]>
Commit | Line | Data |
---|---|---|
c6b4fcba JT |
1 | /* |
2 | * Copyright (C) 2012 Red Hat. All rights reserved. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm.h" | |
b29d4986 | 8 | #include "dm-bio-prison-v2.h" |
b844fe69 | 9 | #include "dm-bio-record.h" |
c6b4fcba JT |
10 | #include "dm-cache-metadata.h" |
11 | ||
12 | #include <linux/dm-io.h> | |
13 | #include <linux/dm-kcopyd.h> | |
0f30af98 | 14 | #include <linux/jiffies.h> |
c6b4fcba JT |
15 | #include <linux/init.h> |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
b29d4986 | 18 | #include <linux/rwsem.h> |
c6b4fcba JT |
19 | #include <linux/slab.h> |
20 | #include <linux/vmalloc.h> | |
21 | ||
22 | #define DM_MSG_PREFIX "cache" | |
23 | ||
24 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle, | |
25 | "A percentage of time allocated for copying to and/or from cache"); | |
26 | ||
27 | /*----------------------------------------------------------------*/ | |
28 | ||
b29d4986 JT |
29 | /* |
30 | * Glossary: | |
31 | * | |
32 | * oblock: index of an origin block | |
33 | * cblock: index of a cache block | |
34 | * promotion: movement of a block from origin to cache | |
35 | * demotion: movement of a block from cache to origin | |
36 | * migration: movement of a block between the origin and cache device, | |
37 | * either direction | |
38 | */ | |
39 | ||
40 | /*----------------------------------------------------------------*/ | |
77289d32 JT |
41 | |
42 | struct io_tracker { | |
43 | spinlock_t lock; | |
44 | ||
45 | /* | |
46 | * Sectors of in-flight IO. | |
47 | */ | |
48 | sector_t in_flight; | |
49 | ||
50 | /* | |
51 | * The time, in jiffies, when this device became idle (if it is | |
52 | * indeed idle). | |
53 | */ | |
54 | unsigned long idle_time; | |
55 | unsigned long last_update_time; | |
56 | }; | |
57 | ||
58 | static void iot_init(struct io_tracker *iot) | |
59 | { | |
60 | spin_lock_init(&iot->lock); | |
61 | iot->in_flight = 0ul; | |
62 | iot->idle_time = 0ul; | |
63 | iot->last_update_time = jiffies; | |
64 | } | |
65 | ||
66 | static bool __iot_idle_for(struct io_tracker *iot, unsigned long jifs) | |
67 | { | |
68 | if (iot->in_flight) | |
69 | return false; | |
70 | ||
71 | return time_after(jiffies, iot->idle_time + jifs); | |
72 | } | |
73 | ||
74 | static bool iot_idle_for(struct io_tracker *iot, unsigned long jifs) | |
75 | { | |
76 | bool r; | |
77 | unsigned long flags; | |
78 | ||
79 | spin_lock_irqsave(&iot->lock, flags); | |
80 | r = __iot_idle_for(iot, jifs); | |
81 | spin_unlock_irqrestore(&iot->lock, flags); | |
82 | ||
83 | return r; | |
84 | } | |
85 | ||
86 | static void iot_io_begin(struct io_tracker *iot, sector_t len) | |
87 | { | |
88 | unsigned long flags; | |
89 | ||
90 | spin_lock_irqsave(&iot->lock, flags); | |
91 | iot->in_flight += len; | |
92 | spin_unlock_irqrestore(&iot->lock, flags); | |
93 | } | |
94 | ||
95 | static void __iot_io_end(struct io_tracker *iot, sector_t len) | |
96 | { | |
97 | iot->in_flight -= len; | |
98 | if (!iot->in_flight) | |
99 | iot->idle_time = jiffies; | |
100 | } | |
101 | ||
102 | static void iot_io_end(struct io_tracker *iot, sector_t len) | |
103 | { | |
104 | unsigned long flags; | |
105 | ||
106 | spin_lock_irqsave(&iot->lock, flags); | |
107 | __iot_io_end(iot, len); | |
108 | spin_unlock_irqrestore(&iot->lock, flags); | |
109 | } | |
110 | ||
111 | /*----------------------------------------------------------------*/ | |
112 | ||
c6b4fcba | 113 | /* |
b29d4986 JT |
114 | * Represents a chunk of future work. 'input' allows continuations to pass |
115 | * values between themselves, typically error values. | |
c6b4fcba | 116 | */ |
b29d4986 JT |
117 | struct continuation { |
118 | struct work_struct ws; | |
119 | int input; | |
120 | }; | |
121 | ||
122 | static inline void init_continuation(struct continuation *k, | |
123 | void (*fn)(struct work_struct *)) | |
124 | { | |
125 | INIT_WORK(&k->ws, fn); | |
126 | k->input = 0; | |
127 | } | |
128 | ||
129 | static inline void queue_continuation(struct workqueue_struct *wq, | |
130 | struct continuation *k) | |
131 | { | |
132 | queue_work(wq, &k->ws); | |
133 | } | |
c6b4fcba JT |
134 | |
135 | /*----------------------------------------------------------------*/ | |
136 | ||
b29d4986 JT |
137 | /* |
138 | * The batcher collects together pieces of work that need a particular | |
139 | * operation to occur before they can proceed (typically a commit). | |
140 | */ | |
141 | struct batcher { | |
142 | /* | |
143 | * The operation that everyone is waiting for. | |
144 | */ | |
145 | int (*commit_op)(void *context); | |
146 | void *commit_context; | |
147 | ||
148 | /* | |
149 | * This is how bios should be issued once the commit op is complete | |
150 | * (accounted_request). | |
151 | */ | |
152 | void (*issue_op)(struct bio *bio, void *context); | |
153 | void *issue_context; | |
154 | ||
155 | /* | |
156 | * Queued work gets put on here after commit. | |
157 | */ | |
158 | struct workqueue_struct *wq; | |
159 | ||
160 | spinlock_t lock; | |
161 | struct list_head work_items; | |
162 | struct bio_list bios; | |
163 | struct work_struct commit_work; | |
164 | ||
165 | bool commit_scheduled; | |
166 | }; | |
167 | ||
168 | static void __commit(struct work_struct *_ws) | |
169 | { | |
170 | struct batcher *b = container_of(_ws, struct batcher, commit_work); | |
171 | ||
172 | int r; | |
173 | unsigned long flags; | |
174 | struct list_head work_items; | |
175 | struct work_struct *ws, *tmp; | |
176 | struct continuation *k; | |
177 | struct bio *bio; | |
178 | struct bio_list bios; | |
179 | ||
180 | INIT_LIST_HEAD(&work_items); | |
181 | bio_list_init(&bios); | |
182 | ||
183 | /* | |
184 | * We have to grab these before the commit_op to avoid a race | |
185 | * condition. | |
186 | */ | |
187 | spin_lock_irqsave(&b->lock, flags); | |
188 | list_splice_init(&b->work_items, &work_items); | |
189 | bio_list_merge(&bios, &b->bios); | |
190 | bio_list_init(&b->bios); | |
191 | b->commit_scheduled = false; | |
192 | spin_unlock_irqrestore(&b->lock, flags); | |
193 | ||
194 | r = b->commit_op(b->commit_context); | |
195 | ||
196 | list_for_each_entry_safe(ws, tmp, &work_items, entry) { | |
197 | k = container_of(ws, struct continuation, ws); | |
198 | k->input = r; | |
199 | INIT_LIST_HEAD(&ws->entry); /* to avoid a WARN_ON */ | |
200 | queue_work(b->wq, ws); | |
201 | } | |
202 | ||
203 | while ((bio = bio_list_pop(&bios))) { | |
204 | if (r) { | |
205 | bio->bi_error = r; | |
206 | bio_endio(bio); | |
207 | } else | |
208 | b->issue_op(bio, b->issue_context); | |
209 | } | |
210 | } | |
211 | ||
212 | static void batcher_init(struct batcher *b, | |
213 | int (*commit_op)(void *), | |
214 | void *commit_context, | |
215 | void (*issue_op)(struct bio *bio, void *), | |
216 | void *issue_context, | |
217 | struct workqueue_struct *wq) | |
218 | { | |
219 | b->commit_op = commit_op; | |
220 | b->commit_context = commit_context; | |
221 | b->issue_op = issue_op; | |
222 | b->issue_context = issue_context; | |
223 | b->wq = wq; | |
224 | ||
225 | spin_lock_init(&b->lock); | |
226 | INIT_LIST_HEAD(&b->work_items); | |
227 | bio_list_init(&b->bios); | |
228 | INIT_WORK(&b->commit_work, __commit); | |
229 | b->commit_scheduled = false; | |
230 | } | |
231 | ||
232 | static void async_commit(struct batcher *b) | |
233 | { | |
234 | queue_work(b->wq, &b->commit_work); | |
235 | } | |
236 | ||
237 | static void continue_after_commit(struct batcher *b, struct continuation *k) | |
238 | { | |
239 | unsigned long flags; | |
240 | bool commit_scheduled; | |
241 | ||
242 | spin_lock_irqsave(&b->lock, flags); | |
243 | commit_scheduled = b->commit_scheduled; | |
244 | list_add_tail(&k->ws.entry, &b->work_items); | |
245 | spin_unlock_irqrestore(&b->lock, flags); | |
246 | ||
247 | if (commit_scheduled) | |
248 | async_commit(b); | |
249 | } | |
250 | ||
251 | /* | |
252 | * Bios are errored if commit failed. | |
253 | */ | |
254 | static void issue_after_commit(struct batcher *b, struct bio *bio) | |
255 | { | |
256 | unsigned long flags; | |
257 | bool commit_scheduled; | |
258 | ||
259 | spin_lock_irqsave(&b->lock, flags); | |
260 | commit_scheduled = b->commit_scheduled; | |
261 | bio_list_add(&b->bios, bio); | |
262 | spin_unlock_irqrestore(&b->lock, flags); | |
263 | ||
264 | if (commit_scheduled) | |
265 | async_commit(b); | |
266 | } | |
267 | ||
268 | /* | |
269 | * Call this if some urgent work is waiting for the commit to complete. | |
270 | */ | |
271 | static void schedule_commit(struct batcher *b) | |
272 | { | |
273 | bool immediate; | |
274 | unsigned long flags; | |
275 | ||
276 | spin_lock_irqsave(&b->lock, flags); | |
277 | immediate = !list_empty(&b->work_items) || !bio_list_empty(&b->bios); | |
278 | b->commit_scheduled = true; | |
279 | spin_unlock_irqrestore(&b->lock, flags); | |
280 | ||
281 | if (immediate) | |
282 | async_commit(b); | |
283 | } | |
284 | ||
c9d28d5d JT |
285 | /* |
286 | * There are a couple of places where we let a bio run, but want to do some | |
287 | * work before calling its endio function. We do this by temporarily | |
288 | * changing the endio fn. | |
289 | */ | |
290 | struct dm_hook_info { | |
291 | bio_end_io_t *bi_end_io; | |
c9d28d5d JT |
292 | }; |
293 | ||
294 | static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio, | |
295 | bio_end_io_t *bi_end_io, void *bi_private) | |
296 | { | |
297 | h->bi_end_io = bio->bi_end_io; | |
c9d28d5d JT |
298 | |
299 | bio->bi_end_io = bi_end_io; | |
300 | bio->bi_private = bi_private; | |
301 | } | |
302 | ||
303 | static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio) | |
304 | { | |
305 | bio->bi_end_io = h->bi_end_io; | |
c9d28d5d JT |
306 | } |
307 | ||
308 | /*----------------------------------------------------------------*/ | |
309 | ||
c6b4fcba JT |
310 | #define MIGRATION_POOL_SIZE 128 |
311 | #define COMMIT_PERIOD HZ | |
312 | #define MIGRATION_COUNT_WINDOW 10 | |
313 | ||
314 | /* | |
05473044 MS |
315 | * The block size of the device holding cache data must be |
316 | * between 32KB and 1GB. | |
c6b4fcba JT |
317 | */ |
318 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT) | |
05473044 | 319 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
c6b4fcba | 320 | |
2ee57d58 | 321 | enum cache_metadata_mode { |
c6b4fcba JT |
322 | CM_WRITE, /* metadata may be changed */ |
323 | CM_READ_ONLY, /* metadata may not be changed */ | |
028ae9f7 | 324 | CM_FAIL |
c6b4fcba JT |
325 | }; |
326 | ||
2ee57d58 JT |
327 | enum cache_io_mode { |
328 | /* | |
329 | * Data is written to cached blocks only. These blocks are marked | |
330 | * dirty. If you lose the cache device you will lose data. | |
331 | * Potential performance increase for both reads and writes. | |
332 | */ | |
333 | CM_IO_WRITEBACK, | |
334 | ||
335 | /* | |
336 | * Data is written to both cache and origin. Blocks are never | |
337 | * dirty. Potential performance benfit for reads only. | |
338 | */ | |
339 | CM_IO_WRITETHROUGH, | |
340 | ||
341 | /* | |
342 | * A degraded mode useful for various cache coherency situations | |
343 | * (eg, rolling back snapshots). Reads and writes always go to the | |
344 | * origin. If a write goes to a cached oblock, then the cache | |
345 | * block is invalidated. | |
346 | */ | |
347 | CM_IO_PASSTHROUGH | |
348 | }; | |
349 | ||
c6b4fcba | 350 | struct cache_features { |
2ee57d58 JT |
351 | enum cache_metadata_mode mode; |
352 | enum cache_io_mode io_mode; | |
629d0a8a | 353 | unsigned metadata_version; |
c6b4fcba JT |
354 | }; |
355 | ||
356 | struct cache_stats { | |
357 | atomic_t read_hit; | |
358 | atomic_t read_miss; | |
359 | atomic_t write_hit; | |
360 | atomic_t write_miss; | |
361 | atomic_t demotion; | |
362 | atomic_t promotion; | |
b29d4986 | 363 | atomic_t writeback; |
c6b4fcba JT |
364 | atomic_t copies_avoided; |
365 | atomic_t cache_cell_clash; | |
366 | atomic_t commit_count; | |
367 | atomic_t discard_count; | |
368 | }; | |
369 | ||
370 | struct cache { | |
371 | struct dm_target *ti; | |
372 | struct dm_target_callbacks callbacks; | |
373 | ||
c9ec5d7c MS |
374 | struct dm_cache_metadata *cmd; |
375 | ||
c6b4fcba JT |
376 | /* |
377 | * Metadata is written to this device. | |
378 | */ | |
379 | struct dm_dev *metadata_dev; | |
380 | ||
381 | /* | |
382 | * The slower of the two data devices. Typically a spindle. | |
383 | */ | |
384 | struct dm_dev *origin_dev; | |
385 | ||
386 | /* | |
387 | * The faster of the two data devices. Typically an SSD. | |
388 | */ | |
389 | struct dm_dev *cache_dev; | |
390 | ||
c6b4fcba JT |
391 | /* |
392 | * Size of the origin device in _complete_ blocks and native sectors. | |
393 | */ | |
394 | dm_oblock_t origin_blocks; | |
395 | sector_t origin_sectors; | |
396 | ||
397 | /* | |
398 | * Size of the cache device in blocks. | |
399 | */ | |
400 | dm_cblock_t cache_size; | |
401 | ||
402 | /* | |
403 | * Fields for converting from sectors to blocks. | |
404 | */ | |
ca763d0a | 405 | sector_t sectors_per_block; |
c6b4fcba JT |
406 | int sectors_per_block_shift; |
407 | ||
c6b4fcba | 408 | spinlock_t lock; |
651f5fa2 | 409 | struct list_head deferred_cells; |
c6b4fcba | 410 | struct bio_list deferred_bios; |
e2e74d61 | 411 | struct bio_list deferred_writethrough_bios; |
c6b4fcba | 412 | sector_t migration_threshold; |
c6b4fcba | 413 | wait_queue_head_t migration_wait; |
a59db676 JT |
414 | atomic_t nr_allocated_migrations; |
415 | ||
416 | /* | |
417 | * The number of in flight migrations that are performing | |
418 | * background io. eg, promotion, writeback. | |
419 | */ | |
420 | atomic_t nr_io_migrations; | |
c6b4fcba | 421 | |
b29d4986 | 422 | struct rw_semaphore quiesce_lock; |
66cb1910 | 423 | |
c6b4fcba JT |
424 | /* |
425 | * cache_size entries, dirty if set | |
426 | */ | |
44fa816b | 427 | atomic_t nr_dirty; |
c6b4fcba JT |
428 | unsigned long *dirty_bitset; |
429 | ||
430 | /* | |
431 | * origin_blocks entries, discarded if set. | |
432 | */ | |
1bad9bc4 | 433 | dm_dblock_t discard_nr_blocks; |
c6b4fcba | 434 | unsigned long *discard_bitset; |
08b18451 | 435 | uint32_t discard_block_size; /* a power of 2 times sectors per block */ |
c9ec5d7c MS |
436 | |
437 | /* | |
438 | * Rather than reconstructing the table line for the status we just | |
439 | * save it and regurgitate. | |
440 | */ | |
441 | unsigned nr_ctr_args; | |
442 | const char **ctr_args; | |
c6b4fcba JT |
443 | |
444 | struct dm_kcopyd_client *copier; | |
445 | struct workqueue_struct *wq; | |
b29d4986 JT |
446 | struct work_struct deferred_bio_worker; |
447 | struct work_struct deferred_writethrough_worker; | |
448 | struct work_struct migration_worker; | |
c6b4fcba | 449 | struct delayed_work waker; |
b29d4986 | 450 | struct dm_bio_prison_v2 *prison; |
c6b4fcba JT |
451 | |
452 | mempool_t *migration_pool; | |
c6b4fcba JT |
453 | |
454 | struct dm_cache_policy *policy; | |
455 | unsigned policy_nr_args; | |
456 | ||
457 | bool need_tick_bio:1; | |
458 | bool sized:1; | |
65790ff9 | 459 | bool invalidate:1; |
c6b4fcba JT |
460 | bool commit_requested:1; |
461 | bool loaded_mappings:1; | |
462 | bool loaded_discards:1; | |
463 | ||
c6b4fcba | 464 | /* |
c9ec5d7c | 465 | * Cache features such as write-through. |
c6b4fcba | 466 | */ |
c9ec5d7c MS |
467 | struct cache_features features; |
468 | ||
469 | struct cache_stats stats; | |
65790ff9 JT |
470 | |
471 | /* | |
472 | * Invalidation fields. | |
473 | */ | |
474 | spinlock_t invalidation_lock; | |
475 | struct list_head invalidation_requests; | |
066dbaa3 JT |
476 | |
477 | struct io_tracker origin_tracker; | |
b29d4986 JT |
478 | |
479 | struct work_struct commit_ws; | |
480 | struct batcher committer; | |
481 | ||
482 | struct rw_semaphore background_work_lock; | |
c6b4fcba JT |
483 | }; |
484 | ||
485 | struct per_bio_data { | |
486 | bool tick:1; | |
487 | unsigned req_nr:2; | |
b29d4986 | 488 | struct dm_bio_prison_cell_v2 *cell; |
c6eda5e8 | 489 | struct dm_hook_info hook_info; |
066dbaa3 | 490 | sector_t len; |
e2e74d61 | 491 | |
19b0092e MS |
492 | /* |
493 | * writethrough fields. These MUST remain at the end of this | |
494 | * structure and the 'cache' member must be the first as it | |
aeed1420 | 495 | * is used to determine the offset of the writethrough fields. |
19b0092e | 496 | */ |
e2e74d61 JT |
497 | struct cache *cache; |
498 | dm_cblock_t cblock; | |
b844fe69 | 499 | struct dm_bio_details bio_details; |
c6b4fcba JT |
500 | }; |
501 | ||
502 | struct dm_cache_migration { | |
b29d4986 | 503 | struct continuation k; |
c6b4fcba JT |
504 | struct cache *cache; |
505 | ||
b29d4986 JT |
506 | struct policy_work *op; |
507 | struct bio *overwrite_bio; | |
508 | struct dm_bio_prison_cell_v2 *cell; | |
c6b4fcba | 509 | |
b29d4986 JT |
510 | dm_cblock_t invalidate_cblock; |
511 | dm_oblock_t invalidate_oblock; | |
c6b4fcba JT |
512 | }; |
513 | ||
b29d4986 JT |
514 | /*----------------------------------------------------------------*/ |
515 | ||
516 | static bool writethrough_mode(struct cache_features *f) | |
517 | { | |
518 | return f->io_mode == CM_IO_WRITETHROUGH; | |
519 | } | |
520 | ||
521 | static bool writeback_mode(struct cache_features *f) | |
522 | { | |
523 | return f->io_mode == CM_IO_WRITEBACK; | |
524 | } | |
525 | ||
526 | static inline bool passthrough_mode(struct cache_features *f) | |
527 | { | |
528 | return unlikely(f->io_mode == CM_IO_PASSTHROUGH); | |
529 | } | |
530 | ||
531 | /*----------------------------------------------------------------*/ | |
532 | ||
533 | static void wake_deferred_bio_worker(struct cache *cache) | |
534 | { | |
535 | queue_work(cache->wq, &cache->deferred_bio_worker); | |
536 | } | |
c6b4fcba | 537 | |
b29d4986 JT |
538 | static void wake_deferred_writethrough_worker(struct cache *cache) |
539 | { | |
540 | queue_work(cache->wq, &cache->deferred_writethrough_worker); | |
541 | } | |
028ae9f7 | 542 | |
b29d4986 | 543 | static void wake_migration_worker(struct cache *cache) |
c6b4fcba | 544 | { |
b29d4986 JT |
545 | if (passthrough_mode(&cache->features)) |
546 | return; | |
547 | ||
548 | queue_work(cache->wq, &cache->migration_worker); | |
c6b4fcba JT |
549 | } |
550 | ||
551 | /*----------------------------------------------------------------*/ | |
552 | ||
b29d4986 | 553 | static struct dm_bio_prison_cell_v2 *alloc_prison_cell(struct cache *cache) |
c6b4fcba | 554 | { |
b29d4986 | 555 | return dm_bio_prison_alloc_cell_v2(cache->prison, GFP_NOWAIT); |
c6b4fcba JT |
556 | } |
557 | ||
b29d4986 | 558 | static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell_v2 *cell) |
c6b4fcba | 559 | { |
b29d4986 | 560 | dm_bio_prison_free_cell_v2(cache->prison, cell); |
c6b4fcba JT |
561 | } |
562 | ||
a59db676 JT |
563 | static struct dm_cache_migration *alloc_migration(struct cache *cache) |
564 | { | |
565 | struct dm_cache_migration *mg; | |
566 | ||
567 | mg = mempool_alloc(cache->migration_pool, GFP_NOWAIT); | |
568 | if (mg) { | |
569 | mg->cache = cache; | |
570 | atomic_inc(&mg->cache->nr_allocated_migrations); | |
571 | } | |
572 | ||
573 | return mg; | |
574 | } | |
575 | ||
576 | static void free_migration(struct dm_cache_migration *mg) | |
577 | { | |
88bf5184 | 578 | struct cache *cache = mg->cache; |
a59db676 | 579 | |
88bf5184 JT |
580 | if (atomic_dec_and_test(&cache->nr_allocated_migrations)) |
581 | wake_up(&cache->migration_wait); | |
582 | ||
583 | mempool_free(mg, cache->migration_pool); | |
a59db676 JT |
584 | } |
585 | ||
b29d4986 | 586 | /*----------------------------------------------------------------*/ |
c6b4fcba | 587 | |
b29d4986 | 588 | static inline dm_oblock_t oblock_succ(dm_oblock_t b) |
c6b4fcba | 589 | { |
b29d4986 | 590 | return to_oblock(from_oblock(b) + 1ull); |
c6b4fcba JT |
591 | } |
592 | ||
b29d4986 | 593 | static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key_v2 *key) |
c6b4fcba | 594 | { |
b29d4986 JT |
595 | key->virtual = 0; |
596 | key->dev = 0; | |
597 | key->block_begin = from_oblock(begin); | |
598 | key->block_end = from_oblock(end); | |
c6b4fcba JT |
599 | } |
600 | ||
601 | /* | |
b29d4986 JT |
602 | * We have two lock levels. Level 0, which is used to prevent WRITEs, and |
603 | * level 1 which prevents *both* READs and WRITEs. | |
c6b4fcba | 604 | */ |
b29d4986 JT |
605 | #define WRITE_LOCK_LEVEL 0 |
606 | #define READ_WRITE_LOCK_LEVEL 1 | |
607 | ||
608 | static unsigned lock_level(struct bio *bio) | |
c6b4fcba | 609 | { |
b29d4986 JT |
610 | return bio_data_dir(bio) == WRITE ? |
611 | WRITE_LOCK_LEVEL : | |
612 | READ_WRITE_LOCK_LEVEL; | |
613 | } | |
c6b4fcba | 614 | |
b29d4986 JT |
615 | /*---------------------------------------------------------------- |
616 | * Per bio data | |
617 | *--------------------------------------------------------------*/ | |
c6b4fcba | 618 | |
b29d4986 JT |
619 | /* |
620 | * If using writeback, leave out struct per_bio_data's writethrough fields. | |
621 | */ | |
622 | #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache)) | |
623 | #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data)) | |
c6b4fcba | 624 | |
b29d4986 JT |
625 | static size_t get_per_bio_data_size(struct cache *cache) |
626 | { | |
627 | return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB; | |
c6b4fcba JT |
628 | } |
629 | ||
b29d4986 | 630 | static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size) |
c6b4fcba | 631 | { |
b29d4986 JT |
632 | struct per_bio_data *pb = dm_per_bio_data(bio, data_size); |
633 | BUG_ON(!pb); | |
634 | return pb; | |
635 | } | |
c6b4fcba | 636 | |
b29d4986 JT |
637 | static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size) |
638 | { | |
639 | struct per_bio_data *pb = get_per_bio_data(bio, data_size); | |
c6b4fcba | 640 | |
b29d4986 JT |
641 | pb->tick = false; |
642 | pb->req_nr = dm_bio_get_target_bio_nr(bio); | |
643 | pb->cell = NULL; | |
644 | pb->len = 0; | |
645 | ||
646 | return pb; | |
c6b4fcba JT |
647 | } |
648 | ||
649 | /*----------------------------------------------------------------*/ | |
650 | ||
b29d4986 | 651 | static void defer_bio(struct cache *cache, struct bio *bio) |
c6b4fcba | 652 | { |
b29d4986 | 653 | unsigned long flags; |
c6b4fcba | 654 | |
b29d4986 JT |
655 | spin_lock_irqsave(&cache->lock, flags); |
656 | bio_list_add(&cache->deferred_bios, bio); | |
657 | spin_unlock_irqrestore(&cache->lock, flags); | |
658 | ||
659 | wake_deferred_bio_worker(cache); | |
660 | } | |
c6b4fcba | 661 | |
b29d4986 | 662 | static void defer_bios(struct cache *cache, struct bio_list *bios) |
c6b4fcba | 663 | { |
b29d4986 | 664 | unsigned long flags; |
c6b4fcba | 665 | |
b29d4986 JT |
666 | spin_lock_irqsave(&cache->lock, flags); |
667 | bio_list_merge(&cache->deferred_bios, bios); | |
668 | bio_list_init(bios); | |
669 | spin_unlock_irqrestore(&cache->lock, flags); | |
c6b4fcba | 670 | |
b29d4986 | 671 | wake_deferred_bio_worker(cache); |
c6b4fcba JT |
672 | } |
673 | ||
b29d4986 JT |
674 | /*----------------------------------------------------------------*/ |
675 | ||
676 | static bool bio_detain_shared(struct cache *cache, dm_oblock_t oblock, struct bio *bio) | |
7ae34e77 | 677 | { |
b29d4986 JT |
678 | bool r; |
679 | size_t pb_size; | |
680 | struct per_bio_data *pb; | |
681 | struct dm_cell_key_v2 key; | |
7ae34e77 | 682 | dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL); |
b29d4986 | 683 | struct dm_bio_prison_cell_v2 *cell_prealloc, *cell; |
7ae34e77 | 684 | |
b29d4986 JT |
685 | cell_prealloc = alloc_prison_cell(cache); /* FIXME: allow wait if calling from worker */ |
686 | if (!cell_prealloc) { | |
687 | defer_bio(cache, bio); | |
688 | return false; | |
689 | } | |
690 | ||
691 | build_key(oblock, end, &key); | |
692 | r = dm_cell_get_v2(cache->prison, &key, lock_level(bio), bio, cell_prealloc, &cell); | |
693 | if (!r) { | |
694 | /* | |
695 | * Failed to get the lock. | |
696 | */ | |
697 | free_prison_cell(cache, cell_prealloc); | |
698 | return r; | |
699 | } | |
c6b4fcba | 700 | |
b29d4986 JT |
701 | if (cell != cell_prealloc) |
702 | free_prison_cell(cache, cell_prealloc); | |
c6b4fcba | 703 | |
b29d4986 JT |
704 | pb_size = get_per_bio_data_size(cache); |
705 | pb = get_per_bio_data(bio, pb_size); | |
706 | pb->cell = cell; | |
c6b4fcba JT |
707 | |
708 | return r; | |
709 | } | |
710 | ||
aeed1420 | 711 | /*----------------------------------------------------------------*/ |
c6b4fcba JT |
712 | |
713 | static bool is_dirty(struct cache *cache, dm_cblock_t b) | |
714 | { | |
715 | return test_bit(from_cblock(b), cache->dirty_bitset); | |
716 | } | |
717 | ||
b29d4986 | 718 | static void set_dirty(struct cache *cache, dm_cblock_t cblock) |
c6b4fcba JT |
719 | { |
720 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 721 | atomic_inc(&cache->nr_dirty); |
b29d4986 | 722 | policy_set_dirty(cache->policy, cblock); |
c6b4fcba JT |
723 | } |
724 | } | |
725 | ||
b29d4986 JT |
726 | /* |
727 | * These two are called when setting after migrations to force the policy | |
728 | * and dirty bitset to be in sync. | |
729 | */ | |
730 | static void force_set_dirty(struct cache *cache, dm_cblock_t cblock) | |
731 | { | |
732 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) | |
733 | atomic_inc(&cache->nr_dirty); | |
734 | policy_set_dirty(cache->policy, cblock); | |
735 | } | |
736 | ||
737 | static void force_clear_dirty(struct cache *cache, dm_cblock_t cblock) | |
c6b4fcba JT |
738 | { |
739 | if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 740 | if (atomic_dec_return(&cache->nr_dirty) == 0) |
c6b4fcba JT |
741 | dm_table_event(cache->ti->table); |
742 | } | |
b29d4986 JT |
743 | |
744 | policy_clear_dirty(cache->policy, cblock); | |
c6b4fcba JT |
745 | } |
746 | ||
747 | /*----------------------------------------------------------------*/ | |
aeed1420 | 748 | |
c6b4fcba JT |
749 | static bool block_size_is_power_of_two(struct cache *cache) |
750 | { | |
751 | return cache->sectors_per_block_shift >= 0; | |
752 | } | |
753 | ||
43aeaa29 MP |
754 | /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */ |
755 | #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6 | |
756 | __always_inline | |
757 | #endif | |
414dd67d JT |
758 | static dm_block_t block_div(dm_block_t b, uint32_t n) |
759 | { | |
760 | do_div(b, n); | |
761 | ||
762 | return b; | |
763 | } | |
764 | ||
7ae34e77 | 765 | static dm_block_t oblocks_per_dblock(struct cache *cache) |
1bad9bc4 | 766 | { |
7ae34e77 | 767 | dm_block_t oblocks = cache->discard_block_size; |
1bad9bc4 | 768 | |
7ae34e77 JT |
769 | if (block_size_is_power_of_two(cache)) |
770 | oblocks >>= cache->sectors_per_block_shift; | |
1bad9bc4 | 771 | else |
7ae34e77 | 772 | oblocks = block_div(oblocks, cache->sectors_per_block); |
1bad9bc4 | 773 | |
7ae34e77 JT |
774 | return oblocks; |
775 | } | |
776 | ||
777 | static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock) | |
778 | { | |
779 | return to_dblock(block_div(from_oblock(oblock), | |
780 | oblocks_per_dblock(cache))); | |
781 | } | |
1bad9bc4 | 782 | |
1bad9bc4 | 783 | static void set_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
784 | { |
785 | unsigned long flags; | |
786 | ||
7ae34e77 | 787 | BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
788 | atomic_inc(&cache->stats.discard_count); |
789 | ||
790 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 791 | set_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
792 | spin_unlock_irqrestore(&cache->lock, flags); |
793 | } | |
794 | ||
1bad9bc4 | 795 | static void clear_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
796 | { |
797 | unsigned long flags; | |
798 | ||
799 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 800 | clear_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
801 | spin_unlock_irqrestore(&cache->lock, flags); |
802 | } | |
803 | ||
1bad9bc4 | 804 | static bool is_discarded(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
805 | { |
806 | int r; | |
807 | unsigned long flags; | |
808 | ||
809 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 810 | r = test_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
811 | spin_unlock_irqrestore(&cache->lock, flags); |
812 | ||
813 | return r; | |
814 | } | |
815 | ||
816 | static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b) | |
817 | { | |
818 | int r; | |
819 | unsigned long flags; | |
820 | ||
821 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 JT |
822 | r = test_bit(from_dblock(oblock_to_dblock(cache, b)), |
823 | cache->discard_bitset); | |
c6b4fcba JT |
824 | spin_unlock_irqrestore(&cache->lock, flags); |
825 | ||
826 | return r; | |
827 | } | |
828 | ||
b29d4986 JT |
829 | /*---------------------------------------------------------------- |
830 | * Remapping | |
831 | *--------------------------------------------------------------*/ | |
832 | static void remap_to_origin(struct cache *cache, struct bio *bio) | |
c6b4fcba | 833 | { |
b29d4986 | 834 | bio->bi_bdev = cache->origin_dev->bdev; |
c6b4fcba JT |
835 | } |
836 | ||
837 | static void remap_to_cache(struct cache *cache, struct bio *bio, | |
838 | dm_cblock_t cblock) | |
839 | { | |
4f024f37 | 840 | sector_t bi_sector = bio->bi_iter.bi_sector; |
e0d849fa | 841 | sector_t block = from_cblock(cblock); |
c6b4fcba JT |
842 | |
843 | bio->bi_bdev = cache->cache_dev->bdev; | |
844 | if (!block_size_is_power_of_two(cache)) | |
4f024f37 | 845 | bio->bi_iter.bi_sector = |
e0d849fa | 846 | (block * cache->sectors_per_block) + |
4f024f37 | 847 | sector_div(bi_sector, cache->sectors_per_block); |
c6b4fcba | 848 | else |
4f024f37 | 849 | bio->bi_iter.bi_sector = |
e0d849fa | 850 | (block << cache->sectors_per_block_shift) | |
4f024f37 | 851 | (bi_sector & (cache->sectors_per_block - 1)); |
c6b4fcba JT |
852 | } |
853 | ||
854 | static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio) | |
855 | { | |
856 | unsigned long flags; | |
19b0092e MS |
857 | size_t pb_data_size = get_per_bio_data_size(cache); |
858 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
859 | |
860 | spin_lock_irqsave(&cache->lock, flags); | |
f73f44eb | 861 | if (cache->need_tick_bio && !op_is_flush(bio->bi_opf) && |
e6047149 | 862 | bio_op(bio) != REQ_OP_DISCARD) { |
c6b4fcba JT |
863 | pb->tick = true; |
864 | cache->need_tick_bio = false; | |
865 | } | |
866 | spin_unlock_irqrestore(&cache->lock, flags); | |
867 | } | |
868 | ||
869 | static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, | |
b29d4986 | 870 | dm_oblock_t oblock) |
c6b4fcba | 871 | { |
b29d4986 | 872 | // FIXME: this is called way too much. |
c6b4fcba JT |
873 | check_if_tick_bio_needed(cache, bio); |
874 | remap_to_origin(cache, bio); | |
875 | if (bio_data_dir(bio) == WRITE) | |
1bad9bc4 | 876 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
877 | } |
878 | ||
879 | static void remap_to_cache_dirty(struct cache *cache, struct bio *bio, | |
880 | dm_oblock_t oblock, dm_cblock_t cblock) | |
881 | { | |
f8e5f01a | 882 | check_if_tick_bio_needed(cache, bio); |
c6b4fcba JT |
883 | remap_to_cache(cache, bio, cblock); |
884 | if (bio_data_dir(bio) == WRITE) { | |
b29d4986 | 885 | set_dirty(cache, cblock); |
1bad9bc4 | 886 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
887 | } |
888 | } | |
889 | ||
890 | static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio) | |
891 | { | |
4f024f37 | 892 | sector_t block_nr = bio->bi_iter.bi_sector; |
c6b4fcba JT |
893 | |
894 | if (!block_size_is_power_of_two(cache)) | |
895 | (void) sector_div(block_nr, cache->sectors_per_block); | |
896 | else | |
897 | block_nr >>= cache->sectors_per_block_shift; | |
898 | ||
899 | return to_oblock(block_nr); | |
900 | } | |
901 | ||
066dbaa3 JT |
902 | static bool accountable_bio(struct cache *cache, struct bio *bio) |
903 | { | |
904 | return ((bio->bi_bdev == cache->origin_dev->bdev) && | |
e6047149 | 905 | bio_op(bio) != REQ_OP_DISCARD); |
066dbaa3 JT |
906 | } |
907 | ||
908 | static void accounted_begin(struct cache *cache, struct bio *bio) | |
909 | { | |
910 | size_t pb_data_size = get_per_bio_data_size(cache); | |
911 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
912 | ||
913 | if (accountable_bio(cache, bio)) { | |
914 | pb->len = bio_sectors(bio); | |
915 | iot_io_begin(&cache->origin_tracker, pb->len); | |
916 | } | |
917 | } | |
918 | ||
919 | static void accounted_complete(struct cache *cache, struct bio *bio) | |
920 | { | |
921 | size_t pb_data_size = get_per_bio_data_size(cache); | |
922 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
923 | ||
924 | iot_io_end(&cache->origin_tracker, pb->len); | |
925 | } | |
926 | ||
927 | static void accounted_request(struct cache *cache, struct bio *bio) | |
928 | { | |
929 | accounted_begin(cache, bio); | |
930 | generic_make_request(bio); | |
931 | } | |
932 | ||
b29d4986 | 933 | static void issue_op(struct bio *bio, void *context) |
8c081b52 | 934 | { |
b29d4986 JT |
935 | struct cache *cache = context; |
936 | accounted_request(cache, bio); | |
8c081b52 JT |
937 | } |
938 | ||
e2e74d61 JT |
939 | static void defer_writethrough_bio(struct cache *cache, struct bio *bio) |
940 | { | |
941 | unsigned long flags; | |
942 | ||
943 | spin_lock_irqsave(&cache->lock, flags); | |
944 | bio_list_add(&cache->deferred_writethrough_bios, bio); | |
945 | spin_unlock_irqrestore(&cache->lock, flags); | |
946 | ||
b29d4986 | 947 | wake_deferred_writethrough_worker(cache); |
e2e74d61 JT |
948 | } |
949 | ||
4246a0b6 | 950 | static void writethrough_endio(struct bio *bio) |
e2e74d61 | 951 | { |
19b0092e | 952 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
c9d28d5d JT |
953 | |
954 | dm_unhook_bio(&pb->hook_info, bio); | |
e2e74d61 | 955 | |
4246a0b6 CH |
956 | if (bio->bi_error) { |
957 | bio_endio(bio); | |
e2e74d61 JT |
958 | return; |
959 | } | |
960 | ||
b844fe69 | 961 | dm_bio_restore(&pb->bio_details, bio); |
e2e74d61 JT |
962 | remap_to_cache(pb->cache, bio, pb->cblock); |
963 | ||
964 | /* | |
965 | * We can't issue this bio directly, since we're in interrupt | |
aeed1420 | 966 | * context. So it gets put on a bio list for processing by the |
e2e74d61 JT |
967 | * worker thread. |
968 | */ | |
969 | defer_writethrough_bio(pb->cache, bio); | |
970 | } | |
971 | ||
972 | /* | |
b29d4986 | 973 | * FIXME: send in parallel, huge latency as is. |
e2e74d61 JT |
974 | * When running in writethrough mode we need to send writes to clean blocks |
975 | * to both the cache and origin devices. In future we'd like to clone the | |
976 | * bio and send them in parallel, but for now we're doing them in | |
977 | * series as this is easier. | |
978 | */ | |
979 | static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio, | |
980 | dm_oblock_t oblock, dm_cblock_t cblock) | |
981 | { | |
19b0092e | 982 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
e2e74d61 JT |
983 | |
984 | pb->cache = cache; | |
985 | pb->cblock = cblock; | |
c9d28d5d | 986 | dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL); |
b844fe69 | 987 | dm_bio_record(&pb->bio_details, bio); |
e2e74d61 JT |
988 | |
989 | remap_to_origin_clear_discard(pb->cache, bio, oblock); | |
990 | } | |
991 | ||
028ae9f7 JT |
992 | /*---------------------------------------------------------------- |
993 | * Failure modes | |
994 | *--------------------------------------------------------------*/ | |
995 | static enum cache_metadata_mode get_cache_mode(struct cache *cache) | |
996 | { | |
997 | return cache->features.mode; | |
998 | } | |
999 | ||
b61d9509 MS |
1000 | static const char *cache_device_name(struct cache *cache) |
1001 | { | |
1002 | return dm_device_name(dm_table_get_md(cache->ti->table)); | |
1003 | } | |
1004 | ||
028ae9f7 JT |
1005 | static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode) |
1006 | { | |
1007 | const char *descs[] = { | |
1008 | "write", | |
1009 | "read-only", | |
1010 | "fail" | |
1011 | }; | |
1012 | ||
1013 | dm_table_event(cache->ti->table); | |
b61d9509 MS |
1014 | DMINFO("%s: switching cache to %s mode", |
1015 | cache_device_name(cache), descs[(int)mode]); | |
028ae9f7 JT |
1016 | } |
1017 | ||
1018 | static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode) | |
1019 | { | |
d14fcf3d | 1020 | bool needs_check; |
028ae9f7 JT |
1021 | enum cache_metadata_mode old_mode = get_cache_mode(cache); |
1022 | ||
d14fcf3d | 1023 | if (dm_cache_metadata_needs_check(cache->cmd, &needs_check)) { |
23cab26d MS |
1024 | DMERR("%s: unable to read needs_check flag, setting failure mode.", |
1025 | cache_device_name(cache)); | |
d14fcf3d JT |
1026 | new_mode = CM_FAIL; |
1027 | } | |
1028 | ||
028ae9f7 | 1029 | if (new_mode == CM_WRITE && needs_check) { |
b61d9509 MS |
1030 | DMERR("%s: unable to switch cache to write mode until repaired.", |
1031 | cache_device_name(cache)); | |
028ae9f7 JT |
1032 | if (old_mode != new_mode) |
1033 | new_mode = old_mode; | |
1034 | else | |
1035 | new_mode = CM_READ_ONLY; | |
1036 | } | |
1037 | ||
1038 | /* Never move out of fail mode */ | |
1039 | if (old_mode == CM_FAIL) | |
1040 | new_mode = CM_FAIL; | |
1041 | ||
1042 | switch (new_mode) { | |
1043 | case CM_FAIL: | |
1044 | case CM_READ_ONLY: | |
1045 | dm_cache_metadata_set_read_only(cache->cmd); | |
1046 | break; | |
1047 | ||
1048 | case CM_WRITE: | |
1049 | dm_cache_metadata_set_read_write(cache->cmd); | |
1050 | break; | |
1051 | } | |
1052 | ||
1053 | cache->features.mode = new_mode; | |
1054 | ||
1055 | if (new_mode != old_mode) | |
1056 | notify_mode_switch(cache, new_mode); | |
1057 | } | |
1058 | ||
1059 | static void abort_transaction(struct cache *cache) | |
1060 | { | |
b61d9509 MS |
1061 | const char *dev_name = cache_device_name(cache); |
1062 | ||
028ae9f7 JT |
1063 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
1064 | return; | |
1065 | ||
1066 | if (dm_cache_metadata_set_needs_check(cache->cmd)) { | |
b61d9509 | 1067 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); |
028ae9f7 JT |
1068 | set_cache_mode(cache, CM_FAIL); |
1069 | } | |
1070 | ||
b61d9509 | 1071 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); |
028ae9f7 | 1072 | if (dm_cache_metadata_abort(cache->cmd)) { |
b61d9509 | 1073 | DMERR("%s: failed to abort metadata transaction", dev_name); |
028ae9f7 JT |
1074 | set_cache_mode(cache, CM_FAIL); |
1075 | } | |
1076 | } | |
1077 | ||
1078 | static void metadata_operation_failed(struct cache *cache, const char *op, int r) | |
1079 | { | |
b61d9509 MS |
1080 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1081 | cache_device_name(cache), op, r); | |
028ae9f7 JT |
1082 | abort_transaction(cache); |
1083 | set_cache_mode(cache, CM_READ_ONLY); | |
1084 | } | |
1085 | ||
b29d4986 JT |
1086 | /*----------------------------------------------------------------*/ |
1087 | ||
1088 | static void load_stats(struct cache *cache) | |
1089 | { | |
1090 | struct dm_cache_statistics stats; | |
1091 | ||
1092 | dm_cache_metadata_get_stats(cache->cmd, &stats); | |
1093 | atomic_set(&cache->stats.read_hit, stats.read_hits); | |
1094 | atomic_set(&cache->stats.read_miss, stats.read_misses); | |
1095 | atomic_set(&cache->stats.write_hit, stats.write_hits); | |
1096 | atomic_set(&cache->stats.write_miss, stats.write_misses); | |
1097 | } | |
1098 | ||
1099 | static void save_stats(struct cache *cache) | |
1100 | { | |
1101 | struct dm_cache_statistics stats; | |
1102 | ||
1103 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
1104 | return; | |
1105 | ||
1106 | stats.read_hits = atomic_read(&cache->stats.read_hit); | |
1107 | stats.read_misses = atomic_read(&cache->stats.read_miss); | |
1108 | stats.write_hits = atomic_read(&cache->stats.write_hit); | |
1109 | stats.write_misses = atomic_read(&cache->stats.write_miss); | |
1110 | ||
1111 | dm_cache_metadata_set_stats(cache->cmd, &stats); | |
1112 | } | |
1113 | ||
1114 | static void update_stats(struct cache_stats *stats, enum policy_operation op) | |
1115 | { | |
1116 | switch (op) { | |
1117 | case POLICY_PROMOTE: | |
1118 | atomic_inc(&stats->promotion); | |
1119 | break; | |
1120 | ||
1121 | case POLICY_DEMOTE: | |
1122 | atomic_inc(&stats->demotion); | |
1123 | break; | |
1124 | ||
1125 | case POLICY_WRITEBACK: | |
1126 | atomic_inc(&stats->writeback); | |
1127 | break; | |
1128 | } | |
1129 | } | |
1130 | ||
c6b4fcba JT |
1131 | /*---------------------------------------------------------------- |
1132 | * Migration processing | |
1133 | * | |
1134 | * Migration covers moving data from the origin device to the cache, or | |
1135 | * vice versa. | |
1136 | *--------------------------------------------------------------*/ | |
b29d4986 | 1137 | |
a59db676 | 1138 | static void inc_io_migrations(struct cache *cache) |
c6b4fcba | 1139 | { |
a59db676 | 1140 | atomic_inc(&cache->nr_io_migrations); |
c6b4fcba JT |
1141 | } |
1142 | ||
a59db676 | 1143 | static void dec_io_migrations(struct cache *cache) |
c6b4fcba | 1144 | { |
a59db676 | 1145 | atomic_dec(&cache->nr_io_migrations); |
c6b4fcba JT |
1146 | } |
1147 | ||
651f5fa2 JT |
1148 | static bool discard_or_flush(struct bio *bio) |
1149 | { | |
f73f44eb | 1150 | return bio_op(bio) == REQ_OP_DISCARD || op_is_flush(bio->bi_opf); |
651f5fa2 JT |
1151 | } |
1152 | ||
b29d4986 JT |
1153 | static void calc_discard_block_range(struct cache *cache, struct bio *bio, |
1154 | dm_dblock_t *b, dm_dblock_t *e) | |
c6b4fcba | 1155 | { |
b29d4986 JT |
1156 | sector_t sb = bio->bi_iter.bi_sector; |
1157 | sector_t se = bio_end_sector(bio); | |
651f5fa2 | 1158 | |
b29d4986 | 1159 | *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size)); |
c6b4fcba | 1160 | |
b29d4986 JT |
1161 | if (se - sb < cache->discard_block_size) |
1162 | *e = *b; | |
1163 | else | |
1164 | *e = to_dblock(block_div(se, cache->discard_block_size)); | |
c6b4fcba JT |
1165 | } |
1166 | ||
b29d4986 | 1167 | /*----------------------------------------------------------------*/ |
651f5fa2 | 1168 | |
b29d4986 | 1169 | static void prevent_background_work(struct cache *cache) |
651f5fa2 | 1170 | { |
b29d4986 JT |
1171 | lockdep_off(); |
1172 | down_write(&cache->background_work_lock); | |
1173 | lockdep_on(); | |
651f5fa2 JT |
1174 | } |
1175 | ||
b29d4986 | 1176 | static void allow_background_work(struct cache *cache) |
c6b4fcba | 1177 | { |
b29d4986 JT |
1178 | lockdep_off(); |
1179 | up_write(&cache->background_work_lock); | |
1180 | lockdep_on(); | |
c6b4fcba JT |
1181 | } |
1182 | ||
b29d4986 | 1183 | static bool background_work_begin(struct cache *cache) |
c6b4fcba | 1184 | { |
b29d4986 | 1185 | bool r; |
c6b4fcba | 1186 | |
b29d4986 JT |
1187 | lockdep_off(); |
1188 | r = down_read_trylock(&cache->background_work_lock); | |
1189 | lockdep_on(); | |
c6b4fcba | 1190 | |
b29d4986 | 1191 | return r; |
c6b4fcba JT |
1192 | } |
1193 | ||
b29d4986 | 1194 | static void background_work_end(struct cache *cache) |
c6b4fcba | 1195 | { |
b29d4986 JT |
1196 | lockdep_off(); |
1197 | up_read(&cache->background_work_lock); | |
1198 | lockdep_on(); | |
1199 | } | |
c6b4fcba | 1200 | |
b29d4986 | 1201 | /*----------------------------------------------------------------*/ |
c6b4fcba | 1202 | |
b29d4986 JT |
1203 | static void quiesce(struct dm_cache_migration *mg, |
1204 | void (*continuation)(struct work_struct *)) | |
1205 | { | |
1206 | init_continuation(&mg->k, continuation); | |
1207 | dm_cell_quiesce_v2(mg->cache->prison, mg->cell, &mg->k.ws); | |
c6b4fcba JT |
1208 | } |
1209 | ||
b29d4986 | 1210 | static struct dm_cache_migration *ws_to_mg(struct work_struct *ws) |
c6b4fcba | 1211 | { |
b29d4986 JT |
1212 | struct continuation *k = container_of(ws, struct continuation, ws); |
1213 | return container_of(k, struct dm_cache_migration, k); | |
c6b4fcba JT |
1214 | } |
1215 | ||
1216 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
1217 | { | |
b29d4986 | 1218 | struct dm_cache_migration *mg = container_of(context, struct dm_cache_migration, k); |
c6b4fcba JT |
1219 | |
1220 | if (read_err || write_err) | |
b29d4986 | 1221 | mg->k.input = -EIO; |
c6b4fcba | 1222 | |
b29d4986 | 1223 | queue_continuation(mg->cache->wq, &mg->k); |
c6b4fcba JT |
1224 | } |
1225 | ||
b29d4986 | 1226 | static int copy(struct dm_cache_migration *mg, bool promote) |
c6b4fcba JT |
1227 | { |
1228 | int r; | |
1229 | struct dm_io_region o_region, c_region; | |
1230 | struct cache *cache = mg->cache; | |
1231 | ||
1232 | o_region.bdev = cache->origin_dev->bdev; | |
b29d4986 | 1233 | o_region.sector = from_oblock(mg->op->oblock) * cache->sectors_per_block; |
c6b4fcba JT |
1234 | o_region.count = cache->sectors_per_block; |
1235 | ||
1236 | c_region.bdev = cache->cache_dev->bdev; | |
b29d4986 | 1237 | c_region.sector = from_cblock(mg->op->cblock) * cache->sectors_per_block; |
c6b4fcba JT |
1238 | c_region.count = cache->sectors_per_block; |
1239 | ||
b29d4986 JT |
1240 | if (promote) |
1241 | r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, &mg->k); | |
1242 | else | |
1243 | r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, &mg->k); | |
c6b4fcba | 1244 | |
b29d4986 JT |
1245 | return r; |
1246 | } | |
1247 | ||
1248 | static void bio_drop_shared_lock(struct cache *cache, struct bio *bio) | |
1249 | { | |
1250 | size_t pb_data_size = get_per_bio_data_size(cache); | |
1251 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1252 | ||
1253 | if (pb->cell && dm_cell_put_v2(cache->prison, pb->cell)) | |
1254 | free_prison_cell(cache, pb->cell); | |
1255 | pb->cell = NULL; | |
c6b4fcba JT |
1256 | } |
1257 | ||
4246a0b6 | 1258 | static void overwrite_endio(struct bio *bio) |
c9d28d5d JT |
1259 | { |
1260 | struct dm_cache_migration *mg = bio->bi_private; | |
1261 | struct cache *cache = mg->cache; | |
1262 | size_t pb_data_size = get_per_bio_data_size(cache); | |
1263 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c9d28d5d | 1264 | |
80ae49aa MS |
1265 | dm_unhook_bio(&pb->hook_info, bio); |
1266 | ||
4246a0b6 | 1267 | if (bio->bi_error) |
b29d4986 | 1268 | mg->k.input = bio->bi_error; |
80ae49aa | 1269 | |
b29d4986 | 1270 | queue_continuation(mg->cache->wq, &mg->k); |
c9d28d5d JT |
1271 | } |
1272 | ||
b29d4986 JT |
1273 | static void overwrite(struct dm_cache_migration *mg, |
1274 | void (*continuation)(struct work_struct *)) | |
c9d28d5d | 1275 | { |
b29d4986 | 1276 | struct bio *bio = mg->overwrite_bio; |
c9d28d5d JT |
1277 | size_t pb_data_size = get_per_bio_data_size(mg->cache); |
1278 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1279 | ||
1280 | dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg); | |
8c081b52 JT |
1281 | |
1282 | /* | |
b29d4986 JT |
1283 | * The overwrite bio is part of the copy operation, as such it does |
1284 | * not set/clear discard or dirty flags. | |
8c081b52 | 1285 | */ |
b29d4986 JT |
1286 | if (mg->op->op == POLICY_PROMOTE) |
1287 | remap_to_cache(mg->cache, bio, mg->op->cblock); | |
1288 | else | |
1289 | remap_to_origin(mg->cache, bio); | |
1290 | ||
1291 | init_continuation(&mg->k, continuation); | |
066dbaa3 | 1292 | accounted_request(mg->cache, bio); |
c9d28d5d JT |
1293 | } |
1294 | ||
b29d4986 JT |
1295 | /* |
1296 | * Migration steps: | |
1297 | * | |
1298 | * 1) exclusive lock preventing WRITEs | |
1299 | * 2) quiesce | |
1300 | * 3) copy or issue overwrite bio | |
1301 | * 4) upgrade to exclusive lock preventing READs and WRITEs | |
1302 | * 5) quiesce | |
1303 | * 6) update metadata and commit | |
1304 | * 7) unlock | |
1305 | */ | |
1306 | static void mg_complete(struct dm_cache_migration *mg, bool success) | |
c9d28d5d | 1307 | { |
b29d4986 JT |
1308 | struct bio_list bios; |
1309 | struct cache *cache = mg->cache; | |
1310 | struct policy_work *op = mg->op; | |
1311 | dm_cblock_t cblock = op->cblock; | |
1312 | ||
1313 | if (success) | |
1314 | update_stats(&cache->stats, op->op); | |
1315 | ||
1316 | switch (op->op) { | |
1317 | case POLICY_PROMOTE: | |
1318 | clear_discard(cache, oblock_to_dblock(cache, op->oblock)); | |
1319 | policy_complete_background_work(cache->policy, op, success); | |
1320 | ||
1321 | if (mg->overwrite_bio) { | |
1322 | if (success) | |
1323 | force_set_dirty(cache, cblock); | |
1324 | else | |
1325 | mg->overwrite_bio->bi_error = (mg->k.input ? : -EIO); | |
1326 | bio_endio(mg->overwrite_bio); | |
1327 | } else { | |
1328 | if (success) | |
1329 | force_clear_dirty(cache, cblock); | |
1330 | dec_io_migrations(cache); | |
1331 | } | |
1332 | break; | |
1333 | ||
1334 | case POLICY_DEMOTE: | |
1335 | /* | |
1336 | * We clear dirty here to update the nr_dirty counter. | |
1337 | */ | |
1338 | if (success) | |
1339 | force_clear_dirty(cache, cblock); | |
1340 | policy_complete_background_work(cache->policy, op, success); | |
1341 | dec_io_migrations(cache); | |
1342 | break; | |
1343 | ||
1344 | case POLICY_WRITEBACK: | |
1345 | if (success) | |
1346 | force_clear_dirty(cache, cblock); | |
1347 | policy_complete_background_work(cache->policy, op, success); | |
1348 | dec_io_migrations(cache); | |
1349 | break; | |
1350 | } | |
1351 | ||
1352 | bio_list_init(&bios); | |
1353 | if (mg->cell) { | |
1354 | if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios)) | |
1355 | free_prison_cell(cache, mg->cell); | |
1356 | } | |
1357 | ||
1358 | free_migration(mg); | |
1359 | defer_bios(cache, &bios); | |
1360 | wake_migration_worker(cache); | |
1361 | ||
1362 | background_work_end(cache); | |
c9d28d5d JT |
1363 | } |
1364 | ||
b29d4986 | 1365 | static void mg_success(struct work_struct *ws) |
c6b4fcba | 1366 | { |
b29d4986 JT |
1367 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1368 | mg_complete(mg, mg->k.input == 0); | |
c6b4fcba JT |
1369 | } |
1370 | ||
b29d4986 | 1371 | static void mg_update_metadata(struct work_struct *ws) |
7ae34e77 | 1372 | { |
b29d4986 JT |
1373 | int r; |
1374 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
c6b4fcba | 1375 | struct cache *cache = mg->cache; |
b29d4986 | 1376 | struct policy_work *op = mg->op; |
c6b4fcba | 1377 | |
b29d4986 JT |
1378 | switch (op->op) { |
1379 | case POLICY_PROMOTE: | |
1380 | r = dm_cache_insert_mapping(cache->cmd, op->cblock, op->oblock); | |
1381 | if (r) { | |
1382 | DMERR_LIMIT("%s: migration failed; couldn't insert mapping", | |
1383 | cache_device_name(cache)); | |
1384 | metadata_operation_failed(cache, "dm_cache_insert_mapping", r); | |
7ae34e77 | 1385 | |
b29d4986 JT |
1386 | mg_complete(mg, false); |
1387 | return; | |
1388 | } | |
1389 | mg_complete(mg, true); | |
1390 | break; | |
c9d28d5d | 1391 | |
b29d4986 JT |
1392 | case POLICY_DEMOTE: |
1393 | r = dm_cache_remove_mapping(cache->cmd, op->cblock); | |
1394 | if (r) { | |
1395 | DMERR_LIMIT("%s: migration failed; couldn't update on disk metadata", | |
1396 | cache_device_name(cache)); | |
1397 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); | |
c6b4fcba | 1398 | |
b29d4986 | 1399 | mg_complete(mg, false); |
c9d28d5d JT |
1400 | return; |
1401 | } | |
c9d28d5d | 1402 | |
b29d4986 JT |
1403 | /* |
1404 | * It would be nice if we only had to commit when a REQ_FLUSH | |
1405 | * comes through. But there's one scenario that we have to | |
1406 | * look out for: | |
1407 | * | |
1408 | * - vblock x in a cache block | |
1409 | * - domotion occurs | |
1410 | * - cache block gets reallocated and over written | |
1411 | * - crash | |
1412 | * | |
1413 | * When we recover, because there was no commit the cache will | |
1414 | * rollback to having the data for vblock x in the cache block. | |
1415 | * But the cache block has since been overwritten, so it'll end | |
1416 | * up pointing to data that was never in 'x' during the history | |
1417 | * of the device. | |
1418 | * | |
1419 | * To avoid this issue we require a commit as part of the | |
1420 | * demotion operation. | |
1421 | */ | |
1422 | init_continuation(&mg->k, mg_success); | |
1423 | continue_after_commit(&cache->committer, &mg->k); | |
1424 | schedule_commit(&cache->committer); | |
1425 | break; | |
1426 | ||
1427 | case POLICY_WRITEBACK: | |
1428 | mg_complete(mg, true); | |
1429 | break; | |
7ae34e77 | 1430 | } |
c6b4fcba JT |
1431 | } |
1432 | ||
b29d4986 | 1433 | static void mg_update_metadata_after_copy(struct work_struct *ws) |
c6b4fcba | 1434 | { |
b29d4986 JT |
1435 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1436 | ||
1437 | /* | |
1438 | * Did the copy succeed? | |
1439 | */ | |
1440 | if (mg->k.input) | |
1441 | mg_complete(mg, false); | |
c6b4fcba | 1442 | else |
b29d4986 | 1443 | mg_update_metadata(ws); |
c6b4fcba JT |
1444 | } |
1445 | ||
b29d4986 | 1446 | static void mg_upgrade_lock(struct work_struct *ws) |
c6b4fcba | 1447 | { |
b29d4986 JT |
1448 | int r; |
1449 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
c6b4fcba | 1450 | |
b29d4986 JT |
1451 | /* |
1452 | * Did the copy succeed? | |
1453 | */ | |
1454 | if (mg->k.input) | |
1455 | mg_complete(mg, false); | |
c6b4fcba | 1456 | |
c9d28d5d | 1457 | else { |
b29d4986 JT |
1458 | /* |
1459 | * Now we want the lock to prevent both reads and writes. | |
1460 | */ | |
1461 | r = dm_cell_lock_promote_v2(mg->cache->prison, mg->cell, | |
1462 | READ_WRITE_LOCK_LEVEL); | |
1463 | if (r < 0) | |
1464 | mg_complete(mg, false); | |
c6b4fcba | 1465 | |
b29d4986 JT |
1466 | else if (r) |
1467 | quiesce(mg, mg_update_metadata); | |
c6b4fcba | 1468 | |
b29d4986 JT |
1469 | else |
1470 | mg_update_metadata(ws); | |
c9d28d5d | 1471 | } |
c6b4fcba JT |
1472 | } |
1473 | ||
b29d4986 | 1474 | static void mg_copy(struct work_struct *ws) |
c6b4fcba | 1475 | { |
b29d4986 JT |
1476 | int r; |
1477 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
c6b4fcba | 1478 | |
b29d4986 JT |
1479 | if (mg->overwrite_bio) { |
1480 | /* | |
1481 | * It's safe to do this here, even though it's new data | |
1482 | * because all IO has been locked out of the block. | |
1483 | * | |
1484 | * mg_lock_writes() already took READ_WRITE_LOCK_LEVEL | |
1485 | * so _not_ using mg_upgrade_lock() as continutation. | |
1486 | */ | |
1487 | overwrite(mg, mg_update_metadata_after_copy); | |
c6b4fcba | 1488 | |
b29d4986 JT |
1489 | } else { |
1490 | struct cache *cache = mg->cache; | |
1491 | struct policy_work *op = mg->op; | |
1492 | bool is_policy_promote = (op->op == POLICY_PROMOTE); | |
c6b4fcba | 1493 | |
b29d4986 JT |
1494 | if ((!is_policy_promote && !is_dirty(cache, op->cblock)) || |
1495 | is_discarded_oblock(cache, op->oblock)) { | |
1496 | mg_upgrade_lock(ws); | |
1497 | return; | |
1498 | } | |
c6b4fcba | 1499 | |
b29d4986 | 1500 | init_continuation(&mg->k, mg_upgrade_lock); |
c6b4fcba | 1501 | |
b29d4986 JT |
1502 | r = copy(mg, is_policy_promote); |
1503 | if (r) { | |
1504 | DMERR_LIMIT("%s: migration copy failed", cache_device_name(cache)); | |
1505 | mg->k.input = -EIO; | |
1506 | mg_complete(mg, false); | |
1507 | } | |
1508 | } | |
c6b4fcba JT |
1509 | } |
1510 | ||
b29d4986 | 1511 | static int mg_lock_writes(struct dm_cache_migration *mg) |
c6b4fcba | 1512 | { |
b29d4986 JT |
1513 | int r; |
1514 | struct dm_cell_key_v2 key; | |
c6b4fcba | 1515 | struct cache *cache = mg->cache; |
b29d4986 | 1516 | struct dm_bio_prison_cell_v2 *prealloc; |
c6b4fcba | 1517 | |
b29d4986 JT |
1518 | prealloc = alloc_prison_cell(cache); |
1519 | if (!prealloc) { | |
1520 | DMERR_LIMIT("%s: alloc_prison_cell failed", cache_device_name(cache)); | |
1521 | mg_complete(mg, false); | |
1522 | return -ENOMEM; | |
1523 | } | |
c6b4fcba | 1524 | |
b29d4986 JT |
1525 | /* |
1526 | * Prevent writes to the block, but allow reads to continue. | |
1527 | * Unless we're using an overwrite bio, in which case we lock | |
1528 | * everything. | |
1529 | */ | |
1530 | build_key(mg->op->oblock, oblock_succ(mg->op->oblock), &key); | |
1531 | r = dm_cell_lock_v2(cache->prison, &key, | |
1532 | mg->overwrite_bio ? READ_WRITE_LOCK_LEVEL : WRITE_LOCK_LEVEL, | |
1533 | prealloc, &mg->cell); | |
1534 | if (r < 0) { | |
1535 | free_prison_cell(cache, prealloc); | |
1536 | mg_complete(mg, false); | |
1537 | return r; | |
1538 | } | |
c6b4fcba | 1539 | |
b29d4986 JT |
1540 | if (mg->cell != prealloc) |
1541 | free_prison_cell(cache, prealloc); | |
c6b4fcba | 1542 | |
b29d4986 JT |
1543 | if (r == 0) |
1544 | mg_copy(&mg->k.ws); | |
1545 | else | |
1546 | quiesce(mg, mg_copy); | |
c6b4fcba | 1547 | |
b29d4986 | 1548 | return 0; |
c6b4fcba JT |
1549 | } |
1550 | ||
b29d4986 | 1551 | static int mg_start(struct cache *cache, struct policy_work *op, struct bio *bio) |
c6b4fcba | 1552 | { |
b29d4986 | 1553 | struct dm_cache_migration *mg; |
c6b4fcba | 1554 | |
b29d4986 JT |
1555 | if (!background_work_begin(cache)) { |
1556 | policy_complete_background_work(cache->policy, op, false); | |
1557 | return -EPERM; | |
1558 | } | |
2ee57d58 | 1559 | |
b29d4986 JT |
1560 | mg = alloc_migration(cache); |
1561 | if (!mg) { | |
1562 | policy_complete_background_work(cache->policy, op, false); | |
1563 | background_work_end(cache); | |
1564 | return -ENOMEM; | |
1565 | } | |
2ee57d58 | 1566 | |
b29d4986 | 1567 | memset(mg, 0, sizeof(*mg)); |
7ae34e77 | 1568 | |
7ae34e77 | 1569 | mg->cache = cache; |
b29d4986 JT |
1570 | mg->op = op; |
1571 | mg->overwrite_bio = bio; | |
1572 | ||
1573 | if (!bio) | |
1574 | inc_io_migrations(cache); | |
7ae34e77 | 1575 | |
b29d4986 | 1576 | return mg_lock_writes(mg); |
7ae34e77 JT |
1577 | } |
1578 | ||
c6b4fcba | 1579 | /*---------------------------------------------------------------- |
b29d4986 | 1580 | * invalidation processing |
c6b4fcba | 1581 | *--------------------------------------------------------------*/ |
c6b4fcba | 1582 | |
b29d4986 | 1583 | static void invalidate_complete(struct dm_cache_migration *mg, bool success) |
c6b4fcba | 1584 | { |
b29d4986 JT |
1585 | struct bio_list bios; |
1586 | struct cache *cache = mg->cache; | |
c6b4fcba | 1587 | |
b29d4986 JT |
1588 | bio_list_init(&bios); |
1589 | if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios)) | |
1590 | free_prison_cell(cache, mg->cell); | |
c6b4fcba | 1591 | |
b29d4986 JT |
1592 | if (!success && mg->overwrite_bio) |
1593 | bio_io_error(mg->overwrite_bio); | |
c6b4fcba | 1594 | |
b29d4986 JT |
1595 | free_migration(mg); |
1596 | defer_bios(cache, &bios); | |
c6b4fcba | 1597 | |
b29d4986 | 1598 | background_work_end(cache); |
c6b4fcba JT |
1599 | } |
1600 | ||
b29d4986 | 1601 | static void invalidate_completed(struct work_struct *ws) |
c6b4fcba | 1602 | { |
b29d4986 JT |
1603 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1604 | invalidate_complete(mg, !mg->k.input); | |
c6b4fcba JT |
1605 | } |
1606 | ||
b29d4986 | 1607 | static int invalidate_cblock(struct cache *cache, dm_cblock_t cblock) |
651f5fa2 | 1608 | { |
b29d4986 JT |
1609 | int r = policy_invalidate_mapping(cache->policy, cblock); |
1610 | if (!r) { | |
1611 | r = dm_cache_remove_mapping(cache->cmd, cblock); | |
1612 | if (r) { | |
1613 | DMERR_LIMIT("%s: invalidation failed; couldn't update on disk metadata", | |
1614 | cache_device_name(cache)); | |
1615 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); | |
651f5fa2 JT |
1616 | } |
1617 | ||
b29d4986 JT |
1618 | } else if (r == -ENODATA) { |
1619 | /* | |
1620 | * Harmless, already unmapped. | |
1621 | */ | |
1622 | r = 0; | |
651f5fa2 | 1623 | |
b29d4986 JT |
1624 | } else |
1625 | DMERR("%s: policy_invalidate_mapping failed", cache_device_name(cache)); | |
2ee57d58 | 1626 | |
b29d4986 | 1627 | return r; |
651f5fa2 JT |
1628 | } |
1629 | ||
b29d4986 | 1630 | static void invalidate_remove(struct work_struct *ws) |
651f5fa2 | 1631 | { |
b29d4986 JT |
1632 | int r; |
1633 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
1634 | struct cache *cache = mg->cache; | |
651f5fa2 | 1635 | |
b29d4986 JT |
1636 | r = invalidate_cblock(cache, mg->invalidate_cblock); |
1637 | if (r) { | |
1638 | invalidate_complete(mg, false); | |
1639 | return; | |
651f5fa2 | 1640 | } |
9153df74 | 1641 | |
b29d4986 JT |
1642 | init_continuation(&mg->k, invalidate_completed); |
1643 | continue_after_commit(&cache->committer, &mg->k); | |
1644 | remap_to_origin_clear_discard(cache, mg->overwrite_bio, mg->invalidate_oblock); | |
1645 | mg->overwrite_bio = NULL; | |
1646 | schedule_commit(&cache->committer); | |
651f5fa2 JT |
1647 | } |
1648 | ||
b29d4986 | 1649 | static int invalidate_lock(struct dm_cache_migration *mg) |
651f5fa2 | 1650 | { |
b29d4986 JT |
1651 | int r; |
1652 | struct dm_cell_key_v2 key; | |
1653 | struct cache *cache = mg->cache; | |
1654 | struct dm_bio_prison_cell_v2 *prealloc; | |
651f5fa2 | 1655 | |
b29d4986 JT |
1656 | prealloc = alloc_prison_cell(cache); |
1657 | if (!prealloc) { | |
1658 | invalidate_complete(mg, false); | |
1659 | return -ENOMEM; | |
651f5fa2 JT |
1660 | } |
1661 | ||
b29d4986 JT |
1662 | build_key(mg->invalidate_oblock, oblock_succ(mg->invalidate_oblock), &key); |
1663 | r = dm_cell_lock_v2(cache->prison, &key, | |
1664 | READ_WRITE_LOCK_LEVEL, prealloc, &mg->cell); | |
1665 | if (r < 0) { | |
1666 | free_prison_cell(cache, prealloc); | |
1667 | invalidate_complete(mg, false); | |
1668 | return r; | |
651f5fa2 | 1669 | } |
9153df74 | 1670 | |
b29d4986 JT |
1671 | if (mg->cell != prealloc) |
1672 | free_prison_cell(cache, prealloc); | |
651f5fa2 | 1673 | |
b29d4986 JT |
1674 | if (r) |
1675 | quiesce(mg, invalidate_remove); | |
651f5fa2 | 1676 | |
b29d4986 JT |
1677 | else { |
1678 | /* | |
1679 | * We can't call invalidate_remove() directly here because we | |
1680 | * might still be in request context. | |
1681 | */ | |
1682 | init_continuation(&mg->k, invalidate_remove); | |
1683 | queue_work(cache->wq, &mg->k.ws); | |
1684 | } | |
fb4100ae | 1685 | |
fb4100ae JT |
1686 | return 0; |
1687 | } | |
1688 | ||
b29d4986 JT |
1689 | static int invalidate_start(struct cache *cache, dm_cblock_t cblock, |
1690 | dm_oblock_t oblock, struct bio *bio) | |
fb4100ae | 1691 | { |
b29d4986 | 1692 | struct dm_cache_migration *mg; |
2ee57d58 | 1693 | |
b29d4986 JT |
1694 | if (!background_work_begin(cache)) |
1695 | return -EPERM; | |
c6b4fcba | 1696 | |
b29d4986 JT |
1697 | mg = alloc_migration(cache); |
1698 | if (!mg) { | |
1699 | background_work_end(cache); | |
1700 | return -ENOMEM; | |
7ae34e77 | 1701 | } |
c6b4fcba | 1702 | |
b29d4986 | 1703 | memset(mg, 0, sizeof(*mg)); |
c6b4fcba | 1704 | |
b29d4986 JT |
1705 | mg->cache = cache; |
1706 | mg->overwrite_bio = bio; | |
1707 | mg->invalidate_cblock = cblock; | |
1708 | mg->invalidate_oblock = oblock; | |
c6b4fcba | 1709 | |
b29d4986 | 1710 | return invalidate_lock(mg); |
c6b4fcba | 1711 | } |
c6b4fcba | 1712 | |
b29d4986 JT |
1713 | /*---------------------------------------------------------------- |
1714 | * bio processing | |
1715 | *--------------------------------------------------------------*/ | |
c6b4fcba | 1716 | |
b29d4986 JT |
1717 | enum busy { |
1718 | IDLE, | |
1719 | MODERATE, | |
1720 | BUSY | |
1721 | }; | |
c6b4fcba | 1722 | |
b29d4986 | 1723 | static enum busy spare_migration_bandwidth(struct cache *cache) |
651f5fa2 | 1724 | { |
b29d4986 | 1725 | bool idle = iot_idle_for(&cache->origin_tracker, HZ); |
a59db676 | 1726 | sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) * |
c6b4fcba | 1727 | cache->sectors_per_block; |
651f5fa2 | 1728 | |
b29d4986 JT |
1729 | if (current_volume <= cache->migration_threshold) |
1730 | return idle ? IDLE : MODERATE; | |
1731 | else | |
1732 | return idle ? MODERATE : BUSY; | |
651f5fa2 JT |
1733 | } |
1734 | ||
c6b4fcba | 1735 | static void inc_hit_counter(struct cache *cache, struct bio *bio) |
c6b4fcba | 1736 | { |
c6b4fcba JT |
1737 | atomic_inc(bio_data_dir(bio) == READ ? |
1738 | &cache->stats.read_hit : &cache->stats.write_hit); | |
c6b4fcba JT |
1739 | } |
1740 | ||
c6b4fcba | 1741 | static void inc_miss_counter(struct cache *cache, struct bio *bio) |
028ae9f7 | 1742 | { |
c6b4fcba JT |
1743 | atomic_inc(bio_data_dir(bio) == READ ? |
1744 | &cache->stats.read_miss : &cache->stats.write_miss); | |
1745 | } | |
028ae9f7 | 1746 | |
fb4100ae | 1747 | /*----------------------------------------------------------------*/ |
028ae9f7 | 1748 | |
b29d4986 | 1749 | static bool bio_writes_complete_block(struct cache *cache, struct bio *bio) |
651f5fa2 | 1750 | { |
b29d4986 JT |
1751 | return (bio_data_dir(bio) == WRITE) && |
1752 | (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT)); | |
028ae9f7 JT |
1753 | } |
1754 | ||
b29d4986 | 1755 | static bool optimisable_bio(struct cache *cache, struct bio *bio, dm_oblock_t block) |
c6b4fcba | 1756 | { |
b29d4986 JT |
1757 | return writeback_mode(&cache->features) && |
1758 | (is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio)); | |
c6b4fcba JT |
1759 | } |
1760 | ||
b29d4986 JT |
1761 | static int map_bio(struct cache *cache, struct bio *bio, dm_oblock_t block, |
1762 | bool *commit_needed) | |
c6b4fcba | 1763 | { |
b29d4986 JT |
1764 | int r, data_dir; |
1765 | bool rb, background_queued; | |
1766 | dm_cblock_t cblock; | |
1767 | size_t pb_data_size = get_per_bio_data_size(cache); | |
1768 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba | 1769 | |
b29d4986 | 1770 | *commit_needed = false; |
c6b4fcba | 1771 | |
b29d4986 JT |
1772 | rb = bio_detain_shared(cache, block, bio); |
1773 | if (!rb) { | |
c6b4fcba | 1774 | /* |
b29d4986 JT |
1775 | * An exclusive lock is held for this block, so we have to |
1776 | * wait. We set the commit_needed flag so the current | |
1777 | * transaction will be committed asap, allowing this lock | |
1778 | * to be dropped. | |
c6b4fcba | 1779 | */ |
b29d4986 JT |
1780 | *commit_needed = true; |
1781 | return DM_MAPIO_SUBMITTED; | |
651f5fa2 | 1782 | } |
9153df74 | 1783 | |
b29d4986 | 1784 | data_dir = bio_data_dir(bio); |
651f5fa2 | 1785 | |
b29d4986 JT |
1786 | if (optimisable_bio(cache, bio, block)) { |
1787 | struct policy_work *op = NULL; | |
fb4100ae | 1788 | |
b29d4986 JT |
1789 | r = policy_lookup_with_work(cache->policy, block, &cblock, data_dir, true, &op); |
1790 | if (unlikely(r && r != -ENOENT)) { | |
1791 | DMERR_LIMIT("%s: policy_lookup_with_work() failed with r = %d", | |
1792 | cache_device_name(cache), r); | |
1793 | bio_io_error(bio); | |
1794 | return DM_MAPIO_SUBMITTED; | |
c6b4fcba JT |
1795 | } |
1796 | ||
b29d4986 JT |
1797 | if (r == -ENOENT && op) { |
1798 | bio_drop_shared_lock(cache, bio); | |
1799 | BUG_ON(op->op != POLICY_PROMOTE); | |
1800 | mg_start(cache, op, bio); | |
1801 | return DM_MAPIO_SUBMITTED; | |
1802 | } | |
1803 | } else { | |
1804 | r = policy_lookup(cache->policy, block, &cblock, data_dir, false, &background_queued); | |
1805 | if (unlikely(r && r != -ENOENT)) { | |
1806 | DMERR_LIMIT("%s: policy_lookup() failed with r = %d", | |
1807 | cache_device_name(cache), r); | |
1808 | bio_io_error(bio); | |
1809 | return DM_MAPIO_SUBMITTED; | |
1810 | } | |
c6b4fcba | 1811 | |
b29d4986 JT |
1812 | if (background_queued) |
1813 | wake_migration_worker(cache); | |
c6b4fcba JT |
1814 | } |
1815 | ||
b29d4986 JT |
1816 | if (r == -ENOENT) { |
1817 | /* | |
1818 | * Miss. | |
1819 | */ | |
1820 | inc_miss_counter(cache, bio); | |
1821 | if (pb->req_nr == 0) { | |
1822 | accounted_begin(cache, bio); | |
1823 | remap_to_origin_clear_discard(cache, bio, block); | |
651f5fa2 | 1824 | |
b29d4986 | 1825 | } else { |
2ee57d58 | 1826 | /* |
b29d4986 JT |
1827 | * This is a duplicate writethrough io that is no |
1828 | * longer needed because the block has been demoted. | |
2ee57d58 | 1829 | */ |
b29d4986 JT |
1830 | bio_endio(bio); |
1831 | return DM_MAPIO_SUBMITTED; | |
1832 | } | |
1833 | } else { | |
1834 | /* | |
1835 | * Hit. | |
1836 | */ | |
1837 | inc_hit_counter(cache, bio); | |
651f5fa2 | 1838 | |
b29d4986 JT |
1839 | /* |
1840 | * Passthrough always maps to the origin, invalidating any | |
1841 | * cache blocks that are written to. | |
1842 | */ | |
1843 | if (passthrough_mode(&cache->features)) { | |
2ee57d58 | 1844 | if (bio_data_dir(bio) == WRITE) { |
b29d4986 | 1845 | bio_drop_shared_lock(cache, bio); |
2ee57d58 | 1846 | atomic_inc(&cache->stats.demotion); |
b29d4986 JT |
1847 | invalidate_start(cache, cblock, block, bio); |
1848 | } else | |
2ee57d58 | 1849 | remap_to_origin_clear_discard(cache, bio, block); |
651f5fa2 | 1850 | |
2ee57d58 | 1851 | } else { |
b29d4986 JT |
1852 | if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) && |
1853 | !is_dirty(cache, cblock)) { | |
1854 | remap_to_origin_then_cache(cache, bio, block, cblock); | |
1855 | accounted_begin(cache, bio); | |
1856 | } else | |
1857 | remap_to_cache_dirty(cache, bio, block, cblock); | |
2ee57d58 | 1858 | } |
c6b4fcba | 1859 | } |
651f5fa2 | 1860 | |
651f5fa2 | 1861 | /* |
b29d4986 | 1862 | * dm core turns FUA requests into a separate payload and FLUSH req. |
651f5fa2 | 1863 | */ |
b29d4986 | 1864 | if (bio->bi_opf & REQ_FUA) { |
651f5fa2 | 1865 | /* |
b29d4986 JT |
1866 | * issue_after_commit will call accounted_begin a second time. So |
1867 | * we call accounted_complete() to avoid double accounting. | |
651f5fa2 | 1868 | */ |
b29d4986 JT |
1869 | accounted_complete(cache, bio); |
1870 | issue_after_commit(&cache->committer, bio); | |
1871 | *commit_needed = true; | |
1872 | return DM_MAPIO_SUBMITTED; | |
651f5fa2 JT |
1873 | } |
1874 | ||
b29d4986 | 1875 | return DM_MAPIO_REMAPPED; |
651f5fa2 JT |
1876 | } |
1877 | ||
b29d4986 | 1878 | static bool process_bio(struct cache *cache, struct bio *bio) |
c6b4fcba | 1879 | { |
b29d4986 | 1880 | bool commit_needed; |
c6b4fcba | 1881 | |
b29d4986 JT |
1882 | if (map_bio(cache, bio, get_bio_block(cache, bio), &commit_needed) == DM_MAPIO_REMAPPED) |
1883 | generic_make_request(bio); | |
c6b4fcba | 1884 | |
b29d4986 | 1885 | return commit_needed; |
c6b4fcba JT |
1886 | } |
1887 | ||
028ae9f7 JT |
1888 | /* |
1889 | * A non-zero return indicates read_only or fail_io mode. | |
1890 | */ | |
1891 | static int commit(struct cache *cache, bool clean_shutdown) | |
e2e74d61 | 1892 | { |
028ae9f7 | 1893 | int r; |
e2e74d61 | 1894 | |
028ae9f7 JT |
1895 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
1896 | return -EINVAL; | |
e2e74d61 | 1897 | |
028ae9f7 JT |
1898 | atomic_inc(&cache->stats.commit_count); |
1899 | r = dm_cache_commit(cache->cmd, clean_shutdown); | |
1900 | if (r) | |
1901 | metadata_operation_failed(cache, "dm_cache_commit", r); | |
e2e74d61 | 1902 | |
028ae9f7 | 1903 | return r; |
e2e74d61 JT |
1904 | } |
1905 | ||
b29d4986 JT |
1906 | /* |
1907 | * Used by the batcher. | |
1908 | */ | |
1909 | static int commit_op(void *context) | |
c6b4fcba | 1910 | { |
b29d4986 | 1911 | struct cache *cache = context; |
c6b4fcba | 1912 | |
b29d4986 JT |
1913 | if (dm_cache_changed_this_transaction(cache->cmd)) |
1914 | return commit(cache, false); | |
c6b4fcba | 1915 | |
b29d4986 | 1916 | return 0; |
c6b4fcba JT |
1917 | } |
1918 | ||
b29d4986 | 1919 | /*----------------------------------------------------------------*/ |
65790ff9 | 1920 | |
b29d4986 | 1921 | static bool process_flush_bio(struct cache *cache, struct bio *bio) |
65790ff9 | 1922 | { |
b29d4986 JT |
1923 | size_t pb_data_size = get_per_bio_data_size(cache); |
1924 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
65790ff9 | 1925 | |
b29d4986 JT |
1926 | if (!pb->req_nr) |
1927 | remap_to_origin(cache, bio); | |
1928 | else | |
1929 | remap_to_cache(cache, bio, 0); | |
65790ff9 | 1930 | |
b29d4986 JT |
1931 | issue_after_commit(&cache->committer, bio); |
1932 | return true; | |
65790ff9 JT |
1933 | } |
1934 | ||
b29d4986 | 1935 | static bool process_discard_bio(struct cache *cache, struct bio *bio) |
65790ff9 | 1936 | { |
b29d4986 | 1937 | dm_dblock_t b, e; |
65790ff9 | 1938 | |
b29d4986 JT |
1939 | // FIXME: do we need to lock the region? Or can we just assume the |
1940 | // user wont be so foolish as to issue discard concurrently with | |
1941 | // other IO? | |
1942 | calc_discard_block_range(cache, bio, &b, &e); | |
1943 | while (b != e) { | |
1944 | set_discard(cache, b); | |
1945 | b = to_dblock(from_dblock(b) + 1); | |
651f5fa2 | 1946 | } |
65790ff9 | 1947 | |
b29d4986 | 1948 | bio_endio(bio); |
65790ff9 | 1949 | |
b29d4986 | 1950 | return false; |
c6b4fcba JT |
1951 | } |
1952 | ||
b29d4986 | 1953 | static void process_deferred_bios(struct work_struct *ws) |
66cb1910 | 1954 | { |
b29d4986 | 1955 | struct cache *cache = container_of(ws, struct cache, deferred_bio_worker); |
66cb1910 | 1956 | |
c6b4fcba | 1957 | unsigned long flags; |
b29d4986 | 1958 | bool commit_needed = false; |
c6b4fcba JT |
1959 | struct bio_list bios; |
1960 | struct bio *bio; | |
66cb1910 | 1961 | |
c6b4fcba | 1962 | bio_list_init(&bios); |
c6b4fcba | 1963 | |
c6b4fcba | 1964 | spin_lock_irqsave(&cache->lock, flags); |
b29d4986 JT |
1965 | bio_list_merge(&bios, &cache->deferred_bios); |
1966 | bio_list_init(&cache->deferred_bios); | |
c6b4fcba | 1967 | spin_unlock_irqrestore(&cache->lock, flags); |
c6b4fcba | 1968 | |
b29d4986 JT |
1969 | while ((bio = bio_list_pop(&bios))) { |
1970 | if (bio->bi_opf & REQ_PREFLUSH) | |
1971 | commit_needed = process_flush_bio(cache, bio) || commit_needed; | |
c6b4fcba | 1972 | |
b29d4986 JT |
1973 | else if (bio_op(bio) == REQ_OP_DISCARD) |
1974 | commit_needed = process_discard_bio(cache, bio) || commit_needed; | |
1975 | ||
1976 | else | |
1977 | commit_needed = process_bio(cache, bio) || commit_needed; | |
1978 | } | |
1979 | ||
1980 | if (commit_needed) | |
1981 | schedule_commit(&cache->committer); | |
c6b4fcba JT |
1982 | } |
1983 | ||
b29d4986 | 1984 | static void process_deferred_writethrough_bios(struct work_struct *ws) |
651f5fa2 | 1985 | { |
b29d4986 JT |
1986 | struct cache *cache = container_of(ws, struct cache, deferred_writethrough_worker); |
1987 | ||
651f5fa2 | 1988 | unsigned long flags; |
e2e74d61 JT |
1989 | struct bio_list bios; |
1990 | struct bio *bio; | |
1991 | ||
1992 | bio_list_init(&bios); | |
651f5fa2 | 1993 | |
651f5fa2 | 1994 | spin_lock_irqsave(&cache->lock, flags); |
e2e74d61 JT |
1995 | bio_list_merge(&bios, &cache->deferred_writethrough_bios); |
1996 | bio_list_init(&cache->deferred_writethrough_bios); | |
651f5fa2 JT |
1997 | spin_unlock_irqrestore(&cache->lock, flags); |
1998 | ||
8c081b52 | 1999 | /* |
b29d4986 | 2000 | * These bios have already been through accounted_begin() |
8c081b52 | 2001 | */ |
e2e74d61 | 2002 | while ((bio = bio_list_pop(&bios))) |
b29d4986 | 2003 | generic_make_request(bio); |
651f5fa2 JT |
2004 | } |
2005 | ||
c6b4fcba JT |
2006 | /*---------------------------------------------------------------- |
2007 | * Main worker loop | |
2008 | *--------------------------------------------------------------*/ | |
651f5fa2 JT |
2009 | |
2010 | static void requeue_deferred_bios(struct cache *cache) | |
c6b4fcba JT |
2011 | { |
2012 | struct bio *bio; | |
2013 | struct bio_list bios; | |
2014 | ||
2015 | bio_list_init(&bios); | |
2016 | bio_list_merge(&bios, &cache->deferred_bios); | |
2017 | bio_list_init(&cache->deferred_bios); | |
2018 | ||
4246a0b6 CH |
2019 | while ((bio = bio_list_pop(&bios))) { |
2020 | bio->bi_error = DM_ENDIO_REQUEUE; | |
2021 | bio_endio(bio); | |
2022 | } | |
c6b4fcba JT |
2023 | } |
2024 | ||
c6b4fcba JT |
2025 | /* |
2026 | * We want to commit periodically so that not too much | |
2027 | * unwritten metadata builds up. | |
2028 | */ | |
2029 | static void do_waker(struct work_struct *ws) | |
2030 | { | |
2031 | struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker); | |
b29d4986 | 2032 | |
fba10109 | 2033 | policy_tick(cache->policy, true); |
b29d4986 JT |
2034 | wake_migration_worker(cache); |
2035 | schedule_commit(&cache->committer); | |
c6b4fcba JT |
2036 | queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD); |
2037 | } | |
2038 | ||
b29d4986 | 2039 | static void check_migrations(struct work_struct *ws) |
c6b4fcba | 2040 | { |
b29d4986 JT |
2041 | int r; |
2042 | struct policy_work *op; | |
2043 | struct cache *cache = container_of(ws, struct cache, migration_worker); | |
2044 | enum busy b; | |
c6b4fcba | 2045 | |
b29d4986 JT |
2046 | for (;;) { |
2047 | b = spare_migration_bandwidth(cache); | |
2048 | if (b == BUSY) | |
2049 | break; | |
c6b4fcba | 2050 | |
b29d4986 JT |
2051 | r = policy_get_background_work(cache->policy, b == IDLE, &op); |
2052 | if (r == -ENODATA) | |
2053 | break; | |
2054 | ||
2055 | if (r) { | |
2056 | DMERR_LIMIT("%s: policy_background_work failed", | |
2057 | cache_device_name(cache)); | |
2058 | break; | |
2059 | } | |
2060 | ||
2061 | r = mg_start(cache, op, NULL); | |
2062 | if (r) | |
2063 | break; | |
2064 | } | |
c6b4fcba JT |
2065 | } |
2066 | ||
2067 | /*---------------------------------------------------------------- | |
2068 | * Target methods | |
2069 | *--------------------------------------------------------------*/ | |
2070 | ||
2071 | /* | |
2072 | * This function gets called on the error paths of the constructor, so we | |
2073 | * have to cope with a partially initialised struct. | |
2074 | */ | |
2075 | static void destroy(struct cache *cache) | |
2076 | { | |
2077 | unsigned i; | |
2078 | ||
6f65985e | 2079 | mempool_destroy(cache->migration_pool); |
c6b4fcba | 2080 | |
c6b4fcba | 2081 | if (cache->prison) |
b29d4986 | 2082 | dm_bio_prison_destroy_v2(cache->prison); |
c6b4fcba JT |
2083 | |
2084 | if (cache->wq) | |
2085 | destroy_workqueue(cache->wq); | |
2086 | ||
2087 | if (cache->dirty_bitset) | |
2088 | free_bitset(cache->dirty_bitset); | |
2089 | ||
2090 | if (cache->discard_bitset) | |
2091 | free_bitset(cache->discard_bitset); | |
2092 | ||
2093 | if (cache->copier) | |
2094 | dm_kcopyd_client_destroy(cache->copier); | |
2095 | ||
2096 | if (cache->cmd) | |
2097 | dm_cache_metadata_close(cache->cmd); | |
2098 | ||
2099 | if (cache->metadata_dev) | |
2100 | dm_put_device(cache->ti, cache->metadata_dev); | |
2101 | ||
2102 | if (cache->origin_dev) | |
2103 | dm_put_device(cache->ti, cache->origin_dev); | |
2104 | ||
2105 | if (cache->cache_dev) | |
2106 | dm_put_device(cache->ti, cache->cache_dev); | |
2107 | ||
2108 | if (cache->policy) | |
2109 | dm_cache_policy_destroy(cache->policy); | |
2110 | ||
2111 | for (i = 0; i < cache->nr_ctr_args ; i++) | |
2112 | kfree(cache->ctr_args[i]); | |
2113 | kfree(cache->ctr_args); | |
2114 | ||
2115 | kfree(cache); | |
2116 | } | |
2117 | ||
2118 | static void cache_dtr(struct dm_target *ti) | |
2119 | { | |
2120 | struct cache *cache = ti->private; | |
2121 | ||
2122 | destroy(cache); | |
2123 | } | |
2124 | ||
2125 | static sector_t get_dev_size(struct dm_dev *dev) | |
2126 | { | |
2127 | return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
2128 | } | |
2129 | ||
2130 | /*----------------------------------------------------------------*/ | |
2131 | ||
2132 | /* | |
2133 | * Construct a cache device mapping. | |
2134 | * | |
2135 | * cache <metadata dev> <cache dev> <origin dev> <block size> | |
2136 | * <#feature args> [<feature arg>]* | |
2137 | * <policy> <#policy args> [<policy arg>]* | |
2138 | * | |
2139 | * metadata dev : fast device holding the persistent metadata | |
2140 | * cache dev : fast device holding cached data blocks | |
2141 | * origin dev : slow device holding original data blocks | |
2142 | * block size : cache unit size in sectors | |
2143 | * | |
2144 | * #feature args : number of feature arguments passed | |
2145 | * feature args : writethrough. (The default is writeback.) | |
2146 | * | |
2147 | * policy : the replacement policy to use | |
2148 | * #policy args : an even number of policy arguments corresponding | |
2149 | * to key/value pairs passed to the policy | |
2150 | * policy args : key/value pairs passed to the policy | |
2151 | * E.g. 'sequential_threshold 1024' | |
2152 | * See cache-policies.txt for details. | |
2153 | * | |
2154 | * Optional feature arguments are: | |
2155 | * writethrough : write through caching that prohibits cache block | |
2156 | * content from being different from origin block content. | |
2157 | * Without this argument, the default behaviour is to write | |
2158 | * back cache block contents later for performance reasons, | |
2159 | * so they may differ from the corresponding origin blocks. | |
2160 | */ | |
2161 | struct cache_args { | |
2162 | struct dm_target *ti; | |
2163 | ||
2164 | struct dm_dev *metadata_dev; | |
2165 | ||
2166 | struct dm_dev *cache_dev; | |
2167 | sector_t cache_sectors; | |
2168 | ||
2169 | struct dm_dev *origin_dev; | |
2170 | sector_t origin_sectors; | |
2171 | ||
2172 | uint32_t block_size; | |
2173 | ||
2174 | const char *policy_name; | |
2175 | int policy_argc; | |
2176 | const char **policy_argv; | |
2177 | ||
2178 | struct cache_features features; | |
2179 | }; | |
2180 | ||
2181 | static void destroy_cache_args(struct cache_args *ca) | |
2182 | { | |
2183 | if (ca->metadata_dev) | |
2184 | dm_put_device(ca->ti, ca->metadata_dev); | |
2185 | ||
2186 | if (ca->cache_dev) | |
2187 | dm_put_device(ca->ti, ca->cache_dev); | |
2188 | ||
2189 | if (ca->origin_dev) | |
2190 | dm_put_device(ca->ti, ca->origin_dev); | |
2191 | ||
2192 | kfree(ca); | |
2193 | } | |
2194 | ||
2195 | static bool at_least_one_arg(struct dm_arg_set *as, char **error) | |
2196 | { | |
2197 | if (!as->argc) { | |
2198 | *error = "Insufficient args"; | |
2199 | return false; | |
2200 | } | |
2201 | ||
2202 | return true; | |
2203 | } | |
2204 | ||
2205 | static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2206 | char **error) | |
2207 | { | |
2208 | int r; | |
2209 | sector_t metadata_dev_size; | |
2210 | char b[BDEVNAME_SIZE]; | |
2211 | ||
2212 | if (!at_least_one_arg(as, error)) | |
2213 | return -EINVAL; | |
2214 | ||
2215 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2216 | &ca->metadata_dev); | |
2217 | if (r) { | |
2218 | *error = "Error opening metadata device"; | |
2219 | return r; | |
2220 | } | |
2221 | ||
2222 | metadata_dev_size = get_dev_size(ca->metadata_dev); | |
2223 | if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING) | |
2224 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
2225 | bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
2226 | ||
2227 | return 0; | |
2228 | } | |
2229 | ||
2230 | static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2231 | char **error) | |
2232 | { | |
2233 | int r; | |
2234 | ||
2235 | if (!at_least_one_arg(as, error)) | |
2236 | return -EINVAL; | |
2237 | ||
2238 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2239 | &ca->cache_dev); | |
2240 | if (r) { | |
2241 | *error = "Error opening cache device"; | |
2242 | return r; | |
2243 | } | |
2244 | ca->cache_sectors = get_dev_size(ca->cache_dev); | |
2245 | ||
2246 | return 0; | |
2247 | } | |
2248 | ||
2249 | static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2250 | char **error) | |
2251 | { | |
2252 | int r; | |
2253 | ||
2254 | if (!at_least_one_arg(as, error)) | |
2255 | return -EINVAL; | |
2256 | ||
2257 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2258 | &ca->origin_dev); | |
2259 | if (r) { | |
2260 | *error = "Error opening origin device"; | |
2261 | return r; | |
2262 | } | |
2263 | ||
2264 | ca->origin_sectors = get_dev_size(ca->origin_dev); | |
2265 | if (ca->ti->len > ca->origin_sectors) { | |
2266 | *error = "Device size larger than cached device"; | |
2267 | return -EINVAL; | |
2268 | } | |
2269 | ||
2270 | return 0; | |
2271 | } | |
2272 | ||
2273 | static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as, | |
2274 | char **error) | |
2275 | { | |
05473044 | 2276 | unsigned long block_size; |
c6b4fcba JT |
2277 | |
2278 | if (!at_least_one_arg(as, error)) | |
2279 | return -EINVAL; | |
2280 | ||
05473044 MS |
2281 | if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size || |
2282 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2283 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
2284 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { | |
c6b4fcba JT |
2285 | *error = "Invalid data block size"; |
2286 | return -EINVAL; | |
2287 | } | |
2288 | ||
05473044 | 2289 | if (block_size > ca->cache_sectors) { |
c6b4fcba JT |
2290 | *error = "Data block size is larger than the cache device"; |
2291 | return -EINVAL; | |
2292 | } | |
2293 | ||
05473044 | 2294 | ca->block_size = block_size; |
c6b4fcba JT |
2295 | |
2296 | return 0; | |
2297 | } | |
2298 | ||
2299 | static void init_features(struct cache_features *cf) | |
2300 | { | |
2301 | cf->mode = CM_WRITE; | |
2ee57d58 | 2302 | cf->io_mode = CM_IO_WRITEBACK; |
629d0a8a | 2303 | cf->metadata_version = 1; |
c6b4fcba JT |
2304 | } |
2305 | ||
2306 | static int parse_features(struct cache_args *ca, struct dm_arg_set *as, | |
2307 | char **error) | |
2308 | { | |
2309 | static struct dm_arg _args[] = { | |
629d0a8a | 2310 | {0, 2, "Invalid number of cache feature arguments"}, |
c6b4fcba JT |
2311 | }; |
2312 | ||
2313 | int r; | |
2314 | unsigned argc; | |
2315 | const char *arg; | |
2316 | struct cache_features *cf = &ca->features; | |
2317 | ||
2318 | init_features(cf); | |
2319 | ||
2320 | r = dm_read_arg_group(_args, as, &argc, error); | |
2321 | if (r) | |
2322 | return -EINVAL; | |
2323 | ||
2324 | while (argc--) { | |
2325 | arg = dm_shift_arg(as); | |
2326 | ||
2327 | if (!strcasecmp(arg, "writeback")) | |
2ee57d58 | 2328 | cf->io_mode = CM_IO_WRITEBACK; |
c6b4fcba JT |
2329 | |
2330 | else if (!strcasecmp(arg, "writethrough")) | |
2ee57d58 JT |
2331 | cf->io_mode = CM_IO_WRITETHROUGH; |
2332 | ||
2333 | else if (!strcasecmp(arg, "passthrough")) | |
2334 | cf->io_mode = CM_IO_PASSTHROUGH; | |
c6b4fcba | 2335 | |
629d0a8a JT |
2336 | else if (!strcasecmp(arg, "metadata2")) |
2337 | cf->metadata_version = 2; | |
2338 | ||
c6b4fcba JT |
2339 | else { |
2340 | *error = "Unrecognised cache feature requested"; | |
2341 | return -EINVAL; | |
2342 | } | |
2343 | } | |
2344 | ||
2345 | return 0; | |
2346 | } | |
2347 | ||
2348 | static int parse_policy(struct cache_args *ca, struct dm_arg_set *as, | |
2349 | char **error) | |
2350 | { | |
2351 | static struct dm_arg _args[] = { | |
2352 | {0, 1024, "Invalid number of policy arguments"}, | |
2353 | }; | |
2354 | ||
2355 | int r; | |
2356 | ||
2357 | if (!at_least_one_arg(as, error)) | |
2358 | return -EINVAL; | |
2359 | ||
2360 | ca->policy_name = dm_shift_arg(as); | |
2361 | ||
2362 | r = dm_read_arg_group(_args, as, &ca->policy_argc, error); | |
2363 | if (r) | |
2364 | return -EINVAL; | |
2365 | ||
2366 | ca->policy_argv = (const char **)as->argv; | |
2367 | dm_consume_args(as, ca->policy_argc); | |
2368 | ||
2369 | return 0; | |
2370 | } | |
2371 | ||
2372 | static int parse_cache_args(struct cache_args *ca, int argc, char **argv, | |
2373 | char **error) | |
2374 | { | |
2375 | int r; | |
2376 | struct dm_arg_set as; | |
2377 | ||
2378 | as.argc = argc; | |
2379 | as.argv = argv; | |
2380 | ||
2381 | r = parse_metadata_dev(ca, &as, error); | |
2382 | if (r) | |
2383 | return r; | |
2384 | ||
2385 | r = parse_cache_dev(ca, &as, error); | |
2386 | if (r) | |
2387 | return r; | |
2388 | ||
2389 | r = parse_origin_dev(ca, &as, error); | |
2390 | if (r) | |
2391 | return r; | |
2392 | ||
2393 | r = parse_block_size(ca, &as, error); | |
2394 | if (r) | |
2395 | return r; | |
2396 | ||
2397 | r = parse_features(ca, &as, error); | |
2398 | if (r) | |
2399 | return r; | |
2400 | ||
2401 | r = parse_policy(ca, &as, error); | |
2402 | if (r) | |
2403 | return r; | |
2404 | ||
2405 | return 0; | |
2406 | } | |
2407 | ||
2408 | /*----------------------------------------------------------------*/ | |
2409 | ||
2410 | static struct kmem_cache *migration_cache; | |
2411 | ||
2c73c471 AK |
2412 | #define NOT_CORE_OPTION 1 |
2413 | ||
2f14f4b5 | 2414 | static int process_config_option(struct cache *cache, const char *key, const char *value) |
2c73c471 AK |
2415 | { |
2416 | unsigned long tmp; | |
2417 | ||
2f14f4b5 JT |
2418 | if (!strcasecmp(key, "migration_threshold")) { |
2419 | if (kstrtoul(value, 10, &tmp)) | |
2c73c471 AK |
2420 | return -EINVAL; |
2421 | ||
2422 | cache->migration_threshold = tmp; | |
2423 | return 0; | |
2424 | } | |
2425 | ||
2426 | return NOT_CORE_OPTION; | |
2427 | } | |
2428 | ||
2f14f4b5 JT |
2429 | static int set_config_value(struct cache *cache, const char *key, const char *value) |
2430 | { | |
2431 | int r = process_config_option(cache, key, value); | |
2432 | ||
2433 | if (r == NOT_CORE_OPTION) | |
2434 | r = policy_set_config_value(cache->policy, key, value); | |
2435 | ||
2436 | if (r) | |
2437 | DMWARN("bad config value for %s: %s", key, value); | |
2438 | ||
2439 | return r; | |
2440 | } | |
2441 | ||
2442 | static int set_config_values(struct cache *cache, int argc, const char **argv) | |
c6b4fcba JT |
2443 | { |
2444 | int r = 0; | |
2445 | ||
2446 | if (argc & 1) { | |
2447 | DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs."); | |
2448 | return -EINVAL; | |
2449 | } | |
2450 | ||
2451 | while (argc) { | |
2f14f4b5 JT |
2452 | r = set_config_value(cache, argv[0], argv[1]); |
2453 | if (r) | |
2454 | break; | |
c6b4fcba JT |
2455 | |
2456 | argc -= 2; | |
2457 | argv += 2; | |
2458 | } | |
2459 | ||
2460 | return r; | |
2461 | } | |
2462 | ||
2463 | static int create_cache_policy(struct cache *cache, struct cache_args *ca, | |
2464 | char **error) | |
2465 | { | |
4cb3e1db MP |
2466 | struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name, |
2467 | cache->cache_size, | |
2468 | cache->origin_sectors, | |
2469 | cache->sectors_per_block); | |
2470 | if (IS_ERR(p)) { | |
c6b4fcba | 2471 | *error = "Error creating cache's policy"; |
4cb3e1db | 2472 | return PTR_ERR(p); |
c6b4fcba | 2473 | } |
4cb3e1db | 2474 | cache->policy = p; |
b29d4986 | 2475 | BUG_ON(!cache->policy); |
c6b4fcba | 2476 | |
2f14f4b5 | 2477 | return 0; |
c6b4fcba JT |
2478 | } |
2479 | ||
08b18451 | 2480 | /* |
2bb812df JT |
2481 | * We want the discard block size to be at least the size of the cache |
2482 | * block size and have no more than 2^14 discard blocks across the origin. | |
08b18451 JT |
2483 | */ |
2484 | #define MAX_DISCARD_BLOCKS (1 << 14) | |
2485 | ||
2486 | static bool too_many_discard_blocks(sector_t discard_block_size, | |
2487 | sector_t origin_size) | |
2488 | { | |
2489 | (void) sector_div(origin_size, discard_block_size); | |
2490 | ||
2491 | return origin_size > MAX_DISCARD_BLOCKS; | |
2492 | } | |
2493 | ||
2494 | static sector_t calculate_discard_block_size(sector_t cache_block_size, | |
2495 | sector_t origin_size) | |
2496 | { | |
2bb812df | 2497 | sector_t discard_block_size = cache_block_size; |
08b18451 JT |
2498 | |
2499 | if (origin_size) | |
2500 | while (too_many_discard_blocks(discard_block_size, origin_size)) | |
2501 | discard_block_size *= 2; | |
2502 | ||
2503 | return discard_block_size; | |
2504 | } | |
2505 | ||
d1d9220c JT |
2506 | static void set_cache_size(struct cache *cache, dm_cblock_t size) |
2507 | { | |
2508 | dm_block_t nr_blocks = from_cblock(size); | |
2509 | ||
2510 | if (nr_blocks > (1 << 20) && cache->cache_size != size) | |
2511 | DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n" | |
2512 | "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n" | |
2513 | "Please consider increasing the cache block size to reduce the overall cache block count.", | |
2514 | (unsigned long long) nr_blocks); | |
2515 | ||
2516 | cache->cache_size = size; | |
2517 | } | |
2518 | ||
b29d4986 JT |
2519 | static int is_congested(struct dm_dev *dev, int bdi_bits) |
2520 | { | |
2521 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
2522 | return bdi_congested(q->backing_dev_info, bdi_bits); | |
2523 | } | |
2524 | ||
2525 | static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2526 | { | |
2527 | struct cache *cache = container_of(cb, struct cache, callbacks); | |
2528 | ||
2529 | return is_congested(cache->origin_dev, bdi_bits) || | |
2530 | is_congested(cache->cache_dev, bdi_bits); | |
2531 | } | |
2532 | ||
f8350daf | 2533 | #define DEFAULT_MIGRATION_THRESHOLD 2048 |
c6b4fcba | 2534 | |
c6b4fcba JT |
2535 | static int cache_create(struct cache_args *ca, struct cache **result) |
2536 | { | |
2537 | int r = 0; | |
2538 | char **error = &ca->ti->error; | |
2539 | struct cache *cache; | |
2540 | struct dm_target *ti = ca->ti; | |
2541 | dm_block_t origin_blocks; | |
2542 | struct dm_cache_metadata *cmd; | |
2543 | bool may_format = ca->features.mode == CM_WRITE; | |
2544 | ||
2545 | cache = kzalloc(sizeof(*cache), GFP_KERNEL); | |
2546 | if (!cache) | |
2547 | return -ENOMEM; | |
2548 | ||
2549 | cache->ti = ca->ti; | |
2550 | ti->private = cache; | |
c6b4fcba JT |
2551 | ti->num_flush_bios = 2; |
2552 | ti->flush_supported = true; | |
2553 | ||
2554 | ti->num_discard_bios = 1; | |
2555 | ti->discards_supported = true; | |
2572629a | 2556 | ti->split_discard_bios = false; |
c6b4fcba | 2557 | |
8c5008fa | 2558 | cache->features = ca->features; |
30187e1d | 2559 | ti->per_io_data_size = get_per_bio_data_size(cache); |
c6b4fcba | 2560 | |
c6b4fcba JT |
2561 | cache->callbacks.congested_fn = cache_is_congested; |
2562 | dm_table_add_target_callbacks(ti->table, &cache->callbacks); | |
2563 | ||
2564 | cache->metadata_dev = ca->metadata_dev; | |
2565 | cache->origin_dev = ca->origin_dev; | |
2566 | cache->cache_dev = ca->cache_dev; | |
2567 | ||
2568 | ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL; | |
2569 | ||
c6b4fcba | 2570 | origin_blocks = cache->origin_sectors = ca->origin_sectors; |
414dd67d | 2571 | origin_blocks = block_div(origin_blocks, ca->block_size); |
c6b4fcba JT |
2572 | cache->origin_blocks = to_oblock(origin_blocks); |
2573 | ||
2574 | cache->sectors_per_block = ca->block_size; | |
2575 | if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) { | |
2576 | r = -EINVAL; | |
2577 | goto bad; | |
2578 | } | |
2579 | ||
2580 | if (ca->block_size & (ca->block_size - 1)) { | |
2581 | dm_block_t cache_size = ca->cache_sectors; | |
2582 | ||
2583 | cache->sectors_per_block_shift = -1; | |
414dd67d | 2584 | cache_size = block_div(cache_size, ca->block_size); |
d1d9220c | 2585 | set_cache_size(cache, to_cblock(cache_size)); |
c6b4fcba JT |
2586 | } else { |
2587 | cache->sectors_per_block_shift = __ffs(ca->block_size); | |
d1d9220c | 2588 | set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift)); |
c6b4fcba JT |
2589 | } |
2590 | ||
2591 | r = create_cache_policy(cache, ca, error); | |
2592 | if (r) | |
2593 | goto bad; | |
2f14f4b5 | 2594 | |
c6b4fcba | 2595 | cache->policy_nr_args = ca->policy_argc; |
2f14f4b5 JT |
2596 | cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD; |
2597 | ||
2598 | r = set_config_values(cache, ca->policy_argc, ca->policy_argv); | |
2599 | if (r) { | |
2600 | *error = "Error setting cache policy's config values"; | |
2601 | goto bad; | |
2602 | } | |
c6b4fcba JT |
2603 | |
2604 | cmd = dm_cache_metadata_open(cache->metadata_dev->bdev, | |
2605 | ca->block_size, may_format, | |
629d0a8a JT |
2606 | dm_cache_policy_get_hint_size(cache->policy), |
2607 | ca->features.metadata_version); | |
c6b4fcba JT |
2608 | if (IS_ERR(cmd)) { |
2609 | *error = "Error creating metadata object"; | |
2610 | r = PTR_ERR(cmd); | |
2611 | goto bad; | |
2612 | } | |
2613 | cache->cmd = cmd; | |
028ae9f7 JT |
2614 | set_cache_mode(cache, CM_WRITE); |
2615 | if (get_cache_mode(cache) != CM_WRITE) { | |
2616 | *error = "Unable to get write access to metadata, please check/repair metadata."; | |
2617 | r = -EINVAL; | |
2618 | goto bad; | |
2619 | } | |
c6b4fcba | 2620 | |
2ee57d58 JT |
2621 | if (passthrough_mode(&cache->features)) { |
2622 | bool all_clean; | |
2623 | ||
2624 | r = dm_cache_metadata_all_clean(cache->cmd, &all_clean); | |
2625 | if (r) { | |
2626 | *error = "dm_cache_metadata_all_clean() failed"; | |
2627 | goto bad; | |
2628 | } | |
2629 | ||
2630 | if (!all_clean) { | |
2631 | *error = "Cannot enter passthrough mode unless all blocks are clean"; | |
2632 | r = -EINVAL; | |
2633 | goto bad; | |
2634 | } | |
b29d4986 JT |
2635 | |
2636 | policy_allow_migrations(cache->policy, false); | |
2ee57d58 JT |
2637 | } |
2638 | ||
c6b4fcba | 2639 | spin_lock_init(&cache->lock); |
651f5fa2 | 2640 | INIT_LIST_HEAD(&cache->deferred_cells); |
c6b4fcba | 2641 | bio_list_init(&cache->deferred_bios); |
e2e74d61 | 2642 | bio_list_init(&cache->deferred_writethrough_bios); |
a59db676 JT |
2643 | atomic_set(&cache->nr_allocated_migrations, 0); |
2644 | atomic_set(&cache->nr_io_migrations, 0); | |
c6b4fcba JT |
2645 | init_waitqueue_head(&cache->migration_wait); |
2646 | ||
fa4d683a | 2647 | r = -ENOMEM; |
44fa816b | 2648 | atomic_set(&cache->nr_dirty, 0); |
c6b4fcba JT |
2649 | cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size)); |
2650 | if (!cache->dirty_bitset) { | |
2651 | *error = "could not allocate dirty bitset"; | |
2652 | goto bad; | |
2653 | } | |
2654 | clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size)); | |
2655 | ||
08b18451 JT |
2656 | cache->discard_block_size = |
2657 | calculate_discard_block_size(cache->sectors_per_block, | |
2658 | cache->origin_sectors); | |
2572629a JT |
2659 | cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors, |
2660 | cache->discard_block_size)); | |
1bad9bc4 | 2661 | cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2662 | if (!cache->discard_bitset) { |
2663 | *error = "could not allocate discard bitset"; | |
2664 | goto bad; | |
2665 | } | |
1bad9bc4 | 2666 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2667 | |
2668 | cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); | |
2669 | if (IS_ERR(cache->copier)) { | |
2670 | *error = "could not create kcopyd client"; | |
2671 | r = PTR_ERR(cache->copier); | |
2672 | goto bad; | |
2673 | } | |
2674 | ||
b29d4986 | 2675 | cache->wq = alloc_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM, 0); |
c6b4fcba JT |
2676 | if (!cache->wq) { |
2677 | *error = "could not create workqueue for metadata object"; | |
2678 | goto bad; | |
2679 | } | |
b29d4986 JT |
2680 | INIT_WORK(&cache->deferred_bio_worker, process_deferred_bios); |
2681 | INIT_WORK(&cache->deferred_writethrough_worker, | |
2682 | process_deferred_writethrough_bios); | |
2683 | INIT_WORK(&cache->migration_worker, check_migrations); | |
c6b4fcba | 2684 | INIT_DELAYED_WORK(&cache->waker, do_waker); |
c6b4fcba | 2685 | |
b29d4986 | 2686 | cache->prison = dm_bio_prison_create_v2(cache->wq); |
c6b4fcba JT |
2687 | if (!cache->prison) { |
2688 | *error = "could not create bio prison"; | |
2689 | goto bad; | |
2690 | } | |
2691 | ||
c6b4fcba JT |
2692 | cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE, |
2693 | migration_cache); | |
2694 | if (!cache->migration_pool) { | |
2695 | *error = "Error creating cache's migration mempool"; | |
2696 | goto bad; | |
2697 | } | |
2698 | ||
c6b4fcba JT |
2699 | cache->need_tick_bio = true; |
2700 | cache->sized = false; | |
65790ff9 | 2701 | cache->invalidate = false; |
c6b4fcba JT |
2702 | cache->commit_requested = false; |
2703 | cache->loaded_mappings = false; | |
2704 | cache->loaded_discards = false; | |
2705 | ||
2706 | load_stats(cache); | |
2707 | ||
2708 | atomic_set(&cache->stats.demotion, 0); | |
2709 | atomic_set(&cache->stats.promotion, 0); | |
2710 | atomic_set(&cache->stats.copies_avoided, 0); | |
2711 | atomic_set(&cache->stats.cache_cell_clash, 0); | |
2712 | atomic_set(&cache->stats.commit_count, 0); | |
2713 | atomic_set(&cache->stats.discard_count, 0); | |
2714 | ||
65790ff9 JT |
2715 | spin_lock_init(&cache->invalidation_lock); |
2716 | INIT_LIST_HEAD(&cache->invalidation_requests); | |
2717 | ||
b29d4986 JT |
2718 | batcher_init(&cache->committer, commit_op, cache, |
2719 | issue_op, cache, cache->wq); | |
066dbaa3 JT |
2720 | iot_init(&cache->origin_tracker); |
2721 | ||
b29d4986 JT |
2722 | init_rwsem(&cache->background_work_lock); |
2723 | prevent_background_work(cache); | |
2724 | ||
c6b4fcba JT |
2725 | *result = cache; |
2726 | return 0; | |
c6b4fcba JT |
2727 | bad: |
2728 | destroy(cache); | |
2729 | return r; | |
2730 | } | |
2731 | ||
2732 | static int copy_ctr_args(struct cache *cache, int argc, const char **argv) | |
2733 | { | |
2734 | unsigned i; | |
2735 | const char **copy; | |
2736 | ||
2737 | copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL); | |
2738 | if (!copy) | |
2739 | return -ENOMEM; | |
2740 | for (i = 0; i < argc; i++) { | |
2741 | copy[i] = kstrdup(argv[i], GFP_KERNEL); | |
2742 | if (!copy[i]) { | |
2743 | while (i--) | |
2744 | kfree(copy[i]); | |
2745 | kfree(copy); | |
2746 | return -ENOMEM; | |
2747 | } | |
2748 | } | |
2749 | ||
2750 | cache->nr_ctr_args = argc; | |
2751 | cache->ctr_args = copy; | |
2752 | ||
2753 | return 0; | |
2754 | } | |
2755 | ||
2756 | static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2757 | { | |
2758 | int r = -EINVAL; | |
2759 | struct cache_args *ca; | |
2760 | struct cache *cache = NULL; | |
2761 | ||
2762 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
2763 | if (!ca) { | |
2764 | ti->error = "Error allocating memory for cache"; | |
2765 | return -ENOMEM; | |
2766 | } | |
2767 | ca->ti = ti; | |
2768 | ||
2769 | r = parse_cache_args(ca, argc, argv, &ti->error); | |
2770 | if (r) | |
2771 | goto out; | |
2772 | ||
2773 | r = cache_create(ca, &cache); | |
617a0b89 HM |
2774 | if (r) |
2775 | goto out; | |
c6b4fcba JT |
2776 | |
2777 | r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3); | |
2778 | if (r) { | |
2779 | destroy(cache); | |
2780 | goto out; | |
2781 | } | |
2782 | ||
2783 | ti->private = cache; | |
c6b4fcba JT |
2784 | out: |
2785 | destroy_cache_args(ca); | |
2786 | return r; | |
2787 | } | |
2788 | ||
651f5fa2 JT |
2789 | /*----------------------------------------------------------------*/ |
2790 | ||
2791 | static int cache_map(struct dm_target *ti, struct bio *bio) | |
c6b4fcba | 2792 | { |
651f5fa2 JT |
2793 | struct cache *cache = ti->private; |
2794 | ||
c6b4fcba | 2795 | int r; |
b29d4986 | 2796 | bool commit_needed; |
c6b4fcba | 2797 | dm_oblock_t block = get_bio_block(cache, bio); |
19b0092e | 2798 | size_t pb_data_size = get_per_bio_data_size(cache); |
c6b4fcba | 2799 | |
b29d4986 | 2800 | init_per_bio_data(bio, pb_data_size); |
e893fba9 | 2801 | if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) { |
c6b4fcba JT |
2802 | /* |
2803 | * This can only occur if the io goes to a partial block at | |
2804 | * the end of the origin device. We don't cache these. | |
2805 | * Just remap to the origin and carry on. | |
2806 | */ | |
e893fba9 | 2807 | remap_to_origin(cache, bio); |
651f5fa2 | 2808 | accounted_begin(cache, bio); |
c6b4fcba JT |
2809 | return DM_MAPIO_REMAPPED; |
2810 | } | |
2811 | ||
651f5fa2 | 2812 | if (discard_or_flush(bio)) { |
c6b4fcba JT |
2813 | defer_bio(cache, bio); |
2814 | return DM_MAPIO_SUBMITTED; | |
2815 | } | |
2816 | ||
b29d4986 JT |
2817 | r = map_bio(cache, bio, block, &commit_needed); |
2818 | if (commit_needed) | |
2819 | schedule_commit(&cache->committer); | |
c6b4fcba | 2820 | |
2ee57d58 | 2821 | return r; |
c6b4fcba JT |
2822 | } |
2823 | ||
2824 | static int cache_end_io(struct dm_target *ti, struct bio *bio, int error) | |
2825 | { | |
2826 | struct cache *cache = ti->private; | |
2827 | unsigned long flags; | |
19b0092e MS |
2828 | size_t pb_data_size = get_per_bio_data_size(cache); |
2829 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
2830 | |
2831 | if (pb->tick) { | |
fba10109 | 2832 | policy_tick(cache->policy, false); |
c6b4fcba JT |
2833 | |
2834 | spin_lock_irqsave(&cache->lock, flags); | |
2835 | cache->need_tick_bio = true; | |
2836 | spin_unlock_irqrestore(&cache->lock, flags); | |
2837 | } | |
2838 | ||
b29d4986 | 2839 | bio_drop_shared_lock(cache, bio); |
066dbaa3 | 2840 | accounted_complete(cache, bio); |
c6b4fcba JT |
2841 | |
2842 | return 0; | |
2843 | } | |
2844 | ||
2845 | static int write_dirty_bitset(struct cache *cache) | |
2846 | { | |
629d0a8a | 2847 | int r; |
c6b4fcba | 2848 | |
028ae9f7 JT |
2849 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
2850 | return -EINVAL; | |
2851 | ||
629d0a8a JT |
2852 | r = dm_cache_set_dirty_bits(cache->cmd, from_cblock(cache->cache_size), cache->dirty_bitset); |
2853 | if (r) | |
2854 | metadata_operation_failed(cache, "dm_cache_set_dirty_bits", r); | |
c6b4fcba | 2855 | |
629d0a8a | 2856 | return r; |
c6b4fcba JT |
2857 | } |
2858 | ||
2859 | static int write_discard_bitset(struct cache *cache) | |
2860 | { | |
2861 | unsigned i, r; | |
2862 | ||
028ae9f7 JT |
2863 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
2864 | return -EINVAL; | |
2865 | ||
1bad9bc4 JT |
2866 | r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size, |
2867 | cache->discard_nr_blocks); | |
c6b4fcba | 2868 | if (r) { |
b61d9509 | 2869 | DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache)); |
028ae9f7 | 2870 | metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r); |
c6b4fcba JT |
2871 | return r; |
2872 | } | |
2873 | ||
1bad9bc4 JT |
2874 | for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) { |
2875 | r = dm_cache_set_discard(cache->cmd, to_dblock(i), | |
2876 | is_discarded(cache, to_dblock(i))); | |
028ae9f7 JT |
2877 | if (r) { |
2878 | metadata_operation_failed(cache, "dm_cache_set_discard", r); | |
c6b4fcba | 2879 | return r; |
028ae9f7 JT |
2880 | } |
2881 | } | |
2882 | ||
2883 | return 0; | |
2884 | } | |
2885 | ||
2886 | static int write_hints(struct cache *cache) | |
2887 | { | |
2888 | int r; | |
2889 | ||
2890 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
2891 | return -EINVAL; | |
2892 | ||
2893 | r = dm_cache_write_hints(cache->cmd, cache->policy); | |
2894 | if (r) { | |
2895 | metadata_operation_failed(cache, "dm_cache_write_hints", r); | |
2896 | return r; | |
c6b4fcba JT |
2897 | } |
2898 | ||
2899 | return 0; | |
2900 | } | |
2901 | ||
c6b4fcba JT |
2902 | /* |
2903 | * returns true on success | |
2904 | */ | |
2905 | static bool sync_metadata(struct cache *cache) | |
2906 | { | |
2907 | int r1, r2, r3, r4; | |
2908 | ||
2909 | r1 = write_dirty_bitset(cache); | |
2910 | if (r1) | |
b61d9509 | 2911 | DMERR("%s: could not write dirty bitset", cache_device_name(cache)); |
c6b4fcba JT |
2912 | |
2913 | r2 = write_discard_bitset(cache); | |
2914 | if (r2) | |
b61d9509 | 2915 | DMERR("%s: could not write discard bitset", cache_device_name(cache)); |
c6b4fcba JT |
2916 | |
2917 | save_stats(cache); | |
2918 | ||
028ae9f7 | 2919 | r3 = write_hints(cache); |
c6b4fcba | 2920 | if (r3) |
b61d9509 | 2921 | DMERR("%s: could not write hints", cache_device_name(cache)); |
c6b4fcba JT |
2922 | |
2923 | /* | |
2924 | * If writing the above metadata failed, we still commit, but don't | |
2925 | * set the clean shutdown flag. This will effectively force every | |
2926 | * dirty bit to be set on reload. | |
2927 | */ | |
028ae9f7 | 2928 | r4 = commit(cache, !r1 && !r2 && !r3); |
c6b4fcba | 2929 | if (r4) |
b61d9509 | 2930 | DMERR("%s: could not write cache metadata", cache_device_name(cache)); |
c6b4fcba JT |
2931 | |
2932 | return !r1 && !r2 && !r3 && !r4; | |
2933 | } | |
2934 | ||
2935 | static void cache_postsuspend(struct dm_target *ti) | |
2936 | { | |
2937 | struct cache *cache = ti->private; | |
2938 | ||
b29d4986 JT |
2939 | prevent_background_work(cache); |
2940 | BUG_ON(atomic_read(&cache->nr_io_migrations)); | |
2941 | ||
2942 | cancel_delayed_work(&cache->waker); | |
2943 | flush_workqueue(cache->wq); | |
2944 | WARN_ON(cache->origin_tracker.in_flight); | |
2945 | ||
2946 | /* | |
2947 | * If it's a flush suspend there won't be any deferred bios, so this | |
2948 | * call is harmless. | |
2949 | */ | |
651f5fa2 | 2950 | requeue_deferred_bios(cache); |
c6b4fcba | 2951 | |
028ae9f7 JT |
2952 | if (get_cache_mode(cache) == CM_WRITE) |
2953 | (void) sync_metadata(cache); | |
c6b4fcba JT |
2954 | } |
2955 | ||
2956 | static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock, | |
2957 | bool dirty, uint32_t hint, bool hint_valid) | |
2958 | { | |
2959 | int r; | |
2960 | struct cache *cache = context; | |
2961 | ||
449b668c JT |
2962 | if (dirty) { |
2963 | set_bit(from_cblock(cblock), cache->dirty_bitset); | |
2964 | atomic_inc(&cache->nr_dirty); | |
2965 | } else | |
2966 | clear_bit(from_cblock(cblock), cache->dirty_bitset); | |
2967 | ||
b29d4986 | 2968 | r = policy_load_mapping(cache->policy, oblock, cblock, dirty, hint, hint_valid); |
c6b4fcba JT |
2969 | if (r) |
2970 | return r; | |
2971 | ||
c6b4fcba JT |
2972 | return 0; |
2973 | } | |
2974 | ||
3e2e1c30 JT |
2975 | /* |
2976 | * The discard block size in the on disk metadata is not | |
2977 | * neccessarily the same as we're currently using. So we have to | |
2978 | * be careful to only set the discarded attribute if we know it | |
2979 | * covers a complete block of the new size. | |
2980 | */ | |
2981 | struct discard_load_info { | |
2982 | struct cache *cache; | |
2983 | ||
2984 | /* | |
2985 | * These blocks are sized using the on disk dblock size, rather | |
2986 | * than the current one. | |
2987 | */ | |
2988 | dm_block_t block_size; | |
2989 | dm_block_t discard_begin, discard_end; | |
2990 | }; | |
2991 | ||
2992 | static void discard_load_info_init(struct cache *cache, | |
2993 | struct discard_load_info *li) | |
2994 | { | |
2995 | li->cache = cache; | |
2996 | li->discard_begin = li->discard_end = 0; | |
2997 | } | |
2998 | ||
2999 | static void set_discard_range(struct discard_load_info *li) | |
3000 | { | |
3001 | sector_t b, e; | |
3002 | ||
3003 | if (li->discard_begin == li->discard_end) | |
3004 | return; | |
3005 | ||
3006 | /* | |
3007 | * Convert to sectors. | |
3008 | */ | |
3009 | b = li->discard_begin * li->block_size; | |
3010 | e = li->discard_end * li->block_size; | |
3011 | ||
3012 | /* | |
3013 | * Then convert back to the current dblock size. | |
3014 | */ | |
3015 | b = dm_sector_div_up(b, li->cache->discard_block_size); | |
3016 | sector_div(e, li->cache->discard_block_size); | |
3017 | ||
3018 | /* | |
3019 | * The origin may have shrunk, so we need to check we're still in | |
3020 | * bounds. | |
3021 | */ | |
3022 | if (e > from_dblock(li->cache->discard_nr_blocks)) | |
3023 | e = from_dblock(li->cache->discard_nr_blocks); | |
3024 | ||
3025 | for (; b < e; b++) | |
3026 | set_discard(li->cache, to_dblock(b)); | |
3027 | } | |
3028 | ||
c6b4fcba | 3029 | static int load_discard(void *context, sector_t discard_block_size, |
1bad9bc4 | 3030 | dm_dblock_t dblock, bool discard) |
c6b4fcba | 3031 | { |
3e2e1c30 | 3032 | struct discard_load_info *li = context; |
c6b4fcba | 3033 | |
3e2e1c30 | 3034 | li->block_size = discard_block_size; |
1bad9bc4 | 3035 | |
3e2e1c30 JT |
3036 | if (discard) { |
3037 | if (from_dblock(dblock) == li->discard_end) | |
3038 | /* | |
3039 | * We're already in a discard range, just extend it. | |
3040 | */ | |
3041 | li->discard_end = li->discard_end + 1ULL; | |
3042 | ||
3043 | else { | |
3044 | /* | |
3045 | * Emit the old range and start a new one. | |
3046 | */ | |
3047 | set_discard_range(li); | |
3048 | li->discard_begin = from_dblock(dblock); | |
3049 | li->discard_end = li->discard_begin + 1ULL; | |
3050 | } | |
3051 | } else { | |
3052 | set_discard_range(li); | |
3053 | li->discard_begin = li->discard_end = 0; | |
3054 | } | |
c6b4fcba JT |
3055 | |
3056 | return 0; | |
3057 | } | |
3058 | ||
f494a9c6 JT |
3059 | static dm_cblock_t get_cache_dev_size(struct cache *cache) |
3060 | { | |
3061 | sector_t size = get_dev_size(cache->cache_dev); | |
3062 | (void) sector_div(size, cache->sectors_per_block); | |
3063 | return to_cblock(size); | |
3064 | } | |
3065 | ||
3066 | static bool can_resize(struct cache *cache, dm_cblock_t new_size) | |
3067 | { | |
3068 | if (from_cblock(new_size) > from_cblock(cache->cache_size)) | |
3069 | return true; | |
3070 | ||
3071 | /* | |
3072 | * We can't drop a dirty block when shrinking the cache. | |
3073 | */ | |
3074 | while (from_cblock(new_size) < from_cblock(cache->cache_size)) { | |
3075 | new_size = to_cblock(from_cblock(new_size) + 1); | |
3076 | if (is_dirty(cache, new_size)) { | |
b61d9509 MS |
3077 | DMERR("%s: unable to shrink cache; cache block %llu is dirty", |
3078 | cache_device_name(cache), | |
f494a9c6 JT |
3079 | (unsigned long long) from_cblock(new_size)); |
3080 | return false; | |
3081 | } | |
3082 | } | |
3083 | ||
3084 | return true; | |
3085 | } | |
3086 | ||
3087 | static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size) | |
3088 | { | |
3089 | int r; | |
3090 | ||
08844800 | 3091 | r = dm_cache_resize(cache->cmd, new_size); |
f494a9c6 | 3092 | if (r) { |
b61d9509 | 3093 | DMERR("%s: could not resize cache metadata", cache_device_name(cache)); |
028ae9f7 | 3094 | metadata_operation_failed(cache, "dm_cache_resize", r); |
f494a9c6 JT |
3095 | return r; |
3096 | } | |
3097 | ||
d1d9220c | 3098 | set_cache_size(cache, new_size); |
f494a9c6 JT |
3099 | |
3100 | return 0; | |
3101 | } | |
3102 | ||
c6b4fcba JT |
3103 | static int cache_preresume(struct dm_target *ti) |
3104 | { | |
3105 | int r = 0; | |
3106 | struct cache *cache = ti->private; | |
f494a9c6 | 3107 | dm_cblock_t csize = get_cache_dev_size(cache); |
c6b4fcba JT |
3108 | |
3109 | /* | |
3110 | * Check to see if the cache has resized. | |
3111 | */ | |
f494a9c6 JT |
3112 | if (!cache->sized) { |
3113 | r = resize_cache_dev(cache, csize); | |
3114 | if (r) | |
c6b4fcba | 3115 | return r; |
c6b4fcba JT |
3116 | |
3117 | cache->sized = true; | |
f494a9c6 JT |
3118 | |
3119 | } else if (csize != cache->cache_size) { | |
3120 | if (!can_resize(cache, csize)) | |
3121 | return -EINVAL; | |
3122 | ||
3123 | r = resize_cache_dev(cache, csize); | |
3124 | if (r) | |
3125 | return r; | |
c6b4fcba JT |
3126 | } |
3127 | ||
3128 | if (!cache->loaded_mappings) { | |
ea2dd8c1 | 3129 | r = dm_cache_load_mappings(cache->cmd, cache->policy, |
c6b4fcba JT |
3130 | load_mapping, cache); |
3131 | if (r) { | |
b61d9509 | 3132 | DMERR("%s: could not load cache mappings", cache_device_name(cache)); |
028ae9f7 | 3133 | metadata_operation_failed(cache, "dm_cache_load_mappings", r); |
c6b4fcba JT |
3134 | return r; |
3135 | } | |
3136 | ||
3137 | cache->loaded_mappings = true; | |
3138 | } | |
3139 | ||
3140 | if (!cache->loaded_discards) { | |
3e2e1c30 JT |
3141 | struct discard_load_info li; |
3142 | ||
3143 | /* | |
3144 | * The discard bitset could have been resized, or the | |
3145 | * discard block size changed. To be safe we start by | |
3146 | * setting every dblock to not discarded. | |
3147 | */ | |
3148 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); | |
3149 | ||
3150 | discard_load_info_init(cache, &li); | |
3151 | r = dm_cache_load_discards(cache->cmd, load_discard, &li); | |
c6b4fcba | 3152 | if (r) { |
b61d9509 | 3153 | DMERR("%s: could not load origin discards", cache_device_name(cache)); |
028ae9f7 | 3154 | metadata_operation_failed(cache, "dm_cache_load_discards", r); |
c6b4fcba JT |
3155 | return r; |
3156 | } | |
3e2e1c30 | 3157 | set_discard_range(&li); |
c6b4fcba JT |
3158 | |
3159 | cache->loaded_discards = true; | |
3160 | } | |
3161 | ||
3162 | return r; | |
3163 | } | |
3164 | ||
3165 | static void cache_resume(struct dm_target *ti) | |
3166 | { | |
3167 | struct cache *cache = ti->private; | |
3168 | ||
3169 | cache->need_tick_bio = true; | |
b29d4986 | 3170 | allow_background_work(cache); |
c6b4fcba JT |
3171 | do_waker(&cache->waker.work); |
3172 | } | |
3173 | ||
3174 | /* | |
3175 | * Status format: | |
3176 | * | |
6a388618 MS |
3177 | * <metadata block size> <#used metadata blocks>/<#total metadata blocks> |
3178 | * <cache block size> <#used cache blocks>/<#total cache blocks> | |
c6b4fcba | 3179 | * <#read hits> <#read misses> <#write hits> <#write misses> |
6a388618 | 3180 | * <#demotions> <#promotions> <#dirty> |
c6b4fcba JT |
3181 | * <#features> <features>* |
3182 | * <#core args> <core args> | |
255eac20 | 3183 | * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check> |
c6b4fcba JT |
3184 | */ |
3185 | static void cache_status(struct dm_target *ti, status_type_t type, | |
3186 | unsigned status_flags, char *result, unsigned maxlen) | |
3187 | { | |
3188 | int r = 0; | |
3189 | unsigned i; | |
3190 | ssize_t sz = 0; | |
3191 | dm_block_t nr_free_blocks_metadata = 0; | |
3192 | dm_block_t nr_blocks_metadata = 0; | |
3193 | char buf[BDEVNAME_SIZE]; | |
3194 | struct cache *cache = ti->private; | |
3195 | dm_cblock_t residency; | |
d14fcf3d | 3196 | bool needs_check; |
c6b4fcba JT |
3197 | |
3198 | switch (type) { | |
3199 | case STATUSTYPE_INFO: | |
028ae9f7 JT |
3200 | if (get_cache_mode(cache) == CM_FAIL) { |
3201 | DMEMIT("Fail"); | |
3202 | break; | |
c6b4fcba JT |
3203 | } |
3204 | ||
028ae9f7 JT |
3205 | /* Commit to ensure statistics aren't out-of-date */ |
3206 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
3207 | (void) commit(cache, false); | |
3208 | ||
b61d9509 | 3209 | r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata); |
c6b4fcba | 3210 | if (r) { |
b61d9509 MS |
3211 | DMERR("%s: dm_cache_get_free_metadata_block_count returned %d", |
3212 | cache_device_name(cache), r); | |
c6b4fcba JT |
3213 | goto err; |
3214 | } | |
3215 | ||
3216 | r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata); | |
3217 | if (r) { | |
b61d9509 MS |
3218 | DMERR("%s: dm_cache_get_metadata_dev_size returned %d", |
3219 | cache_device_name(cache), r); | |
c6b4fcba JT |
3220 | goto err; |
3221 | } | |
3222 | ||
3223 | residency = policy_residency(cache->policy); | |
3224 | ||
ca763d0a | 3225 | DMEMIT("%u %llu/%llu %llu %llu/%llu %u %u %u %u %u %u %lu ", |
895b47d7 | 3226 | (unsigned)DM_CACHE_METADATA_BLOCK_SIZE, |
c6b4fcba JT |
3227 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
3228 | (unsigned long long)nr_blocks_metadata, | |
ca763d0a | 3229 | (unsigned long long)cache->sectors_per_block, |
6a388618 MS |
3230 | (unsigned long long) from_cblock(residency), |
3231 | (unsigned long long) from_cblock(cache->cache_size), | |
c6b4fcba JT |
3232 | (unsigned) atomic_read(&cache->stats.read_hit), |
3233 | (unsigned) atomic_read(&cache->stats.read_miss), | |
3234 | (unsigned) atomic_read(&cache->stats.write_hit), | |
3235 | (unsigned) atomic_read(&cache->stats.write_miss), | |
3236 | (unsigned) atomic_read(&cache->stats.demotion), | |
3237 | (unsigned) atomic_read(&cache->stats.promotion), | |
44fa816b | 3238 | (unsigned long) atomic_read(&cache->nr_dirty)); |
c6b4fcba | 3239 | |
629d0a8a JT |
3240 | if (cache->features.metadata_version == 2) |
3241 | DMEMIT("2 metadata2 "); | |
3242 | else | |
3243 | DMEMIT("1 "); | |
3244 | ||
2ee57d58 | 3245 | if (writethrough_mode(&cache->features)) |
629d0a8a | 3246 | DMEMIT("writethrough "); |
2ee57d58 JT |
3247 | |
3248 | else if (passthrough_mode(&cache->features)) | |
629d0a8a | 3249 | DMEMIT("passthrough "); |
2ee57d58 JT |
3250 | |
3251 | else if (writeback_mode(&cache->features)) | |
629d0a8a | 3252 | DMEMIT("writeback "); |
2ee57d58 JT |
3253 | |
3254 | else { | |
b61d9509 MS |
3255 | DMERR("%s: internal error: unknown io mode: %d", |
3256 | cache_device_name(cache), (int) cache->features.io_mode); | |
2ee57d58 JT |
3257 | goto err; |
3258 | } | |
c6b4fcba JT |
3259 | |
3260 | DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold); | |
2e68c4e6 MS |
3261 | |
3262 | DMEMIT("%s ", dm_cache_policy_get_name(cache->policy)); | |
c6b4fcba | 3263 | if (sz < maxlen) { |
028ae9f7 | 3264 | r = policy_emit_config_values(cache->policy, result, maxlen, &sz); |
c6b4fcba | 3265 | if (r) |
b61d9509 MS |
3266 | DMERR("%s: policy_emit_config_values returned %d", |
3267 | cache_device_name(cache), r); | |
c6b4fcba JT |
3268 | } |
3269 | ||
028ae9f7 JT |
3270 | if (get_cache_mode(cache) == CM_READ_ONLY) |
3271 | DMEMIT("ro "); | |
3272 | else | |
3273 | DMEMIT("rw "); | |
3274 | ||
d14fcf3d JT |
3275 | r = dm_cache_metadata_needs_check(cache->cmd, &needs_check); |
3276 | ||
3277 | if (r || needs_check) | |
255eac20 MS |
3278 | DMEMIT("needs_check "); |
3279 | else | |
3280 | DMEMIT("- "); | |
3281 | ||
c6b4fcba JT |
3282 | break; |
3283 | ||
3284 | case STATUSTYPE_TABLE: | |
3285 | format_dev_t(buf, cache->metadata_dev->bdev->bd_dev); | |
3286 | DMEMIT("%s ", buf); | |
3287 | format_dev_t(buf, cache->cache_dev->bdev->bd_dev); | |
3288 | DMEMIT("%s ", buf); | |
3289 | format_dev_t(buf, cache->origin_dev->bdev->bd_dev); | |
3290 | DMEMIT("%s", buf); | |
3291 | ||
3292 | for (i = 0; i < cache->nr_ctr_args - 1; i++) | |
3293 | DMEMIT(" %s", cache->ctr_args[i]); | |
3294 | if (cache->nr_ctr_args) | |
3295 | DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]); | |
3296 | } | |
3297 | ||
3298 | return; | |
3299 | ||
3300 | err: | |
3301 | DMEMIT("Error"); | |
3302 | } | |
3303 | ||
b29d4986 JT |
3304 | /* |
3305 | * Defines a range of cblocks, begin to (end - 1) are in the range. end is | |
3306 | * the one-past-the-end value. | |
3307 | */ | |
3308 | struct cblock_range { | |
3309 | dm_cblock_t begin; | |
3310 | dm_cblock_t end; | |
3311 | }; | |
3312 | ||
c6b4fcba | 3313 | /* |
65790ff9 JT |
3314 | * A cache block range can take two forms: |
3315 | * | |
3316 | * i) A single cblock, eg. '3456' | |
b29d4986 | 3317 | * ii) A begin and end cblock with a dash between, eg. 123-234 |
65790ff9 JT |
3318 | */ |
3319 | static int parse_cblock_range(struct cache *cache, const char *str, | |
3320 | struct cblock_range *result) | |
3321 | { | |
3322 | char dummy; | |
3323 | uint64_t b, e; | |
3324 | int r; | |
3325 | ||
3326 | /* | |
3327 | * Try and parse form (ii) first. | |
3328 | */ | |
3329 | r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy); | |
3330 | if (r < 0) | |
3331 | return r; | |
3332 | ||
3333 | if (r == 2) { | |
3334 | result->begin = to_cblock(b); | |
3335 | result->end = to_cblock(e); | |
3336 | return 0; | |
3337 | } | |
3338 | ||
3339 | /* | |
3340 | * That didn't work, try form (i). | |
3341 | */ | |
3342 | r = sscanf(str, "%llu%c", &b, &dummy); | |
3343 | if (r < 0) | |
3344 | return r; | |
3345 | ||
3346 | if (r == 1) { | |
3347 | result->begin = to_cblock(b); | |
3348 | result->end = to_cblock(from_cblock(result->begin) + 1u); | |
3349 | return 0; | |
3350 | } | |
3351 | ||
b61d9509 | 3352 | DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str); |
65790ff9 JT |
3353 | return -EINVAL; |
3354 | } | |
3355 | ||
3356 | static int validate_cblock_range(struct cache *cache, struct cblock_range *range) | |
3357 | { | |
3358 | uint64_t b = from_cblock(range->begin); | |
3359 | uint64_t e = from_cblock(range->end); | |
3360 | uint64_t n = from_cblock(cache->cache_size); | |
3361 | ||
3362 | if (b >= n) { | |
b61d9509 MS |
3363 | DMERR("%s: begin cblock out of range: %llu >= %llu", |
3364 | cache_device_name(cache), b, n); | |
65790ff9 JT |
3365 | return -EINVAL; |
3366 | } | |
3367 | ||
3368 | if (e > n) { | |
b61d9509 MS |
3369 | DMERR("%s: end cblock out of range: %llu > %llu", |
3370 | cache_device_name(cache), e, n); | |
65790ff9 JT |
3371 | return -EINVAL; |
3372 | } | |
3373 | ||
3374 | if (b >= e) { | |
b61d9509 MS |
3375 | DMERR("%s: invalid cblock range: %llu >= %llu", |
3376 | cache_device_name(cache), b, e); | |
65790ff9 JT |
3377 | return -EINVAL; |
3378 | } | |
3379 | ||
3380 | return 0; | |
3381 | } | |
3382 | ||
b29d4986 JT |
3383 | static inline dm_cblock_t cblock_succ(dm_cblock_t b) |
3384 | { | |
3385 | return to_cblock(from_cblock(b) + 1); | |
3386 | } | |
3387 | ||
65790ff9 JT |
3388 | static int request_invalidation(struct cache *cache, struct cblock_range *range) |
3389 | { | |
b29d4986 | 3390 | int r = 0; |
65790ff9 | 3391 | |
b29d4986 JT |
3392 | /* |
3393 | * We don't need to do any locking here because we know we're in | |
3394 | * passthrough mode. There's is potential for a race between an | |
3395 | * invalidation triggered by an io and an invalidation message. This | |
3396 | * is harmless, we must not worry if the policy call fails. | |
3397 | */ | |
3398 | while (range->begin != range->end) { | |
3399 | r = invalidate_cblock(cache, range->begin); | |
3400 | if (r) | |
3401 | return r; | |
65790ff9 | 3402 | |
b29d4986 JT |
3403 | range->begin = cblock_succ(range->begin); |
3404 | } | |
65790ff9 | 3405 | |
b29d4986 JT |
3406 | cache->commit_requested = true; |
3407 | return r; | |
65790ff9 JT |
3408 | } |
3409 | ||
3410 | static int process_invalidate_cblocks_message(struct cache *cache, unsigned count, | |
3411 | const char **cblock_ranges) | |
3412 | { | |
3413 | int r = 0; | |
3414 | unsigned i; | |
3415 | struct cblock_range range; | |
3416 | ||
3417 | if (!passthrough_mode(&cache->features)) { | |
b61d9509 MS |
3418 | DMERR("%s: cache has to be in passthrough mode for invalidation", |
3419 | cache_device_name(cache)); | |
65790ff9 JT |
3420 | return -EPERM; |
3421 | } | |
3422 | ||
3423 | for (i = 0; i < count; i++) { | |
3424 | r = parse_cblock_range(cache, cblock_ranges[i], &range); | |
3425 | if (r) | |
3426 | break; | |
3427 | ||
3428 | r = validate_cblock_range(cache, &range); | |
3429 | if (r) | |
3430 | break; | |
3431 | ||
3432 | /* | |
3433 | * Pass begin and end origin blocks to the worker and wake it. | |
3434 | */ | |
3435 | r = request_invalidation(cache, &range); | |
3436 | if (r) | |
3437 | break; | |
3438 | } | |
3439 | ||
3440 | return r; | |
3441 | } | |
3442 | ||
3443 | /* | |
3444 | * Supports | |
3445 | * "<key> <value>" | |
3446 | * and | |
3447 | * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]* | |
c6b4fcba JT |
3448 | * |
3449 | * The key migration_threshold is supported by the cache target core. | |
3450 | */ | |
3451 | static int cache_message(struct dm_target *ti, unsigned argc, char **argv) | |
3452 | { | |
c6b4fcba JT |
3453 | struct cache *cache = ti->private; |
3454 | ||
65790ff9 JT |
3455 | if (!argc) |
3456 | return -EINVAL; | |
3457 | ||
028ae9f7 | 3458 | if (get_cache_mode(cache) >= CM_READ_ONLY) { |
b61d9509 MS |
3459 | DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode", |
3460 | cache_device_name(cache)); | |
028ae9f7 JT |
3461 | return -EOPNOTSUPP; |
3462 | } | |
3463 | ||
7b6b2bc9 | 3464 | if (!strcasecmp(argv[0], "invalidate_cblocks")) |
65790ff9 JT |
3465 | return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1); |
3466 | ||
c6b4fcba JT |
3467 | if (argc != 2) |
3468 | return -EINVAL; | |
3469 | ||
2f14f4b5 | 3470 | return set_config_value(cache, argv[0], argv[1]); |
c6b4fcba JT |
3471 | } |
3472 | ||
3473 | static int cache_iterate_devices(struct dm_target *ti, | |
3474 | iterate_devices_callout_fn fn, void *data) | |
3475 | { | |
3476 | int r = 0; | |
3477 | struct cache *cache = ti->private; | |
3478 | ||
3479 | r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data); | |
3480 | if (!r) | |
3481 | r = fn(ti, cache->origin_dev, 0, ti->len, data); | |
3482 | ||
3483 | return r; | |
3484 | } | |
3485 | ||
c6b4fcba JT |
3486 | static void set_discard_limits(struct cache *cache, struct queue_limits *limits) |
3487 | { | |
3488 | /* | |
3489 | * FIXME: these limits may be incompatible with the cache device | |
3490 | */ | |
7ae34e77 JT |
3491 | limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024, |
3492 | cache->origin_sectors); | |
1bad9bc4 | 3493 | limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT; |
c6b4fcba JT |
3494 | } |
3495 | ||
3496 | static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3497 | { | |
3498 | struct cache *cache = ti->private; | |
f6109372 | 3499 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
c6b4fcba | 3500 | |
f6109372 MS |
3501 | /* |
3502 | * If the system-determined stacked limits are compatible with the | |
3503 | * cache's blocksize (io_opt is a factor) do not override them. | |
3504 | */ | |
3505 | if (io_opt_sectors < cache->sectors_per_block || | |
3506 | do_div(io_opt_sectors, cache->sectors_per_block)) { | |
b0246530 | 3507 | blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT); |
f6109372 MS |
3508 | blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT); |
3509 | } | |
c6b4fcba JT |
3510 | set_discard_limits(cache, limits); |
3511 | } | |
3512 | ||
3513 | /*----------------------------------------------------------------*/ | |
3514 | ||
3515 | static struct target_type cache_target = { | |
3516 | .name = "cache", | |
b29d4986 | 3517 | .version = {2, 0, 0}, |
c6b4fcba JT |
3518 | .module = THIS_MODULE, |
3519 | .ctr = cache_ctr, | |
3520 | .dtr = cache_dtr, | |
3521 | .map = cache_map, | |
3522 | .end_io = cache_end_io, | |
3523 | .postsuspend = cache_postsuspend, | |
3524 | .preresume = cache_preresume, | |
3525 | .resume = cache_resume, | |
3526 | .status = cache_status, | |
3527 | .message = cache_message, | |
3528 | .iterate_devices = cache_iterate_devices, | |
c6b4fcba JT |
3529 | .io_hints = cache_io_hints, |
3530 | }; | |
3531 | ||
3532 | static int __init dm_cache_init(void) | |
3533 | { | |
3534 | int r; | |
3535 | ||
3536 | r = dm_register_target(&cache_target); | |
3537 | if (r) { | |
3538 | DMERR("cache target registration failed: %d", r); | |
3539 | return r; | |
3540 | } | |
3541 | ||
3542 | migration_cache = KMEM_CACHE(dm_cache_migration, 0); | |
3543 | if (!migration_cache) { | |
3544 | dm_unregister_target(&cache_target); | |
3545 | return -ENOMEM; | |
3546 | } | |
3547 | ||
3548 | return 0; | |
3549 | } | |
3550 | ||
3551 | static void __exit dm_cache_exit(void) | |
3552 | { | |
3553 | dm_unregister_target(&cache_target); | |
3554 | kmem_cache_destroy(migration_cache); | |
3555 | } | |
3556 | ||
3557 | module_init(dm_cache_init); | |
3558 | module_exit(dm_cache_exit); | |
3559 | ||
3560 | MODULE_DESCRIPTION(DM_NAME " cache target"); | |
3561 | MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>"); | |
3562 | MODULE_LICENSE("GPL"); |