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 | 10 | #include "dm-cache-metadata.h" |
dc4fa29f | 11 | #include "dm-io-tracker.h" |
c6b4fcba JT |
12 | |
13 | #include <linux/dm-io.h> | |
14 | #include <linux/dm-kcopyd.h> | |
0f30af98 | 15 | #include <linux/jiffies.h> |
c6b4fcba JT |
16 | #include <linux/init.h> |
17 | #include <linux/mempool.h> | |
18 | #include <linux/module.h> | |
b29d4986 | 19 | #include <linux/rwsem.h> |
c6b4fcba JT |
20 | #include <linux/slab.h> |
21 | #include <linux/vmalloc.h> | |
22 | ||
23 | #define DM_MSG_PREFIX "cache" | |
24 | ||
25 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle, | |
26 | "A percentage of time allocated for copying to and/or from cache"); | |
27 | ||
28 | /*----------------------------------------------------------------*/ | |
29 | ||
b29d4986 JT |
30 | /* |
31 | * Glossary: | |
32 | * | |
33 | * oblock: index of an origin block | |
34 | * cblock: index of a cache block | |
35 | * promotion: movement of a block from origin to cache | |
36 | * demotion: movement of a block from cache to origin | |
37 | * migration: movement of a block between the origin and cache device, | |
38 | * either direction | |
39 | */ | |
40 | ||
41 | /*----------------------------------------------------------------*/ | |
77289d32 | 42 | |
c6b4fcba | 43 | /* |
b29d4986 JT |
44 | * Represents a chunk of future work. 'input' allows continuations to pass |
45 | * values between themselves, typically error values. | |
c6b4fcba | 46 | */ |
b29d4986 JT |
47 | struct continuation { |
48 | struct work_struct ws; | |
4e4cbee9 | 49 | blk_status_t input; |
b29d4986 JT |
50 | }; |
51 | ||
52 | static inline void init_continuation(struct continuation *k, | |
53 | void (*fn)(struct work_struct *)) | |
54 | { | |
55 | INIT_WORK(&k->ws, fn); | |
56 | k->input = 0; | |
57 | } | |
58 | ||
59 | static inline void queue_continuation(struct workqueue_struct *wq, | |
60 | struct continuation *k) | |
61 | { | |
62 | queue_work(wq, &k->ws); | |
63 | } | |
c6b4fcba JT |
64 | |
65 | /*----------------------------------------------------------------*/ | |
66 | ||
b29d4986 JT |
67 | /* |
68 | * The batcher collects together pieces of work that need a particular | |
69 | * operation to occur before they can proceed (typically a commit). | |
70 | */ | |
71 | struct batcher { | |
72 | /* | |
73 | * The operation that everyone is waiting for. | |
74 | */ | |
4e4cbee9 | 75 | blk_status_t (*commit_op)(void *context); |
b29d4986 JT |
76 | void *commit_context; |
77 | ||
78 | /* | |
79 | * This is how bios should be issued once the commit op is complete | |
80 | * (accounted_request). | |
81 | */ | |
82 | void (*issue_op)(struct bio *bio, void *context); | |
83 | void *issue_context; | |
84 | ||
85 | /* | |
86 | * Queued work gets put on here after commit. | |
87 | */ | |
88 | struct workqueue_struct *wq; | |
89 | ||
90 | spinlock_t lock; | |
91 | struct list_head work_items; | |
92 | struct bio_list bios; | |
93 | struct work_struct commit_work; | |
94 | ||
95 | bool commit_scheduled; | |
96 | }; | |
97 | ||
98 | static void __commit(struct work_struct *_ws) | |
99 | { | |
100 | struct batcher *b = container_of(_ws, struct batcher, commit_work); | |
4e4cbee9 | 101 | blk_status_t r; |
b29d4986 JT |
102 | struct list_head work_items; |
103 | struct work_struct *ws, *tmp; | |
104 | struct continuation *k; | |
105 | struct bio *bio; | |
106 | struct bio_list bios; | |
107 | ||
108 | INIT_LIST_HEAD(&work_items); | |
109 | bio_list_init(&bios); | |
110 | ||
111 | /* | |
112 | * We have to grab these before the commit_op to avoid a race | |
113 | * condition. | |
114 | */ | |
26b924b9 | 115 | spin_lock_irq(&b->lock); |
b29d4986 JT |
116 | list_splice_init(&b->work_items, &work_items); |
117 | bio_list_merge(&bios, &b->bios); | |
118 | bio_list_init(&b->bios); | |
119 | b->commit_scheduled = false; | |
26b924b9 | 120 | spin_unlock_irq(&b->lock); |
b29d4986 JT |
121 | |
122 | r = b->commit_op(b->commit_context); | |
123 | ||
124 | list_for_each_entry_safe(ws, tmp, &work_items, entry) { | |
125 | k = container_of(ws, struct continuation, ws); | |
126 | k->input = r; | |
127 | INIT_LIST_HEAD(&ws->entry); /* to avoid a WARN_ON */ | |
128 | queue_work(b->wq, ws); | |
129 | } | |
130 | ||
131 | while ((bio = bio_list_pop(&bios))) { | |
132 | if (r) { | |
4e4cbee9 | 133 | bio->bi_status = r; |
b29d4986 JT |
134 | bio_endio(bio); |
135 | } else | |
136 | b->issue_op(bio, b->issue_context); | |
137 | } | |
138 | } | |
139 | ||
140 | static void batcher_init(struct batcher *b, | |
4e4cbee9 | 141 | blk_status_t (*commit_op)(void *), |
b29d4986 JT |
142 | void *commit_context, |
143 | void (*issue_op)(struct bio *bio, void *), | |
144 | void *issue_context, | |
145 | struct workqueue_struct *wq) | |
146 | { | |
147 | b->commit_op = commit_op; | |
148 | b->commit_context = commit_context; | |
149 | b->issue_op = issue_op; | |
150 | b->issue_context = issue_context; | |
151 | b->wq = wq; | |
152 | ||
153 | spin_lock_init(&b->lock); | |
154 | INIT_LIST_HEAD(&b->work_items); | |
155 | bio_list_init(&b->bios); | |
156 | INIT_WORK(&b->commit_work, __commit); | |
157 | b->commit_scheduled = false; | |
158 | } | |
159 | ||
160 | static void async_commit(struct batcher *b) | |
161 | { | |
162 | queue_work(b->wq, &b->commit_work); | |
163 | } | |
164 | ||
165 | static void continue_after_commit(struct batcher *b, struct continuation *k) | |
166 | { | |
b29d4986 JT |
167 | bool commit_scheduled; |
168 | ||
26b924b9 | 169 | spin_lock_irq(&b->lock); |
b29d4986 JT |
170 | commit_scheduled = b->commit_scheduled; |
171 | list_add_tail(&k->ws.entry, &b->work_items); | |
26b924b9 | 172 | spin_unlock_irq(&b->lock); |
b29d4986 JT |
173 | |
174 | if (commit_scheduled) | |
175 | async_commit(b); | |
176 | } | |
177 | ||
178 | /* | |
179 | * Bios are errored if commit failed. | |
180 | */ | |
181 | static void issue_after_commit(struct batcher *b, struct bio *bio) | |
182 | { | |
b29d4986 JT |
183 | bool commit_scheduled; |
184 | ||
26b924b9 | 185 | spin_lock_irq(&b->lock); |
b29d4986 JT |
186 | commit_scheduled = b->commit_scheduled; |
187 | bio_list_add(&b->bios, bio); | |
26b924b9 | 188 | spin_unlock_irq(&b->lock); |
b29d4986 JT |
189 | |
190 | if (commit_scheduled) | |
191 | async_commit(b); | |
192 | } | |
193 | ||
194 | /* | |
195 | * Call this if some urgent work is waiting for the commit to complete. | |
196 | */ | |
197 | static void schedule_commit(struct batcher *b) | |
198 | { | |
199 | bool immediate; | |
b29d4986 | 200 | |
26b924b9 | 201 | spin_lock_irq(&b->lock); |
b29d4986 JT |
202 | immediate = !list_empty(&b->work_items) || !bio_list_empty(&b->bios); |
203 | b->commit_scheduled = true; | |
26b924b9 | 204 | spin_unlock_irq(&b->lock); |
b29d4986 JT |
205 | |
206 | if (immediate) | |
207 | async_commit(b); | |
208 | } | |
209 | ||
c9d28d5d JT |
210 | /* |
211 | * There are a couple of places where we let a bio run, but want to do some | |
212 | * work before calling its endio function. We do this by temporarily | |
213 | * changing the endio fn. | |
214 | */ | |
215 | struct dm_hook_info { | |
216 | bio_end_io_t *bi_end_io; | |
c9d28d5d JT |
217 | }; |
218 | ||
219 | static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio, | |
220 | bio_end_io_t *bi_end_io, void *bi_private) | |
221 | { | |
222 | h->bi_end_io = bio->bi_end_io; | |
c9d28d5d JT |
223 | |
224 | bio->bi_end_io = bi_end_io; | |
225 | bio->bi_private = bi_private; | |
226 | } | |
227 | ||
228 | static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio) | |
229 | { | |
230 | bio->bi_end_io = h->bi_end_io; | |
c9d28d5d JT |
231 | } |
232 | ||
233 | /*----------------------------------------------------------------*/ | |
234 | ||
c6b4fcba JT |
235 | #define MIGRATION_POOL_SIZE 128 |
236 | #define COMMIT_PERIOD HZ | |
237 | #define MIGRATION_COUNT_WINDOW 10 | |
238 | ||
239 | /* | |
05473044 MS |
240 | * The block size of the device holding cache data must be |
241 | * between 32KB and 1GB. | |
c6b4fcba JT |
242 | */ |
243 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT) | |
05473044 | 244 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
c6b4fcba | 245 | |
2ee57d58 | 246 | enum cache_metadata_mode { |
c6b4fcba JT |
247 | CM_WRITE, /* metadata may be changed */ |
248 | CM_READ_ONLY, /* metadata may not be changed */ | |
028ae9f7 | 249 | CM_FAIL |
c6b4fcba JT |
250 | }; |
251 | ||
2ee57d58 JT |
252 | enum cache_io_mode { |
253 | /* | |
254 | * Data is written to cached blocks only. These blocks are marked | |
255 | * dirty. If you lose the cache device you will lose data. | |
256 | * Potential performance increase for both reads and writes. | |
257 | */ | |
258 | CM_IO_WRITEBACK, | |
259 | ||
260 | /* | |
261 | * Data is written to both cache and origin. Blocks are never | |
262 | * dirty. Potential performance benfit for reads only. | |
263 | */ | |
264 | CM_IO_WRITETHROUGH, | |
265 | ||
266 | /* | |
267 | * A degraded mode useful for various cache coherency situations | |
268 | * (eg, rolling back snapshots). Reads and writes always go to the | |
269 | * origin. If a write goes to a cached oblock, then the cache | |
270 | * block is invalidated. | |
271 | */ | |
272 | CM_IO_PASSTHROUGH | |
273 | }; | |
274 | ||
c6b4fcba | 275 | struct cache_features { |
2ee57d58 JT |
276 | enum cache_metadata_mode mode; |
277 | enum cache_io_mode io_mode; | |
629d0a8a | 278 | unsigned metadata_version; |
de7180ff | 279 | bool discard_passdown:1; |
c6b4fcba JT |
280 | }; |
281 | ||
282 | struct cache_stats { | |
283 | atomic_t read_hit; | |
284 | atomic_t read_miss; | |
285 | atomic_t write_hit; | |
286 | atomic_t write_miss; | |
287 | atomic_t demotion; | |
288 | atomic_t promotion; | |
b29d4986 | 289 | atomic_t writeback; |
c6b4fcba JT |
290 | atomic_t copies_avoided; |
291 | atomic_t cache_cell_clash; | |
292 | atomic_t commit_count; | |
293 | atomic_t discard_count; | |
294 | }; | |
295 | ||
296 | struct cache { | |
297 | struct dm_target *ti; | |
72d711c8 MS |
298 | spinlock_t lock; |
299 | ||
300 | /* | |
301 | * Fields for converting from sectors to blocks. | |
302 | */ | |
303 | int sectors_per_block_shift; | |
304 | sector_t sectors_per_block; | |
c6b4fcba | 305 | |
c9ec5d7c MS |
306 | struct dm_cache_metadata *cmd; |
307 | ||
c6b4fcba JT |
308 | /* |
309 | * Metadata is written to this device. | |
310 | */ | |
311 | struct dm_dev *metadata_dev; | |
312 | ||
313 | /* | |
314 | * The slower of the two data devices. Typically a spindle. | |
315 | */ | |
316 | struct dm_dev *origin_dev; | |
317 | ||
318 | /* | |
319 | * The faster of the two data devices. Typically an SSD. | |
320 | */ | |
321 | struct dm_dev *cache_dev; | |
322 | ||
c6b4fcba JT |
323 | /* |
324 | * Size of the origin device in _complete_ blocks and native sectors. | |
325 | */ | |
326 | dm_oblock_t origin_blocks; | |
327 | sector_t origin_sectors; | |
328 | ||
329 | /* | |
330 | * Size of the cache device in blocks. | |
331 | */ | |
332 | dm_cblock_t cache_size; | |
333 | ||
334 | /* | |
72d711c8 | 335 | * Invalidation fields. |
c6b4fcba | 336 | */ |
72d711c8 MS |
337 | spinlock_t invalidation_lock; |
338 | struct list_head invalidation_requests; | |
c6b4fcba | 339 | |
c6b4fcba | 340 | sector_t migration_threshold; |
c6b4fcba | 341 | wait_queue_head_t migration_wait; |
a59db676 JT |
342 | atomic_t nr_allocated_migrations; |
343 | ||
344 | /* | |
345 | * The number of in flight migrations that are performing | |
346 | * background io. eg, promotion, writeback. | |
347 | */ | |
348 | atomic_t nr_io_migrations; | |
c6b4fcba | 349 | |
72d711c8 MS |
350 | struct bio_list deferred_bios; |
351 | ||
b29d4986 | 352 | struct rw_semaphore quiesce_lock; |
66cb1910 | 353 | |
c6b4fcba JT |
354 | /* |
355 | * origin_blocks entries, discarded if set. | |
356 | */ | |
1bad9bc4 | 357 | dm_dblock_t discard_nr_blocks; |
c6b4fcba | 358 | unsigned long *discard_bitset; |
08b18451 | 359 | uint32_t discard_block_size; /* a power of 2 times sectors per block */ |
c9ec5d7c MS |
360 | |
361 | /* | |
362 | * Rather than reconstructing the table line for the status we just | |
363 | * save it and regurgitate. | |
364 | */ | |
365 | unsigned nr_ctr_args; | |
366 | const char **ctr_args; | |
c6b4fcba JT |
367 | |
368 | struct dm_kcopyd_client *copier; | |
b29d4986 | 369 | struct work_struct deferred_bio_worker; |
b29d4986 | 370 | struct work_struct migration_worker; |
72d711c8 | 371 | struct workqueue_struct *wq; |
c6b4fcba | 372 | struct delayed_work waker; |
b29d4986 | 373 | struct dm_bio_prison_v2 *prison; |
c6b4fcba | 374 | |
72d711c8 MS |
375 | /* |
376 | * cache_size entries, dirty if set | |
377 | */ | |
378 | unsigned long *dirty_bitset; | |
379 | atomic_t nr_dirty; | |
c6b4fcba | 380 | |
c6b4fcba | 381 | unsigned policy_nr_args; |
72d711c8 MS |
382 | struct dm_cache_policy *policy; |
383 | ||
384 | /* | |
385 | * Cache features such as write-through. | |
386 | */ | |
387 | struct cache_features features; | |
388 | ||
389 | struct cache_stats stats; | |
c6b4fcba JT |
390 | |
391 | bool need_tick_bio:1; | |
392 | bool sized:1; | |
65790ff9 | 393 | bool invalidate:1; |
c6b4fcba JT |
394 | bool commit_requested:1; |
395 | bool loaded_mappings:1; | |
396 | bool loaded_discards:1; | |
397 | ||
72d711c8 | 398 | struct rw_semaphore background_work_lock; |
65790ff9 | 399 | |
72d711c8 MS |
400 | struct batcher committer; |
401 | struct work_struct commit_ws; | |
066dbaa3 | 402 | |
dc4fa29f | 403 | struct dm_io_tracker tracker; |
b29d4986 | 404 | |
72d711c8 | 405 | mempool_t migration_pool; |
b29d4986 | 406 | |
72d711c8 | 407 | struct bio_set bs; |
c6b4fcba JT |
408 | }; |
409 | ||
410 | struct per_bio_data { | |
411 | bool tick:1; | |
412 | unsigned req_nr:2; | |
b29d4986 | 413 | struct dm_bio_prison_cell_v2 *cell; |
c6eda5e8 | 414 | struct dm_hook_info hook_info; |
066dbaa3 | 415 | sector_t len; |
c6b4fcba JT |
416 | }; |
417 | ||
418 | struct dm_cache_migration { | |
b29d4986 | 419 | struct continuation k; |
c6b4fcba JT |
420 | struct cache *cache; |
421 | ||
b29d4986 JT |
422 | struct policy_work *op; |
423 | struct bio *overwrite_bio; | |
424 | struct dm_bio_prison_cell_v2 *cell; | |
c6b4fcba | 425 | |
b29d4986 JT |
426 | dm_cblock_t invalidate_cblock; |
427 | dm_oblock_t invalidate_oblock; | |
c6b4fcba JT |
428 | }; |
429 | ||
b29d4986 JT |
430 | /*----------------------------------------------------------------*/ |
431 | ||
8e3c3827 | 432 | static bool writethrough_mode(struct cache *cache) |
b29d4986 | 433 | { |
8e3c3827 | 434 | return cache->features.io_mode == CM_IO_WRITETHROUGH; |
b29d4986 JT |
435 | } |
436 | ||
8e3c3827 | 437 | static bool writeback_mode(struct cache *cache) |
b29d4986 | 438 | { |
8e3c3827 | 439 | return cache->features.io_mode == CM_IO_WRITEBACK; |
b29d4986 JT |
440 | } |
441 | ||
8e3c3827 | 442 | static inline bool passthrough_mode(struct cache *cache) |
b29d4986 | 443 | { |
8e3c3827 | 444 | return unlikely(cache->features.io_mode == CM_IO_PASSTHROUGH); |
b29d4986 JT |
445 | } |
446 | ||
447 | /*----------------------------------------------------------------*/ | |
448 | ||
449 | static void wake_deferred_bio_worker(struct cache *cache) | |
450 | { | |
451 | queue_work(cache->wq, &cache->deferred_bio_worker); | |
452 | } | |
c6b4fcba | 453 | |
b29d4986 | 454 | static void wake_migration_worker(struct cache *cache) |
c6b4fcba | 455 | { |
8e3c3827 | 456 | if (passthrough_mode(cache)) |
b29d4986 JT |
457 | return; |
458 | ||
459 | queue_work(cache->wq, &cache->migration_worker); | |
c6b4fcba JT |
460 | } |
461 | ||
462 | /*----------------------------------------------------------------*/ | |
463 | ||
b29d4986 | 464 | static struct dm_bio_prison_cell_v2 *alloc_prison_cell(struct cache *cache) |
c6b4fcba | 465 | { |
13bd677a | 466 | return dm_bio_prison_alloc_cell_v2(cache->prison, GFP_NOIO); |
c6b4fcba JT |
467 | } |
468 | ||
b29d4986 | 469 | static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell_v2 *cell) |
c6b4fcba | 470 | { |
b29d4986 | 471 | dm_bio_prison_free_cell_v2(cache->prison, cell); |
c6b4fcba JT |
472 | } |
473 | ||
a59db676 JT |
474 | static struct dm_cache_migration *alloc_migration(struct cache *cache) |
475 | { | |
476 | struct dm_cache_migration *mg; | |
477 | ||
13bd677a | 478 | mg = mempool_alloc(&cache->migration_pool, GFP_NOIO); |
ef7afb36 MS |
479 | |
480 | memset(mg, 0, sizeof(*mg)); | |
481 | ||
482 | mg->cache = cache; | |
483 | atomic_inc(&cache->nr_allocated_migrations); | |
a59db676 JT |
484 | |
485 | return mg; | |
486 | } | |
487 | ||
488 | static void free_migration(struct dm_cache_migration *mg) | |
489 | { | |
88bf5184 | 490 | struct cache *cache = mg->cache; |
a59db676 | 491 | |
88bf5184 JT |
492 | if (atomic_dec_and_test(&cache->nr_allocated_migrations)) |
493 | wake_up(&cache->migration_wait); | |
494 | ||
6f1c819c | 495 | mempool_free(mg, &cache->migration_pool); |
a59db676 JT |
496 | } |
497 | ||
b29d4986 | 498 | /*----------------------------------------------------------------*/ |
c6b4fcba | 499 | |
b29d4986 | 500 | static inline dm_oblock_t oblock_succ(dm_oblock_t b) |
c6b4fcba | 501 | { |
b29d4986 | 502 | return to_oblock(from_oblock(b) + 1ull); |
c6b4fcba JT |
503 | } |
504 | ||
b29d4986 | 505 | static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key_v2 *key) |
c6b4fcba | 506 | { |
b29d4986 JT |
507 | key->virtual = 0; |
508 | key->dev = 0; | |
509 | key->block_begin = from_oblock(begin); | |
510 | key->block_end = from_oblock(end); | |
c6b4fcba JT |
511 | } |
512 | ||
513 | /* | |
b29d4986 JT |
514 | * We have two lock levels. Level 0, which is used to prevent WRITEs, and |
515 | * level 1 which prevents *both* READs and WRITEs. | |
c6b4fcba | 516 | */ |
b29d4986 JT |
517 | #define WRITE_LOCK_LEVEL 0 |
518 | #define READ_WRITE_LOCK_LEVEL 1 | |
519 | ||
520 | static unsigned lock_level(struct bio *bio) | |
c6b4fcba | 521 | { |
b29d4986 JT |
522 | return bio_data_dir(bio) == WRITE ? |
523 | WRITE_LOCK_LEVEL : | |
524 | READ_WRITE_LOCK_LEVEL; | |
525 | } | |
c6b4fcba | 526 | |
b29d4986 JT |
527 | /*---------------------------------------------------------------- |
528 | * Per bio data | |
529 | *--------------------------------------------------------------*/ | |
c6b4fcba | 530 | |
693b960e | 531 | static struct per_bio_data *get_per_bio_data(struct bio *bio) |
b29d4986 | 532 | { |
693b960e | 533 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); |
b29d4986 JT |
534 | BUG_ON(!pb); |
535 | return pb; | |
536 | } | |
c6b4fcba | 537 | |
693b960e | 538 | static struct per_bio_data *init_per_bio_data(struct bio *bio) |
b29d4986 | 539 | { |
693b960e | 540 | struct per_bio_data *pb = get_per_bio_data(bio); |
c6b4fcba | 541 | |
b29d4986 JT |
542 | pb->tick = false; |
543 | pb->req_nr = dm_bio_get_target_bio_nr(bio); | |
544 | pb->cell = NULL; | |
545 | pb->len = 0; | |
546 | ||
547 | return pb; | |
c6b4fcba JT |
548 | } |
549 | ||
550 | /*----------------------------------------------------------------*/ | |
551 | ||
b29d4986 | 552 | static void defer_bio(struct cache *cache, struct bio *bio) |
c6b4fcba | 553 | { |
26b924b9 | 554 | spin_lock_irq(&cache->lock); |
b29d4986 | 555 | bio_list_add(&cache->deferred_bios, bio); |
26b924b9 | 556 | spin_unlock_irq(&cache->lock); |
b29d4986 JT |
557 | |
558 | wake_deferred_bio_worker(cache); | |
559 | } | |
c6b4fcba | 560 | |
b29d4986 | 561 | static void defer_bios(struct cache *cache, struct bio_list *bios) |
c6b4fcba | 562 | { |
26b924b9 | 563 | spin_lock_irq(&cache->lock); |
b29d4986 JT |
564 | bio_list_merge(&cache->deferred_bios, bios); |
565 | bio_list_init(bios); | |
26b924b9 | 566 | spin_unlock_irq(&cache->lock); |
c6b4fcba | 567 | |
b29d4986 | 568 | wake_deferred_bio_worker(cache); |
c6b4fcba JT |
569 | } |
570 | ||
b29d4986 JT |
571 | /*----------------------------------------------------------------*/ |
572 | ||
573 | static bool bio_detain_shared(struct cache *cache, dm_oblock_t oblock, struct bio *bio) | |
7ae34e77 | 574 | { |
b29d4986 | 575 | bool r; |
b29d4986 JT |
576 | struct per_bio_data *pb; |
577 | struct dm_cell_key_v2 key; | |
7ae34e77 | 578 | dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL); |
b29d4986 | 579 | struct dm_bio_prison_cell_v2 *cell_prealloc, *cell; |
7ae34e77 | 580 | |
b29d4986 | 581 | cell_prealloc = alloc_prison_cell(cache); /* FIXME: allow wait if calling from worker */ |
b29d4986 JT |
582 | |
583 | build_key(oblock, end, &key); | |
584 | r = dm_cell_get_v2(cache->prison, &key, lock_level(bio), bio, cell_prealloc, &cell); | |
585 | if (!r) { | |
586 | /* | |
587 | * Failed to get the lock. | |
588 | */ | |
589 | free_prison_cell(cache, cell_prealloc); | |
590 | return r; | |
591 | } | |
c6b4fcba | 592 | |
b29d4986 JT |
593 | if (cell != cell_prealloc) |
594 | free_prison_cell(cache, cell_prealloc); | |
c6b4fcba | 595 | |
693b960e | 596 | pb = get_per_bio_data(bio); |
b29d4986 | 597 | pb->cell = cell; |
c6b4fcba JT |
598 | |
599 | return r; | |
600 | } | |
601 | ||
aeed1420 | 602 | /*----------------------------------------------------------------*/ |
c6b4fcba JT |
603 | |
604 | static bool is_dirty(struct cache *cache, dm_cblock_t b) | |
605 | { | |
606 | return test_bit(from_cblock(b), cache->dirty_bitset); | |
607 | } | |
608 | ||
b29d4986 | 609 | static void set_dirty(struct cache *cache, dm_cblock_t cblock) |
c6b4fcba JT |
610 | { |
611 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 612 | atomic_inc(&cache->nr_dirty); |
b29d4986 | 613 | policy_set_dirty(cache->policy, cblock); |
c6b4fcba JT |
614 | } |
615 | } | |
616 | ||
b29d4986 JT |
617 | /* |
618 | * These two are called when setting after migrations to force the policy | |
619 | * and dirty bitset to be in sync. | |
620 | */ | |
621 | static void force_set_dirty(struct cache *cache, dm_cblock_t cblock) | |
622 | { | |
623 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) | |
624 | atomic_inc(&cache->nr_dirty); | |
625 | policy_set_dirty(cache->policy, cblock); | |
626 | } | |
627 | ||
628 | static void force_clear_dirty(struct cache *cache, dm_cblock_t cblock) | |
c6b4fcba JT |
629 | { |
630 | if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 631 | if (atomic_dec_return(&cache->nr_dirty) == 0) |
c6b4fcba JT |
632 | dm_table_event(cache->ti->table); |
633 | } | |
b29d4986 JT |
634 | |
635 | policy_clear_dirty(cache->policy, cblock); | |
c6b4fcba JT |
636 | } |
637 | ||
638 | /*----------------------------------------------------------------*/ | |
aeed1420 | 639 | |
c6b4fcba JT |
640 | static bool block_size_is_power_of_two(struct cache *cache) |
641 | { | |
642 | return cache->sectors_per_block_shift >= 0; | |
643 | } | |
644 | ||
414dd67d JT |
645 | static dm_block_t block_div(dm_block_t b, uint32_t n) |
646 | { | |
647 | do_div(b, n); | |
648 | ||
649 | return b; | |
650 | } | |
651 | ||
7ae34e77 | 652 | static dm_block_t oblocks_per_dblock(struct cache *cache) |
1bad9bc4 | 653 | { |
7ae34e77 | 654 | dm_block_t oblocks = cache->discard_block_size; |
1bad9bc4 | 655 | |
7ae34e77 JT |
656 | if (block_size_is_power_of_two(cache)) |
657 | oblocks >>= cache->sectors_per_block_shift; | |
1bad9bc4 | 658 | else |
7ae34e77 | 659 | oblocks = block_div(oblocks, cache->sectors_per_block); |
1bad9bc4 | 660 | |
7ae34e77 JT |
661 | return oblocks; |
662 | } | |
663 | ||
664 | static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock) | |
665 | { | |
666 | return to_dblock(block_div(from_oblock(oblock), | |
667 | oblocks_per_dblock(cache))); | |
668 | } | |
1bad9bc4 | 669 | |
1bad9bc4 | 670 | static void set_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba | 671 | { |
7ae34e77 | 672 | BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
673 | atomic_inc(&cache->stats.discard_count); |
674 | ||
26b924b9 | 675 | spin_lock_irq(&cache->lock); |
1bad9bc4 | 676 | set_bit(from_dblock(b), cache->discard_bitset); |
26b924b9 | 677 | spin_unlock_irq(&cache->lock); |
c6b4fcba JT |
678 | } |
679 | ||
1bad9bc4 | 680 | static void clear_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba | 681 | { |
26b924b9 | 682 | spin_lock_irq(&cache->lock); |
1bad9bc4 | 683 | clear_bit(from_dblock(b), cache->discard_bitset); |
26b924b9 | 684 | spin_unlock_irq(&cache->lock); |
c6b4fcba JT |
685 | } |
686 | ||
1bad9bc4 | 687 | static bool is_discarded(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
688 | { |
689 | int r; | |
26b924b9 | 690 | spin_lock_irq(&cache->lock); |
1bad9bc4 | 691 | r = test_bit(from_dblock(b), cache->discard_bitset); |
26b924b9 | 692 | spin_unlock_irq(&cache->lock); |
c6b4fcba JT |
693 | |
694 | return r; | |
695 | } | |
696 | ||
697 | static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b) | |
698 | { | |
699 | int r; | |
26b924b9 | 700 | spin_lock_irq(&cache->lock); |
1bad9bc4 JT |
701 | r = test_bit(from_dblock(oblock_to_dblock(cache, b)), |
702 | cache->discard_bitset); | |
26b924b9 | 703 | spin_unlock_irq(&cache->lock); |
c6b4fcba JT |
704 | |
705 | return r; | |
706 | } | |
707 | ||
b29d4986 JT |
708 | /*---------------------------------------------------------------- |
709 | * Remapping | |
710 | *--------------------------------------------------------------*/ | |
711 | static void remap_to_origin(struct cache *cache, struct bio *bio) | |
c6b4fcba | 712 | { |
74d46992 | 713 | bio_set_dev(bio, cache->origin_dev->bdev); |
c6b4fcba JT |
714 | } |
715 | ||
716 | static void remap_to_cache(struct cache *cache, struct bio *bio, | |
717 | dm_cblock_t cblock) | |
718 | { | |
4f024f37 | 719 | sector_t bi_sector = bio->bi_iter.bi_sector; |
e0d849fa | 720 | sector_t block = from_cblock(cblock); |
c6b4fcba | 721 | |
74d46992 | 722 | bio_set_dev(bio, cache->cache_dev->bdev); |
c6b4fcba | 723 | if (!block_size_is_power_of_two(cache)) |
4f024f37 | 724 | bio->bi_iter.bi_sector = |
e0d849fa | 725 | (block * cache->sectors_per_block) + |
4f024f37 | 726 | sector_div(bi_sector, cache->sectors_per_block); |
c6b4fcba | 727 | else |
4f024f37 | 728 | bio->bi_iter.bi_sector = |
e0d849fa | 729 | (block << cache->sectors_per_block_shift) | |
4f024f37 | 730 | (bi_sector & (cache->sectors_per_block - 1)); |
c6b4fcba JT |
731 | } |
732 | ||
733 | static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio) | |
734 | { | |
693b960e | 735 | struct per_bio_data *pb; |
c6b4fcba | 736 | |
26b924b9 | 737 | spin_lock_irq(&cache->lock); |
f73f44eb | 738 | if (cache->need_tick_bio && !op_is_flush(bio->bi_opf) && |
e6047149 | 739 | bio_op(bio) != REQ_OP_DISCARD) { |
693b960e | 740 | pb = get_per_bio_data(bio); |
c6b4fcba JT |
741 | pb->tick = true; |
742 | cache->need_tick_bio = false; | |
743 | } | |
26b924b9 | 744 | spin_unlock_irq(&cache->lock); |
c6b4fcba JT |
745 | } |
746 | ||
2df3bae9 MS |
747 | static void __remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, |
748 | dm_oblock_t oblock, bool bio_has_pbd) | |
c6b4fcba | 749 | { |
2df3bae9 MS |
750 | if (bio_has_pbd) |
751 | check_if_tick_bio_needed(cache, bio); | |
c6b4fcba JT |
752 | remap_to_origin(cache, bio); |
753 | if (bio_data_dir(bio) == WRITE) | |
1bad9bc4 | 754 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
755 | } |
756 | ||
2df3bae9 MS |
757 | static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, |
758 | dm_oblock_t oblock) | |
759 | { | |
760 | // FIXME: check_if_tick_bio_needed() is called way too much through this interface | |
761 | __remap_to_origin_clear_discard(cache, bio, oblock, true); | |
762 | } | |
763 | ||
c6b4fcba JT |
764 | static void remap_to_cache_dirty(struct cache *cache, struct bio *bio, |
765 | dm_oblock_t oblock, dm_cblock_t cblock) | |
766 | { | |
f8e5f01a | 767 | check_if_tick_bio_needed(cache, bio); |
c6b4fcba JT |
768 | remap_to_cache(cache, bio, cblock); |
769 | if (bio_data_dir(bio) == WRITE) { | |
b29d4986 | 770 | set_dirty(cache, cblock); |
1bad9bc4 | 771 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
772 | } |
773 | } | |
774 | ||
775 | static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio) | |
776 | { | |
4f024f37 | 777 | sector_t block_nr = bio->bi_iter.bi_sector; |
c6b4fcba JT |
778 | |
779 | if (!block_size_is_power_of_two(cache)) | |
780 | (void) sector_div(block_nr, cache->sectors_per_block); | |
781 | else | |
782 | block_nr >>= cache->sectors_per_block_shift; | |
783 | ||
784 | return to_oblock(block_nr); | |
785 | } | |
786 | ||
066dbaa3 JT |
787 | static bool accountable_bio(struct cache *cache, struct bio *bio) |
788 | { | |
701e03e4 | 789 | return bio_op(bio) != REQ_OP_DISCARD; |
066dbaa3 JT |
790 | } |
791 | ||
792 | static void accounted_begin(struct cache *cache, struct bio *bio) | |
793 | { | |
693b960e | 794 | struct per_bio_data *pb; |
066dbaa3 JT |
795 | |
796 | if (accountable_bio(cache, bio)) { | |
693b960e | 797 | pb = get_per_bio_data(bio); |
066dbaa3 | 798 | pb->len = bio_sectors(bio); |
dc4fa29f | 799 | dm_iot_io_begin(&cache->tracker, pb->len); |
066dbaa3 JT |
800 | } |
801 | } | |
802 | ||
803 | static void accounted_complete(struct cache *cache, struct bio *bio) | |
804 | { | |
693b960e | 805 | struct per_bio_data *pb = get_per_bio_data(bio); |
066dbaa3 | 806 | |
dc4fa29f | 807 | dm_iot_io_end(&cache->tracker, pb->len); |
066dbaa3 JT |
808 | } |
809 | ||
810 | static void accounted_request(struct cache *cache, struct bio *bio) | |
811 | { | |
812 | accounted_begin(cache, bio); | |
ed00aabd | 813 | submit_bio_noacct(bio); |
066dbaa3 JT |
814 | } |
815 | ||
b29d4986 | 816 | static void issue_op(struct bio *bio, void *context) |
8c081b52 | 817 | { |
b29d4986 JT |
818 | struct cache *cache = context; |
819 | accounted_request(cache, bio); | |
8c081b52 JT |
820 | } |
821 | ||
e2e74d61 JT |
822 | /* |
823 | * When running in writethrough mode we need to send writes to clean blocks | |
2df3bae9 | 824 | * to both the cache and origin devices. Clone the bio and send them in parallel. |
e2e74d61 | 825 | */ |
2df3bae9 MS |
826 | static void remap_to_origin_and_cache(struct cache *cache, struct bio *bio, |
827 | dm_oblock_t oblock, dm_cblock_t cblock) | |
e2e74d61 | 828 | { |
6f1c819c | 829 | struct bio *origin_bio = bio_clone_fast(bio, GFP_NOIO, &cache->bs); |
2df3bae9 MS |
830 | |
831 | BUG_ON(!origin_bio); | |
e2e74d61 | 832 | |
2df3bae9 MS |
833 | bio_chain(origin_bio, bio); |
834 | /* | |
835 | * Passing false to __remap_to_origin_clear_discard() skips | |
836 | * all code that might use per_bio_data (since clone doesn't have it) | |
837 | */ | |
838 | __remap_to_origin_clear_discard(cache, origin_bio, oblock, false); | |
839 | submit_bio(origin_bio); | |
e2e74d61 | 840 | |
2df3bae9 | 841 | remap_to_cache(cache, bio, cblock); |
e2e74d61 JT |
842 | } |
843 | ||
028ae9f7 JT |
844 | /*---------------------------------------------------------------- |
845 | * Failure modes | |
846 | *--------------------------------------------------------------*/ | |
847 | static enum cache_metadata_mode get_cache_mode(struct cache *cache) | |
848 | { | |
849 | return cache->features.mode; | |
850 | } | |
851 | ||
b61d9509 MS |
852 | static const char *cache_device_name(struct cache *cache) |
853 | { | |
d4a512ed | 854 | return dm_table_device_name(cache->ti->table); |
b61d9509 MS |
855 | } |
856 | ||
028ae9f7 JT |
857 | static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode) |
858 | { | |
859 | const char *descs[] = { | |
860 | "write", | |
861 | "read-only", | |
862 | "fail" | |
863 | }; | |
864 | ||
865 | dm_table_event(cache->ti->table); | |
b61d9509 MS |
866 | DMINFO("%s: switching cache to %s mode", |
867 | cache_device_name(cache), descs[(int)mode]); | |
028ae9f7 JT |
868 | } |
869 | ||
870 | static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode) | |
871 | { | |
d14fcf3d | 872 | bool needs_check; |
028ae9f7 JT |
873 | enum cache_metadata_mode old_mode = get_cache_mode(cache); |
874 | ||
d14fcf3d | 875 | if (dm_cache_metadata_needs_check(cache->cmd, &needs_check)) { |
23cab26d MS |
876 | DMERR("%s: unable to read needs_check flag, setting failure mode.", |
877 | cache_device_name(cache)); | |
d14fcf3d JT |
878 | new_mode = CM_FAIL; |
879 | } | |
880 | ||
028ae9f7 | 881 | if (new_mode == CM_WRITE && needs_check) { |
b61d9509 MS |
882 | DMERR("%s: unable to switch cache to write mode until repaired.", |
883 | cache_device_name(cache)); | |
028ae9f7 JT |
884 | if (old_mode != new_mode) |
885 | new_mode = old_mode; | |
886 | else | |
887 | new_mode = CM_READ_ONLY; | |
888 | } | |
889 | ||
890 | /* Never move out of fail mode */ | |
891 | if (old_mode == CM_FAIL) | |
892 | new_mode = CM_FAIL; | |
893 | ||
894 | switch (new_mode) { | |
895 | case CM_FAIL: | |
896 | case CM_READ_ONLY: | |
897 | dm_cache_metadata_set_read_only(cache->cmd); | |
898 | break; | |
899 | ||
900 | case CM_WRITE: | |
901 | dm_cache_metadata_set_read_write(cache->cmd); | |
902 | break; | |
903 | } | |
904 | ||
905 | cache->features.mode = new_mode; | |
906 | ||
907 | if (new_mode != old_mode) | |
908 | notify_mode_switch(cache, new_mode); | |
909 | } | |
910 | ||
911 | static void abort_transaction(struct cache *cache) | |
912 | { | |
b61d9509 MS |
913 | const char *dev_name = cache_device_name(cache); |
914 | ||
028ae9f7 JT |
915 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
916 | return; | |
917 | ||
918 | if (dm_cache_metadata_set_needs_check(cache->cmd)) { | |
b61d9509 | 919 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); |
028ae9f7 JT |
920 | set_cache_mode(cache, CM_FAIL); |
921 | } | |
922 | ||
b61d9509 | 923 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); |
028ae9f7 | 924 | if (dm_cache_metadata_abort(cache->cmd)) { |
b61d9509 | 925 | DMERR("%s: failed to abort metadata transaction", dev_name); |
028ae9f7 JT |
926 | set_cache_mode(cache, CM_FAIL); |
927 | } | |
928 | } | |
929 | ||
930 | static void metadata_operation_failed(struct cache *cache, const char *op, int r) | |
931 | { | |
b61d9509 MS |
932 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
933 | cache_device_name(cache), op, r); | |
028ae9f7 JT |
934 | abort_transaction(cache); |
935 | set_cache_mode(cache, CM_READ_ONLY); | |
936 | } | |
937 | ||
b29d4986 JT |
938 | /*----------------------------------------------------------------*/ |
939 | ||
940 | static void load_stats(struct cache *cache) | |
941 | { | |
942 | struct dm_cache_statistics stats; | |
943 | ||
944 | dm_cache_metadata_get_stats(cache->cmd, &stats); | |
945 | atomic_set(&cache->stats.read_hit, stats.read_hits); | |
946 | atomic_set(&cache->stats.read_miss, stats.read_misses); | |
947 | atomic_set(&cache->stats.write_hit, stats.write_hits); | |
948 | atomic_set(&cache->stats.write_miss, stats.write_misses); | |
949 | } | |
950 | ||
951 | static void save_stats(struct cache *cache) | |
952 | { | |
953 | struct dm_cache_statistics stats; | |
954 | ||
955 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
956 | return; | |
957 | ||
958 | stats.read_hits = atomic_read(&cache->stats.read_hit); | |
959 | stats.read_misses = atomic_read(&cache->stats.read_miss); | |
960 | stats.write_hits = atomic_read(&cache->stats.write_hit); | |
961 | stats.write_misses = atomic_read(&cache->stats.write_miss); | |
962 | ||
963 | dm_cache_metadata_set_stats(cache->cmd, &stats); | |
964 | } | |
965 | ||
966 | static void update_stats(struct cache_stats *stats, enum policy_operation op) | |
967 | { | |
968 | switch (op) { | |
969 | case POLICY_PROMOTE: | |
970 | atomic_inc(&stats->promotion); | |
971 | break; | |
972 | ||
973 | case POLICY_DEMOTE: | |
974 | atomic_inc(&stats->demotion); | |
975 | break; | |
976 | ||
977 | case POLICY_WRITEBACK: | |
978 | atomic_inc(&stats->writeback); | |
979 | break; | |
980 | } | |
981 | } | |
982 | ||
c6b4fcba JT |
983 | /*---------------------------------------------------------------- |
984 | * Migration processing | |
985 | * | |
986 | * Migration covers moving data from the origin device to the cache, or | |
987 | * vice versa. | |
988 | *--------------------------------------------------------------*/ | |
b29d4986 | 989 | |
a59db676 | 990 | static void inc_io_migrations(struct cache *cache) |
c6b4fcba | 991 | { |
a59db676 | 992 | atomic_inc(&cache->nr_io_migrations); |
c6b4fcba JT |
993 | } |
994 | ||
a59db676 | 995 | static void dec_io_migrations(struct cache *cache) |
c6b4fcba | 996 | { |
a59db676 | 997 | atomic_dec(&cache->nr_io_migrations); |
c6b4fcba JT |
998 | } |
999 | ||
651f5fa2 JT |
1000 | static bool discard_or_flush(struct bio *bio) |
1001 | { | |
f73f44eb | 1002 | return bio_op(bio) == REQ_OP_DISCARD || op_is_flush(bio->bi_opf); |
651f5fa2 JT |
1003 | } |
1004 | ||
b29d4986 JT |
1005 | static void calc_discard_block_range(struct cache *cache, struct bio *bio, |
1006 | dm_dblock_t *b, dm_dblock_t *e) | |
c6b4fcba | 1007 | { |
b29d4986 JT |
1008 | sector_t sb = bio->bi_iter.bi_sector; |
1009 | sector_t se = bio_end_sector(bio); | |
651f5fa2 | 1010 | |
b29d4986 | 1011 | *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size)); |
c6b4fcba | 1012 | |
b29d4986 JT |
1013 | if (se - sb < cache->discard_block_size) |
1014 | *e = *b; | |
1015 | else | |
1016 | *e = to_dblock(block_div(se, cache->discard_block_size)); | |
c6b4fcba JT |
1017 | } |
1018 | ||
b29d4986 | 1019 | /*----------------------------------------------------------------*/ |
651f5fa2 | 1020 | |
b29d4986 | 1021 | static void prevent_background_work(struct cache *cache) |
651f5fa2 | 1022 | { |
b29d4986 JT |
1023 | lockdep_off(); |
1024 | down_write(&cache->background_work_lock); | |
1025 | lockdep_on(); | |
651f5fa2 JT |
1026 | } |
1027 | ||
b29d4986 | 1028 | static void allow_background_work(struct cache *cache) |
c6b4fcba | 1029 | { |
b29d4986 JT |
1030 | lockdep_off(); |
1031 | up_write(&cache->background_work_lock); | |
1032 | lockdep_on(); | |
c6b4fcba JT |
1033 | } |
1034 | ||
b29d4986 | 1035 | static bool background_work_begin(struct cache *cache) |
c6b4fcba | 1036 | { |
b29d4986 | 1037 | bool r; |
c6b4fcba | 1038 | |
b29d4986 JT |
1039 | lockdep_off(); |
1040 | r = down_read_trylock(&cache->background_work_lock); | |
1041 | lockdep_on(); | |
c6b4fcba | 1042 | |
b29d4986 | 1043 | return r; |
c6b4fcba JT |
1044 | } |
1045 | ||
b29d4986 | 1046 | static void background_work_end(struct cache *cache) |
c6b4fcba | 1047 | { |
b29d4986 JT |
1048 | lockdep_off(); |
1049 | up_read(&cache->background_work_lock); | |
1050 | lockdep_on(); | |
1051 | } | |
c6b4fcba | 1052 | |
b29d4986 | 1053 | /*----------------------------------------------------------------*/ |
c6b4fcba | 1054 | |
d1260e2a JT |
1055 | static bool bio_writes_complete_block(struct cache *cache, struct bio *bio) |
1056 | { | |
1057 | return (bio_data_dir(bio) == WRITE) && | |
1058 | (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT)); | |
1059 | } | |
1060 | ||
1061 | static bool optimisable_bio(struct cache *cache, struct bio *bio, dm_oblock_t block) | |
1062 | { | |
8e3c3827 | 1063 | return writeback_mode(cache) && |
d1260e2a JT |
1064 | (is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio)); |
1065 | } | |
1066 | ||
b29d4986 JT |
1067 | static void quiesce(struct dm_cache_migration *mg, |
1068 | void (*continuation)(struct work_struct *)) | |
1069 | { | |
1070 | init_continuation(&mg->k, continuation); | |
1071 | dm_cell_quiesce_v2(mg->cache->prison, mg->cell, &mg->k.ws); | |
c6b4fcba JT |
1072 | } |
1073 | ||
b29d4986 | 1074 | static struct dm_cache_migration *ws_to_mg(struct work_struct *ws) |
c6b4fcba | 1075 | { |
b29d4986 JT |
1076 | struct continuation *k = container_of(ws, struct continuation, ws); |
1077 | return container_of(k, struct dm_cache_migration, k); | |
c6b4fcba JT |
1078 | } |
1079 | ||
1080 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
1081 | { | |
b29d4986 | 1082 | struct dm_cache_migration *mg = container_of(context, struct dm_cache_migration, k); |
c6b4fcba JT |
1083 | |
1084 | if (read_err || write_err) | |
4e4cbee9 | 1085 | mg->k.input = BLK_STS_IOERR; |
c6b4fcba | 1086 | |
b29d4986 | 1087 | queue_continuation(mg->cache->wq, &mg->k); |
c6b4fcba JT |
1088 | } |
1089 | ||
7209049d | 1090 | static void copy(struct dm_cache_migration *mg, bool promote) |
c6b4fcba | 1091 | { |
c6b4fcba JT |
1092 | struct dm_io_region o_region, c_region; |
1093 | struct cache *cache = mg->cache; | |
1094 | ||
1095 | o_region.bdev = cache->origin_dev->bdev; | |
b29d4986 | 1096 | o_region.sector = from_oblock(mg->op->oblock) * cache->sectors_per_block; |
c6b4fcba JT |
1097 | o_region.count = cache->sectors_per_block; |
1098 | ||
1099 | c_region.bdev = cache->cache_dev->bdev; | |
b29d4986 | 1100 | c_region.sector = from_cblock(mg->op->cblock) * cache->sectors_per_block; |
c6b4fcba JT |
1101 | c_region.count = cache->sectors_per_block; |
1102 | ||
b29d4986 | 1103 | if (promote) |
7209049d | 1104 | dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, &mg->k); |
b29d4986 | 1105 | else |
7209049d | 1106 | dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, &mg->k); |
b29d4986 JT |
1107 | } |
1108 | ||
1109 | static void bio_drop_shared_lock(struct cache *cache, struct bio *bio) | |
1110 | { | |
693b960e | 1111 | struct per_bio_data *pb = get_per_bio_data(bio); |
b29d4986 JT |
1112 | |
1113 | if (pb->cell && dm_cell_put_v2(cache->prison, pb->cell)) | |
1114 | free_prison_cell(cache, pb->cell); | |
1115 | pb->cell = NULL; | |
c6b4fcba JT |
1116 | } |
1117 | ||
4246a0b6 | 1118 | static void overwrite_endio(struct bio *bio) |
c9d28d5d JT |
1119 | { |
1120 | struct dm_cache_migration *mg = bio->bi_private; | |
1121 | struct cache *cache = mg->cache; | |
693b960e | 1122 | struct per_bio_data *pb = get_per_bio_data(bio); |
c9d28d5d | 1123 | |
80ae49aa MS |
1124 | dm_unhook_bio(&pb->hook_info, bio); |
1125 | ||
4e4cbee9 CH |
1126 | if (bio->bi_status) |
1127 | mg->k.input = bio->bi_status; | |
80ae49aa | 1128 | |
693b960e | 1129 | queue_continuation(cache->wq, &mg->k); |
c9d28d5d JT |
1130 | } |
1131 | ||
b29d4986 JT |
1132 | static void overwrite(struct dm_cache_migration *mg, |
1133 | void (*continuation)(struct work_struct *)) | |
c9d28d5d | 1134 | { |
b29d4986 | 1135 | struct bio *bio = mg->overwrite_bio; |
693b960e | 1136 | struct per_bio_data *pb = get_per_bio_data(bio); |
c9d28d5d JT |
1137 | |
1138 | dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg); | |
8c081b52 JT |
1139 | |
1140 | /* | |
b29d4986 JT |
1141 | * The overwrite bio is part of the copy operation, as such it does |
1142 | * not set/clear discard or dirty flags. | |
8c081b52 | 1143 | */ |
b29d4986 JT |
1144 | if (mg->op->op == POLICY_PROMOTE) |
1145 | remap_to_cache(mg->cache, bio, mg->op->cblock); | |
1146 | else | |
1147 | remap_to_origin(mg->cache, bio); | |
1148 | ||
1149 | init_continuation(&mg->k, continuation); | |
066dbaa3 | 1150 | accounted_request(mg->cache, bio); |
c9d28d5d JT |
1151 | } |
1152 | ||
b29d4986 JT |
1153 | /* |
1154 | * Migration steps: | |
1155 | * | |
1156 | * 1) exclusive lock preventing WRITEs | |
1157 | * 2) quiesce | |
1158 | * 3) copy or issue overwrite bio | |
1159 | * 4) upgrade to exclusive lock preventing READs and WRITEs | |
1160 | * 5) quiesce | |
1161 | * 6) update metadata and commit | |
1162 | * 7) unlock | |
1163 | */ | |
1164 | static void mg_complete(struct dm_cache_migration *mg, bool success) | |
c9d28d5d | 1165 | { |
b29d4986 JT |
1166 | struct bio_list bios; |
1167 | struct cache *cache = mg->cache; | |
1168 | struct policy_work *op = mg->op; | |
1169 | dm_cblock_t cblock = op->cblock; | |
1170 | ||
1171 | if (success) | |
1172 | update_stats(&cache->stats, op->op); | |
1173 | ||
1174 | switch (op->op) { | |
1175 | case POLICY_PROMOTE: | |
1176 | clear_discard(cache, oblock_to_dblock(cache, op->oblock)); | |
1177 | policy_complete_background_work(cache->policy, op, success); | |
1178 | ||
1179 | if (mg->overwrite_bio) { | |
1180 | if (success) | |
1181 | force_set_dirty(cache, cblock); | |
4e4cbee9 CH |
1182 | else if (mg->k.input) |
1183 | mg->overwrite_bio->bi_status = mg->k.input; | |
b29d4986 | 1184 | else |
4e4cbee9 | 1185 | mg->overwrite_bio->bi_status = BLK_STS_IOERR; |
b29d4986 JT |
1186 | bio_endio(mg->overwrite_bio); |
1187 | } else { | |
1188 | if (success) | |
1189 | force_clear_dirty(cache, cblock); | |
1190 | dec_io_migrations(cache); | |
1191 | } | |
1192 | break; | |
1193 | ||
1194 | case POLICY_DEMOTE: | |
1195 | /* | |
1196 | * We clear dirty here to update the nr_dirty counter. | |
1197 | */ | |
1198 | if (success) | |
1199 | force_clear_dirty(cache, cblock); | |
1200 | policy_complete_background_work(cache->policy, op, success); | |
1201 | dec_io_migrations(cache); | |
1202 | break; | |
1203 | ||
1204 | case POLICY_WRITEBACK: | |
1205 | if (success) | |
1206 | force_clear_dirty(cache, cblock); | |
1207 | policy_complete_background_work(cache->policy, op, success); | |
1208 | dec_io_migrations(cache); | |
1209 | break; | |
1210 | } | |
1211 | ||
1212 | bio_list_init(&bios); | |
1213 | if (mg->cell) { | |
1214 | if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios)) | |
1215 | free_prison_cell(cache, mg->cell); | |
1216 | } | |
1217 | ||
1218 | free_migration(mg); | |
1219 | defer_bios(cache, &bios); | |
1220 | wake_migration_worker(cache); | |
1221 | ||
1222 | background_work_end(cache); | |
c9d28d5d JT |
1223 | } |
1224 | ||
b29d4986 | 1225 | static void mg_success(struct work_struct *ws) |
c6b4fcba | 1226 | { |
b29d4986 JT |
1227 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1228 | mg_complete(mg, mg->k.input == 0); | |
c6b4fcba JT |
1229 | } |
1230 | ||
b29d4986 | 1231 | static void mg_update_metadata(struct work_struct *ws) |
7ae34e77 | 1232 | { |
b29d4986 JT |
1233 | int r; |
1234 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
c6b4fcba | 1235 | struct cache *cache = mg->cache; |
b29d4986 | 1236 | struct policy_work *op = mg->op; |
c6b4fcba | 1237 | |
b29d4986 JT |
1238 | switch (op->op) { |
1239 | case POLICY_PROMOTE: | |
1240 | r = dm_cache_insert_mapping(cache->cmd, op->cblock, op->oblock); | |
1241 | if (r) { | |
1242 | DMERR_LIMIT("%s: migration failed; couldn't insert mapping", | |
1243 | cache_device_name(cache)); | |
1244 | metadata_operation_failed(cache, "dm_cache_insert_mapping", r); | |
7ae34e77 | 1245 | |
b29d4986 JT |
1246 | mg_complete(mg, false); |
1247 | return; | |
1248 | } | |
1249 | mg_complete(mg, true); | |
1250 | break; | |
c9d28d5d | 1251 | |
b29d4986 JT |
1252 | case POLICY_DEMOTE: |
1253 | r = dm_cache_remove_mapping(cache->cmd, op->cblock); | |
1254 | if (r) { | |
1255 | DMERR_LIMIT("%s: migration failed; couldn't update on disk metadata", | |
1256 | cache_device_name(cache)); | |
1257 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); | |
c6b4fcba | 1258 | |
b29d4986 | 1259 | mg_complete(mg, false); |
c9d28d5d JT |
1260 | return; |
1261 | } | |
c9d28d5d | 1262 | |
b29d4986 JT |
1263 | /* |
1264 | * It would be nice if we only had to commit when a REQ_FLUSH | |
1265 | * comes through. But there's one scenario that we have to | |
1266 | * look out for: | |
1267 | * | |
1268 | * - vblock x in a cache block | |
1269 | * - domotion occurs | |
1270 | * - cache block gets reallocated and over written | |
1271 | * - crash | |
1272 | * | |
1273 | * When we recover, because there was no commit the cache will | |
1274 | * rollback to having the data for vblock x in the cache block. | |
1275 | * But the cache block has since been overwritten, so it'll end | |
1276 | * up pointing to data that was never in 'x' during the history | |
1277 | * of the device. | |
1278 | * | |
1279 | * To avoid this issue we require a commit as part of the | |
1280 | * demotion operation. | |
1281 | */ | |
1282 | init_continuation(&mg->k, mg_success); | |
1283 | continue_after_commit(&cache->committer, &mg->k); | |
1284 | schedule_commit(&cache->committer); | |
1285 | break; | |
1286 | ||
1287 | case POLICY_WRITEBACK: | |
1288 | mg_complete(mg, true); | |
1289 | break; | |
7ae34e77 | 1290 | } |
c6b4fcba JT |
1291 | } |
1292 | ||
b29d4986 | 1293 | static void mg_update_metadata_after_copy(struct work_struct *ws) |
c6b4fcba | 1294 | { |
b29d4986 JT |
1295 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1296 | ||
1297 | /* | |
1298 | * Did the copy succeed? | |
1299 | */ | |
1300 | if (mg->k.input) | |
1301 | mg_complete(mg, false); | |
c6b4fcba | 1302 | else |
b29d4986 | 1303 | mg_update_metadata(ws); |
c6b4fcba JT |
1304 | } |
1305 | ||
b29d4986 | 1306 | static void mg_upgrade_lock(struct work_struct *ws) |
c6b4fcba | 1307 | { |
b29d4986 JT |
1308 | int r; |
1309 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
c6b4fcba | 1310 | |
b29d4986 JT |
1311 | /* |
1312 | * Did the copy succeed? | |
1313 | */ | |
1314 | if (mg->k.input) | |
1315 | mg_complete(mg, false); | |
c6b4fcba | 1316 | |
c9d28d5d | 1317 | else { |
b29d4986 JT |
1318 | /* |
1319 | * Now we want the lock to prevent both reads and writes. | |
1320 | */ | |
1321 | r = dm_cell_lock_promote_v2(mg->cache->prison, mg->cell, | |
1322 | READ_WRITE_LOCK_LEVEL); | |
1323 | if (r < 0) | |
1324 | mg_complete(mg, false); | |
c6b4fcba | 1325 | |
b29d4986 JT |
1326 | else if (r) |
1327 | quiesce(mg, mg_update_metadata); | |
c6b4fcba | 1328 | |
b29d4986 JT |
1329 | else |
1330 | mg_update_metadata(ws); | |
c9d28d5d | 1331 | } |
c6b4fcba JT |
1332 | } |
1333 | ||
d1260e2a JT |
1334 | static void mg_full_copy(struct work_struct *ws) |
1335 | { | |
1336 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
1337 | struct cache *cache = mg->cache; | |
1338 | struct policy_work *op = mg->op; | |
1339 | bool is_policy_promote = (op->op == POLICY_PROMOTE); | |
1340 | ||
1341 | if ((!is_policy_promote && !is_dirty(cache, op->cblock)) || | |
1342 | is_discarded_oblock(cache, op->oblock)) { | |
1343 | mg_upgrade_lock(ws); | |
1344 | return; | |
1345 | } | |
1346 | ||
1347 | init_continuation(&mg->k, mg_upgrade_lock); | |
7209049d | 1348 | copy(mg, is_policy_promote); |
d1260e2a JT |
1349 | } |
1350 | ||
b29d4986 | 1351 | static void mg_copy(struct work_struct *ws) |
c6b4fcba | 1352 | { |
b29d4986 | 1353 | struct dm_cache_migration *mg = ws_to_mg(ws); |
c6b4fcba | 1354 | |
b29d4986 | 1355 | if (mg->overwrite_bio) { |
d1260e2a JT |
1356 | /* |
1357 | * No exclusive lock was held when we last checked if the bio | |
1358 | * was optimisable. So we have to check again in case things | |
1359 | * have changed (eg, the block may no longer be discarded). | |
1360 | */ | |
1361 | if (!optimisable_bio(mg->cache, mg->overwrite_bio, mg->op->oblock)) { | |
1362 | /* | |
1363 | * Fallback to a real full copy after doing some tidying up. | |
1364 | */ | |
1365 | bool rb = bio_detain_shared(mg->cache, mg->op->oblock, mg->overwrite_bio); | |
1366 | BUG_ON(rb); /* An exclussive lock must _not_ be held for this block */ | |
1367 | mg->overwrite_bio = NULL; | |
1368 | inc_io_migrations(mg->cache); | |
1369 | mg_full_copy(ws); | |
1370 | return; | |
1371 | } | |
1372 | ||
b29d4986 JT |
1373 | /* |
1374 | * It's safe to do this here, even though it's new data | |
1375 | * because all IO has been locked out of the block. | |
1376 | * | |
1377 | * mg_lock_writes() already took READ_WRITE_LOCK_LEVEL | |
1378 | * so _not_ using mg_upgrade_lock() as continutation. | |
1379 | */ | |
1380 | overwrite(mg, mg_update_metadata_after_copy); | |
c6b4fcba | 1381 | |
d1260e2a JT |
1382 | } else |
1383 | mg_full_copy(ws); | |
c6b4fcba JT |
1384 | } |
1385 | ||
b29d4986 | 1386 | static int mg_lock_writes(struct dm_cache_migration *mg) |
c6b4fcba | 1387 | { |
b29d4986 JT |
1388 | int r; |
1389 | struct dm_cell_key_v2 key; | |
c6b4fcba | 1390 | struct cache *cache = mg->cache; |
b29d4986 | 1391 | struct dm_bio_prison_cell_v2 *prealloc; |
c6b4fcba | 1392 | |
b29d4986 | 1393 | prealloc = alloc_prison_cell(cache); |
c6b4fcba | 1394 | |
b29d4986 JT |
1395 | /* |
1396 | * Prevent writes to the block, but allow reads to continue. | |
1397 | * Unless we're using an overwrite bio, in which case we lock | |
1398 | * everything. | |
1399 | */ | |
1400 | build_key(mg->op->oblock, oblock_succ(mg->op->oblock), &key); | |
1401 | r = dm_cell_lock_v2(cache->prison, &key, | |
1402 | mg->overwrite_bio ? READ_WRITE_LOCK_LEVEL : WRITE_LOCK_LEVEL, | |
1403 | prealloc, &mg->cell); | |
1404 | if (r < 0) { | |
1405 | free_prison_cell(cache, prealloc); | |
1406 | mg_complete(mg, false); | |
1407 | return r; | |
1408 | } | |
c6b4fcba | 1409 | |
b29d4986 JT |
1410 | if (mg->cell != prealloc) |
1411 | free_prison_cell(cache, prealloc); | |
c6b4fcba | 1412 | |
b29d4986 JT |
1413 | if (r == 0) |
1414 | mg_copy(&mg->k.ws); | |
1415 | else | |
1416 | quiesce(mg, mg_copy); | |
c6b4fcba | 1417 | |
b29d4986 | 1418 | return 0; |
c6b4fcba JT |
1419 | } |
1420 | ||
b29d4986 | 1421 | static int mg_start(struct cache *cache, struct policy_work *op, struct bio *bio) |
c6b4fcba | 1422 | { |
b29d4986 | 1423 | struct dm_cache_migration *mg; |
c6b4fcba | 1424 | |
b29d4986 JT |
1425 | if (!background_work_begin(cache)) { |
1426 | policy_complete_background_work(cache->policy, op, false); | |
1427 | return -EPERM; | |
1428 | } | |
2ee57d58 | 1429 | |
b29d4986 | 1430 | mg = alloc_migration(cache); |
2ee57d58 | 1431 | |
b29d4986 JT |
1432 | mg->op = op; |
1433 | mg->overwrite_bio = bio; | |
1434 | ||
1435 | if (!bio) | |
1436 | inc_io_migrations(cache); | |
7ae34e77 | 1437 | |
b29d4986 | 1438 | return mg_lock_writes(mg); |
7ae34e77 JT |
1439 | } |
1440 | ||
c6b4fcba | 1441 | /*---------------------------------------------------------------- |
b29d4986 | 1442 | * invalidation processing |
c6b4fcba | 1443 | *--------------------------------------------------------------*/ |
c6b4fcba | 1444 | |
b29d4986 | 1445 | static void invalidate_complete(struct dm_cache_migration *mg, bool success) |
c6b4fcba | 1446 | { |
b29d4986 JT |
1447 | struct bio_list bios; |
1448 | struct cache *cache = mg->cache; | |
c6b4fcba | 1449 | |
b29d4986 JT |
1450 | bio_list_init(&bios); |
1451 | if (dm_cell_unlock_v2(cache->prison, mg->cell, &bios)) | |
1452 | free_prison_cell(cache, mg->cell); | |
c6b4fcba | 1453 | |
b29d4986 JT |
1454 | if (!success && mg->overwrite_bio) |
1455 | bio_io_error(mg->overwrite_bio); | |
c6b4fcba | 1456 | |
b29d4986 JT |
1457 | free_migration(mg); |
1458 | defer_bios(cache, &bios); | |
c6b4fcba | 1459 | |
b29d4986 | 1460 | background_work_end(cache); |
c6b4fcba JT |
1461 | } |
1462 | ||
b29d4986 | 1463 | static void invalidate_completed(struct work_struct *ws) |
c6b4fcba | 1464 | { |
b29d4986 JT |
1465 | struct dm_cache_migration *mg = ws_to_mg(ws); |
1466 | invalidate_complete(mg, !mg->k.input); | |
c6b4fcba JT |
1467 | } |
1468 | ||
b29d4986 | 1469 | static int invalidate_cblock(struct cache *cache, dm_cblock_t cblock) |
651f5fa2 | 1470 | { |
b29d4986 JT |
1471 | int r = policy_invalidate_mapping(cache->policy, cblock); |
1472 | if (!r) { | |
1473 | r = dm_cache_remove_mapping(cache->cmd, cblock); | |
1474 | if (r) { | |
1475 | DMERR_LIMIT("%s: invalidation failed; couldn't update on disk metadata", | |
1476 | cache_device_name(cache)); | |
1477 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); | |
651f5fa2 JT |
1478 | } |
1479 | ||
b29d4986 JT |
1480 | } else if (r == -ENODATA) { |
1481 | /* | |
1482 | * Harmless, already unmapped. | |
1483 | */ | |
1484 | r = 0; | |
651f5fa2 | 1485 | |
b29d4986 JT |
1486 | } else |
1487 | DMERR("%s: policy_invalidate_mapping failed", cache_device_name(cache)); | |
2ee57d58 | 1488 | |
b29d4986 | 1489 | return r; |
651f5fa2 JT |
1490 | } |
1491 | ||
b29d4986 | 1492 | static void invalidate_remove(struct work_struct *ws) |
651f5fa2 | 1493 | { |
b29d4986 JT |
1494 | int r; |
1495 | struct dm_cache_migration *mg = ws_to_mg(ws); | |
1496 | struct cache *cache = mg->cache; | |
651f5fa2 | 1497 | |
b29d4986 JT |
1498 | r = invalidate_cblock(cache, mg->invalidate_cblock); |
1499 | if (r) { | |
1500 | invalidate_complete(mg, false); | |
1501 | return; | |
651f5fa2 | 1502 | } |
9153df74 | 1503 | |
b29d4986 JT |
1504 | init_continuation(&mg->k, invalidate_completed); |
1505 | continue_after_commit(&cache->committer, &mg->k); | |
1506 | remap_to_origin_clear_discard(cache, mg->overwrite_bio, mg->invalidate_oblock); | |
1507 | mg->overwrite_bio = NULL; | |
1508 | schedule_commit(&cache->committer); | |
651f5fa2 JT |
1509 | } |
1510 | ||
b29d4986 | 1511 | static int invalidate_lock(struct dm_cache_migration *mg) |
651f5fa2 | 1512 | { |
b29d4986 JT |
1513 | int r; |
1514 | struct dm_cell_key_v2 key; | |
1515 | struct cache *cache = mg->cache; | |
1516 | struct dm_bio_prison_cell_v2 *prealloc; | |
651f5fa2 | 1517 | |
b29d4986 | 1518 | prealloc = alloc_prison_cell(cache); |
651f5fa2 | 1519 | |
b29d4986 JT |
1520 | build_key(mg->invalidate_oblock, oblock_succ(mg->invalidate_oblock), &key); |
1521 | r = dm_cell_lock_v2(cache->prison, &key, | |
1522 | READ_WRITE_LOCK_LEVEL, prealloc, &mg->cell); | |
1523 | if (r < 0) { | |
1524 | free_prison_cell(cache, prealloc); | |
1525 | invalidate_complete(mg, false); | |
1526 | return r; | |
651f5fa2 | 1527 | } |
9153df74 | 1528 | |
b29d4986 JT |
1529 | if (mg->cell != prealloc) |
1530 | free_prison_cell(cache, prealloc); | |
651f5fa2 | 1531 | |
b29d4986 JT |
1532 | if (r) |
1533 | quiesce(mg, invalidate_remove); | |
651f5fa2 | 1534 | |
b29d4986 JT |
1535 | else { |
1536 | /* | |
1537 | * We can't call invalidate_remove() directly here because we | |
1538 | * might still be in request context. | |
1539 | */ | |
1540 | init_continuation(&mg->k, invalidate_remove); | |
1541 | queue_work(cache->wq, &mg->k.ws); | |
1542 | } | |
fb4100ae | 1543 | |
fb4100ae JT |
1544 | return 0; |
1545 | } | |
1546 | ||
b29d4986 JT |
1547 | static int invalidate_start(struct cache *cache, dm_cblock_t cblock, |
1548 | dm_oblock_t oblock, struct bio *bio) | |
fb4100ae | 1549 | { |
b29d4986 | 1550 | struct dm_cache_migration *mg; |
2ee57d58 | 1551 | |
b29d4986 JT |
1552 | if (!background_work_begin(cache)) |
1553 | return -EPERM; | |
c6b4fcba | 1554 | |
b29d4986 | 1555 | mg = alloc_migration(cache); |
c6b4fcba | 1556 | |
b29d4986 JT |
1557 | mg->overwrite_bio = bio; |
1558 | mg->invalidate_cblock = cblock; | |
1559 | mg->invalidate_oblock = oblock; | |
c6b4fcba | 1560 | |
b29d4986 | 1561 | return invalidate_lock(mg); |
c6b4fcba | 1562 | } |
c6b4fcba | 1563 | |
b29d4986 JT |
1564 | /*---------------------------------------------------------------- |
1565 | * bio processing | |
1566 | *--------------------------------------------------------------*/ | |
c6b4fcba | 1567 | |
b29d4986 JT |
1568 | enum busy { |
1569 | IDLE, | |
b29d4986 JT |
1570 | BUSY |
1571 | }; | |
c6b4fcba | 1572 | |
b29d4986 | 1573 | static enum busy spare_migration_bandwidth(struct cache *cache) |
651f5fa2 | 1574 | { |
dc4fa29f | 1575 | bool idle = dm_iot_idle_for(&cache->tracker, HZ); |
a59db676 | 1576 | sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) * |
c6b4fcba | 1577 | cache->sectors_per_block; |
651f5fa2 | 1578 | |
49b7f768 JT |
1579 | if (idle && current_volume <= cache->migration_threshold) |
1580 | return IDLE; | |
b29d4986 | 1581 | else |
49b7f768 | 1582 | return BUSY; |
651f5fa2 JT |
1583 | } |
1584 | ||
c6b4fcba | 1585 | static void inc_hit_counter(struct cache *cache, struct bio *bio) |
c6b4fcba | 1586 | { |
c6b4fcba JT |
1587 | atomic_inc(bio_data_dir(bio) == READ ? |
1588 | &cache->stats.read_hit : &cache->stats.write_hit); | |
c6b4fcba JT |
1589 | } |
1590 | ||
c6b4fcba | 1591 | static void inc_miss_counter(struct cache *cache, struct bio *bio) |
028ae9f7 | 1592 | { |
c6b4fcba JT |
1593 | atomic_inc(bio_data_dir(bio) == READ ? |
1594 | &cache->stats.read_miss : &cache->stats.write_miss); | |
1595 | } | |
028ae9f7 | 1596 | |
fb4100ae | 1597 | /*----------------------------------------------------------------*/ |
028ae9f7 | 1598 | |
b29d4986 JT |
1599 | static int map_bio(struct cache *cache, struct bio *bio, dm_oblock_t block, |
1600 | bool *commit_needed) | |
c6b4fcba | 1601 | { |
b29d4986 JT |
1602 | int r, data_dir; |
1603 | bool rb, background_queued; | |
1604 | dm_cblock_t cblock; | |
c6b4fcba | 1605 | |
b29d4986 | 1606 | *commit_needed = false; |
c6b4fcba | 1607 | |
b29d4986 JT |
1608 | rb = bio_detain_shared(cache, block, bio); |
1609 | if (!rb) { | |
c6b4fcba | 1610 | /* |
b29d4986 JT |
1611 | * An exclusive lock is held for this block, so we have to |
1612 | * wait. We set the commit_needed flag so the current | |
1613 | * transaction will be committed asap, allowing this lock | |
1614 | * to be dropped. | |
c6b4fcba | 1615 | */ |
b29d4986 JT |
1616 | *commit_needed = true; |
1617 | return DM_MAPIO_SUBMITTED; | |
651f5fa2 | 1618 | } |
9153df74 | 1619 | |
b29d4986 | 1620 | data_dir = bio_data_dir(bio); |
651f5fa2 | 1621 | |
b29d4986 JT |
1622 | if (optimisable_bio(cache, bio, block)) { |
1623 | struct policy_work *op = NULL; | |
fb4100ae | 1624 | |
b29d4986 JT |
1625 | r = policy_lookup_with_work(cache->policy, block, &cblock, data_dir, true, &op); |
1626 | if (unlikely(r && r != -ENOENT)) { | |
1627 | DMERR_LIMIT("%s: policy_lookup_with_work() failed with r = %d", | |
1628 | cache_device_name(cache), r); | |
1629 | bio_io_error(bio); | |
1630 | return DM_MAPIO_SUBMITTED; | |
c6b4fcba JT |
1631 | } |
1632 | ||
b29d4986 JT |
1633 | if (r == -ENOENT && op) { |
1634 | bio_drop_shared_lock(cache, bio); | |
1635 | BUG_ON(op->op != POLICY_PROMOTE); | |
1636 | mg_start(cache, op, bio); | |
1637 | return DM_MAPIO_SUBMITTED; | |
1638 | } | |
1639 | } else { | |
1640 | r = policy_lookup(cache->policy, block, &cblock, data_dir, false, &background_queued); | |
1641 | if (unlikely(r && r != -ENOENT)) { | |
1642 | DMERR_LIMIT("%s: policy_lookup() failed with r = %d", | |
1643 | cache_device_name(cache), r); | |
1644 | bio_io_error(bio); | |
1645 | return DM_MAPIO_SUBMITTED; | |
1646 | } | |
c6b4fcba | 1647 | |
b29d4986 JT |
1648 | if (background_queued) |
1649 | wake_migration_worker(cache); | |
c6b4fcba JT |
1650 | } |
1651 | ||
b29d4986 | 1652 | if (r == -ENOENT) { |
693b960e MS |
1653 | struct per_bio_data *pb = get_per_bio_data(bio); |
1654 | ||
b29d4986 JT |
1655 | /* |
1656 | * Miss. | |
1657 | */ | |
1658 | inc_miss_counter(cache, bio); | |
1659 | if (pb->req_nr == 0) { | |
1660 | accounted_begin(cache, bio); | |
1661 | remap_to_origin_clear_discard(cache, bio, block); | |
b29d4986 | 1662 | } else { |
2ee57d58 | 1663 | /* |
b29d4986 JT |
1664 | * This is a duplicate writethrough io that is no |
1665 | * longer needed because the block has been demoted. | |
2ee57d58 | 1666 | */ |
b29d4986 JT |
1667 | bio_endio(bio); |
1668 | return DM_MAPIO_SUBMITTED; | |
1669 | } | |
1670 | } else { | |
1671 | /* | |
1672 | * Hit. | |
1673 | */ | |
1674 | inc_hit_counter(cache, bio); | |
651f5fa2 | 1675 | |
b29d4986 JT |
1676 | /* |
1677 | * Passthrough always maps to the origin, invalidating any | |
1678 | * cache blocks that are written to. | |
1679 | */ | |
8e3c3827 | 1680 | if (passthrough_mode(cache)) { |
2ee57d58 | 1681 | if (bio_data_dir(bio) == WRITE) { |
b29d4986 | 1682 | bio_drop_shared_lock(cache, bio); |
2ee57d58 | 1683 | atomic_inc(&cache->stats.demotion); |
b29d4986 JT |
1684 | invalidate_start(cache, cblock, block, bio); |
1685 | } else | |
2ee57d58 | 1686 | remap_to_origin_clear_discard(cache, bio, block); |
2ee57d58 | 1687 | } else { |
8e3c3827 | 1688 | if (bio_data_dir(bio) == WRITE && writethrough_mode(cache) && |
b29d4986 | 1689 | !is_dirty(cache, cblock)) { |
2df3bae9 | 1690 | remap_to_origin_and_cache(cache, bio, block, cblock); |
b29d4986 JT |
1691 | accounted_begin(cache, bio); |
1692 | } else | |
1693 | remap_to_cache_dirty(cache, bio, block, cblock); | |
2ee57d58 | 1694 | } |
c6b4fcba | 1695 | } |
651f5fa2 | 1696 | |
651f5fa2 | 1697 | /* |
b29d4986 | 1698 | * dm core turns FUA requests into a separate payload and FLUSH req. |
651f5fa2 | 1699 | */ |
b29d4986 | 1700 | if (bio->bi_opf & REQ_FUA) { |
651f5fa2 | 1701 | /* |
b29d4986 JT |
1702 | * issue_after_commit will call accounted_begin a second time. So |
1703 | * we call accounted_complete() to avoid double accounting. | |
651f5fa2 | 1704 | */ |
b29d4986 JT |
1705 | accounted_complete(cache, bio); |
1706 | issue_after_commit(&cache->committer, bio); | |
1707 | *commit_needed = true; | |
1708 | return DM_MAPIO_SUBMITTED; | |
651f5fa2 JT |
1709 | } |
1710 | ||
b29d4986 | 1711 | return DM_MAPIO_REMAPPED; |
651f5fa2 JT |
1712 | } |
1713 | ||
b29d4986 | 1714 | static bool process_bio(struct cache *cache, struct bio *bio) |
c6b4fcba | 1715 | { |
b29d4986 | 1716 | bool commit_needed; |
c6b4fcba | 1717 | |
b29d4986 | 1718 | if (map_bio(cache, bio, get_bio_block(cache, bio), &commit_needed) == DM_MAPIO_REMAPPED) |
ed00aabd | 1719 | submit_bio_noacct(bio); |
c6b4fcba | 1720 | |
b29d4986 | 1721 | return commit_needed; |
c6b4fcba JT |
1722 | } |
1723 | ||
028ae9f7 JT |
1724 | /* |
1725 | * A non-zero return indicates read_only or fail_io mode. | |
1726 | */ | |
1727 | static int commit(struct cache *cache, bool clean_shutdown) | |
e2e74d61 | 1728 | { |
028ae9f7 | 1729 | int r; |
e2e74d61 | 1730 | |
028ae9f7 JT |
1731 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
1732 | return -EINVAL; | |
e2e74d61 | 1733 | |
028ae9f7 JT |
1734 | atomic_inc(&cache->stats.commit_count); |
1735 | r = dm_cache_commit(cache->cmd, clean_shutdown); | |
1736 | if (r) | |
1737 | metadata_operation_failed(cache, "dm_cache_commit", r); | |
e2e74d61 | 1738 | |
028ae9f7 | 1739 | return r; |
e2e74d61 JT |
1740 | } |
1741 | ||
b29d4986 JT |
1742 | /* |
1743 | * Used by the batcher. | |
1744 | */ | |
4e4cbee9 | 1745 | static blk_status_t commit_op(void *context) |
c6b4fcba | 1746 | { |
b29d4986 | 1747 | struct cache *cache = context; |
c6b4fcba | 1748 | |
b29d4986 | 1749 | if (dm_cache_changed_this_transaction(cache->cmd)) |
4e4cbee9 | 1750 | return errno_to_blk_status(commit(cache, false)); |
c6b4fcba | 1751 | |
b29d4986 | 1752 | return 0; |
c6b4fcba JT |
1753 | } |
1754 | ||
b29d4986 | 1755 | /*----------------------------------------------------------------*/ |
65790ff9 | 1756 | |
b29d4986 | 1757 | static bool process_flush_bio(struct cache *cache, struct bio *bio) |
65790ff9 | 1758 | { |
693b960e | 1759 | struct per_bio_data *pb = get_per_bio_data(bio); |
65790ff9 | 1760 | |
b29d4986 JT |
1761 | if (!pb->req_nr) |
1762 | remap_to_origin(cache, bio); | |
1763 | else | |
1764 | remap_to_cache(cache, bio, 0); | |
65790ff9 | 1765 | |
b29d4986 JT |
1766 | issue_after_commit(&cache->committer, bio); |
1767 | return true; | |
65790ff9 JT |
1768 | } |
1769 | ||
b29d4986 | 1770 | static bool process_discard_bio(struct cache *cache, struct bio *bio) |
65790ff9 | 1771 | { |
b29d4986 | 1772 | dm_dblock_t b, e; |
65790ff9 | 1773 | |
b29d4986 JT |
1774 | // FIXME: do we need to lock the region? Or can we just assume the |
1775 | // user wont be so foolish as to issue discard concurrently with | |
1776 | // other IO? | |
1777 | calc_discard_block_range(cache, bio, &b, &e); | |
1778 | while (b != e) { | |
1779 | set_discard(cache, b); | |
1780 | b = to_dblock(from_dblock(b) + 1); | |
651f5fa2 | 1781 | } |
65790ff9 | 1782 | |
de7180ff MS |
1783 | if (cache->features.discard_passdown) { |
1784 | remap_to_origin(cache, bio); | |
ed00aabd | 1785 | submit_bio_noacct(bio); |
de7180ff MS |
1786 | } else |
1787 | bio_endio(bio); | |
65790ff9 | 1788 | |
b29d4986 | 1789 | return false; |
c6b4fcba JT |
1790 | } |
1791 | ||
b29d4986 | 1792 | static void process_deferred_bios(struct work_struct *ws) |
66cb1910 | 1793 | { |
b29d4986 | 1794 | struct cache *cache = container_of(ws, struct cache, deferred_bio_worker); |
66cb1910 | 1795 | |
b29d4986 | 1796 | bool commit_needed = false; |
c6b4fcba JT |
1797 | struct bio_list bios; |
1798 | struct bio *bio; | |
66cb1910 | 1799 | |
c6b4fcba | 1800 | bio_list_init(&bios); |
c6b4fcba | 1801 | |
26b924b9 | 1802 | spin_lock_irq(&cache->lock); |
b29d4986 JT |
1803 | bio_list_merge(&bios, &cache->deferred_bios); |
1804 | bio_list_init(&cache->deferred_bios); | |
26b924b9 | 1805 | spin_unlock_irq(&cache->lock); |
c6b4fcba | 1806 | |
b29d4986 JT |
1807 | while ((bio = bio_list_pop(&bios))) { |
1808 | if (bio->bi_opf & REQ_PREFLUSH) | |
1809 | commit_needed = process_flush_bio(cache, bio) || commit_needed; | |
c6b4fcba | 1810 | |
b29d4986 JT |
1811 | else if (bio_op(bio) == REQ_OP_DISCARD) |
1812 | commit_needed = process_discard_bio(cache, bio) || commit_needed; | |
1813 | ||
1814 | else | |
1815 | commit_needed = process_bio(cache, bio) || commit_needed; | |
1816 | } | |
1817 | ||
1818 | if (commit_needed) | |
1819 | schedule_commit(&cache->committer); | |
c6b4fcba JT |
1820 | } |
1821 | ||
c6b4fcba JT |
1822 | /*---------------------------------------------------------------- |
1823 | * Main worker loop | |
1824 | *--------------------------------------------------------------*/ | |
651f5fa2 JT |
1825 | |
1826 | static void requeue_deferred_bios(struct cache *cache) | |
c6b4fcba JT |
1827 | { |
1828 | struct bio *bio; | |
1829 | struct bio_list bios; | |
1830 | ||
1831 | bio_list_init(&bios); | |
1832 | bio_list_merge(&bios, &cache->deferred_bios); | |
1833 | bio_list_init(&cache->deferred_bios); | |
1834 | ||
4246a0b6 | 1835 | while ((bio = bio_list_pop(&bios))) { |
4e4cbee9 | 1836 | bio->bi_status = BLK_STS_DM_REQUEUE; |
4246a0b6 CH |
1837 | bio_endio(bio); |
1838 | } | |
c6b4fcba JT |
1839 | } |
1840 | ||
c6b4fcba JT |
1841 | /* |
1842 | * We want to commit periodically so that not too much | |
1843 | * unwritten metadata builds up. | |
1844 | */ | |
1845 | static void do_waker(struct work_struct *ws) | |
1846 | { | |
1847 | struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker); | |
b29d4986 | 1848 | |
fba10109 | 1849 | policy_tick(cache->policy, true); |
b29d4986 JT |
1850 | wake_migration_worker(cache); |
1851 | schedule_commit(&cache->committer); | |
c6b4fcba JT |
1852 | queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD); |
1853 | } | |
1854 | ||
b29d4986 | 1855 | static void check_migrations(struct work_struct *ws) |
c6b4fcba | 1856 | { |
b29d4986 JT |
1857 | int r; |
1858 | struct policy_work *op; | |
1859 | struct cache *cache = container_of(ws, struct cache, migration_worker); | |
1860 | enum busy b; | |
c6b4fcba | 1861 | |
b29d4986 JT |
1862 | for (;;) { |
1863 | b = spare_migration_bandwidth(cache); | |
c6b4fcba | 1864 | |
b29d4986 JT |
1865 | r = policy_get_background_work(cache->policy, b == IDLE, &op); |
1866 | if (r == -ENODATA) | |
1867 | break; | |
1868 | ||
1869 | if (r) { | |
1870 | DMERR_LIMIT("%s: policy_background_work failed", | |
1871 | cache_device_name(cache)); | |
1872 | break; | |
1873 | } | |
1874 | ||
1875 | r = mg_start(cache, op, NULL); | |
1876 | if (r) | |
1877 | break; | |
1878 | } | |
c6b4fcba JT |
1879 | } |
1880 | ||
1881 | /*---------------------------------------------------------------- | |
1882 | * Target methods | |
1883 | *--------------------------------------------------------------*/ | |
1884 | ||
1885 | /* | |
1886 | * This function gets called on the error paths of the constructor, so we | |
1887 | * have to cope with a partially initialised struct. | |
1888 | */ | |
1889 | static void destroy(struct cache *cache) | |
1890 | { | |
1891 | unsigned i; | |
1892 | ||
6f1c819c | 1893 | mempool_exit(&cache->migration_pool); |
c6b4fcba | 1894 | |
c6b4fcba | 1895 | if (cache->prison) |
b29d4986 | 1896 | dm_bio_prison_destroy_v2(cache->prison); |
c6b4fcba JT |
1897 | |
1898 | if (cache->wq) | |
1899 | destroy_workqueue(cache->wq); | |
1900 | ||
1901 | if (cache->dirty_bitset) | |
1902 | free_bitset(cache->dirty_bitset); | |
1903 | ||
1904 | if (cache->discard_bitset) | |
1905 | free_bitset(cache->discard_bitset); | |
1906 | ||
1907 | if (cache->copier) | |
1908 | dm_kcopyd_client_destroy(cache->copier); | |
1909 | ||
1910 | if (cache->cmd) | |
1911 | dm_cache_metadata_close(cache->cmd); | |
1912 | ||
1913 | if (cache->metadata_dev) | |
1914 | dm_put_device(cache->ti, cache->metadata_dev); | |
1915 | ||
1916 | if (cache->origin_dev) | |
1917 | dm_put_device(cache->ti, cache->origin_dev); | |
1918 | ||
1919 | if (cache->cache_dev) | |
1920 | dm_put_device(cache->ti, cache->cache_dev); | |
1921 | ||
1922 | if (cache->policy) | |
1923 | dm_cache_policy_destroy(cache->policy); | |
1924 | ||
1925 | for (i = 0; i < cache->nr_ctr_args ; i++) | |
1926 | kfree(cache->ctr_args[i]); | |
1927 | kfree(cache->ctr_args); | |
1928 | ||
6f1c819c | 1929 | bioset_exit(&cache->bs); |
2df3bae9 | 1930 | |
c6b4fcba JT |
1931 | kfree(cache); |
1932 | } | |
1933 | ||
1934 | static void cache_dtr(struct dm_target *ti) | |
1935 | { | |
1936 | struct cache *cache = ti->private; | |
1937 | ||
1938 | destroy(cache); | |
1939 | } | |
1940 | ||
1941 | static sector_t get_dev_size(struct dm_dev *dev) | |
1942 | { | |
1943 | return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
1944 | } | |
1945 | ||
1946 | /*----------------------------------------------------------------*/ | |
1947 | ||
1948 | /* | |
1949 | * Construct a cache device mapping. | |
1950 | * | |
1951 | * cache <metadata dev> <cache dev> <origin dev> <block size> | |
1952 | * <#feature args> [<feature arg>]* | |
1953 | * <policy> <#policy args> [<policy arg>]* | |
1954 | * | |
1955 | * metadata dev : fast device holding the persistent metadata | |
1956 | * cache dev : fast device holding cached data blocks | |
1957 | * origin dev : slow device holding original data blocks | |
1958 | * block size : cache unit size in sectors | |
1959 | * | |
1960 | * #feature args : number of feature arguments passed | |
1961 | * feature args : writethrough. (The default is writeback.) | |
1962 | * | |
1963 | * policy : the replacement policy to use | |
1964 | * #policy args : an even number of policy arguments corresponding | |
1965 | * to key/value pairs passed to the policy | |
1966 | * policy args : key/value pairs passed to the policy | |
1967 | * E.g. 'sequential_threshold 1024' | |
1968 | * See cache-policies.txt for details. | |
1969 | * | |
1970 | * Optional feature arguments are: | |
1971 | * writethrough : write through caching that prohibits cache block | |
1972 | * content from being different from origin block content. | |
1973 | * Without this argument, the default behaviour is to write | |
1974 | * back cache block contents later for performance reasons, | |
1975 | * so they may differ from the corresponding origin blocks. | |
1976 | */ | |
1977 | struct cache_args { | |
1978 | struct dm_target *ti; | |
1979 | ||
1980 | struct dm_dev *metadata_dev; | |
1981 | ||
1982 | struct dm_dev *cache_dev; | |
1983 | sector_t cache_sectors; | |
1984 | ||
1985 | struct dm_dev *origin_dev; | |
1986 | sector_t origin_sectors; | |
1987 | ||
1988 | uint32_t block_size; | |
1989 | ||
1990 | const char *policy_name; | |
1991 | int policy_argc; | |
1992 | const char **policy_argv; | |
1993 | ||
1994 | struct cache_features features; | |
1995 | }; | |
1996 | ||
1997 | static void destroy_cache_args(struct cache_args *ca) | |
1998 | { | |
1999 | if (ca->metadata_dev) | |
2000 | dm_put_device(ca->ti, ca->metadata_dev); | |
2001 | ||
2002 | if (ca->cache_dev) | |
2003 | dm_put_device(ca->ti, ca->cache_dev); | |
2004 | ||
2005 | if (ca->origin_dev) | |
2006 | dm_put_device(ca->ti, ca->origin_dev); | |
2007 | ||
2008 | kfree(ca); | |
2009 | } | |
2010 | ||
2011 | static bool at_least_one_arg(struct dm_arg_set *as, char **error) | |
2012 | { | |
2013 | if (!as->argc) { | |
2014 | *error = "Insufficient args"; | |
2015 | return false; | |
2016 | } | |
2017 | ||
2018 | return true; | |
2019 | } | |
2020 | ||
2021 | static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2022 | char **error) | |
2023 | { | |
2024 | int r; | |
2025 | sector_t metadata_dev_size; | |
2026 | char b[BDEVNAME_SIZE]; | |
2027 | ||
2028 | if (!at_least_one_arg(as, error)) | |
2029 | return -EINVAL; | |
2030 | ||
2031 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2032 | &ca->metadata_dev); | |
2033 | if (r) { | |
2034 | *error = "Error opening metadata device"; | |
2035 | return r; | |
2036 | } | |
2037 | ||
2038 | metadata_dev_size = get_dev_size(ca->metadata_dev); | |
2039 | if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING) | |
2040 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
2041 | bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
2042 | ||
2043 | return 0; | |
2044 | } | |
2045 | ||
2046 | static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2047 | char **error) | |
2048 | { | |
2049 | int r; | |
2050 | ||
2051 | if (!at_least_one_arg(as, error)) | |
2052 | return -EINVAL; | |
2053 | ||
2054 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2055 | &ca->cache_dev); | |
2056 | if (r) { | |
2057 | *error = "Error opening cache device"; | |
2058 | return r; | |
2059 | } | |
2060 | ca->cache_sectors = get_dev_size(ca->cache_dev); | |
2061 | ||
2062 | return 0; | |
2063 | } | |
2064 | ||
2065 | static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2066 | char **error) | |
2067 | { | |
2068 | int r; | |
2069 | ||
2070 | if (!at_least_one_arg(as, error)) | |
2071 | return -EINVAL; | |
2072 | ||
2073 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2074 | &ca->origin_dev); | |
2075 | if (r) { | |
2076 | *error = "Error opening origin device"; | |
2077 | return r; | |
2078 | } | |
2079 | ||
2080 | ca->origin_sectors = get_dev_size(ca->origin_dev); | |
2081 | if (ca->ti->len > ca->origin_sectors) { | |
2082 | *error = "Device size larger than cached device"; | |
2083 | return -EINVAL; | |
2084 | } | |
2085 | ||
2086 | return 0; | |
2087 | } | |
2088 | ||
2089 | static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as, | |
2090 | char **error) | |
2091 | { | |
05473044 | 2092 | unsigned long block_size; |
c6b4fcba JT |
2093 | |
2094 | if (!at_least_one_arg(as, error)) | |
2095 | return -EINVAL; | |
2096 | ||
05473044 MS |
2097 | if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size || |
2098 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2099 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
2100 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { | |
c6b4fcba JT |
2101 | *error = "Invalid data block size"; |
2102 | return -EINVAL; | |
2103 | } | |
2104 | ||
05473044 | 2105 | if (block_size > ca->cache_sectors) { |
c6b4fcba JT |
2106 | *error = "Data block size is larger than the cache device"; |
2107 | return -EINVAL; | |
2108 | } | |
2109 | ||
05473044 | 2110 | ca->block_size = block_size; |
c6b4fcba JT |
2111 | |
2112 | return 0; | |
2113 | } | |
2114 | ||
2115 | static void init_features(struct cache_features *cf) | |
2116 | { | |
2117 | cf->mode = CM_WRITE; | |
2ee57d58 | 2118 | cf->io_mode = CM_IO_WRITEBACK; |
629d0a8a | 2119 | cf->metadata_version = 1; |
de7180ff | 2120 | cf->discard_passdown = true; |
c6b4fcba JT |
2121 | } |
2122 | ||
2123 | static int parse_features(struct cache_args *ca, struct dm_arg_set *as, | |
2124 | char **error) | |
2125 | { | |
5916a22b | 2126 | static const struct dm_arg _args[] = { |
de7180ff | 2127 | {0, 3, "Invalid number of cache feature arguments"}, |
c6b4fcba JT |
2128 | }; |
2129 | ||
af9313c3 | 2130 | int r, mode_ctr = 0; |
c6b4fcba JT |
2131 | unsigned argc; |
2132 | const char *arg; | |
2133 | struct cache_features *cf = &ca->features; | |
2134 | ||
2135 | init_features(cf); | |
2136 | ||
2137 | r = dm_read_arg_group(_args, as, &argc, error); | |
2138 | if (r) | |
2139 | return -EINVAL; | |
2140 | ||
2141 | while (argc--) { | |
2142 | arg = dm_shift_arg(as); | |
2143 | ||
af9313c3 | 2144 | if (!strcasecmp(arg, "writeback")) { |
2ee57d58 | 2145 | cf->io_mode = CM_IO_WRITEBACK; |
af9313c3 JP |
2146 | mode_ctr++; |
2147 | } | |
c6b4fcba | 2148 | |
af9313c3 | 2149 | else if (!strcasecmp(arg, "writethrough")) { |
2ee57d58 | 2150 | cf->io_mode = CM_IO_WRITETHROUGH; |
af9313c3 JP |
2151 | mode_ctr++; |
2152 | } | |
2ee57d58 | 2153 | |
af9313c3 | 2154 | else if (!strcasecmp(arg, "passthrough")) { |
2ee57d58 | 2155 | cf->io_mode = CM_IO_PASSTHROUGH; |
af9313c3 JP |
2156 | mode_ctr++; |
2157 | } | |
c6b4fcba | 2158 | |
629d0a8a JT |
2159 | else if (!strcasecmp(arg, "metadata2")) |
2160 | cf->metadata_version = 2; | |
2161 | ||
de7180ff MS |
2162 | else if (!strcasecmp(arg, "no_discard_passdown")) |
2163 | cf->discard_passdown = false; | |
2164 | ||
c6b4fcba JT |
2165 | else { |
2166 | *error = "Unrecognised cache feature requested"; | |
2167 | return -EINVAL; | |
2168 | } | |
2169 | } | |
2170 | ||
af9313c3 JP |
2171 | if (mode_ctr > 1) { |
2172 | *error = "Duplicate cache io_mode features requested"; | |
2173 | return -EINVAL; | |
2174 | } | |
2175 | ||
c6b4fcba JT |
2176 | return 0; |
2177 | } | |
2178 | ||
2179 | static int parse_policy(struct cache_args *ca, struct dm_arg_set *as, | |
2180 | char **error) | |
2181 | { | |
5916a22b | 2182 | static const struct dm_arg _args[] = { |
c6b4fcba JT |
2183 | {0, 1024, "Invalid number of policy arguments"}, |
2184 | }; | |
2185 | ||
2186 | int r; | |
2187 | ||
2188 | if (!at_least_one_arg(as, error)) | |
2189 | return -EINVAL; | |
2190 | ||
2191 | ca->policy_name = dm_shift_arg(as); | |
2192 | ||
2193 | r = dm_read_arg_group(_args, as, &ca->policy_argc, error); | |
2194 | if (r) | |
2195 | return -EINVAL; | |
2196 | ||
2197 | ca->policy_argv = (const char **)as->argv; | |
2198 | dm_consume_args(as, ca->policy_argc); | |
2199 | ||
2200 | return 0; | |
2201 | } | |
2202 | ||
2203 | static int parse_cache_args(struct cache_args *ca, int argc, char **argv, | |
2204 | char **error) | |
2205 | { | |
2206 | int r; | |
2207 | struct dm_arg_set as; | |
2208 | ||
2209 | as.argc = argc; | |
2210 | as.argv = argv; | |
2211 | ||
2212 | r = parse_metadata_dev(ca, &as, error); | |
2213 | if (r) | |
2214 | return r; | |
2215 | ||
2216 | r = parse_cache_dev(ca, &as, error); | |
2217 | if (r) | |
2218 | return r; | |
2219 | ||
2220 | r = parse_origin_dev(ca, &as, error); | |
2221 | if (r) | |
2222 | return r; | |
2223 | ||
2224 | r = parse_block_size(ca, &as, error); | |
2225 | if (r) | |
2226 | return r; | |
2227 | ||
2228 | r = parse_features(ca, &as, error); | |
2229 | if (r) | |
2230 | return r; | |
2231 | ||
2232 | r = parse_policy(ca, &as, error); | |
2233 | if (r) | |
2234 | return r; | |
2235 | ||
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | /*----------------------------------------------------------------*/ | |
2240 | ||
2241 | static struct kmem_cache *migration_cache; | |
2242 | ||
2c73c471 AK |
2243 | #define NOT_CORE_OPTION 1 |
2244 | ||
2f14f4b5 | 2245 | static int process_config_option(struct cache *cache, const char *key, const char *value) |
2c73c471 AK |
2246 | { |
2247 | unsigned long tmp; | |
2248 | ||
2f14f4b5 JT |
2249 | if (!strcasecmp(key, "migration_threshold")) { |
2250 | if (kstrtoul(value, 10, &tmp)) | |
2c73c471 AK |
2251 | return -EINVAL; |
2252 | ||
2253 | cache->migration_threshold = tmp; | |
2254 | return 0; | |
2255 | } | |
2256 | ||
2257 | return NOT_CORE_OPTION; | |
2258 | } | |
2259 | ||
2f14f4b5 JT |
2260 | static int set_config_value(struct cache *cache, const char *key, const char *value) |
2261 | { | |
2262 | int r = process_config_option(cache, key, value); | |
2263 | ||
2264 | if (r == NOT_CORE_OPTION) | |
2265 | r = policy_set_config_value(cache->policy, key, value); | |
2266 | ||
2267 | if (r) | |
2268 | DMWARN("bad config value for %s: %s", key, value); | |
2269 | ||
2270 | return r; | |
2271 | } | |
2272 | ||
2273 | static int set_config_values(struct cache *cache, int argc, const char **argv) | |
c6b4fcba JT |
2274 | { |
2275 | int r = 0; | |
2276 | ||
2277 | if (argc & 1) { | |
2278 | DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs."); | |
2279 | return -EINVAL; | |
2280 | } | |
2281 | ||
2282 | while (argc) { | |
2f14f4b5 JT |
2283 | r = set_config_value(cache, argv[0], argv[1]); |
2284 | if (r) | |
2285 | break; | |
c6b4fcba JT |
2286 | |
2287 | argc -= 2; | |
2288 | argv += 2; | |
2289 | } | |
2290 | ||
2291 | return r; | |
2292 | } | |
2293 | ||
2294 | static int create_cache_policy(struct cache *cache, struct cache_args *ca, | |
2295 | char **error) | |
2296 | { | |
4cb3e1db MP |
2297 | struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name, |
2298 | cache->cache_size, | |
2299 | cache->origin_sectors, | |
2300 | cache->sectors_per_block); | |
2301 | if (IS_ERR(p)) { | |
c6b4fcba | 2302 | *error = "Error creating cache's policy"; |
4cb3e1db | 2303 | return PTR_ERR(p); |
c6b4fcba | 2304 | } |
4cb3e1db | 2305 | cache->policy = p; |
b29d4986 | 2306 | BUG_ON(!cache->policy); |
c6b4fcba | 2307 | |
2f14f4b5 | 2308 | return 0; |
c6b4fcba JT |
2309 | } |
2310 | ||
08b18451 | 2311 | /* |
2bb812df JT |
2312 | * We want the discard block size to be at least the size of the cache |
2313 | * block size and have no more than 2^14 discard blocks across the origin. | |
08b18451 JT |
2314 | */ |
2315 | #define MAX_DISCARD_BLOCKS (1 << 14) | |
2316 | ||
2317 | static bool too_many_discard_blocks(sector_t discard_block_size, | |
2318 | sector_t origin_size) | |
2319 | { | |
2320 | (void) sector_div(origin_size, discard_block_size); | |
2321 | ||
2322 | return origin_size > MAX_DISCARD_BLOCKS; | |
2323 | } | |
2324 | ||
2325 | static sector_t calculate_discard_block_size(sector_t cache_block_size, | |
2326 | sector_t origin_size) | |
2327 | { | |
2bb812df | 2328 | sector_t discard_block_size = cache_block_size; |
08b18451 JT |
2329 | |
2330 | if (origin_size) | |
2331 | while (too_many_discard_blocks(discard_block_size, origin_size)) | |
2332 | discard_block_size *= 2; | |
2333 | ||
2334 | return discard_block_size; | |
2335 | } | |
2336 | ||
d1d9220c JT |
2337 | static void set_cache_size(struct cache *cache, dm_cblock_t size) |
2338 | { | |
2339 | dm_block_t nr_blocks = from_cblock(size); | |
2340 | ||
2341 | if (nr_blocks > (1 << 20) && cache->cache_size != size) | |
2342 | DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n" | |
2343 | "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n" | |
2344 | "Please consider increasing the cache block size to reduce the overall cache block count.", | |
2345 | (unsigned long long) nr_blocks); | |
2346 | ||
2347 | cache->cache_size = size; | |
2348 | } | |
2349 | ||
f8350daf | 2350 | #define DEFAULT_MIGRATION_THRESHOLD 2048 |
c6b4fcba | 2351 | |
c6b4fcba JT |
2352 | static int cache_create(struct cache_args *ca, struct cache **result) |
2353 | { | |
2354 | int r = 0; | |
2355 | char **error = &ca->ti->error; | |
2356 | struct cache *cache; | |
2357 | struct dm_target *ti = ca->ti; | |
2358 | dm_block_t origin_blocks; | |
2359 | struct dm_cache_metadata *cmd; | |
2360 | bool may_format = ca->features.mode == CM_WRITE; | |
2361 | ||
2362 | cache = kzalloc(sizeof(*cache), GFP_KERNEL); | |
2363 | if (!cache) | |
2364 | return -ENOMEM; | |
2365 | ||
2366 | cache->ti = ca->ti; | |
2367 | ti->private = cache; | |
c6b4fcba JT |
2368 | ti->num_flush_bios = 2; |
2369 | ti->flush_supported = true; | |
2370 | ||
2371 | ti->num_discard_bios = 1; | |
2372 | ti->discards_supported = true; | |
c6b4fcba | 2373 | |
693b960e | 2374 | ti->per_io_data_size = sizeof(struct per_bio_data); |
c6b4fcba | 2375 | |
693b960e | 2376 | cache->features = ca->features; |
2df3bae9 MS |
2377 | if (writethrough_mode(cache)) { |
2378 | /* Create bioset for writethrough bios issued to origin */ | |
6f1c819c KO |
2379 | r = bioset_init(&cache->bs, BIO_POOL_SIZE, 0, 0); |
2380 | if (r) | |
2df3bae9 MS |
2381 | goto bad; |
2382 | } | |
2383 | ||
c6b4fcba JT |
2384 | cache->metadata_dev = ca->metadata_dev; |
2385 | cache->origin_dev = ca->origin_dev; | |
2386 | cache->cache_dev = ca->cache_dev; | |
2387 | ||
2388 | ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL; | |
2389 | ||
c6b4fcba | 2390 | origin_blocks = cache->origin_sectors = ca->origin_sectors; |
414dd67d | 2391 | origin_blocks = block_div(origin_blocks, ca->block_size); |
c6b4fcba JT |
2392 | cache->origin_blocks = to_oblock(origin_blocks); |
2393 | ||
2394 | cache->sectors_per_block = ca->block_size; | |
2395 | if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) { | |
2396 | r = -EINVAL; | |
2397 | goto bad; | |
2398 | } | |
2399 | ||
2400 | if (ca->block_size & (ca->block_size - 1)) { | |
2401 | dm_block_t cache_size = ca->cache_sectors; | |
2402 | ||
2403 | cache->sectors_per_block_shift = -1; | |
414dd67d | 2404 | cache_size = block_div(cache_size, ca->block_size); |
d1d9220c | 2405 | set_cache_size(cache, to_cblock(cache_size)); |
c6b4fcba JT |
2406 | } else { |
2407 | cache->sectors_per_block_shift = __ffs(ca->block_size); | |
d1d9220c | 2408 | set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift)); |
c6b4fcba JT |
2409 | } |
2410 | ||
2411 | r = create_cache_policy(cache, ca, error); | |
2412 | if (r) | |
2413 | goto bad; | |
2f14f4b5 | 2414 | |
c6b4fcba | 2415 | cache->policy_nr_args = ca->policy_argc; |
2f14f4b5 JT |
2416 | cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD; |
2417 | ||
2418 | r = set_config_values(cache, ca->policy_argc, ca->policy_argv); | |
2419 | if (r) { | |
2420 | *error = "Error setting cache policy's config values"; | |
2421 | goto bad; | |
2422 | } | |
c6b4fcba JT |
2423 | |
2424 | cmd = dm_cache_metadata_open(cache->metadata_dev->bdev, | |
2425 | ca->block_size, may_format, | |
629d0a8a JT |
2426 | dm_cache_policy_get_hint_size(cache->policy), |
2427 | ca->features.metadata_version); | |
c6b4fcba JT |
2428 | if (IS_ERR(cmd)) { |
2429 | *error = "Error creating metadata object"; | |
2430 | r = PTR_ERR(cmd); | |
2431 | goto bad; | |
2432 | } | |
2433 | cache->cmd = cmd; | |
028ae9f7 JT |
2434 | set_cache_mode(cache, CM_WRITE); |
2435 | if (get_cache_mode(cache) != CM_WRITE) { | |
2436 | *error = "Unable to get write access to metadata, please check/repair metadata."; | |
2437 | r = -EINVAL; | |
2438 | goto bad; | |
2439 | } | |
c6b4fcba | 2440 | |
8e3c3827 | 2441 | if (passthrough_mode(cache)) { |
2ee57d58 JT |
2442 | bool all_clean; |
2443 | ||
2444 | r = dm_cache_metadata_all_clean(cache->cmd, &all_clean); | |
2445 | if (r) { | |
2446 | *error = "dm_cache_metadata_all_clean() failed"; | |
2447 | goto bad; | |
2448 | } | |
2449 | ||
2450 | if (!all_clean) { | |
2451 | *error = "Cannot enter passthrough mode unless all blocks are clean"; | |
2452 | r = -EINVAL; | |
2453 | goto bad; | |
2454 | } | |
b29d4986 JT |
2455 | |
2456 | policy_allow_migrations(cache->policy, false); | |
2ee57d58 JT |
2457 | } |
2458 | ||
c6b4fcba JT |
2459 | spin_lock_init(&cache->lock); |
2460 | bio_list_init(&cache->deferred_bios); | |
a59db676 JT |
2461 | atomic_set(&cache->nr_allocated_migrations, 0); |
2462 | atomic_set(&cache->nr_io_migrations, 0); | |
c6b4fcba JT |
2463 | init_waitqueue_head(&cache->migration_wait); |
2464 | ||
fa4d683a | 2465 | r = -ENOMEM; |
44fa816b | 2466 | atomic_set(&cache->nr_dirty, 0); |
c6b4fcba JT |
2467 | cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size)); |
2468 | if (!cache->dirty_bitset) { | |
2469 | *error = "could not allocate dirty bitset"; | |
2470 | goto bad; | |
2471 | } | |
2472 | clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size)); | |
2473 | ||
08b18451 JT |
2474 | cache->discard_block_size = |
2475 | calculate_discard_block_size(cache->sectors_per_block, | |
2476 | cache->origin_sectors); | |
2572629a JT |
2477 | cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors, |
2478 | cache->discard_block_size)); | |
1bad9bc4 | 2479 | cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2480 | if (!cache->discard_bitset) { |
2481 | *error = "could not allocate discard bitset"; | |
2482 | goto bad; | |
2483 | } | |
1bad9bc4 | 2484 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2485 | |
2486 | cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); | |
2487 | if (IS_ERR(cache->copier)) { | |
2488 | *error = "could not create kcopyd client"; | |
2489 | r = PTR_ERR(cache->copier); | |
2490 | goto bad; | |
2491 | } | |
2492 | ||
b29d4986 | 2493 | cache->wq = alloc_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM, 0); |
c6b4fcba JT |
2494 | if (!cache->wq) { |
2495 | *error = "could not create workqueue for metadata object"; | |
2496 | goto bad; | |
2497 | } | |
b29d4986 | 2498 | INIT_WORK(&cache->deferred_bio_worker, process_deferred_bios); |
b29d4986 | 2499 | INIT_WORK(&cache->migration_worker, check_migrations); |
c6b4fcba | 2500 | INIT_DELAYED_WORK(&cache->waker, do_waker); |
c6b4fcba | 2501 | |
b29d4986 | 2502 | cache->prison = dm_bio_prison_create_v2(cache->wq); |
c6b4fcba JT |
2503 | if (!cache->prison) { |
2504 | *error = "could not create bio prison"; | |
2505 | goto bad; | |
2506 | } | |
2507 | ||
6f1c819c KO |
2508 | r = mempool_init_slab_pool(&cache->migration_pool, MIGRATION_POOL_SIZE, |
2509 | migration_cache); | |
2510 | if (r) { | |
c6b4fcba JT |
2511 | *error = "Error creating cache's migration mempool"; |
2512 | goto bad; | |
2513 | } | |
2514 | ||
c6b4fcba JT |
2515 | cache->need_tick_bio = true; |
2516 | cache->sized = false; | |
65790ff9 | 2517 | cache->invalidate = false; |
c6b4fcba JT |
2518 | cache->commit_requested = false; |
2519 | cache->loaded_mappings = false; | |
2520 | cache->loaded_discards = false; | |
2521 | ||
2522 | load_stats(cache); | |
2523 | ||
2524 | atomic_set(&cache->stats.demotion, 0); | |
2525 | atomic_set(&cache->stats.promotion, 0); | |
2526 | atomic_set(&cache->stats.copies_avoided, 0); | |
2527 | atomic_set(&cache->stats.cache_cell_clash, 0); | |
2528 | atomic_set(&cache->stats.commit_count, 0); | |
2529 | atomic_set(&cache->stats.discard_count, 0); | |
2530 | ||
65790ff9 JT |
2531 | spin_lock_init(&cache->invalidation_lock); |
2532 | INIT_LIST_HEAD(&cache->invalidation_requests); | |
2533 | ||
b29d4986 JT |
2534 | batcher_init(&cache->committer, commit_op, cache, |
2535 | issue_op, cache, cache->wq); | |
dc4fa29f | 2536 | dm_iot_init(&cache->tracker); |
066dbaa3 | 2537 | |
b29d4986 JT |
2538 | init_rwsem(&cache->background_work_lock); |
2539 | prevent_background_work(cache); | |
2540 | ||
c6b4fcba JT |
2541 | *result = cache; |
2542 | return 0; | |
c6b4fcba JT |
2543 | bad: |
2544 | destroy(cache); | |
2545 | return r; | |
2546 | } | |
2547 | ||
2548 | static int copy_ctr_args(struct cache *cache, int argc, const char **argv) | |
2549 | { | |
2550 | unsigned i; | |
2551 | const char **copy; | |
2552 | ||
2553 | copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL); | |
2554 | if (!copy) | |
2555 | return -ENOMEM; | |
2556 | for (i = 0; i < argc; i++) { | |
2557 | copy[i] = kstrdup(argv[i], GFP_KERNEL); | |
2558 | if (!copy[i]) { | |
2559 | while (i--) | |
2560 | kfree(copy[i]); | |
2561 | kfree(copy); | |
2562 | return -ENOMEM; | |
2563 | } | |
2564 | } | |
2565 | ||
2566 | cache->nr_ctr_args = argc; | |
2567 | cache->ctr_args = copy; | |
2568 | ||
2569 | return 0; | |
2570 | } | |
2571 | ||
2572 | static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2573 | { | |
2574 | int r = -EINVAL; | |
2575 | struct cache_args *ca; | |
2576 | struct cache *cache = NULL; | |
2577 | ||
2578 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
2579 | if (!ca) { | |
2580 | ti->error = "Error allocating memory for cache"; | |
2581 | return -ENOMEM; | |
2582 | } | |
2583 | ca->ti = ti; | |
2584 | ||
2585 | r = parse_cache_args(ca, argc, argv, &ti->error); | |
2586 | if (r) | |
2587 | goto out; | |
2588 | ||
2589 | r = cache_create(ca, &cache); | |
617a0b89 HM |
2590 | if (r) |
2591 | goto out; | |
c6b4fcba JT |
2592 | |
2593 | r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3); | |
2594 | if (r) { | |
2595 | destroy(cache); | |
2596 | goto out; | |
2597 | } | |
2598 | ||
2599 | ti->private = cache; | |
c6b4fcba JT |
2600 | out: |
2601 | destroy_cache_args(ca); | |
2602 | return r; | |
2603 | } | |
2604 | ||
651f5fa2 JT |
2605 | /*----------------------------------------------------------------*/ |
2606 | ||
2607 | static int cache_map(struct dm_target *ti, struct bio *bio) | |
c6b4fcba | 2608 | { |
651f5fa2 JT |
2609 | struct cache *cache = ti->private; |
2610 | ||
c6b4fcba | 2611 | int r; |
b29d4986 | 2612 | bool commit_needed; |
c6b4fcba | 2613 | dm_oblock_t block = get_bio_block(cache, bio); |
c6b4fcba | 2614 | |
693b960e | 2615 | init_per_bio_data(bio); |
e893fba9 | 2616 | if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) { |
c6b4fcba JT |
2617 | /* |
2618 | * This can only occur if the io goes to a partial block at | |
2619 | * the end of the origin device. We don't cache these. | |
2620 | * Just remap to the origin and carry on. | |
2621 | */ | |
e893fba9 | 2622 | remap_to_origin(cache, bio); |
651f5fa2 | 2623 | accounted_begin(cache, bio); |
c6b4fcba JT |
2624 | return DM_MAPIO_REMAPPED; |
2625 | } | |
2626 | ||
651f5fa2 | 2627 | if (discard_or_flush(bio)) { |
c6b4fcba JT |
2628 | defer_bio(cache, bio); |
2629 | return DM_MAPIO_SUBMITTED; | |
2630 | } | |
2631 | ||
b29d4986 JT |
2632 | r = map_bio(cache, bio, block, &commit_needed); |
2633 | if (commit_needed) | |
2634 | schedule_commit(&cache->committer); | |
c6b4fcba | 2635 | |
2ee57d58 | 2636 | return r; |
c6b4fcba JT |
2637 | } |
2638 | ||
693b960e | 2639 | static int cache_end_io(struct dm_target *ti, struct bio *bio, blk_status_t *error) |
c6b4fcba JT |
2640 | { |
2641 | struct cache *cache = ti->private; | |
2642 | unsigned long flags; | |
693b960e | 2643 | struct per_bio_data *pb = get_per_bio_data(bio); |
c6b4fcba JT |
2644 | |
2645 | if (pb->tick) { | |
fba10109 | 2646 | policy_tick(cache->policy, false); |
c6b4fcba JT |
2647 | |
2648 | spin_lock_irqsave(&cache->lock, flags); | |
2649 | cache->need_tick_bio = true; | |
2650 | spin_unlock_irqrestore(&cache->lock, flags); | |
2651 | } | |
2652 | ||
b29d4986 | 2653 | bio_drop_shared_lock(cache, bio); |
066dbaa3 | 2654 | accounted_complete(cache, bio); |
c6b4fcba | 2655 | |
1be56909 | 2656 | return DM_ENDIO_DONE; |
c6b4fcba JT |
2657 | } |
2658 | ||
2659 | static int write_dirty_bitset(struct cache *cache) | |
2660 | { | |
629d0a8a | 2661 | int r; |
c6b4fcba | 2662 | |
028ae9f7 JT |
2663 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
2664 | return -EINVAL; | |
2665 | ||
629d0a8a JT |
2666 | r = dm_cache_set_dirty_bits(cache->cmd, from_cblock(cache->cache_size), cache->dirty_bitset); |
2667 | if (r) | |
2668 | metadata_operation_failed(cache, "dm_cache_set_dirty_bits", r); | |
c6b4fcba | 2669 | |
629d0a8a | 2670 | return r; |
c6b4fcba JT |
2671 | } |
2672 | ||
2673 | static int write_discard_bitset(struct cache *cache) | |
2674 | { | |
2675 | unsigned i, r; | |
2676 | ||
028ae9f7 JT |
2677 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
2678 | return -EINVAL; | |
2679 | ||
1bad9bc4 JT |
2680 | r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size, |
2681 | cache->discard_nr_blocks); | |
c6b4fcba | 2682 | if (r) { |
b61d9509 | 2683 | DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache)); |
028ae9f7 | 2684 | metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r); |
c6b4fcba JT |
2685 | return r; |
2686 | } | |
2687 | ||
1bad9bc4 JT |
2688 | for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) { |
2689 | r = dm_cache_set_discard(cache->cmd, to_dblock(i), | |
2690 | is_discarded(cache, to_dblock(i))); | |
028ae9f7 JT |
2691 | if (r) { |
2692 | metadata_operation_failed(cache, "dm_cache_set_discard", r); | |
c6b4fcba | 2693 | return r; |
028ae9f7 JT |
2694 | } |
2695 | } | |
2696 | ||
2697 | return 0; | |
2698 | } | |
2699 | ||
2700 | static int write_hints(struct cache *cache) | |
2701 | { | |
2702 | int r; | |
2703 | ||
2704 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
2705 | return -EINVAL; | |
2706 | ||
2707 | r = dm_cache_write_hints(cache->cmd, cache->policy); | |
2708 | if (r) { | |
2709 | metadata_operation_failed(cache, "dm_cache_write_hints", r); | |
2710 | return r; | |
c6b4fcba JT |
2711 | } |
2712 | ||
2713 | return 0; | |
2714 | } | |
2715 | ||
c6b4fcba JT |
2716 | /* |
2717 | * returns true on success | |
2718 | */ | |
2719 | static bool sync_metadata(struct cache *cache) | |
2720 | { | |
2721 | int r1, r2, r3, r4; | |
2722 | ||
2723 | r1 = write_dirty_bitset(cache); | |
2724 | if (r1) | |
b61d9509 | 2725 | DMERR("%s: could not write dirty bitset", cache_device_name(cache)); |
c6b4fcba JT |
2726 | |
2727 | r2 = write_discard_bitset(cache); | |
2728 | if (r2) | |
b61d9509 | 2729 | DMERR("%s: could not write discard bitset", cache_device_name(cache)); |
c6b4fcba JT |
2730 | |
2731 | save_stats(cache); | |
2732 | ||
028ae9f7 | 2733 | r3 = write_hints(cache); |
c6b4fcba | 2734 | if (r3) |
b61d9509 | 2735 | DMERR("%s: could not write hints", cache_device_name(cache)); |
c6b4fcba JT |
2736 | |
2737 | /* | |
2738 | * If writing the above metadata failed, we still commit, but don't | |
2739 | * set the clean shutdown flag. This will effectively force every | |
2740 | * dirty bit to be set on reload. | |
2741 | */ | |
028ae9f7 | 2742 | r4 = commit(cache, !r1 && !r2 && !r3); |
c6b4fcba | 2743 | if (r4) |
b61d9509 | 2744 | DMERR("%s: could not write cache metadata", cache_device_name(cache)); |
c6b4fcba JT |
2745 | |
2746 | return !r1 && !r2 && !r3 && !r4; | |
2747 | } | |
2748 | ||
2749 | static void cache_postsuspend(struct dm_target *ti) | |
2750 | { | |
2751 | struct cache *cache = ti->private; | |
2752 | ||
b29d4986 JT |
2753 | prevent_background_work(cache); |
2754 | BUG_ON(atomic_read(&cache->nr_io_migrations)); | |
2755 | ||
7cdf6a0a MP |
2756 | cancel_delayed_work_sync(&cache->waker); |
2757 | drain_workqueue(cache->wq); | |
701e03e4 | 2758 | WARN_ON(cache->tracker.in_flight); |
b29d4986 JT |
2759 | |
2760 | /* | |
2761 | * If it's a flush suspend there won't be any deferred bios, so this | |
2762 | * call is harmless. | |
2763 | */ | |
651f5fa2 | 2764 | requeue_deferred_bios(cache); |
c6b4fcba | 2765 | |
028ae9f7 JT |
2766 | if (get_cache_mode(cache) == CM_WRITE) |
2767 | (void) sync_metadata(cache); | |
c6b4fcba JT |
2768 | } |
2769 | ||
2770 | static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock, | |
2771 | bool dirty, uint32_t hint, bool hint_valid) | |
2772 | { | |
c6b4fcba JT |
2773 | struct cache *cache = context; |
2774 | ||
449b668c JT |
2775 | if (dirty) { |
2776 | set_bit(from_cblock(cblock), cache->dirty_bitset); | |
2777 | atomic_inc(&cache->nr_dirty); | |
2778 | } else | |
2779 | clear_bit(from_cblock(cblock), cache->dirty_bitset); | |
2780 | ||
b7770923 | 2781 | return policy_load_mapping(cache->policy, oblock, cblock, dirty, hint, hint_valid); |
c6b4fcba JT |
2782 | } |
2783 | ||
3e2e1c30 JT |
2784 | /* |
2785 | * The discard block size in the on disk metadata is not | |
2786 | * neccessarily the same as we're currently using. So we have to | |
2787 | * be careful to only set the discarded attribute if we know it | |
2788 | * covers a complete block of the new size. | |
2789 | */ | |
2790 | struct discard_load_info { | |
2791 | struct cache *cache; | |
2792 | ||
2793 | /* | |
2794 | * These blocks are sized using the on disk dblock size, rather | |
2795 | * than the current one. | |
2796 | */ | |
2797 | dm_block_t block_size; | |
2798 | dm_block_t discard_begin, discard_end; | |
2799 | }; | |
2800 | ||
2801 | static void discard_load_info_init(struct cache *cache, | |
2802 | struct discard_load_info *li) | |
2803 | { | |
2804 | li->cache = cache; | |
2805 | li->discard_begin = li->discard_end = 0; | |
2806 | } | |
2807 | ||
2808 | static void set_discard_range(struct discard_load_info *li) | |
2809 | { | |
2810 | sector_t b, e; | |
2811 | ||
2812 | if (li->discard_begin == li->discard_end) | |
2813 | return; | |
2814 | ||
2815 | /* | |
2816 | * Convert to sectors. | |
2817 | */ | |
2818 | b = li->discard_begin * li->block_size; | |
2819 | e = li->discard_end * li->block_size; | |
2820 | ||
2821 | /* | |
2822 | * Then convert back to the current dblock size. | |
2823 | */ | |
2824 | b = dm_sector_div_up(b, li->cache->discard_block_size); | |
2825 | sector_div(e, li->cache->discard_block_size); | |
2826 | ||
2827 | /* | |
2828 | * The origin may have shrunk, so we need to check we're still in | |
2829 | * bounds. | |
2830 | */ | |
2831 | if (e > from_dblock(li->cache->discard_nr_blocks)) | |
2832 | e = from_dblock(li->cache->discard_nr_blocks); | |
2833 | ||
2834 | for (; b < e; b++) | |
2835 | set_discard(li->cache, to_dblock(b)); | |
2836 | } | |
2837 | ||
c6b4fcba | 2838 | static int load_discard(void *context, sector_t discard_block_size, |
1bad9bc4 | 2839 | dm_dblock_t dblock, bool discard) |
c6b4fcba | 2840 | { |
3e2e1c30 | 2841 | struct discard_load_info *li = context; |
c6b4fcba | 2842 | |
3e2e1c30 | 2843 | li->block_size = discard_block_size; |
1bad9bc4 | 2844 | |
3e2e1c30 JT |
2845 | if (discard) { |
2846 | if (from_dblock(dblock) == li->discard_end) | |
2847 | /* | |
2848 | * We're already in a discard range, just extend it. | |
2849 | */ | |
2850 | li->discard_end = li->discard_end + 1ULL; | |
2851 | ||
2852 | else { | |
2853 | /* | |
2854 | * Emit the old range and start a new one. | |
2855 | */ | |
2856 | set_discard_range(li); | |
2857 | li->discard_begin = from_dblock(dblock); | |
2858 | li->discard_end = li->discard_begin + 1ULL; | |
2859 | } | |
2860 | } else { | |
2861 | set_discard_range(li); | |
2862 | li->discard_begin = li->discard_end = 0; | |
2863 | } | |
c6b4fcba JT |
2864 | |
2865 | return 0; | |
2866 | } | |
2867 | ||
f494a9c6 JT |
2868 | static dm_cblock_t get_cache_dev_size(struct cache *cache) |
2869 | { | |
2870 | sector_t size = get_dev_size(cache->cache_dev); | |
2871 | (void) sector_div(size, cache->sectors_per_block); | |
2872 | return to_cblock(size); | |
2873 | } | |
2874 | ||
2875 | static bool can_resize(struct cache *cache, dm_cblock_t new_size) | |
2876 | { | |
5d07384a MS |
2877 | if (from_cblock(new_size) > from_cblock(cache->cache_size)) { |
2878 | if (cache->sized) { | |
2879 | DMERR("%s: unable to extend cache due to missing cache table reload", | |
2880 | cache_device_name(cache)); | |
2881 | return false; | |
2882 | } | |
2883 | } | |
f494a9c6 JT |
2884 | |
2885 | /* | |
2886 | * We can't drop a dirty block when shrinking the cache. | |
2887 | */ | |
2888 | while (from_cblock(new_size) < from_cblock(cache->cache_size)) { | |
2889 | new_size = to_cblock(from_cblock(new_size) + 1); | |
2890 | if (is_dirty(cache, new_size)) { | |
b61d9509 MS |
2891 | DMERR("%s: unable to shrink cache; cache block %llu is dirty", |
2892 | cache_device_name(cache), | |
f494a9c6 JT |
2893 | (unsigned long long) from_cblock(new_size)); |
2894 | return false; | |
2895 | } | |
2896 | } | |
2897 | ||
2898 | return true; | |
2899 | } | |
2900 | ||
2901 | static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size) | |
2902 | { | |
2903 | int r; | |
2904 | ||
08844800 | 2905 | r = dm_cache_resize(cache->cmd, new_size); |
f494a9c6 | 2906 | if (r) { |
b61d9509 | 2907 | DMERR("%s: could not resize cache metadata", cache_device_name(cache)); |
028ae9f7 | 2908 | metadata_operation_failed(cache, "dm_cache_resize", r); |
f494a9c6 JT |
2909 | return r; |
2910 | } | |
2911 | ||
d1d9220c | 2912 | set_cache_size(cache, new_size); |
f494a9c6 JT |
2913 | |
2914 | return 0; | |
2915 | } | |
2916 | ||
c6b4fcba JT |
2917 | static int cache_preresume(struct dm_target *ti) |
2918 | { | |
2919 | int r = 0; | |
2920 | struct cache *cache = ti->private; | |
f494a9c6 | 2921 | dm_cblock_t csize = get_cache_dev_size(cache); |
c6b4fcba JT |
2922 | |
2923 | /* | |
2924 | * Check to see if the cache has resized. | |
2925 | */ | |
f494a9c6 JT |
2926 | if (!cache->sized) { |
2927 | r = resize_cache_dev(cache, csize); | |
2928 | if (r) | |
c6b4fcba | 2929 | return r; |
c6b4fcba JT |
2930 | |
2931 | cache->sized = true; | |
f494a9c6 JT |
2932 | |
2933 | } else if (csize != cache->cache_size) { | |
2934 | if (!can_resize(cache, csize)) | |
2935 | return -EINVAL; | |
2936 | ||
2937 | r = resize_cache_dev(cache, csize); | |
2938 | if (r) | |
2939 | return r; | |
c6b4fcba JT |
2940 | } |
2941 | ||
2942 | if (!cache->loaded_mappings) { | |
ea2dd8c1 | 2943 | r = dm_cache_load_mappings(cache->cmd, cache->policy, |
c6b4fcba JT |
2944 | load_mapping, cache); |
2945 | if (r) { | |
b61d9509 | 2946 | DMERR("%s: could not load cache mappings", cache_device_name(cache)); |
028ae9f7 | 2947 | metadata_operation_failed(cache, "dm_cache_load_mappings", r); |
c6b4fcba JT |
2948 | return r; |
2949 | } | |
2950 | ||
2951 | cache->loaded_mappings = true; | |
2952 | } | |
2953 | ||
2954 | if (!cache->loaded_discards) { | |
3e2e1c30 JT |
2955 | struct discard_load_info li; |
2956 | ||
2957 | /* | |
2958 | * The discard bitset could have been resized, or the | |
2959 | * discard block size changed. To be safe we start by | |
2960 | * setting every dblock to not discarded. | |
2961 | */ | |
2962 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); | |
2963 | ||
2964 | discard_load_info_init(cache, &li); | |
2965 | r = dm_cache_load_discards(cache->cmd, load_discard, &li); | |
c6b4fcba | 2966 | if (r) { |
b61d9509 | 2967 | DMERR("%s: could not load origin discards", cache_device_name(cache)); |
028ae9f7 | 2968 | metadata_operation_failed(cache, "dm_cache_load_discards", r); |
c6b4fcba JT |
2969 | return r; |
2970 | } | |
3e2e1c30 | 2971 | set_discard_range(&li); |
c6b4fcba JT |
2972 | |
2973 | cache->loaded_discards = true; | |
2974 | } | |
2975 | ||
2976 | return r; | |
2977 | } | |
2978 | ||
2979 | static void cache_resume(struct dm_target *ti) | |
2980 | { | |
2981 | struct cache *cache = ti->private; | |
2982 | ||
2983 | cache->need_tick_bio = true; | |
b29d4986 | 2984 | allow_background_work(cache); |
c6b4fcba JT |
2985 | do_waker(&cache->waker.work); |
2986 | } | |
2987 | ||
de7180ff MS |
2988 | static void emit_flags(struct cache *cache, char *result, |
2989 | unsigned maxlen, ssize_t *sz_ptr) | |
2990 | { | |
2991 | ssize_t sz = *sz_ptr; | |
2992 | struct cache_features *cf = &cache->features; | |
2993 | unsigned count = (cf->metadata_version == 2) + !cf->discard_passdown + 1; | |
2994 | ||
2995 | DMEMIT("%u ", count); | |
2996 | ||
2997 | if (cf->metadata_version == 2) | |
2998 | DMEMIT("metadata2 "); | |
2999 | ||
3000 | if (writethrough_mode(cache)) | |
3001 | DMEMIT("writethrough "); | |
3002 | ||
3003 | else if (passthrough_mode(cache)) | |
3004 | DMEMIT("passthrough "); | |
3005 | ||
3006 | else if (writeback_mode(cache)) | |
3007 | DMEMIT("writeback "); | |
3008 | ||
3009 | else { | |
3010 | DMEMIT("unknown "); | |
3011 | DMERR("%s: internal error: unknown io mode: %d", | |
3012 | cache_device_name(cache), (int) cf->io_mode); | |
3013 | } | |
3014 | ||
3015 | if (!cf->discard_passdown) | |
3016 | DMEMIT("no_discard_passdown "); | |
3017 | ||
3018 | *sz_ptr = sz; | |
3019 | } | |
3020 | ||
c6b4fcba JT |
3021 | /* |
3022 | * Status format: | |
3023 | * | |
6a388618 MS |
3024 | * <metadata block size> <#used metadata blocks>/<#total metadata blocks> |
3025 | * <cache block size> <#used cache blocks>/<#total cache blocks> | |
c6b4fcba | 3026 | * <#read hits> <#read misses> <#write hits> <#write misses> |
6a388618 | 3027 | * <#demotions> <#promotions> <#dirty> |
c6b4fcba JT |
3028 | * <#features> <features>* |
3029 | * <#core args> <core args> | |
255eac20 | 3030 | * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check> |
c6b4fcba JT |
3031 | */ |
3032 | static void cache_status(struct dm_target *ti, status_type_t type, | |
3033 | unsigned status_flags, char *result, unsigned maxlen) | |
3034 | { | |
3035 | int r = 0; | |
3036 | unsigned i; | |
3037 | ssize_t sz = 0; | |
3038 | dm_block_t nr_free_blocks_metadata = 0; | |
3039 | dm_block_t nr_blocks_metadata = 0; | |
3040 | char buf[BDEVNAME_SIZE]; | |
3041 | struct cache *cache = ti->private; | |
3042 | dm_cblock_t residency; | |
d14fcf3d | 3043 | bool needs_check; |
c6b4fcba JT |
3044 | |
3045 | switch (type) { | |
3046 | case STATUSTYPE_INFO: | |
028ae9f7 JT |
3047 | if (get_cache_mode(cache) == CM_FAIL) { |
3048 | DMEMIT("Fail"); | |
3049 | break; | |
c6b4fcba JT |
3050 | } |
3051 | ||
028ae9f7 JT |
3052 | /* Commit to ensure statistics aren't out-of-date */ |
3053 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
3054 | (void) commit(cache, false); | |
3055 | ||
b61d9509 | 3056 | r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata); |
c6b4fcba | 3057 | if (r) { |
b61d9509 MS |
3058 | DMERR("%s: dm_cache_get_free_metadata_block_count returned %d", |
3059 | cache_device_name(cache), r); | |
c6b4fcba JT |
3060 | goto err; |
3061 | } | |
3062 | ||
3063 | r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata); | |
3064 | if (r) { | |
b61d9509 MS |
3065 | DMERR("%s: dm_cache_get_metadata_dev_size returned %d", |
3066 | cache_device_name(cache), r); | |
c6b4fcba JT |
3067 | goto err; |
3068 | } | |
3069 | ||
3070 | residency = policy_residency(cache->policy); | |
3071 | ||
ca763d0a | 3072 | DMEMIT("%u %llu/%llu %llu %llu/%llu %u %u %u %u %u %u %lu ", |
895b47d7 | 3073 | (unsigned)DM_CACHE_METADATA_BLOCK_SIZE, |
c6b4fcba JT |
3074 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
3075 | (unsigned long long)nr_blocks_metadata, | |
ca763d0a | 3076 | (unsigned long long)cache->sectors_per_block, |
6a388618 MS |
3077 | (unsigned long long) from_cblock(residency), |
3078 | (unsigned long long) from_cblock(cache->cache_size), | |
c6b4fcba JT |
3079 | (unsigned) atomic_read(&cache->stats.read_hit), |
3080 | (unsigned) atomic_read(&cache->stats.read_miss), | |
3081 | (unsigned) atomic_read(&cache->stats.write_hit), | |
3082 | (unsigned) atomic_read(&cache->stats.write_miss), | |
3083 | (unsigned) atomic_read(&cache->stats.demotion), | |
3084 | (unsigned) atomic_read(&cache->stats.promotion), | |
44fa816b | 3085 | (unsigned long) atomic_read(&cache->nr_dirty)); |
c6b4fcba | 3086 | |
de7180ff | 3087 | emit_flags(cache, result, maxlen, &sz); |
c6b4fcba JT |
3088 | |
3089 | DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold); | |
2e68c4e6 MS |
3090 | |
3091 | DMEMIT("%s ", dm_cache_policy_get_name(cache->policy)); | |
c6b4fcba | 3092 | if (sz < maxlen) { |
028ae9f7 | 3093 | r = policy_emit_config_values(cache->policy, result, maxlen, &sz); |
c6b4fcba | 3094 | if (r) |
b61d9509 MS |
3095 | DMERR("%s: policy_emit_config_values returned %d", |
3096 | cache_device_name(cache), r); | |
c6b4fcba JT |
3097 | } |
3098 | ||
028ae9f7 JT |
3099 | if (get_cache_mode(cache) == CM_READ_ONLY) |
3100 | DMEMIT("ro "); | |
3101 | else | |
3102 | DMEMIT("rw "); | |
3103 | ||
d14fcf3d JT |
3104 | r = dm_cache_metadata_needs_check(cache->cmd, &needs_check); |
3105 | ||
3106 | if (r || needs_check) | |
255eac20 MS |
3107 | DMEMIT("needs_check "); |
3108 | else | |
3109 | DMEMIT("- "); | |
3110 | ||
c6b4fcba JT |
3111 | break; |
3112 | ||
3113 | case STATUSTYPE_TABLE: | |
3114 | format_dev_t(buf, cache->metadata_dev->bdev->bd_dev); | |
3115 | DMEMIT("%s ", buf); | |
3116 | format_dev_t(buf, cache->cache_dev->bdev->bd_dev); | |
3117 | DMEMIT("%s ", buf); | |
3118 | format_dev_t(buf, cache->origin_dev->bdev->bd_dev); | |
3119 | DMEMIT("%s", buf); | |
3120 | ||
3121 | for (i = 0; i < cache->nr_ctr_args - 1; i++) | |
3122 | DMEMIT(" %s", cache->ctr_args[i]); | |
3123 | if (cache->nr_ctr_args) | |
3124 | DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]); | |
8ec45662 TS |
3125 | break; |
3126 | ||
3127 | case STATUSTYPE_IMA: | |
3128 | DMEMIT_TARGET_NAME_VERSION(ti->type); | |
3129 | if (get_cache_mode(cache) == CM_FAIL) | |
3130 | DMEMIT(",metadata_mode=fail"); | |
3131 | else if (get_cache_mode(cache) == CM_READ_ONLY) | |
3132 | DMEMIT(",metadata_mode=ro"); | |
3133 | else | |
3134 | DMEMIT(",metadata_mode=rw"); | |
3135 | ||
3136 | format_dev_t(buf, cache->metadata_dev->bdev->bd_dev); | |
3137 | DMEMIT(",cache_metadata_device=%s", buf); | |
3138 | format_dev_t(buf, cache->cache_dev->bdev->bd_dev); | |
3139 | DMEMIT(",cache_device=%s", buf); | |
3140 | format_dev_t(buf, cache->origin_dev->bdev->bd_dev); | |
3141 | DMEMIT(",cache_origin_device=%s", buf); | |
3142 | DMEMIT(",writethrough=%c", writethrough_mode(cache) ? 'y' : 'n'); | |
3143 | DMEMIT(",writeback=%c", writeback_mode(cache) ? 'y' : 'n'); | |
3144 | DMEMIT(",passthrough=%c", passthrough_mode(cache) ? 'y' : 'n'); | |
3145 | DMEMIT(",metadata2=%c", cache->features.metadata_version == 2 ? 'y' : 'n'); | |
3146 | DMEMIT(",no_discard_passdown=%c", cache->features.discard_passdown ? 'n' : 'y'); | |
3147 | DMEMIT(";"); | |
3148 | break; | |
c6b4fcba JT |
3149 | } |
3150 | ||
3151 | return; | |
3152 | ||
3153 | err: | |
3154 | DMEMIT("Error"); | |
3155 | } | |
3156 | ||
b29d4986 JT |
3157 | /* |
3158 | * Defines a range of cblocks, begin to (end - 1) are in the range. end is | |
3159 | * the one-past-the-end value. | |
3160 | */ | |
3161 | struct cblock_range { | |
3162 | dm_cblock_t begin; | |
3163 | dm_cblock_t end; | |
3164 | }; | |
3165 | ||
c6b4fcba | 3166 | /* |
65790ff9 JT |
3167 | * A cache block range can take two forms: |
3168 | * | |
3169 | * i) A single cblock, eg. '3456' | |
b29d4986 | 3170 | * ii) A begin and end cblock with a dash between, eg. 123-234 |
65790ff9 JT |
3171 | */ |
3172 | static int parse_cblock_range(struct cache *cache, const char *str, | |
3173 | struct cblock_range *result) | |
3174 | { | |
3175 | char dummy; | |
3176 | uint64_t b, e; | |
3177 | int r; | |
3178 | ||
3179 | /* | |
3180 | * Try and parse form (ii) first. | |
3181 | */ | |
3182 | r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy); | |
3183 | if (r < 0) | |
3184 | return r; | |
3185 | ||
3186 | if (r == 2) { | |
3187 | result->begin = to_cblock(b); | |
3188 | result->end = to_cblock(e); | |
3189 | return 0; | |
3190 | } | |
3191 | ||
3192 | /* | |
3193 | * That didn't work, try form (i). | |
3194 | */ | |
3195 | r = sscanf(str, "%llu%c", &b, &dummy); | |
3196 | if (r < 0) | |
3197 | return r; | |
3198 | ||
3199 | if (r == 1) { | |
3200 | result->begin = to_cblock(b); | |
3201 | result->end = to_cblock(from_cblock(result->begin) + 1u); | |
3202 | return 0; | |
3203 | } | |
3204 | ||
b61d9509 | 3205 | DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str); |
65790ff9 JT |
3206 | return -EINVAL; |
3207 | } | |
3208 | ||
3209 | static int validate_cblock_range(struct cache *cache, struct cblock_range *range) | |
3210 | { | |
3211 | uint64_t b = from_cblock(range->begin); | |
3212 | uint64_t e = from_cblock(range->end); | |
3213 | uint64_t n = from_cblock(cache->cache_size); | |
3214 | ||
3215 | if (b >= n) { | |
b61d9509 MS |
3216 | DMERR("%s: begin cblock out of range: %llu >= %llu", |
3217 | cache_device_name(cache), b, n); | |
65790ff9 JT |
3218 | return -EINVAL; |
3219 | } | |
3220 | ||
3221 | if (e > n) { | |
b61d9509 MS |
3222 | DMERR("%s: end cblock out of range: %llu > %llu", |
3223 | cache_device_name(cache), e, n); | |
65790ff9 JT |
3224 | return -EINVAL; |
3225 | } | |
3226 | ||
3227 | if (b >= e) { | |
b61d9509 MS |
3228 | DMERR("%s: invalid cblock range: %llu >= %llu", |
3229 | cache_device_name(cache), b, e); | |
65790ff9 JT |
3230 | return -EINVAL; |
3231 | } | |
3232 | ||
3233 | return 0; | |
3234 | } | |
3235 | ||
b29d4986 JT |
3236 | static inline dm_cblock_t cblock_succ(dm_cblock_t b) |
3237 | { | |
3238 | return to_cblock(from_cblock(b) + 1); | |
3239 | } | |
3240 | ||
65790ff9 JT |
3241 | static int request_invalidation(struct cache *cache, struct cblock_range *range) |
3242 | { | |
b29d4986 | 3243 | int r = 0; |
65790ff9 | 3244 | |
b29d4986 JT |
3245 | /* |
3246 | * We don't need to do any locking here because we know we're in | |
3247 | * passthrough mode. There's is potential for a race between an | |
3248 | * invalidation triggered by an io and an invalidation message. This | |
3249 | * is harmless, we must not worry if the policy call fails. | |
3250 | */ | |
3251 | while (range->begin != range->end) { | |
3252 | r = invalidate_cblock(cache, range->begin); | |
3253 | if (r) | |
3254 | return r; | |
65790ff9 | 3255 | |
b29d4986 JT |
3256 | range->begin = cblock_succ(range->begin); |
3257 | } | |
65790ff9 | 3258 | |
b29d4986 JT |
3259 | cache->commit_requested = true; |
3260 | return r; | |
65790ff9 JT |
3261 | } |
3262 | ||
3263 | static int process_invalidate_cblocks_message(struct cache *cache, unsigned count, | |
3264 | const char **cblock_ranges) | |
3265 | { | |
3266 | int r = 0; | |
3267 | unsigned i; | |
3268 | struct cblock_range range; | |
3269 | ||
8e3c3827 | 3270 | if (!passthrough_mode(cache)) { |
b61d9509 MS |
3271 | DMERR("%s: cache has to be in passthrough mode for invalidation", |
3272 | cache_device_name(cache)); | |
65790ff9 JT |
3273 | return -EPERM; |
3274 | } | |
3275 | ||
3276 | for (i = 0; i < count; i++) { | |
3277 | r = parse_cblock_range(cache, cblock_ranges[i], &range); | |
3278 | if (r) | |
3279 | break; | |
3280 | ||
3281 | r = validate_cblock_range(cache, &range); | |
3282 | if (r) | |
3283 | break; | |
3284 | ||
3285 | /* | |
3286 | * Pass begin and end origin blocks to the worker and wake it. | |
3287 | */ | |
3288 | r = request_invalidation(cache, &range); | |
3289 | if (r) | |
3290 | break; | |
3291 | } | |
3292 | ||
3293 | return r; | |
3294 | } | |
3295 | ||
3296 | /* | |
3297 | * Supports | |
3298 | * "<key> <value>" | |
3299 | * and | |
3300 | * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]* | |
c6b4fcba JT |
3301 | * |
3302 | * The key migration_threshold is supported by the cache target core. | |
3303 | */ | |
1eb5fa84 MS |
3304 | static int cache_message(struct dm_target *ti, unsigned argc, char **argv, |
3305 | char *result, unsigned maxlen) | |
c6b4fcba | 3306 | { |
c6b4fcba JT |
3307 | struct cache *cache = ti->private; |
3308 | ||
65790ff9 JT |
3309 | if (!argc) |
3310 | return -EINVAL; | |
3311 | ||
028ae9f7 | 3312 | if (get_cache_mode(cache) >= CM_READ_ONLY) { |
b61d9509 MS |
3313 | DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode", |
3314 | cache_device_name(cache)); | |
028ae9f7 JT |
3315 | return -EOPNOTSUPP; |
3316 | } | |
3317 | ||
7b6b2bc9 | 3318 | if (!strcasecmp(argv[0], "invalidate_cblocks")) |
65790ff9 JT |
3319 | return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1); |
3320 | ||
c6b4fcba JT |
3321 | if (argc != 2) |
3322 | return -EINVAL; | |
3323 | ||
2f14f4b5 | 3324 | return set_config_value(cache, argv[0], argv[1]); |
c6b4fcba JT |
3325 | } |
3326 | ||
3327 | static int cache_iterate_devices(struct dm_target *ti, | |
3328 | iterate_devices_callout_fn fn, void *data) | |
3329 | { | |
3330 | int r = 0; | |
3331 | struct cache *cache = ti->private; | |
3332 | ||
3333 | r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data); | |
3334 | if (!r) | |
3335 | r = fn(ti, cache->origin_dev, 0, ti->len, data); | |
3336 | ||
3337 | return r; | |
3338 | } | |
3339 | ||
de7180ff MS |
3340 | static bool origin_dev_supports_discard(struct block_device *origin_bdev) |
3341 | { | |
3342 | struct request_queue *q = bdev_get_queue(origin_bdev); | |
3343 | ||
63508e38 | 3344 | return blk_queue_discard(q); |
de7180ff MS |
3345 | } |
3346 | ||
3347 | /* | |
3348 | * If discard_passdown was enabled verify that the origin device | |
3349 | * supports discards. Disable discard_passdown if not. | |
3350 | */ | |
3351 | static void disable_passdown_if_not_supported(struct cache *cache) | |
3352 | { | |
3353 | struct block_device *origin_bdev = cache->origin_dev->bdev; | |
3354 | struct queue_limits *origin_limits = &bdev_get_queue(origin_bdev)->limits; | |
3355 | const char *reason = NULL; | |
3356 | char buf[BDEVNAME_SIZE]; | |
3357 | ||
3358 | if (!cache->features.discard_passdown) | |
3359 | return; | |
3360 | ||
3361 | if (!origin_dev_supports_discard(origin_bdev)) | |
3362 | reason = "discard unsupported"; | |
3363 | ||
3364 | else if (origin_limits->max_discard_sectors < cache->sectors_per_block) | |
3365 | reason = "max discard sectors smaller than a block"; | |
3366 | ||
3367 | if (reason) { | |
3368 | DMWARN("Origin device (%s) %s: Disabling discard passdown.", | |
3369 | bdevname(origin_bdev, buf), reason); | |
3370 | cache->features.discard_passdown = false; | |
3371 | } | |
3372 | } | |
3373 | ||
c6b4fcba JT |
3374 | static void set_discard_limits(struct cache *cache, struct queue_limits *limits) |
3375 | { | |
de7180ff MS |
3376 | struct block_device *origin_bdev = cache->origin_dev->bdev; |
3377 | struct queue_limits *origin_limits = &bdev_get_queue(origin_bdev)->limits; | |
3378 | ||
3379 | if (!cache->features.discard_passdown) { | |
3380 | /* No passdown is done so setting own virtual limits */ | |
3381 | limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024, | |
3382 | cache->origin_sectors); | |
3383 | limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT; | |
3384 | return; | |
3385 | } | |
3386 | ||
c6b4fcba | 3387 | /* |
de7180ff MS |
3388 | * cache_iterate_devices() is stacking both origin and fast device limits |
3389 | * but discards aren't passed to fast device, so inherit origin's limits. | |
c6b4fcba | 3390 | */ |
de7180ff MS |
3391 | limits->max_discard_sectors = origin_limits->max_discard_sectors; |
3392 | limits->max_hw_discard_sectors = origin_limits->max_hw_discard_sectors; | |
3393 | limits->discard_granularity = origin_limits->discard_granularity; | |
3394 | limits->discard_alignment = origin_limits->discard_alignment; | |
3395 | limits->discard_misaligned = origin_limits->discard_misaligned; | |
c6b4fcba JT |
3396 | } |
3397 | ||
3398 | static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3399 | { | |
3400 | struct cache *cache = ti->private; | |
f6109372 | 3401 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
c6b4fcba | 3402 | |
f6109372 MS |
3403 | /* |
3404 | * If the system-determined stacked limits are compatible with the | |
3405 | * cache's blocksize (io_opt is a factor) do not override them. | |
3406 | */ | |
3407 | if (io_opt_sectors < cache->sectors_per_block || | |
3408 | do_div(io_opt_sectors, cache->sectors_per_block)) { | |
b0246530 | 3409 | blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT); |
f6109372 MS |
3410 | blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT); |
3411 | } | |
de7180ff MS |
3412 | |
3413 | disable_passdown_if_not_supported(cache); | |
c6b4fcba JT |
3414 | set_discard_limits(cache, limits); |
3415 | } | |
3416 | ||
3417 | /*----------------------------------------------------------------*/ | |
3418 | ||
3419 | static struct target_type cache_target = { | |
3420 | .name = "cache", | |
636be424 | 3421 | .version = {2, 2, 0}, |
c6b4fcba JT |
3422 | .module = THIS_MODULE, |
3423 | .ctr = cache_ctr, | |
3424 | .dtr = cache_dtr, | |
3425 | .map = cache_map, | |
3426 | .end_io = cache_end_io, | |
3427 | .postsuspend = cache_postsuspend, | |
3428 | .preresume = cache_preresume, | |
3429 | .resume = cache_resume, | |
3430 | .status = cache_status, | |
3431 | .message = cache_message, | |
3432 | .iterate_devices = cache_iterate_devices, | |
c6b4fcba JT |
3433 | .io_hints = cache_io_hints, |
3434 | }; | |
3435 | ||
3436 | static int __init dm_cache_init(void) | |
3437 | { | |
3438 | int r; | |
3439 | ||
c6b4fcba | 3440 | migration_cache = KMEM_CACHE(dm_cache_migration, 0); |
c7cd5550 | 3441 | if (!migration_cache) |
c6b4fcba | 3442 | return -ENOMEM; |
c6b4fcba | 3443 | |
7e6358d2 | 3444 | r = dm_register_target(&cache_target); |
3445 | if (r) { | |
3446 | DMERR("cache target registration failed: %d", r); | |
c7cd5550 | 3447 | kmem_cache_destroy(migration_cache); |
7e6358d2 | 3448 | return r; |
3449 | } | |
3450 | ||
c6b4fcba JT |
3451 | return 0; |
3452 | } | |
3453 | ||
3454 | static void __exit dm_cache_exit(void) | |
3455 | { | |
3456 | dm_unregister_target(&cache_target); | |
3457 | kmem_cache_destroy(migration_cache); | |
3458 | } | |
3459 | ||
3460 | module_init(dm_cache_init); | |
3461 | module_exit(dm_cache_exit); | |
3462 | ||
3463 | MODULE_DESCRIPTION(DM_NAME " cache target"); | |
3464 | MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>"); | |
3465 | MODULE_LICENSE("GPL"); |