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" | |
8 | #include "dm-bio-prison.h" | |
b844fe69 | 9 | #include "dm-bio-record.h" |
c6b4fcba JT |
10 | #include "dm-cache-metadata.h" |
11 | ||
12 | #include <linux/dm-io.h> | |
13 | #include <linux/dm-kcopyd.h> | |
0f30af98 | 14 | #include <linux/jiffies.h> |
c6b4fcba JT |
15 | #include <linux/init.h> |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/vmalloc.h> | |
20 | ||
21 | #define DM_MSG_PREFIX "cache" | |
22 | ||
23 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(cache_copy_throttle, | |
24 | "A percentage of time allocated for copying to and/or from cache"); | |
25 | ||
26 | /*----------------------------------------------------------------*/ | |
27 | ||
77289d32 JT |
28 | #define IOT_RESOLUTION 4 |
29 | ||
30 | struct io_tracker { | |
31 | spinlock_t lock; | |
32 | ||
33 | /* | |
34 | * Sectors of in-flight IO. | |
35 | */ | |
36 | sector_t in_flight; | |
37 | ||
38 | /* | |
39 | * The time, in jiffies, when this device became idle (if it is | |
40 | * indeed idle). | |
41 | */ | |
42 | unsigned long idle_time; | |
43 | unsigned long last_update_time; | |
44 | }; | |
45 | ||
46 | static void iot_init(struct io_tracker *iot) | |
47 | { | |
48 | spin_lock_init(&iot->lock); | |
49 | iot->in_flight = 0ul; | |
50 | iot->idle_time = 0ul; | |
51 | iot->last_update_time = jiffies; | |
52 | } | |
53 | ||
54 | static bool __iot_idle_for(struct io_tracker *iot, unsigned long jifs) | |
55 | { | |
56 | if (iot->in_flight) | |
57 | return false; | |
58 | ||
59 | return time_after(jiffies, iot->idle_time + jifs); | |
60 | } | |
61 | ||
62 | static bool iot_idle_for(struct io_tracker *iot, unsigned long jifs) | |
63 | { | |
64 | bool r; | |
65 | unsigned long flags; | |
66 | ||
67 | spin_lock_irqsave(&iot->lock, flags); | |
68 | r = __iot_idle_for(iot, jifs); | |
69 | spin_unlock_irqrestore(&iot->lock, flags); | |
70 | ||
71 | return r; | |
72 | } | |
73 | ||
74 | static void iot_io_begin(struct io_tracker *iot, sector_t len) | |
75 | { | |
76 | unsigned long flags; | |
77 | ||
78 | spin_lock_irqsave(&iot->lock, flags); | |
79 | iot->in_flight += len; | |
80 | spin_unlock_irqrestore(&iot->lock, flags); | |
81 | } | |
82 | ||
83 | static void __iot_io_end(struct io_tracker *iot, sector_t len) | |
84 | { | |
85 | iot->in_flight -= len; | |
86 | if (!iot->in_flight) | |
87 | iot->idle_time = jiffies; | |
88 | } | |
89 | ||
90 | static void iot_io_end(struct io_tracker *iot, sector_t len) | |
91 | { | |
92 | unsigned long flags; | |
93 | ||
94 | spin_lock_irqsave(&iot->lock, flags); | |
95 | __iot_io_end(iot, len); | |
96 | spin_unlock_irqrestore(&iot->lock, flags); | |
97 | } | |
98 | ||
99 | /*----------------------------------------------------------------*/ | |
100 | ||
c6b4fcba JT |
101 | /* |
102 | * Glossary: | |
103 | * | |
104 | * oblock: index of an origin block | |
105 | * cblock: index of a cache block | |
106 | * promotion: movement of a block from origin to cache | |
107 | * demotion: movement of a block from cache to origin | |
108 | * migration: movement of a block between the origin and cache device, | |
109 | * either direction | |
110 | */ | |
111 | ||
112 | /*----------------------------------------------------------------*/ | |
113 | ||
c9d28d5d JT |
114 | /* |
115 | * There are a couple of places where we let a bio run, but want to do some | |
116 | * work before calling its endio function. We do this by temporarily | |
117 | * changing the endio fn. | |
118 | */ | |
119 | struct dm_hook_info { | |
120 | bio_end_io_t *bi_end_io; | |
121 | void *bi_private; | |
122 | }; | |
123 | ||
124 | static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio, | |
125 | bio_end_io_t *bi_end_io, void *bi_private) | |
126 | { | |
127 | h->bi_end_io = bio->bi_end_io; | |
128 | h->bi_private = bio->bi_private; | |
129 | ||
130 | bio->bi_end_io = bi_end_io; | |
131 | bio->bi_private = bi_private; | |
132 | } | |
133 | ||
134 | static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio) | |
135 | { | |
136 | bio->bi_end_io = h->bi_end_io; | |
137 | bio->bi_private = h->bi_private; | |
138 | } | |
139 | ||
140 | /*----------------------------------------------------------------*/ | |
141 | ||
c6b4fcba JT |
142 | #define MIGRATION_POOL_SIZE 128 |
143 | #define COMMIT_PERIOD HZ | |
144 | #define MIGRATION_COUNT_WINDOW 10 | |
145 | ||
146 | /* | |
05473044 MS |
147 | * The block size of the device holding cache data must be |
148 | * between 32KB and 1GB. | |
c6b4fcba JT |
149 | */ |
150 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT) | |
05473044 | 151 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
c6b4fcba | 152 | |
2ee57d58 | 153 | enum cache_metadata_mode { |
c6b4fcba JT |
154 | CM_WRITE, /* metadata may be changed */ |
155 | CM_READ_ONLY, /* metadata may not be changed */ | |
028ae9f7 | 156 | CM_FAIL |
c6b4fcba JT |
157 | }; |
158 | ||
2ee57d58 JT |
159 | enum cache_io_mode { |
160 | /* | |
161 | * Data is written to cached blocks only. These blocks are marked | |
162 | * dirty. If you lose the cache device you will lose data. | |
163 | * Potential performance increase for both reads and writes. | |
164 | */ | |
165 | CM_IO_WRITEBACK, | |
166 | ||
167 | /* | |
168 | * Data is written to both cache and origin. Blocks are never | |
169 | * dirty. Potential performance benfit for reads only. | |
170 | */ | |
171 | CM_IO_WRITETHROUGH, | |
172 | ||
173 | /* | |
174 | * A degraded mode useful for various cache coherency situations | |
175 | * (eg, rolling back snapshots). Reads and writes always go to the | |
176 | * origin. If a write goes to a cached oblock, then the cache | |
177 | * block is invalidated. | |
178 | */ | |
179 | CM_IO_PASSTHROUGH | |
180 | }; | |
181 | ||
c6b4fcba | 182 | struct cache_features { |
2ee57d58 JT |
183 | enum cache_metadata_mode mode; |
184 | enum cache_io_mode io_mode; | |
c6b4fcba JT |
185 | }; |
186 | ||
187 | struct cache_stats { | |
188 | atomic_t read_hit; | |
189 | atomic_t read_miss; | |
190 | atomic_t write_hit; | |
191 | atomic_t write_miss; | |
192 | atomic_t demotion; | |
193 | atomic_t promotion; | |
194 | atomic_t copies_avoided; | |
195 | atomic_t cache_cell_clash; | |
196 | atomic_t commit_count; | |
197 | atomic_t discard_count; | |
198 | }; | |
199 | ||
65790ff9 JT |
200 | /* |
201 | * Defines a range of cblocks, begin to (end - 1) are in the range. end is | |
202 | * the one-past-the-end value. | |
203 | */ | |
204 | struct cblock_range { | |
205 | dm_cblock_t begin; | |
206 | dm_cblock_t end; | |
207 | }; | |
208 | ||
209 | struct invalidation_request { | |
210 | struct list_head list; | |
211 | struct cblock_range *cblocks; | |
212 | ||
213 | atomic_t complete; | |
214 | int err; | |
215 | ||
216 | wait_queue_head_t result_wait; | |
217 | }; | |
218 | ||
c6b4fcba JT |
219 | struct cache { |
220 | struct dm_target *ti; | |
221 | struct dm_target_callbacks callbacks; | |
222 | ||
c9ec5d7c MS |
223 | struct dm_cache_metadata *cmd; |
224 | ||
c6b4fcba JT |
225 | /* |
226 | * Metadata is written to this device. | |
227 | */ | |
228 | struct dm_dev *metadata_dev; | |
229 | ||
230 | /* | |
231 | * The slower of the two data devices. Typically a spindle. | |
232 | */ | |
233 | struct dm_dev *origin_dev; | |
234 | ||
235 | /* | |
236 | * The faster of the two data devices. Typically an SSD. | |
237 | */ | |
238 | struct dm_dev *cache_dev; | |
239 | ||
c6b4fcba JT |
240 | /* |
241 | * Size of the origin device in _complete_ blocks and native sectors. | |
242 | */ | |
243 | dm_oblock_t origin_blocks; | |
244 | sector_t origin_sectors; | |
245 | ||
246 | /* | |
247 | * Size of the cache device in blocks. | |
248 | */ | |
249 | dm_cblock_t cache_size; | |
250 | ||
251 | /* | |
252 | * Fields for converting from sectors to blocks. | |
253 | */ | |
254 | uint32_t sectors_per_block; | |
255 | int sectors_per_block_shift; | |
256 | ||
c6b4fcba | 257 | spinlock_t lock; |
651f5fa2 | 258 | struct list_head deferred_cells; |
c6b4fcba JT |
259 | struct bio_list deferred_bios; |
260 | struct bio_list deferred_flush_bios; | |
e2e74d61 | 261 | struct bio_list deferred_writethrough_bios; |
c6b4fcba JT |
262 | struct list_head quiesced_migrations; |
263 | struct list_head completed_migrations; | |
264 | struct list_head need_commit_migrations; | |
265 | sector_t migration_threshold; | |
c6b4fcba | 266 | wait_queue_head_t migration_wait; |
a59db676 JT |
267 | atomic_t nr_allocated_migrations; |
268 | ||
269 | /* | |
270 | * The number of in flight migrations that are performing | |
271 | * background io. eg, promotion, writeback. | |
272 | */ | |
273 | atomic_t nr_io_migrations; | |
c6b4fcba | 274 | |
66cb1910 | 275 | wait_queue_head_t quiescing_wait; |
238f8363 | 276 | atomic_t quiescing; |
66cb1910 JT |
277 | atomic_t quiescing_ack; |
278 | ||
c6b4fcba JT |
279 | /* |
280 | * cache_size entries, dirty if set | |
281 | */ | |
44fa816b | 282 | atomic_t nr_dirty; |
c6b4fcba JT |
283 | unsigned long *dirty_bitset; |
284 | ||
285 | /* | |
286 | * origin_blocks entries, discarded if set. | |
287 | */ | |
1bad9bc4 | 288 | dm_dblock_t discard_nr_blocks; |
c6b4fcba | 289 | unsigned long *discard_bitset; |
08b18451 | 290 | uint32_t discard_block_size; /* a power of 2 times sectors per block */ |
c9ec5d7c MS |
291 | |
292 | /* | |
293 | * Rather than reconstructing the table line for the status we just | |
294 | * save it and regurgitate. | |
295 | */ | |
296 | unsigned nr_ctr_args; | |
297 | const char **ctr_args; | |
c6b4fcba JT |
298 | |
299 | struct dm_kcopyd_client *copier; | |
300 | struct workqueue_struct *wq; | |
301 | struct work_struct worker; | |
302 | ||
303 | struct delayed_work waker; | |
304 | unsigned long last_commit_jiffies; | |
305 | ||
306 | struct dm_bio_prison *prison; | |
307 | struct dm_deferred_set *all_io_ds; | |
308 | ||
309 | mempool_t *migration_pool; | |
c6b4fcba JT |
310 | |
311 | struct dm_cache_policy *policy; | |
312 | unsigned policy_nr_args; | |
313 | ||
314 | bool need_tick_bio:1; | |
315 | bool sized:1; | |
65790ff9 | 316 | bool invalidate:1; |
c6b4fcba JT |
317 | bool commit_requested:1; |
318 | bool loaded_mappings:1; | |
319 | bool loaded_discards:1; | |
320 | ||
c6b4fcba | 321 | /* |
c9ec5d7c | 322 | * Cache features such as write-through. |
c6b4fcba | 323 | */ |
c9ec5d7c MS |
324 | struct cache_features features; |
325 | ||
326 | struct cache_stats stats; | |
65790ff9 JT |
327 | |
328 | /* | |
329 | * Invalidation fields. | |
330 | */ | |
331 | spinlock_t invalidation_lock; | |
332 | struct list_head invalidation_requests; | |
066dbaa3 JT |
333 | |
334 | struct io_tracker origin_tracker; | |
c6b4fcba JT |
335 | }; |
336 | ||
337 | struct per_bio_data { | |
338 | bool tick:1; | |
339 | unsigned req_nr:2; | |
340 | struct dm_deferred_entry *all_io_entry; | |
c6eda5e8 | 341 | struct dm_hook_info hook_info; |
066dbaa3 | 342 | sector_t len; |
e2e74d61 | 343 | |
19b0092e MS |
344 | /* |
345 | * writethrough fields. These MUST remain at the end of this | |
346 | * structure and the 'cache' member must be the first as it | |
aeed1420 | 347 | * is used to determine the offset of the writethrough fields. |
19b0092e | 348 | */ |
e2e74d61 JT |
349 | struct cache *cache; |
350 | dm_cblock_t cblock; | |
b844fe69 | 351 | struct dm_bio_details bio_details; |
c6b4fcba JT |
352 | }; |
353 | ||
354 | struct dm_cache_migration { | |
355 | struct list_head list; | |
356 | struct cache *cache; | |
357 | ||
358 | unsigned long start_jiffies; | |
359 | dm_oblock_t old_oblock; | |
360 | dm_oblock_t new_oblock; | |
361 | dm_cblock_t cblock; | |
362 | ||
363 | bool err:1; | |
7ae34e77 | 364 | bool discard:1; |
c6b4fcba JT |
365 | bool writeback:1; |
366 | bool demote:1; | |
367 | bool promote:1; | |
c9d28d5d | 368 | bool requeue_holder:1; |
65790ff9 | 369 | bool invalidate:1; |
c6b4fcba JT |
370 | |
371 | struct dm_bio_prison_cell *old_ocell; | |
372 | struct dm_bio_prison_cell *new_ocell; | |
373 | }; | |
374 | ||
375 | /* | |
376 | * Processing a bio in the worker thread may require these memory | |
377 | * allocations. We prealloc to avoid deadlocks (the same worker thread | |
378 | * frees them back to the mempool). | |
379 | */ | |
380 | struct prealloc { | |
381 | struct dm_cache_migration *mg; | |
382 | struct dm_bio_prison_cell *cell1; | |
383 | struct dm_bio_prison_cell *cell2; | |
384 | }; | |
385 | ||
028ae9f7 JT |
386 | static enum cache_metadata_mode get_cache_mode(struct cache *cache); |
387 | ||
c6b4fcba JT |
388 | static void wake_worker(struct cache *cache) |
389 | { | |
390 | queue_work(cache->wq, &cache->worker); | |
391 | } | |
392 | ||
393 | /*----------------------------------------------------------------*/ | |
394 | ||
395 | static struct dm_bio_prison_cell *alloc_prison_cell(struct cache *cache) | |
396 | { | |
397 | /* FIXME: change to use a local slab. */ | |
398 | return dm_bio_prison_alloc_cell(cache->prison, GFP_NOWAIT); | |
399 | } | |
400 | ||
401 | static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell *cell) | |
402 | { | |
403 | dm_bio_prison_free_cell(cache->prison, cell); | |
404 | } | |
405 | ||
a59db676 JT |
406 | static struct dm_cache_migration *alloc_migration(struct cache *cache) |
407 | { | |
408 | struct dm_cache_migration *mg; | |
409 | ||
410 | mg = mempool_alloc(cache->migration_pool, GFP_NOWAIT); | |
411 | if (mg) { | |
412 | mg->cache = cache; | |
413 | atomic_inc(&mg->cache->nr_allocated_migrations); | |
414 | } | |
415 | ||
416 | return mg; | |
417 | } | |
418 | ||
419 | static void free_migration(struct dm_cache_migration *mg) | |
420 | { | |
88bf5184 | 421 | struct cache *cache = mg->cache; |
a59db676 | 422 | |
88bf5184 JT |
423 | if (atomic_dec_and_test(&cache->nr_allocated_migrations)) |
424 | wake_up(&cache->migration_wait); | |
425 | ||
426 | mempool_free(mg, cache->migration_pool); | |
3508e659 | 427 | wake_worker(cache); |
a59db676 JT |
428 | } |
429 | ||
c6b4fcba JT |
430 | static int prealloc_data_structs(struct cache *cache, struct prealloc *p) |
431 | { | |
432 | if (!p->mg) { | |
a59db676 | 433 | p->mg = alloc_migration(cache); |
c6b4fcba JT |
434 | if (!p->mg) |
435 | return -ENOMEM; | |
436 | } | |
437 | ||
438 | if (!p->cell1) { | |
439 | p->cell1 = alloc_prison_cell(cache); | |
440 | if (!p->cell1) | |
441 | return -ENOMEM; | |
442 | } | |
443 | ||
444 | if (!p->cell2) { | |
445 | p->cell2 = alloc_prison_cell(cache); | |
446 | if (!p->cell2) | |
447 | return -ENOMEM; | |
448 | } | |
449 | ||
450 | return 0; | |
451 | } | |
452 | ||
453 | static void prealloc_free_structs(struct cache *cache, struct prealloc *p) | |
454 | { | |
455 | if (p->cell2) | |
456 | free_prison_cell(cache, p->cell2); | |
457 | ||
458 | if (p->cell1) | |
459 | free_prison_cell(cache, p->cell1); | |
460 | ||
461 | if (p->mg) | |
a59db676 | 462 | free_migration(p->mg); |
c6b4fcba JT |
463 | } |
464 | ||
465 | static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p) | |
466 | { | |
467 | struct dm_cache_migration *mg = p->mg; | |
468 | ||
469 | BUG_ON(!mg); | |
470 | p->mg = NULL; | |
471 | ||
472 | return mg; | |
473 | } | |
474 | ||
475 | /* | |
476 | * You must have a cell within the prealloc struct to return. If not this | |
477 | * function will BUG() rather than returning NULL. | |
478 | */ | |
479 | static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p) | |
480 | { | |
481 | struct dm_bio_prison_cell *r = NULL; | |
482 | ||
483 | if (p->cell1) { | |
484 | r = p->cell1; | |
485 | p->cell1 = NULL; | |
486 | ||
487 | } else if (p->cell2) { | |
488 | r = p->cell2; | |
489 | p->cell2 = NULL; | |
490 | } else | |
491 | BUG(); | |
492 | ||
493 | return r; | |
494 | } | |
495 | ||
496 | /* | |
497 | * You can't have more than two cells in a prealloc struct. BUG() will be | |
498 | * called if you try and overfill. | |
499 | */ | |
500 | static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell) | |
501 | { | |
502 | if (!p->cell2) | |
503 | p->cell2 = cell; | |
504 | ||
505 | else if (!p->cell1) | |
506 | p->cell1 = cell; | |
507 | ||
508 | else | |
509 | BUG(); | |
510 | } | |
511 | ||
512 | /*----------------------------------------------------------------*/ | |
513 | ||
7ae34e77 | 514 | static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key *key) |
c6b4fcba JT |
515 | { |
516 | key->virtual = 0; | |
517 | key->dev = 0; | |
7ae34e77 JT |
518 | key->block_begin = from_oblock(begin); |
519 | key->block_end = from_oblock(end); | |
c6b4fcba JT |
520 | } |
521 | ||
522 | /* | |
523 | * The caller hands in a preallocated cell, and a free function for it. | |
524 | * The cell will be freed if there's an error, or if it wasn't used because | |
525 | * a cell with that key already exists. | |
526 | */ | |
527 | typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell); | |
528 | ||
7ae34e77 JT |
529 | static int bio_detain_range(struct cache *cache, dm_oblock_t oblock_begin, dm_oblock_t oblock_end, |
530 | struct bio *bio, struct dm_bio_prison_cell *cell_prealloc, | |
531 | cell_free_fn free_fn, void *free_context, | |
532 | struct dm_bio_prison_cell **cell_result) | |
c6b4fcba JT |
533 | { |
534 | int r; | |
535 | struct dm_cell_key key; | |
536 | ||
7ae34e77 | 537 | build_key(oblock_begin, oblock_end, &key); |
c6b4fcba JT |
538 | r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result); |
539 | if (r) | |
540 | free_fn(free_context, cell_prealloc); | |
541 | ||
542 | return r; | |
543 | } | |
544 | ||
7ae34e77 JT |
545 | static int bio_detain(struct cache *cache, dm_oblock_t oblock, |
546 | struct bio *bio, struct dm_bio_prison_cell *cell_prealloc, | |
547 | cell_free_fn free_fn, void *free_context, | |
548 | struct dm_bio_prison_cell **cell_result) | |
549 | { | |
550 | dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL); | |
551 | return bio_detain_range(cache, oblock, end, bio, | |
552 | cell_prealloc, free_fn, free_context, cell_result); | |
553 | } | |
554 | ||
c6b4fcba JT |
555 | static int get_cell(struct cache *cache, |
556 | dm_oblock_t oblock, | |
557 | struct prealloc *structs, | |
558 | struct dm_bio_prison_cell **cell_result) | |
559 | { | |
560 | int r; | |
561 | struct dm_cell_key key; | |
562 | struct dm_bio_prison_cell *cell_prealloc; | |
563 | ||
564 | cell_prealloc = prealloc_get_cell(structs); | |
565 | ||
7ae34e77 | 566 | build_key(oblock, to_oblock(from_oblock(oblock) + 1ULL), &key); |
c6b4fcba JT |
567 | r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result); |
568 | if (r) | |
569 | prealloc_put_cell(structs, cell_prealloc); | |
570 | ||
571 | return r; | |
572 | } | |
573 | ||
aeed1420 | 574 | /*----------------------------------------------------------------*/ |
c6b4fcba JT |
575 | |
576 | static bool is_dirty(struct cache *cache, dm_cblock_t b) | |
577 | { | |
578 | return test_bit(from_cblock(b), cache->dirty_bitset); | |
579 | } | |
580 | ||
581 | static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock) | |
582 | { | |
583 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 584 | atomic_inc(&cache->nr_dirty); |
c6b4fcba JT |
585 | policy_set_dirty(cache->policy, oblock); |
586 | } | |
587 | } | |
588 | ||
589 | static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock) | |
590 | { | |
591 | if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
592 | policy_clear_dirty(cache->policy, oblock); | |
44fa816b | 593 | if (atomic_dec_return(&cache->nr_dirty) == 0) |
c6b4fcba JT |
594 | dm_table_event(cache->ti->table); |
595 | } | |
596 | } | |
597 | ||
598 | /*----------------------------------------------------------------*/ | |
aeed1420 | 599 | |
c6b4fcba JT |
600 | static bool block_size_is_power_of_two(struct cache *cache) |
601 | { | |
602 | return cache->sectors_per_block_shift >= 0; | |
603 | } | |
604 | ||
43aeaa29 MP |
605 | /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */ |
606 | #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6 | |
607 | __always_inline | |
608 | #endif | |
414dd67d JT |
609 | static dm_block_t block_div(dm_block_t b, uint32_t n) |
610 | { | |
611 | do_div(b, n); | |
612 | ||
613 | return b; | |
614 | } | |
615 | ||
7ae34e77 | 616 | static dm_block_t oblocks_per_dblock(struct cache *cache) |
1bad9bc4 | 617 | { |
7ae34e77 | 618 | dm_block_t oblocks = cache->discard_block_size; |
1bad9bc4 | 619 | |
7ae34e77 JT |
620 | if (block_size_is_power_of_two(cache)) |
621 | oblocks >>= cache->sectors_per_block_shift; | |
1bad9bc4 | 622 | else |
7ae34e77 | 623 | oblocks = block_div(oblocks, cache->sectors_per_block); |
1bad9bc4 | 624 | |
7ae34e77 JT |
625 | return oblocks; |
626 | } | |
627 | ||
628 | static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock) | |
629 | { | |
630 | return to_dblock(block_div(from_oblock(oblock), | |
631 | oblocks_per_dblock(cache))); | |
632 | } | |
1bad9bc4 | 633 | |
7ae34e77 JT |
634 | static dm_oblock_t dblock_to_oblock(struct cache *cache, dm_dblock_t dblock) |
635 | { | |
636 | return to_oblock(from_dblock(dblock) * oblocks_per_dblock(cache)); | |
1bad9bc4 JT |
637 | } |
638 | ||
639 | static void set_discard(struct cache *cache, dm_dblock_t b) | |
c6b4fcba JT |
640 | { |
641 | unsigned long flags; | |
642 | ||
7ae34e77 | 643 | BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
644 | atomic_inc(&cache->stats.discard_count); |
645 | ||
646 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 647 | set_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
648 | spin_unlock_irqrestore(&cache->lock, flags); |
649 | } | |
650 | ||
1bad9bc4 | 651 | static void clear_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
652 | { |
653 | unsigned long flags; | |
654 | ||
655 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 656 | clear_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
657 | spin_unlock_irqrestore(&cache->lock, flags); |
658 | } | |
659 | ||
1bad9bc4 | 660 | static bool is_discarded(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
661 | { |
662 | int r; | |
663 | unsigned long flags; | |
664 | ||
665 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 666 | r = test_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
667 | spin_unlock_irqrestore(&cache->lock, flags); |
668 | ||
669 | return r; | |
670 | } | |
671 | ||
672 | static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b) | |
673 | { | |
674 | int r; | |
675 | unsigned long flags; | |
676 | ||
677 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 JT |
678 | r = test_bit(from_dblock(oblock_to_dblock(cache, b)), |
679 | cache->discard_bitset); | |
c6b4fcba JT |
680 | spin_unlock_irqrestore(&cache->lock, flags); |
681 | ||
682 | return r; | |
683 | } | |
684 | ||
685 | /*----------------------------------------------------------------*/ | |
686 | ||
687 | static void load_stats(struct cache *cache) | |
688 | { | |
689 | struct dm_cache_statistics stats; | |
690 | ||
691 | dm_cache_metadata_get_stats(cache->cmd, &stats); | |
692 | atomic_set(&cache->stats.read_hit, stats.read_hits); | |
693 | atomic_set(&cache->stats.read_miss, stats.read_misses); | |
694 | atomic_set(&cache->stats.write_hit, stats.write_hits); | |
695 | atomic_set(&cache->stats.write_miss, stats.write_misses); | |
696 | } | |
697 | ||
698 | static void save_stats(struct cache *cache) | |
699 | { | |
700 | struct dm_cache_statistics stats; | |
701 | ||
028ae9f7 JT |
702 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
703 | return; | |
704 | ||
c6b4fcba JT |
705 | stats.read_hits = atomic_read(&cache->stats.read_hit); |
706 | stats.read_misses = atomic_read(&cache->stats.read_miss); | |
707 | stats.write_hits = atomic_read(&cache->stats.write_hit); | |
708 | stats.write_misses = atomic_read(&cache->stats.write_miss); | |
709 | ||
710 | dm_cache_metadata_set_stats(cache->cmd, &stats); | |
711 | } | |
712 | ||
713 | /*---------------------------------------------------------------- | |
714 | * Per bio data | |
715 | *--------------------------------------------------------------*/ | |
19b0092e MS |
716 | |
717 | /* | |
718 | * If using writeback, leave out struct per_bio_data's writethrough fields. | |
719 | */ | |
720 | #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache)) | |
721 | #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data)) | |
722 | ||
2ee57d58 JT |
723 | static bool writethrough_mode(struct cache_features *f) |
724 | { | |
725 | return f->io_mode == CM_IO_WRITETHROUGH; | |
726 | } | |
727 | ||
728 | static bool writeback_mode(struct cache_features *f) | |
729 | { | |
730 | return f->io_mode == CM_IO_WRITEBACK; | |
731 | } | |
732 | ||
733 | static bool passthrough_mode(struct cache_features *f) | |
734 | { | |
735 | return f->io_mode == CM_IO_PASSTHROUGH; | |
736 | } | |
737 | ||
19b0092e MS |
738 | static size_t get_per_bio_data_size(struct cache *cache) |
739 | { | |
2ee57d58 | 740 | return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB; |
19b0092e MS |
741 | } |
742 | ||
743 | static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size) | |
c6b4fcba | 744 | { |
19b0092e | 745 | struct per_bio_data *pb = dm_per_bio_data(bio, data_size); |
c6b4fcba JT |
746 | BUG_ON(!pb); |
747 | return pb; | |
748 | } | |
749 | ||
19b0092e | 750 | static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size) |
c6b4fcba | 751 | { |
19b0092e | 752 | struct per_bio_data *pb = get_per_bio_data(bio, data_size); |
c6b4fcba JT |
753 | |
754 | pb->tick = false; | |
755 | pb->req_nr = dm_bio_get_target_bio_nr(bio); | |
756 | pb->all_io_entry = NULL; | |
066dbaa3 | 757 | pb->len = 0; |
c6b4fcba JT |
758 | |
759 | return pb; | |
760 | } | |
761 | ||
762 | /*---------------------------------------------------------------- | |
763 | * Remapping | |
764 | *--------------------------------------------------------------*/ | |
765 | static void remap_to_origin(struct cache *cache, struct bio *bio) | |
766 | { | |
767 | bio->bi_bdev = cache->origin_dev->bdev; | |
768 | } | |
769 | ||
770 | static void remap_to_cache(struct cache *cache, struct bio *bio, | |
771 | dm_cblock_t cblock) | |
772 | { | |
4f024f37 | 773 | sector_t bi_sector = bio->bi_iter.bi_sector; |
e0d849fa | 774 | sector_t block = from_cblock(cblock); |
c6b4fcba JT |
775 | |
776 | bio->bi_bdev = cache->cache_dev->bdev; | |
777 | if (!block_size_is_power_of_two(cache)) | |
4f024f37 | 778 | bio->bi_iter.bi_sector = |
e0d849fa | 779 | (block * cache->sectors_per_block) + |
4f024f37 | 780 | sector_div(bi_sector, cache->sectors_per_block); |
c6b4fcba | 781 | else |
4f024f37 | 782 | bio->bi_iter.bi_sector = |
e0d849fa | 783 | (block << cache->sectors_per_block_shift) | |
4f024f37 | 784 | (bi_sector & (cache->sectors_per_block - 1)); |
c6b4fcba JT |
785 | } |
786 | ||
787 | static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio) | |
788 | { | |
789 | unsigned long flags; | |
19b0092e MS |
790 | size_t pb_data_size = get_per_bio_data_size(cache); |
791 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
792 | |
793 | spin_lock_irqsave(&cache->lock, flags); | |
794 | if (cache->need_tick_bio && | |
795 | !(bio->bi_rw & (REQ_FUA | REQ_FLUSH | REQ_DISCARD))) { | |
796 | pb->tick = true; | |
797 | cache->need_tick_bio = false; | |
798 | } | |
799 | spin_unlock_irqrestore(&cache->lock, flags); | |
800 | } | |
801 | ||
802 | static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, | |
803 | dm_oblock_t oblock) | |
804 | { | |
805 | check_if_tick_bio_needed(cache, bio); | |
806 | remap_to_origin(cache, bio); | |
807 | if (bio_data_dir(bio) == WRITE) | |
1bad9bc4 | 808 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
809 | } |
810 | ||
811 | static void remap_to_cache_dirty(struct cache *cache, struct bio *bio, | |
812 | dm_oblock_t oblock, dm_cblock_t cblock) | |
813 | { | |
f8e5f01a | 814 | check_if_tick_bio_needed(cache, bio); |
c6b4fcba JT |
815 | remap_to_cache(cache, bio, cblock); |
816 | if (bio_data_dir(bio) == WRITE) { | |
817 | set_dirty(cache, oblock, cblock); | |
1bad9bc4 | 818 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
819 | } |
820 | } | |
821 | ||
822 | static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio) | |
823 | { | |
4f024f37 | 824 | sector_t block_nr = bio->bi_iter.bi_sector; |
c6b4fcba JT |
825 | |
826 | if (!block_size_is_power_of_two(cache)) | |
827 | (void) sector_div(block_nr, cache->sectors_per_block); | |
828 | else | |
829 | block_nr >>= cache->sectors_per_block_shift; | |
830 | ||
831 | return to_oblock(block_nr); | |
832 | } | |
833 | ||
834 | static int bio_triggers_commit(struct cache *cache, struct bio *bio) | |
835 | { | |
836 | return bio->bi_rw & (REQ_FLUSH | REQ_FUA); | |
837 | } | |
838 | ||
8c081b52 JT |
839 | /* |
840 | * You must increment the deferred set whilst the prison cell is held. To | |
841 | * encourage this, we ask for 'cell' to be passed in. | |
842 | */ | |
843 | static void inc_ds(struct cache *cache, struct bio *bio, | |
844 | struct dm_bio_prison_cell *cell) | |
845 | { | |
846 | size_t pb_data_size = get_per_bio_data_size(cache); | |
847 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
848 | ||
849 | BUG_ON(!cell); | |
850 | BUG_ON(pb->all_io_entry); | |
851 | ||
852 | pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds); | |
853 | } | |
854 | ||
066dbaa3 JT |
855 | static bool accountable_bio(struct cache *cache, struct bio *bio) |
856 | { | |
857 | return ((bio->bi_bdev == cache->origin_dev->bdev) && | |
858 | !(bio->bi_rw & REQ_DISCARD)); | |
859 | } | |
860 | ||
861 | static void accounted_begin(struct cache *cache, struct bio *bio) | |
862 | { | |
863 | size_t pb_data_size = get_per_bio_data_size(cache); | |
864 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
865 | ||
866 | if (accountable_bio(cache, bio)) { | |
867 | pb->len = bio_sectors(bio); | |
868 | iot_io_begin(&cache->origin_tracker, pb->len); | |
869 | } | |
870 | } | |
871 | ||
872 | static void accounted_complete(struct cache *cache, struct bio *bio) | |
873 | { | |
874 | size_t pb_data_size = get_per_bio_data_size(cache); | |
875 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
876 | ||
877 | iot_io_end(&cache->origin_tracker, pb->len); | |
878 | } | |
879 | ||
880 | static void accounted_request(struct cache *cache, struct bio *bio) | |
881 | { | |
882 | accounted_begin(cache, bio); | |
883 | generic_make_request(bio); | |
884 | } | |
885 | ||
c6b4fcba JT |
886 | static void issue(struct cache *cache, struct bio *bio) |
887 | { | |
888 | unsigned long flags; | |
889 | ||
890 | if (!bio_triggers_commit(cache, bio)) { | |
066dbaa3 | 891 | accounted_request(cache, bio); |
c6b4fcba JT |
892 | return; |
893 | } | |
894 | ||
895 | /* | |
896 | * Batch together any bios that trigger commits and then issue a | |
897 | * single commit for them in do_worker(). | |
898 | */ | |
899 | spin_lock_irqsave(&cache->lock, flags); | |
900 | cache->commit_requested = true; | |
901 | bio_list_add(&cache->deferred_flush_bios, bio); | |
902 | spin_unlock_irqrestore(&cache->lock, flags); | |
903 | } | |
904 | ||
8c081b52 JT |
905 | static void inc_and_issue(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell *cell) |
906 | { | |
907 | inc_ds(cache, bio, cell); | |
908 | issue(cache, bio); | |
909 | } | |
910 | ||
e2e74d61 JT |
911 | static void defer_writethrough_bio(struct cache *cache, struct bio *bio) |
912 | { | |
913 | unsigned long flags; | |
914 | ||
915 | spin_lock_irqsave(&cache->lock, flags); | |
916 | bio_list_add(&cache->deferred_writethrough_bios, bio); | |
917 | spin_unlock_irqrestore(&cache->lock, flags); | |
918 | ||
919 | wake_worker(cache); | |
920 | } | |
921 | ||
922 | static void writethrough_endio(struct bio *bio, int err) | |
923 | { | |
19b0092e | 924 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
c9d28d5d JT |
925 | |
926 | dm_unhook_bio(&pb->hook_info, bio); | |
e2e74d61 JT |
927 | |
928 | if (err) { | |
929 | bio_endio(bio, err); | |
930 | return; | |
931 | } | |
932 | ||
b844fe69 | 933 | dm_bio_restore(&pb->bio_details, bio); |
e2e74d61 JT |
934 | remap_to_cache(pb->cache, bio, pb->cblock); |
935 | ||
936 | /* | |
937 | * We can't issue this bio directly, since we're in interrupt | |
aeed1420 | 938 | * context. So it gets put on a bio list for processing by the |
e2e74d61 JT |
939 | * worker thread. |
940 | */ | |
941 | defer_writethrough_bio(pb->cache, bio); | |
942 | } | |
943 | ||
944 | /* | |
945 | * When running in writethrough mode we need to send writes to clean blocks | |
946 | * to both the cache and origin devices. In future we'd like to clone the | |
947 | * bio and send them in parallel, but for now we're doing them in | |
948 | * series as this is easier. | |
949 | */ | |
950 | static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio, | |
951 | dm_oblock_t oblock, dm_cblock_t cblock) | |
952 | { | |
19b0092e | 953 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
e2e74d61 JT |
954 | |
955 | pb->cache = cache; | |
956 | pb->cblock = cblock; | |
c9d28d5d | 957 | dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL); |
b844fe69 | 958 | dm_bio_record(&pb->bio_details, bio); |
e2e74d61 JT |
959 | |
960 | remap_to_origin_clear_discard(pb->cache, bio, oblock); | |
961 | } | |
962 | ||
028ae9f7 JT |
963 | /*---------------------------------------------------------------- |
964 | * Failure modes | |
965 | *--------------------------------------------------------------*/ | |
966 | static enum cache_metadata_mode get_cache_mode(struct cache *cache) | |
967 | { | |
968 | return cache->features.mode; | |
969 | } | |
970 | ||
b61d9509 MS |
971 | static const char *cache_device_name(struct cache *cache) |
972 | { | |
973 | return dm_device_name(dm_table_get_md(cache->ti->table)); | |
974 | } | |
975 | ||
028ae9f7 JT |
976 | static void notify_mode_switch(struct cache *cache, enum cache_metadata_mode mode) |
977 | { | |
978 | const char *descs[] = { | |
979 | "write", | |
980 | "read-only", | |
981 | "fail" | |
982 | }; | |
983 | ||
984 | dm_table_event(cache->ti->table); | |
b61d9509 MS |
985 | DMINFO("%s: switching cache to %s mode", |
986 | cache_device_name(cache), descs[(int)mode]); | |
028ae9f7 JT |
987 | } |
988 | ||
989 | static void set_cache_mode(struct cache *cache, enum cache_metadata_mode new_mode) | |
990 | { | |
991 | bool needs_check = dm_cache_metadata_needs_check(cache->cmd); | |
992 | enum cache_metadata_mode old_mode = get_cache_mode(cache); | |
993 | ||
994 | if (new_mode == CM_WRITE && needs_check) { | |
b61d9509 MS |
995 | DMERR("%s: unable to switch cache to write mode until repaired.", |
996 | cache_device_name(cache)); | |
028ae9f7 JT |
997 | if (old_mode != new_mode) |
998 | new_mode = old_mode; | |
999 | else | |
1000 | new_mode = CM_READ_ONLY; | |
1001 | } | |
1002 | ||
1003 | /* Never move out of fail mode */ | |
1004 | if (old_mode == CM_FAIL) | |
1005 | new_mode = CM_FAIL; | |
1006 | ||
1007 | switch (new_mode) { | |
1008 | case CM_FAIL: | |
1009 | case CM_READ_ONLY: | |
1010 | dm_cache_metadata_set_read_only(cache->cmd); | |
1011 | break; | |
1012 | ||
1013 | case CM_WRITE: | |
1014 | dm_cache_metadata_set_read_write(cache->cmd); | |
1015 | break; | |
1016 | } | |
1017 | ||
1018 | cache->features.mode = new_mode; | |
1019 | ||
1020 | if (new_mode != old_mode) | |
1021 | notify_mode_switch(cache, new_mode); | |
1022 | } | |
1023 | ||
1024 | static void abort_transaction(struct cache *cache) | |
1025 | { | |
b61d9509 MS |
1026 | const char *dev_name = cache_device_name(cache); |
1027 | ||
028ae9f7 JT |
1028 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
1029 | return; | |
1030 | ||
1031 | if (dm_cache_metadata_set_needs_check(cache->cmd)) { | |
b61d9509 | 1032 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); |
028ae9f7 JT |
1033 | set_cache_mode(cache, CM_FAIL); |
1034 | } | |
1035 | ||
b61d9509 | 1036 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); |
028ae9f7 | 1037 | if (dm_cache_metadata_abort(cache->cmd)) { |
b61d9509 | 1038 | DMERR("%s: failed to abort metadata transaction", dev_name); |
028ae9f7 JT |
1039 | set_cache_mode(cache, CM_FAIL); |
1040 | } | |
1041 | } | |
1042 | ||
1043 | static void metadata_operation_failed(struct cache *cache, const char *op, int r) | |
1044 | { | |
b61d9509 MS |
1045 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1046 | cache_device_name(cache), op, r); | |
028ae9f7 JT |
1047 | abort_transaction(cache); |
1048 | set_cache_mode(cache, CM_READ_ONLY); | |
1049 | } | |
1050 | ||
c6b4fcba JT |
1051 | /*---------------------------------------------------------------- |
1052 | * Migration processing | |
1053 | * | |
1054 | * Migration covers moving data from the origin device to the cache, or | |
1055 | * vice versa. | |
1056 | *--------------------------------------------------------------*/ | |
a59db676 | 1057 | static void inc_io_migrations(struct cache *cache) |
c6b4fcba | 1058 | { |
a59db676 | 1059 | atomic_inc(&cache->nr_io_migrations); |
c6b4fcba JT |
1060 | } |
1061 | ||
a59db676 | 1062 | static void dec_io_migrations(struct cache *cache) |
c6b4fcba | 1063 | { |
a59db676 | 1064 | atomic_dec(&cache->nr_io_migrations); |
c6b4fcba JT |
1065 | } |
1066 | ||
651f5fa2 JT |
1067 | static void __cell_release(struct cache *cache, struct dm_bio_prison_cell *cell, |
1068 | bool holder, struct bio_list *bios) | |
c6b4fcba JT |
1069 | { |
1070 | (holder ? dm_cell_release : dm_cell_release_no_holder) | |
651f5fa2 | 1071 | (cache->prison, cell, bios); |
c6b4fcba JT |
1072 | free_prison_cell(cache, cell); |
1073 | } | |
1074 | ||
651f5fa2 JT |
1075 | static bool discard_or_flush(struct bio *bio) |
1076 | { | |
1077 | return bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_DISCARD); | |
1078 | } | |
1079 | ||
1080 | static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell) | |
1081 | { | |
1082 | if (discard_or_flush(cell->holder)) | |
1083 | /* | |
1084 | * We have to handle these bios | |
1085 | * individually. | |
1086 | */ | |
1087 | __cell_release(cache, cell, true, &cache->deferred_bios); | |
1088 | ||
1089 | else | |
1090 | list_add_tail(&cell->user_list, &cache->deferred_cells); | |
1091 | } | |
1092 | ||
1093 | static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, bool holder) | |
c6b4fcba JT |
1094 | { |
1095 | unsigned long flags; | |
1096 | ||
651f5fa2 JT |
1097 | if (!holder && dm_cell_promote_or_release(cache->prison, cell)) { |
1098 | /* | |
1099 | * There was no prisoner to promote to holder, the | |
1100 | * cell has been released. | |
1101 | */ | |
1102 | free_prison_cell(cache, cell); | |
1103 | return; | |
1104 | } | |
1105 | ||
c6b4fcba | 1106 | spin_lock_irqsave(&cache->lock, flags); |
651f5fa2 | 1107 | __cell_defer(cache, cell); |
c6b4fcba JT |
1108 | spin_unlock_irqrestore(&cache->lock, flags); |
1109 | ||
1110 | wake_worker(cache); | |
1111 | } | |
1112 | ||
651f5fa2 JT |
1113 | static void cell_error_with_code(struct cache *cache, struct dm_bio_prison_cell *cell, int err) |
1114 | { | |
1115 | dm_cell_error(cache->prison, cell, err); | |
1116 | dm_bio_prison_free_cell(cache->prison, cell); | |
1117 | } | |
1118 | ||
1119 | static void cell_requeue(struct cache *cache, struct dm_bio_prison_cell *cell) | |
1120 | { | |
1121 | cell_error_with_code(cache, cell, DM_ENDIO_REQUEUE); | |
1122 | } | |
1123 | ||
a59db676 | 1124 | static void free_io_migration(struct dm_cache_migration *mg) |
c6b4fcba | 1125 | { |
a59db676 | 1126 | dec_io_migrations(mg->cache); |
c6b4fcba JT |
1127 | free_migration(mg); |
1128 | } | |
1129 | ||
1130 | static void migration_failure(struct dm_cache_migration *mg) | |
1131 | { | |
1132 | struct cache *cache = mg->cache; | |
b61d9509 | 1133 | const char *dev_name = cache_device_name(cache); |
c6b4fcba JT |
1134 | |
1135 | if (mg->writeback) { | |
b61d9509 | 1136 | DMERR_LIMIT("%s: writeback failed; couldn't copy block", dev_name); |
c6b4fcba JT |
1137 | set_dirty(cache, mg->old_oblock, mg->cblock); |
1138 | cell_defer(cache, mg->old_ocell, false); | |
1139 | ||
1140 | } else if (mg->demote) { | |
b61d9509 | 1141 | DMERR_LIMIT("%s: demotion failed; couldn't copy block", dev_name); |
c6b4fcba JT |
1142 | policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock); |
1143 | ||
80f659f3 | 1144 | cell_defer(cache, mg->old_ocell, mg->promote ? false : true); |
c6b4fcba | 1145 | if (mg->promote) |
80f659f3 | 1146 | cell_defer(cache, mg->new_ocell, true); |
c6b4fcba | 1147 | } else { |
b61d9509 | 1148 | DMERR_LIMIT("%s: promotion failed; couldn't copy block", dev_name); |
c6b4fcba | 1149 | policy_remove_mapping(cache->policy, mg->new_oblock); |
80f659f3 | 1150 | cell_defer(cache, mg->new_ocell, true); |
c6b4fcba JT |
1151 | } |
1152 | ||
a59db676 | 1153 | free_io_migration(mg); |
c6b4fcba JT |
1154 | } |
1155 | ||
1156 | static void migration_success_pre_commit(struct dm_cache_migration *mg) | |
1157 | { | |
028ae9f7 | 1158 | int r; |
c6b4fcba JT |
1159 | unsigned long flags; |
1160 | struct cache *cache = mg->cache; | |
1161 | ||
1162 | if (mg->writeback) { | |
c6b4fcba | 1163 | clear_dirty(cache, mg->old_oblock, mg->cblock); |
40aa978e | 1164 | cell_defer(cache, mg->old_ocell, false); |
a59db676 | 1165 | free_io_migration(mg); |
c6b4fcba JT |
1166 | return; |
1167 | ||
1168 | } else if (mg->demote) { | |
028ae9f7 JT |
1169 | r = dm_cache_remove_mapping(cache->cmd, mg->cblock); |
1170 | if (r) { | |
b61d9509 MS |
1171 | DMERR_LIMIT("%s: demotion failed; couldn't update on disk metadata", |
1172 | cache_device_name(cache)); | |
028ae9f7 | 1173 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); |
c6b4fcba JT |
1174 | policy_force_mapping(cache->policy, mg->new_oblock, |
1175 | mg->old_oblock); | |
1176 | if (mg->promote) | |
1177 | cell_defer(cache, mg->new_ocell, true); | |
a59db676 | 1178 | free_io_migration(mg); |
c6b4fcba JT |
1179 | return; |
1180 | } | |
1181 | } else { | |
028ae9f7 JT |
1182 | r = dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock); |
1183 | if (r) { | |
b61d9509 MS |
1184 | DMERR_LIMIT("%s: promotion failed; couldn't update on disk metadata", |
1185 | cache_device_name(cache)); | |
028ae9f7 | 1186 | metadata_operation_failed(cache, "dm_cache_insert_mapping", r); |
c6b4fcba | 1187 | policy_remove_mapping(cache->policy, mg->new_oblock); |
a59db676 | 1188 | free_io_migration(mg); |
c6b4fcba JT |
1189 | return; |
1190 | } | |
1191 | } | |
1192 | ||
1193 | spin_lock_irqsave(&cache->lock, flags); | |
1194 | list_add_tail(&mg->list, &cache->need_commit_migrations); | |
1195 | cache->commit_requested = true; | |
1196 | spin_unlock_irqrestore(&cache->lock, flags); | |
1197 | } | |
1198 | ||
1199 | static void migration_success_post_commit(struct dm_cache_migration *mg) | |
1200 | { | |
1201 | unsigned long flags; | |
1202 | struct cache *cache = mg->cache; | |
1203 | ||
1204 | if (mg->writeback) { | |
b61d9509 MS |
1205 | DMWARN_LIMIT("%s: writeback unexpectedly triggered commit", |
1206 | cache_device_name(cache)); | |
c6b4fcba JT |
1207 | return; |
1208 | ||
1209 | } else if (mg->demote) { | |
80f659f3 | 1210 | cell_defer(cache, mg->old_ocell, mg->promote ? false : true); |
c6b4fcba JT |
1211 | |
1212 | if (mg->promote) { | |
1213 | mg->demote = false; | |
1214 | ||
1215 | spin_lock_irqsave(&cache->lock, flags); | |
1216 | list_add_tail(&mg->list, &cache->quiesced_migrations); | |
1217 | spin_unlock_irqrestore(&cache->lock, flags); | |
1218 | ||
65790ff9 JT |
1219 | } else { |
1220 | if (mg->invalidate) | |
1221 | policy_remove_mapping(cache->policy, mg->old_oblock); | |
a59db676 | 1222 | free_io_migration(mg); |
65790ff9 | 1223 | } |
c6b4fcba JT |
1224 | |
1225 | } else { | |
1e32134a JT |
1226 | if (mg->requeue_holder) { |
1227 | clear_dirty(cache, mg->new_oblock, mg->cblock); | |
c9d28d5d | 1228 | cell_defer(cache, mg->new_ocell, true); |
1e32134a JT |
1229 | } else { |
1230 | /* | |
1231 | * The block was promoted via an overwrite, so it's dirty. | |
1232 | */ | |
1233 | set_dirty(cache, mg->new_oblock, mg->cblock); | |
c9d28d5d JT |
1234 | bio_endio(mg->new_ocell->holder, 0); |
1235 | cell_defer(cache, mg->new_ocell, false); | |
1236 | } | |
a59db676 | 1237 | free_io_migration(mg); |
c6b4fcba JT |
1238 | } |
1239 | } | |
1240 | ||
1241 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
1242 | { | |
1243 | unsigned long flags; | |
1244 | struct dm_cache_migration *mg = (struct dm_cache_migration *) context; | |
1245 | struct cache *cache = mg->cache; | |
1246 | ||
1247 | if (read_err || write_err) | |
1248 | mg->err = true; | |
1249 | ||
1250 | spin_lock_irqsave(&cache->lock, flags); | |
1251 | list_add_tail(&mg->list, &cache->completed_migrations); | |
1252 | spin_unlock_irqrestore(&cache->lock, flags); | |
1253 | ||
1254 | wake_worker(cache); | |
1255 | } | |
1256 | ||
7ae34e77 | 1257 | static void issue_copy(struct dm_cache_migration *mg) |
c6b4fcba JT |
1258 | { |
1259 | int r; | |
1260 | struct dm_io_region o_region, c_region; | |
1261 | struct cache *cache = mg->cache; | |
8b9d9666 | 1262 | sector_t cblock = from_cblock(mg->cblock); |
c6b4fcba JT |
1263 | |
1264 | o_region.bdev = cache->origin_dev->bdev; | |
1265 | o_region.count = cache->sectors_per_block; | |
1266 | ||
1267 | c_region.bdev = cache->cache_dev->bdev; | |
8b9d9666 | 1268 | c_region.sector = cblock * cache->sectors_per_block; |
c6b4fcba JT |
1269 | c_region.count = cache->sectors_per_block; |
1270 | ||
1271 | if (mg->writeback || mg->demote) { | |
1272 | /* demote */ | |
1273 | o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block; | |
1274 | r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg); | |
1275 | } else { | |
1276 | /* promote */ | |
1277 | o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block; | |
1278 | r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg); | |
1279 | } | |
1280 | ||
2c2263c9 | 1281 | if (r < 0) { |
b61d9509 | 1282 | DMERR_LIMIT("%s: issuing migration failed", cache_device_name(cache)); |
c6b4fcba | 1283 | migration_failure(mg); |
2c2263c9 | 1284 | } |
c6b4fcba JT |
1285 | } |
1286 | ||
c9d28d5d JT |
1287 | static void overwrite_endio(struct bio *bio, int err) |
1288 | { | |
1289 | struct dm_cache_migration *mg = bio->bi_private; | |
1290 | struct cache *cache = mg->cache; | |
1291 | size_t pb_data_size = get_per_bio_data_size(cache); | |
1292 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1293 | unsigned long flags; | |
1294 | ||
80ae49aa MS |
1295 | dm_unhook_bio(&pb->hook_info, bio); |
1296 | ||
c9d28d5d JT |
1297 | if (err) |
1298 | mg->err = true; | |
1299 | ||
80ae49aa MS |
1300 | mg->requeue_holder = false; |
1301 | ||
c9d28d5d JT |
1302 | spin_lock_irqsave(&cache->lock, flags); |
1303 | list_add_tail(&mg->list, &cache->completed_migrations); | |
c9d28d5d JT |
1304 | spin_unlock_irqrestore(&cache->lock, flags); |
1305 | ||
1306 | wake_worker(cache); | |
1307 | } | |
1308 | ||
1309 | static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio) | |
1310 | { | |
1311 | size_t pb_data_size = get_per_bio_data_size(mg->cache); | |
1312 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1313 | ||
1314 | dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg); | |
1315 | remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock); | |
8c081b52 JT |
1316 | |
1317 | /* | |
1318 | * No need to inc_ds() here, since the cell will be held for the | |
1319 | * duration of the io. | |
1320 | */ | |
066dbaa3 | 1321 | accounted_request(mg->cache, bio); |
c9d28d5d JT |
1322 | } |
1323 | ||
1324 | static bool bio_writes_complete_block(struct cache *cache, struct bio *bio) | |
1325 | { | |
1326 | return (bio_data_dir(bio) == WRITE) && | |
4f024f37 | 1327 | (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT)); |
c9d28d5d JT |
1328 | } |
1329 | ||
c6b4fcba JT |
1330 | static void avoid_copy(struct dm_cache_migration *mg) |
1331 | { | |
1332 | atomic_inc(&mg->cache->stats.copies_avoided); | |
1333 | migration_success_pre_commit(mg); | |
1334 | } | |
1335 | ||
7ae34e77 JT |
1336 | static void calc_discard_block_range(struct cache *cache, struct bio *bio, |
1337 | dm_dblock_t *b, dm_dblock_t *e) | |
1338 | { | |
1339 | sector_t sb = bio->bi_iter.bi_sector; | |
1340 | sector_t se = bio_end_sector(bio); | |
1341 | ||
1342 | *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size)); | |
1343 | ||
1344 | if (se - sb < cache->discard_block_size) | |
1345 | *e = *b; | |
1346 | else | |
1347 | *e = to_dblock(block_div(se, cache->discard_block_size)); | |
1348 | } | |
1349 | ||
1350 | static void issue_discard(struct dm_cache_migration *mg) | |
1351 | { | |
1352 | dm_dblock_t b, e; | |
1353 | struct bio *bio = mg->new_ocell->holder; | |
1354 | ||
1355 | calc_discard_block_range(mg->cache, bio, &b, &e); | |
1356 | while (b != e) { | |
1357 | set_discard(mg->cache, b); | |
1358 | b = to_dblock(from_dblock(b) + 1); | |
1359 | } | |
1360 | ||
1361 | bio_endio(bio, 0); | |
1362 | cell_defer(mg->cache, mg->new_ocell, false); | |
1363 | free_migration(mg); | |
1364 | } | |
1365 | ||
1366 | static void issue_copy_or_discard(struct dm_cache_migration *mg) | |
c6b4fcba JT |
1367 | { |
1368 | bool avoid; | |
1369 | struct cache *cache = mg->cache; | |
1370 | ||
7ae34e77 JT |
1371 | if (mg->discard) { |
1372 | issue_discard(mg); | |
1373 | return; | |
1374 | } | |
1375 | ||
c6b4fcba JT |
1376 | if (mg->writeback || mg->demote) |
1377 | avoid = !is_dirty(cache, mg->cblock) || | |
1378 | is_discarded_oblock(cache, mg->old_oblock); | |
c9d28d5d JT |
1379 | else { |
1380 | struct bio *bio = mg->new_ocell->holder; | |
1381 | ||
c6b4fcba JT |
1382 | avoid = is_discarded_oblock(cache, mg->new_oblock); |
1383 | ||
f29a3147 JT |
1384 | if (writeback_mode(&cache->features) && |
1385 | !avoid && bio_writes_complete_block(cache, bio)) { | |
c9d28d5d JT |
1386 | issue_overwrite(mg, bio); |
1387 | return; | |
1388 | } | |
1389 | } | |
1390 | ||
7ae34e77 | 1391 | avoid ? avoid_copy(mg) : issue_copy(mg); |
c6b4fcba JT |
1392 | } |
1393 | ||
1394 | static void complete_migration(struct dm_cache_migration *mg) | |
1395 | { | |
1396 | if (mg->err) | |
1397 | migration_failure(mg); | |
1398 | else | |
1399 | migration_success_pre_commit(mg); | |
1400 | } | |
1401 | ||
1402 | static void process_migrations(struct cache *cache, struct list_head *head, | |
1403 | void (*fn)(struct dm_cache_migration *)) | |
1404 | { | |
1405 | unsigned long flags; | |
1406 | struct list_head list; | |
1407 | struct dm_cache_migration *mg, *tmp; | |
1408 | ||
1409 | INIT_LIST_HEAD(&list); | |
1410 | spin_lock_irqsave(&cache->lock, flags); | |
1411 | list_splice_init(head, &list); | |
1412 | spin_unlock_irqrestore(&cache->lock, flags); | |
1413 | ||
1414 | list_for_each_entry_safe(mg, tmp, &list, list) | |
1415 | fn(mg); | |
1416 | } | |
1417 | ||
1418 | static void __queue_quiesced_migration(struct dm_cache_migration *mg) | |
1419 | { | |
1420 | list_add_tail(&mg->list, &mg->cache->quiesced_migrations); | |
1421 | } | |
1422 | ||
1423 | static void queue_quiesced_migration(struct dm_cache_migration *mg) | |
1424 | { | |
1425 | unsigned long flags; | |
1426 | struct cache *cache = mg->cache; | |
1427 | ||
1428 | spin_lock_irqsave(&cache->lock, flags); | |
1429 | __queue_quiesced_migration(mg); | |
1430 | spin_unlock_irqrestore(&cache->lock, flags); | |
1431 | ||
1432 | wake_worker(cache); | |
1433 | } | |
1434 | ||
1435 | static void queue_quiesced_migrations(struct cache *cache, struct list_head *work) | |
1436 | { | |
1437 | unsigned long flags; | |
1438 | struct dm_cache_migration *mg, *tmp; | |
1439 | ||
1440 | spin_lock_irqsave(&cache->lock, flags); | |
1441 | list_for_each_entry_safe(mg, tmp, work, list) | |
1442 | __queue_quiesced_migration(mg); | |
1443 | spin_unlock_irqrestore(&cache->lock, flags); | |
1444 | ||
1445 | wake_worker(cache); | |
1446 | } | |
1447 | ||
1448 | static void check_for_quiesced_migrations(struct cache *cache, | |
1449 | struct per_bio_data *pb) | |
1450 | { | |
1451 | struct list_head work; | |
1452 | ||
1453 | if (!pb->all_io_entry) | |
1454 | return; | |
1455 | ||
1456 | INIT_LIST_HEAD(&work); | |
8c081b52 | 1457 | dm_deferred_entry_dec(pb->all_io_entry, &work); |
c6b4fcba JT |
1458 | |
1459 | if (!list_empty(&work)) | |
1460 | queue_quiesced_migrations(cache, &work); | |
1461 | } | |
1462 | ||
1463 | static void quiesce_migration(struct dm_cache_migration *mg) | |
1464 | { | |
1465 | if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list)) | |
1466 | queue_quiesced_migration(mg); | |
1467 | } | |
1468 | ||
1469 | static void promote(struct cache *cache, struct prealloc *structs, | |
1470 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1471 | struct dm_bio_prison_cell *cell) | |
1472 | { | |
1473 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1474 | ||
1475 | mg->err = false; | |
7ae34e77 | 1476 | mg->discard = false; |
c6b4fcba JT |
1477 | mg->writeback = false; |
1478 | mg->demote = false; | |
1479 | mg->promote = true; | |
c9d28d5d | 1480 | mg->requeue_holder = true; |
65790ff9 | 1481 | mg->invalidate = false; |
c6b4fcba JT |
1482 | mg->cache = cache; |
1483 | mg->new_oblock = oblock; | |
1484 | mg->cblock = cblock; | |
1485 | mg->old_ocell = NULL; | |
1486 | mg->new_ocell = cell; | |
1487 | mg->start_jiffies = jiffies; | |
1488 | ||
a59db676 | 1489 | inc_io_migrations(cache); |
c6b4fcba JT |
1490 | quiesce_migration(mg); |
1491 | } | |
1492 | ||
1493 | static void writeback(struct cache *cache, struct prealloc *structs, | |
1494 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1495 | struct dm_bio_prison_cell *cell) | |
1496 | { | |
1497 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1498 | ||
1499 | mg->err = false; | |
7ae34e77 | 1500 | mg->discard = false; |
c6b4fcba JT |
1501 | mg->writeback = true; |
1502 | mg->demote = false; | |
1503 | mg->promote = false; | |
c9d28d5d | 1504 | mg->requeue_holder = true; |
65790ff9 | 1505 | mg->invalidate = false; |
c6b4fcba JT |
1506 | mg->cache = cache; |
1507 | mg->old_oblock = oblock; | |
1508 | mg->cblock = cblock; | |
1509 | mg->old_ocell = cell; | |
1510 | mg->new_ocell = NULL; | |
1511 | mg->start_jiffies = jiffies; | |
1512 | ||
a59db676 | 1513 | inc_io_migrations(cache); |
c6b4fcba JT |
1514 | quiesce_migration(mg); |
1515 | } | |
1516 | ||
1517 | static void demote_then_promote(struct cache *cache, struct prealloc *structs, | |
1518 | dm_oblock_t old_oblock, dm_oblock_t new_oblock, | |
1519 | dm_cblock_t cblock, | |
1520 | struct dm_bio_prison_cell *old_ocell, | |
1521 | struct dm_bio_prison_cell *new_ocell) | |
1522 | { | |
1523 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1524 | ||
1525 | mg->err = false; | |
7ae34e77 | 1526 | mg->discard = false; |
c6b4fcba JT |
1527 | mg->writeback = false; |
1528 | mg->demote = true; | |
1529 | mg->promote = true; | |
c9d28d5d | 1530 | mg->requeue_holder = true; |
65790ff9 | 1531 | mg->invalidate = false; |
c6b4fcba JT |
1532 | mg->cache = cache; |
1533 | mg->old_oblock = old_oblock; | |
1534 | mg->new_oblock = new_oblock; | |
1535 | mg->cblock = cblock; | |
1536 | mg->old_ocell = old_ocell; | |
1537 | mg->new_ocell = new_ocell; | |
1538 | mg->start_jiffies = jiffies; | |
1539 | ||
a59db676 | 1540 | inc_io_migrations(cache); |
c6b4fcba JT |
1541 | quiesce_migration(mg); |
1542 | } | |
1543 | ||
2ee57d58 JT |
1544 | /* |
1545 | * Invalidate a cache entry. No writeback occurs; any changes in the cache | |
1546 | * block are thrown away. | |
1547 | */ | |
1548 | static void invalidate(struct cache *cache, struct prealloc *structs, | |
1549 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1550 | struct dm_bio_prison_cell *cell) | |
1551 | { | |
1552 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1553 | ||
1554 | mg->err = false; | |
7ae34e77 | 1555 | mg->discard = false; |
2ee57d58 JT |
1556 | mg->writeback = false; |
1557 | mg->demote = true; | |
1558 | mg->promote = false; | |
1559 | mg->requeue_holder = true; | |
65790ff9 | 1560 | mg->invalidate = true; |
2ee57d58 JT |
1561 | mg->cache = cache; |
1562 | mg->old_oblock = oblock; | |
1563 | mg->cblock = cblock; | |
1564 | mg->old_ocell = cell; | |
1565 | mg->new_ocell = NULL; | |
1566 | mg->start_jiffies = jiffies; | |
1567 | ||
a59db676 | 1568 | inc_io_migrations(cache); |
2ee57d58 JT |
1569 | quiesce_migration(mg); |
1570 | } | |
1571 | ||
7ae34e77 JT |
1572 | static void discard(struct cache *cache, struct prealloc *structs, |
1573 | struct dm_bio_prison_cell *cell) | |
1574 | { | |
1575 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1576 | ||
1577 | mg->err = false; | |
1578 | mg->discard = true; | |
1579 | mg->writeback = false; | |
1580 | mg->demote = false; | |
1581 | mg->promote = false; | |
1582 | mg->requeue_holder = false; | |
1583 | mg->invalidate = false; | |
1584 | mg->cache = cache; | |
1585 | mg->old_ocell = NULL; | |
1586 | mg->new_ocell = cell; | |
1587 | mg->start_jiffies = jiffies; | |
1588 | ||
1589 | quiesce_migration(mg); | |
1590 | } | |
1591 | ||
c6b4fcba JT |
1592 | /*---------------------------------------------------------------- |
1593 | * bio processing | |
1594 | *--------------------------------------------------------------*/ | |
1595 | static void defer_bio(struct cache *cache, struct bio *bio) | |
1596 | { | |
1597 | unsigned long flags; | |
1598 | ||
1599 | spin_lock_irqsave(&cache->lock, flags); | |
1600 | bio_list_add(&cache->deferred_bios, bio); | |
1601 | spin_unlock_irqrestore(&cache->lock, flags); | |
1602 | ||
1603 | wake_worker(cache); | |
1604 | } | |
1605 | ||
1606 | static void process_flush_bio(struct cache *cache, struct bio *bio) | |
1607 | { | |
19b0092e MS |
1608 | size_t pb_data_size = get_per_bio_data_size(cache); |
1609 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba | 1610 | |
4f024f37 | 1611 | BUG_ON(bio->bi_iter.bi_size); |
c6b4fcba JT |
1612 | if (!pb->req_nr) |
1613 | remap_to_origin(cache, bio); | |
1614 | else | |
1615 | remap_to_cache(cache, bio, 0); | |
1616 | ||
8c081b52 JT |
1617 | /* |
1618 | * REQ_FLUSH is not directed at any particular block so we don't | |
1619 | * need to inc_ds(). REQ_FUA's are split into a write + REQ_FLUSH | |
1620 | * by dm-core. | |
1621 | */ | |
c6b4fcba JT |
1622 | issue(cache, bio); |
1623 | } | |
1624 | ||
7ae34e77 JT |
1625 | static void process_discard_bio(struct cache *cache, struct prealloc *structs, |
1626 | struct bio *bio) | |
c6b4fcba | 1627 | { |
7ae34e77 JT |
1628 | int r; |
1629 | dm_dblock_t b, e; | |
1630 | struct dm_bio_prison_cell *cell_prealloc, *new_ocell; | |
c6b4fcba | 1631 | |
7ae34e77 JT |
1632 | calc_discard_block_range(cache, bio, &b, &e); |
1633 | if (b == e) { | |
1634 | bio_endio(bio, 0); | |
1635 | return; | |
1636 | } | |
c6b4fcba | 1637 | |
7ae34e77 JT |
1638 | cell_prealloc = prealloc_get_cell(structs); |
1639 | r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc, | |
1640 | (cell_free_fn) prealloc_put_cell, | |
1641 | structs, &new_ocell); | |
1642 | if (r > 0) | |
1643 | return; | |
c6b4fcba | 1644 | |
7ae34e77 | 1645 | discard(cache, structs, new_ocell); |
c6b4fcba JT |
1646 | } |
1647 | ||
1648 | static bool spare_migration_bandwidth(struct cache *cache) | |
1649 | { | |
a59db676 | 1650 | sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) * |
c6b4fcba JT |
1651 | cache->sectors_per_block; |
1652 | return current_volume < cache->migration_threshold; | |
1653 | } | |
1654 | ||
c6b4fcba JT |
1655 | static void inc_hit_counter(struct cache *cache, struct bio *bio) |
1656 | { | |
1657 | atomic_inc(bio_data_dir(bio) == READ ? | |
1658 | &cache->stats.read_hit : &cache->stats.write_hit); | |
1659 | } | |
1660 | ||
1661 | static void inc_miss_counter(struct cache *cache, struct bio *bio) | |
1662 | { | |
1663 | atomic_inc(bio_data_dir(bio) == READ ? | |
1664 | &cache->stats.read_miss : &cache->stats.write_miss); | |
1665 | } | |
1666 | ||
fb4100ae JT |
1667 | /*----------------------------------------------------------------*/ |
1668 | ||
651f5fa2 JT |
1669 | struct inc_detail { |
1670 | struct cache *cache; | |
1671 | struct bio_list bios_for_issue; | |
1672 | struct bio_list unhandled_bios; | |
1673 | bool any_writes; | |
1674 | }; | |
1675 | ||
1676 | static void inc_fn(void *context, struct dm_bio_prison_cell *cell) | |
1677 | { | |
1678 | struct bio *bio; | |
1679 | struct inc_detail *detail = context; | |
1680 | struct cache *cache = detail->cache; | |
1681 | ||
1682 | inc_ds(cache, cell->holder, cell); | |
1683 | if (bio_data_dir(cell->holder) == WRITE) | |
1684 | detail->any_writes = true; | |
1685 | ||
1686 | while ((bio = bio_list_pop(&cell->bios))) { | |
1687 | if (discard_or_flush(bio)) { | |
1688 | bio_list_add(&detail->unhandled_bios, bio); | |
1689 | continue; | |
1690 | } | |
1691 | ||
1692 | if (bio_data_dir(bio) == WRITE) | |
1693 | detail->any_writes = true; | |
1694 | ||
1695 | bio_list_add(&detail->bios_for_issue, bio); | |
1696 | inc_ds(cache, bio, cell); | |
1697 | } | |
1698 | } | |
1699 | ||
1700 | // FIXME: refactor these two | |
1701 | static void remap_cell_to_origin_clear_discard(struct cache *cache, | |
1702 | struct dm_bio_prison_cell *cell, | |
1703 | dm_oblock_t oblock, bool issue_holder) | |
1704 | { | |
1705 | struct bio *bio; | |
1706 | unsigned long flags; | |
1707 | struct inc_detail detail; | |
1708 | ||
1709 | detail.cache = cache; | |
1710 | bio_list_init(&detail.bios_for_issue); | |
1711 | bio_list_init(&detail.unhandled_bios); | |
1712 | detail.any_writes = false; | |
1713 | ||
1714 | spin_lock_irqsave(&cache->lock, flags); | |
1715 | dm_cell_visit_release(cache->prison, inc_fn, &detail, cell); | |
1716 | bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios); | |
1717 | spin_unlock_irqrestore(&cache->lock, flags); | |
1718 | ||
1719 | remap_to_origin(cache, cell->holder); | |
1720 | if (issue_holder) | |
1721 | issue(cache, cell->holder); | |
1722 | else | |
1723 | accounted_begin(cache, cell->holder); | |
1724 | ||
1725 | if (detail.any_writes) | |
1726 | clear_discard(cache, oblock_to_dblock(cache, oblock)); | |
1727 | ||
1728 | while ((bio = bio_list_pop(&detail.bios_for_issue))) { | |
1729 | remap_to_origin(cache, bio); | |
1730 | issue(cache, bio); | |
1731 | } | |
1732 | } | |
1733 | ||
1734 | static void remap_cell_to_cache_dirty(struct cache *cache, struct dm_bio_prison_cell *cell, | |
1735 | dm_oblock_t oblock, dm_cblock_t cblock, bool issue_holder) | |
1736 | { | |
1737 | struct bio *bio; | |
1738 | unsigned long flags; | |
1739 | struct inc_detail detail; | |
1740 | ||
1741 | detail.cache = cache; | |
1742 | bio_list_init(&detail.bios_for_issue); | |
1743 | bio_list_init(&detail.unhandled_bios); | |
1744 | detail.any_writes = false; | |
1745 | ||
1746 | spin_lock_irqsave(&cache->lock, flags); | |
1747 | dm_cell_visit_release(cache->prison, inc_fn, &detail, cell); | |
1748 | bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios); | |
1749 | spin_unlock_irqrestore(&cache->lock, flags); | |
1750 | ||
1751 | remap_to_cache(cache, cell->holder, cblock); | |
1752 | if (issue_holder) | |
1753 | issue(cache, cell->holder); | |
1754 | else | |
1755 | accounted_begin(cache, cell->holder); | |
1756 | ||
1757 | if (detail.any_writes) { | |
1758 | set_dirty(cache, oblock, cblock); | |
1759 | clear_discard(cache, oblock_to_dblock(cache, oblock)); | |
1760 | } | |
1761 | ||
1762 | while ((bio = bio_list_pop(&detail.bios_for_issue))) { | |
1763 | remap_to_cache(cache, bio, cblock); | |
1764 | issue(cache, bio); | |
1765 | } | |
1766 | } | |
1767 | ||
1768 | /*----------------------------------------------------------------*/ | |
1769 | ||
fb4100ae JT |
1770 | struct old_oblock_lock { |
1771 | struct policy_locker locker; | |
1772 | struct cache *cache; | |
1773 | struct prealloc *structs; | |
1774 | struct dm_bio_prison_cell *cell; | |
1775 | }; | |
1776 | ||
1777 | static int null_locker(struct policy_locker *locker, dm_oblock_t b) | |
1778 | { | |
1779 | /* This should never be called */ | |
1780 | BUG(); | |
1781 | return 0; | |
1782 | } | |
1783 | ||
1784 | static int cell_locker(struct policy_locker *locker, dm_oblock_t b) | |
1785 | { | |
1786 | struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker); | |
1787 | struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs); | |
1788 | ||
1789 | return bio_detain(l->cache, b, NULL, cell_prealloc, | |
1790 | (cell_free_fn) prealloc_put_cell, | |
1791 | l->structs, &l->cell); | |
1792 | } | |
1793 | ||
651f5fa2 JT |
1794 | static void process_cell(struct cache *cache, struct prealloc *structs, |
1795 | struct dm_bio_prison_cell *new_ocell) | |
c6b4fcba JT |
1796 | { |
1797 | int r; | |
1798 | bool release_cell = true; | |
651f5fa2 | 1799 | struct bio *bio = new_ocell->holder; |
c6b4fcba | 1800 | dm_oblock_t block = get_bio_block(cache, bio); |
c6b4fcba | 1801 | struct policy_result lookup_result; |
2ee57d58 | 1802 | bool passthrough = passthrough_mode(&cache->features); |
40775257 | 1803 | bool fast_promotion, can_migrate; |
fb4100ae | 1804 | struct old_oblock_lock ool; |
c6b4fcba | 1805 | |
40775257 JT |
1806 | fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio); |
1807 | can_migrate = !passthrough && (fast_promotion || spare_migration_bandwidth(cache)); | |
43c32bf2 | 1808 | |
fb4100ae JT |
1809 | ool.locker.fn = cell_locker; |
1810 | ool.cache = cache; | |
1811 | ool.structs = structs; | |
1812 | ool.cell = NULL; | |
40775257 | 1813 | r = policy_map(cache->policy, block, true, can_migrate, fast_promotion, |
fb4100ae | 1814 | bio, &ool.locker, &lookup_result); |
c6b4fcba JT |
1815 | |
1816 | if (r == -EWOULDBLOCK) | |
1817 | /* migration has been denied */ | |
1818 | lookup_result.op = POLICY_MISS; | |
1819 | ||
1820 | switch (lookup_result.op) { | |
1821 | case POLICY_HIT: | |
2ee57d58 JT |
1822 | if (passthrough) { |
1823 | inc_miss_counter(cache, bio); | |
c6b4fcba | 1824 | |
2ee57d58 JT |
1825 | /* |
1826 | * Passthrough always maps to the origin, | |
1827 | * invalidating any cache blocks that are written | |
1828 | * to. | |
1829 | */ | |
1830 | ||
1831 | if (bio_data_dir(bio) == WRITE) { | |
1832 | atomic_inc(&cache->stats.demotion); | |
1833 | invalidate(cache, structs, block, lookup_result.cblock, new_ocell); | |
1834 | release_cell = false; | |
1835 | ||
1836 | } else { | |
1837 | /* FIXME: factor out issue_origin() */ | |
2ee57d58 | 1838 | remap_to_origin_clear_discard(cache, bio, block); |
8c081b52 | 1839 | inc_and_issue(cache, bio, new_ocell); |
2ee57d58 JT |
1840 | } |
1841 | } else { | |
1842 | inc_hit_counter(cache, bio); | |
1843 | ||
1844 | if (bio_data_dir(bio) == WRITE && | |
1845 | writethrough_mode(&cache->features) && | |
1846 | !is_dirty(cache, lookup_result.cblock)) { | |
2ee57d58 | 1847 | remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock); |
8c081b52 JT |
1848 | inc_and_issue(cache, bio, new_ocell); |
1849 | ||
651f5fa2 JT |
1850 | } else { |
1851 | remap_cell_to_cache_dirty(cache, new_ocell, block, lookup_result.cblock, true); | |
1852 | release_cell = false; | |
8c081b52 | 1853 | } |
2ee57d58 | 1854 | } |
c6b4fcba | 1855 | |
c6b4fcba JT |
1856 | break; |
1857 | ||
1858 | case POLICY_MISS: | |
1859 | inc_miss_counter(cache, bio); | |
651f5fa2 JT |
1860 | remap_cell_to_origin_clear_discard(cache, new_ocell, block, true); |
1861 | release_cell = false; | |
c6b4fcba JT |
1862 | break; |
1863 | ||
1864 | case POLICY_NEW: | |
1865 | atomic_inc(&cache->stats.promotion); | |
1866 | promote(cache, structs, block, lookup_result.cblock, new_ocell); | |
1867 | release_cell = false; | |
1868 | break; | |
1869 | ||
1870 | case POLICY_REPLACE: | |
c6b4fcba JT |
1871 | atomic_inc(&cache->stats.demotion); |
1872 | atomic_inc(&cache->stats.promotion); | |
c6b4fcba JT |
1873 | demote_then_promote(cache, structs, lookup_result.old_oblock, |
1874 | block, lookup_result.cblock, | |
fb4100ae | 1875 | ool.cell, new_ocell); |
c6b4fcba JT |
1876 | release_cell = false; |
1877 | break; | |
1878 | ||
1879 | default: | |
b61d9509 MS |
1880 | DMERR_LIMIT("%s: %s: erroring bio, unknown policy op: %u", |
1881 | cache_device_name(cache), __func__, | |
c6b4fcba JT |
1882 | (unsigned) lookup_result.op); |
1883 | bio_io_error(bio); | |
1884 | } | |
1885 | ||
1886 | if (release_cell) | |
1887 | cell_defer(cache, new_ocell, false); | |
1888 | } | |
1889 | ||
651f5fa2 JT |
1890 | static void process_bio(struct cache *cache, struct prealloc *structs, |
1891 | struct bio *bio) | |
1892 | { | |
1893 | int r; | |
1894 | dm_oblock_t block = get_bio_block(cache, bio); | |
1895 | struct dm_bio_prison_cell *cell_prealloc, *new_ocell; | |
1896 | ||
1897 | /* | |
1898 | * Check to see if that block is currently migrating. | |
1899 | */ | |
1900 | cell_prealloc = prealloc_get_cell(structs); | |
1901 | r = bio_detain(cache, block, bio, cell_prealloc, | |
1902 | (cell_free_fn) prealloc_put_cell, | |
1903 | structs, &new_ocell); | |
1904 | if (r > 0) | |
1905 | return; | |
1906 | ||
1907 | process_cell(cache, structs, new_ocell); | |
1908 | } | |
1909 | ||
c6b4fcba JT |
1910 | static int need_commit_due_to_time(struct cache *cache) |
1911 | { | |
651f5fa2 JT |
1912 | return jiffies < cache->last_commit_jiffies || |
1913 | jiffies > cache->last_commit_jiffies + COMMIT_PERIOD; | |
c6b4fcba JT |
1914 | } |
1915 | ||
028ae9f7 JT |
1916 | /* |
1917 | * A non-zero return indicates read_only or fail_io mode. | |
1918 | */ | |
1919 | static int commit(struct cache *cache, bool clean_shutdown) | |
1920 | { | |
1921 | int r; | |
1922 | ||
1923 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
1924 | return -EINVAL; | |
1925 | ||
1926 | atomic_inc(&cache->stats.commit_count); | |
1927 | r = dm_cache_commit(cache->cmd, clean_shutdown); | |
1928 | if (r) | |
1929 | metadata_operation_failed(cache, "dm_cache_commit", r); | |
1930 | ||
1931 | return r; | |
1932 | } | |
1933 | ||
c6b4fcba JT |
1934 | static int commit_if_needed(struct cache *cache) |
1935 | { | |
ffcbcb67 HM |
1936 | int r = 0; |
1937 | ||
1938 | if ((cache->commit_requested || need_commit_due_to_time(cache)) && | |
1939 | dm_cache_changed_this_transaction(cache->cmd)) { | |
028ae9f7 | 1940 | r = commit(cache, false); |
c6b4fcba | 1941 | cache->commit_requested = false; |
ffcbcb67 | 1942 | cache->last_commit_jiffies = jiffies; |
c6b4fcba JT |
1943 | } |
1944 | ||
ffcbcb67 | 1945 | return r; |
c6b4fcba JT |
1946 | } |
1947 | ||
1948 | static void process_deferred_bios(struct cache *cache) | |
1949 | { | |
665022d7 | 1950 | bool prealloc_used = false; |
c6b4fcba JT |
1951 | unsigned long flags; |
1952 | struct bio_list bios; | |
1953 | struct bio *bio; | |
1954 | struct prealloc structs; | |
1955 | ||
1956 | memset(&structs, 0, sizeof(structs)); | |
1957 | bio_list_init(&bios); | |
1958 | ||
1959 | spin_lock_irqsave(&cache->lock, flags); | |
1960 | bio_list_merge(&bios, &cache->deferred_bios); | |
1961 | bio_list_init(&cache->deferred_bios); | |
1962 | spin_unlock_irqrestore(&cache->lock, flags); | |
1963 | ||
1964 | while (!bio_list_empty(&bios)) { | |
1965 | /* | |
1966 | * If we've got no free migration structs, and processing | |
1967 | * this bio might require one, we pause until there are some | |
1968 | * prepared mappings to process. | |
1969 | */ | |
795e633a | 1970 | prealloc_used = true; |
c6b4fcba JT |
1971 | if (prealloc_data_structs(cache, &structs)) { |
1972 | spin_lock_irqsave(&cache->lock, flags); | |
1973 | bio_list_merge(&cache->deferred_bios, &bios); | |
1974 | spin_unlock_irqrestore(&cache->lock, flags); | |
1975 | break; | |
1976 | } | |
1977 | ||
1978 | bio = bio_list_pop(&bios); | |
1979 | ||
1980 | if (bio->bi_rw & REQ_FLUSH) | |
1981 | process_flush_bio(cache, bio); | |
1982 | else if (bio->bi_rw & REQ_DISCARD) | |
7ae34e77 | 1983 | process_discard_bio(cache, &structs, bio); |
c6b4fcba JT |
1984 | else |
1985 | process_bio(cache, &structs, bio); | |
1986 | } | |
1987 | ||
665022d7 MS |
1988 | if (prealloc_used) |
1989 | prealloc_free_structs(cache, &structs); | |
c6b4fcba JT |
1990 | } |
1991 | ||
651f5fa2 JT |
1992 | static void process_deferred_cells(struct cache *cache) |
1993 | { | |
665022d7 | 1994 | bool prealloc_used = false; |
651f5fa2 JT |
1995 | unsigned long flags; |
1996 | struct dm_bio_prison_cell *cell, *tmp; | |
1997 | struct list_head cells; | |
1998 | struct prealloc structs; | |
1999 | ||
2000 | memset(&structs, 0, sizeof(structs)); | |
2001 | ||
2002 | INIT_LIST_HEAD(&cells); | |
2003 | ||
2004 | spin_lock_irqsave(&cache->lock, flags); | |
2005 | list_splice_init(&cache->deferred_cells, &cells); | |
2006 | spin_unlock_irqrestore(&cache->lock, flags); | |
2007 | ||
2008 | list_for_each_entry_safe(cell, tmp, &cells, user_list) { | |
2009 | /* | |
2010 | * If we've got no free migration structs, and processing | |
2011 | * this bio might require one, we pause until there are some | |
2012 | * prepared mappings to process. | |
2013 | */ | |
795e633a | 2014 | prealloc_used = true; |
651f5fa2 JT |
2015 | if (prealloc_data_structs(cache, &structs)) { |
2016 | spin_lock_irqsave(&cache->lock, flags); | |
2017 | list_splice(&cells, &cache->deferred_cells); | |
2018 | spin_unlock_irqrestore(&cache->lock, flags); | |
2019 | break; | |
2020 | } | |
2021 | ||
2022 | process_cell(cache, &structs, cell); | |
2023 | } | |
2024 | ||
665022d7 MS |
2025 | if (prealloc_used) |
2026 | prealloc_free_structs(cache, &structs); | |
651f5fa2 JT |
2027 | } |
2028 | ||
c6b4fcba JT |
2029 | static void process_deferred_flush_bios(struct cache *cache, bool submit_bios) |
2030 | { | |
2031 | unsigned long flags; | |
2032 | struct bio_list bios; | |
2033 | struct bio *bio; | |
2034 | ||
2035 | bio_list_init(&bios); | |
2036 | ||
2037 | spin_lock_irqsave(&cache->lock, flags); | |
2038 | bio_list_merge(&bios, &cache->deferred_flush_bios); | |
2039 | bio_list_init(&cache->deferred_flush_bios); | |
2040 | spin_unlock_irqrestore(&cache->lock, flags); | |
2041 | ||
8c081b52 JT |
2042 | /* |
2043 | * These bios have already been through inc_ds() | |
2044 | */ | |
c6b4fcba | 2045 | while ((bio = bio_list_pop(&bios))) |
066dbaa3 | 2046 | submit_bios ? accounted_request(cache, bio) : bio_io_error(bio); |
c6b4fcba JT |
2047 | } |
2048 | ||
e2e74d61 JT |
2049 | static void process_deferred_writethrough_bios(struct cache *cache) |
2050 | { | |
2051 | unsigned long flags; | |
2052 | struct bio_list bios; | |
2053 | struct bio *bio; | |
2054 | ||
2055 | bio_list_init(&bios); | |
2056 | ||
2057 | spin_lock_irqsave(&cache->lock, flags); | |
2058 | bio_list_merge(&bios, &cache->deferred_writethrough_bios); | |
2059 | bio_list_init(&cache->deferred_writethrough_bios); | |
2060 | spin_unlock_irqrestore(&cache->lock, flags); | |
2061 | ||
8c081b52 JT |
2062 | /* |
2063 | * These bios have already been through inc_ds() | |
2064 | */ | |
e2e74d61 | 2065 | while ((bio = bio_list_pop(&bios))) |
066dbaa3 | 2066 | accounted_request(cache, bio); |
e2e74d61 JT |
2067 | } |
2068 | ||
c6b4fcba JT |
2069 | static void writeback_some_dirty_blocks(struct cache *cache) |
2070 | { | |
665022d7 | 2071 | bool prealloc_used = false; |
c6b4fcba JT |
2072 | dm_oblock_t oblock; |
2073 | dm_cblock_t cblock; | |
2074 | struct prealloc structs; | |
2075 | struct dm_bio_prison_cell *old_ocell; | |
20f6814b | 2076 | bool busy = !iot_idle_for(&cache->origin_tracker, HZ); |
c6b4fcba JT |
2077 | |
2078 | memset(&structs, 0, sizeof(structs)); | |
2079 | ||
2080 | while (spare_migration_bandwidth(cache)) { | |
e782eff5 MS |
2081 | if (policy_writeback_work(cache->policy, &oblock, &cblock, busy)) |
2082 | break; /* no work to do */ | |
c6b4fcba | 2083 | |
795e633a | 2084 | prealloc_used = true; |
e782eff5 MS |
2085 | if (prealloc_data_structs(cache, &structs) || |
2086 | get_cell(cache, oblock, &structs, &old_ocell)) { | |
c6b4fcba JT |
2087 | policy_set_dirty(cache->policy, oblock); |
2088 | break; | |
2089 | } | |
2090 | ||
2091 | writeback(cache, &structs, oblock, cblock, old_ocell); | |
2092 | } | |
2093 | ||
665022d7 MS |
2094 | if (prealloc_used) |
2095 | prealloc_free_structs(cache, &structs); | |
c6b4fcba JT |
2096 | } |
2097 | ||
65790ff9 JT |
2098 | /*---------------------------------------------------------------- |
2099 | * Invalidations. | |
2100 | * Dropping something from the cache *without* writing back. | |
2101 | *--------------------------------------------------------------*/ | |
2102 | ||
2103 | static void process_invalidation_request(struct cache *cache, struct invalidation_request *req) | |
2104 | { | |
2105 | int r = 0; | |
2106 | uint64_t begin = from_cblock(req->cblocks->begin); | |
2107 | uint64_t end = from_cblock(req->cblocks->end); | |
2108 | ||
2109 | while (begin != end) { | |
2110 | r = policy_remove_cblock(cache->policy, to_cblock(begin)); | |
2111 | if (!r) { | |
2112 | r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin)); | |
028ae9f7 JT |
2113 | if (r) { |
2114 | metadata_operation_failed(cache, "dm_cache_remove_mapping", r); | |
65790ff9 | 2115 | break; |
028ae9f7 | 2116 | } |
65790ff9 JT |
2117 | |
2118 | } else if (r == -ENODATA) { | |
2119 | /* harmless, already unmapped */ | |
2120 | r = 0; | |
2121 | ||
2122 | } else { | |
b61d9509 | 2123 | DMERR("%s: policy_remove_cblock failed", cache_device_name(cache)); |
65790ff9 JT |
2124 | break; |
2125 | } | |
2126 | ||
2127 | begin++; | |
2128 | } | |
2129 | ||
2130 | cache->commit_requested = true; | |
2131 | ||
2132 | req->err = r; | |
2133 | atomic_set(&req->complete, 1); | |
2134 | ||
2135 | wake_up(&req->result_wait); | |
2136 | } | |
2137 | ||
2138 | static void process_invalidation_requests(struct cache *cache) | |
2139 | { | |
2140 | struct list_head list; | |
2141 | struct invalidation_request *req, *tmp; | |
2142 | ||
2143 | INIT_LIST_HEAD(&list); | |
2144 | spin_lock(&cache->invalidation_lock); | |
2145 | list_splice_init(&cache->invalidation_requests, &list); | |
2146 | spin_unlock(&cache->invalidation_lock); | |
2147 | ||
2148 | list_for_each_entry_safe (req, tmp, &list, list) | |
2149 | process_invalidation_request(cache, req); | |
2150 | } | |
2151 | ||
c6b4fcba JT |
2152 | /*---------------------------------------------------------------- |
2153 | * Main worker loop | |
2154 | *--------------------------------------------------------------*/ | |
66cb1910 | 2155 | static bool is_quiescing(struct cache *cache) |
c6b4fcba | 2156 | { |
238f8363 | 2157 | return atomic_read(&cache->quiescing); |
c6b4fcba JT |
2158 | } |
2159 | ||
66cb1910 JT |
2160 | static void ack_quiescing(struct cache *cache) |
2161 | { | |
2162 | if (is_quiescing(cache)) { | |
2163 | atomic_inc(&cache->quiescing_ack); | |
2164 | wake_up(&cache->quiescing_wait); | |
2165 | } | |
2166 | } | |
2167 | ||
2168 | static void wait_for_quiescing_ack(struct cache *cache) | |
2169 | { | |
2170 | wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack)); | |
2171 | } | |
2172 | ||
2173 | static void start_quiescing(struct cache *cache) | |
c6b4fcba | 2174 | { |
238f8363 | 2175 | atomic_inc(&cache->quiescing); |
66cb1910 | 2176 | wait_for_quiescing_ack(cache); |
c6b4fcba JT |
2177 | } |
2178 | ||
66cb1910 | 2179 | static void stop_quiescing(struct cache *cache) |
c6b4fcba | 2180 | { |
238f8363 | 2181 | atomic_set(&cache->quiescing, 0); |
66cb1910 | 2182 | atomic_set(&cache->quiescing_ack, 0); |
c6b4fcba JT |
2183 | } |
2184 | ||
2185 | static void wait_for_migrations(struct cache *cache) | |
2186 | { | |
a59db676 | 2187 | wait_event(cache->migration_wait, !atomic_read(&cache->nr_allocated_migrations)); |
c6b4fcba JT |
2188 | } |
2189 | ||
2190 | static void stop_worker(struct cache *cache) | |
2191 | { | |
2192 | cancel_delayed_work(&cache->waker); | |
2193 | flush_workqueue(cache->wq); | |
2194 | } | |
2195 | ||
651f5fa2 JT |
2196 | static void requeue_deferred_cells(struct cache *cache) |
2197 | { | |
2198 | unsigned long flags; | |
2199 | struct list_head cells; | |
2200 | struct dm_bio_prison_cell *cell, *tmp; | |
2201 | ||
2202 | INIT_LIST_HEAD(&cells); | |
2203 | spin_lock_irqsave(&cache->lock, flags); | |
2204 | list_splice_init(&cache->deferred_cells, &cells); | |
2205 | spin_unlock_irqrestore(&cache->lock, flags); | |
2206 | ||
2207 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
2208 | cell_requeue(cache, cell); | |
2209 | } | |
2210 | ||
2211 | static void requeue_deferred_bios(struct cache *cache) | |
c6b4fcba JT |
2212 | { |
2213 | struct bio *bio; | |
2214 | struct bio_list bios; | |
2215 | ||
2216 | bio_list_init(&bios); | |
2217 | bio_list_merge(&bios, &cache->deferred_bios); | |
2218 | bio_list_init(&cache->deferred_bios); | |
2219 | ||
2220 | while ((bio = bio_list_pop(&bios))) | |
2221 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
2222 | } | |
2223 | ||
2224 | static int more_work(struct cache *cache) | |
2225 | { | |
2226 | if (is_quiescing(cache)) | |
2227 | return !list_empty(&cache->quiesced_migrations) || | |
2228 | !list_empty(&cache->completed_migrations) || | |
2229 | !list_empty(&cache->need_commit_migrations); | |
2230 | else | |
2231 | return !bio_list_empty(&cache->deferred_bios) || | |
651f5fa2 | 2232 | !list_empty(&cache->deferred_cells) || |
c6b4fcba | 2233 | !bio_list_empty(&cache->deferred_flush_bios) || |
e2e74d61 | 2234 | !bio_list_empty(&cache->deferred_writethrough_bios) || |
c6b4fcba JT |
2235 | !list_empty(&cache->quiesced_migrations) || |
2236 | !list_empty(&cache->completed_migrations) || | |
65790ff9 JT |
2237 | !list_empty(&cache->need_commit_migrations) || |
2238 | cache->invalidate; | |
c6b4fcba JT |
2239 | } |
2240 | ||
2241 | static void do_worker(struct work_struct *ws) | |
2242 | { | |
2243 | struct cache *cache = container_of(ws, struct cache, worker); | |
2244 | ||
2245 | do { | |
66cb1910 JT |
2246 | if (!is_quiescing(cache)) { |
2247 | writeback_some_dirty_blocks(cache); | |
2248 | process_deferred_writethrough_bios(cache); | |
c6b4fcba | 2249 | process_deferred_bios(cache); |
651f5fa2 | 2250 | process_deferred_cells(cache); |
65790ff9 | 2251 | process_invalidation_requests(cache); |
66cb1910 | 2252 | } |
c6b4fcba | 2253 | |
7ae34e77 | 2254 | process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard); |
c6b4fcba JT |
2255 | process_migrations(cache, &cache->completed_migrations, complete_migration); |
2256 | ||
c6b4fcba JT |
2257 | if (commit_if_needed(cache)) { |
2258 | process_deferred_flush_bios(cache, false); | |
304affaa | 2259 | process_migrations(cache, &cache->need_commit_migrations, migration_failure); |
c6b4fcba JT |
2260 | } else { |
2261 | process_deferred_flush_bios(cache, true); | |
2262 | process_migrations(cache, &cache->need_commit_migrations, | |
2263 | migration_success_post_commit); | |
2264 | } | |
66cb1910 JT |
2265 | |
2266 | ack_quiescing(cache); | |
2267 | ||
c6b4fcba JT |
2268 | } while (more_work(cache)); |
2269 | } | |
2270 | ||
2271 | /* | |
2272 | * We want to commit periodically so that not too much | |
2273 | * unwritten metadata builds up. | |
2274 | */ | |
2275 | static void do_waker(struct work_struct *ws) | |
2276 | { | |
2277 | struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker); | |
fba10109 | 2278 | policy_tick(cache->policy, true); |
c6b4fcba JT |
2279 | wake_worker(cache); |
2280 | queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD); | |
2281 | } | |
2282 | ||
2283 | /*----------------------------------------------------------------*/ | |
2284 | ||
2285 | static int is_congested(struct dm_dev *dev, int bdi_bits) | |
2286 | { | |
2287 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
2288 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
2289 | } | |
2290 | ||
2291 | static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2292 | { | |
2293 | struct cache *cache = container_of(cb, struct cache, callbacks); | |
2294 | ||
2295 | return is_congested(cache->origin_dev, bdi_bits) || | |
2296 | is_congested(cache->cache_dev, bdi_bits); | |
2297 | } | |
2298 | ||
2299 | /*---------------------------------------------------------------- | |
2300 | * Target methods | |
2301 | *--------------------------------------------------------------*/ | |
2302 | ||
2303 | /* | |
2304 | * This function gets called on the error paths of the constructor, so we | |
2305 | * have to cope with a partially initialised struct. | |
2306 | */ | |
2307 | static void destroy(struct cache *cache) | |
2308 | { | |
2309 | unsigned i; | |
2310 | ||
c6b4fcba JT |
2311 | if (cache->migration_pool) |
2312 | mempool_destroy(cache->migration_pool); | |
2313 | ||
2314 | if (cache->all_io_ds) | |
2315 | dm_deferred_set_destroy(cache->all_io_ds); | |
2316 | ||
2317 | if (cache->prison) | |
2318 | dm_bio_prison_destroy(cache->prison); | |
2319 | ||
2320 | if (cache->wq) | |
2321 | destroy_workqueue(cache->wq); | |
2322 | ||
2323 | if (cache->dirty_bitset) | |
2324 | free_bitset(cache->dirty_bitset); | |
2325 | ||
2326 | if (cache->discard_bitset) | |
2327 | free_bitset(cache->discard_bitset); | |
2328 | ||
2329 | if (cache->copier) | |
2330 | dm_kcopyd_client_destroy(cache->copier); | |
2331 | ||
2332 | if (cache->cmd) | |
2333 | dm_cache_metadata_close(cache->cmd); | |
2334 | ||
2335 | if (cache->metadata_dev) | |
2336 | dm_put_device(cache->ti, cache->metadata_dev); | |
2337 | ||
2338 | if (cache->origin_dev) | |
2339 | dm_put_device(cache->ti, cache->origin_dev); | |
2340 | ||
2341 | if (cache->cache_dev) | |
2342 | dm_put_device(cache->ti, cache->cache_dev); | |
2343 | ||
2344 | if (cache->policy) | |
2345 | dm_cache_policy_destroy(cache->policy); | |
2346 | ||
2347 | for (i = 0; i < cache->nr_ctr_args ; i++) | |
2348 | kfree(cache->ctr_args[i]); | |
2349 | kfree(cache->ctr_args); | |
2350 | ||
2351 | kfree(cache); | |
2352 | } | |
2353 | ||
2354 | static void cache_dtr(struct dm_target *ti) | |
2355 | { | |
2356 | struct cache *cache = ti->private; | |
2357 | ||
2358 | destroy(cache); | |
2359 | } | |
2360 | ||
2361 | static sector_t get_dev_size(struct dm_dev *dev) | |
2362 | { | |
2363 | return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
2364 | } | |
2365 | ||
2366 | /*----------------------------------------------------------------*/ | |
2367 | ||
2368 | /* | |
2369 | * Construct a cache device mapping. | |
2370 | * | |
2371 | * cache <metadata dev> <cache dev> <origin dev> <block size> | |
2372 | * <#feature args> [<feature arg>]* | |
2373 | * <policy> <#policy args> [<policy arg>]* | |
2374 | * | |
2375 | * metadata dev : fast device holding the persistent metadata | |
2376 | * cache dev : fast device holding cached data blocks | |
2377 | * origin dev : slow device holding original data blocks | |
2378 | * block size : cache unit size in sectors | |
2379 | * | |
2380 | * #feature args : number of feature arguments passed | |
2381 | * feature args : writethrough. (The default is writeback.) | |
2382 | * | |
2383 | * policy : the replacement policy to use | |
2384 | * #policy args : an even number of policy arguments corresponding | |
2385 | * to key/value pairs passed to the policy | |
2386 | * policy args : key/value pairs passed to the policy | |
2387 | * E.g. 'sequential_threshold 1024' | |
2388 | * See cache-policies.txt for details. | |
2389 | * | |
2390 | * Optional feature arguments are: | |
2391 | * writethrough : write through caching that prohibits cache block | |
2392 | * content from being different from origin block content. | |
2393 | * Without this argument, the default behaviour is to write | |
2394 | * back cache block contents later for performance reasons, | |
2395 | * so they may differ from the corresponding origin blocks. | |
2396 | */ | |
2397 | struct cache_args { | |
2398 | struct dm_target *ti; | |
2399 | ||
2400 | struct dm_dev *metadata_dev; | |
2401 | ||
2402 | struct dm_dev *cache_dev; | |
2403 | sector_t cache_sectors; | |
2404 | ||
2405 | struct dm_dev *origin_dev; | |
2406 | sector_t origin_sectors; | |
2407 | ||
2408 | uint32_t block_size; | |
2409 | ||
2410 | const char *policy_name; | |
2411 | int policy_argc; | |
2412 | const char **policy_argv; | |
2413 | ||
2414 | struct cache_features features; | |
2415 | }; | |
2416 | ||
2417 | static void destroy_cache_args(struct cache_args *ca) | |
2418 | { | |
2419 | if (ca->metadata_dev) | |
2420 | dm_put_device(ca->ti, ca->metadata_dev); | |
2421 | ||
2422 | if (ca->cache_dev) | |
2423 | dm_put_device(ca->ti, ca->cache_dev); | |
2424 | ||
2425 | if (ca->origin_dev) | |
2426 | dm_put_device(ca->ti, ca->origin_dev); | |
2427 | ||
2428 | kfree(ca); | |
2429 | } | |
2430 | ||
2431 | static bool at_least_one_arg(struct dm_arg_set *as, char **error) | |
2432 | { | |
2433 | if (!as->argc) { | |
2434 | *error = "Insufficient args"; | |
2435 | return false; | |
2436 | } | |
2437 | ||
2438 | return true; | |
2439 | } | |
2440 | ||
2441 | static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2442 | char **error) | |
2443 | { | |
2444 | int r; | |
2445 | sector_t metadata_dev_size; | |
2446 | char b[BDEVNAME_SIZE]; | |
2447 | ||
2448 | if (!at_least_one_arg(as, error)) | |
2449 | return -EINVAL; | |
2450 | ||
2451 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2452 | &ca->metadata_dev); | |
2453 | if (r) { | |
2454 | *error = "Error opening metadata device"; | |
2455 | return r; | |
2456 | } | |
2457 | ||
2458 | metadata_dev_size = get_dev_size(ca->metadata_dev); | |
2459 | if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING) | |
2460 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
2461 | bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
2462 | ||
2463 | return 0; | |
2464 | } | |
2465 | ||
2466 | static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2467 | char **error) | |
2468 | { | |
2469 | int r; | |
2470 | ||
2471 | if (!at_least_one_arg(as, error)) | |
2472 | return -EINVAL; | |
2473 | ||
2474 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2475 | &ca->cache_dev); | |
2476 | if (r) { | |
2477 | *error = "Error opening cache device"; | |
2478 | return r; | |
2479 | } | |
2480 | ca->cache_sectors = get_dev_size(ca->cache_dev); | |
2481 | ||
2482 | return 0; | |
2483 | } | |
2484 | ||
2485 | static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2486 | char **error) | |
2487 | { | |
2488 | int r; | |
2489 | ||
2490 | if (!at_least_one_arg(as, error)) | |
2491 | return -EINVAL; | |
2492 | ||
2493 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2494 | &ca->origin_dev); | |
2495 | if (r) { | |
2496 | *error = "Error opening origin device"; | |
2497 | return r; | |
2498 | } | |
2499 | ||
2500 | ca->origin_sectors = get_dev_size(ca->origin_dev); | |
2501 | if (ca->ti->len > ca->origin_sectors) { | |
2502 | *error = "Device size larger than cached device"; | |
2503 | return -EINVAL; | |
2504 | } | |
2505 | ||
2506 | return 0; | |
2507 | } | |
2508 | ||
2509 | static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as, | |
2510 | char **error) | |
2511 | { | |
05473044 | 2512 | unsigned long block_size; |
c6b4fcba JT |
2513 | |
2514 | if (!at_least_one_arg(as, error)) | |
2515 | return -EINVAL; | |
2516 | ||
05473044 MS |
2517 | if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size || |
2518 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2519 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
2520 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { | |
c6b4fcba JT |
2521 | *error = "Invalid data block size"; |
2522 | return -EINVAL; | |
2523 | } | |
2524 | ||
05473044 | 2525 | if (block_size > ca->cache_sectors) { |
c6b4fcba JT |
2526 | *error = "Data block size is larger than the cache device"; |
2527 | return -EINVAL; | |
2528 | } | |
2529 | ||
05473044 | 2530 | ca->block_size = block_size; |
c6b4fcba JT |
2531 | |
2532 | return 0; | |
2533 | } | |
2534 | ||
2535 | static void init_features(struct cache_features *cf) | |
2536 | { | |
2537 | cf->mode = CM_WRITE; | |
2ee57d58 | 2538 | cf->io_mode = CM_IO_WRITEBACK; |
c6b4fcba JT |
2539 | } |
2540 | ||
2541 | static int parse_features(struct cache_args *ca, struct dm_arg_set *as, | |
2542 | char **error) | |
2543 | { | |
2544 | static struct dm_arg _args[] = { | |
2545 | {0, 1, "Invalid number of cache feature arguments"}, | |
2546 | }; | |
2547 | ||
2548 | int r; | |
2549 | unsigned argc; | |
2550 | const char *arg; | |
2551 | struct cache_features *cf = &ca->features; | |
2552 | ||
2553 | init_features(cf); | |
2554 | ||
2555 | r = dm_read_arg_group(_args, as, &argc, error); | |
2556 | if (r) | |
2557 | return -EINVAL; | |
2558 | ||
2559 | while (argc--) { | |
2560 | arg = dm_shift_arg(as); | |
2561 | ||
2562 | if (!strcasecmp(arg, "writeback")) | |
2ee57d58 | 2563 | cf->io_mode = CM_IO_WRITEBACK; |
c6b4fcba JT |
2564 | |
2565 | else if (!strcasecmp(arg, "writethrough")) | |
2ee57d58 JT |
2566 | cf->io_mode = CM_IO_WRITETHROUGH; |
2567 | ||
2568 | else if (!strcasecmp(arg, "passthrough")) | |
2569 | cf->io_mode = CM_IO_PASSTHROUGH; | |
c6b4fcba JT |
2570 | |
2571 | else { | |
2572 | *error = "Unrecognised cache feature requested"; | |
2573 | return -EINVAL; | |
2574 | } | |
2575 | } | |
2576 | ||
2577 | return 0; | |
2578 | } | |
2579 | ||
2580 | static int parse_policy(struct cache_args *ca, struct dm_arg_set *as, | |
2581 | char **error) | |
2582 | { | |
2583 | static struct dm_arg _args[] = { | |
2584 | {0, 1024, "Invalid number of policy arguments"}, | |
2585 | }; | |
2586 | ||
2587 | int r; | |
2588 | ||
2589 | if (!at_least_one_arg(as, error)) | |
2590 | return -EINVAL; | |
2591 | ||
2592 | ca->policy_name = dm_shift_arg(as); | |
2593 | ||
2594 | r = dm_read_arg_group(_args, as, &ca->policy_argc, error); | |
2595 | if (r) | |
2596 | return -EINVAL; | |
2597 | ||
2598 | ca->policy_argv = (const char **)as->argv; | |
2599 | dm_consume_args(as, ca->policy_argc); | |
2600 | ||
2601 | return 0; | |
2602 | } | |
2603 | ||
2604 | static int parse_cache_args(struct cache_args *ca, int argc, char **argv, | |
2605 | char **error) | |
2606 | { | |
2607 | int r; | |
2608 | struct dm_arg_set as; | |
2609 | ||
2610 | as.argc = argc; | |
2611 | as.argv = argv; | |
2612 | ||
2613 | r = parse_metadata_dev(ca, &as, error); | |
2614 | if (r) | |
2615 | return r; | |
2616 | ||
2617 | r = parse_cache_dev(ca, &as, error); | |
2618 | if (r) | |
2619 | return r; | |
2620 | ||
2621 | r = parse_origin_dev(ca, &as, error); | |
2622 | if (r) | |
2623 | return r; | |
2624 | ||
2625 | r = parse_block_size(ca, &as, error); | |
2626 | if (r) | |
2627 | return r; | |
2628 | ||
2629 | r = parse_features(ca, &as, error); | |
2630 | if (r) | |
2631 | return r; | |
2632 | ||
2633 | r = parse_policy(ca, &as, error); | |
2634 | if (r) | |
2635 | return r; | |
2636 | ||
2637 | return 0; | |
2638 | } | |
2639 | ||
2640 | /*----------------------------------------------------------------*/ | |
2641 | ||
2642 | static struct kmem_cache *migration_cache; | |
2643 | ||
2c73c471 AK |
2644 | #define NOT_CORE_OPTION 1 |
2645 | ||
2f14f4b5 | 2646 | static int process_config_option(struct cache *cache, const char *key, const char *value) |
2c73c471 AK |
2647 | { |
2648 | unsigned long tmp; | |
2649 | ||
2f14f4b5 JT |
2650 | if (!strcasecmp(key, "migration_threshold")) { |
2651 | if (kstrtoul(value, 10, &tmp)) | |
2c73c471 AK |
2652 | return -EINVAL; |
2653 | ||
2654 | cache->migration_threshold = tmp; | |
2655 | return 0; | |
2656 | } | |
2657 | ||
2658 | return NOT_CORE_OPTION; | |
2659 | } | |
2660 | ||
2f14f4b5 JT |
2661 | static int set_config_value(struct cache *cache, const char *key, const char *value) |
2662 | { | |
2663 | int r = process_config_option(cache, key, value); | |
2664 | ||
2665 | if (r == NOT_CORE_OPTION) | |
2666 | r = policy_set_config_value(cache->policy, key, value); | |
2667 | ||
2668 | if (r) | |
2669 | DMWARN("bad config value for %s: %s", key, value); | |
2670 | ||
2671 | return r; | |
2672 | } | |
2673 | ||
2674 | static int set_config_values(struct cache *cache, int argc, const char **argv) | |
c6b4fcba JT |
2675 | { |
2676 | int r = 0; | |
2677 | ||
2678 | if (argc & 1) { | |
2679 | DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs."); | |
2680 | return -EINVAL; | |
2681 | } | |
2682 | ||
2683 | while (argc) { | |
2f14f4b5 JT |
2684 | r = set_config_value(cache, argv[0], argv[1]); |
2685 | if (r) | |
2686 | break; | |
c6b4fcba JT |
2687 | |
2688 | argc -= 2; | |
2689 | argv += 2; | |
2690 | } | |
2691 | ||
2692 | return r; | |
2693 | } | |
2694 | ||
2695 | static int create_cache_policy(struct cache *cache, struct cache_args *ca, | |
2696 | char **error) | |
2697 | { | |
4cb3e1db MP |
2698 | struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name, |
2699 | cache->cache_size, | |
2700 | cache->origin_sectors, | |
2701 | cache->sectors_per_block); | |
2702 | if (IS_ERR(p)) { | |
c6b4fcba | 2703 | *error = "Error creating cache's policy"; |
4cb3e1db | 2704 | return PTR_ERR(p); |
c6b4fcba | 2705 | } |
4cb3e1db | 2706 | cache->policy = p; |
c6b4fcba | 2707 | |
2f14f4b5 | 2708 | return 0; |
c6b4fcba JT |
2709 | } |
2710 | ||
08b18451 | 2711 | /* |
2bb812df JT |
2712 | * We want the discard block size to be at least the size of the cache |
2713 | * block size and have no more than 2^14 discard blocks across the origin. | |
08b18451 JT |
2714 | */ |
2715 | #define MAX_DISCARD_BLOCKS (1 << 14) | |
2716 | ||
2717 | static bool too_many_discard_blocks(sector_t discard_block_size, | |
2718 | sector_t origin_size) | |
2719 | { | |
2720 | (void) sector_div(origin_size, discard_block_size); | |
2721 | ||
2722 | return origin_size > MAX_DISCARD_BLOCKS; | |
2723 | } | |
2724 | ||
2725 | static sector_t calculate_discard_block_size(sector_t cache_block_size, | |
2726 | sector_t origin_size) | |
2727 | { | |
2bb812df | 2728 | sector_t discard_block_size = cache_block_size; |
08b18451 JT |
2729 | |
2730 | if (origin_size) | |
2731 | while (too_many_discard_blocks(discard_block_size, origin_size)) | |
2732 | discard_block_size *= 2; | |
2733 | ||
2734 | return discard_block_size; | |
2735 | } | |
2736 | ||
d1d9220c JT |
2737 | static void set_cache_size(struct cache *cache, dm_cblock_t size) |
2738 | { | |
2739 | dm_block_t nr_blocks = from_cblock(size); | |
2740 | ||
2741 | if (nr_blocks > (1 << 20) && cache->cache_size != size) | |
2742 | DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n" | |
2743 | "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n" | |
2744 | "Please consider increasing the cache block size to reduce the overall cache block count.", | |
2745 | (unsigned long long) nr_blocks); | |
2746 | ||
2747 | cache->cache_size = size; | |
2748 | } | |
2749 | ||
f8350daf | 2750 | #define DEFAULT_MIGRATION_THRESHOLD 2048 |
c6b4fcba | 2751 | |
c6b4fcba JT |
2752 | static int cache_create(struct cache_args *ca, struct cache **result) |
2753 | { | |
2754 | int r = 0; | |
2755 | char **error = &ca->ti->error; | |
2756 | struct cache *cache; | |
2757 | struct dm_target *ti = ca->ti; | |
2758 | dm_block_t origin_blocks; | |
2759 | struct dm_cache_metadata *cmd; | |
2760 | bool may_format = ca->features.mode == CM_WRITE; | |
2761 | ||
2762 | cache = kzalloc(sizeof(*cache), GFP_KERNEL); | |
2763 | if (!cache) | |
2764 | return -ENOMEM; | |
2765 | ||
2766 | cache->ti = ca->ti; | |
2767 | ti->private = cache; | |
c6b4fcba JT |
2768 | ti->num_flush_bios = 2; |
2769 | ti->flush_supported = true; | |
2770 | ||
2771 | ti->num_discard_bios = 1; | |
2772 | ti->discards_supported = true; | |
2773 | ti->discard_zeroes_data_unsupported = true; | |
2572629a | 2774 | ti->split_discard_bios = false; |
c6b4fcba | 2775 | |
8c5008fa | 2776 | cache->features = ca->features; |
19b0092e | 2777 | ti->per_bio_data_size = get_per_bio_data_size(cache); |
c6b4fcba | 2778 | |
c6b4fcba JT |
2779 | cache->callbacks.congested_fn = cache_is_congested; |
2780 | dm_table_add_target_callbacks(ti->table, &cache->callbacks); | |
2781 | ||
2782 | cache->metadata_dev = ca->metadata_dev; | |
2783 | cache->origin_dev = ca->origin_dev; | |
2784 | cache->cache_dev = ca->cache_dev; | |
2785 | ||
2786 | ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL; | |
2787 | ||
2788 | /* FIXME: factor out this whole section */ | |
2789 | origin_blocks = cache->origin_sectors = ca->origin_sectors; | |
414dd67d | 2790 | origin_blocks = block_div(origin_blocks, ca->block_size); |
c6b4fcba JT |
2791 | cache->origin_blocks = to_oblock(origin_blocks); |
2792 | ||
2793 | cache->sectors_per_block = ca->block_size; | |
2794 | if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) { | |
2795 | r = -EINVAL; | |
2796 | goto bad; | |
2797 | } | |
2798 | ||
2799 | if (ca->block_size & (ca->block_size - 1)) { | |
2800 | dm_block_t cache_size = ca->cache_sectors; | |
2801 | ||
2802 | cache->sectors_per_block_shift = -1; | |
414dd67d | 2803 | cache_size = block_div(cache_size, ca->block_size); |
d1d9220c | 2804 | set_cache_size(cache, to_cblock(cache_size)); |
c6b4fcba JT |
2805 | } else { |
2806 | cache->sectors_per_block_shift = __ffs(ca->block_size); | |
d1d9220c | 2807 | set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift)); |
c6b4fcba JT |
2808 | } |
2809 | ||
2810 | r = create_cache_policy(cache, ca, error); | |
2811 | if (r) | |
2812 | goto bad; | |
2f14f4b5 | 2813 | |
c6b4fcba | 2814 | cache->policy_nr_args = ca->policy_argc; |
2f14f4b5 JT |
2815 | cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD; |
2816 | ||
2817 | r = set_config_values(cache, ca->policy_argc, ca->policy_argv); | |
2818 | if (r) { | |
2819 | *error = "Error setting cache policy's config values"; | |
2820 | goto bad; | |
2821 | } | |
c6b4fcba JT |
2822 | |
2823 | cmd = dm_cache_metadata_open(cache->metadata_dev->bdev, | |
2824 | ca->block_size, may_format, | |
2825 | dm_cache_policy_get_hint_size(cache->policy)); | |
2826 | if (IS_ERR(cmd)) { | |
2827 | *error = "Error creating metadata object"; | |
2828 | r = PTR_ERR(cmd); | |
2829 | goto bad; | |
2830 | } | |
2831 | cache->cmd = cmd; | |
028ae9f7 JT |
2832 | set_cache_mode(cache, CM_WRITE); |
2833 | if (get_cache_mode(cache) != CM_WRITE) { | |
2834 | *error = "Unable to get write access to metadata, please check/repair metadata."; | |
2835 | r = -EINVAL; | |
2836 | goto bad; | |
2837 | } | |
c6b4fcba | 2838 | |
2ee57d58 JT |
2839 | if (passthrough_mode(&cache->features)) { |
2840 | bool all_clean; | |
2841 | ||
2842 | r = dm_cache_metadata_all_clean(cache->cmd, &all_clean); | |
2843 | if (r) { | |
2844 | *error = "dm_cache_metadata_all_clean() failed"; | |
2845 | goto bad; | |
2846 | } | |
2847 | ||
2848 | if (!all_clean) { | |
2849 | *error = "Cannot enter passthrough mode unless all blocks are clean"; | |
2850 | r = -EINVAL; | |
2851 | goto bad; | |
2852 | } | |
2853 | } | |
2854 | ||
c6b4fcba | 2855 | spin_lock_init(&cache->lock); |
651f5fa2 | 2856 | INIT_LIST_HEAD(&cache->deferred_cells); |
c6b4fcba JT |
2857 | bio_list_init(&cache->deferred_bios); |
2858 | bio_list_init(&cache->deferred_flush_bios); | |
e2e74d61 | 2859 | bio_list_init(&cache->deferred_writethrough_bios); |
c6b4fcba JT |
2860 | INIT_LIST_HEAD(&cache->quiesced_migrations); |
2861 | INIT_LIST_HEAD(&cache->completed_migrations); | |
2862 | INIT_LIST_HEAD(&cache->need_commit_migrations); | |
a59db676 JT |
2863 | atomic_set(&cache->nr_allocated_migrations, 0); |
2864 | atomic_set(&cache->nr_io_migrations, 0); | |
c6b4fcba JT |
2865 | init_waitqueue_head(&cache->migration_wait); |
2866 | ||
66cb1910 | 2867 | init_waitqueue_head(&cache->quiescing_wait); |
238f8363 | 2868 | atomic_set(&cache->quiescing, 0); |
66cb1910 JT |
2869 | atomic_set(&cache->quiescing_ack, 0); |
2870 | ||
fa4d683a | 2871 | r = -ENOMEM; |
44fa816b | 2872 | atomic_set(&cache->nr_dirty, 0); |
c6b4fcba JT |
2873 | cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size)); |
2874 | if (!cache->dirty_bitset) { | |
2875 | *error = "could not allocate dirty bitset"; | |
2876 | goto bad; | |
2877 | } | |
2878 | clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size)); | |
2879 | ||
08b18451 JT |
2880 | cache->discard_block_size = |
2881 | calculate_discard_block_size(cache->sectors_per_block, | |
2882 | cache->origin_sectors); | |
2572629a JT |
2883 | cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors, |
2884 | cache->discard_block_size)); | |
1bad9bc4 | 2885 | cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2886 | if (!cache->discard_bitset) { |
2887 | *error = "could not allocate discard bitset"; | |
2888 | goto bad; | |
2889 | } | |
1bad9bc4 | 2890 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2891 | |
2892 | cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); | |
2893 | if (IS_ERR(cache->copier)) { | |
2894 | *error = "could not create kcopyd client"; | |
2895 | r = PTR_ERR(cache->copier); | |
2896 | goto bad; | |
2897 | } | |
2898 | ||
2899 | cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2900 | if (!cache->wq) { | |
2901 | *error = "could not create workqueue for metadata object"; | |
2902 | goto bad; | |
2903 | } | |
2904 | INIT_WORK(&cache->worker, do_worker); | |
2905 | INIT_DELAYED_WORK(&cache->waker, do_waker); | |
2906 | cache->last_commit_jiffies = jiffies; | |
2907 | ||
a195db2d | 2908 | cache->prison = dm_bio_prison_create(); |
c6b4fcba JT |
2909 | if (!cache->prison) { |
2910 | *error = "could not create bio prison"; | |
2911 | goto bad; | |
2912 | } | |
2913 | ||
2914 | cache->all_io_ds = dm_deferred_set_create(); | |
2915 | if (!cache->all_io_ds) { | |
2916 | *error = "could not create all_io deferred set"; | |
2917 | goto bad; | |
2918 | } | |
2919 | ||
2920 | cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE, | |
2921 | migration_cache); | |
2922 | if (!cache->migration_pool) { | |
2923 | *error = "Error creating cache's migration mempool"; | |
2924 | goto bad; | |
2925 | } | |
2926 | ||
c6b4fcba JT |
2927 | cache->need_tick_bio = true; |
2928 | cache->sized = false; | |
65790ff9 | 2929 | cache->invalidate = false; |
c6b4fcba JT |
2930 | cache->commit_requested = false; |
2931 | cache->loaded_mappings = false; | |
2932 | cache->loaded_discards = false; | |
2933 | ||
2934 | load_stats(cache); | |
2935 | ||
2936 | atomic_set(&cache->stats.demotion, 0); | |
2937 | atomic_set(&cache->stats.promotion, 0); | |
2938 | atomic_set(&cache->stats.copies_avoided, 0); | |
2939 | atomic_set(&cache->stats.cache_cell_clash, 0); | |
2940 | atomic_set(&cache->stats.commit_count, 0); | |
2941 | atomic_set(&cache->stats.discard_count, 0); | |
2942 | ||
65790ff9 JT |
2943 | spin_lock_init(&cache->invalidation_lock); |
2944 | INIT_LIST_HEAD(&cache->invalidation_requests); | |
2945 | ||
066dbaa3 JT |
2946 | iot_init(&cache->origin_tracker); |
2947 | ||
c6b4fcba JT |
2948 | *result = cache; |
2949 | return 0; | |
2950 | ||
2951 | bad: | |
2952 | destroy(cache); | |
2953 | return r; | |
2954 | } | |
2955 | ||
2956 | static int copy_ctr_args(struct cache *cache, int argc, const char **argv) | |
2957 | { | |
2958 | unsigned i; | |
2959 | const char **copy; | |
2960 | ||
2961 | copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL); | |
2962 | if (!copy) | |
2963 | return -ENOMEM; | |
2964 | for (i = 0; i < argc; i++) { | |
2965 | copy[i] = kstrdup(argv[i], GFP_KERNEL); | |
2966 | if (!copy[i]) { | |
2967 | while (i--) | |
2968 | kfree(copy[i]); | |
2969 | kfree(copy); | |
2970 | return -ENOMEM; | |
2971 | } | |
2972 | } | |
2973 | ||
2974 | cache->nr_ctr_args = argc; | |
2975 | cache->ctr_args = copy; | |
2976 | ||
2977 | return 0; | |
2978 | } | |
2979 | ||
2980 | static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2981 | { | |
2982 | int r = -EINVAL; | |
2983 | struct cache_args *ca; | |
2984 | struct cache *cache = NULL; | |
2985 | ||
2986 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
2987 | if (!ca) { | |
2988 | ti->error = "Error allocating memory for cache"; | |
2989 | return -ENOMEM; | |
2990 | } | |
2991 | ca->ti = ti; | |
2992 | ||
2993 | r = parse_cache_args(ca, argc, argv, &ti->error); | |
2994 | if (r) | |
2995 | goto out; | |
2996 | ||
2997 | r = cache_create(ca, &cache); | |
617a0b89 HM |
2998 | if (r) |
2999 | goto out; | |
c6b4fcba JT |
3000 | |
3001 | r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3); | |
3002 | if (r) { | |
3003 | destroy(cache); | |
3004 | goto out; | |
3005 | } | |
3006 | ||
3007 | ti->private = cache; | |
3008 | ||
3009 | out: | |
3010 | destroy_cache_args(ca); | |
3011 | return r; | |
3012 | } | |
3013 | ||
651f5fa2 JT |
3014 | /*----------------------------------------------------------------*/ |
3015 | ||
3016 | static int cache_map(struct dm_target *ti, struct bio *bio) | |
c6b4fcba | 3017 | { |
651f5fa2 JT |
3018 | struct cache *cache = ti->private; |
3019 | ||
c6b4fcba | 3020 | int r; |
651f5fa2 | 3021 | struct dm_bio_prison_cell *cell = NULL; |
c6b4fcba | 3022 | dm_oblock_t block = get_bio_block(cache, bio); |
19b0092e | 3023 | size_t pb_data_size = get_per_bio_data_size(cache); |
c6b4fcba | 3024 | bool can_migrate = false; |
40775257 | 3025 | bool fast_promotion; |
c6b4fcba | 3026 | struct policy_result lookup_result; |
e893fba9 | 3027 | struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size); |
fb4100ae JT |
3028 | struct old_oblock_lock ool; |
3029 | ||
3030 | ool.locker.fn = null_locker; | |
c6b4fcba | 3031 | |
e893fba9 | 3032 | if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) { |
c6b4fcba JT |
3033 | /* |
3034 | * This can only occur if the io goes to a partial block at | |
3035 | * the end of the origin device. We don't cache these. | |
3036 | * Just remap to the origin and carry on. | |
3037 | */ | |
e893fba9 | 3038 | remap_to_origin(cache, bio); |
651f5fa2 | 3039 | accounted_begin(cache, bio); |
c6b4fcba JT |
3040 | return DM_MAPIO_REMAPPED; |
3041 | } | |
3042 | ||
651f5fa2 | 3043 | if (discard_or_flush(bio)) { |
c6b4fcba JT |
3044 | defer_bio(cache, bio); |
3045 | return DM_MAPIO_SUBMITTED; | |
3046 | } | |
3047 | ||
3048 | /* | |
3049 | * Check to see if that block is currently migrating. | |
3050 | */ | |
651f5fa2 JT |
3051 | cell = alloc_prison_cell(cache); |
3052 | if (!cell) { | |
c6b4fcba JT |
3053 | defer_bio(cache, bio); |
3054 | return DM_MAPIO_SUBMITTED; | |
3055 | } | |
3056 | ||
651f5fa2 | 3057 | r = bio_detain(cache, block, bio, cell, |
c6b4fcba | 3058 | (cell_free_fn) free_prison_cell, |
651f5fa2 | 3059 | cache, &cell); |
c6b4fcba JT |
3060 | if (r) { |
3061 | if (r < 0) | |
3062 | defer_bio(cache, bio); | |
3063 | ||
3064 | return DM_MAPIO_SUBMITTED; | |
3065 | } | |
3066 | ||
40775257 | 3067 | fast_promotion = is_discarded_oblock(cache, block) || bio_writes_complete_block(cache, bio); |
c6b4fcba | 3068 | |
40775257 | 3069 | r = policy_map(cache->policy, block, false, can_migrate, fast_promotion, |
fb4100ae | 3070 | bio, &ool.locker, &lookup_result); |
c6b4fcba | 3071 | if (r == -EWOULDBLOCK) { |
651f5fa2 | 3072 | cell_defer(cache, cell, true); |
c6b4fcba JT |
3073 | return DM_MAPIO_SUBMITTED; |
3074 | ||
3075 | } else if (r) { | |
b61d9509 MS |
3076 | DMERR_LIMIT("%s: Unexpected return from cache replacement policy: %d", |
3077 | cache_device_name(cache), r); | |
651f5fa2 | 3078 | cell_defer(cache, cell, false); |
c6b4fcba JT |
3079 | bio_io_error(bio); |
3080 | return DM_MAPIO_SUBMITTED; | |
3081 | } | |
3082 | ||
2ee57d58 | 3083 | r = DM_MAPIO_REMAPPED; |
c6b4fcba JT |
3084 | switch (lookup_result.op) { |
3085 | case POLICY_HIT: | |
2ee57d58 JT |
3086 | if (passthrough_mode(&cache->features)) { |
3087 | if (bio_data_dir(bio) == WRITE) { | |
3088 | /* | |
3089 | * We need to invalidate this block, so | |
3090 | * defer for the worker thread. | |
3091 | */ | |
651f5fa2 | 3092 | cell_defer(cache, cell, true); |
2ee57d58 JT |
3093 | r = DM_MAPIO_SUBMITTED; |
3094 | ||
3095 | } else { | |
2ee57d58 JT |
3096 | inc_miss_counter(cache, bio); |
3097 | remap_to_origin_clear_discard(cache, bio, block); | |
651f5fa2 JT |
3098 | accounted_begin(cache, bio); |
3099 | inc_ds(cache, bio, cell); | |
3100 | // FIXME: we want to remap hits or misses straight | |
3101 | // away rather than passing over to the worker. | |
3102 | cell_defer(cache, cell, false); | |
2ee57d58 | 3103 | } |
c6b4fcba | 3104 | |
2ee57d58 JT |
3105 | } else { |
3106 | inc_hit_counter(cache, bio); | |
2ee57d58 | 3107 | if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) && |
651f5fa2 | 3108 | !is_dirty(cache, lookup_result.cblock)) { |
2ee57d58 | 3109 | remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock); |
651f5fa2 JT |
3110 | accounted_begin(cache, bio); |
3111 | inc_ds(cache, bio, cell); | |
3112 | cell_defer(cache, cell, false); | |
3113 | ||
3114 | } else | |
3115 | remap_cell_to_cache_dirty(cache, cell, block, lookup_result.cblock, false); | |
2ee57d58 | 3116 | } |
c6b4fcba JT |
3117 | break; |
3118 | ||
3119 | case POLICY_MISS: | |
3120 | inc_miss_counter(cache, bio); | |
c6b4fcba JT |
3121 | if (pb->req_nr != 0) { |
3122 | /* | |
3123 | * This is a duplicate writethrough io that is no | |
3124 | * longer needed because the block has been demoted. | |
3125 | */ | |
3126 | bio_endio(bio, 0); | |
651f5fa2 JT |
3127 | // FIXME: remap everything as a miss |
3128 | cell_defer(cache, cell, false); | |
8c081b52 JT |
3129 | r = DM_MAPIO_SUBMITTED; |
3130 | ||
3131 | } else | |
651f5fa2 | 3132 | remap_cell_to_origin_clear_discard(cache, cell, block, false); |
c6b4fcba JT |
3133 | break; |
3134 | ||
3135 | default: | |
b61d9509 MS |
3136 | DMERR_LIMIT("%s: %s: erroring bio: unknown policy op: %u", |
3137 | cache_device_name(cache), __func__, | |
c6b4fcba | 3138 | (unsigned) lookup_result.op); |
651f5fa2 | 3139 | cell_defer(cache, cell, false); |
c6b4fcba | 3140 | bio_io_error(bio); |
2ee57d58 | 3141 | r = DM_MAPIO_SUBMITTED; |
c6b4fcba JT |
3142 | } |
3143 | ||
2ee57d58 | 3144 | return r; |
c6b4fcba JT |
3145 | } |
3146 | ||
3147 | static int cache_end_io(struct dm_target *ti, struct bio *bio, int error) | |
3148 | { | |
3149 | struct cache *cache = ti->private; | |
3150 | unsigned long flags; | |
19b0092e MS |
3151 | size_t pb_data_size = get_per_bio_data_size(cache); |
3152 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
3153 | |
3154 | if (pb->tick) { | |
fba10109 | 3155 | policy_tick(cache->policy, false); |
c6b4fcba JT |
3156 | |
3157 | spin_lock_irqsave(&cache->lock, flags); | |
3158 | cache->need_tick_bio = true; | |
3159 | spin_unlock_irqrestore(&cache->lock, flags); | |
3160 | } | |
3161 | ||
3162 | check_for_quiesced_migrations(cache, pb); | |
066dbaa3 | 3163 | accounted_complete(cache, bio); |
c6b4fcba JT |
3164 | |
3165 | return 0; | |
3166 | } | |
3167 | ||
3168 | static int write_dirty_bitset(struct cache *cache) | |
3169 | { | |
3170 | unsigned i, r; | |
3171 | ||
028ae9f7 JT |
3172 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
3173 | return -EINVAL; | |
3174 | ||
c6b4fcba JT |
3175 | for (i = 0; i < from_cblock(cache->cache_size); i++) { |
3176 | r = dm_cache_set_dirty(cache->cmd, to_cblock(i), | |
3177 | is_dirty(cache, to_cblock(i))); | |
028ae9f7 JT |
3178 | if (r) { |
3179 | metadata_operation_failed(cache, "dm_cache_set_dirty", r); | |
c6b4fcba | 3180 | return r; |
028ae9f7 | 3181 | } |
c6b4fcba JT |
3182 | } |
3183 | ||
3184 | return 0; | |
3185 | } | |
3186 | ||
3187 | static int write_discard_bitset(struct cache *cache) | |
3188 | { | |
3189 | unsigned i, r; | |
3190 | ||
028ae9f7 JT |
3191 | if (get_cache_mode(cache) >= CM_READ_ONLY) |
3192 | return -EINVAL; | |
3193 | ||
1bad9bc4 JT |
3194 | r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size, |
3195 | cache->discard_nr_blocks); | |
c6b4fcba | 3196 | if (r) { |
b61d9509 | 3197 | DMERR("%s: could not resize on-disk discard bitset", cache_device_name(cache)); |
028ae9f7 | 3198 | metadata_operation_failed(cache, "dm_cache_discard_bitset_resize", r); |
c6b4fcba JT |
3199 | return r; |
3200 | } | |
3201 | ||
1bad9bc4 JT |
3202 | for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) { |
3203 | r = dm_cache_set_discard(cache->cmd, to_dblock(i), | |
3204 | is_discarded(cache, to_dblock(i))); | |
028ae9f7 JT |
3205 | if (r) { |
3206 | metadata_operation_failed(cache, "dm_cache_set_discard", r); | |
c6b4fcba | 3207 | return r; |
028ae9f7 JT |
3208 | } |
3209 | } | |
3210 | ||
3211 | return 0; | |
3212 | } | |
3213 | ||
3214 | static int write_hints(struct cache *cache) | |
3215 | { | |
3216 | int r; | |
3217 | ||
3218 | if (get_cache_mode(cache) >= CM_READ_ONLY) | |
3219 | return -EINVAL; | |
3220 | ||
3221 | r = dm_cache_write_hints(cache->cmd, cache->policy); | |
3222 | if (r) { | |
3223 | metadata_operation_failed(cache, "dm_cache_write_hints", r); | |
3224 | return r; | |
c6b4fcba JT |
3225 | } |
3226 | ||
3227 | return 0; | |
3228 | } | |
3229 | ||
c6b4fcba JT |
3230 | /* |
3231 | * returns true on success | |
3232 | */ | |
3233 | static bool sync_metadata(struct cache *cache) | |
3234 | { | |
3235 | int r1, r2, r3, r4; | |
3236 | ||
3237 | r1 = write_dirty_bitset(cache); | |
3238 | if (r1) | |
b61d9509 | 3239 | DMERR("%s: could not write dirty bitset", cache_device_name(cache)); |
c6b4fcba JT |
3240 | |
3241 | r2 = write_discard_bitset(cache); | |
3242 | if (r2) | |
b61d9509 | 3243 | DMERR("%s: could not write discard bitset", cache_device_name(cache)); |
c6b4fcba JT |
3244 | |
3245 | save_stats(cache); | |
3246 | ||
028ae9f7 | 3247 | r3 = write_hints(cache); |
c6b4fcba | 3248 | if (r3) |
b61d9509 | 3249 | DMERR("%s: could not write hints", cache_device_name(cache)); |
c6b4fcba JT |
3250 | |
3251 | /* | |
3252 | * If writing the above metadata failed, we still commit, but don't | |
3253 | * set the clean shutdown flag. This will effectively force every | |
3254 | * dirty bit to be set on reload. | |
3255 | */ | |
028ae9f7 | 3256 | r4 = commit(cache, !r1 && !r2 && !r3); |
c6b4fcba | 3257 | if (r4) |
b61d9509 | 3258 | DMERR("%s: could not write cache metadata", cache_device_name(cache)); |
c6b4fcba JT |
3259 | |
3260 | return !r1 && !r2 && !r3 && !r4; | |
3261 | } | |
3262 | ||
3263 | static void cache_postsuspend(struct dm_target *ti) | |
3264 | { | |
3265 | struct cache *cache = ti->private; | |
3266 | ||
3267 | start_quiescing(cache); | |
3268 | wait_for_migrations(cache); | |
3269 | stop_worker(cache); | |
651f5fa2 JT |
3270 | requeue_deferred_bios(cache); |
3271 | requeue_deferred_cells(cache); | |
c6b4fcba JT |
3272 | stop_quiescing(cache); |
3273 | ||
028ae9f7 JT |
3274 | if (get_cache_mode(cache) == CM_WRITE) |
3275 | (void) sync_metadata(cache); | |
c6b4fcba JT |
3276 | } |
3277 | ||
3278 | static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock, | |
3279 | bool dirty, uint32_t hint, bool hint_valid) | |
3280 | { | |
3281 | int r; | |
3282 | struct cache *cache = context; | |
3283 | ||
3284 | r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid); | |
3285 | if (r) | |
3286 | return r; | |
3287 | ||
3288 | if (dirty) | |
3289 | set_dirty(cache, oblock, cblock); | |
3290 | else | |
3291 | clear_dirty(cache, oblock, cblock); | |
3292 | ||
3293 | return 0; | |
3294 | } | |
3295 | ||
3e2e1c30 JT |
3296 | /* |
3297 | * The discard block size in the on disk metadata is not | |
3298 | * neccessarily the same as we're currently using. So we have to | |
3299 | * be careful to only set the discarded attribute if we know it | |
3300 | * covers a complete block of the new size. | |
3301 | */ | |
3302 | struct discard_load_info { | |
3303 | struct cache *cache; | |
3304 | ||
3305 | /* | |
3306 | * These blocks are sized using the on disk dblock size, rather | |
3307 | * than the current one. | |
3308 | */ | |
3309 | dm_block_t block_size; | |
3310 | dm_block_t discard_begin, discard_end; | |
3311 | }; | |
3312 | ||
3313 | static void discard_load_info_init(struct cache *cache, | |
3314 | struct discard_load_info *li) | |
3315 | { | |
3316 | li->cache = cache; | |
3317 | li->discard_begin = li->discard_end = 0; | |
3318 | } | |
3319 | ||
3320 | static void set_discard_range(struct discard_load_info *li) | |
3321 | { | |
3322 | sector_t b, e; | |
3323 | ||
3324 | if (li->discard_begin == li->discard_end) | |
3325 | return; | |
3326 | ||
3327 | /* | |
3328 | * Convert to sectors. | |
3329 | */ | |
3330 | b = li->discard_begin * li->block_size; | |
3331 | e = li->discard_end * li->block_size; | |
3332 | ||
3333 | /* | |
3334 | * Then convert back to the current dblock size. | |
3335 | */ | |
3336 | b = dm_sector_div_up(b, li->cache->discard_block_size); | |
3337 | sector_div(e, li->cache->discard_block_size); | |
3338 | ||
3339 | /* | |
3340 | * The origin may have shrunk, so we need to check we're still in | |
3341 | * bounds. | |
3342 | */ | |
3343 | if (e > from_dblock(li->cache->discard_nr_blocks)) | |
3344 | e = from_dblock(li->cache->discard_nr_blocks); | |
3345 | ||
3346 | for (; b < e; b++) | |
3347 | set_discard(li->cache, to_dblock(b)); | |
3348 | } | |
3349 | ||
c6b4fcba | 3350 | static int load_discard(void *context, sector_t discard_block_size, |
1bad9bc4 | 3351 | dm_dblock_t dblock, bool discard) |
c6b4fcba | 3352 | { |
3e2e1c30 | 3353 | struct discard_load_info *li = context; |
c6b4fcba | 3354 | |
3e2e1c30 | 3355 | li->block_size = discard_block_size; |
1bad9bc4 | 3356 | |
3e2e1c30 JT |
3357 | if (discard) { |
3358 | if (from_dblock(dblock) == li->discard_end) | |
3359 | /* | |
3360 | * We're already in a discard range, just extend it. | |
3361 | */ | |
3362 | li->discard_end = li->discard_end + 1ULL; | |
3363 | ||
3364 | else { | |
3365 | /* | |
3366 | * Emit the old range and start a new one. | |
3367 | */ | |
3368 | set_discard_range(li); | |
3369 | li->discard_begin = from_dblock(dblock); | |
3370 | li->discard_end = li->discard_begin + 1ULL; | |
3371 | } | |
3372 | } else { | |
3373 | set_discard_range(li); | |
3374 | li->discard_begin = li->discard_end = 0; | |
3375 | } | |
c6b4fcba JT |
3376 | |
3377 | return 0; | |
3378 | } | |
3379 | ||
f494a9c6 JT |
3380 | static dm_cblock_t get_cache_dev_size(struct cache *cache) |
3381 | { | |
3382 | sector_t size = get_dev_size(cache->cache_dev); | |
3383 | (void) sector_div(size, cache->sectors_per_block); | |
3384 | return to_cblock(size); | |
3385 | } | |
3386 | ||
3387 | static bool can_resize(struct cache *cache, dm_cblock_t new_size) | |
3388 | { | |
3389 | if (from_cblock(new_size) > from_cblock(cache->cache_size)) | |
3390 | return true; | |
3391 | ||
3392 | /* | |
3393 | * We can't drop a dirty block when shrinking the cache. | |
3394 | */ | |
3395 | while (from_cblock(new_size) < from_cblock(cache->cache_size)) { | |
3396 | new_size = to_cblock(from_cblock(new_size) + 1); | |
3397 | if (is_dirty(cache, new_size)) { | |
b61d9509 MS |
3398 | DMERR("%s: unable to shrink cache; cache block %llu is dirty", |
3399 | cache_device_name(cache), | |
f494a9c6 JT |
3400 | (unsigned long long) from_cblock(new_size)); |
3401 | return false; | |
3402 | } | |
3403 | } | |
3404 | ||
3405 | return true; | |
3406 | } | |
3407 | ||
3408 | static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size) | |
3409 | { | |
3410 | int r; | |
3411 | ||
08844800 | 3412 | r = dm_cache_resize(cache->cmd, new_size); |
f494a9c6 | 3413 | if (r) { |
b61d9509 | 3414 | DMERR("%s: could not resize cache metadata", cache_device_name(cache)); |
028ae9f7 | 3415 | metadata_operation_failed(cache, "dm_cache_resize", r); |
f494a9c6 JT |
3416 | return r; |
3417 | } | |
3418 | ||
d1d9220c | 3419 | set_cache_size(cache, new_size); |
f494a9c6 JT |
3420 | |
3421 | return 0; | |
3422 | } | |
3423 | ||
c6b4fcba JT |
3424 | static int cache_preresume(struct dm_target *ti) |
3425 | { | |
3426 | int r = 0; | |
3427 | struct cache *cache = ti->private; | |
f494a9c6 | 3428 | dm_cblock_t csize = get_cache_dev_size(cache); |
c6b4fcba JT |
3429 | |
3430 | /* | |
3431 | * Check to see if the cache has resized. | |
3432 | */ | |
f494a9c6 JT |
3433 | if (!cache->sized) { |
3434 | r = resize_cache_dev(cache, csize); | |
3435 | if (r) | |
c6b4fcba | 3436 | return r; |
c6b4fcba JT |
3437 | |
3438 | cache->sized = true; | |
f494a9c6 JT |
3439 | |
3440 | } else if (csize != cache->cache_size) { | |
3441 | if (!can_resize(cache, csize)) | |
3442 | return -EINVAL; | |
3443 | ||
3444 | r = resize_cache_dev(cache, csize); | |
3445 | if (r) | |
3446 | return r; | |
c6b4fcba JT |
3447 | } |
3448 | ||
3449 | if (!cache->loaded_mappings) { | |
ea2dd8c1 | 3450 | r = dm_cache_load_mappings(cache->cmd, cache->policy, |
c6b4fcba JT |
3451 | load_mapping, cache); |
3452 | if (r) { | |
b61d9509 | 3453 | DMERR("%s: could not load cache mappings", cache_device_name(cache)); |
028ae9f7 | 3454 | metadata_operation_failed(cache, "dm_cache_load_mappings", r); |
c6b4fcba JT |
3455 | return r; |
3456 | } | |
3457 | ||
3458 | cache->loaded_mappings = true; | |
3459 | } | |
3460 | ||
3461 | if (!cache->loaded_discards) { | |
3e2e1c30 JT |
3462 | struct discard_load_info li; |
3463 | ||
3464 | /* | |
3465 | * The discard bitset could have been resized, or the | |
3466 | * discard block size changed. To be safe we start by | |
3467 | * setting every dblock to not discarded. | |
3468 | */ | |
3469 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); | |
3470 | ||
3471 | discard_load_info_init(cache, &li); | |
3472 | r = dm_cache_load_discards(cache->cmd, load_discard, &li); | |
c6b4fcba | 3473 | if (r) { |
b61d9509 | 3474 | DMERR("%s: could not load origin discards", cache_device_name(cache)); |
028ae9f7 | 3475 | metadata_operation_failed(cache, "dm_cache_load_discards", r); |
c6b4fcba JT |
3476 | return r; |
3477 | } | |
3e2e1c30 | 3478 | set_discard_range(&li); |
c6b4fcba JT |
3479 | |
3480 | cache->loaded_discards = true; | |
3481 | } | |
3482 | ||
3483 | return r; | |
3484 | } | |
3485 | ||
3486 | static void cache_resume(struct dm_target *ti) | |
3487 | { | |
3488 | struct cache *cache = ti->private; | |
3489 | ||
3490 | cache->need_tick_bio = true; | |
3491 | do_waker(&cache->waker.work); | |
3492 | } | |
3493 | ||
3494 | /* | |
3495 | * Status format: | |
3496 | * | |
6a388618 MS |
3497 | * <metadata block size> <#used metadata blocks>/<#total metadata blocks> |
3498 | * <cache block size> <#used cache blocks>/<#total cache blocks> | |
c6b4fcba | 3499 | * <#read hits> <#read misses> <#write hits> <#write misses> |
6a388618 | 3500 | * <#demotions> <#promotions> <#dirty> |
c6b4fcba JT |
3501 | * <#features> <features>* |
3502 | * <#core args> <core args> | |
255eac20 | 3503 | * <policy name> <#policy args> <policy args>* <cache metadata mode> <needs_check> |
c6b4fcba JT |
3504 | */ |
3505 | static void cache_status(struct dm_target *ti, status_type_t type, | |
3506 | unsigned status_flags, char *result, unsigned maxlen) | |
3507 | { | |
3508 | int r = 0; | |
3509 | unsigned i; | |
3510 | ssize_t sz = 0; | |
3511 | dm_block_t nr_free_blocks_metadata = 0; | |
3512 | dm_block_t nr_blocks_metadata = 0; | |
3513 | char buf[BDEVNAME_SIZE]; | |
3514 | struct cache *cache = ti->private; | |
3515 | dm_cblock_t residency; | |
3516 | ||
3517 | switch (type) { | |
3518 | case STATUSTYPE_INFO: | |
028ae9f7 JT |
3519 | if (get_cache_mode(cache) == CM_FAIL) { |
3520 | DMEMIT("Fail"); | |
3521 | break; | |
c6b4fcba JT |
3522 | } |
3523 | ||
028ae9f7 JT |
3524 | /* Commit to ensure statistics aren't out-of-date */ |
3525 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
3526 | (void) commit(cache, false); | |
3527 | ||
b61d9509 | 3528 | r = dm_cache_get_free_metadata_block_count(cache->cmd, &nr_free_blocks_metadata); |
c6b4fcba | 3529 | if (r) { |
b61d9509 MS |
3530 | DMERR("%s: dm_cache_get_free_metadata_block_count returned %d", |
3531 | cache_device_name(cache), r); | |
c6b4fcba JT |
3532 | goto err; |
3533 | } | |
3534 | ||
3535 | r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata); | |
3536 | if (r) { | |
b61d9509 MS |
3537 | DMERR("%s: dm_cache_get_metadata_dev_size returned %d", |
3538 | cache_device_name(cache), r); | |
c6b4fcba JT |
3539 | goto err; |
3540 | } | |
3541 | ||
3542 | residency = policy_residency(cache->policy); | |
3543 | ||
44fa816b | 3544 | DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ", |
895b47d7 | 3545 | (unsigned)DM_CACHE_METADATA_BLOCK_SIZE, |
c6b4fcba JT |
3546 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
3547 | (unsigned long long)nr_blocks_metadata, | |
6a388618 MS |
3548 | cache->sectors_per_block, |
3549 | (unsigned long long) from_cblock(residency), | |
3550 | (unsigned long long) from_cblock(cache->cache_size), | |
c6b4fcba JT |
3551 | (unsigned) atomic_read(&cache->stats.read_hit), |
3552 | (unsigned) atomic_read(&cache->stats.read_miss), | |
3553 | (unsigned) atomic_read(&cache->stats.write_hit), | |
3554 | (unsigned) atomic_read(&cache->stats.write_miss), | |
3555 | (unsigned) atomic_read(&cache->stats.demotion), | |
3556 | (unsigned) atomic_read(&cache->stats.promotion), | |
44fa816b | 3557 | (unsigned long) atomic_read(&cache->nr_dirty)); |
c6b4fcba | 3558 | |
2ee57d58 | 3559 | if (writethrough_mode(&cache->features)) |
c6b4fcba | 3560 | DMEMIT("1 writethrough "); |
2ee57d58 JT |
3561 | |
3562 | else if (passthrough_mode(&cache->features)) | |
3563 | DMEMIT("1 passthrough "); | |
3564 | ||
3565 | else if (writeback_mode(&cache->features)) | |
3566 | DMEMIT("1 writeback "); | |
3567 | ||
3568 | else { | |
b61d9509 MS |
3569 | DMERR("%s: internal error: unknown io mode: %d", |
3570 | cache_device_name(cache), (int) cache->features.io_mode); | |
2ee57d58 JT |
3571 | goto err; |
3572 | } | |
c6b4fcba JT |
3573 | |
3574 | DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold); | |
2e68c4e6 MS |
3575 | |
3576 | DMEMIT("%s ", dm_cache_policy_get_name(cache->policy)); | |
c6b4fcba | 3577 | if (sz < maxlen) { |
028ae9f7 | 3578 | r = policy_emit_config_values(cache->policy, result, maxlen, &sz); |
c6b4fcba | 3579 | if (r) |
b61d9509 MS |
3580 | DMERR("%s: policy_emit_config_values returned %d", |
3581 | cache_device_name(cache), r); | |
c6b4fcba JT |
3582 | } |
3583 | ||
028ae9f7 JT |
3584 | if (get_cache_mode(cache) == CM_READ_ONLY) |
3585 | DMEMIT("ro "); | |
3586 | else | |
3587 | DMEMIT("rw "); | |
3588 | ||
255eac20 MS |
3589 | if (dm_cache_metadata_needs_check(cache->cmd)) |
3590 | DMEMIT("needs_check "); | |
3591 | else | |
3592 | DMEMIT("- "); | |
3593 | ||
c6b4fcba JT |
3594 | break; |
3595 | ||
3596 | case STATUSTYPE_TABLE: | |
3597 | format_dev_t(buf, cache->metadata_dev->bdev->bd_dev); | |
3598 | DMEMIT("%s ", buf); | |
3599 | format_dev_t(buf, cache->cache_dev->bdev->bd_dev); | |
3600 | DMEMIT("%s ", buf); | |
3601 | format_dev_t(buf, cache->origin_dev->bdev->bd_dev); | |
3602 | DMEMIT("%s", buf); | |
3603 | ||
3604 | for (i = 0; i < cache->nr_ctr_args - 1; i++) | |
3605 | DMEMIT(" %s", cache->ctr_args[i]); | |
3606 | if (cache->nr_ctr_args) | |
3607 | DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]); | |
3608 | } | |
3609 | ||
3610 | return; | |
3611 | ||
3612 | err: | |
3613 | DMEMIT("Error"); | |
3614 | } | |
3615 | ||
c6b4fcba | 3616 | /* |
65790ff9 JT |
3617 | * A cache block range can take two forms: |
3618 | * | |
3619 | * i) A single cblock, eg. '3456' | |
3620 | * ii) A begin and end cblock with dots between, eg. 123-234 | |
3621 | */ | |
3622 | static int parse_cblock_range(struct cache *cache, const char *str, | |
3623 | struct cblock_range *result) | |
3624 | { | |
3625 | char dummy; | |
3626 | uint64_t b, e; | |
3627 | int r; | |
3628 | ||
3629 | /* | |
3630 | * Try and parse form (ii) first. | |
3631 | */ | |
3632 | r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy); | |
3633 | if (r < 0) | |
3634 | return r; | |
3635 | ||
3636 | if (r == 2) { | |
3637 | result->begin = to_cblock(b); | |
3638 | result->end = to_cblock(e); | |
3639 | return 0; | |
3640 | } | |
3641 | ||
3642 | /* | |
3643 | * That didn't work, try form (i). | |
3644 | */ | |
3645 | r = sscanf(str, "%llu%c", &b, &dummy); | |
3646 | if (r < 0) | |
3647 | return r; | |
3648 | ||
3649 | if (r == 1) { | |
3650 | result->begin = to_cblock(b); | |
3651 | result->end = to_cblock(from_cblock(result->begin) + 1u); | |
3652 | return 0; | |
3653 | } | |
3654 | ||
b61d9509 | 3655 | DMERR("%s: invalid cblock range '%s'", cache_device_name(cache), str); |
65790ff9 JT |
3656 | return -EINVAL; |
3657 | } | |
3658 | ||
3659 | static int validate_cblock_range(struct cache *cache, struct cblock_range *range) | |
3660 | { | |
3661 | uint64_t b = from_cblock(range->begin); | |
3662 | uint64_t e = from_cblock(range->end); | |
3663 | uint64_t n = from_cblock(cache->cache_size); | |
3664 | ||
3665 | if (b >= n) { | |
b61d9509 MS |
3666 | DMERR("%s: begin cblock out of range: %llu >= %llu", |
3667 | cache_device_name(cache), b, n); | |
65790ff9 JT |
3668 | return -EINVAL; |
3669 | } | |
3670 | ||
3671 | if (e > n) { | |
b61d9509 MS |
3672 | DMERR("%s: end cblock out of range: %llu > %llu", |
3673 | cache_device_name(cache), e, n); | |
65790ff9 JT |
3674 | return -EINVAL; |
3675 | } | |
3676 | ||
3677 | if (b >= e) { | |
b61d9509 MS |
3678 | DMERR("%s: invalid cblock range: %llu >= %llu", |
3679 | cache_device_name(cache), b, e); | |
65790ff9 JT |
3680 | return -EINVAL; |
3681 | } | |
3682 | ||
3683 | return 0; | |
3684 | } | |
3685 | ||
3686 | static int request_invalidation(struct cache *cache, struct cblock_range *range) | |
3687 | { | |
3688 | struct invalidation_request req; | |
3689 | ||
3690 | INIT_LIST_HEAD(&req.list); | |
3691 | req.cblocks = range; | |
3692 | atomic_set(&req.complete, 0); | |
3693 | req.err = 0; | |
3694 | init_waitqueue_head(&req.result_wait); | |
3695 | ||
3696 | spin_lock(&cache->invalidation_lock); | |
3697 | list_add(&req.list, &cache->invalidation_requests); | |
3698 | spin_unlock(&cache->invalidation_lock); | |
3699 | wake_worker(cache); | |
3700 | ||
3701 | wait_event(req.result_wait, atomic_read(&req.complete)); | |
3702 | return req.err; | |
3703 | } | |
3704 | ||
3705 | static int process_invalidate_cblocks_message(struct cache *cache, unsigned count, | |
3706 | const char **cblock_ranges) | |
3707 | { | |
3708 | int r = 0; | |
3709 | unsigned i; | |
3710 | struct cblock_range range; | |
3711 | ||
3712 | if (!passthrough_mode(&cache->features)) { | |
b61d9509 MS |
3713 | DMERR("%s: cache has to be in passthrough mode for invalidation", |
3714 | cache_device_name(cache)); | |
65790ff9 JT |
3715 | return -EPERM; |
3716 | } | |
3717 | ||
3718 | for (i = 0; i < count; i++) { | |
3719 | r = parse_cblock_range(cache, cblock_ranges[i], &range); | |
3720 | if (r) | |
3721 | break; | |
3722 | ||
3723 | r = validate_cblock_range(cache, &range); | |
3724 | if (r) | |
3725 | break; | |
3726 | ||
3727 | /* | |
3728 | * Pass begin and end origin blocks to the worker and wake it. | |
3729 | */ | |
3730 | r = request_invalidation(cache, &range); | |
3731 | if (r) | |
3732 | break; | |
3733 | } | |
3734 | ||
3735 | return r; | |
3736 | } | |
3737 | ||
3738 | /* | |
3739 | * Supports | |
3740 | * "<key> <value>" | |
3741 | * and | |
3742 | * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]* | |
c6b4fcba JT |
3743 | * |
3744 | * The key migration_threshold is supported by the cache target core. | |
3745 | */ | |
3746 | static int cache_message(struct dm_target *ti, unsigned argc, char **argv) | |
3747 | { | |
c6b4fcba JT |
3748 | struct cache *cache = ti->private; |
3749 | ||
65790ff9 JT |
3750 | if (!argc) |
3751 | return -EINVAL; | |
3752 | ||
028ae9f7 | 3753 | if (get_cache_mode(cache) >= CM_READ_ONLY) { |
b61d9509 MS |
3754 | DMERR("%s: unable to service cache target messages in READ_ONLY or FAIL mode", |
3755 | cache_device_name(cache)); | |
028ae9f7 JT |
3756 | return -EOPNOTSUPP; |
3757 | } | |
3758 | ||
7b6b2bc9 | 3759 | if (!strcasecmp(argv[0], "invalidate_cblocks")) |
65790ff9 JT |
3760 | return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1); |
3761 | ||
c6b4fcba JT |
3762 | if (argc != 2) |
3763 | return -EINVAL; | |
3764 | ||
2f14f4b5 | 3765 | return set_config_value(cache, argv[0], argv[1]); |
c6b4fcba JT |
3766 | } |
3767 | ||
3768 | static int cache_iterate_devices(struct dm_target *ti, | |
3769 | iterate_devices_callout_fn fn, void *data) | |
3770 | { | |
3771 | int r = 0; | |
3772 | struct cache *cache = ti->private; | |
3773 | ||
3774 | r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data); | |
3775 | if (!r) | |
3776 | r = fn(ti, cache->origin_dev, 0, ti->len, data); | |
3777 | ||
3778 | return r; | |
3779 | } | |
3780 | ||
3781 | /* | |
3782 | * We assume I/O is going to the origin (which is the volume | |
3783 | * more likely to have restrictions e.g. by being striped). | |
3784 | * (Looking up the exact location of the data would be expensive | |
3785 | * and could always be out of date by the time the bio is submitted.) | |
3786 | */ | |
3787 | static int cache_bvec_merge(struct dm_target *ti, | |
3788 | struct bvec_merge_data *bvm, | |
3789 | struct bio_vec *biovec, int max_size) | |
3790 | { | |
3791 | struct cache *cache = ti->private; | |
3792 | struct request_queue *q = bdev_get_queue(cache->origin_dev->bdev); | |
3793 | ||
3794 | if (!q->merge_bvec_fn) | |
3795 | return max_size; | |
3796 | ||
3797 | bvm->bi_bdev = cache->origin_dev->bdev; | |
3798 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3799 | } | |
3800 | ||
3801 | static void set_discard_limits(struct cache *cache, struct queue_limits *limits) | |
3802 | { | |
3803 | /* | |
3804 | * FIXME: these limits may be incompatible with the cache device | |
3805 | */ | |
7ae34e77 JT |
3806 | limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024, |
3807 | cache->origin_sectors); | |
1bad9bc4 | 3808 | limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT; |
c6b4fcba JT |
3809 | } |
3810 | ||
3811 | static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3812 | { | |
3813 | struct cache *cache = ti->private; | |
f6109372 | 3814 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
c6b4fcba | 3815 | |
f6109372 MS |
3816 | /* |
3817 | * If the system-determined stacked limits are compatible with the | |
3818 | * cache's blocksize (io_opt is a factor) do not override them. | |
3819 | */ | |
3820 | if (io_opt_sectors < cache->sectors_per_block || | |
3821 | do_div(io_opt_sectors, cache->sectors_per_block)) { | |
b0246530 | 3822 | blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT); |
f6109372 MS |
3823 | blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT); |
3824 | } | |
c6b4fcba JT |
3825 | set_discard_limits(cache, limits); |
3826 | } | |
3827 | ||
3828 | /*----------------------------------------------------------------*/ | |
3829 | ||
3830 | static struct target_type cache_target = { | |
3831 | .name = "cache", | |
255eac20 | 3832 | .version = {1, 8, 0}, |
c6b4fcba JT |
3833 | .module = THIS_MODULE, |
3834 | .ctr = cache_ctr, | |
3835 | .dtr = cache_dtr, | |
3836 | .map = cache_map, | |
3837 | .end_io = cache_end_io, | |
3838 | .postsuspend = cache_postsuspend, | |
3839 | .preresume = cache_preresume, | |
3840 | .resume = cache_resume, | |
3841 | .status = cache_status, | |
3842 | .message = cache_message, | |
3843 | .iterate_devices = cache_iterate_devices, | |
3844 | .merge = cache_bvec_merge, | |
3845 | .io_hints = cache_io_hints, | |
3846 | }; | |
3847 | ||
3848 | static int __init dm_cache_init(void) | |
3849 | { | |
3850 | int r; | |
3851 | ||
3852 | r = dm_register_target(&cache_target); | |
3853 | if (r) { | |
3854 | DMERR("cache target registration failed: %d", r); | |
3855 | return r; | |
3856 | } | |
3857 | ||
3858 | migration_cache = KMEM_CACHE(dm_cache_migration, 0); | |
3859 | if (!migration_cache) { | |
3860 | dm_unregister_target(&cache_target); | |
3861 | return -ENOMEM; | |
3862 | } | |
3863 | ||
3864 | return 0; | |
3865 | } | |
3866 | ||
3867 | static void __exit dm_cache_exit(void) | |
3868 | { | |
3869 | dm_unregister_target(&cache_target); | |
3870 | kmem_cache_destroy(migration_cache); | |
3871 | } | |
3872 | ||
3873 | module_init(dm_cache_init); | |
3874 | module_exit(dm_cache_exit); | |
3875 | ||
3876 | MODULE_DESCRIPTION(DM_NAME " cache target"); | |
3877 | MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>"); | |
3878 | MODULE_LICENSE("GPL"); |