
3289 if (r)
3290 return r;
3291
3292 if (dirty)
3293 set_dirty(cache, oblock, cblock);
3294 else
3295 clear_dirty(cache, oblock, cblock);
3296
3297 return 0;
3298}
3299
Joe Thornber3e2e1c32014-11-24 14:06:22 +00003300/*
3301 * The discard block size in the on disk metadata is not
3302 * neccessarily the same as we're currently using. So we have to
3303 * be careful to only set the discarded attribute if we know it
3304 * covers a complete block of the new size.
3305 */
3306struct discard_load_info {
3307 struct cache *cache;
3308
3309 /*
3310 * These blocks are sized using the on disk dblock size, rather
3311 * than the current one.
3312 */
3313 dm_block_t block_size;
3314 dm_block_t discard_begin, discard_end;
3315};
3316
3317static void discard_load_info_init(struct cache *cache,
3318 struct discard_load_info *li)
3319{
3320 li->cache = cache;
3321 li->discard_begin = li->discard_end = 0;
3322}
3323
3324static void set_discard_range(struct discard_load_info *li)
3325{
3326 sector_t b, e;
3327
3328 if (li->discard_begin == li->discard_end)
3329 return;
3330
3331 /*
3332 * Convert to sectors.
3333 */
3334 b = li->discard_begin * li->block_size;
3335 e = li->discard_end * li->block_size;
3336
3337 /*
3338 * Then convert back to the current dblock size.
3339 */
3340 b = dm_sector_div_up(b, li->cache->discard_block_size);
3341 sector_div(e, li->cache->discard_block_size);
3342
3343 /*
3344 * The origin may have shrunk, so we need to check we're still in
3345 * bounds.
3346 */
3347 if (e > from_dblock(li->cache->discard_nr_blocks))
3348 e = from_dblock(li->cache->discard_nr_blocks);
3349
3350 for (; b < e; b++)
3351 set_discard(li->cache, to_dblock(b));
3352}
3353
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003354static int load_discard(void *context, sector_t discard_block_size,
Joe Thornber1bad9bc2014-11-07 14:47:07 +00003355 dm_dblock_t dblock, bool discard)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003356{
Joe Thornber3e2e1c32014-11-24 14:06:22 +00003357 struct discard_load_info *li = context;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003358
Joe Thornber3e2e1c32014-11-24 14:06:22 +00003359 li->block_size = discard_block_size;
Joe Thornber1bad9bc2014-11-07 14:47:07 +00003360
Joe Thornber3e2e1c32014-11-24 14:06:22 +00003361 if (discard) {
3362 if (from_dblock(dblock) == li->discard_end)
3363 /*
3364 * We're already in a discard range, just extend it.
3365 */
3366 li->discard_end = li->discard_end + 1ULL;
3367
3368 else {
3369 /*
3370 * Emit the old range and start a new one.
3371 */
3372 set_discard_range(li);
3373 li->discard_begin = from_dblock(dblock);
3374 li->discard_end = li->discard_begin + 1ULL;
3375 }
3376 } else {
3377 set_discard_range(li);
3378 li->discard_begin = li->discard_end = 0;
3379 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003380
3381 return 0;
3382}
3383
Joe Thornberf494a9c2013-10-31 13:55:49 -04003384static dm_cblock_t get_cache_dev_size(struct cache *cache)
3385{
3386 sector_t size = get_dev_size(cache->cache_dev);
3387 (void) sector_div(size, cache->sectors_per_block);
3388 return to_cblock(size);
3389}
3390
3391static bool can_resize(struct cache *cache, dm_cblock_t new_size)
3392{
3393 if (from_cblock(new_size) > from_cblock(cache->cache_size))
3394 return true;
3395
3396 /*
3397 * We can't drop a dirty block when shrinking the cache.
3398 */
3399 while (from_cblock(new_size) < from_cblock(cache->cache_size)) {
3400 new_size = to_cblock(from_cblock(new_size) + 1);
3401 if (is_dirty(cache, new_size)) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003402 DMERR("%s: unable to shrink cache; cache block %llu is dirty",
3403 cache_device_name(cache),
Joe Thornberf494a9c2013-10-31 13:55:49 -04003404 (unsigned long long) from_cblock(new_size));
3405 return false;
3406 }
3407 }
3408
3409 return true;
3410}
3411
3412static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size)
3413{
3414 int r;
3415
Vincent Pelletier08844802013-11-30 12:58:42 +01003416 r = dm_cache_resize(cache->cmd, new_size);
Joe Thornberf494a9c2013-10-31 13:55:49 -04003417 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003418 DMERR("%s: could not resize cache metadata", cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -04003419 metadata_operation_failed(cache, "dm_cache_resize", r);
Joe Thornberf494a9c2013-10-31 13:55:49 -04003420 return r;
3421 }
3422
Joe Thornberd1d92202014-11-11 11:58:32 +00003423 set_cache_size(cache, new_size);
Joe Thornberf494a9c2013-10-31 13:55:49 -04003424
3425 return 0;
3426}
3427
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003428static int cache_preresume(struct dm_target *ti)
3429{
3430 int r = 0;
3431 struct cache *cache = ti->private;
Joe Thornberf494a9c2013-10-31 13:55:49 -04003432 dm_cblock_t csize = get_cache_dev_size(cache);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003433
3434 /*
3435 * Check to see if the cache has resized.
3436 */
Joe Thornberf494a9c2013-10-31 13:55:49 -04003437 if (!cache->sized) {
3438 r = resize_cache_dev(cache, csize);
3439 if (r)
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003440 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003441
3442 cache->sized = true;
Joe Thornberf494a9c2013-10-31 13:55:49 -04003443
3444 } else if (csize != cache->cache_size) {
3445 if (!can_resize(cache, csize))
3446 return -EINVAL;
3447
3448 r = resize_cache_dev(cache, csize);
3449 if (r)
3450 return r;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003451 }
3452
3453 if (!cache->loaded_mappings) {
Mike Snitzerea2dd8c2013-03-20 17:21:28 +00003454 r = dm_cache_load_mappings(cache->cmd, cache->policy,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003455 load_mapping, cache);
3456 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003457 DMERR("%s: could not load cache mappings", cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -04003458 metadata_operation_failed(cache, "dm_cache_load_mappings", r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003459 return r;
3460 }
3461
3462 cache->loaded_mappings = true;
3463 }
3464
3465 if (!cache->loaded_discards) {
Joe Thornber3e2e1c32014-11-24 14:06:22 +00003466 struct discard_load_info li;
3467
3468 /*
3469 * The discard bitset could have been resized, or the
3470 * discard block size changed. To be safe we start by
3471 * setting every dblock to not discarded.
3472 */
3473 clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks));
3474
3475 discard_load_info_init(cache, &li);
3476 r = dm_cache_load_discards(cache->cmd, load_discard, &li);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003477 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003478 DMERR("%s: could not load origin discards", cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -04003479 metadata_operation_failed(cache, "dm_cache_load_discards", r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003480 return r;
3481 }
Joe Thornber3e2e1c32014-11-24 14:06:22 +00003482 set_discard_range(&li);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003483
3484 cache->loaded_discards = true;
3485 }
3486
3487 return r;
3488}
3489
3490static void cache_resume(struct dm_target *ti)
3491{
3492 struct cache *cache = ti->private;
3493
3494 cache->need_tick_bio = true;
3495 do_waker(&cache->waker.work);
3496}
3497
3498/*
3499 * Status format:
3500 *
Mike Snitzer6a388612014-01-09 16:04:12 -05003501 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
3502 * <cache block size> <#used cache blocks>/<#total cache blocks>
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003503 * <#read hits> <#read misses> <#write hits> <#write misses>
Mike Snitzer6a388612014-01-09 16:04:12 -05003504 * <#demotions> <#promotions> <#dirty>
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003505 * <#features> <features>*
3506 * <#core args> <core args>
Mike Snitzer255eac22015-07-15 11:42:59 -04003507 * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check>
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003508 */
3509static void cache_status(struct dm_target *ti, status_type_t type,
3510 unsigned status_flags, char *result, unsigned maxlen)
3511{
3512 int r = 0;
3513 unsigned i;
3514 ssize_t sz = 0;
3515 dm_block_t nr_free_blocks_metadata = 0;
3516 dm_block_t nr_blocks_metadata = 0;
3517 char buf[BDEVNAME_SIZE];
3518 struct cache *cache = ti->private;
3519 dm_cblock_t residency;
Joe Thornberd14fcf32016-03-10 16:20:58 +00003520 bool needs_check;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003521
3522 switch (type) {
3523 case STATUSTYPE_INFO:
Joe Thornber028ae9f2015-04-22 16:42:35 -04003524 if (get_cache_mode(cache) == CM_FAIL) {
3525 DMEMIT("Fail");
3526 break;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003527 }
3528
Joe Thornber028ae9f2015-04-22 16:42:35 -04003529 /* Commit to ensure statistics aren't out-of-date */
3530 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3531 (void) commit(cache, false);
3532
Mike Snitzerb61d9502015-04-22 17:25:56 -04003533 r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003534 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003535 DMERR("%s: dm_cache_get_free_metadata_block_count returned %d",
3536 cache_device_name(cache), r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003537 goto err;
3538 }
3539
3540 r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata);
3541 if (r) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003542 DMERR("%s: dm_cache_get_metadata_dev_size returned %d",
3543 cache_device_name(cache), r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003544 goto err;
3545 }
3546
3547 residency = policy_residency(cache->policy);
3548
Anssi Hannula44fa8162014-08-01 11:55:47 -04003549 DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ",
Mike Snitzer895b47d2014-07-14 15:37:18 -04003550 (unsigned)DM_CACHE_METADATA_BLOCK_SIZE,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003551 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3552 (unsigned long long)nr_blocks_metadata,
Mike Snitzer6a388612014-01-09 16:04:12 -05003553 cache->sectors_per_block,
3554 (unsigned long long) from_cblock(residency),
3555 (unsigned long long) from_cblock(cache->cache_size),
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003556 (unsigned) atomic_read(&cache->stats.read_hit),
3557 (unsigned) atomic_read(&cache->stats.read_miss),
3558 (unsigned) atomic_read(&cache->stats.write_hit),
3559 (unsigned) atomic_read(&cache->stats.write_miss),
3560 (unsigned) atomic_read(&cache->stats.demotion),
3561 (unsigned) atomic_read(&cache->stats.promotion),
Anssi Hannula44fa8162014-08-01 11:55:47 -04003562 (unsigned long) atomic_read(&cache->nr_dirty));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003563
Joe Thornber2ee57d52013-10-24 14:10:29 -04003564 if (writethrough_mode(&cache->features))
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003565 DMEMIT("1 writethrough ");
Joe Thornber2ee57d52013-10-24 14:10:29 -04003566
3567 else if (passthrough_mode(&cache->features))
3568 DMEMIT("1 passthrough ");
3569
3570 else if (writeback_mode(&cache->features))
3571 DMEMIT("1 writeback ");
3572
3573 else {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003574 DMERR("%s: internal error: unknown io mode: %d",
3575 cache_device_name(cache), (int) cache->features.io_mode);
Joe Thornber2ee57d52013-10-24 14:10:29 -04003576 goto err;
3577 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003578
3579 DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
Mike Snitzer2e68c4e2014-01-15 21:06:55 -05003580
3581 DMEMIT("%s ", dm_cache_policy_get_name(cache->policy));
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003582 if (sz < maxlen) {
Joe Thornber028ae9f2015-04-22 16:42:35 -04003583 r = policy_emit_config_values(cache->policy, result, maxlen, &sz);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003584 if (r)
Mike Snitzerb61d9502015-04-22 17:25:56 -04003585 DMERR("%s: policy_emit_config_values returned %d",
3586 cache_device_name(cache), r);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003587 }
3588
Joe Thornber028ae9f2015-04-22 16:42:35 -04003589 if (get_cache_mode(cache) == CM_READ_ONLY)
3590 DMEMIT("ro ");
3591 else
3592 DMEMIT("rw ");
3593
Joe Thornberd14fcf32016-03-10 16:20:58 +00003594 r = dm_cache_metadata_needs_check(cache->cmd, &needs_check);
3595
3596 if (r || needs_check)
Mike Snitzer255eac22015-07-15 11:42:59 -04003597 DMEMIT("needs_check ");
3598 else
3599 DMEMIT("- ");
3600
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003601 break;
3602
3603 case STATUSTYPE_TABLE:
3604 format_dev_t(buf, cache->metadata_dev->bdev->bd_dev);
3605 DMEMIT("%s ", buf);
3606 format_dev_t(buf, cache->cache_dev->bdev->bd_dev);
3607 DMEMIT("%s ", buf);
3608 format_dev_t(buf, cache->origin_dev->bdev->bd_dev);
3609 DMEMIT("%s", buf);
3610
3611 for (i = 0; i < cache->nr_ctr_args - 1; i++)
3612 DMEMIT(" %s", cache->ctr_args[i]);
3613 if (cache->nr_ctr_args)
3614 DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]);
3615 }
3616
3617 return;
3618
3619err:
3620 DMEMIT("Error");
3621}
3622
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003623/*
Joe Thornber65790ff2013-11-08 16:39:50 +00003624 * A cache block range can take two forms:
3625 *
3626 * i) A single cblock, eg. '3456'
3627 * ii) A begin and end cblock with dots between, eg. 123-234
3628 */
3629static int parse_cblock_range(struct cache *cache, const char *str,
3630 struct cblock_range *result)
3631{
3632 char dummy;
3633 uint64_t b, e;
3634 int r;
3635
3636 /*
3637 * Try and parse form (ii) first.
3638 */
3639 r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy);
3640 if (r < 0)
3641 return r;
3642
3643 if (r == 2) {
3644 result->begin = to_cblock(b);
3645 result->end = to_cblock(e);
3646 return 0;
3647 }
3648
3649 /*
3650 * That didn't work, try form (i).
3651 */
3652 r = sscanf(str, "%llu%c", &b, &dummy);
3653 if (r < 0)
3654 return r;
3655
3656 if (r == 1) {
3657 result->begin = to_cblock(b);
3658 result->end = to_cblock(from_cblock(result->begin) + 1u);
3659 return 0;
3660 }
3661
Mike Snitzerb61d9502015-04-22 17:25:56 -04003662 DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str);
Joe Thornber65790ff2013-11-08 16:39:50 +00003663 return -EINVAL;
3664}
3665
3666static int validate_cblock_range(struct cache *cache, struct cblock_range *range)
3667{
3668 uint64_t b = from_cblock(range->begin);
3669 uint64_t e = from_cblock(range->end);
3670 uint64_t n = from_cblock(cache->cache_size);
3671
3672 if (b >= n) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003673 DMERR("%s: begin cblock out of range: %llu >= %llu",
3674 cache_device_name(cache), b, n);
Joe Thornber65790ff2013-11-08 16:39:50 +00003675 return -EINVAL;
3676 }
3677
3678 if (e > n) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003679 DMERR("%s: end cblock out of range: %llu > %llu",
3680 cache_device_name(cache), e, n);
Joe Thornber65790ff2013-11-08 16:39:50 +00003681 return -EINVAL;
3682 }
3683
3684 if (b >= e) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003685 DMERR("%s: invalid cblock range: %llu >= %llu",
3686 cache_device_name(cache), b, e);
Joe Thornber65790ff2013-11-08 16:39:50 +00003687 return -EINVAL;
3688 }
3689
3690 return 0;
3691}
3692
3693static int request_invalidation(struct cache *cache, struct cblock_range *range)
3694{
3695 struct invalidation_request req;
3696
3697 INIT_LIST_HEAD(&req.list);
3698 req.cblocks = range;
3699 atomic_set(&req.complete, 0);
3700 req.err = 0;
3701 init_waitqueue_head(&req.result_wait);
3702
3703 spin_lock(&cache->invalidation_lock);
3704 list_add(&req.list, &cache->invalidation_requests);
3705 spin_unlock(&cache->invalidation_lock);
3706 wake_worker(cache);
3707
3708 wait_event(req.result_wait, atomic_read(&req.complete));
3709 return req.err;
3710}
3711
3712static int process_invalidate_cblocks_message(struct cache *cache, unsigned count,
3713 const char **cblock_ranges)
3714{
3715 int r = 0;
3716 unsigned i;
3717 struct cblock_range range;
3718
3719 if (!passthrough_mode(&cache->features)) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003720 DMERR("%s: cache has to be in passthrough mode for invalidation",
3721 cache_device_name(cache));
Joe Thornber65790ff2013-11-08 16:39:50 +00003722 return -EPERM;
3723 }
3724
3725 for (i = 0; i < count; i++) {
3726 r = parse_cblock_range(cache, cblock_ranges[i], &range);
3727 if (r)
3728 break;
3729
3730 r = validate_cblock_range(cache, &range);
3731 if (r)
3732 break;
3733
3734 /*
3735 * Pass begin and end origin blocks to the worker and wake it.
3736 */
3737 r = request_invalidation(cache, &range);
3738 if (r)
3739 break;
3740 }
3741
3742 return r;
3743}
3744
3745/*
3746 * Supports
3747 * "<key> <value>"
3748 * and
3749 * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]*
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003750 *
3751 * The key migration_threshold is supported by the cache target core.
3752 */
3753static int cache_message(struct dm_target *ti, unsigned argc, char **argv)
3754{
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003755 struct cache *cache = ti->private;
3756
Joe Thornber65790ff2013-11-08 16:39:50 +00003757 if (!argc)
3758 return -EINVAL;
3759
Joe Thornber028ae9f2015-04-22 16:42:35 -04003760 if (get_cache_mode(cache) >= CM_READ_ONLY) {
Mike Snitzerb61d9502015-04-22 17:25:56 -04003761 DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode",
3762 cache_device_name(cache));
Joe Thornber028ae9f2015-04-22 16:42:35 -04003763 return -EOPNOTSUPP;
3764 }
3765
Mike Snitzer7b6b2bc2013-11-12 12:17:43 -05003766 if (!strcasecmp(argv[0], "invalidate_cblocks"))
Joe Thornber65790ff2013-11-08 16:39:50 +00003767 return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1);
3768
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003769 if (argc != 2)
3770 return -EINVAL;
3771
Joe Thornber2f14f4b2013-05-10 14:37:21 +01003772 return set_config_value(cache, argv[0], argv[1]);
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003773}
3774
3775static int cache_iterate_devices(struct dm_target *ti,
3776 iterate_devices_callout_fn fn, void *data)
3777{
3778 int r = 0;
3779 struct cache *cache = ti->private;
3780
3781 r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data);
3782 if (!r)
3783 r = fn(ti, cache->origin_dev, 0, ti->len, data);
3784
3785 return r;
3786}
3787
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003788static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
3789{
3790 /*
3791 * FIXME: these limits may be incompatible with the cache device
3792 */
Joe Thornber7ae34e72014-11-06 10:18:04 +00003793 limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024,
3794 cache->origin_sectors);
Joe Thornber1bad9bc2014-11-07 14:47:07 +00003795 limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003796}
3797
3798static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
3799{
3800 struct cache *cache = ti->private;
Mike Snitzerf6109372013-08-20 15:02:41 -04003801 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003802
Mike Snitzerf6109372013-08-20 15:02:41 -04003803 /*
3804 * If the system-determined stacked limits are compatible with the
3805 * cache's blocksize (io_opt is a factor) do not override them.
3806 */
3807 if (io_opt_sectors < cache->sectors_per_block ||
3808 do_div(io_opt_sectors, cache->sectors_per_block)) {
Mike Snitzerb0246532014-07-19 13:25:46 -04003809 blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT);
Mike Snitzerf6109372013-08-20 15:02:41 -04003810 blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
3811 }
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003812 set_discard_limits(cache, limits);
3813}
3814
3815/*----------------------------------------------------------------*/
3816
3817static struct target_type cache_target = {
3818 .name = "cache",
Mike Snitzer843f0f22016-03-10 13:40:26 -05003819 .version = {1, 9, 0},
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003820 .module = THIS_MODULE,
3821 .ctr = cache_ctr,
3822 .dtr = cache_dtr,
3823 .map = cache_map,
3824 .end_io = cache_end_io,
3825 .postsuspend = cache_postsuspend,
3826 .preresume = cache_preresume,
3827 .resume = cache_resume,
3828 .status = cache_status,
3829 .message = cache_message,
3830 .iterate_devices = cache_iterate_devices,
Joe Thornberc6b4fcb2013-03-01 22:45:51 +00003831 .io_hints = cache_io_hints,
3832};
3833
3834static int __init dm_cache_init(void)
3835{
3836 int r;
3837
3838 r = dm_register_target(&cache_target);
3839 if (r) {
3840 DMERR("cache target registration failed: %d", r);
3841 return r;
3842 }
3843
3844 migration_cache = KMEM_CACHE(dm_cache_migration, 0);
3845 if (!migration_cache) {
3846 dm_unregister_target(&cache_target);
3847 return -ENOMEM;
3848 }
3849
3850 return 0;
3851}
3852
3853static void __exit dm_cache_exit(void)
3854{
3855 dm_unregister_target(&cache_target);
3856 kmem_cache_destroy(migration_cache);
3857}
3858
3859module_init(dm_cache_init);
3860module_exit(dm_cache_exit);
3861
3862MODULE_DESCRIPTION(DM_NAME " cache target");
3863MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
3864MODULE_LICENSE("GPL");