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 | { | |
072792dc JT |
97 | if (!len) |
98 | return; | |
99 | ||
77289d32 JT |
100 | iot->in_flight -= len; |
101 | if (!iot->in_flight) | |
102 | iot->idle_time = jiffies; | |
103 | } | |
104 | ||
105 | static void iot_io_end(struct io_tracker *iot, sector_t len) | |
106 | { | |
107 | unsigned long flags; | |
108 | ||
109 | spin_lock_irqsave(&iot->lock, flags); | |
110 | __iot_io_end(iot, len); | |
111 | spin_unlock_irqrestore(&iot->lock, flags); | |
112 | } | |
113 | ||
114 | /*----------------------------------------------------------------*/ | |
115 | ||
c6b4fcba | 116 | /* |
b29d4986 JT |
117 | * Represents a chunk of future work. 'input' allows continuations to pass |
118 | * values between themselves, typically error values. | |
c6b4fcba | 119 | */ |
b29d4986 JT |
120 | struct continuation { |
121 | struct work_struct ws; | |
4e4cbee9 | 122 | blk_status_t input; |
b29d4986 JT |
123 | }; |
124 | ||
125 | static inline void init_continuation(struct continuation *k, | |
126 | void (*fn)(struct work_struct *)) | |
127 | { | |
128 | INIT_WORK(&k->ws, fn); | |
129 | k->input = 0; | |
130 | } | |
131 | ||
132 | static inline void queue_continuation(struct workqueue_struct *wq, | |
133 | struct continuation *k) | |
134 | { | |
135 | queue_work(wq, &k->ws); | |
136 | } | |
c6b4fcba JT |
137 | |
138 | /*----------------------------------------------------------------*/ | |
139 | ||
b29d4986 JT |
140 | /* |
141 | * The batcher collects together pieces of work that need a particular | |
142 | * operation to occur before they can proceed (typically a commit). | |
143 | */ | |
144 | struct batcher { | |
145 | /* | |
146 | * The operation that everyone is waiting for. | |
147 | */ | |
4e4cbee9 | 148 | blk_status_t (*commit_op)(void *context); |
b29d4986 JT |
149 | void *commit_context; |
150 | ||
151 | /* | |
152 | * This is how bios should be issued once the commit op is complete | |
153 | * (accounted_request). | |
154 | */ | |
155 | void (*issue_op)(struct bio *bio, void *context); | |
156 | void *issue_context; | |
157 | ||
158 | /* | |
159 | * Queued work gets put on here after commit. | |
160 | */ | |
161 | struct workqueue_struct *wq; | |
162 | ||
163 | spinlock_t lock; | |
164 | struct list_head work_items; | |
165 | struct bio_list bios; | |
166 | struct work_struct commit_work; | |
167 | ||
168 | bool commit_scheduled; | |
169 | }; | |
170 | ||
171 | static void __commit(struct work_struct *_ws) | |
172 | { | |
173 | struct batcher *b = container_of(_ws, struct batcher, commit_work); | |
4e4cbee9 | 174 | blk_status_t r; |
b29d4986 JT |
175 | unsigned long flags; |
176 | struct list_head work_items; | |
177 | struct work_struct *ws, *tmp; | |
178 | struct continuation *k; | |
179 | struct bio *bio; | |
180 | struct bio_list bios; | |
181 | ||
182 | INIT_LIST_HEAD(&work_items); | |
183 | bio_list_init(&bios); | |
184 | ||
185 | /* | |
186 | * We have to grab these before the commit_op to avoid a race | |
187 | * condition. | |
188 | */ | |
189 | spin_lock_irqsave(&b->lock, flags); | |
190 | list_splice_init(&b->work_items, &work_items); | |
191 | bio_list_merge(&bios, &b->bios); | |
192 | bio_list_init(&b->bios); | |
193 | b->commit_scheduled = false; | |
194 | spin_unlock_irqrestore(&b->lock, flags); | |
195 | ||
196 | r = b->commit_op(b->commit_context); | |
197 | ||
198 | list_for_each_entry_safe(ws, tmp, &work_items, entry) { | |
199 | k = container_of(ws, struct continuation, ws); | |
200 | k->input = r; | |
201 | INIT_LIST_HEAD(&ws->entry); /* to avoid a WARN_ON */ | |
202 | queue_work(b->wq, ws); | |
203 | } | |
204 | ||
205 | while ((bio = bio_list_pop(&bios))) { | |
206 | if (r) { | |
4e4cbee9 | 207 | bio->bi_status = r; |
b29d4986 JT |
208 | bio_endio(bio); |
209 | } else | |
210 | b->issue_op(bio, b->issue_context); | |
211 | } | |
212 | } | |
213 | ||
214 | static void batcher_init(struct batcher *b, | |
4e4cbee9 | 215 | blk_status_t (*commit_op)(void *), |
b29d4986 JT |
216 | void *commit_context, |
217 | void (*issue_op)(struct bio *bio, void *), | |
218 | void *issue_context, | |
219 | struct workqueue_struct *wq) | |
220 | { | |
221 | b->commit_op = commit_op; | |
222 | b->commit_context = commit_context; | |
223 | b->issue_op = issue_op; | |
224 | b->issue_context = issue_context; | |
225 | b->wq = wq; | |
226 | ||
227 | spin_lock_init(&b->lock); | |
228 | INIT_LIST_HEAD(&b->work_items); | |
229 | bio_list_init(&b->bios); | |
230 | INIT_WORK(&b->commit_work, __commit); | |
231 | b->commit_scheduled = false; | |
232 | } | |
233 | ||
234 | static void async_commit(struct batcher *b) | |
235 | { | |
236 | queue_work(b->wq, &b->commit_work); | |
237 | } | |
238 | ||
239 | static void continue_after_commit(struct batcher *b, struct continuation *k) | |
240 | { | |
241 | unsigned long flags; | |
242 | bool commit_scheduled; | |
243 | ||
244 | spin_lock_irqsave(&b->lock, flags); | |
245 | commit_scheduled = b->commit_scheduled; | |
246 | list_add_tail(&k->ws.entry, &b->work_items); | |
247 | spin_unlock_irqrestore(&b->lock, flags); | |
248 | ||
249 | if (commit_scheduled) | |
250 | async_commit(b); | |
251 | } | |
252 | ||
253 | /* | |
254 | * Bios are errored if commit failed. | |
255 | */ | |
256 | static void issue_after_commit(struct batcher *b, struct bio *bio) | |
257 | { | |
258 | unsigned long flags; | |
259 | bool commit_scheduled; | |
260 | ||
261 | spin_lock_irqsave(&b->lock, flags); | |
262 | commit_scheduled = b->commit_scheduled; | |
263 | bio_list_add(&b->bios, bio); | |
264 | spin_unlock_irqrestore(&b->lock, flags); | |
265 | ||
266 | if (commit_scheduled) | |
267 | async_commit(b); | |
268 | } | |
269 | ||
270 | /* | |
271 | * Call this if some urgent work is waiting for the commit to complete. | |
272 | */ | |
273 | static void schedule_commit(struct batcher *b) | |
274 | { | |
275 | bool immediate; | |
276 | unsigned long flags; | |
277 | ||
278 | spin_lock_irqsave(&b->lock, flags); | |
279 | immediate = !list_empty(&b->work_items) || !bio_list_empty(&b->bios); | |
280 | b->commit_scheduled = true; | |
281 | spin_unlock_irqrestore(&b->lock, flags); | |
282 | ||
283 | if (immediate) | |
284 | async_commit(b); | |
285 | } | |
286 | ||
c9d28d5d JT |
287 | /* |
288 | * There are a couple of places where we let a bio run, but want to do some | |
289 | * work before calling its endio function. We do this by temporarily | |
290 | * changing the endio fn. | |
291 | */ | |
292 | struct dm_hook_info { | |
293 | bio_end_io_t *bi_end_io; | |
c9d28d5d JT |
294 | }; |
295 | ||
296 | static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio, | |
297 | bio_end_io_t *bi_end_io, void *bi_private) | |
298 | { | |
299 | h->bi_end_io = bio->bi_end_io; | |
c9d28d5d JT |
300 | |
301 | bio->bi_end_io = bi_end_io; | |
302 | bio->bi_private = bi_private; | |
303 | } | |
304 | ||
305 | static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio) | |
306 | { | |
307 | bio->bi_end_io = h->bi_end_io; | |
c9d28d5d JT |
308 | } |
309 | ||
310 | /*----------------------------------------------------------------*/ | |
311 | ||
c6b4fcba JT |
312 | #define MIGRATION_POOL_SIZE 128 |
313 | #define COMMIT_PERIOD HZ | |
314 | #define MIGRATION_COUNT_WINDOW 10 | |
315 | ||
316 | /* | |
05473044 MS |
317 | * The block size of the device holding cache data must be |
318 | * between 32KB and 1GB. | |
c6b4fcba JT |
319 | */ |
320 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT) | |
05473044 | 321 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
c6b4fcba | 322 | |
2ee57d58 | 323 | enum cache_metadata_mode { |
c6b4fcba JT |
324 | CM_WRITE, /* metadata may be changed */ |
325 | CM_READ_ONLY, /* metadata may not be changed */ | |
028ae9f7 | 326 | CM_FAIL |
c6b4fcba JT |
327 | }; |
328 | ||
2ee57d58 JT |
329 | enum cache_io_mode { |
330 | /* | |
331 | * Data is written to cached blocks only. These blocks are marked | |
332 | * dirty. If you lose the cache device you will lose data. | |
333 | * Potential performance increase for both reads and writes. | |
334 | */ | |
335 | CM_IO_WRITEBACK, | |
336 | ||
337 | /* | |
338 | * Data is written to both cache and origin. Blocks are never | |
339 | * dirty. Potential performance benfit for reads only. | |
340 | */ | |
341 | CM_IO_WRITETHROUGH, | |
342 | ||
343 | /* | |
344 | * A degraded mode useful for various cache coherency situations | |
345 | * (eg, rolling back snapshots). Reads and writes always go to the | |
346 | * origin. If a write goes to a cached oblock, then the cache | |
347 | * block is invalidated. | |
348 | */ | |
349 | CM_IO_PASSTHROUGH | |
350 | }; | |
351 | ||
c6b4fcba | 352 | struct cache_features { |
2ee57d58 JT |
353 | enum cache_metadata_mode mode; |
354 | enum cache_io_mode io_mode; | |
629d0a8a | 355 | unsigned metadata_version; |
de7180ff | 356 | bool discard_passdown:1; |
c6b4fcba JT |
357 | }; |
358 | ||
359 | struct cache_stats { | |
360 | atomic_t read_hit; | |
361 | atomic_t read_miss; | |
362 | atomic_t write_hit; | |
363 | atomic_t write_miss; | |
364 | atomic_t demotion; | |
365 | atomic_t promotion; | |
b29d4986 | 366 | atomic_t writeback; |
c6b4fcba JT |
367 | atomic_t copies_avoided; |
368 | atomic_t cache_cell_clash; | |
369 | atomic_t commit_count; | |
370 | atomic_t discard_count; | |
371 | }; | |
372 | ||
373 | struct cache { | |
374 | struct dm_target *ti; | |
72d711c8 MS |
375 | spinlock_t lock; |
376 | ||
377 | /* | |
378 | * Fields for converting from sectors to blocks. | |
379 | */ | |
380 | int sectors_per_block_shift; | |
381 | sector_t sectors_per_block; | |
c6b4fcba | 382 | |
c9ec5d7c MS |
383 | struct dm_cache_metadata *cmd; |
384 | ||
c6b4fcba JT |
385 | /* |
386 | * Metadata is written to this device. | |
387 | */ | |
388 | struct dm_dev *metadata_dev; | |
389 | ||
390 | /* | |
391 | * The slower of the two data devices. Typically a spindle. | |
392 | */ | |
393 | struct dm_dev *origin_dev; | |
394 | ||
395 | /* | |
396 | * The faster of the two data devices. Typically an SSD. | |
397 | */ | |
398 | struct dm_dev *cache_dev; | |
399 | ||
c6b4fcba JT |
400 | /* |
401 | * Size of the origin device in _complete_ blocks and native sectors. | |
402 | */ | |
403 | dm_oblock_t origin_blocks; | |
404 | sector_t origin_sectors; | |
405 | ||
406 | /* | |
407 | * Size of the cache device in blocks. | |
408 | */ | |
409 | dm_cblock_t cache_size; | |
410 | ||
411 | /* | |
72d711c8 | 412 | * Invalidation fields. |
c6b4fcba | 413 | */ |
72d711c8 MS |
414 | spinlock_t invalidation_lock; |
415 | struct list_head invalidation_requests; | |
c6b4fcba | 416 | |
c6b4fcba | 417 | sector_t migration_threshold; |
c6b4fcba | 418 | wait_queue_head_t migration_wait; |
a59db676 JT |
419 | atomic_t nr_allocated_migrations; |
420 | ||
421 | /* | |
422 | * The number of in flight migrations that are performing | |
423 | * background io. eg, promotion, writeback. | |
424 | */ | |
425 | atomic_t nr_io_migrations; | |
c6b4fcba | 426 | |
72d711c8 MS |
427 | struct bio_list deferred_bios; |
428 | ||
b29d4986 | 429 | struct rw_semaphore quiesce_lock; |
66cb1910 | 430 | |
72d711c8 | 431 | struct dm_target_callbacks callbacks; |
c6b4fcba JT |
432 | |
433 | /* | |
434 | * origin_blocks entries, discarded if set. | |
435 | */ | |
1bad9bc4 | 436 | dm_dblock_t discard_nr_blocks; |
c6b4fcba | 437 | unsigned long *discard_bitset; |
08b18451 | 438 | uint32_t discard_block_size; /* a power of 2 times sectors per block */ |
c9ec5d7c MS |
439 | |
440 | /* | |
441 | * Rather than reconstructing the table line for the status we just | |
442 | * save it and regurgitate. | |
443 | */ | |
444 | unsigned nr_ctr_args; | |
445 | const char **ctr_args; | |
c6b4fcba JT |
446 | |
447 | struct dm_kcopyd_client *copier; | |
b29d4986 | 448 | struct work_struct deferred_bio_worker; |
b29d4986 | 449 | struct work_struct migration_worker; |
72d711c8 | 450 | struct workqueue_struct *wq; |
c6b4fcba | 451 | struct delayed_work waker; |
b29d4986 | 452 | struct dm_bio_prison_v2 *prison; |
c6b4fcba | 453 | |
72d711c8 MS |
454 | /* |
455 | * cache_size entries, dirty if set | |
456 | */ | |
457 | unsigned long *dirty_bitset; | |
458 | atomic_t nr_dirty; | |
c6b4fcba | 459 | |
c6b4fcba | 460 | unsigned policy_nr_args; |
72d711c8 MS |
461 | struct dm_cache_policy *policy; |
462 | ||
463 | /* | |
464 | * Cache features such as write-through. | |
465 | */ | |
466 | struct cache_features features; | |
467 | ||
468 | struct cache_stats stats; | |
c6b4fcba JT |
469 | |
470 | bool need_tick_bio:1; | |
471 | bool sized:1; | |
65790ff9 | 472 | bool invalidate:1; |
c6b4fcba JT |
473 | bool commit_requested:1; |
474 | bool loaded_mappings:1; | |
475 | bool loaded_discards:1; | |
476 | ||
72d711c8 | 477 | struct rw_semaphore background_work_lock; |
65790ff9 | 478 | |
72d711c8 MS |
479 | struct batcher committer; |
480 | struct work_struct commit_ws; | |
066dbaa3 | 481 | |
701e03e4 | 482 | struct io_tracker tracker; |
b29d4986 | 483 | |
72d711c8 | 484 | mempool_t migration_pool; |
b29d4986 | 485 | |
72d711c8 | 486 | struct bio_set bs; |
c6b4fcba JT |
487 | }; |
488 | ||
489 | struct per_bio_data { | |
490 | bool tick:1; | |
491 | unsigned req_nr:2; | |
b29d4986 | 492 | struct dm_bio_prison_cell_v2 *cell; |
c6eda5e8 | 493 | struct dm_hook_info hook_info; |
066dbaa3 | 494 | sector_t len; |
c6b4fcba JT |
495 | }; |
496 | ||
497 | struct dm_cache_migration { | |
b29d4986 | 498 | struct continuation k; |
c6b4fcba JT |
499 | struct cache *cache; |
500 | ||
b29d4986 JT |
501 | struct policy_work *op; |
502 | struct bio *overwrite_bio; | |
503 | struct dm_bio_prison_cell_v2 *cell; | |
c6b4fcba | 504 | |
b29d4986 JT |
505 | dm_cblock_t invalidate_cblock; |
506 | dm_oblock_t invalidate_oblock; | |
c6b4fcba JT |
507 | }; |
508 | ||
b29d4986 JT |
509 | /*----------------------------------------------------------------*/ |
510 | ||
8e3c3827 | 511 | static bool writethrough_mode(struct cache *cache) |
b29d4986 | 512 | { |
8e3c3827 | 513 | return cache->features.io_mode == CM_IO_WRITETHROUGH; |
b29d4986 JT |
514 | } |
515 | ||
8e3c3827 | 516 | static bool writeback_mode(struct cache *cache) |
b29d4986 | 517 | { |
8e3c3827 | 518 | return cache->features.io_mode == CM_IO_WRITEBACK; |
b29d4986 JT |
519 | } |
520 | ||
8e3c3827 | 521 | static inline bool passthrough_mode(struct cache *cache) |
b29d4986 | 522 | { |
8e3c3827 | 523 | return unlikely(cache->features.io_mode == CM_IO_PASSTHROUGH); |
b29d4986 JT |
524 | } |
525 | ||
526 | /*----------------------------------------------------------------*/ | |
527 | ||
528 | static void wake_deferred_bio_worker(struct cache *cache) | |
529 | { | |
530 | queue_work(cache->wq, &cache->deferred_bio_worker); | |
531 | } | |
c6b4fcba | 532 | |
b29d4986 | 533 | static void wake_migration_worker(struct cache *cache) |
c6b4fcba | 534 | { |
8e3c3827 | 535 | if (passthrough_mode(cache)) |
b29d4986 JT |
536 | return; |
537 | ||
538 | queue_work(cache->wq, &cache->migration_worker); | |
c6b4fcba JT |
539 | } |
540 | ||
541 | /*----------------------------------------------------------------*/ | |
542 | ||
b29d4986 | 543 | static struct dm_bio_prison_cell_v2 *alloc_prison_cell(struct cache *cache) |
c6b4fcba | 544 | { |
b29d4986 | 545 | return dm_bio_prison_alloc_cell_v2(cache->prison, GFP_NOWAIT); |
c6b4fcba JT |
546 | } |
547 | ||
b29d4986 | 548 | static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell_v2 *cell) |
c6b4fcba | 549 | { |
b29d4986 | 550 | dm_bio_prison_free_cell_v2(cache->prison, cell); |
c6b4fcba JT |
551 | } |
552 | ||
a59db676 JT |
553 | static struct dm_cache_migration *alloc_migration(struct cache *cache) |
554 | { | |
555 | struct dm_cache_migration *mg; | |
556 | ||
6f1c819c | 557 | mg = mempool_alloc(&cache->migration_pool, GFP_NOWAIT); |
ef7afb36 MS |
558 | if (!mg) |
559 | return NULL; | |
560 | ||
561 | memset(mg, 0, sizeof(*mg)); | |
562 | ||
563 | mg->cache = cache; | |
564 | atomic_inc(&cache->nr_allocated_migrations); | |
a59db676 JT |
565 | |
566 | return mg; | |
567 | } | |
568 | ||
569 | static void free_migration(struct dm_cache_migration *mg) | |
570 | { | |
88bf5184 | 571 | struct cache *cache = mg->cache; |
a59db676 | 572 | |
88bf5184 JT |
573 | if (atomic_dec_and_test(&cache->nr_allocated_migrations)) |
574 | wake_up(&cache->migration_wait); | |
575 | ||
6f1c819c | 576 | mempool_free(mg, &cache->migration_pool); |
a59db676 JT |
577 | } |
578 | ||
b29d4986 | 579 | /*----------------------------------------------------------------*/ |
c6b4fcba | 580 | |
b29d4986 | 581 | static inline dm_oblock_t oblock_succ(dm_oblock_t b) |
c6b4fcba | 582 | { |
b29d4986 | 583 | return to_oblock(from_oblock(b) + 1ull); |
c6b4fcba JT |
584 | } |
585 | ||
b29d4986 | 586 | static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key_v2 *key) |
c6b4fcba | 587 | { |
b29d4986 JT |
588 | key->virtual = 0; |
589 | key->dev = 0; | |
590 | key->block_begin = from_oblock(begin); | |
591 | key->block_end = from_oblock(end); | |
c6b4fcba JT |
592 | } |
593 | ||
594 | /* | |
b29d4986 JT |
595 | * We have two lock levels. Level 0, which is used to prevent WRITEs, and |
596 | * level 1 which prevents *both* READs and WRITEs. | |
c6b4fcba | 597 | */ |
b29d4986 JT |
598 | #define WRITE_LOCK_LEVEL 0 |
599 | #define READ_WRITE_LOCK_LEVEL 1 | |
600 | ||
601 | static unsigned lock_level(struct bio *bio) | |
c6b4fcba | 602 | { |
b29d4986 JT |
603 | return bio_data_dir(bio) == WRITE ? |
604 | WRITE_LOCK_LEVEL : | |
605 | READ_WRITE_LOCK_LEVEL; | |
606 | } | |
c6b4fcba | 607 | |
b29d4986 JT |
608 | /*---------------------------------------------------------------- |
609 | * Per bio data | |
610 | *--------------------------------------------------------------*/ | |
c6b4fcba | 611 | |
693b960e | 612 | static struct per_bio_data *get_per_bio_data(struct bio *bio) |
b29d4986 | 613 | { |
693b960e | 614 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); |
b29d4986 JT |
615 | BUG_ON(!pb); |
616 | return pb; | |
617 | } | |
c6b4fcba | 618 | |
693b960e | 619 | static struct per_bio_data *init_per_bio_data(struct bio *bio) |
b29d4986 | 620 | { |
693b960e | 621 | struct per_bio_data *pb = get_per_bio_data(bio); |
c6b4fcba | 622 | |
b29d4986 JT |
623 | pb->tick = false; |
624 | pb->req_nr = dm_bio_get_target_bio_nr(bio); | |
625 | pb->cell = NULL; | |
626 | pb->len = 0; | |
627 | ||
628 | return pb; | |
c6b4fcba JT |
629 | } |
630 | ||
631 | /*----------------------------------------------------------------*/ | |
632 | ||
b29d4986 | 633 | static void defer_bio(struct cache *cache, struct bio *bio) |
c6b4fcba | 634 | { |
b29d4986 | 635 | unsigned long flags; |
c6b4fcba | 636 | |
b29d4986 JT |
637 | spin_lock_irqsave(&cache->lock, flags); |
638 | bio_list_add(&cache->deferred_bios, bio); | |
639 | spin_unlock_irqrestore(&cache->lock, flags); | |
640 | ||
641 | wake_deferred_bio_worker(cache); | |
642 | } | |
c6b4fcba | 643 | |
b29d4986 | 644 | static void defer_bios(struct cache *cache, struct bio_list *bios) |
c6b4fcba | 645 | { |
b29d4986 | 646 | unsigned long flags; |
c6b4fcba | 647 | |
b29d4986 JT |
648 | spin_lock_irqsave(&cache->lock, flags); |
649 | bio_list_merge(&cache->deferred_bios, bios); | |
650 | bio_list_init(bios); | |
651 | spin_unlock_irqrestore(&cache->lock, flags); | |
c6b4fcba | 652 | |
b29d4986 | 653 | wake_deferred_bio_worker(cache); |
c6b4fcba JT |
654 | } |
655 | ||
b29d4986 JT |
656 | /*----------------------------------------------------------------*/ |
657 | ||
658 | static bool bio_detain_shared(struct cache *cache, dm_oblock_t oblock, struct bio *bio) | |
7ae34e77 | 659 | { |
b29d4986 | 660 | bool r; |
b29d4986 JT |
661 | struct per_bio_data *pb; |
662 | struct dm_cell_key_v2 key; | |
7ae34e77 | 663 | dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL); |
b29d4986 | 664 | struct dm_bio_prison_cell_v2 *cell_prealloc, *cell; |
7ae34e77 | 665 | |
b29d4986 JT |
666 | cell_prealloc = alloc_prison_cell(cache); /* FIXME: allow wait if calling from worker */ |
667 | if (!cell_prealloc) { | |
668 | defer_bio(cache, bio); | |
669 | return false; | |
670 | } | |
671 | ||
672 | build_key(oblock, end, &key); | |
673 | r = dm_cell_get_v2(cache->prison, &key, lock_level(bio), bio, cell_prealloc, &cell); | |
674 | if (!r) { | |
675 | /* | |
676 | * Failed to get the lock. | |
677 | */ | |
678 | free_prison_cell(cache, cell_prealloc); | |
679 | return r; | |
680 | } | |
c6b4fcba | 681 | |
b29d4986 JT |
682 | if (cell != cell_prealloc) |
683 | free_prison_cell(cache, cell_prealloc); | |
c6b4fcba | 684 | |
693b960e | 685 | pb = get_per_bio_data(bio); |
b29d4986 | 686 | pb->cell = cell; |
c6b4fcba JT |
687 | |
688 | return r; | |
689 | } | |
690 | ||
aeed1420 | 691 | /*----------------------------------------------------------------*/ |
c6b4fcba JT |
692 | |
693 | static bool is_dirty(struct cache *cache, dm_cblock_t b) | |
694 | { | |
695 | return test_bit(from_cblock(b), cache->dirty_bitset); | |
696 | } | |
697 | ||
b29d4986 | 698 | static void set_dirty(struct cache *cache, dm_cblock_t cblock) |
c6b4fcba JT |
699 | { |
700 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 701 | atomic_inc(&cache->nr_dirty); |
b29d4986 | 702 | policy_set_dirty(cache->policy, cblock); |
c6b4fcba JT |
703 | } |
704 | } | |
705 | ||
b29d4986 JT |
706 | /* |
707 | * These two are called when setting after migrations to force the policy | |
708 | * and dirty bitset to be in sync. | |
709 | */ | |
710 | static void force_set_dirty(struct cache *cache, dm_cblock_t cblock) | |
711 | { | |
712 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) | |
713 | atomic_inc(&cache->nr_dirty); | |
714 | policy_set_dirty(cache->policy, cblock); | |
715 | } | |
716 | ||
717 | static void force_clear_dirty(struct cache *cache, dm_cblock_t cblock) | |
c6b4fcba JT |
718 | { |
719 | if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 720 | if (atomic_dec_return(&cache->nr_dirty) == 0) |
c6b4fcba JT |
721 | dm_table_event(cache->ti->table); |
722 | } | |
b29d4986 JT |
723 | |
724 | policy_clear_dirty(cache->policy, cblock); | |
c6b4fcba JT |
725 | } |
726 | ||
727 | /*----------------------------------------------------------------*/ | |
aeed1420 | 728 | |
c6b4fcba JT |
729 | static bool block_size_is_power_of_two(struct cache *cache) |
730 | { | |
731 | return cache->sectors_per_block_shift >= 0; | |
732 | } | |
733 | ||
43aeaa29 MP |
734 | /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */ |
735 | #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6 | |
736 | __always_inline | |
737 | #endif | |
414dd67d JT |
738 | static dm_block_t block_div(dm_block_t b, uint32_t n) |
739 | { | |
740 | do_div(b, n); | |
741 | ||
742 | return b; | |
743 | } | |
744 | ||
7ae34e77 | 745 | static dm_block_t oblocks_per_dblock(struct cache *cache) |
1bad9bc4 | 746 | { |
7ae34e77 | 747 | dm_block_t oblocks = cache->discard_block_size; |
1bad9bc4 | 748 | |
7ae34e77 JT |
749 | if (block_size_is_power_of_two(cache)) |
750 | oblocks >>= cache->sectors_per_block_shift; | |
1bad9bc4 | 751 | else |
7ae34e77 | 752 | oblocks = block_div(oblocks, cache->sectors_per_block); |
1bad9bc4 | 753 | |
7ae34e77 JT |
754 | return oblocks; |
755 | } | |
756 | ||
757 | static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock) | |
758 | { | |
759 | return to_dblock(block_div(from_oblock(oblock), | |
760 | oblocks_per_dblock(cache))); | |
761 | } | |
1bad9bc4 | 762 | |
1bad9bc4 | 763 | static void set_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
764 | { |
765 | unsigned long flags; | |
766 | ||
7ae34e77 | 767 | BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
768 | atomic_inc(&cache->stats.discard_count); |
769 | ||
770 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 771 | set_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
772 | spin_unlock_irqrestore(&cache->lock, flags); |
773 | } | |
774 | ||
1bad9bc4 | 775 | static void clear_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
776 | { |
777 | unsigned long flags; | |
778 | ||
779 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 780 | clear_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
781 | spin_unlock_irqrestore(&cache->lock, flags); |
782 | } | |
783 | ||
1bad9bc4 | 784 | static bool is_discarded(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
785 | { |
786 | int r; | |
787 | unsigned long flags; | |
788 | ||
789 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 790 | r = test_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
791 | spin_unlock_irqrestore(&cache->lock, flags); |
792 | ||
793 | return r; | |
794 | } | |
795 | ||
796 | static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b) | |
797 | { | |
798 | int r; | |
799 | unsigned long flags; | |
800 | ||
801 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 JT |
802 | r = test_bit(from_dblock(oblock_to_dblock(cache, b)), |
803 | cache->discard_bitset); | |
c6b4fcba JT |
804 | spin_unlock_irqrestore(&cache->lock, flags); |
805 | ||
806 | return r; | |
807 | } | |
808 | ||
b29d4986 JT |
809 | /*---------------------------------------------------------------- |
810 | * Remapping | |
811 | *--------------------------------------------------------------*/ | |
812 | static void remap_to_origin(struct cache *cache, struct bio *bio) | |
c6b4fcba | 813 | { |
74d46992 | 814 | bio_set_dev(bio, cache->origin_dev->bdev); |
c6b4fcba JT |
815 | } |
816 | ||
817 | static void remap_to_cache(struct cache *cache, struct bio *bio, | |
818 | dm_cblock_t cblock) | |
819 | { | |
4f024f37 | 820 | sector_t bi_sector = bio->bi_iter.bi_sector; |
e0d849fa | 821 | sector_t block = from_cblock(cblock); |
c6b4fcba | 822 | |
74d46992 | 823 | bio_set_dev(bio, cache->cache_dev->bdev); |
c6b4fcba | 824 | if (!block_size_is_power_of_two(cache)) |
4f024f37 | 825 | bio->bi_iter.bi_sector = |
e0d849fa | 826 | (block * cache->sectors_per_block) + |
4f024f37 | 827 | sector_div(bi_sector, cache->sectors_per_block); |
c6b4fcba | 828 | else |
4f024f37 | 829 | bio->bi_iter.bi_sector = |
e0d849fa | 830 | (block << cache->sectors_per_block_shift) | |
4f024f37 | 831 | (bi_sector & (cache->sectors_per_block - 1)); |
c6b4fcba JT |
832 | } |
833 | ||
834 | static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio) | |
835 | { | |
836 | unsigned long flags; | |
693b960e | 837 | struct per_bio_data *pb; |
c6b4fcba JT |
838 | |
839 | spin_lock_irqsave(&cache->lock, flags); | |
f73f44eb | 840 | if (cache->need_tick_bio && !op_is_flush(bio->bi_opf) && |
e6047149 | 841 | bio_op(bio) != REQ_OP_DISCARD) { |
693b960e | 842 | pb = get_per_bio_data(bio); |
c6b4fcba JT |
843 | pb->tick = true; |
844 | cache->need_tick_bio = false; | |
845 | } | |
846 | spin_unlock_irqrestore(&cache->lock, flags); | |
847 | } | |
848 | ||
2df3bae9 MS |
849 | static void __remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, |
850 | dm_oblock_t oblock, bool bio_has_pbd) | |
c6b4fcba | 851 | { |
2df3bae9 MS |
852 | if (bio_has_pbd) |
853 | check_if_tick_bio_needed(cache, bio); | |
c6b4fcba JT |
854 | remap_to_origin(cache, bio); |
855 | if (bio_data_dir(bio) == WRITE) | |
1bad9bc4 | 856 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
857 | } |
858 | ||
2df3bae9 MS |
859 | static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, |
860 | dm_oblock_t oblock) | |
861 | { | |
862 | // FIXME: check_if_tick_bio_needed() is called way too much through this interface | |
863 | __remap_to_origin_clear_discard(cache, bio, oblock, true); | |
864 | } | |
865 | ||
c6b4fcba JT |
866 | static void remap_to_cache_dirty(struct cache *cache, struct bio *bio, |
867 | dm_oblock_t oblock, dm_cblock_t cblock) | |
868 | { | |
f8e5f01a | 869 | check_if_tick_bio_needed(cache, bio); |
c6b4fcba JT |
870 | remap_to_cache(cache, bio, cblock); |
871 | if (bio_data_dir(bio) == WRITE) { | |
b29d4986 | 872 | set_dirty(cache, cblock); |
1bad9bc4 | 873 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
874 | } |
875 | } | |
876 | ||
877 | static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio) | |
878 | { | |
4f024f37 | 879 | sector_t block_nr = bio->bi_iter.bi_sector; |
c6b4fcba JT |
880 | |
881 | if (!block_size_is_power_of_two(cache)) | |
882 | (void) sector_div(block_nr, cache->sectors_per_block); | |
883 | else | |
884 | block_nr >>= cache->sectors_per_block_shift; | |
885 | ||
886 | return to_oblock(block_nr); | |
887 | } | |
888 | ||
066dbaa3 JT |
889 | static bool accountable_bio(struct cache *cache, struct bio *bio) |
890 | { | |
701e03e4 | 891 | return bio_op(bio) != REQ_OP_DISCARD; |
066dbaa3 JT |
892 | } |
893 | ||
894 | static void accounted_begin(struct cache *cache, struct bio *bio) | |
895 | { | |
693b960e | 896 | struct per_bio_data *pb; |
066dbaa3 JT |
897 | |
898 | if (accountable_bio(cache, bio)) { | |
693b960e | 899 | pb = get_per_bio_data(bio); |
066dbaa3 | 900 | pb->len = bio_sectors(bio); |
701e03e4 | 901 | iot_io_begin(&cache->tracker, pb->len); |
066dbaa3 JT |
902 | } |
903 | } | |
904 | ||
905 | static void accounted_complete(struct cache *cache, struct bio *bio) | |
906 | { | |
693b960e | 907 | struct per_bio_data *pb = get_per_bio_data(bio); |
066dbaa3 | 908 | |
701e03e4 | 909 | iot_io_end(&cache->tracker, pb->len); |
066dbaa3 JT |
910 | } |
911 | ||
912 | static void accounted_request(struct cache *cache, struct bio *bio) | |
913 | { | |
914 | accounted_begin(cache, bio); | |
915 | generic_make_request(bio); | |
916 | } | |
917 | ||
b29d4986 | 918 | static void issue_op(struct bio *bio, void *context) |
8c081b52 | 919 | { |
b29d4986 JT |
920 | struct cache *cache = context; |
921 | accounted_request(cache, bio); | |
8c081b52 JT |
922 | } |
923 | ||
e2e74d61 JT |
924 | /* |
925 | * When running in writethrough mode we need to send writes to clean blocks | |
2df3bae9 | 926 | * to both the cache and origin devices. Clone the bio and send them in parallel. |
e2e74d61 | 927 | */ |
2df3bae9 MS |
928 | static void remap_to_origin_and_cache(struct cache *cache, struct bio *bio, |
929 | dm_oblock_t oblock, dm_cblock_t cblock) | |
e2e74d61 | 930 | { |
6f1c819c | 931 | struct bio *origin_bio = bio_clone_fast(bio, GFP_NOIO, &cache->bs); |
2df3bae9 MS |
932 | |
933 | BUG_ON(!origin_bio); | |
e2e74d61 | 934 | |
2df3bae9 MS |
935 | bio_chain(origin_bio, bio); |
936 | /* | |
937 | * Passing false to __remap_to_origin_clear_discard() skips | |
938 | * all code that might use per_bio_data (since clone doesn't have it) | |
939 | */ | |
940 | __remap_to_origin_clear_discard(cache, origin_bio, oblock, false); | |
941 | submit_bio(origin_bio); | |
e2e74d61 | 942 | |
2df3bae9 | 943 | remap_to_cache(cache, bio, cblock); |
e2e74d61 JT |
944 | } |
945 | ||
028ae9f7 JT |
946 | /*---------------------------------------------------------------- |
947 | * Failure modes | |
948 | *--------------------------------------------------------------*/ | |
949 | static enum cache_metadata_mode get_cache_mode(struct cache *cache) | |
950 | { | |
951 | return cache->features.mode; | |
952 | } | |
953 | ||
b61d9509 MS |
954 | static const char *cache_device_name(struct cache *cache) |
955 | { | |
956 | return dm_device_name(dm_table_get_md(cache->ti->table)); | |
957 | } | |
958 | ||
028ae9f7 JT |
959 | static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode) |
960 | { | |
961 | const char *descs[] = { | |
962 | "write", | |
963 | "read-only", | |
964 | "fail" | |
965 | }; | |
966 | ||
967 | dm_table_event(cache->ti->table); | |
b61d9509 MS |
968 | DMINFO("%s: switching cache to %s mode", |
969 | cache_device_name(cache), descs[(int)mode]); | |
028ae9f7 JT |
970 | } |
971 | ||
972 | static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode) | |
973 | { | |
d14fcf3d | 974 | bool needs_check; |
028ae9f7 JT |
975 | enum cache_metadata_mode old_mode = get_cache_mode(cache); |
976 | ||
d14fcf3d | 977 | if (dm_cache_metadata_needs_check(cache->cmd, &needs_check)) { |
23cab26d MS |
978 | DMERR("%s: unable to read needs_check flag, setting failure mode.", |
979 | cache_device_name(cache)); | |
d14fcf3d JT |
980 | new_mode = CM_FAIL; |
981 | } | |
982 | ||
028ae9f7 | 983 | if (new_mode == CM_WRITE && needs_check) { |
b61d9509 MS |
984 | DMERR("%s: unable to switch cache to write mode until repaired.", |
985 | cache_device_name(cache)); | |
028ae9f7 JT |
986 | if (old_mode != new_mode) |
987 | new_mode = old_mode; | |
988 | else | |
989 | new_mode = CM_READ_ONLY; | |
990 | } | |
991 | ||
992 | /* Never move out of fail mode */ | |
993 | if (old_mode == CM_FAIL) | |
994 | new_mode = CM_FAIL; | |
995 | ||
996 | switch (new_mode) { | |
997 | case CM_FAIL: | |
998 | case CM_READ_ONLY: | |
999 | dm_cache_metadata_set_read_only(cache->cmd); | |
1000 | break; | |
1001 | ||
1002 | case CM_WRITE: | |
1003 | dm_cache_metadata_set_read_write(cache->cmd); | |
1004 | break; | |
1005 | } | |
1006 | ||
1007 | cache->features.mode = new_mode; | |
1008 | ||
1009 | if (new_mode != old_mode) | |
1010 | notify_mode_switch(cache, new_mode); | |
1011 | } | |
1012 | ||
1013 | static void abort_transaction(struct cache *cache) | |
1014 | { | |
b61d9509 MS |
1015 | const char *dev_name = cache_device_name(cache); |
1016 | ||
028ae9f7 JT |
1017 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
1018 | return; | |
1019 | ||
1020 | if (dm_cache_metadata_set_needs_check(cache->cmd)) { | |
b61d9509 | 1021 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); |
028ae9f7 JT |
1022 | set_cache_mode(cache, CM_FAIL); |
1023 | } | |
1024 | ||
b61d9509 | 1025 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); |
028ae9f7 | 1026 | if (dm_cache_metadata_abort(cache->cmd)) { |
b61d9509 | 1027 | DMERR("%s: failed to abort metadata transaction", dev_name); |
028ae9f7 JT |
1028 | set_cache_mode(cache, CM_FAIL); |
1029 | } | |
1030 | } | |
1031 | ||
1032 | static void metadata_operation_failed(struct cache *cache, const char *op, int r) | |
1033 | { | |
b61d9509 MS |
1034 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1035 | cache_device_name(cache), op, r); | |
028ae9f7 JT |
1036 | abort_transaction(cache); |
1037 | set_cache_mode(cache, CM_READ_ONLY); | |
1038 | } | |
1039 | ||
b29d4986 JT |
1040 | /*----------------------------------------------------------------*/ |
1041 | ||
1042 | static void load_stats(struct cache *cache) | |
1043 | { | |
1044 | struct dm_cache_statistics stats; | |
1045 | ||
1046 | dm_cache_metadata_get_stats(cache->cmd, &stats); | |
1047 | atomic_set(&cache->stats.read_hit, stats.read_hits); | |
1048 | atomic_set(&cache->stats.read_miss, stats.read_misses); | |
1049 | atomic_set(&cache->stats.write_hit, stats.write_hits); | |
1050 | atomic_set(&cache->stats.write_miss, stats.write_misses); | |
1051 | } | |
1052 | ||
1053 | static void save_stats(struct cache *cache) | |
1054 | { | |
1055 | struct dm_cache_statistics stats; | |
1056 | ||
1057 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
1058 | return; | |
1059 | ||
1060 | stats.read_hits = atomic_read(&cache->stats.read_hit); | |
1061 | stats.read_misses = atomic_read(&cache->stats.read_miss); | |
1062 | stats.write_hits = atomic_read(&cache->stats.write_hit); | |
1063 | stats.write_misses = atomic_read(&cache->stats.write_miss); | |
1064 | ||
1065 | dm_cache_metadata_set_stats(cache->cmd, &stats); | |
1066 | } | |
1067 | ||
1068 | static void update_stats(struct cache_stats *stats, enum policy_operation op) | |
1069 | { | |
1070 | switch (op) { | |
1071 | case POLICY_PROMOTE: | |
1072 | atomic_inc(&stats->promotion); | |
1073 | break; | |
1074 | ||
1075 | case POLICY_DEMOTE: | |
1076 | atomic_inc(&stats->demotion); | |
1077 | break; | |
1078 | ||
1079 | case POLICY_WRITEBACK: | |
1080 | atomic_inc(&stats->writeback); | |
1081 | break; | |
1082 | } | |
1083 | } | |
1084 | ||
c6b4fcba JT |
1085 | /*---------------------------------------------------------------- |
1086 | * Migration processing | |
1087 | * | |
1088 | * Migration covers moving data from the origin device to the cache, or | |
1089 | * vice versa. | |
1090 | *--------------------------------------------------------------*/ | |
b29d4986 | 1091 | |
a59db676 | 1092 | static void inc_io_migrations(struct cache *cache) |
c6b4fcba | 1093 | { |
a59db676 | 1094 | atomic_inc(&cache->nr_io_migrations); |
c6b4fcba JT |
1095 | } |
1096 | ||
a59db676 | 1097 | static void dec_io_migrations(struct cache *cache) |
c6b4fcba | 1098 | { |
a59db676 | 1099 | atomic_dec(&cache->nr_io_migrations); |
c6b4fcba JT |
1100 | } |
1101 | ||
651f5fa2 JT |
1102 | static bool discard_or_flush(struct bio *bio) |
1103 | { | |
f73f44eb | 1104 | return bio_op(bio) == REQ_OP_DISCARD || op_is_flush(bio->bi_opf); |
651f5fa2 JT |
1105 | } |
1106 | ||
b29d4986 JT |
1107 | static void calc_discard_block_range(struct cache *cache, struct bio *bio, |
1108 | dm_dblock_t *b, dm_dblock_t *e) | |
c6b4fcba | 1109 | { |
b29d4986 JT |
1110 | sector_t sb = bio->bi_iter.bi_sector; |
1111 | sector_t se = bio_end_sector(bio); | |
651f5fa2 | 1112 | |
b29d4986 | 1113 | *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size)); |
c6b4fcba | 1114 | |
b29d4986 JT |
1115 | if (se - sb < cache->discard_block_size) |
1116 | *e = *b; | |
1117 | else | |
1118 | *e = to_dblock(block_div(se, cache->discard_block_size)); | |
c6b4fcba JT |
1119 | } |
1120 | ||
b29d4986 | 1121 | /*----------------------------------------------------------------*/ |
651f5fa2 | 1122 | |
b29d4986 | 1123 | static void prevent_background_work(struct cache *cache) |
651f5fa2 | 1124 | { |
b29d4986 JT |
1125 | lockdep_off(); |
1126 | down_write(&cache->background_work_lock); | |
1127 | lockdep_on(); | |
651f5fa2 JT |
1128 | } |
1129 | ||
b29d4986 | 1130 | static void allow_background_work(struct cache *cache) |
c6b4fcba | 1131 | { |
b29d4986 JT |
1132 | lockdep_off(); |
1133 | up_write(&cache->background_work_lock); | |
1134 | lockdep_on(); | |
c6b4fcba JT |
1135 | } |
1136 | ||
b29d4986 | 1137 | static bool background_work_begin(struct cache *cache) |
c6b4fcba | 1138 | { |
b29d4986 | 1139 | bool r; |
c6b4fcba | 1140 | |
b29d4986 JT |
1141 | lockdep_off(); |
1142 | r = down_read_trylock(&cache->background_work_lock); | |
1143 | lockdep_on(); | |
c6b4fcba | 1144 | |
b29d4986 | 1145 | return r; |
c6b4fcba JT |
1146 | } |
1147 | ||
b29d4986 | 1148 | static void background_work_end(struct cache *cache) |
c6b4fcba | 1149 | { |
b29d4986 JT |
1150 | lockdep_off(); |
1151 | up_read(&cache->background_work_lock); | |
1152 | lockdep_on(); | |
1153 | } | |
c6b4fcba | 1154 | |
b29d4986 | 1155 | /*----------------------------------------------------------------*/ |
c6b4fcba | 1156 | |
d1260e2a JT |
1157 | static bool bio_writes_complete_block(struct cache *cache, struct bio *bio) |
1158 | { | |
1159 | return (bio_data_dir(bio) == WRITE) && | |
1160 | (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT)); | |
1161 | } | |
1162 | ||
1163 | static bool optimisable_bio(struct cache *cache, struct bio *bio, dm_oblock_t block) | |
1164 | { | |
8e3c3827 | 1165 | return writeback_mode(cache) && |
d1260e2a JT |
1166 | (is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio)); |
1167 | } | |
1168 | ||
b29d4986 JT |
1169 | static void quiesce(struct dm_cache_migration *mg, |
1170 | void (*continuation)(struct work_struct *)) | |
1171 | { | |
1172 | init_continuation(&mg->k, continuation); | |
1173 | dm_cell_quiesce_v2(mg->cache->prison, mg->cell, &mg->k.ws); | |
c6b4fcba JT |
1174 | } |
1175 | ||
b29d4986 | 1176 | static struct dm_cache_migration *ws_to_mg(struct work_struct *ws) |
c6b4fcba | 1177 | { |
b29d4986 JT |
1178 | struct continuation *k = container_of(ws, struct continuation, ws); |
1179 | return container_of(k, struct dm_cache_migration, k); | |
c6b4fcba JT |
1180 | } |
1181 | ||
1182 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
1183 | { | |
b29d4986 | 1184 | struct dm_cache_migration *mg = container_of(context, struct dm_cache_migration, k); |
c6b4fcba JT |
1185 | |
1186 | if (read_err || write_err) | |
4e4cbee9 | 1187 | mg->k.input = BLK_STS_IOERR; |
c6b4fcba | 1188 | |
b29d4986 | 1189 | queue_continuation(mg->cache->wq, &mg->k); |
c6b4fcba JT |
1190 | } |
1191 | ||
7209049d | 1192 | static void copy(struct dm_cache_migration *mg, bool promote) |
c6b4fcba | 1193 | { |
c6b4fcba JT |
1194 | struct dm_io_region o_region, c_region; |
1195 | struct cache *cache = mg->cache; | |
1196 | ||
1197 | o_region.bdev = cache->origin_dev->bdev; | |
b29d4986 | 1198 | o_region.sector = from_oblock(mg->op->oblock) * cache->sectors_per_block; |
c6b4fcba JT |
1199 | o_region.count = cache->sectors_per_block; |
1200 | ||
1201 | c_region.bdev = cache->cache_dev->bdev; | |
b29d4986 | 1202 | c_region.sector = from_cblock(mg->op->cblock) * cache->sectors_per_block; |
c6b4fcba JT |
1203 | c_region.count = cache->sectors_per_block; |
1204 | ||
b29d4986 | 1205 | if (promote) |
7209049d | 1206 | dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, &mg->k); |
b29d4986 | 1207 | else |
7209049d | 1208 | dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, &mg->k); |
b29d4986 JT |
1209 | } |
1210 | ||
1211 | static void bio_drop_shared_lock(struct cache *cache, struct bio *bio) | |
1212 | { | |
693b960e | 1213 | struct per_bio_data *pb = get_per_bio_data(bio); |
b29d4986 JT |
1214 | |
1215 | if (pb->cell && dm_cell_put_v2(cache->prison, pb->cell)) | |
1216 | free_prison_cell(cache, pb->cell); | |
1217 | pb->cell = NULL; | |
c6b4fcba JT |
1218 | } |
1219 | ||
4246a0b6 | 1220 | static void overwrite_endio(struct bio *bio) |
c9d28d5d JT |
1221 | { |
1222 | struct dm_cache_migration *mg = bio->bi_private; | |
1223 | struct cache *cache = mg->cache; | |
693b960e | 1224 | struct per_bio_data *pb = get_per_bio_data(bio); |
c9d28d5d | 1225 | |
80ae49aa MS |
1226 | dm_unhook_bio(&pb->hook_info, bio); |
1227 | ||
4e4cbee9 CH |
1228 | if (bio->bi_status) |
1229 | mg->k.input = bio->bi_status; | |
80ae49aa | 1230 | |
693b960e | 1231 | queue_continuation(cache->wq, &mg->k); |
c9d28d5d JT |
1232 | } |
1233 | ||
b29d4986 JT |
1234 | static void overwrite(struct dm_cache_migration *mg, |
1235 | void (*continuation)(struct work_struct *)) | |
c9d28d5d | 1236 | { |
b29d4986 | 1237 | struct bio *bio = mg->overwrite_bio; |
693b960e | 1238 | struct per_bio_data *pb = get_per_bio_data(bio); |
c9d28d5d JT |
1239 | |
1240 | dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg); | |
8c081b52 JT |
1241 | |
1242 | /* | |
b29d4986 JT |
1243 | * The overwrite bio is part of the copy operation, as such it does |
1244 | * not set/clear discard or dirty flags. | |
8c081b52 | 1245 | */ |
b29d4986 JT |
1246 | if (mg->op->op == POLICY_PROMOTE) |
1247 | remap_to_cache(mg->cache, bio, mg->op->cblock); | |
1248 | else | |
1249 | remap_to_origin(mg->cache, bio); | |
1250 | ||
1251 | init_continuation(&mg->k, continuation); | |
066dbaa3 | 1252 | accounted_request(mg->cache, bio); |
c9d28d5d JT |
1253 | } |
1254 | ||
b29d4986 JT |
1255 | /* |
1256 | * Migration steps: | |
1257 | * | |
1258 | * 1) exclusive lock preventing WRITEs | |
1259 | * 2) quiesce | |
1260 | * 3) copy or issue overwrite bio | |
1261 | * 4) upgrade to exclusive lock preventing READs and WRITEs | |
1262 | * 5) quiesce | |
1263 | * 6) update metadata and commit | |
1264 | * 7) unlock | |
1265 | */ | |
1266 | static void mg_complete(struct dm_cache_migration *mg, bool success) | |
c9d28d5d | 1267 | { |
b29d4986 JT |
1268 | struct bio_list bios; |
1269 | struct cache *cache = mg->cache; | |
1270 | struct policy_work *op = mg->op; | |
1271 | dm_cblock_t cblock = op->cblock; | |
1272 | ||
1273 | if (success) | |
1274 | update_stats(&cache->stats, op->op); | |
1275 | ||
1276 | switch (op->op) { | |
1277 | case POLICY_PROMOTE: | |
1278 | clear_discard(cache, oblock_to_dblock(cache, op->oblock)); | |
1279 | policy_complete_background_work(cache->policy, op, success); | |
1280 | ||
1281 | if (mg->overwrite_bio) { | |
1282 | if (success) | |
1283 | force_set_dirty(cache, cblock); | |
4e4cbee9 CH |
1284 | else if (mg->k.input) |
1285 | mg->overwrite_bio->bi_status = mg->k.input; | |
b29d4986 | 1286 | else |
4e4cbee9 | 1287 | mg->overwrite_bio->bi_status = BLK_STS_IOERR; |
b29d4986 JT |
1288 | bio_endio(mg->overwrite_bio); |
1289 | } else { | |
1290 | if (success) | |
1291 | force_clear_dirty(cache, cblock); | |
1292 | dec_io_migrations(cache); | |
1293 | } | |
1294 | break; | |
1295 | ||
1296 | case POLICY_DEMOTE: | |
1297 | /* | |
1298 | * We clear dirty here to update the nr_dirty counter. | |
1299 | */ | |
1300 | if (success) | |
1301 | force_clear_dirty(cache, cblock); | |
1302 | policy_complete_background_work(cache->policy, op, success); | |
1303 | dec_io_migrations(cache); | |
1304 | break; | |
1305 | ||
1306 | case POLICY_WRITEBACK: | |
1307 | if (success) | |
1308 | force_clear_dirty(cache, cblock); | |
1309 | policy_complete_background_work(cache->policy, op, success); | |
1310 | dec_io_migrations(cache); | |
1311 | break; | |
1312 | } | |
1313 | ||
1314 | bio_list_init(&bios); | |
1315 | if (mg->cell) { | |
1316 | if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios)) | |
1317 | free_prison_cell(cache, mg->cell); | |
1318 | } | |
1319 | ||
1320 | free_migration(mg); | |
1321 | defer_bios(cache, &bios); | |
1322 | wake_migration_worker(cache); | |
1323 | ||
1324 | background_work_end(cache); | |
c9d28d5d JT |
1325 | } |
1326 | ||
b29d4986 | 1327 | static void mg_success(struct work_struct *ws) |
c6b4fcba | 1328 | { |
b29d4986 JT |
1329 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1330 | mg_complete(mg, mg->k.input == 0); | |
c6b4fcba JT |
1331 | } |
1332 | ||
b29d4986 | 1333 | static void mg_update_metadata(struct work_struct *ws) |
7ae34e77 | 1334 | { |
b29d4986 JT |
1335 | int r; |
1336 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
c6b4fcba | 1337 | struct cache *cache = mg->cache; |
b29d4986 | 1338 | struct policy_work *op = mg->op; |
c6b4fcba | 1339 | |
b29d4986 JT |
1340 | switch (op->op) { |
1341 | case POLICY_PROMOTE: | |
1342 | r = dm_cache_insert_mapping(cache->cmd, op->cblock, op->oblock); | |
1343 | if (r) { | |
1344 | DMERR_LIMIT("%s: migration failed; couldn't insert mapping", | |
1345 | cache_device_name(cache)); | |
1346 | metadata_operation_failed(cache, "dm_cache_insert_mapping", r); | |
7ae34e77 | 1347 | |
b29d4986 JT |
1348 | mg_complete(mg, false); |
1349 | return; | |
1350 | } | |
1351 | mg_complete(mg, true); | |
1352 | break; | |
c9d28d5d | 1353 | |
b29d4986 JT |
1354 | case POLICY_DEMOTE: |
1355 | r = dm_cache_remove_mapping(cache->cmd, op->cblock); | |
1356 | if (r) { | |
1357 | DMERR_LIMIT("%s: migration failed; couldn't update on disk metadata", | |
1358 | cache_device_name(cache)); | |
1359 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); | |
c6b4fcba | 1360 | |
b29d4986 | 1361 | mg_complete(mg, false); |
c9d28d5d JT |
1362 | return; |
1363 | } | |
c9d28d5d | 1364 | |
b29d4986 JT |
1365 | /* |
1366 | * It would be nice if we only had to commit when a REQ_FLUSH | |
1367 | * comes through. But there's one scenario that we have to | |
1368 | * look out for: | |
1369 | * | |
1370 | * - vblock x in a cache block | |
1371 | * - domotion occurs | |
1372 | * - cache block gets reallocated and over written | |
1373 | * - crash | |
1374 | * | |
1375 | * When we recover, because there was no commit the cache will | |
1376 | * rollback to having the data for vblock x in the cache block. | |
1377 | * But the cache block has since been overwritten, so it'll end | |
1378 | * up pointing to data that was never in 'x' during the history | |
1379 | * of the device. | |
1380 | * | |
1381 | * To avoid this issue we require a commit as part of the | |
1382 | * demotion operation. | |
1383 | */ | |
1384 | init_continuation(&mg->k, mg_success); | |
1385 | continue_after_commit(&cache->committer, &mg->k); | |
1386 | schedule_commit(&cache->committer); | |
1387 | break; | |
1388 | ||
1389 | case POLICY_WRITEBACK: | |
1390 | mg_complete(mg, true); | |
1391 | break; | |
7ae34e77 | 1392 | } |
c6b4fcba JT |
1393 | } |
1394 | ||
b29d4986 | 1395 | static void mg_update_metadata_after_copy(struct work_struct *ws) |
c6b4fcba | 1396 | { |
b29d4986 JT |
1397 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1398 | ||
1399 | /* | |
1400 | * Did the copy succeed? | |
1401 | */ | |
1402 | if (mg->k.input) | |
1403 | mg_complete(mg, false); | |
c6b4fcba | 1404 | else |
b29d4986 | 1405 | mg_update_metadata(ws); |
c6b4fcba JT |
1406 | } |
1407 | ||
b29d4986 | 1408 | static void mg_upgrade_lock(struct work_struct *ws) |
c6b4fcba | 1409 | { |
b29d4986 JT |
1410 | int r; |
1411 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
c6b4fcba | 1412 | |
b29d4986 JT |
1413 | /* |
1414 | * Did the copy succeed? | |
1415 | */ | |
1416 | if (mg->k.input) | |
1417 | mg_complete(mg, false); | |
c6b4fcba | 1418 | |
c9d28d5d | 1419 | else { |
b29d4986 JT |
1420 | /* |
1421 | * Now we want the lock to prevent both reads and writes. | |
1422 | */ | |
1423 | r = dm_cell_lock_promote_v2(mg->cache->prison, mg->cell, | |
1424 | READ_WRITE_LOCK_LEVEL); | |
1425 | if (r < 0) | |
1426 | mg_complete(mg, false); | |
c6b4fcba | 1427 | |
b29d4986 JT |
1428 | else if (r) |
1429 | quiesce(mg, mg_update_metadata); | |
c6b4fcba | 1430 | |
b29d4986 JT |
1431 | else |
1432 | mg_update_metadata(ws); | |
c9d28d5d | 1433 | } |
c6b4fcba JT |
1434 | } |
1435 | ||
d1260e2a JT |
1436 | static void mg_full_copy(struct work_struct *ws) |
1437 | { | |
1438 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
1439 | struct cache *cache = mg->cache; | |
1440 | struct policy_work *op = mg->op; | |
1441 | bool is_policy_promote = (op->op == POLICY_PROMOTE); | |
1442 | ||
1443 | if ((!is_policy_promote && !is_dirty(cache, op->cblock)) || | |
1444 | is_discarded_oblock(cache, op->oblock)) { | |
1445 | mg_upgrade_lock(ws); | |
1446 | return; | |
1447 | } | |
1448 | ||
1449 | init_continuation(&mg->k, mg_upgrade_lock); | |
7209049d | 1450 | copy(mg, is_policy_promote); |
d1260e2a JT |
1451 | } |
1452 | ||
b29d4986 | 1453 | static void mg_copy(struct work_struct *ws) |
c6b4fcba | 1454 | { |
b29d4986 | 1455 | struct dm_cache_migration *mg = ws_to_mg(ws); |
c6b4fcba | 1456 | |
b29d4986 | 1457 | if (mg->overwrite_bio) { |
d1260e2a JT |
1458 | /* |
1459 | * No exclusive lock was held when we last checked if the bio | |
1460 | * was optimisable. So we have to check again in case things | |
1461 | * have changed (eg, the block may no longer be discarded). | |
1462 | */ | |
1463 | if (!optimisable_bio(mg->cache, mg->overwrite_bio, mg->op->oblock)) { | |
1464 | /* | |
1465 | * Fallback to a real full copy after doing some tidying up. | |
1466 | */ | |
1467 | bool rb = bio_detain_shared(mg->cache, mg->op->oblock, mg->overwrite_bio); | |
1468 | BUG_ON(rb); /* An exclussive lock must _not_ be held for this block */ | |
1469 | mg->overwrite_bio = NULL; | |
1470 | inc_io_migrations(mg->cache); | |
1471 | mg_full_copy(ws); | |
1472 | return; | |
1473 | } | |
1474 | ||
b29d4986 JT |
1475 | /* |
1476 | * It's safe to do this here, even though it's new data | |
1477 | * because all IO has been locked out of the block. | |
1478 | * | |
1479 | * mg_lock_writes() already took READ_WRITE_LOCK_LEVEL | |
1480 | * so _not_ using mg_upgrade_lock() as continutation. | |
1481 | */ | |
1482 | overwrite(mg, mg_update_metadata_after_copy); | |
c6b4fcba | 1483 | |
d1260e2a JT |
1484 | } else |
1485 | mg_full_copy(ws); | |
c6b4fcba JT |
1486 | } |
1487 | ||
b29d4986 | 1488 | static int mg_lock_writes(struct dm_cache_migration *mg) |
c6b4fcba | 1489 | { |
b29d4986 JT |
1490 | int r; |
1491 | struct dm_cell_key_v2 key; | |
c6b4fcba | 1492 | struct cache *cache = mg->cache; |
b29d4986 | 1493 | struct dm_bio_prison_cell_v2 *prealloc; |
c6b4fcba | 1494 | |
b29d4986 JT |
1495 | prealloc = alloc_prison_cell(cache); |
1496 | if (!prealloc) { | |
1497 | DMERR_LIMIT("%s: alloc_prison_cell failed", cache_device_name(cache)); | |
1498 | mg_complete(mg, false); | |
1499 | return -ENOMEM; | |
1500 | } | |
c6b4fcba | 1501 | |
b29d4986 JT |
1502 | /* |
1503 | * Prevent writes to the block, but allow reads to continue. | |
1504 | * Unless we're using an overwrite bio, in which case we lock | |
1505 | * everything. | |
1506 | */ | |
1507 | build_key(mg->op->oblock, oblock_succ(mg->op->oblock), &key); | |
1508 | r = dm_cell_lock_v2(cache->prison, &key, | |
1509 | mg->overwrite_bio ? READ_WRITE_LOCK_LEVEL : WRITE_LOCK_LEVEL, | |
1510 | prealloc, &mg->cell); | |
1511 | if (r < 0) { | |
1512 | free_prison_cell(cache, prealloc); | |
1513 | mg_complete(mg, false); | |
1514 | return r; | |
1515 | } | |
c6b4fcba | 1516 | |
b29d4986 JT |
1517 | if (mg->cell != prealloc) |
1518 | free_prison_cell(cache, prealloc); | |
c6b4fcba | 1519 | |
b29d4986 JT |
1520 | if (r == 0) |
1521 | mg_copy(&mg->k.ws); | |
1522 | else | |
1523 | quiesce(mg, mg_copy); | |
c6b4fcba | 1524 | |
b29d4986 | 1525 | return 0; |
c6b4fcba JT |
1526 | } |
1527 | ||
b29d4986 | 1528 | static int mg_start(struct cache *cache, struct policy_work *op, struct bio *bio) |
c6b4fcba | 1529 | { |
b29d4986 | 1530 | struct dm_cache_migration *mg; |
c6b4fcba | 1531 | |
b29d4986 JT |
1532 | if (!background_work_begin(cache)) { |
1533 | policy_complete_background_work(cache->policy, op, false); | |
1534 | return -EPERM; | |
1535 | } | |
2ee57d58 | 1536 | |
b29d4986 JT |
1537 | mg = alloc_migration(cache); |
1538 | if (!mg) { | |
1539 | policy_complete_background_work(cache->policy, op, false); | |
1540 | background_work_end(cache); | |
1541 | return -ENOMEM; | |
1542 | } | |
2ee57d58 | 1543 | |
b29d4986 JT |
1544 | mg->op = op; |
1545 | mg->overwrite_bio = bio; | |
1546 | ||
1547 | if (!bio) | |
1548 | inc_io_migrations(cache); | |
7ae34e77 | 1549 | |
b29d4986 | 1550 | return mg_lock_writes(mg); |
7ae34e77 JT |
1551 | } |
1552 | ||
c6b4fcba | 1553 | /*---------------------------------------------------------------- |
b29d4986 | 1554 | * invalidation processing |
c6b4fcba | 1555 | *--------------------------------------------------------------*/ |
c6b4fcba | 1556 | |
b29d4986 | 1557 | static void invalidate_complete(struct dm_cache_migration *mg, bool success) |
c6b4fcba | 1558 | { |
b29d4986 JT |
1559 | struct bio_list bios; |
1560 | struct cache *cache = mg->cache; | |
c6b4fcba | 1561 | |
b29d4986 JT |
1562 | bio_list_init(&bios); |
1563 | if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios)) | |
1564 | free_prison_cell(cache, mg->cell); | |
c6b4fcba | 1565 | |
b29d4986 JT |
1566 | if (!success && mg->overwrite_bio) |
1567 | bio_io_error(mg->overwrite_bio); | |
c6b4fcba | 1568 | |
b29d4986 JT |
1569 | free_migration(mg); |
1570 | defer_bios(cache, &bios); | |
c6b4fcba | 1571 | |
b29d4986 | 1572 | background_work_end(cache); |
c6b4fcba JT |
1573 | } |
1574 | ||
b29d4986 | 1575 | static void invalidate_completed(struct work_struct *ws) |
c6b4fcba | 1576 | { |
b29d4986 JT |
1577 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1578 | invalidate_complete(mg, !mg->k.input); | |
c6b4fcba JT |
1579 | } |
1580 | ||
b29d4986 | 1581 | static int invalidate_cblock(struct cache *cache, dm_cblock_t cblock) |
651f5fa2 | 1582 | { |
b29d4986 JT |
1583 | int r = policy_invalidate_mapping(cache->policy, cblock); |
1584 | if (!r) { | |
1585 | r = dm_cache_remove_mapping(cache->cmd, cblock); | |
1586 | if (r) { | |
1587 | DMERR_LIMIT("%s: invalidation failed; couldn't update on disk metadata", | |
1588 | cache_device_name(cache)); | |
1589 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); | |
651f5fa2 JT |
1590 | } |
1591 | ||
b29d4986 JT |
1592 | } else if (r == -ENODATA) { |
1593 | /* | |
1594 | * Harmless, already unmapped. | |
1595 | */ | |
1596 | r = 0; | |
651f5fa2 | 1597 | |
b29d4986 JT |
1598 | } else |
1599 | DMERR("%s: policy_invalidate_mapping failed", cache_device_name(cache)); | |
2ee57d58 | 1600 | |
b29d4986 | 1601 | return r; |
651f5fa2 JT |
1602 | } |
1603 | ||
b29d4986 | 1604 | static void invalidate_remove(struct work_struct *ws) |
651f5fa2 | 1605 | { |
b29d4986 JT |
1606 | int r; |
1607 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
1608 | struct cache *cache = mg->cache; | |
651f5fa2 | 1609 | |
b29d4986 JT |
1610 | r = invalidate_cblock(cache, mg->invalidate_cblock); |
1611 | if (r) { | |
1612 | invalidate_complete(mg, false); | |
1613 | return; | |
651f5fa2 | 1614 | } |
9153df74 | 1615 | |
b29d4986 JT |
1616 | init_continuation(&mg->k, invalidate_completed); |
1617 | continue_after_commit(&cache->committer, &mg->k); | |
1618 | remap_to_origin_clear_discard(cache, mg->overwrite_bio, mg->invalidate_oblock); | |
1619 | mg->overwrite_bio = NULL; | |
1620 | schedule_commit(&cache->committer); | |
651f5fa2 JT |
1621 | } |
1622 | ||
b29d4986 | 1623 | static int invalidate_lock(struct dm_cache_migration *mg) |
651f5fa2 | 1624 | { |
b29d4986 JT |
1625 | int r; |
1626 | struct dm_cell_key_v2 key; | |
1627 | struct cache *cache = mg->cache; | |
1628 | struct dm_bio_prison_cell_v2 *prealloc; | |
651f5fa2 | 1629 | |
b29d4986 JT |
1630 | prealloc = alloc_prison_cell(cache); |
1631 | if (!prealloc) { | |
1632 | invalidate_complete(mg, false); | |
1633 | return -ENOMEM; | |
651f5fa2 JT |
1634 | } |
1635 | ||
b29d4986 JT |
1636 | build_key(mg->invalidate_oblock, oblock_succ(mg->invalidate_oblock), &key); |
1637 | r = dm_cell_lock_v2(cache->prison, &key, | |
1638 | READ_WRITE_LOCK_LEVEL, prealloc, &mg->cell); | |
1639 | if (r < 0) { | |
1640 | free_prison_cell(cache, prealloc); | |
1641 | invalidate_complete(mg, false); | |
1642 | return r; | |
651f5fa2 | 1643 | } |
9153df74 | 1644 | |
b29d4986 JT |
1645 | if (mg->cell != prealloc) |
1646 | free_prison_cell(cache, prealloc); | |
651f5fa2 | 1647 | |
b29d4986 JT |
1648 | if (r) |
1649 | quiesce(mg, invalidate_remove); | |
651f5fa2 | 1650 | |
b29d4986 JT |
1651 | else { |
1652 | /* | |
1653 | * We can't call invalidate_remove() directly here because we | |
1654 | * might still be in request context. | |
1655 | */ | |
1656 | init_continuation(&mg->k, invalidate_remove); | |
1657 | queue_work(cache->wq, &mg->k.ws); | |
1658 | } | |
fb4100ae | 1659 | |
fb4100ae JT |
1660 | return 0; |
1661 | } | |
1662 | ||
b29d4986 JT |
1663 | static int invalidate_start(struct cache *cache, dm_cblock_t cblock, |
1664 | dm_oblock_t oblock, struct bio *bio) | |
fb4100ae | 1665 | { |
b29d4986 | 1666 | struct dm_cache_migration *mg; |
2ee57d58 | 1667 | |
b29d4986 JT |
1668 | if (!background_work_begin(cache)) |
1669 | return -EPERM; | |
c6b4fcba | 1670 | |
b29d4986 JT |
1671 | mg = alloc_migration(cache); |
1672 | if (!mg) { | |
1673 | background_work_end(cache); | |
1674 | return -ENOMEM; | |
7ae34e77 | 1675 | } |
c6b4fcba | 1676 | |
b29d4986 JT |
1677 | mg->overwrite_bio = bio; |
1678 | mg->invalidate_cblock = cblock; | |
1679 | mg->invalidate_oblock = oblock; | |
c6b4fcba | 1680 | |
b29d4986 | 1681 | return invalidate_lock(mg); |
c6b4fcba | 1682 | } |
c6b4fcba | 1683 | |
b29d4986 JT |
1684 | /*---------------------------------------------------------------- |
1685 | * bio processing | |
1686 | *--------------------------------------------------------------*/ | |
c6b4fcba | 1687 | |
b29d4986 JT |
1688 | enum busy { |
1689 | IDLE, | |
b29d4986 JT |
1690 | BUSY |
1691 | }; | |
c6b4fcba | 1692 | |
b29d4986 | 1693 | static enum busy spare_migration_bandwidth(struct cache *cache) |
651f5fa2 | 1694 | { |
701e03e4 | 1695 | bool idle = iot_idle_for(&cache->tracker, HZ); |
a59db676 | 1696 | sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) * |
c6b4fcba | 1697 | cache->sectors_per_block; |
651f5fa2 | 1698 | |
49b7f768 JT |
1699 | if (idle && current_volume <= cache->migration_threshold) |
1700 | return IDLE; | |
b29d4986 | 1701 | else |
49b7f768 | 1702 | return BUSY; |
651f5fa2 JT |
1703 | } |
1704 | ||
c6b4fcba | 1705 | static void inc_hit_counter(struct cache *cache, struct bio *bio) |
c6b4fcba | 1706 | { |
c6b4fcba JT |
1707 | atomic_inc(bio_data_dir(bio) == READ ? |
1708 | &cache->stats.read_hit : &cache->stats.write_hit); | |
c6b4fcba JT |
1709 | } |
1710 | ||
c6b4fcba | 1711 | static void inc_miss_counter(struct cache *cache, struct bio *bio) |
028ae9f7 | 1712 | { |
c6b4fcba JT |
1713 | atomic_inc(bio_data_dir(bio) == READ ? |
1714 | &cache->stats.read_miss : &cache->stats.write_miss); | |
1715 | } | |
028ae9f7 | 1716 | |
fb4100ae | 1717 | /*----------------------------------------------------------------*/ |
028ae9f7 | 1718 | |
b29d4986 JT |
1719 | static int map_bio(struct cache *cache, struct bio *bio, dm_oblock_t block, |
1720 | bool *commit_needed) | |
c6b4fcba | 1721 | { |
b29d4986 JT |
1722 | int r, data_dir; |
1723 | bool rb, background_queued; | |
1724 | dm_cblock_t cblock; | |
c6b4fcba | 1725 | |
b29d4986 | 1726 | *commit_needed = false; |
c6b4fcba | 1727 | |
b29d4986 JT |
1728 | rb = bio_detain_shared(cache, block, bio); |
1729 | if (!rb) { | |
c6b4fcba | 1730 | /* |
b29d4986 JT |
1731 | * An exclusive lock is held for this block, so we have to |
1732 | * wait. We set the commit_needed flag so the current | |
1733 | * transaction will be committed asap, allowing this lock | |
1734 | * to be dropped. | |
c6b4fcba | 1735 | */ |
b29d4986 JT |
1736 | *commit_needed = true; |
1737 | return DM_MAPIO_SUBMITTED; | |
651f5fa2 | 1738 | } |
9153df74 | 1739 | |
b29d4986 | 1740 | data_dir = bio_data_dir(bio); |
651f5fa2 | 1741 | |
b29d4986 JT |
1742 | if (optimisable_bio(cache, bio, block)) { |
1743 | struct policy_work *op = NULL; | |
fb4100ae | 1744 | |
b29d4986 JT |
1745 | r = policy_lookup_with_work(cache->policy, block, &cblock, data_dir, true, &op); |
1746 | if (unlikely(r && r != -ENOENT)) { | |
1747 | DMERR_LIMIT("%s: policy_lookup_with_work() failed with r = %d", | |
1748 | cache_device_name(cache), r); | |
1749 | bio_io_error(bio); | |
1750 | return DM_MAPIO_SUBMITTED; | |
c6b4fcba JT |
1751 | } |
1752 | ||
b29d4986 JT |
1753 | if (r == -ENOENT && op) { |
1754 | bio_drop_shared_lock(cache, bio); | |
1755 | BUG_ON(op->op != POLICY_PROMOTE); | |
1756 | mg_start(cache, op, bio); | |
1757 | return DM_MAPIO_SUBMITTED; | |
1758 | } | |
1759 | } else { | |
1760 | r = policy_lookup(cache->policy, block, &cblock, data_dir, false, &background_queued); | |
1761 | if (unlikely(r && r != -ENOENT)) { | |
1762 | DMERR_LIMIT("%s: policy_lookup() failed with r = %d", | |
1763 | cache_device_name(cache), r); | |
1764 | bio_io_error(bio); | |
1765 | return DM_MAPIO_SUBMITTED; | |
1766 | } | |
c6b4fcba | 1767 | |
b29d4986 JT |
1768 | if (background_queued) |
1769 | wake_migration_worker(cache); | |
c6b4fcba JT |
1770 | } |
1771 | ||
b29d4986 | 1772 | if (r == -ENOENT) { |
693b960e MS |
1773 | struct per_bio_data *pb = get_per_bio_data(bio); |
1774 | ||
b29d4986 JT |
1775 | /* |
1776 | * Miss. | |
1777 | */ | |
1778 | inc_miss_counter(cache, bio); | |
1779 | if (pb->req_nr == 0) { | |
1780 | accounted_begin(cache, bio); | |
1781 | remap_to_origin_clear_discard(cache, bio, block); | |
b29d4986 | 1782 | } else { |
2ee57d58 | 1783 | /* |
b29d4986 JT |
1784 | * This is a duplicate writethrough io that is no |
1785 | * longer needed because the block has been demoted. | |
2ee57d58 | 1786 | */ |
b29d4986 JT |
1787 | bio_endio(bio); |
1788 | return DM_MAPIO_SUBMITTED; | |
1789 | } | |
1790 | } else { | |
1791 | /* | |
1792 | * Hit. | |
1793 | */ | |
1794 | inc_hit_counter(cache, bio); | |
651f5fa2 | 1795 | |
b29d4986 JT |
1796 | /* |
1797 | * Passthrough always maps to the origin, invalidating any | |
1798 | * cache blocks that are written to. | |
1799 | */ | |
8e3c3827 | 1800 | if (passthrough_mode(cache)) { |
2ee57d58 | 1801 | if (bio_data_dir(bio) == WRITE) { |
b29d4986 | 1802 | bio_drop_shared_lock(cache, bio); |
2ee57d58 | 1803 | atomic_inc(&cache->stats.demotion); |
b29d4986 JT |
1804 | invalidate_start(cache, cblock, block, bio); |
1805 | } else | |
2ee57d58 | 1806 | remap_to_origin_clear_discard(cache, bio, block); |
2ee57d58 | 1807 | } else { |
8e3c3827 | 1808 | if (bio_data_dir(bio) == WRITE && writethrough_mode(cache) && |
b29d4986 | 1809 | !is_dirty(cache, cblock)) { |
2df3bae9 | 1810 | remap_to_origin_and_cache(cache, bio, block, cblock); |
b29d4986 JT |
1811 | accounted_begin(cache, bio); |
1812 | } else | |
1813 | remap_to_cache_dirty(cache, bio, block, cblock); | |
2ee57d58 | 1814 | } |
c6b4fcba | 1815 | } |
651f5fa2 | 1816 | |
651f5fa2 | 1817 | /* |
b29d4986 | 1818 | * dm core turns FUA requests into a separate payload and FLUSH req. |
651f5fa2 | 1819 | */ |
b29d4986 | 1820 | if (bio->bi_opf & REQ_FUA) { |
651f5fa2 | 1821 | /* |
b29d4986 JT |
1822 | * issue_after_commit will call accounted_begin a second time. So |
1823 | * we call accounted_complete() to avoid double accounting. | |
651f5fa2 | 1824 | */ |
b29d4986 JT |
1825 | accounted_complete(cache, bio); |
1826 | issue_after_commit(&cache->committer, bio); | |
1827 | *commit_needed = true; | |
1828 | return DM_MAPIO_SUBMITTED; | |
651f5fa2 JT |
1829 | } |
1830 | ||
b29d4986 | 1831 | return DM_MAPIO_REMAPPED; |
651f5fa2 JT |
1832 | } |
1833 | ||
b29d4986 | 1834 | static bool process_bio(struct cache *cache, struct bio *bio) |
c6b4fcba | 1835 | { |
b29d4986 | 1836 | bool commit_needed; |
c6b4fcba | 1837 | |
b29d4986 JT |
1838 | if (map_bio(cache, bio, get_bio_block(cache, bio), &commit_needed) == DM_MAPIO_REMAPPED) |
1839 | generic_make_request(bio); | |
c6b4fcba | 1840 | |
b29d4986 | 1841 | return commit_needed; |
c6b4fcba JT |
1842 | } |
1843 | ||
028ae9f7 JT |
1844 | /* |
1845 | * A non-zero return indicates read_only or fail_io mode. | |
1846 | */ | |
1847 | static int commit(struct cache *cache, bool clean_shutdown) | |
e2e74d61 | 1848 | { |
028ae9f7 | 1849 | int r; |
e2e74d61 | 1850 | |
028ae9f7 JT |
1851 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
1852 | return -EINVAL; | |
e2e74d61 | 1853 | |
028ae9f7 JT |
1854 | atomic_inc(&cache->stats.commit_count); |
1855 | r = dm_cache_commit(cache->cmd, clean_shutdown); | |
1856 | if (r) | |
1857 | metadata_operation_failed(cache, "dm_cache_commit", r); | |
e2e74d61 | 1858 | |
028ae9f7 | 1859 | return r; |
e2e74d61 JT |
1860 | } |
1861 | ||
b29d4986 JT |
1862 | /* |
1863 | * Used by the batcher. | |
1864 | */ | |
4e4cbee9 | 1865 | static blk_status_t commit_op(void *context) |
c6b4fcba | 1866 | { |
b29d4986 | 1867 | struct cache *cache = context; |
c6b4fcba | 1868 | |
b29d4986 | 1869 | if (dm_cache_changed_this_transaction(cache->cmd)) |
4e4cbee9 | 1870 | return errno_to_blk_status(commit(cache, false)); |
c6b4fcba | 1871 | |
b29d4986 | 1872 | return 0; |
c6b4fcba JT |
1873 | } |
1874 | ||
b29d4986 | 1875 | /*----------------------------------------------------------------*/ |
65790ff9 | 1876 | |
b29d4986 | 1877 | static bool process_flush_bio(struct cache *cache, struct bio *bio) |
65790ff9 | 1878 | { |
693b960e | 1879 | struct per_bio_data *pb = get_per_bio_data(bio); |
65790ff9 | 1880 | |
b29d4986 JT |
1881 | if (!pb->req_nr) |
1882 | remap_to_origin(cache, bio); | |
1883 | else | |
1884 | remap_to_cache(cache, bio, 0); | |
65790ff9 | 1885 | |
b29d4986 JT |
1886 | issue_after_commit(&cache->committer, bio); |
1887 | return true; | |
65790ff9 JT |
1888 | } |
1889 | ||
b29d4986 | 1890 | static bool process_discard_bio(struct cache *cache, struct bio *bio) |
65790ff9 | 1891 | { |
b29d4986 | 1892 | dm_dblock_t b, e; |
65790ff9 | 1893 | |
b29d4986 JT |
1894 | // FIXME: do we need to lock the region? Or can we just assume the |
1895 | // user wont be so foolish as to issue discard concurrently with | |
1896 | // other IO? | |
1897 | calc_discard_block_range(cache, bio, &b, &e); | |
1898 | while (b != e) { | |
1899 | set_discard(cache, b); | |
1900 | b = to_dblock(from_dblock(b) + 1); | |
651f5fa2 | 1901 | } |
65790ff9 | 1902 | |
de7180ff MS |
1903 | if (cache->features.discard_passdown) { |
1904 | remap_to_origin(cache, bio); | |
1905 | generic_make_request(bio); | |
1906 | } else | |
1907 | bio_endio(bio); | |
65790ff9 | 1908 | |
b29d4986 | 1909 | return false; |
c6b4fcba JT |
1910 | } |
1911 | ||
b29d4986 | 1912 | static void process_deferred_bios(struct work_struct *ws) |
66cb1910 | 1913 | { |
b29d4986 | 1914 | struct cache *cache = container_of(ws, struct cache, deferred_bio_worker); |
66cb1910 | 1915 | |
c6b4fcba | 1916 | unsigned long flags; |
b29d4986 | 1917 | bool commit_needed = false; |
c6b4fcba JT |
1918 | struct bio_list bios; |
1919 | struct bio *bio; | |
66cb1910 | 1920 | |
c6b4fcba | 1921 | bio_list_init(&bios); |
c6b4fcba | 1922 | |
c6b4fcba | 1923 | spin_lock_irqsave(&cache->lock, flags); |
b29d4986 JT |
1924 | bio_list_merge(&bios, &cache->deferred_bios); |
1925 | bio_list_init(&cache->deferred_bios); | |
c6b4fcba | 1926 | spin_unlock_irqrestore(&cache->lock, flags); |
c6b4fcba | 1927 | |
b29d4986 JT |
1928 | while ((bio = bio_list_pop(&bios))) { |
1929 | if (bio->bi_opf & REQ_PREFLUSH) | |
1930 | commit_needed = process_flush_bio(cache, bio) || commit_needed; | |
c6b4fcba | 1931 | |
b29d4986 JT |
1932 | else if (bio_op(bio) == REQ_OP_DISCARD) |
1933 | commit_needed = process_discard_bio(cache, bio) || commit_needed; | |
1934 | ||
1935 | else | |
1936 | commit_needed = process_bio(cache, bio) || commit_needed; | |
1937 | } | |
1938 | ||
1939 | if (commit_needed) | |
1940 | schedule_commit(&cache->committer); | |
c6b4fcba JT |
1941 | } |
1942 | ||
c6b4fcba JT |
1943 | /*---------------------------------------------------------------- |
1944 | * Main worker loop | |
1945 | *--------------------------------------------------------------*/ | |
651f5fa2 JT |
1946 | |
1947 | static void requeue_deferred_bios(struct cache *cache) | |
c6b4fcba JT |
1948 | { |
1949 | struct bio *bio; | |
1950 | struct bio_list bios; | |
1951 | ||
1952 | bio_list_init(&bios); | |
1953 | bio_list_merge(&bios, &cache->deferred_bios); | |
1954 | bio_list_init(&cache->deferred_bios); | |
1955 | ||
4246a0b6 | 1956 | while ((bio = bio_list_pop(&bios))) { |
4e4cbee9 | 1957 | bio->bi_status = BLK_STS_DM_REQUEUE; |
4246a0b6 CH |
1958 | bio_endio(bio); |
1959 | } | |
c6b4fcba JT |
1960 | } |
1961 | ||
c6b4fcba JT |
1962 | /* |
1963 | * We want to commit periodically so that not too much | |
1964 | * unwritten metadata builds up. | |
1965 | */ | |
1966 | static void do_waker(struct work_struct *ws) | |
1967 | { | |
1968 | struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker); | |
b29d4986 | 1969 | |
fba10109 | 1970 | policy_tick(cache->policy, true); |
b29d4986 JT |
1971 | wake_migration_worker(cache); |
1972 | schedule_commit(&cache->committer); | |
c6b4fcba JT |
1973 | queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD); |
1974 | } | |
1975 | ||
b29d4986 | 1976 | static void check_migrations(struct work_struct *ws) |
c6b4fcba | 1977 | { |
b29d4986 JT |
1978 | int r; |
1979 | struct policy_work *op; | |
1980 | struct cache *cache = container_of(ws, struct cache, migration_worker); | |
1981 | enum busy b; | |
c6b4fcba | 1982 | |
b29d4986 JT |
1983 | for (;;) { |
1984 | b = spare_migration_bandwidth(cache); | |
c6b4fcba | 1985 | |
b29d4986 JT |
1986 | r = policy_get_background_work(cache->policy, b == IDLE, &op); |
1987 | if (r == -ENODATA) | |
1988 | break; | |
1989 | ||
1990 | if (r) { | |
1991 | DMERR_LIMIT("%s: policy_background_work failed", | |
1992 | cache_device_name(cache)); | |
1993 | break; | |
1994 | } | |
1995 | ||
1996 | r = mg_start(cache, op, NULL); | |
1997 | if (r) | |
1998 | break; | |
1999 | } | |
c6b4fcba JT |
2000 | } |
2001 | ||
2002 | /*---------------------------------------------------------------- | |
2003 | * Target methods | |
2004 | *--------------------------------------------------------------*/ | |
2005 | ||
2006 | /* | |
2007 | * This function gets called on the error paths of the constructor, so we | |
2008 | * have to cope with a partially initialised struct. | |
2009 | */ | |
2010 | static void destroy(struct cache *cache) | |
2011 | { | |
2012 | unsigned i; | |
2013 | ||
6f1c819c | 2014 | mempool_exit(&cache->migration_pool); |
c6b4fcba | 2015 | |
c6b4fcba | 2016 | if (cache->prison) |
b29d4986 | 2017 | dm_bio_prison_destroy_v2(cache->prison); |
c6b4fcba JT |
2018 | |
2019 | if (cache->wq) | |
2020 | destroy_workqueue(cache->wq); | |
2021 | ||
2022 | if (cache->dirty_bitset) | |
2023 | free_bitset(cache->dirty_bitset); | |
2024 | ||
2025 | if (cache->discard_bitset) | |
2026 | free_bitset(cache->discard_bitset); | |
2027 | ||
2028 | if (cache->copier) | |
2029 | dm_kcopyd_client_destroy(cache->copier); | |
2030 | ||
2031 | if (cache->cmd) | |
2032 | dm_cache_metadata_close(cache->cmd); | |
2033 | ||
2034 | if (cache->metadata_dev) | |
2035 | dm_put_device(cache->ti, cache->metadata_dev); | |
2036 | ||
2037 | if (cache->origin_dev) | |
2038 | dm_put_device(cache->ti, cache->origin_dev); | |
2039 | ||
2040 | if (cache->cache_dev) | |
2041 | dm_put_device(cache->ti, cache->cache_dev); | |
2042 | ||
2043 | if (cache->policy) | |
2044 | dm_cache_policy_destroy(cache->policy); | |
2045 | ||
2046 | for (i = 0; i < cache->nr_ctr_args ; i++) | |
2047 | kfree(cache->ctr_args[i]); | |
2048 | kfree(cache->ctr_args); | |
2049 | ||
6f1c819c | 2050 | bioset_exit(&cache->bs); |
2df3bae9 | 2051 | |
c6b4fcba JT |
2052 | kfree(cache); |
2053 | } | |
2054 | ||
2055 | static void cache_dtr(struct dm_target *ti) | |
2056 | { | |
2057 | struct cache *cache = ti->private; | |
2058 | ||
2059 | destroy(cache); | |
2060 | } | |
2061 | ||
2062 | static sector_t get_dev_size(struct dm_dev *dev) | |
2063 | { | |
2064 | return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
2065 | } | |
2066 | ||
2067 | /*----------------------------------------------------------------*/ | |
2068 | ||
2069 | /* | |
2070 | * Construct a cache device mapping. | |
2071 | * | |
2072 | * cache <metadata dev> <cache dev> <origin dev> <block size> | |
2073 | * <#feature args> [<feature arg>]* | |
2074 | * <policy> <#policy args> [<policy arg>]* | |
2075 | * | |
2076 | * metadata dev : fast device holding the persistent metadata | |
2077 | * cache dev : fast device holding cached data blocks | |
2078 | * origin dev : slow device holding original data blocks | |
2079 | * block size : cache unit size in sectors | |
2080 | * | |
2081 | * #feature args : number of feature arguments passed | |
2082 | * feature args : writethrough. (The default is writeback.) | |
2083 | * | |
2084 | * policy : the replacement policy to use | |
2085 | * #policy args : an even number of policy arguments corresponding | |
2086 | * to key/value pairs passed to the policy | |
2087 | * policy args : key/value pairs passed to the policy | |
2088 | * E.g. 'sequential_threshold 1024' | |
2089 | * See cache-policies.txt for details. | |
2090 | * | |
2091 | * Optional feature arguments are: | |
2092 | * writethrough : write through caching that prohibits cache block | |
2093 | * content from being different from origin block content. | |
2094 | * Without this argument, the default behaviour is to write | |
2095 | * back cache block contents later for performance reasons, | |
2096 | * so they may differ from the corresponding origin blocks. | |
2097 | */ | |
2098 | struct cache_args { | |
2099 | struct dm_target *ti; | |
2100 | ||
2101 | struct dm_dev *metadata_dev; | |
2102 | ||
2103 | struct dm_dev *cache_dev; | |
2104 | sector_t cache_sectors; | |
2105 | ||
2106 | struct dm_dev *origin_dev; | |
2107 | sector_t origin_sectors; | |
2108 | ||
2109 | uint32_t block_size; | |
2110 | ||
2111 | const char *policy_name; | |
2112 | int policy_argc; | |
2113 | const char **policy_argv; | |
2114 | ||
2115 | struct cache_features features; | |
2116 | }; | |
2117 | ||
2118 | static void destroy_cache_args(struct cache_args *ca) | |
2119 | { | |
2120 | if (ca->metadata_dev) | |
2121 | dm_put_device(ca->ti, ca->metadata_dev); | |
2122 | ||
2123 | if (ca->cache_dev) | |
2124 | dm_put_device(ca->ti, ca->cache_dev); | |
2125 | ||
2126 | if (ca->origin_dev) | |
2127 | dm_put_device(ca->ti, ca->origin_dev); | |
2128 | ||
2129 | kfree(ca); | |
2130 | } | |
2131 | ||
2132 | static bool at_least_one_arg(struct dm_arg_set *as, char **error) | |
2133 | { | |
2134 | if (!as->argc) { | |
2135 | *error = "Insufficient args"; | |
2136 | return false; | |
2137 | } | |
2138 | ||
2139 | return true; | |
2140 | } | |
2141 | ||
2142 | static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2143 | char **error) | |
2144 | { | |
2145 | int r; | |
2146 | sector_t metadata_dev_size; | |
2147 | char b[BDEVNAME_SIZE]; | |
2148 | ||
2149 | if (!at_least_one_arg(as, error)) | |
2150 | return -EINVAL; | |
2151 | ||
2152 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2153 | &ca->metadata_dev); | |
2154 | if (r) { | |
2155 | *error = "Error opening metadata device"; | |
2156 | return r; | |
2157 | } | |
2158 | ||
2159 | metadata_dev_size = get_dev_size(ca->metadata_dev); | |
2160 | if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING) | |
2161 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
2162 | bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
2163 | ||
2164 | return 0; | |
2165 | } | |
2166 | ||
2167 | static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2168 | char **error) | |
2169 | { | |
2170 | int r; | |
2171 | ||
2172 | if (!at_least_one_arg(as, error)) | |
2173 | return -EINVAL; | |
2174 | ||
2175 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2176 | &ca->cache_dev); | |
2177 | if (r) { | |
2178 | *error = "Error opening cache device"; | |
2179 | return r; | |
2180 | } | |
2181 | ca->cache_sectors = get_dev_size(ca->cache_dev); | |
2182 | ||
2183 | return 0; | |
2184 | } | |
2185 | ||
2186 | static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2187 | char **error) | |
2188 | { | |
2189 | int r; | |
2190 | ||
2191 | if (!at_least_one_arg(as, error)) | |
2192 | return -EINVAL; | |
2193 | ||
2194 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2195 | &ca->origin_dev); | |
2196 | if (r) { | |
2197 | *error = "Error opening origin device"; | |
2198 | return r; | |
2199 | } | |
2200 | ||
2201 | ca->origin_sectors = get_dev_size(ca->origin_dev); | |
2202 | if (ca->ti->len > ca->origin_sectors) { | |
2203 | *error = "Device size larger than cached device"; | |
2204 | return -EINVAL; | |
2205 | } | |
2206 | ||
2207 | return 0; | |
2208 | } | |
2209 | ||
2210 | static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as, | |
2211 | char **error) | |
2212 | { | |
05473044 | 2213 | unsigned long block_size; |
c6b4fcba JT |
2214 | |
2215 | if (!at_least_one_arg(as, error)) | |
2216 | return -EINVAL; | |
2217 | ||
05473044 MS |
2218 | if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size || |
2219 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2220 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
2221 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { | |
c6b4fcba JT |
2222 | *error = "Invalid data block size"; |
2223 | return -EINVAL; | |
2224 | } | |
2225 | ||
05473044 | 2226 | if (block_size > ca->cache_sectors) { |
c6b4fcba JT |
2227 | *error = "Data block size is larger than the cache device"; |
2228 | return -EINVAL; | |
2229 | } | |
2230 | ||
05473044 | 2231 | ca->block_size = block_size; |
c6b4fcba JT |
2232 | |
2233 | return 0; | |
2234 | } | |
2235 | ||
2236 | static void init_features(struct cache_features *cf) | |
2237 | { | |
2238 | cf->mode = CM_WRITE; | |
2ee57d58 | 2239 | cf->io_mode = CM_IO_WRITEBACK; |
629d0a8a | 2240 | cf->metadata_version = 1; |
de7180ff | 2241 | cf->discard_passdown = true; |
c6b4fcba JT |
2242 | } |
2243 | ||
2244 | static int parse_features(struct cache_args *ca, struct dm_arg_set *as, | |
2245 | char **error) | |
2246 | { | |
5916a22b | 2247 | static const struct dm_arg _args[] = { |
de7180ff | 2248 | {0, 3, "Invalid number of cache feature arguments"}, |
c6b4fcba JT |
2249 | }; |
2250 | ||
af9313c3 | 2251 | int r, mode_ctr = 0; |
c6b4fcba JT |
2252 | unsigned argc; |
2253 | const char *arg; | |
2254 | struct cache_features *cf = &ca->features; | |
2255 | ||
2256 | init_features(cf); | |
2257 | ||
2258 | r = dm_read_arg_group(_args, as, &argc, error); | |
2259 | if (r) | |
2260 | return -EINVAL; | |
2261 | ||
2262 | while (argc--) { | |
2263 | arg = dm_shift_arg(as); | |
2264 | ||
af9313c3 | 2265 | if (!strcasecmp(arg, "writeback")) { |
2ee57d58 | 2266 | cf->io_mode = CM_IO_WRITEBACK; |
af9313c3 JP |
2267 | mode_ctr++; |
2268 | } | |
c6b4fcba | 2269 | |
af9313c3 | 2270 | else if (!strcasecmp(arg, "writethrough")) { |
2ee57d58 | 2271 | cf->io_mode = CM_IO_WRITETHROUGH; |
af9313c3 JP |
2272 | mode_ctr++; |
2273 | } | |
2ee57d58 | 2274 | |
af9313c3 | 2275 | else if (!strcasecmp(arg, "passthrough")) { |
2ee57d58 | 2276 | cf->io_mode = CM_IO_PASSTHROUGH; |
af9313c3 JP |
2277 | mode_ctr++; |
2278 | } | |
c6b4fcba | 2279 | |
629d0a8a JT |
2280 | else if (!strcasecmp(arg, "metadata2")) |
2281 | cf->metadata_version = 2; | |
2282 | ||
de7180ff MS |
2283 | else if (!strcasecmp(arg, "no_discard_passdown")) |
2284 | cf->discard_passdown = false; | |
2285 | ||
c6b4fcba JT |
2286 | else { |
2287 | *error = "Unrecognised cache feature requested"; | |
2288 | return -EINVAL; | |
2289 | } | |
2290 | } | |
2291 | ||
af9313c3 JP |
2292 | if (mode_ctr > 1) { |
2293 | *error = "Duplicate cache io_mode features requested"; | |
2294 | return -EINVAL; | |
2295 | } | |
2296 | ||
c6b4fcba JT |
2297 | return 0; |
2298 | } | |
2299 | ||
2300 | static int parse_policy(struct cache_args *ca, struct dm_arg_set *as, | |
2301 | char **error) | |
2302 | { | |
5916a22b | 2303 | static const struct dm_arg _args[] = { |
c6b4fcba JT |
2304 | {0, 1024, "Invalid number of policy arguments"}, |
2305 | }; | |
2306 | ||
2307 | int r; | |
2308 | ||
2309 | if (!at_least_one_arg(as, error)) | |
2310 | return -EINVAL; | |
2311 | ||
2312 | ca->policy_name = dm_shift_arg(as); | |
2313 | ||
2314 | r = dm_read_arg_group(_args, as, &ca->policy_argc, error); | |
2315 | if (r) | |
2316 | return -EINVAL; | |
2317 | ||
2318 | ca->policy_argv = (const char **)as->argv; | |
2319 | dm_consume_args(as, ca->policy_argc); | |
2320 | ||
2321 | return 0; | |
2322 | } | |
2323 | ||
2324 | static int parse_cache_args(struct cache_args *ca, int argc, char **argv, | |
2325 | char **error) | |
2326 | { | |
2327 | int r; | |
2328 | struct dm_arg_set as; | |
2329 | ||
2330 | as.argc = argc; | |
2331 | as.argv = argv; | |
2332 | ||
2333 | r = parse_metadata_dev(ca, &as, error); | |
2334 | if (r) | |
2335 | return r; | |
2336 | ||
2337 | r = parse_cache_dev(ca, &as, error); | |
2338 | if (r) | |
2339 | return r; | |
2340 | ||
2341 | r = parse_origin_dev(ca, &as, error); | |
2342 | if (r) | |
2343 | return r; | |
2344 | ||
2345 | r = parse_block_size(ca, &as, error); | |
2346 | if (r) | |
2347 | return r; | |
2348 | ||
2349 | r = parse_features(ca, &as, error); | |
2350 | if (r) | |
2351 | return r; | |
2352 | ||
2353 | r = parse_policy(ca, &as, error); | |
2354 | if (r) | |
2355 | return r; | |
2356 | ||
2357 | return 0; | |
2358 | } | |
2359 | ||
2360 | /*----------------------------------------------------------------*/ | |
2361 | ||
2362 | static struct kmem_cache *migration_cache; | |
2363 | ||
2c73c471 AK |
2364 | #define NOT_CORE_OPTION 1 |
2365 | ||
2f14f4b5 | 2366 | static int process_config_option(struct cache *cache, const char *key, const char *value) |
2c73c471 AK |
2367 | { |
2368 | unsigned long tmp; | |
2369 | ||
2f14f4b5 JT |
2370 | if (!strcasecmp(key, "migration_threshold")) { |
2371 | if (kstrtoul(value, 10, &tmp)) | |
2c73c471 AK |
2372 | return -EINVAL; |
2373 | ||
2374 | cache->migration_threshold = tmp; | |
2375 | return 0; | |
2376 | } | |
2377 | ||
2378 | return NOT_CORE_OPTION; | |
2379 | } | |
2380 | ||
2f14f4b5 JT |
2381 | static int set_config_value(struct cache *cache, const char *key, const char *value) |
2382 | { | |
2383 | int r = process_config_option(cache, key, value); | |
2384 | ||
2385 | if (r == NOT_CORE_OPTION) | |
2386 | r = policy_set_config_value(cache->policy, key, value); | |
2387 | ||
2388 | if (r) | |
2389 | DMWARN("bad config value for %s: %s", key, value); | |
2390 | ||
2391 | return r; | |
2392 | } | |
2393 | ||
2394 | static int set_config_values(struct cache *cache, int argc, const char **argv) | |
c6b4fcba JT |
2395 | { |
2396 | int r = 0; | |
2397 | ||
2398 | if (argc & 1) { | |
2399 | DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs."); | |
2400 | return -EINVAL; | |
2401 | } | |
2402 | ||
2403 | while (argc) { | |
2f14f4b5 JT |
2404 | r = set_config_value(cache, argv[0], argv[1]); |
2405 | if (r) | |
2406 | break; | |
c6b4fcba JT |
2407 | |
2408 | argc -= 2; | |
2409 | argv += 2; | |
2410 | } | |
2411 | ||
2412 | return r; | |
2413 | } | |
2414 | ||
2415 | static int create_cache_policy(struct cache *cache, struct cache_args *ca, | |
2416 | char **error) | |
2417 | { | |
4cb3e1db MP |
2418 | struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name, |
2419 | cache->cache_size, | |
2420 | cache->origin_sectors, | |
2421 | cache->sectors_per_block); | |
2422 | if (IS_ERR(p)) { | |
c6b4fcba | 2423 | *error = "Error creating cache's policy"; |
4cb3e1db | 2424 | return PTR_ERR(p); |
c6b4fcba | 2425 | } |
4cb3e1db | 2426 | cache->policy = p; |
b29d4986 | 2427 | BUG_ON(!cache->policy); |
c6b4fcba | 2428 | |
2f14f4b5 | 2429 | return 0; |
c6b4fcba JT |
2430 | } |
2431 | ||
08b18451 | 2432 | /* |
2bb812df JT |
2433 | * We want the discard block size to be at least the size of the cache |
2434 | * block size and have no more than 2^14 discard blocks across the origin. | |
08b18451 JT |
2435 | */ |
2436 | #define MAX_DISCARD_BLOCKS (1 << 14) | |
2437 | ||
2438 | static bool too_many_discard_blocks(sector_t discard_block_size, | |
2439 | sector_t origin_size) | |
2440 | { | |
2441 | (void) sector_div(origin_size, discard_block_size); | |
2442 | ||
2443 | return origin_size > MAX_DISCARD_BLOCKS; | |
2444 | } | |
2445 | ||
2446 | static sector_t calculate_discard_block_size(sector_t cache_block_size, | |
2447 | sector_t origin_size) | |
2448 | { | |
2bb812df | 2449 | sector_t discard_block_size = cache_block_size; |
08b18451 JT |
2450 | |
2451 | if (origin_size) | |
2452 | while (too_many_discard_blocks(discard_block_size, origin_size)) | |
2453 | discard_block_size *= 2; | |
2454 | ||
2455 | return discard_block_size; | |
2456 | } | |
2457 | ||
d1d9220c JT |
2458 | static void set_cache_size(struct cache *cache, dm_cblock_t size) |
2459 | { | |
2460 | dm_block_t nr_blocks = from_cblock(size); | |
2461 | ||
2462 | if (nr_blocks > (1 << 20) && cache->cache_size != size) | |
2463 | DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n" | |
2464 | "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n" | |
2465 | "Please consider increasing the cache block size to reduce the overall cache block count.", | |
2466 | (unsigned long long) nr_blocks); | |
2467 | ||
2468 | cache->cache_size = size; | |
2469 | } | |
2470 | ||
b29d4986 JT |
2471 | static int is_congested(struct dm_dev *dev, int bdi_bits) |
2472 | { | |
2473 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
2474 | return bdi_congested(q->backing_dev_info, bdi_bits); | |
2475 | } | |
2476 | ||
2477 | static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2478 | { | |
2479 | struct cache *cache = container_of(cb, struct cache, callbacks); | |
2480 | ||
2481 | return is_congested(cache->origin_dev, bdi_bits) || | |
2482 | is_congested(cache->cache_dev, bdi_bits); | |
2483 | } | |
2484 | ||
f8350daf | 2485 | #define DEFAULT_MIGRATION_THRESHOLD 2048 |
c6b4fcba | 2486 | |
c6b4fcba JT |
2487 | static int cache_create(struct cache_args *ca, struct cache **result) |
2488 | { | |
2489 | int r = 0; | |
2490 | char **error = &ca->ti->error; | |
2491 | struct cache *cache; | |
2492 | struct dm_target *ti = ca->ti; | |
2493 | dm_block_t origin_blocks; | |
2494 | struct dm_cache_metadata *cmd; | |
2495 | bool may_format = ca->features.mode == CM_WRITE; | |
2496 | ||
2497 | cache = kzalloc(sizeof(*cache), GFP_KERNEL); | |
2498 | if (!cache) | |
2499 | return -ENOMEM; | |
2500 | ||
2501 | cache->ti = ca->ti; | |
2502 | ti->private = cache; | |
c6b4fcba JT |
2503 | ti->num_flush_bios = 2; |
2504 | ti->flush_supported = true; | |
2505 | ||
2506 | ti->num_discard_bios = 1; | |
2507 | ti->discards_supported = true; | |
c6b4fcba | 2508 | |
693b960e | 2509 | ti->per_io_data_size = sizeof(struct per_bio_data); |
c6b4fcba | 2510 | |
693b960e | 2511 | cache->features = ca->features; |
2df3bae9 MS |
2512 | if (writethrough_mode(cache)) { |
2513 | /* Create bioset for writethrough bios issued to origin */ | |
6f1c819c KO |
2514 | r = bioset_init(&cache->bs, BIO_POOL_SIZE, 0, 0); |
2515 | if (r) | |
2df3bae9 MS |
2516 | goto bad; |
2517 | } | |
2518 | ||
c6b4fcba JT |
2519 | cache->callbacks.congested_fn = cache_is_congested; |
2520 | dm_table_add_target_callbacks(ti->table, &cache->callbacks); | |
2521 | ||
2522 | cache->metadata_dev = ca->metadata_dev; | |
2523 | cache->origin_dev = ca->origin_dev; | |
2524 | cache->cache_dev = ca->cache_dev; | |
2525 | ||
2526 | ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL; | |
2527 | ||
c6b4fcba | 2528 | origin_blocks = cache->origin_sectors = ca->origin_sectors; |
414dd67d | 2529 | origin_blocks = block_div(origin_blocks, ca->block_size); |
c6b4fcba JT |
2530 | cache->origin_blocks = to_oblock(origin_blocks); |
2531 | ||
2532 | cache->sectors_per_block = ca->block_size; | |
2533 | if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) { | |
2534 | r = -EINVAL; | |
2535 | goto bad; | |
2536 | } | |
2537 | ||
2538 | if (ca->block_size & (ca->block_size - 1)) { | |
2539 | dm_block_t cache_size = ca->cache_sectors; | |
2540 | ||
2541 | cache->sectors_per_block_shift = -1; | |
414dd67d | 2542 | cache_size = block_div(cache_size, ca->block_size); |
d1d9220c | 2543 | set_cache_size(cache, to_cblock(cache_size)); |
c6b4fcba JT |
2544 | } else { |
2545 | cache->sectors_per_block_shift = __ffs(ca->block_size); | |
d1d9220c | 2546 | set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift)); |
c6b4fcba JT |
2547 | } |
2548 | ||
2549 | r = create_cache_policy(cache, ca, error); | |
2550 | if (r) | |
2551 | goto bad; | |
2f14f4b5 | 2552 | |
c6b4fcba | 2553 | cache->policy_nr_args = ca->policy_argc; |
2f14f4b5 JT |
2554 | cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD; |
2555 | ||
2556 | r = set_config_values(cache, ca->policy_argc, ca->policy_argv); | |
2557 | if (r) { | |
2558 | *error = "Error setting cache policy's config values"; | |
2559 | goto bad; | |
2560 | } | |
c6b4fcba JT |
2561 | |
2562 | cmd = dm_cache_metadata_open(cache->metadata_dev->bdev, | |
2563 | ca->block_size, may_format, | |
629d0a8a JT |
2564 | dm_cache_policy_get_hint_size(cache->policy), |
2565 | ca->features.metadata_version); | |
c6b4fcba JT |
2566 | if (IS_ERR(cmd)) { |
2567 | *error = "Error creating metadata object"; | |
2568 | r = PTR_ERR(cmd); | |
2569 | goto bad; | |
2570 | } | |
2571 | cache->cmd = cmd; | |
028ae9f7 JT |
2572 | set_cache_mode(cache, CM_WRITE); |
2573 | if (get_cache_mode(cache) != CM_WRITE) { | |
2574 | *error = "Unable to get write access to metadata, please check/repair metadata."; | |
2575 | r = -EINVAL; | |
2576 | goto bad; | |
2577 | } | |
c6b4fcba | 2578 | |
8e3c3827 | 2579 | if (passthrough_mode(cache)) { |
2ee57d58 JT |
2580 | bool all_clean; |
2581 | ||
2582 | r = dm_cache_metadata_all_clean(cache->cmd, &all_clean); | |
2583 | if (r) { | |
2584 | *error = "dm_cache_metadata_all_clean() failed"; | |
2585 | goto bad; | |
2586 | } | |
2587 | ||
2588 | if (!all_clean) { | |
2589 | *error = "Cannot enter passthrough mode unless all blocks are clean"; | |
2590 | r = -EINVAL; | |
2591 | goto bad; | |
2592 | } | |
b29d4986 JT |
2593 | |
2594 | policy_allow_migrations(cache->policy, false); | |
2ee57d58 JT |
2595 | } |
2596 | ||
c6b4fcba JT |
2597 | spin_lock_init(&cache->lock); |
2598 | bio_list_init(&cache->deferred_bios); | |
a59db676 JT |
2599 | atomic_set(&cache->nr_allocated_migrations, 0); |
2600 | atomic_set(&cache->nr_io_migrations, 0); | |
c6b4fcba JT |
2601 | init_waitqueue_head(&cache->migration_wait); |
2602 | ||
fa4d683a | 2603 | r = -ENOMEM; |
44fa816b | 2604 | atomic_set(&cache->nr_dirty, 0); |
c6b4fcba JT |
2605 | cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size)); |
2606 | if (!cache->dirty_bitset) { | |
2607 | *error = "could not allocate dirty bitset"; | |
2608 | goto bad; | |
2609 | } | |
2610 | clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size)); | |
2611 | ||
08b18451 JT |
2612 | cache->discard_block_size = |
2613 | calculate_discard_block_size(cache->sectors_per_block, | |
2614 | cache->origin_sectors); | |
2572629a JT |
2615 | cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors, |
2616 | cache->discard_block_size)); | |
1bad9bc4 | 2617 | cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2618 | if (!cache->discard_bitset) { |
2619 | *error = "could not allocate discard bitset"; | |
2620 | goto bad; | |
2621 | } | |
1bad9bc4 | 2622 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2623 | |
2624 | cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); | |
2625 | if (IS_ERR(cache->copier)) { | |
2626 | *error = "could not create kcopyd client"; | |
2627 | r = PTR_ERR(cache->copier); | |
2628 | goto bad; | |
2629 | } | |
2630 | ||
b29d4986 | 2631 | cache->wq = alloc_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM, 0); |
c6b4fcba JT |
2632 | if (!cache->wq) { |
2633 | *error = "could not create workqueue for metadata object"; | |
2634 | goto bad; | |
2635 | } | |
b29d4986 | 2636 | INIT_WORK(&cache->deferred_bio_worker, process_deferred_bios); |
b29d4986 | 2637 | INIT_WORK(&cache->migration_worker, check_migrations); |
c6b4fcba | 2638 | INIT_DELAYED_WORK(&cache->waker, do_waker); |
c6b4fcba | 2639 | |
b29d4986 | 2640 | cache->prison = dm_bio_prison_create_v2(cache->wq); |
c6b4fcba JT |
2641 | if (!cache->prison) { |
2642 | *error = "could not create bio prison"; | |
2643 | goto bad; | |
2644 | } | |
2645 | ||
6f1c819c KO |
2646 | r = mempool_init_slab_pool(&cache->migration_pool, MIGRATION_POOL_SIZE, |
2647 | migration_cache); | |
2648 | if (r) { | |
c6b4fcba JT |
2649 | *error = "Error creating cache's migration mempool"; |
2650 | goto bad; | |
2651 | } | |
2652 | ||
c6b4fcba JT |
2653 | cache->need_tick_bio = true; |
2654 | cache->sized = false; | |
65790ff9 | 2655 | cache->invalidate = false; |
c6b4fcba JT |
2656 | cache->commit_requested = false; |
2657 | cache->loaded_mappings = false; | |
2658 | cache->loaded_discards = false; | |
2659 | ||
2660 | load_stats(cache); | |
2661 | ||
2662 | atomic_set(&cache->stats.demotion, 0); | |
2663 | atomic_set(&cache->stats.promotion, 0); | |
2664 | atomic_set(&cache->stats.copies_avoided, 0); | |
2665 | atomic_set(&cache->stats.cache_cell_clash, 0); | |
2666 | atomic_set(&cache->stats.commit_count, 0); | |
2667 | atomic_set(&cache->stats.discard_count, 0); | |
2668 | ||
65790ff9 JT |
2669 | spin_lock_init(&cache->invalidation_lock); |
2670 | INIT_LIST_HEAD(&cache->invalidation_requests); | |
2671 | ||
b29d4986 JT |
2672 | batcher_init(&cache->committer, commit_op, cache, |
2673 | issue_op, cache, cache->wq); | |
701e03e4 | 2674 | iot_init(&cache->tracker); |
066dbaa3 | 2675 | |
b29d4986 JT |
2676 | init_rwsem(&cache->background_work_lock); |
2677 | prevent_background_work(cache); | |
2678 | ||
c6b4fcba JT |
2679 | *result = cache; |
2680 | return 0; | |
c6b4fcba JT |
2681 | bad: |
2682 | destroy(cache); | |
2683 | return r; | |
2684 | } | |
2685 | ||
2686 | static int copy_ctr_args(struct cache *cache, int argc, const char **argv) | |
2687 | { | |
2688 | unsigned i; | |
2689 | const char **copy; | |
2690 | ||
2691 | copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL); | |
2692 | if (!copy) | |
2693 | return -ENOMEM; | |
2694 | for (i = 0; i < argc; i++) { | |
2695 | copy[i] = kstrdup(argv[i], GFP_KERNEL); | |
2696 | if (!copy[i]) { | |
2697 | while (i--) | |
2698 | kfree(copy[i]); | |
2699 | kfree(copy); | |
2700 | return -ENOMEM; | |
2701 | } | |
2702 | } | |
2703 | ||
2704 | cache->nr_ctr_args = argc; | |
2705 | cache->ctr_args = copy; | |
2706 | ||
2707 | return 0; | |
2708 | } | |
2709 | ||
2710 | static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2711 | { | |
2712 | int r = -EINVAL; | |
2713 | struct cache_args *ca; | |
2714 | struct cache *cache = NULL; | |
2715 | ||
2716 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
2717 | if (!ca) { | |
2718 | ti->error = "Error allocating memory for cache"; | |
2719 | return -ENOMEM; | |
2720 | } | |
2721 | ca->ti = ti; | |
2722 | ||
2723 | r = parse_cache_args(ca, argc, argv, &ti->error); | |
2724 | if (r) | |
2725 | goto out; | |
2726 | ||
2727 | r = cache_create(ca, &cache); | |
617a0b89 HM |
2728 | if (r) |
2729 | goto out; | |
c6b4fcba JT |
2730 | |
2731 | r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3); | |
2732 | if (r) { | |
2733 | destroy(cache); | |
2734 | goto out; | |
2735 | } | |
2736 | ||
2737 | ti->private = cache; | |
c6b4fcba JT |
2738 | out: |
2739 | destroy_cache_args(ca); | |
2740 | return r; | |
2741 | } | |
2742 | ||
651f5fa2 JT |
2743 | /*----------------------------------------------------------------*/ |
2744 | ||
2745 | static int cache_map(struct dm_target *ti, struct bio *bio) | |
c6b4fcba | 2746 | { |
651f5fa2 JT |
2747 | struct cache *cache = ti->private; |
2748 | ||
c6b4fcba | 2749 | int r; |
b29d4986 | 2750 | bool commit_needed; |
c6b4fcba | 2751 | dm_oblock_t block = get_bio_block(cache, bio); |
c6b4fcba | 2752 | |
693b960e | 2753 | init_per_bio_data(bio); |
e893fba9 | 2754 | if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) { |
c6b4fcba JT |
2755 | /* |
2756 | * This can only occur if the io goes to a partial block at | |
2757 | * the end of the origin device. We don't cache these. | |
2758 | * Just remap to the origin and carry on. | |
2759 | */ | |
e893fba9 | 2760 | remap_to_origin(cache, bio); |
651f5fa2 | 2761 | accounted_begin(cache, bio); |
c6b4fcba JT |
2762 | return DM_MAPIO_REMAPPED; |
2763 | } | |
2764 | ||
651f5fa2 | 2765 | if (discard_or_flush(bio)) { |
c6b4fcba JT |
2766 | defer_bio(cache, bio); |
2767 | return DM_MAPIO_SUBMITTED; | |
2768 | } | |
2769 | ||
b29d4986 JT |
2770 | r = map_bio(cache, bio, block, &commit_needed); |
2771 | if (commit_needed) | |
2772 | schedule_commit(&cache->committer); | |
c6b4fcba | 2773 | |
2ee57d58 | 2774 | return r; |
c6b4fcba JT |
2775 | } |
2776 | ||
693b960e | 2777 | static int cache_end_io(struct dm_target *ti, struct bio *bio, blk_status_t *error) |
c6b4fcba JT |
2778 | { |
2779 | struct cache *cache = ti->private; | |
2780 | unsigned long flags; | |
693b960e | 2781 | struct per_bio_data *pb = get_per_bio_data(bio); |
c6b4fcba JT |
2782 | |
2783 | if (pb->tick) { | |
fba10109 | 2784 | policy_tick(cache->policy, false); |
c6b4fcba JT |
2785 | |
2786 | spin_lock_irqsave(&cache->lock, flags); | |
2787 | cache->need_tick_bio = true; | |
2788 | spin_unlock_irqrestore(&cache->lock, flags); | |
2789 | } | |
2790 | ||
b29d4986 | 2791 | bio_drop_shared_lock(cache, bio); |
066dbaa3 | 2792 | accounted_complete(cache, bio); |
c6b4fcba | 2793 | |
1be56909 | 2794 | return DM_ENDIO_DONE; |
c6b4fcba JT |
2795 | } |
2796 | ||
2797 | static int write_dirty_bitset(struct cache *cache) | |
2798 | { | |
629d0a8a | 2799 | int r; |
c6b4fcba | 2800 | |
028ae9f7 JT |
2801 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
2802 | return -EINVAL; | |
2803 | ||
629d0a8a JT |
2804 | r = dm_cache_set_dirty_bits(cache->cmd, from_cblock(cache->cache_size), cache->dirty_bitset); |
2805 | if (r) | |
2806 | metadata_operation_failed(cache, "dm_cache_set_dirty_bits", r); | |
c6b4fcba | 2807 | |
629d0a8a | 2808 | return r; |
c6b4fcba JT |
2809 | } |
2810 | ||
2811 | static int write_discard_bitset(struct cache *cache) | |
2812 | { | |
2813 | unsigned i, r; | |
2814 | ||
028ae9f7 JT |
2815 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
2816 | return -EINVAL; | |
2817 | ||
1bad9bc4 JT |
2818 | r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size, |
2819 | cache->discard_nr_blocks); | |
c6b4fcba | 2820 | if (r) { |
b61d9509 | 2821 | DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache)); |
028ae9f7 | 2822 | metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r); |
c6b4fcba JT |
2823 | return r; |
2824 | } | |
2825 | ||
1bad9bc4 JT |
2826 | for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) { |
2827 | r = dm_cache_set_discard(cache->cmd, to_dblock(i), | |
2828 | is_discarded(cache, to_dblock(i))); | |
028ae9f7 JT |
2829 | if (r) { |
2830 | metadata_operation_failed(cache, "dm_cache_set_discard", r); | |
c6b4fcba | 2831 | return r; |
028ae9f7 JT |
2832 | } |
2833 | } | |
2834 | ||
2835 | return 0; | |
2836 | } | |
2837 | ||
2838 | static int write_hints(struct cache *cache) | |
2839 | { | |
2840 | int r; | |
2841 | ||
2842 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
2843 | return -EINVAL; | |
2844 | ||
2845 | r = dm_cache_write_hints(cache->cmd, cache->policy); | |
2846 | if (r) { | |
2847 | metadata_operation_failed(cache, "dm_cache_write_hints", r); | |
2848 | return r; | |
c6b4fcba JT |
2849 | } |
2850 | ||
2851 | return 0; | |
2852 | } | |
2853 | ||
c6b4fcba JT |
2854 | /* |
2855 | * returns true on success | |
2856 | */ | |
2857 | static bool sync_metadata(struct cache *cache) | |
2858 | { | |
2859 | int r1, r2, r3, r4; | |
2860 | ||
2861 | r1 = write_dirty_bitset(cache); | |
2862 | if (r1) | |
b61d9509 | 2863 | DMERR("%s: could not write dirty bitset", cache_device_name(cache)); |
c6b4fcba JT |
2864 | |
2865 | r2 = write_discard_bitset(cache); | |
2866 | if (r2) | |
b61d9509 | 2867 | DMERR("%s: could not write discard bitset", cache_device_name(cache)); |
c6b4fcba JT |
2868 | |
2869 | save_stats(cache); | |
2870 | ||
028ae9f7 | 2871 | r3 = write_hints(cache); |
c6b4fcba | 2872 | if (r3) |
b61d9509 | 2873 | DMERR("%s: could not write hints", cache_device_name(cache)); |
c6b4fcba JT |
2874 | |
2875 | /* | |
2876 | * If writing the above metadata failed, we still commit, but don't | |
2877 | * set the clean shutdown flag. This will effectively force every | |
2878 | * dirty bit to be set on reload. | |
2879 | */ | |
028ae9f7 | 2880 | r4 = commit(cache, !r1 && !r2 && !r3); |
c6b4fcba | 2881 | if (r4) |
b61d9509 | 2882 | DMERR("%s: could not write cache metadata", cache_device_name(cache)); |
c6b4fcba JT |
2883 | |
2884 | return !r1 && !r2 && !r3 && !r4; | |
2885 | } | |
2886 | ||
2887 | static void cache_postsuspend(struct dm_target *ti) | |
2888 | { | |
2889 | struct cache *cache = ti->private; | |
2890 | ||
b29d4986 JT |
2891 | prevent_background_work(cache); |
2892 | BUG_ON(atomic_read(&cache->nr_io_migrations)); | |
2893 | ||
2894 | cancel_delayed_work(&cache->waker); | |
2895 | flush_workqueue(cache->wq); | |
701e03e4 | 2896 | WARN_ON(cache->tracker.in_flight); |
b29d4986 JT |
2897 | |
2898 | /* | |
2899 | * If it's a flush suspend there won't be any deferred bios, so this | |
2900 | * call is harmless. | |
2901 | */ | |
651f5fa2 | 2902 | requeue_deferred_bios(cache); |
c6b4fcba | 2903 | |
028ae9f7 JT |
2904 | if (get_cache_mode(cache) == CM_WRITE) |
2905 | (void) sync_metadata(cache); | |
c6b4fcba JT |
2906 | } |
2907 | ||
2908 | static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock, | |
2909 | bool dirty, uint32_t hint, bool hint_valid) | |
2910 | { | |
2911 | int r; | |
2912 | struct cache *cache = context; | |
2913 | ||
449b668c JT |
2914 | if (dirty) { |
2915 | set_bit(from_cblock(cblock), cache->dirty_bitset); | |
2916 | atomic_inc(&cache->nr_dirty); | |
2917 | } else | |
2918 | clear_bit(from_cblock(cblock), cache->dirty_bitset); | |
2919 | ||
b29d4986 | 2920 | r = policy_load_mapping(cache->policy, oblock, cblock, dirty, hint, hint_valid); |
c6b4fcba JT |
2921 | if (r) |
2922 | return r; | |
2923 | ||
c6b4fcba JT |
2924 | return 0; |
2925 | } | |
2926 | ||
3e2e1c30 JT |
2927 | /* |
2928 | * The discard block size in the on disk metadata is not | |
2929 | * neccessarily the same as we're currently using. So we have to | |
2930 | * be careful to only set the discarded attribute if we know it | |
2931 | * covers a complete block of the new size. | |
2932 | */ | |
2933 | struct discard_load_info { | |
2934 | struct cache *cache; | |
2935 | ||
2936 | /* | |
2937 | * These blocks are sized using the on disk dblock size, rather | |
2938 | * than the current one. | |
2939 | */ | |
2940 | dm_block_t block_size; | |
2941 | dm_block_t discard_begin, discard_end; | |
2942 | }; | |
2943 | ||
2944 | static void discard_load_info_init(struct cache *cache, | |
2945 | struct discard_load_info *li) | |
2946 | { | |
2947 | li->cache = cache; | |
2948 | li->discard_begin = li->discard_end = 0; | |
2949 | } | |
2950 | ||
2951 | static void set_discard_range(struct discard_load_info *li) | |
2952 | { | |
2953 | sector_t b, e; | |
2954 | ||
2955 | if (li->discard_begin == li->discard_end) | |
2956 | return; | |
2957 | ||
2958 | /* | |
2959 | * Convert to sectors. | |
2960 | */ | |
2961 | b = li->discard_begin * li->block_size; | |
2962 | e = li->discard_end * li->block_size; | |
2963 | ||
2964 | /* | |
2965 | * Then convert back to the current dblock size. | |
2966 | */ | |
2967 | b = dm_sector_div_up(b, li->cache->discard_block_size); | |
2968 | sector_div(e, li->cache->discard_block_size); | |
2969 | ||
2970 | /* | |
2971 | * The origin may have shrunk, so we need to check we're still in | |
2972 | * bounds. | |
2973 | */ | |
2974 | if (e > from_dblock(li->cache->discard_nr_blocks)) | |
2975 | e = from_dblock(li->cache->discard_nr_blocks); | |
2976 | ||
2977 | for (; b < e; b++) | |
2978 | set_discard(li->cache, to_dblock(b)); | |
2979 | } | |
2980 | ||
c6b4fcba | 2981 | static int load_discard(void *context, sector_t discard_block_size, |
1bad9bc4 | 2982 | dm_dblock_t dblock, bool discard) |
c6b4fcba | 2983 | { |
3e2e1c30 | 2984 | struct discard_load_info *li = context; |
c6b4fcba | 2985 | |
3e2e1c30 | 2986 | li->block_size = discard_block_size; |
1bad9bc4 | 2987 | |
3e2e1c30 JT |
2988 | if (discard) { |
2989 | if (from_dblock(dblock) == li->discard_end) | |
2990 | /* | |
2991 | * We're already in a discard range, just extend it. | |
2992 | */ | |
2993 | li->discard_end = li->discard_end + 1ULL; | |
2994 | ||
2995 | else { | |
2996 | /* | |
2997 | * Emit the old range and start a new one. | |
2998 | */ | |
2999 | set_discard_range(li); | |
3000 | li->discard_begin = from_dblock(dblock); | |
3001 | li->discard_end = li->discard_begin + 1ULL; | |
3002 | } | |
3003 | } else { | |
3004 | set_discard_range(li); | |
3005 | li->discard_begin = li->discard_end = 0; | |
3006 | } | |
c6b4fcba JT |
3007 | |
3008 | return 0; | |
3009 | } | |
3010 | ||
f494a9c6 JT |
3011 | static dm_cblock_t get_cache_dev_size(struct cache *cache) |
3012 | { | |
3013 | sector_t size = get_dev_size(cache->cache_dev); | |
3014 | (void) sector_div(size, cache->sectors_per_block); | |
3015 | return to_cblock(size); | |
3016 | } | |
3017 | ||
3018 | static bool can_resize(struct cache *cache, dm_cblock_t new_size) | |
3019 | { | |
5d07384a MS |
3020 | if (from_cblock(new_size) > from_cblock(cache->cache_size)) { |
3021 | if (cache->sized) { | |
3022 | DMERR("%s: unable to extend cache due to missing cache table reload", | |
3023 | cache_device_name(cache)); | |
3024 | return false; | |
3025 | } | |
3026 | } | |
f494a9c6 JT |
3027 | |
3028 | /* | |
3029 | * We can't drop a dirty block when shrinking the cache. | |
3030 | */ | |
3031 | while (from_cblock(new_size) < from_cblock(cache->cache_size)) { | |
3032 | new_size = to_cblock(from_cblock(new_size) + 1); | |
3033 | if (is_dirty(cache, new_size)) { | |
b61d9509 MS |
3034 | DMERR("%s: unable to shrink cache; cache block %llu is dirty", |
3035 | cache_device_name(cache), | |
f494a9c6 JT |
3036 | (unsigned long long) from_cblock(new_size)); |
3037 | return false; | |
3038 | } | |
3039 | } | |
3040 | ||
3041 | return true; | |
3042 | } | |
3043 | ||
3044 | static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size) | |
3045 | { | |
3046 | int r; | |
3047 | ||
08844800 | 3048 | r = dm_cache_resize(cache->cmd, new_size); |
f494a9c6 | 3049 | if (r) { |
b61d9509 | 3050 | DMERR("%s: could not resize cache metadata", cache_device_name(cache)); |
028ae9f7 | 3051 | metadata_operation_failed(cache, "dm_cache_resize", r); |
f494a9c6 JT |
3052 | return r; |
3053 | } | |
3054 | ||
d1d9220c | 3055 | set_cache_size(cache, new_size); |
f494a9c6 JT |
3056 | |
3057 | return 0; | |
3058 | } | |
3059 | ||
c6b4fcba JT |
3060 | static int cache_preresume(struct dm_target *ti) |
3061 | { | |
3062 | int r = 0; | |
3063 | struct cache *cache = ti->private; | |
f494a9c6 | 3064 | dm_cblock_t csize = get_cache_dev_size(cache); |
c6b4fcba JT |
3065 | |
3066 | /* | |
3067 | * Check to see if the cache has resized. | |
3068 | */ | |
f494a9c6 JT |
3069 | if (!cache->sized) { |
3070 | r = resize_cache_dev(cache, csize); | |
3071 | if (r) | |
c6b4fcba | 3072 | return r; |
c6b4fcba JT |
3073 | |
3074 | cache->sized = true; | |
f494a9c6 JT |
3075 | |
3076 | } else if (csize != cache->cache_size) { | |
3077 | if (!can_resize(cache, csize)) | |
3078 | return -EINVAL; | |
3079 | ||
3080 | r = resize_cache_dev(cache, csize); | |
3081 | if (r) | |
3082 | return r; | |
c6b4fcba JT |
3083 | } |
3084 | ||
3085 | if (!cache->loaded_mappings) { | |
ea2dd8c1 | 3086 | r = dm_cache_load_mappings(cache->cmd, cache->policy, |
c6b4fcba JT |
3087 | load_mapping, cache); |
3088 | if (r) { | |
b61d9509 | 3089 | DMERR("%s: could not load cache mappings", cache_device_name(cache)); |
028ae9f7 | 3090 | metadata_operation_failed(cache, "dm_cache_load_mappings", r); |
c6b4fcba JT |
3091 | return r; |
3092 | } | |
3093 | ||
3094 | cache->loaded_mappings = true; | |
3095 | } | |
3096 | ||
3097 | if (!cache->loaded_discards) { | |
3e2e1c30 JT |
3098 | struct discard_load_info li; |
3099 | ||
3100 | /* | |
3101 | * The discard bitset could have been resized, or the | |
3102 | * discard block size changed. To be safe we start by | |
3103 | * setting every dblock to not discarded. | |
3104 | */ | |
3105 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); | |
3106 | ||
3107 | discard_load_info_init(cache, &li); | |
3108 | r = dm_cache_load_discards(cache->cmd, load_discard, &li); | |
c6b4fcba | 3109 | if (r) { |
b61d9509 | 3110 | DMERR("%s: could not load origin discards", cache_device_name(cache)); |
028ae9f7 | 3111 | metadata_operation_failed(cache, "dm_cache_load_discards", r); |
c6b4fcba JT |
3112 | return r; |
3113 | } | |
3e2e1c30 | 3114 | set_discard_range(&li); |
c6b4fcba JT |
3115 | |
3116 | cache->loaded_discards = true; | |
3117 | } | |
3118 | ||
3119 | return r; | |
3120 | } | |
3121 | ||
3122 | static void cache_resume(struct dm_target *ti) | |
3123 | { | |
3124 | struct cache *cache = ti->private; | |
3125 | ||
3126 | cache->need_tick_bio = true; | |
b29d4986 | 3127 | allow_background_work(cache); |
c6b4fcba JT |
3128 | do_waker(&cache->waker.work); |
3129 | } | |
3130 | ||
de7180ff MS |
3131 | static void emit_flags(struct cache *cache, char *result, |
3132 | unsigned maxlen, ssize_t *sz_ptr) | |
3133 | { | |
3134 | ssize_t sz = *sz_ptr; | |
3135 | struct cache_features *cf = &cache->features; | |
3136 | unsigned count = (cf->metadata_version == 2) + !cf->discard_passdown + 1; | |
3137 | ||
3138 | DMEMIT("%u ", count); | |
3139 | ||
3140 | if (cf->metadata_version == 2) | |
3141 | DMEMIT("metadata2 "); | |
3142 | ||
3143 | if (writethrough_mode(cache)) | |
3144 | DMEMIT("writethrough "); | |
3145 | ||
3146 | else if (passthrough_mode(cache)) | |
3147 | DMEMIT("passthrough "); | |
3148 | ||
3149 | else if (writeback_mode(cache)) | |
3150 | DMEMIT("writeback "); | |
3151 | ||
3152 | else { | |
3153 | DMEMIT("unknown "); | |
3154 | DMERR("%s: internal error: unknown io mode: %d", | |
3155 | cache_device_name(cache), (int) cf->io_mode); | |
3156 | } | |
3157 | ||
3158 | if (!cf->discard_passdown) | |
3159 | DMEMIT("no_discard_passdown "); | |
3160 | ||
3161 | *sz_ptr = sz; | |
3162 | } | |
3163 | ||
c6b4fcba JT |
3164 | /* |
3165 | * Status format: | |
3166 | * | |
6a388618 MS |
3167 | * <metadata block size> <#used metadata blocks>/<#total metadata blocks> |
3168 | * <cache block size> <#used cache blocks>/<#total cache blocks> | |
c6b4fcba | 3169 | * <#read hits> <#read misses> <#write hits> <#write misses> |
6a388618 | 3170 | * <#demotions> <#promotions> <#dirty> |
c6b4fcba JT |
3171 | * <#features> <features>* |
3172 | * <#core args> <core args> | |
255eac20 | 3173 | * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check> |
c6b4fcba JT |
3174 | */ |
3175 | static void cache_status(struct dm_target *ti, status_type_t type, | |
3176 | unsigned status_flags, char *result, unsigned maxlen) | |
3177 | { | |
3178 | int r = 0; | |
3179 | unsigned i; | |
3180 | ssize_t sz = 0; | |
3181 | dm_block_t nr_free_blocks_metadata = 0; | |
3182 | dm_block_t nr_blocks_metadata = 0; | |
3183 | char buf[BDEVNAME_SIZE]; | |
3184 | struct cache *cache = ti->private; | |
3185 | dm_cblock_t residency; | |
d14fcf3d | 3186 | bool needs_check; |
c6b4fcba JT |
3187 | |
3188 | switch (type) { | |
3189 | case STATUSTYPE_INFO: | |
028ae9f7 JT |
3190 | if (get_cache_mode(cache) == CM_FAIL) { |
3191 | DMEMIT("Fail"); | |
3192 | break; | |
c6b4fcba JT |
3193 | } |
3194 | ||
028ae9f7 JT |
3195 | /* Commit to ensure statistics aren't out-of-date */ |
3196 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
3197 | (void) commit(cache, false); | |
3198 | ||
b61d9509 | 3199 | r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata); |
c6b4fcba | 3200 | if (r) { |
b61d9509 MS |
3201 | DMERR("%s: dm_cache_get_free_metadata_block_count returned %d", |
3202 | cache_device_name(cache), r); | |
c6b4fcba JT |
3203 | goto err; |
3204 | } | |
3205 | ||
3206 | r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata); | |
3207 | if (r) { | |
b61d9509 MS |
3208 | DMERR("%s: dm_cache_get_metadata_dev_size returned %d", |
3209 | cache_device_name(cache), r); | |
c6b4fcba JT |
3210 | goto err; |
3211 | } | |
3212 | ||
3213 | residency = policy_residency(cache->policy); | |
3214 | ||
ca763d0a | 3215 | DMEMIT("%u %llu/%llu %llu %llu/%llu %u %u %u %u %u %u %lu ", |
895b47d7 | 3216 | (unsigned)DM_CACHE_METADATA_BLOCK_SIZE, |
c6b4fcba JT |
3217 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
3218 | (unsigned long long)nr_blocks_metadata, | |
ca763d0a | 3219 | (unsigned long long)cache->sectors_per_block, |
6a388618 MS |
3220 | (unsigned long long) from_cblock(residency), |
3221 | (unsigned long long) from_cblock(cache->cache_size), | |
c6b4fcba JT |
3222 | (unsigned) atomic_read(&cache->stats.read_hit), |
3223 | (unsigned) atomic_read(&cache->stats.read_miss), | |
3224 | (unsigned) atomic_read(&cache->stats.write_hit), | |
3225 | (unsigned) atomic_read(&cache->stats.write_miss), | |
3226 | (unsigned) atomic_read(&cache->stats.demotion), | |
3227 | (unsigned) atomic_read(&cache->stats.promotion), | |
44fa816b | 3228 | (unsigned long) atomic_read(&cache->nr_dirty)); |
c6b4fcba | 3229 | |
de7180ff | 3230 | emit_flags(cache, result, maxlen, &sz); |
c6b4fcba JT |
3231 | |
3232 | DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold); | |
2e68c4e6 MS |
3233 | |
3234 | DMEMIT("%s ", dm_cache_policy_get_name(cache->policy)); | |
c6b4fcba | 3235 | if (sz < maxlen) { |
028ae9f7 | 3236 | r = policy_emit_config_values(cache->policy, result, maxlen, &sz); |
c6b4fcba | 3237 | if (r) |
b61d9509 MS |
3238 | DMERR("%s: policy_emit_config_values returned %d", |
3239 | cache_device_name(cache), r); | |
c6b4fcba JT |
3240 | } |
3241 | ||
028ae9f7 JT |
3242 | if (get_cache_mode(cache) == CM_READ_ONLY) |
3243 | DMEMIT("ro "); | |
3244 | else | |
3245 | DMEMIT("rw "); | |
3246 | ||
d14fcf3d JT |
3247 | r = dm_cache_metadata_needs_check(cache->cmd, &needs_check); |
3248 | ||
3249 | if (r || needs_check) | |
255eac20 MS |
3250 | DMEMIT("needs_check "); |
3251 | else | |
3252 | DMEMIT("- "); | |
3253 | ||
c6b4fcba JT |
3254 | break; |
3255 | ||
3256 | case STATUSTYPE_TABLE: | |
3257 | format_dev_t(buf, cache->metadata_dev->bdev->bd_dev); | |
3258 | DMEMIT("%s ", buf); | |
3259 | format_dev_t(buf, cache->cache_dev->bdev->bd_dev); | |
3260 | DMEMIT("%s ", buf); | |
3261 | format_dev_t(buf, cache->origin_dev->bdev->bd_dev); | |
3262 | DMEMIT("%s", buf); | |
3263 | ||
3264 | for (i = 0; i < cache->nr_ctr_args - 1; i++) | |
3265 | DMEMIT(" %s", cache->ctr_args[i]); | |
3266 | if (cache->nr_ctr_args) | |
3267 | DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]); | |
3268 | } | |
3269 | ||
3270 | return; | |
3271 | ||
3272 | err: | |
3273 | DMEMIT("Error"); | |
3274 | } | |
3275 | ||
b29d4986 JT |
3276 | /* |
3277 | * Defines a range of cblocks, begin to (end - 1) are in the range. end is | |
3278 | * the one-past-the-end value. | |
3279 | */ | |
3280 | struct cblock_range { | |
3281 | dm_cblock_t begin; | |
3282 | dm_cblock_t end; | |
3283 | }; | |
3284 | ||
c6b4fcba | 3285 | /* |
65790ff9 JT |
3286 | * A cache block range can take two forms: |
3287 | * | |
3288 | * i) A single cblock, eg. '3456' | |
b29d4986 | 3289 | * ii) A begin and end cblock with a dash between, eg. 123-234 |
65790ff9 JT |
3290 | */ |
3291 | static int parse_cblock_range(struct cache *cache, const char *str, | |
3292 | struct cblock_range *result) | |
3293 | { | |
3294 | char dummy; | |
3295 | uint64_t b, e; | |
3296 | int r; | |
3297 | ||
3298 | /* | |
3299 | * Try and parse form (ii) first. | |
3300 | */ | |
3301 | r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy); | |
3302 | if (r < 0) | |
3303 | return r; | |
3304 | ||
3305 | if (r == 2) { | |
3306 | result->begin = to_cblock(b); | |
3307 | result->end = to_cblock(e); | |
3308 | return 0; | |
3309 | } | |
3310 | ||
3311 | /* | |
3312 | * That didn't work, try form (i). | |
3313 | */ | |
3314 | r = sscanf(str, "%llu%c", &b, &dummy); | |
3315 | if (r < 0) | |
3316 | return r; | |
3317 | ||
3318 | if (r == 1) { | |
3319 | result->begin = to_cblock(b); | |
3320 | result->end = to_cblock(from_cblock(result->begin) + 1u); | |
3321 | return 0; | |
3322 | } | |
3323 | ||
b61d9509 | 3324 | DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str); |
65790ff9 JT |
3325 | return -EINVAL; |
3326 | } | |
3327 | ||
3328 | static int validate_cblock_range(struct cache *cache, struct cblock_range *range) | |
3329 | { | |
3330 | uint64_t b = from_cblock(range->begin); | |
3331 | uint64_t e = from_cblock(range->end); | |
3332 | uint64_t n = from_cblock(cache->cache_size); | |
3333 | ||
3334 | if (b >= n) { | |
b61d9509 MS |
3335 | DMERR("%s: begin cblock out of range: %llu >= %llu", |
3336 | cache_device_name(cache), b, n); | |
65790ff9 JT |
3337 | return -EINVAL; |
3338 | } | |
3339 | ||
3340 | if (e > n) { | |
b61d9509 MS |
3341 | DMERR("%s: end cblock out of range: %llu > %llu", |
3342 | cache_device_name(cache), e, n); | |
65790ff9 JT |
3343 | return -EINVAL; |
3344 | } | |
3345 | ||
3346 | if (b >= e) { | |
b61d9509 MS |
3347 | DMERR("%s: invalid cblock range: %llu >= %llu", |
3348 | cache_device_name(cache), b, e); | |
65790ff9 JT |
3349 | return -EINVAL; |
3350 | } | |
3351 | ||
3352 | return 0; | |
3353 | } | |
3354 | ||
b29d4986 JT |
3355 | static inline dm_cblock_t cblock_succ(dm_cblock_t b) |
3356 | { | |
3357 | return to_cblock(from_cblock(b) + 1); | |
3358 | } | |
3359 | ||
65790ff9 JT |
3360 | static int request_invalidation(struct cache *cache, struct cblock_range *range) |
3361 | { | |
b29d4986 | 3362 | int r = 0; |
65790ff9 | 3363 | |
b29d4986 JT |
3364 | /* |
3365 | * We don't need to do any locking here because we know we're in | |
3366 | * passthrough mode. There's is potential for a race between an | |
3367 | * invalidation triggered by an io and an invalidation message. This | |
3368 | * is harmless, we must not worry if the policy call fails. | |
3369 | */ | |
3370 | while (range->begin != range->end) { | |
3371 | r = invalidate_cblock(cache, range->begin); | |
3372 | if (r) | |
3373 | return r; | |
65790ff9 | 3374 | |
b29d4986 JT |
3375 | range->begin = cblock_succ(range->begin); |
3376 | } | |
65790ff9 | 3377 | |
b29d4986 JT |
3378 | cache->commit_requested = true; |
3379 | return r; | |
65790ff9 JT |
3380 | } |
3381 | ||
3382 | static int process_invalidate_cblocks_message(struct cache *cache, unsigned count, | |
3383 | const char **cblock_ranges) | |
3384 | { | |
3385 | int r = 0; | |
3386 | unsigned i; | |
3387 | struct cblock_range range; | |
3388 | ||
8e3c3827 | 3389 | if (!passthrough_mode(cache)) { |
b61d9509 MS |
3390 | DMERR("%s: cache has to be in passthrough mode for invalidation", |
3391 | cache_device_name(cache)); | |
65790ff9 JT |
3392 | return -EPERM; |
3393 | } | |
3394 | ||
3395 | for (i = 0; i < count; i++) { | |
3396 | r = parse_cblock_range(cache, cblock_ranges[i], &range); | |
3397 | if (r) | |
3398 | break; | |
3399 | ||
3400 | r = validate_cblock_range(cache, &range); | |
3401 | if (r) | |
3402 | break; | |
3403 | ||
3404 | /* | |
3405 | * Pass begin and end origin blocks to the worker and wake it. | |
3406 | */ | |
3407 | r = request_invalidation(cache, &range); | |
3408 | if (r) | |
3409 | break; | |
3410 | } | |
3411 | ||
3412 | return r; | |
3413 | } | |
3414 | ||
3415 | /* | |
3416 | * Supports | |
3417 | * "<key> <value>" | |
3418 | * and | |
3419 | * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]* | |
c6b4fcba JT |
3420 | * |
3421 | * The key migration_threshold is supported by the cache target core. | |
3422 | */ | |
1eb5fa84 MS |
3423 | static int cache_message(struct dm_target *ti, unsigned argc, char **argv, |
3424 | char *result, unsigned maxlen) | |
c6b4fcba | 3425 | { |
c6b4fcba JT |
3426 | struct cache *cache = ti->private; |
3427 | ||
65790ff9 JT |
3428 | if (!argc) |
3429 | return -EINVAL; | |
3430 | ||
028ae9f7 | 3431 | if (get_cache_mode(cache) >= CM_READ_ONLY) { |
b61d9509 MS |
3432 | DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode", |
3433 | cache_device_name(cache)); | |
028ae9f7 JT |
3434 | return -EOPNOTSUPP; |
3435 | } | |
3436 | ||
7b6b2bc9 | 3437 | if (!strcasecmp(argv[0], "invalidate_cblocks")) |
65790ff9 JT |
3438 | return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1); |
3439 | ||
c6b4fcba JT |
3440 | if (argc != 2) |
3441 | return -EINVAL; | |
3442 | ||
2f14f4b5 | 3443 | return set_config_value(cache, argv[0], argv[1]); |
c6b4fcba JT |
3444 | } |
3445 | ||
3446 | static int cache_iterate_devices(struct dm_target *ti, | |
3447 | iterate_devices_callout_fn fn, void *data) | |
3448 | { | |
3449 | int r = 0; | |
3450 | struct cache *cache = ti->private; | |
3451 | ||
3452 | r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data); | |
3453 | if (!r) | |
3454 | r = fn(ti, cache->origin_dev, 0, ti->len, data); | |
3455 | ||
3456 | return r; | |
3457 | } | |
3458 | ||
de7180ff MS |
3459 | static bool origin_dev_supports_discard(struct block_device *origin_bdev) |
3460 | { | |
3461 | struct request_queue *q = bdev_get_queue(origin_bdev); | |
3462 | ||
3463 | return q && blk_queue_discard(q); | |
3464 | } | |
3465 | ||
3466 | /* | |
3467 | * If discard_passdown was enabled verify that the origin device | |
3468 | * supports discards. Disable discard_passdown if not. | |
3469 | */ | |
3470 | static void disable_passdown_if_not_supported(struct cache *cache) | |
3471 | { | |
3472 | struct block_device *origin_bdev = cache->origin_dev->bdev; | |
3473 | struct queue_limits *origin_limits = &bdev_get_queue(origin_bdev)->limits; | |
3474 | const char *reason = NULL; | |
3475 | char buf[BDEVNAME_SIZE]; | |
3476 | ||
3477 | if (!cache->features.discard_passdown) | |
3478 | return; | |
3479 | ||
3480 | if (!origin_dev_supports_discard(origin_bdev)) | |
3481 | reason = "discard unsupported"; | |
3482 | ||
3483 | else if (origin_limits->max_discard_sectors < cache->sectors_per_block) | |
3484 | reason = "max discard sectors smaller than a block"; | |
3485 | ||
3486 | if (reason) { | |
3487 | DMWARN("Origin device (%s) %s: Disabling discard passdown.", | |
3488 | bdevname(origin_bdev, buf), reason); | |
3489 | cache->features.discard_passdown = false; | |
3490 | } | |
3491 | } | |
3492 | ||
c6b4fcba JT |
3493 | static void set_discard_limits(struct cache *cache, struct queue_limits *limits) |
3494 | { | |
de7180ff MS |
3495 | struct block_device *origin_bdev = cache->origin_dev->bdev; |
3496 | struct queue_limits *origin_limits = &bdev_get_queue(origin_bdev)->limits; | |
3497 | ||
3498 | if (!cache->features.discard_passdown) { | |
3499 | /* No passdown is done so setting own virtual limits */ | |
3500 | limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024, | |
3501 | cache->origin_sectors); | |
3502 | limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT; | |
3503 | return; | |
3504 | } | |
3505 | ||
c6b4fcba | 3506 | /* |
de7180ff MS |
3507 | * cache_iterate_devices() is stacking both origin and fast device limits |
3508 | * but discards aren't passed to fast device, so inherit origin's limits. | |
c6b4fcba | 3509 | */ |
de7180ff MS |
3510 | limits->max_discard_sectors = origin_limits->max_discard_sectors; |
3511 | limits->max_hw_discard_sectors = origin_limits->max_hw_discard_sectors; | |
3512 | limits->discard_granularity = origin_limits->discard_granularity; | |
3513 | limits->discard_alignment = origin_limits->discard_alignment; | |
3514 | limits->discard_misaligned = origin_limits->discard_misaligned; | |
c6b4fcba JT |
3515 | } |
3516 | ||
3517 | static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3518 | { | |
3519 | struct cache *cache = ti->private; | |
f6109372 | 3520 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
c6b4fcba | 3521 | |
f6109372 MS |
3522 | /* |
3523 | * If the system-determined stacked limits are compatible with the | |
3524 | * cache's blocksize (io_opt is a factor) do not override them. | |
3525 | */ | |
3526 | if (io_opt_sectors < cache->sectors_per_block || | |
3527 | do_div(io_opt_sectors, cache->sectors_per_block)) { | |
b0246530 | 3528 | blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT); |
f6109372 MS |
3529 | blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT); |
3530 | } | |
de7180ff MS |
3531 | |
3532 | disable_passdown_if_not_supported(cache); | |
c6b4fcba JT |
3533 | set_discard_limits(cache, limits); |
3534 | } | |
3535 | ||
3536 | /*----------------------------------------------------------------*/ | |
3537 | ||
3538 | static struct target_type cache_target = { | |
3539 | .name = "cache", | |
de7180ff | 3540 | .version = {2, 1, 0}, |
c6b4fcba JT |
3541 | .module = THIS_MODULE, |
3542 | .ctr = cache_ctr, | |
3543 | .dtr = cache_dtr, | |
3544 | .map = cache_map, | |
3545 | .end_io = cache_end_io, | |
3546 | .postsuspend = cache_postsuspend, | |
3547 | .preresume = cache_preresume, | |
3548 | .resume = cache_resume, | |
3549 | .status = cache_status, | |
3550 | .message = cache_message, | |
3551 | .iterate_devices = cache_iterate_devices, | |
c6b4fcba JT |
3552 | .io_hints = cache_io_hints, |
3553 | }; | |
3554 | ||
3555 | static int __init dm_cache_init(void) | |
3556 | { | |
3557 | int r; | |
3558 | ||
c6b4fcba | 3559 | migration_cache = KMEM_CACHE(dm_cache_migration, 0); |
c7cd5550 | 3560 | if (!migration_cache) |
c6b4fcba | 3561 | return -ENOMEM; |
c6b4fcba | 3562 | |
7e6358d2 | 3563 | r = dm_register_target(&cache_target); |
3564 | if (r) { | |
3565 | DMERR("cache target registration failed: %d", r); | |
c7cd5550 | 3566 | kmem_cache_destroy(migration_cache); |
7e6358d2 | 3567 | return r; |
3568 | } | |
3569 | ||
c6b4fcba JT |
3570 | return 0; |
3571 | } | |
3572 | ||
3573 | static void __exit dm_cache_exit(void) | |
3574 | { | |
3575 | dm_unregister_target(&cache_target); | |
3576 | kmem_cache_destroy(migration_cache); | |
3577 | } | |
3578 | ||
3579 | module_init(dm_cache_init); | |
3580 | module_exit(dm_cache_exit); | |
3581 | ||
3582 | MODULE_DESCRIPTION(DM_NAME " cache target"); | |
3583 | MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>"); | |
3584 | MODULE_LICENSE("GPL"); |