]>
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 JT |
152 | |
153 | /* | |
154 | * FIXME: the cache is read/write for the time being. | |
155 | */ | |
2ee57d58 | 156 | enum cache_metadata_mode { |
c6b4fcba JT |
157 | CM_WRITE, /* metadata may be changed */ |
158 | CM_READ_ONLY, /* metadata may not be changed */ | |
159 | }; | |
160 | ||
2ee57d58 JT |
161 | enum cache_io_mode { |
162 | /* | |
163 | * Data is written to cached blocks only. These blocks are marked | |
164 | * dirty. If you lose the cache device you will lose data. | |
165 | * Potential performance increase for both reads and writes. | |
166 | */ | |
167 | CM_IO_WRITEBACK, | |
168 | ||
169 | /* | |
170 | * Data is written to both cache and origin. Blocks are never | |
171 | * dirty. Potential performance benfit for reads only. | |
172 | */ | |
173 | CM_IO_WRITETHROUGH, | |
174 | ||
175 | /* | |
176 | * A degraded mode useful for various cache coherency situations | |
177 | * (eg, rolling back snapshots). Reads and writes always go to the | |
178 | * origin. If a write goes to a cached oblock, then the cache | |
179 | * block is invalidated. | |
180 | */ | |
181 | CM_IO_PASSTHROUGH | |
182 | }; | |
183 | ||
c6b4fcba | 184 | struct cache_features { |
2ee57d58 JT |
185 | enum cache_metadata_mode mode; |
186 | enum cache_io_mode io_mode; | |
c6b4fcba JT |
187 | }; |
188 | ||
189 | struct cache_stats { | |
190 | atomic_t read_hit; | |
191 | atomic_t read_miss; | |
192 | atomic_t write_hit; | |
193 | atomic_t write_miss; | |
194 | atomic_t demotion; | |
195 | atomic_t promotion; | |
196 | atomic_t copies_avoided; | |
197 | atomic_t cache_cell_clash; | |
198 | atomic_t commit_count; | |
199 | atomic_t discard_count; | |
200 | }; | |
201 | ||
65790ff9 JT |
202 | /* |
203 | * Defines a range of cblocks, begin to (end - 1) are in the range. end is | |
204 | * the one-past-the-end value. | |
205 | */ | |
206 | struct cblock_range { | |
207 | dm_cblock_t begin; | |
208 | dm_cblock_t end; | |
209 | }; | |
210 | ||
211 | struct invalidation_request { | |
212 | struct list_head list; | |
213 | struct cblock_range *cblocks; | |
214 | ||
215 | atomic_t complete; | |
216 | int err; | |
217 | ||
218 | wait_queue_head_t result_wait; | |
219 | }; | |
220 | ||
c6b4fcba JT |
221 | struct cache { |
222 | struct dm_target *ti; | |
223 | struct dm_target_callbacks callbacks; | |
224 | ||
c9ec5d7c MS |
225 | struct dm_cache_metadata *cmd; |
226 | ||
c6b4fcba JT |
227 | /* |
228 | * Metadata is written to this device. | |
229 | */ | |
230 | struct dm_dev *metadata_dev; | |
231 | ||
232 | /* | |
233 | * The slower of the two data devices. Typically a spindle. | |
234 | */ | |
235 | struct dm_dev *origin_dev; | |
236 | ||
237 | /* | |
238 | * The faster of the two data devices. Typically an SSD. | |
239 | */ | |
240 | struct dm_dev *cache_dev; | |
241 | ||
c6b4fcba JT |
242 | /* |
243 | * Size of the origin device in _complete_ blocks and native sectors. | |
244 | */ | |
245 | dm_oblock_t origin_blocks; | |
246 | sector_t origin_sectors; | |
247 | ||
248 | /* | |
249 | * Size of the cache device in blocks. | |
250 | */ | |
251 | dm_cblock_t cache_size; | |
252 | ||
253 | /* | |
254 | * Fields for converting from sectors to blocks. | |
255 | */ | |
256 | uint32_t sectors_per_block; | |
257 | int sectors_per_block_shift; | |
258 | ||
c6b4fcba | 259 | spinlock_t lock; |
651f5fa2 | 260 | struct list_head deferred_cells; |
c6b4fcba JT |
261 | struct bio_list deferred_bios; |
262 | struct bio_list deferred_flush_bios; | |
e2e74d61 | 263 | struct bio_list deferred_writethrough_bios; |
c6b4fcba JT |
264 | struct list_head quiesced_migrations; |
265 | struct list_head completed_migrations; | |
266 | struct list_head need_commit_migrations; | |
267 | sector_t migration_threshold; | |
c6b4fcba | 268 | wait_queue_head_t migration_wait; |
a59db676 JT |
269 | atomic_t nr_allocated_migrations; |
270 | ||
271 | /* | |
272 | * The number of in flight migrations that are performing | |
273 | * background io. eg, promotion, writeback. | |
274 | */ | |
275 | atomic_t nr_io_migrations; | |
c6b4fcba | 276 | |
66cb1910 | 277 | wait_queue_head_t quiescing_wait; |
238f8363 | 278 | atomic_t quiescing; |
66cb1910 JT |
279 | atomic_t quiescing_ack; |
280 | ||
c6b4fcba JT |
281 | /* |
282 | * cache_size entries, dirty if set | |
283 | */ | |
44fa816b | 284 | atomic_t nr_dirty; |
c6b4fcba JT |
285 | unsigned long *dirty_bitset; |
286 | ||
287 | /* | |
288 | * origin_blocks entries, discarded if set. | |
289 | */ | |
1bad9bc4 | 290 | dm_dblock_t discard_nr_blocks; |
c6b4fcba | 291 | unsigned long *discard_bitset; |
08b18451 | 292 | uint32_t discard_block_size; /* a power of 2 times sectors per block */ |
c9ec5d7c MS |
293 | |
294 | /* | |
295 | * Rather than reconstructing the table line for the status we just | |
296 | * save it and regurgitate. | |
297 | */ | |
298 | unsigned nr_ctr_args; | |
299 | const char **ctr_args; | |
c6b4fcba JT |
300 | |
301 | struct dm_kcopyd_client *copier; | |
302 | struct workqueue_struct *wq; | |
303 | struct work_struct worker; | |
304 | ||
305 | struct delayed_work waker; | |
306 | unsigned long last_commit_jiffies; | |
307 | ||
308 | struct dm_bio_prison *prison; | |
309 | struct dm_deferred_set *all_io_ds; | |
310 | ||
311 | mempool_t *migration_pool; | |
c6b4fcba JT |
312 | |
313 | struct dm_cache_policy *policy; | |
314 | unsigned policy_nr_args; | |
315 | ||
316 | bool need_tick_bio:1; | |
317 | bool sized:1; | |
65790ff9 | 318 | bool invalidate:1; |
c6b4fcba JT |
319 | bool commit_requested:1; |
320 | bool loaded_mappings:1; | |
321 | bool loaded_discards:1; | |
322 | ||
c6b4fcba | 323 | /* |
c9ec5d7c | 324 | * Cache features such as write-through. |
c6b4fcba | 325 | */ |
c9ec5d7c MS |
326 | struct cache_features features; |
327 | ||
328 | struct cache_stats stats; | |
65790ff9 JT |
329 | |
330 | /* | |
331 | * Invalidation fields. | |
332 | */ | |
333 | spinlock_t invalidation_lock; | |
334 | struct list_head invalidation_requests; | |
066dbaa3 JT |
335 | |
336 | struct io_tracker origin_tracker; | |
c6b4fcba JT |
337 | }; |
338 | ||
339 | struct per_bio_data { | |
340 | bool tick:1; | |
341 | unsigned req_nr:2; | |
342 | struct dm_deferred_entry *all_io_entry; | |
c6eda5e8 | 343 | struct dm_hook_info hook_info; |
066dbaa3 | 344 | sector_t len; |
e2e74d61 | 345 | |
19b0092e MS |
346 | /* |
347 | * writethrough fields. These MUST remain at the end of this | |
348 | * structure and the 'cache' member must be the first as it | |
aeed1420 | 349 | * is used to determine the offset of the writethrough fields. |
19b0092e | 350 | */ |
e2e74d61 JT |
351 | struct cache *cache; |
352 | dm_cblock_t cblock; | |
b844fe69 | 353 | struct dm_bio_details bio_details; |
c6b4fcba JT |
354 | }; |
355 | ||
356 | struct dm_cache_migration { | |
357 | struct list_head list; | |
358 | struct cache *cache; | |
359 | ||
360 | unsigned long start_jiffies; | |
361 | dm_oblock_t old_oblock; | |
362 | dm_oblock_t new_oblock; | |
363 | dm_cblock_t cblock; | |
364 | ||
365 | bool err:1; | |
7ae34e77 | 366 | bool discard:1; |
c6b4fcba JT |
367 | bool writeback:1; |
368 | bool demote:1; | |
369 | bool promote:1; | |
c9d28d5d | 370 | bool requeue_holder:1; |
65790ff9 | 371 | bool invalidate:1; |
c6b4fcba JT |
372 | |
373 | struct dm_bio_prison_cell *old_ocell; | |
374 | struct dm_bio_prison_cell *new_ocell; | |
375 | }; | |
376 | ||
377 | /* | |
378 | * Processing a bio in the worker thread may require these memory | |
379 | * allocations. We prealloc to avoid deadlocks (the same worker thread | |
380 | * frees them back to the mempool). | |
381 | */ | |
382 | struct prealloc { | |
383 | struct dm_cache_migration *mg; | |
384 | struct dm_bio_prison_cell *cell1; | |
385 | struct dm_bio_prison_cell *cell2; | |
386 | }; | |
387 | ||
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 | { | |
421 | if (atomic_dec_and_test(&mg->cache->nr_allocated_migrations)) | |
422 | wake_up(&mg->cache->migration_wait); | |
423 | ||
424 | mempool_free(mg, mg->cache->migration_pool); | |
425 | } | |
426 | ||
c6b4fcba JT |
427 | static int prealloc_data_structs(struct cache *cache, struct prealloc *p) |
428 | { | |
429 | if (!p->mg) { | |
a59db676 | 430 | p->mg = alloc_migration(cache); |
c6b4fcba JT |
431 | if (!p->mg) |
432 | return -ENOMEM; | |
433 | } | |
434 | ||
435 | if (!p->cell1) { | |
436 | p->cell1 = alloc_prison_cell(cache); | |
437 | if (!p->cell1) | |
438 | return -ENOMEM; | |
439 | } | |
440 | ||
441 | if (!p->cell2) { | |
442 | p->cell2 = alloc_prison_cell(cache); | |
443 | if (!p->cell2) | |
444 | return -ENOMEM; | |
445 | } | |
446 | ||
447 | return 0; | |
448 | } | |
449 | ||
450 | static void prealloc_free_structs(struct cache *cache, struct prealloc *p) | |
451 | { | |
452 | if (p->cell2) | |
453 | free_prison_cell(cache, p->cell2); | |
454 | ||
455 | if (p->cell1) | |
456 | free_prison_cell(cache, p->cell1); | |
457 | ||
458 | if (p->mg) | |
a59db676 | 459 | free_migration(p->mg); |
c6b4fcba JT |
460 | } |
461 | ||
462 | static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p) | |
463 | { | |
464 | struct dm_cache_migration *mg = p->mg; | |
465 | ||
466 | BUG_ON(!mg); | |
467 | p->mg = NULL; | |
468 | ||
469 | return mg; | |
470 | } | |
471 | ||
472 | /* | |
473 | * You must have a cell within the prealloc struct to return. If not this | |
474 | * function will BUG() rather than returning NULL. | |
475 | */ | |
476 | static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p) | |
477 | { | |
478 | struct dm_bio_prison_cell *r = NULL; | |
479 | ||
480 | if (p->cell1) { | |
481 | r = p->cell1; | |
482 | p->cell1 = NULL; | |
483 | ||
484 | } else if (p->cell2) { | |
485 | r = p->cell2; | |
486 | p->cell2 = NULL; | |
487 | } else | |
488 | BUG(); | |
489 | ||
490 | return r; | |
491 | } | |
492 | ||
493 | /* | |
494 | * You can't have more than two cells in a prealloc struct. BUG() will be | |
495 | * called if you try and overfill. | |
496 | */ | |
497 | static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell) | |
498 | { | |
499 | if (!p->cell2) | |
500 | p->cell2 = cell; | |
501 | ||
502 | else if (!p->cell1) | |
503 | p->cell1 = cell; | |
504 | ||
505 | else | |
506 | BUG(); | |
507 | } | |
508 | ||
509 | /*----------------------------------------------------------------*/ | |
510 | ||
7ae34e77 | 511 | static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key *key) |
c6b4fcba JT |
512 | { |
513 | key->virtual = 0; | |
514 | key->dev = 0; | |
7ae34e77 JT |
515 | key->block_begin = from_oblock(begin); |
516 | key->block_end = from_oblock(end); | |
c6b4fcba JT |
517 | } |
518 | ||
519 | /* | |
520 | * The caller hands in a preallocated cell, and a free function for it. | |
521 | * The cell will be freed if there's an error, or if it wasn't used because | |
522 | * a cell with that key already exists. | |
523 | */ | |
524 | typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell); | |
525 | ||
7ae34e77 JT |
526 | static int bio_detain_range(struct cache *cache, dm_oblock_t oblock_begin, dm_oblock_t oblock_end, |
527 | struct bio *bio, struct dm_bio_prison_cell *cell_prealloc, | |
528 | cell_free_fn free_fn, void *free_context, | |
529 | struct dm_bio_prison_cell **cell_result) | |
c6b4fcba JT |
530 | { |
531 | int r; | |
532 | struct dm_cell_key key; | |
533 | ||
7ae34e77 | 534 | build_key(oblock_begin, oblock_end, &key); |
c6b4fcba JT |
535 | r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result); |
536 | if (r) | |
537 | free_fn(free_context, cell_prealloc); | |
538 | ||
539 | return r; | |
540 | } | |
541 | ||
7ae34e77 JT |
542 | static int bio_detain(struct cache *cache, dm_oblock_t oblock, |
543 | struct bio *bio, struct dm_bio_prison_cell *cell_prealloc, | |
544 | cell_free_fn free_fn, void *free_context, | |
545 | struct dm_bio_prison_cell **cell_result) | |
546 | { | |
547 | dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL); | |
548 | return bio_detain_range(cache, oblock, end, bio, | |
549 | cell_prealloc, free_fn, free_context, cell_result); | |
550 | } | |
551 | ||
c6b4fcba JT |
552 | static int get_cell(struct cache *cache, |
553 | dm_oblock_t oblock, | |
554 | struct prealloc *structs, | |
555 | struct dm_bio_prison_cell **cell_result) | |
556 | { | |
557 | int r; | |
558 | struct dm_cell_key key; | |
559 | struct dm_bio_prison_cell *cell_prealloc; | |
560 | ||
561 | cell_prealloc = prealloc_get_cell(structs); | |
562 | ||
7ae34e77 | 563 | build_key(oblock, to_oblock(from_oblock(oblock) + 1ULL), &key); |
c6b4fcba JT |
564 | r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result); |
565 | if (r) | |
566 | prealloc_put_cell(structs, cell_prealloc); | |
567 | ||
568 | return r; | |
569 | } | |
570 | ||
aeed1420 | 571 | /*----------------------------------------------------------------*/ |
c6b4fcba JT |
572 | |
573 | static bool is_dirty(struct cache *cache, dm_cblock_t b) | |
574 | { | |
575 | return test_bit(from_cblock(b), cache->dirty_bitset); | |
576 | } | |
577 | ||
578 | static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock) | |
579 | { | |
580 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 581 | atomic_inc(&cache->nr_dirty); |
c6b4fcba JT |
582 | policy_set_dirty(cache->policy, oblock); |
583 | } | |
584 | } | |
585 | ||
586 | static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock) | |
587 | { | |
588 | if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
589 | policy_clear_dirty(cache->policy, oblock); | |
44fa816b | 590 | if (atomic_dec_return(&cache->nr_dirty) == 0) |
c6b4fcba JT |
591 | dm_table_event(cache->ti->table); |
592 | } | |
593 | } | |
594 | ||
595 | /*----------------------------------------------------------------*/ | |
aeed1420 | 596 | |
c6b4fcba JT |
597 | static bool block_size_is_power_of_two(struct cache *cache) |
598 | { | |
599 | return cache->sectors_per_block_shift >= 0; | |
600 | } | |
601 | ||
43aeaa29 MP |
602 | /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */ |
603 | #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6 | |
604 | __always_inline | |
605 | #endif | |
414dd67d JT |
606 | static dm_block_t block_div(dm_block_t b, uint32_t n) |
607 | { | |
608 | do_div(b, n); | |
609 | ||
610 | return b; | |
611 | } | |
612 | ||
7ae34e77 | 613 | static dm_block_t oblocks_per_dblock(struct cache *cache) |
1bad9bc4 | 614 | { |
7ae34e77 | 615 | dm_block_t oblocks = cache->discard_block_size; |
1bad9bc4 | 616 | |
7ae34e77 JT |
617 | if (block_size_is_power_of_two(cache)) |
618 | oblocks >>= cache->sectors_per_block_shift; | |
1bad9bc4 | 619 | else |
7ae34e77 | 620 | oblocks = block_div(oblocks, cache->sectors_per_block); |
1bad9bc4 | 621 | |
7ae34e77 JT |
622 | return oblocks; |
623 | } | |
624 | ||
625 | static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock) | |
626 | { | |
627 | return to_dblock(block_div(from_oblock(oblock), | |
628 | oblocks_per_dblock(cache))); | |
629 | } | |
1bad9bc4 | 630 | |
7ae34e77 JT |
631 | static dm_oblock_t dblock_to_oblock(struct cache *cache, dm_dblock_t dblock) |
632 | { | |
633 | return to_oblock(from_dblock(dblock) * oblocks_per_dblock(cache)); | |
1bad9bc4 JT |
634 | } |
635 | ||
636 | static void set_discard(struct cache *cache, dm_dblock_t b) | |
c6b4fcba JT |
637 | { |
638 | unsigned long flags; | |
639 | ||
7ae34e77 | 640 | BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
641 | atomic_inc(&cache->stats.discard_count); |
642 | ||
643 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 644 | set_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
645 | spin_unlock_irqrestore(&cache->lock, flags); |
646 | } | |
647 | ||
1bad9bc4 | 648 | static void clear_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
649 | { |
650 | unsigned long flags; | |
651 | ||
652 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 653 | clear_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
654 | spin_unlock_irqrestore(&cache->lock, flags); |
655 | } | |
656 | ||
1bad9bc4 | 657 | static bool is_discarded(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
658 | { |
659 | int r; | |
660 | unsigned long flags; | |
661 | ||
662 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 663 | r = test_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
664 | spin_unlock_irqrestore(&cache->lock, flags); |
665 | ||
666 | return r; | |
667 | } | |
668 | ||
669 | static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b) | |
670 | { | |
671 | int r; | |
672 | unsigned long flags; | |
673 | ||
674 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 JT |
675 | r = test_bit(from_dblock(oblock_to_dblock(cache, b)), |
676 | cache->discard_bitset); | |
c6b4fcba JT |
677 | spin_unlock_irqrestore(&cache->lock, flags); |
678 | ||
679 | return r; | |
680 | } | |
681 | ||
682 | /*----------------------------------------------------------------*/ | |
683 | ||
684 | static void load_stats(struct cache *cache) | |
685 | { | |
686 | struct dm_cache_statistics stats; | |
687 | ||
688 | dm_cache_metadata_get_stats(cache->cmd, &stats); | |
689 | atomic_set(&cache->stats.read_hit, stats.read_hits); | |
690 | atomic_set(&cache->stats.read_miss, stats.read_misses); | |
691 | atomic_set(&cache->stats.write_hit, stats.write_hits); | |
692 | atomic_set(&cache->stats.write_miss, stats.write_misses); | |
693 | } | |
694 | ||
695 | static void save_stats(struct cache *cache) | |
696 | { | |
697 | struct dm_cache_statistics stats; | |
698 | ||
699 | stats.read_hits = atomic_read(&cache->stats.read_hit); | |
700 | stats.read_misses = atomic_read(&cache->stats.read_miss); | |
701 | stats.write_hits = atomic_read(&cache->stats.write_hit); | |
702 | stats.write_misses = atomic_read(&cache->stats.write_miss); | |
703 | ||
704 | dm_cache_metadata_set_stats(cache->cmd, &stats); | |
705 | } | |
706 | ||
707 | /*---------------------------------------------------------------- | |
708 | * Per bio data | |
709 | *--------------------------------------------------------------*/ | |
19b0092e MS |
710 | |
711 | /* | |
712 | * If using writeback, leave out struct per_bio_data's writethrough fields. | |
713 | */ | |
714 | #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache)) | |
715 | #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data)) | |
716 | ||
2ee57d58 JT |
717 | static bool writethrough_mode(struct cache_features *f) |
718 | { | |
719 | return f->io_mode == CM_IO_WRITETHROUGH; | |
720 | } | |
721 | ||
722 | static bool writeback_mode(struct cache_features *f) | |
723 | { | |
724 | return f->io_mode == CM_IO_WRITEBACK; | |
725 | } | |
726 | ||
727 | static bool passthrough_mode(struct cache_features *f) | |
728 | { | |
729 | return f->io_mode == CM_IO_PASSTHROUGH; | |
730 | } | |
731 | ||
19b0092e MS |
732 | static size_t get_per_bio_data_size(struct cache *cache) |
733 | { | |
2ee57d58 | 734 | return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB; |
19b0092e MS |
735 | } |
736 | ||
737 | static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size) | |
c6b4fcba | 738 | { |
19b0092e | 739 | struct per_bio_data *pb = dm_per_bio_data(bio, data_size); |
c6b4fcba JT |
740 | BUG_ON(!pb); |
741 | return pb; | |
742 | } | |
743 | ||
19b0092e | 744 | static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size) |
c6b4fcba | 745 | { |
19b0092e | 746 | struct per_bio_data *pb = get_per_bio_data(bio, data_size); |
c6b4fcba JT |
747 | |
748 | pb->tick = false; | |
749 | pb->req_nr = dm_bio_get_target_bio_nr(bio); | |
750 | pb->all_io_entry = NULL; | |
066dbaa3 | 751 | pb->len = 0; |
c6b4fcba JT |
752 | |
753 | return pb; | |
754 | } | |
755 | ||
756 | /*---------------------------------------------------------------- | |
757 | * Remapping | |
758 | *--------------------------------------------------------------*/ | |
759 | static void remap_to_origin(struct cache *cache, struct bio *bio) | |
760 | { | |
761 | bio->bi_bdev = cache->origin_dev->bdev; | |
762 | } | |
763 | ||
764 | static void remap_to_cache(struct cache *cache, struct bio *bio, | |
765 | dm_cblock_t cblock) | |
766 | { | |
4f024f37 | 767 | sector_t bi_sector = bio->bi_iter.bi_sector; |
e0d849fa | 768 | sector_t block = from_cblock(cblock); |
c6b4fcba JT |
769 | |
770 | bio->bi_bdev = cache->cache_dev->bdev; | |
771 | if (!block_size_is_power_of_two(cache)) | |
4f024f37 | 772 | bio->bi_iter.bi_sector = |
e0d849fa | 773 | (block * cache->sectors_per_block) + |
4f024f37 | 774 | sector_div(bi_sector, cache->sectors_per_block); |
c6b4fcba | 775 | else |
4f024f37 | 776 | bio->bi_iter.bi_sector = |
e0d849fa | 777 | (block << cache->sectors_per_block_shift) | |
4f024f37 | 778 | (bi_sector & (cache->sectors_per_block - 1)); |
c6b4fcba JT |
779 | } |
780 | ||
781 | static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio) | |
782 | { | |
783 | unsigned long flags; | |
19b0092e MS |
784 | size_t pb_data_size = get_per_bio_data_size(cache); |
785 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
786 | |
787 | spin_lock_irqsave(&cache->lock, flags); | |
788 | if (cache->need_tick_bio && | |
789 | !(bio->bi_rw & (REQ_FUA | REQ_FLUSH | REQ_DISCARD))) { | |
790 | pb->tick = true; | |
791 | cache->need_tick_bio = false; | |
792 | } | |
793 | spin_unlock_irqrestore(&cache->lock, flags); | |
794 | } | |
795 | ||
796 | static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, | |
797 | dm_oblock_t oblock) | |
798 | { | |
799 | check_if_tick_bio_needed(cache, bio); | |
800 | remap_to_origin(cache, bio); | |
801 | if (bio_data_dir(bio) == WRITE) | |
1bad9bc4 | 802 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
803 | } |
804 | ||
805 | static void remap_to_cache_dirty(struct cache *cache, struct bio *bio, | |
806 | dm_oblock_t oblock, dm_cblock_t cblock) | |
807 | { | |
f8e5f01a | 808 | check_if_tick_bio_needed(cache, bio); |
c6b4fcba JT |
809 | remap_to_cache(cache, bio, cblock); |
810 | if (bio_data_dir(bio) == WRITE) { | |
811 | set_dirty(cache, oblock, cblock); | |
1bad9bc4 | 812 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
813 | } |
814 | } | |
815 | ||
816 | static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio) | |
817 | { | |
4f024f37 | 818 | sector_t block_nr = bio->bi_iter.bi_sector; |
c6b4fcba JT |
819 | |
820 | if (!block_size_is_power_of_two(cache)) | |
821 | (void) sector_div(block_nr, cache->sectors_per_block); | |
822 | else | |
823 | block_nr >>= cache->sectors_per_block_shift; | |
824 | ||
825 | return to_oblock(block_nr); | |
826 | } | |
827 | ||
828 | static int bio_triggers_commit(struct cache *cache, struct bio *bio) | |
829 | { | |
830 | return bio->bi_rw & (REQ_FLUSH | REQ_FUA); | |
831 | } | |
832 | ||
8c081b52 JT |
833 | /* |
834 | * You must increment the deferred set whilst the prison cell is held. To | |
835 | * encourage this, we ask for 'cell' to be passed in. | |
836 | */ | |
837 | static void inc_ds(struct cache *cache, struct bio *bio, | |
838 | struct dm_bio_prison_cell *cell) | |
839 | { | |
840 | size_t pb_data_size = get_per_bio_data_size(cache); | |
841 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
842 | ||
843 | BUG_ON(!cell); | |
844 | BUG_ON(pb->all_io_entry); | |
845 | ||
846 | pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds); | |
847 | } | |
848 | ||
066dbaa3 JT |
849 | static bool accountable_bio(struct cache *cache, struct bio *bio) |
850 | { | |
851 | return ((bio->bi_bdev == cache->origin_dev->bdev) && | |
852 | !(bio->bi_rw & REQ_DISCARD)); | |
853 | } | |
854 | ||
855 | static void accounted_begin(struct cache *cache, struct bio *bio) | |
856 | { | |
857 | size_t pb_data_size = get_per_bio_data_size(cache); | |
858 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
859 | ||
860 | if (accountable_bio(cache, bio)) { | |
861 | pb->len = bio_sectors(bio); | |
862 | iot_io_begin(&cache->origin_tracker, pb->len); | |
863 | } | |
864 | } | |
865 | ||
866 | static void accounted_complete(struct cache *cache, struct bio *bio) | |
867 | { | |
868 | size_t pb_data_size = get_per_bio_data_size(cache); | |
869 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
870 | ||
871 | iot_io_end(&cache->origin_tracker, pb->len); | |
872 | } | |
873 | ||
874 | static void accounted_request(struct cache *cache, struct bio *bio) | |
875 | { | |
876 | accounted_begin(cache, bio); | |
877 | generic_make_request(bio); | |
878 | } | |
879 | ||
c6b4fcba JT |
880 | static void issue(struct cache *cache, struct bio *bio) |
881 | { | |
882 | unsigned long flags; | |
883 | ||
884 | if (!bio_triggers_commit(cache, bio)) { | |
066dbaa3 | 885 | accounted_request(cache, bio); |
c6b4fcba JT |
886 | return; |
887 | } | |
888 | ||
889 | /* | |
890 | * Batch together any bios that trigger commits and then issue a | |
891 | * single commit for them in do_worker(). | |
892 | */ | |
893 | spin_lock_irqsave(&cache->lock, flags); | |
894 | cache->commit_requested = true; | |
895 | bio_list_add(&cache->deferred_flush_bios, bio); | |
896 | spin_unlock_irqrestore(&cache->lock, flags); | |
897 | } | |
898 | ||
8c081b52 JT |
899 | static void inc_and_issue(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell *cell) |
900 | { | |
901 | inc_ds(cache, bio, cell); | |
902 | issue(cache, bio); | |
903 | } | |
904 | ||
e2e74d61 JT |
905 | static void defer_writethrough_bio(struct cache *cache, struct bio *bio) |
906 | { | |
907 | unsigned long flags; | |
908 | ||
909 | spin_lock_irqsave(&cache->lock, flags); | |
910 | bio_list_add(&cache->deferred_writethrough_bios, bio); | |
911 | spin_unlock_irqrestore(&cache->lock, flags); | |
912 | ||
913 | wake_worker(cache); | |
914 | } | |
915 | ||
916 | static void writethrough_endio(struct bio *bio, int err) | |
917 | { | |
19b0092e | 918 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
c9d28d5d JT |
919 | |
920 | dm_unhook_bio(&pb->hook_info, bio); | |
e2e74d61 JT |
921 | |
922 | if (err) { | |
923 | bio_endio(bio, err); | |
924 | return; | |
925 | } | |
926 | ||
b844fe69 | 927 | dm_bio_restore(&pb->bio_details, bio); |
e2e74d61 JT |
928 | remap_to_cache(pb->cache, bio, pb->cblock); |
929 | ||
930 | /* | |
931 | * We can't issue this bio directly, since we're in interrupt | |
aeed1420 | 932 | * context. So it gets put on a bio list for processing by the |
e2e74d61 JT |
933 | * worker thread. |
934 | */ | |
935 | defer_writethrough_bio(pb->cache, bio); | |
936 | } | |
937 | ||
938 | /* | |
939 | * When running in writethrough mode we need to send writes to clean blocks | |
940 | * to both the cache and origin devices. In future we'd like to clone the | |
941 | * bio and send them in parallel, but for now we're doing them in | |
942 | * series as this is easier. | |
943 | */ | |
944 | static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio, | |
945 | dm_oblock_t oblock, dm_cblock_t cblock) | |
946 | { | |
19b0092e | 947 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
e2e74d61 JT |
948 | |
949 | pb->cache = cache; | |
950 | pb->cblock = cblock; | |
c9d28d5d | 951 | dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL); |
b844fe69 | 952 | dm_bio_record(&pb->bio_details, bio); |
e2e74d61 JT |
953 | |
954 | remap_to_origin_clear_discard(pb->cache, bio, oblock); | |
955 | } | |
956 | ||
c6b4fcba JT |
957 | /*---------------------------------------------------------------- |
958 | * Migration processing | |
959 | * | |
960 | * Migration covers moving data from the origin device to the cache, or | |
961 | * vice versa. | |
962 | *--------------------------------------------------------------*/ | |
a59db676 | 963 | static void inc_io_migrations(struct cache *cache) |
c6b4fcba | 964 | { |
a59db676 | 965 | atomic_inc(&cache->nr_io_migrations); |
c6b4fcba JT |
966 | } |
967 | ||
a59db676 | 968 | static void dec_io_migrations(struct cache *cache) |
c6b4fcba | 969 | { |
a59db676 | 970 | atomic_dec(&cache->nr_io_migrations); |
c6b4fcba JT |
971 | } |
972 | ||
651f5fa2 JT |
973 | static void __cell_release(struct cache *cache, struct dm_bio_prison_cell *cell, |
974 | bool holder, struct bio_list *bios) | |
c6b4fcba JT |
975 | { |
976 | (holder ? dm_cell_release : dm_cell_release_no_holder) | |
651f5fa2 | 977 | (cache->prison, cell, bios); |
c6b4fcba JT |
978 | free_prison_cell(cache, cell); |
979 | } | |
980 | ||
651f5fa2 JT |
981 | static bool discard_or_flush(struct bio *bio) |
982 | { | |
983 | return bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_DISCARD); | |
984 | } | |
985 | ||
986 | static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell) | |
987 | { | |
988 | if (discard_or_flush(cell->holder)) | |
989 | /* | |
990 | * We have to handle these bios | |
991 | * individually. | |
992 | */ | |
993 | __cell_release(cache, cell, true, &cache->deferred_bios); | |
994 | ||
995 | else | |
996 | list_add_tail(&cell->user_list, &cache->deferred_cells); | |
997 | } | |
998 | ||
999 | static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, bool holder) | |
c6b4fcba JT |
1000 | { |
1001 | unsigned long flags; | |
1002 | ||
651f5fa2 JT |
1003 | if (!holder && dm_cell_promote_or_release(cache->prison, cell)) { |
1004 | /* | |
1005 | * There was no prisoner to promote to holder, the | |
1006 | * cell has been released. | |
1007 | */ | |
1008 | free_prison_cell(cache, cell); | |
1009 | return; | |
1010 | } | |
1011 | ||
c6b4fcba | 1012 | spin_lock_irqsave(&cache->lock, flags); |
651f5fa2 | 1013 | __cell_defer(cache, cell); |
c6b4fcba JT |
1014 | spin_unlock_irqrestore(&cache->lock, flags); |
1015 | ||
1016 | wake_worker(cache); | |
1017 | } | |
1018 | ||
651f5fa2 JT |
1019 | static void cell_error_with_code(struct cache *cache, struct dm_bio_prison_cell *cell, int err) |
1020 | { | |
1021 | dm_cell_error(cache->prison, cell, err); | |
1022 | dm_bio_prison_free_cell(cache->prison, cell); | |
1023 | } | |
1024 | ||
1025 | static void cell_requeue(struct cache *cache, struct dm_bio_prison_cell *cell) | |
1026 | { | |
1027 | cell_error_with_code(cache, cell, DM_ENDIO_REQUEUE); | |
1028 | } | |
1029 | ||
a59db676 | 1030 | static void free_io_migration(struct dm_cache_migration *mg) |
c6b4fcba | 1031 | { |
a59db676 | 1032 | dec_io_migrations(mg->cache); |
c6b4fcba JT |
1033 | free_migration(mg); |
1034 | } | |
1035 | ||
1036 | static void migration_failure(struct dm_cache_migration *mg) | |
1037 | { | |
1038 | struct cache *cache = mg->cache; | |
1039 | ||
1040 | if (mg->writeback) { | |
1041 | DMWARN_LIMIT("writeback failed; couldn't copy block"); | |
1042 | set_dirty(cache, mg->old_oblock, mg->cblock); | |
1043 | cell_defer(cache, mg->old_ocell, false); | |
1044 | ||
1045 | } else if (mg->demote) { | |
1046 | DMWARN_LIMIT("demotion failed; couldn't copy block"); | |
1047 | policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock); | |
1048 | ||
80f659f3 | 1049 | cell_defer(cache, mg->old_ocell, mg->promote ? false : true); |
c6b4fcba | 1050 | if (mg->promote) |
80f659f3 | 1051 | cell_defer(cache, mg->new_ocell, true); |
c6b4fcba JT |
1052 | } else { |
1053 | DMWARN_LIMIT("promotion failed; couldn't copy block"); | |
1054 | policy_remove_mapping(cache->policy, mg->new_oblock); | |
80f659f3 | 1055 | cell_defer(cache, mg->new_ocell, true); |
c6b4fcba JT |
1056 | } |
1057 | ||
a59db676 | 1058 | free_io_migration(mg); |
c6b4fcba JT |
1059 | } |
1060 | ||
1061 | static void migration_success_pre_commit(struct dm_cache_migration *mg) | |
1062 | { | |
1063 | unsigned long flags; | |
1064 | struct cache *cache = mg->cache; | |
1065 | ||
1066 | if (mg->writeback) { | |
c6b4fcba | 1067 | clear_dirty(cache, mg->old_oblock, mg->cblock); |
40aa978e | 1068 | cell_defer(cache, mg->old_ocell, false); |
a59db676 | 1069 | free_io_migration(mg); |
c6b4fcba JT |
1070 | return; |
1071 | ||
1072 | } else if (mg->demote) { | |
1073 | if (dm_cache_remove_mapping(cache->cmd, mg->cblock)) { | |
1074 | DMWARN_LIMIT("demotion failed; couldn't update on disk metadata"); | |
1075 | policy_force_mapping(cache->policy, mg->new_oblock, | |
1076 | mg->old_oblock); | |
1077 | if (mg->promote) | |
1078 | cell_defer(cache, mg->new_ocell, true); | |
a59db676 | 1079 | free_io_migration(mg); |
c6b4fcba JT |
1080 | return; |
1081 | } | |
1082 | } else { | |
1083 | if (dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock)) { | |
1084 | DMWARN_LIMIT("promotion failed; couldn't update on disk metadata"); | |
1085 | policy_remove_mapping(cache->policy, mg->new_oblock); | |
a59db676 | 1086 | free_io_migration(mg); |
c6b4fcba JT |
1087 | return; |
1088 | } | |
1089 | } | |
1090 | ||
1091 | spin_lock_irqsave(&cache->lock, flags); | |
1092 | list_add_tail(&mg->list, &cache->need_commit_migrations); | |
1093 | cache->commit_requested = true; | |
1094 | spin_unlock_irqrestore(&cache->lock, flags); | |
1095 | } | |
1096 | ||
1097 | static void migration_success_post_commit(struct dm_cache_migration *mg) | |
1098 | { | |
1099 | unsigned long flags; | |
1100 | struct cache *cache = mg->cache; | |
1101 | ||
1102 | if (mg->writeback) { | |
1103 | DMWARN("writeback unexpectedly triggered commit"); | |
1104 | return; | |
1105 | ||
1106 | } else if (mg->demote) { | |
80f659f3 | 1107 | cell_defer(cache, mg->old_ocell, mg->promote ? false : true); |
c6b4fcba JT |
1108 | |
1109 | if (mg->promote) { | |
1110 | mg->demote = false; | |
1111 | ||
1112 | spin_lock_irqsave(&cache->lock, flags); | |
1113 | list_add_tail(&mg->list, &cache->quiesced_migrations); | |
1114 | spin_unlock_irqrestore(&cache->lock, flags); | |
1115 | ||
65790ff9 JT |
1116 | } else { |
1117 | if (mg->invalidate) | |
1118 | policy_remove_mapping(cache->policy, mg->old_oblock); | |
a59db676 | 1119 | free_io_migration(mg); |
65790ff9 | 1120 | } |
c6b4fcba JT |
1121 | |
1122 | } else { | |
1e32134a JT |
1123 | if (mg->requeue_holder) { |
1124 | clear_dirty(cache, mg->new_oblock, mg->cblock); | |
c9d28d5d | 1125 | cell_defer(cache, mg->new_ocell, true); |
1e32134a JT |
1126 | } else { |
1127 | /* | |
1128 | * The block was promoted via an overwrite, so it's dirty. | |
1129 | */ | |
1130 | set_dirty(cache, mg->new_oblock, mg->cblock); | |
c9d28d5d JT |
1131 | bio_endio(mg->new_ocell->holder, 0); |
1132 | cell_defer(cache, mg->new_ocell, false); | |
1133 | } | |
a59db676 | 1134 | free_io_migration(mg); |
c6b4fcba JT |
1135 | } |
1136 | } | |
1137 | ||
1138 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
1139 | { | |
1140 | unsigned long flags; | |
1141 | struct dm_cache_migration *mg = (struct dm_cache_migration *) context; | |
1142 | struct cache *cache = mg->cache; | |
1143 | ||
1144 | if (read_err || write_err) | |
1145 | mg->err = true; | |
1146 | ||
1147 | spin_lock_irqsave(&cache->lock, flags); | |
1148 | list_add_tail(&mg->list, &cache->completed_migrations); | |
1149 | spin_unlock_irqrestore(&cache->lock, flags); | |
1150 | ||
1151 | wake_worker(cache); | |
1152 | } | |
1153 | ||
7ae34e77 | 1154 | static void issue_copy(struct dm_cache_migration *mg) |
c6b4fcba JT |
1155 | { |
1156 | int r; | |
1157 | struct dm_io_region o_region, c_region; | |
1158 | struct cache *cache = mg->cache; | |
8b9d9666 | 1159 | sector_t cblock = from_cblock(mg->cblock); |
c6b4fcba JT |
1160 | |
1161 | o_region.bdev = cache->origin_dev->bdev; | |
1162 | o_region.count = cache->sectors_per_block; | |
1163 | ||
1164 | c_region.bdev = cache->cache_dev->bdev; | |
8b9d9666 | 1165 | c_region.sector = cblock * cache->sectors_per_block; |
c6b4fcba JT |
1166 | c_region.count = cache->sectors_per_block; |
1167 | ||
1168 | if (mg->writeback || mg->demote) { | |
1169 | /* demote */ | |
1170 | o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block; | |
1171 | r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg); | |
1172 | } else { | |
1173 | /* promote */ | |
1174 | o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block; | |
1175 | r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg); | |
1176 | } | |
1177 | ||
2c2263c9 HM |
1178 | if (r < 0) { |
1179 | DMERR_LIMIT("issuing migration failed"); | |
c6b4fcba | 1180 | migration_failure(mg); |
2c2263c9 | 1181 | } |
c6b4fcba JT |
1182 | } |
1183 | ||
c9d28d5d JT |
1184 | static void overwrite_endio(struct bio *bio, int err) |
1185 | { | |
1186 | struct dm_cache_migration *mg = bio->bi_private; | |
1187 | struct cache *cache = mg->cache; | |
1188 | size_t pb_data_size = get_per_bio_data_size(cache); | |
1189 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1190 | unsigned long flags; | |
1191 | ||
80ae49aa MS |
1192 | dm_unhook_bio(&pb->hook_info, bio); |
1193 | ||
c9d28d5d JT |
1194 | if (err) |
1195 | mg->err = true; | |
1196 | ||
80ae49aa MS |
1197 | mg->requeue_holder = false; |
1198 | ||
c9d28d5d JT |
1199 | spin_lock_irqsave(&cache->lock, flags); |
1200 | list_add_tail(&mg->list, &cache->completed_migrations); | |
c9d28d5d JT |
1201 | spin_unlock_irqrestore(&cache->lock, flags); |
1202 | ||
1203 | wake_worker(cache); | |
1204 | } | |
1205 | ||
1206 | static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio) | |
1207 | { | |
1208 | size_t pb_data_size = get_per_bio_data_size(mg->cache); | |
1209 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1210 | ||
1211 | dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg); | |
1212 | remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock); | |
8c081b52 JT |
1213 | |
1214 | /* | |
1215 | * No need to inc_ds() here, since the cell will be held for the | |
1216 | * duration of the io. | |
1217 | */ | |
066dbaa3 | 1218 | accounted_request(mg->cache, bio); |
c9d28d5d JT |
1219 | } |
1220 | ||
1221 | static bool bio_writes_complete_block(struct cache *cache, struct bio *bio) | |
1222 | { | |
1223 | return (bio_data_dir(bio) == WRITE) && | |
4f024f37 | 1224 | (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT)); |
c9d28d5d JT |
1225 | } |
1226 | ||
c6b4fcba JT |
1227 | static void avoid_copy(struct dm_cache_migration *mg) |
1228 | { | |
1229 | atomic_inc(&mg->cache->stats.copies_avoided); | |
1230 | migration_success_pre_commit(mg); | |
1231 | } | |
1232 | ||
7ae34e77 JT |
1233 | static void calc_discard_block_range(struct cache *cache, struct bio *bio, |
1234 | dm_dblock_t *b, dm_dblock_t *e) | |
1235 | { | |
1236 | sector_t sb = bio->bi_iter.bi_sector; | |
1237 | sector_t se = bio_end_sector(bio); | |
1238 | ||
1239 | *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size)); | |
1240 | ||
1241 | if (se - sb < cache->discard_block_size) | |
1242 | *e = *b; | |
1243 | else | |
1244 | *e = to_dblock(block_div(se, cache->discard_block_size)); | |
1245 | } | |
1246 | ||
1247 | static void issue_discard(struct dm_cache_migration *mg) | |
1248 | { | |
1249 | dm_dblock_t b, e; | |
1250 | struct bio *bio = mg->new_ocell->holder; | |
1251 | ||
1252 | calc_discard_block_range(mg->cache, bio, &b, &e); | |
1253 | while (b != e) { | |
1254 | set_discard(mg->cache, b); | |
1255 | b = to_dblock(from_dblock(b) + 1); | |
1256 | } | |
1257 | ||
1258 | bio_endio(bio, 0); | |
1259 | cell_defer(mg->cache, mg->new_ocell, false); | |
1260 | free_migration(mg); | |
1261 | } | |
1262 | ||
1263 | static void issue_copy_or_discard(struct dm_cache_migration *mg) | |
c6b4fcba JT |
1264 | { |
1265 | bool avoid; | |
1266 | struct cache *cache = mg->cache; | |
1267 | ||
7ae34e77 JT |
1268 | if (mg->discard) { |
1269 | issue_discard(mg); | |
1270 | return; | |
1271 | } | |
1272 | ||
c6b4fcba JT |
1273 | if (mg->writeback || mg->demote) |
1274 | avoid = !is_dirty(cache, mg->cblock) || | |
1275 | is_discarded_oblock(cache, mg->old_oblock); | |
c9d28d5d JT |
1276 | else { |
1277 | struct bio *bio = mg->new_ocell->holder; | |
1278 | ||
c6b4fcba JT |
1279 | avoid = is_discarded_oblock(cache, mg->new_oblock); |
1280 | ||
f29a3147 JT |
1281 | if (writeback_mode(&cache->features) && |
1282 | !avoid && bio_writes_complete_block(cache, bio)) { | |
c9d28d5d JT |
1283 | issue_overwrite(mg, bio); |
1284 | return; | |
1285 | } | |
1286 | } | |
1287 | ||
7ae34e77 | 1288 | avoid ? avoid_copy(mg) : issue_copy(mg); |
c6b4fcba JT |
1289 | } |
1290 | ||
1291 | static void complete_migration(struct dm_cache_migration *mg) | |
1292 | { | |
1293 | if (mg->err) | |
1294 | migration_failure(mg); | |
1295 | else | |
1296 | migration_success_pre_commit(mg); | |
1297 | } | |
1298 | ||
1299 | static void process_migrations(struct cache *cache, struct list_head *head, | |
1300 | void (*fn)(struct dm_cache_migration *)) | |
1301 | { | |
1302 | unsigned long flags; | |
1303 | struct list_head list; | |
1304 | struct dm_cache_migration *mg, *tmp; | |
1305 | ||
1306 | INIT_LIST_HEAD(&list); | |
1307 | spin_lock_irqsave(&cache->lock, flags); | |
1308 | list_splice_init(head, &list); | |
1309 | spin_unlock_irqrestore(&cache->lock, flags); | |
1310 | ||
1311 | list_for_each_entry_safe(mg, tmp, &list, list) | |
1312 | fn(mg); | |
1313 | } | |
1314 | ||
1315 | static void __queue_quiesced_migration(struct dm_cache_migration *mg) | |
1316 | { | |
1317 | list_add_tail(&mg->list, &mg->cache->quiesced_migrations); | |
1318 | } | |
1319 | ||
1320 | static void queue_quiesced_migration(struct dm_cache_migration *mg) | |
1321 | { | |
1322 | unsigned long flags; | |
1323 | struct cache *cache = mg->cache; | |
1324 | ||
1325 | spin_lock_irqsave(&cache->lock, flags); | |
1326 | __queue_quiesced_migration(mg); | |
1327 | spin_unlock_irqrestore(&cache->lock, flags); | |
1328 | ||
1329 | wake_worker(cache); | |
1330 | } | |
1331 | ||
1332 | static void queue_quiesced_migrations(struct cache *cache, struct list_head *work) | |
1333 | { | |
1334 | unsigned long flags; | |
1335 | struct dm_cache_migration *mg, *tmp; | |
1336 | ||
1337 | spin_lock_irqsave(&cache->lock, flags); | |
1338 | list_for_each_entry_safe(mg, tmp, work, list) | |
1339 | __queue_quiesced_migration(mg); | |
1340 | spin_unlock_irqrestore(&cache->lock, flags); | |
1341 | ||
1342 | wake_worker(cache); | |
1343 | } | |
1344 | ||
1345 | static void check_for_quiesced_migrations(struct cache *cache, | |
1346 | struct per_bio_data *pb) | |
1347 | { | |
1348 | struct list_head work; | |
1349 | ||
1350 | if (!pb->all_io_entry) | |
1351 | return; | |
1352 | ||
1353 | INIT_LIST_HEAD(&work); | |
8c081b52 | 1354 | dm_deferred_entry_dec(pb->all_io_entry, &work); |
c6b4fcba JT |
1355 | |
1356 | if (!list_empty(&work)) | |
1357 | queue_quiesced_migrations(cache, &work); | |
1358 | } | |
1359 | ||
1360 | static void quiesce_migration(struct dm_cache_migration *mg) | |
1361 | { | |
1362 | if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list)) | |
1363 | queue_quiesced_migration(mg); | |
1364 | } | |
1365 | ||
1366 | static void promote(struct cache *cache, struct prealloc *structs, | |
1367 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1368 | struct dm_bio_prison_cell *cell) | |
1369 | { | |
1370 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1371 | ||
1372 | mg->err = false; | |
7ae34e77 | 1373 | mg->discard = false; |
c6b4fcba JT |
1374 | mg->writeback = false; |
1375 | mg->demote = false; | |
1376 | mg->promote = true; | |
c9d28d5d | 1377 | mg->requeue_holder = true; |
65790ff9 | 1378 | mg->invalidate = false; |
c6b4fcba JT |
1379 | mg->cache = cache; |
1380 | mg->new_oblock = oblock; | |
1381 | mg->cblock = cblock; | |
1382 | mg->old_ocell = NULL; | |
1383 | mg->new_ocell = cell; | |
1384 | mg->start_jiffies = jiffies; | |
1385 | ||
a59db676 | 1386 | inc_io_migrations(cache); |
c6b4fcba JT |
1387 | quiesce_migration(mg); |
1388 | } | |
1389 | ||
1390 | static void writeback(struct cache *cache, struct prealloc *structs, | |
1391 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1392 | struct dm_bio_prison_cell *cell) | |
1393 | { | |
1394 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1395 | ||
1396 | mg->err = false; | |
7ae34e77 | 1397 | mg->discard = false; |
c6b4fcba JT |
1398 | mg->writeback = true; |
1399 | mg->demote = false; | |
1400 | mg->promote = false; | |
c9d28d5d | 1401 | mg->requeue_holder = true; |
65790ff9 | 1402 | mg->invalidate = false; |
c6b4fcba JT |
1403 | mg->cache = cache; |
1404 | mg->old_oblock = oblock; | |
1405 | mg->cblock = cblock; | |
1406 | mg->old_ocell = cell; | |
1407 | mg->new_ocell = NULL; | |
1408 | mg->start_jiffies = jiffies; | |
1409 | ||
a59db676 | 1410 | inc_io_migrations(cache); |
c6b4fcba JT |
1411 | quiesce_migration(mg); |
1412 | } | |
1413 | ||
1414 | static void demote_then_promote(struct cache *cache, struct prealloc *structs, | |
1415 | dm_oblock_t old_oblock, dm_oblock_t new_oblock, | |
1416 | dm_cblock_t cblock, | |
1417 | struct dm_bio_prison_cell *old_ocell, | |
1418 | struct dm_bio_prison_cell *new_ocell) | |
1419 | { | |
1420 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1421 | ||
1422 | mg->err = false; | |
7ae34e77 | 1423 | mg->discard = false; |
c6b4fcba JT |
1424 | mg->writeback = false; |
1425 | mg->demote = true; | |
1426 | mg->promote = true; | |
c9d28d5d | 1427 | mg->requeue_holder = true; |
65790ff9 | 1428 | mg->invalidate = false; |
c6b4fcba JT |
1429 | mg->cache = cache; |
1430 | mg->old_oblock = old_oblock; | |
1431 | mg->new_oblock = new_oblock; | |
1432 | mg->cblock = cblock; | |
1433 | mg->old_ocell = old_ocell; | |
1434 | mg->new_ocell = new_ocell; | |
1435 | mg->start_jiffies = jiffies; | |
1436 | ||
a59db676 | 1437 | inc_io_migrations(cache); |
c6b4fcba JT |
1438 | quiesce_migration(mg); |
1439 | } | |
1440 | ||
2ee57d58 JT |
1441 | /* |
1442 | * Invalidate a cache entry. No writeback occurs; any changes in the cache | |
1443 | * block are thrown away. | |
1444 | */ | |
1445 | static void invalidate(struct cache *cache, struct prealloc *structs, | |
1446 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1447 | struct dm_bio_prison_cell *cell) | |
1448 | { | |
1449 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1450 | ||
1451 | mg->err = false; | |
7ae34e77 | 1452 | mg->discard = false; |
2ee57d58 JT |
1453 | mg->writeback = false; |
1454 | mg->demote = true; | |
1455 | mg->promote = false; | |
1456 | mg->requeue_holder = true; | |
65790ff9 | 1457 | mg->invalidate = true; |
2ee57d58 JT |
1458 | mg->cache = cache; |
1459 | mg->old_oblock = oblock; | |
1460 | mg->cblock = cblock; | |
1461 | mg->old_ocell = cell; | |
1462 | mg->new_ocell = NULL; | |
1463 | mg->start_jiffies = jiffies; | |
1464 | ||
a59db676 | 1465 | inc_io_migrations(cache); |
2ee57d58 JT |
1466 | quiesce_migration(mg); |
1467 | } | |
1468 | ||
7ae34e77 JT |
1469 | static void discard(struct cache *cache, struct prealloc *structs, |
1470 | struct dm_bio_prison_cell *cell) | |
1471 | { | |
1472 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1473 | ||
1474 | mg->err = false; | |
1475 | mg->discard = true; | |
1476 | mg->writeback = false; | |
1477 | mg->demote = false; | |
1478 | mg->promote = false; | |
1479 | mg->requeue_holder = false; | |
1480 | mg->invalidate = false; | |
1481 | mg->cache = cache; | |
1482 | mg->old_ocell = NULL; | |
1483 | mg->new_ocell = cell; | |
1484 | mg->start_jiffies = jiffies; | |
1485 | ||
1486 | quiesce_migration(mg); | |
1487 | } | |
1488 | ||
c6b4fcba JT |
1489 | /*---------------------------------------------------------------- |
1490 | * bio processing | |
1491 | *--------------------------------------------------------------*/ | |
1492 | static void defer_bio(struct cache *cache, struct bio *bio) | |
1493 | { | |
1494 | unsigned long flags; | |
1495 | ||
1496 | spin_lock_irqsave(&cache->lock, flags); | |
1497 | bio_list_add(&cache->deferred_bios, bio); | |
1498 | spin_unlock_irqrestore(&cache->lock, flags); | |
1499 | ||
1500 | wake_worker(cache); | |
1501 | } | |
1502 | ||
1503 | static void process_flush_bio(struct cache *cache, struct bio *bio) | |
1504 | { | |
19b0092e MS |
1505 | size_t pb_data_size = get_per_bio_data_size(cache); |
1506 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba | 1507 | |
4f024f37 | 1508 | BUG_ON(bio->bi_iter.bi_size); |
c6b4fcba JT |
1509 | if (!pb->req_nr) |
1510 | remap_to_origin(cache, bio); | |
1511 | else | |
1512 | remap_to_cache(cache, bio, 0); | |
1513 | ||
8c081b52 JT |
1514 | /* |
1515 | * REQ_FLUSH is not directed at any particular block so we don't | |
1516 | * need to inc_ds(). REQ_FUA's are split into a write + REQ_FLUSH | |
1517 | * by dm-core. | |
1518 | */ | |
c6b4fcba JT |
1519 | issue(cache, bio); |
1520 | } | |
1521 | ||
7ae34e77 JT |
1522 | static void process_discard_bio(struct cache *cache, struct prealloc *structs, |
1523 | struct bio *bio) | |
c6b4fcba | 1524 | { |
7ae34e77 JT |
1525 | int r; |
1526 | dm_dblock_t b, e; | |
1527 | struct dm_bio_prison_cell *cell_prealloc, *new_ocell; | |
c6b4fcba | 1528 | |
7ae34e77 JT |
1529 | calc_discard_block_range(cache, bio, &b, &e); |
1530 | if (b == e) { | |
1531 | bio_endio(bio, 0); | |
1532 | return; | |
1533 | } | |
c6b4fcba | 1534 | |
7ae34e77 JT |
1535 | cell_prealloc = prealloc_get_cell(structs); |
1536 | r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc, | |
1537 | (cell_free_fn) prealloc_put_cell, | |
1538 | structs, &new_ocell); | |
1539 | if (r > 0) | |
1540 | return; | |
c6b4fcba | 1541 | |
7ae34e77 | 1542 | discard(cache, structs, new_ocell); |
c6b4fcba JT |
1543 | } |
1544 | ||
1545 | static bool spare_migration_bandwidth(struct cache *cache) | |
1546 | { | |
a59db676 | 1547 | sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) * |
c6b4fcba JT |
1548 | cache->sectors_per_block; |
1549 | return current_volume < cache->migration_threshold; | |
1550 | } | |
1551 | ||
c6b4fcba JT |
1552 | static void inc_hit_counter(struct cache *cache, struct bio *bio) |
1553 | { | |
1554 | atomic_inc(bio_data_dir(bio) == READ ? | |
1555 | &cache->stats.read_hit : &cache->stats.write_hit); | |
1556 | } | |
1557 | ||
1558 | static void inc_miss_counter(struct cache *cache, struct bio *bio) | |
1559 | { | |
1560 | atomic_inc(bio_data_dir(bio) == READ ? | |
1561 | &cache->stats.read_miss : &cache->stats.write_miss); | |
1562 | } | |
1563 | ||
fb4100ae JT |
1564 | /*----------------------------------------------------------------*/ |
1565 | ||
651f5fa2 JT |
1566 | struct inc_detail { |
1567 | struct cache *cache; | |
1568 | struct bio_list bios_for_issue; | |
1569 | struct bio_list unhandled_bios; | |
1570 | bool any_writes; | |
1571 | }; | |
1572 | ||
1573 | static void inc_fn(void *context, struct dm_bio_prison_cell *cell) | |
1574 | { | |
1575 | struct bio *bio; | |
1576 | struct inc_detail *detail = context; | |
1577 | struct cache *cache = detail->cache; | |
1578 | ||
1579 | inc_ds(cache, cell->holder, cell); | |
1580 | if (bio_data_dir(cell->holder) == WRITE) | |
1581 | detail->any_writes = true; | |
1582 | ||
1583 | while ((bio = bio_list_pop(&cell->bios))) { | |
1584 | if (discard_or_flush(bio)) { | |
1585 | bio_list_add(&detail->unhandled_bios, bio); | |
1586 | continue; | |
1587 | } | |
1588 | ||
1589 | if (bio_data_dir(bio) == WRITE) | |
1590 | detail->any_writes = true; | |
1591 | ||
1592 | bio_list_add(&detail->bios_for_issue, bio); | |
1593 | inc_ds(cache, bio, cell); | |
1594 | } | |
1595 | } | |
1596 | ||
1597 | // FIXME: refactor these two | |
1598 | static void remap_cell_to_origin_clear_discard(struct cache *cache, | |
1599 | struct dm_bio_prison_cell *cell, | |
1600 | dm_oblock_t oblock, bool issue_holder) | |
1601 | { | |
1602 | struct bio *bio; | |
1603 | unsigned long flags; | |
1604 | struct inc_detail detail; | |
1605 | ||
1606 | detail.cache = cache; | |
1607 | bio_list_init(&detail.bios_for_issue); | |
1608 | bio_list_init(&detail.unhandled_bios); | |
1609 | detail.any_writes = false; | |
1610 | ||
1611 | spin_lock_irqsave(&cache->lock, flags); | |
1612 | dm_cell_visit_release(cache->prison, inc_fn, &detail, cell); | |
1613 | bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios); | |
1614 | spin_unlock_irqrestore(&cache->lock, flags); | |
1615 | ||
1616 | remap_to_origin(cache, cell->holder); | |
1617 | if (issue_holder) | |
1618 | issue(cache, cell->holder); | |
1619 | else | |
1620 | accounted_begin(cache, cell->holder); | |
1621 | ||
1622 | if (detail.any_writes) | |
1623 | clear_discard(cache, oblock_to_dblock(cache, oblock)); | |
1624 | ||
1625 | while ((bio = bio_list_pop(&detail.bios_for_issue))) { | |
1626 | remap_to_origin(cache, bio); | |
1627 | issue(cache, bio); | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | static void remap_cell_to_cache_dirty(struct cache *cache, struct dm_bio_prison_cell *cell, | |
1632 | dm_oblock_t oblock, dm_cblock_t cblock, bool issue_holder) | |
1633 | { | |
1634 | struct bio *bio; | |
1635 | unsigned long flags; | |
1636 | struct inc_detail detail; | |
1637 | ||
1638 | detail.cache = cache; | |
1639 | bio_list_init(&detail.bios_for_issue); | |
1640 | bio_list_init(&detail.unhandled_bios); | |
1641 | detail.any_writes = false; | |
1642 | ||
1643 | spin_lock_irqsave(&cache->lock, flags); | |
1644 | dm_cell_visit_release(cache->prison, inc_fn, &detail, cell); | |
1645 | bio_list_merge(&cache->deferred_bios, &detail.unhandled_bios); | |
1646 | spin_unlock_irqrestore(&cache->lock, flags); | |
1647 | ||
1648 | remap_to_cache(cache, cell->holder, cblock); | |
1649 | if (issue_holder) | |
1650 | issue(cache, cell->holder); | |
1651 | else | |
1652 | accounted_begin(cache, cell->holder); | |
1653 | ||
1654 | if (detail.any_writes) { | |
1655 | set_dirty(cache, oblock, cblock); | |
1656 | clear_discard(cache, oblock_to_dblock(cache, oblock)); | |
1657 | } | |
1658 | ||
1659 | while ((bio = bio_list_pop(&detail.bios_for_issue))) { | |
1660 | remap_to_cache(cache, bio, cblock); | |
1661 | issue(cache, bio); | |
1662 | } | |
1663 | } | |
1664 | ||
1665 | /*----------------------------------------------------------------*/ | |
1666 | ||
fb4100ae JT |
1667 | struct old_oblock_lock { |
1668 | struct policy_locker locker; | |
1669 | struct cache *cache; | |
1670 | struct prealloc *structs; | |
1671 | struct dm_bio_prison_cell *cell; | |
1672 | }; | |
1673 | ||
1674 | static int null_locker(struct policy_locker *locker, dm_oblock_t b) | |
1675 | { | |
1676 | /* This should never be called */ | |
1677 | BUG(); | |
1678 | return 0; | |
1679 | } | |
1680 | ||
1681 | static int cell_locker(struct policy_locker *locker, dm_oblock_t b) | |
1682 | { | |
1683 | struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker); | |
1684 | struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs); | |
1685 | ||
1686 | return bio_detain(l->cache, b, NULL, cell_prealloc, | |
1687 | (cell_free_fn) prealloc_put_cell, | |
1688 | l->structs, &l->cell); | |
1689 | } | |
1690 | ||
651f5fa2 JT |
1691 | static void process_cell(struct cache *cache, struct prealloc *structs, |
1692 | struct dm_bio_prison_cell *new_ocell) | |
c6b4fcba JT |
1693 | { |
1694 | int r; | |
1695 | bool release_cell = true; | |
651f5fa2 | 1696 | struct bio *bio = new_ocell->holder; |
c6b4fcba | 1697 | dm_oblock_t block = get_bio_block(cache, bio); |
c6b4fcba | 1698 | struct policy_result lookup_result; |
2ee57d58 | 1699 | bool passthrough = passthrough_mode(&cache->features); |
43c32bf2 | 1700 | bool discarded_block, can_migrate; |
fb4100ae | 1701 | struct old_oblock_lock ool; |
c6b4fcba | 1702 | |
43c32bf2 JT |
1703 | discarded_block = is_discarded_oblock(cache, block); |
1704 | can_migrate = !passthrough && (discarded_block || spare_migration_bandwidth(cache)); | |
1705 | ||
fb4100ae JT |
1706 | ool.locker.fn = cell_locker; |
1707 | ool.cache = cache; | |
1708 | ool.structs = structs; | |
1709 | ool.cell = NULL; | |
c6b4fcba | 1710 | r = policy_map(cache->policy, block, true, can_migrate, discarded_block, |
fb4100ae | 1711 | bio, &ool.locker, &lookup_result); |
c6b4fcba JT |
1712 | |
1713 | if (r == -EWOULDBLOCK) | |
1714 | /* migration has been denied */ | |
1715 | lookup_result.op = POLICY_MISS; | |
1716 | ||
1717 | switch (lookup_result.op) { | |
1718 | case POLICY_HIT: | |
2ee57d58 JT |
1719 | if (passthrough) { |
1720 | inc_miss_counter(cache, bio); | |
c6b4fcba | 1721 | |
2ee57d58 JT |
1722 | /* |
1723 | * Passthrough always maps to the origin, | |
1724 | * invalidating any cache blocks that are written | |
1725 | * to. | |
1726 | */ | |
1727 | ||
1728 | if (bio_data_dir(bio) == WRITE) { | |
1729 | atomic_inc(&cache->stats.demotion); | |
1730 | invalidate(cache, structs, block, lookup_result.cblock, new_ocell); | |
1731 | release_cell = false; | |
1732 | ||
1733 | } else { | |
1734 | /* FIXME: factor out issue_origin() */ | |
2ee57d58 | 1735 | remap_to_origin_clear_discard(cache, bio, block); |
8c081b52 | 1736 | inc_and_issue(cache, bio, new_ocell); |
2ee57d58 JT |
1737 | } |
1738 | } else { | |
1739 | inc_hit_counter(cache, bio); | |
1740 | ||
1741 | if (bio_data_dir(bio) == WRITE && | |
1742 | writethrough_mode(&cache->features) && | |
1743 | !is_dirty(cache, lookup_result.cblock)) { | |
2ee57d58 | 1744 | remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock); |
8c081b52 JT |
1745 | inc_and_issue(cache, bio, new_ocell); |
1746 | ||
651f5fa2 JT |
1747 | } else { |
1748 | remap_cell_to_cache_dirty(cache, new_ocell, block, lookup_result.cblock, true); | |
1749 | release_cell = false; | |
8c081b52 | 1750 | } |
2ee57d58 | 1751 | } |
c6b4fcba | 1752 | |
c6b4fcba JT |
1753 | break; |
1754 | ||
1755 | case POLICY_MISS: | |
1756 | inc_miss_counter(cache, bio); | |
651f5fa2 JT |
1757 | remap_cell_to_origin_clear_discard(cache, new_ocell, block, true); |
1758 | release_cell = false; | |
c6b4fcba JT |
1759 | break; |
1760 | ||
1761 | case POLICY_NEW: | |
1762 | atomic_inc(&cache->stats.promotion); | |
1763 | promote(cache, structs, block, lookup_result.cblock, new_ocell); | |
1764 | release_cell = false; | |
1765 | break; | |
1766 | ||
1767 | case POLICY_REPLACE: | |
c6b4fcba JT |
1768 | atomic_inc(&cache->stats.demotion); |
1769 | atomic_inc(&cache->stats.promotion); | |
c6b4fcba JT |
1770 | demote_then_promote(cache, structs, lookup_result.old_oblock, |
1771 | block, lookup_result.cblock, | |
fb4100ae | 1772 | ool.cell, new_ocell); |
c6b4fcba JT |
1773 | release_cell = false; |
1774 | break; | |
1775 | ||
1776 | default: | |
1777 | DMERR_LIMIT("%s: erroring bio, unknown policy op: %u", __func__, | |
1778 | (unsigned) lookup_result.op); | |
1779 | bio_io_error(bio); | |
1780 | } | |
1781 | ||
1782 | if (release_cell) | |
1783 | cell_defer(cache, new_ocell, false); | |
1784 | } | |
1785 | ||
651f5fa2 JT |
1786 | static void process_bio(struct cache *cache, struct prealloc *structs, |
1787 | struct bio *bio) | |
1788 | { | |
1789 | int r; | |
1790 | dm_oblock_t block = get_bio_block(cache, bio); | |
1791 | struct dm_bio_prison_cell *cell_prealloc, *new_ocell; | |
1792 | ||
1793 | /* | |
1794 | * Check to see if that block is currently migrating. | |
1795 | */ | |
1796 | cell_prealloc = prealloc_get_cell(structs); | |
1797 | r = bio_detain(cache, block, bio, cell_prealloc, | |
1798 | (cell_free_fn) prealloc_put_cell, | |
1799 | structs, &new_ocell); | |
1800 | if (r > 0) | |
1801 | return; | |
1802 | ||
1803 | process_cell(cache, structs, new_ocell); | |
1804 | } | |
1805 | ||
c6b4fcba JT |
1806 | static int need_commit_due_to_time(struct cache *cache) |
1807 | { | |
651f5fa2 JT |
1808 | return jiffies < cache->last_commit_jiffies || |
1809 | jiffies > cache->last_commit_jiffies + COMMIT_PERIOD; | |
c6b4fcba JT |
1810 | } |
1811 | ||
1812 | static int commit_if_needed(struct cache *cache) | |
1813 | { | |
ffcbcb67 HM |
1814 | int r = 0; |
1815 | ||
1816 | if ((cache->commit_requested || need_commit_due_to_time(cache)) && | |
1817 | dm_cache_changed_this_transaction(cache->cmd)) { | |
c6b4fcba | 1818 | atomic_inc(&cache->stats.commit_count); |
c6b4fcba | 1819 | cache->commit_requested = false; |
ffcbcb67 HM |
1820 | r = dm_cache_commit(cache->cmd, false); |
1821 | cache->last_commit_jiffies = jiffies; | |
c6b4fcba JT |
1822 | } |
1823 | ||
ffcbcb67 | 1824 | return r; |
c6b4fcba JT |
1825 | } |
1826 | ||
1827 | static void process_deferred_bios(struct cache *cache) | |
1828 | { | |
1829 | unsigned long flags; | |
1830 | struct bio_list bios; | |
1831 | struct bio *bio; | |
1832 | struct prealloc structs; | |
1833 | ||
1834 | memset(&structs, 0, sizeof(structs)); | |
1835 | bio_list_init(&bios); | |
1836 | ||
1837 | spin_lock_irqsave(&cache->lock, flags); | |
1838 | bio_list_merge(&bios, &cache->deferred_bios); | |
1839 | bio_list_init(&cache->deferred_bios); | |
1840 | spin_unlock_irqrestore(&cache->lock, flags); | |
1841 | ||
1842 | while (!bio_list_empty(&bios)) { | |
1843 | /* | |
1844 | * If we've got no free migration structs, and processing | |
1845 | * this bio might require one, we pause until there are some | |
1846 | * prepared mappings to process. | |
1847 | */ | |
1848 | if (prealloc_data_structs(cache, &structs)) { | |
1849 | spin_lock_irqsave(&cache->lock, flags); | |
1850 | bio_list_merge(&cache->deferred_bios, &bios); | |
1851 | spin_unlock_irqrestore(&cache->lock, flags); | |
1852 | break; | |
1853 | } | |
1854 | ||
1855 | bio = bio_list_pop(&bios); | |
1856 | ||
1857 | if (bio->bi_rw & REQ_FLUSH) | |
1858 | process_flush_bio(cache, bio); | |
1859 | else if (bio->bi_rw & REQ_DISCARD) | |
7ae34e77 | 1860 | process_discard_bio(cache, &structs, bio); |
c6b4fcba JT |
1861 | else |
1862 | process_bio(cache, &structs, bio); | |
1863 | } | |
1864 | ||
1865 | prealloc_free_structs(cache, &structs); | |
1866 | } | |
1867 | ||
651f5fa2 JT |
1868 | static void process_deferred_cells(struct cache *cache) |
1869 | { | |
1870 | unsigned long flags; | |
1871 | struct dm_bio_prison_cell *cell, *tmp; | |
1872 | struct list_head cells; | |
1873 | struct prealloc structs; | |
1874 | ||
1875 | memset(&structs, 0, sizeof(structs)); | |
1876 | ||
1877 | INIT_LIST_HEAD(&cells); | |
1878 | ||
1879 | spin_lock_irqsave(&cache->lock, flags); | |
1880 | list_splice_init(&cache->deferred_cells, &cells); | |
1881 | spin_unlock_irqrestore(&cache->lock, flags); | |
1882 | ||
1883 | list_for_each_entry_safe(cell, tmp, &cells, user_list) { | |
1884 | /* | |
1885 | * If we've got no free migration structs, and processing | |
1886 | * this bio might require one, we pause until there are some | |
1887 | * prepared mappings to process. | |
1888 | */ | |
1889 | if (prealloc_data_structs(cache, &structs)) { | |
1890 | spin_lock_irqsave(&cache->lock, flags); | |
1891 | list_splice(&cells, &cache->deferred_cells); | |
1892 | spin_unlock_irqrestore(&cache->lock, flags); | |
1893 | break; | |
1894 | } | |
1895 | ||
1896 | process_cell(cache, &structs, cell); | |
1897 | } | |
1898 | ||
1899 | prealloc_free_structs(cache, &structs); | |
1900 | } | |
1901 | ||
c6b4fcba JT |
1902 | static void process_deferred_flush_bios(struct cache *cache, bool submit_bios) |
1903 | { | |
1904 | unsigned long flags; | |
1905 | struct bio_list bios; | |
1906 | struct bio *bio; | |
1907 | ||
1908 | bio_list_init(&bios); | |
1909 | ||
1910 | spin_lock_irqsave(&cache->lock, flags); | |
1911 | bio_list_merge(&bios, &cache->deferred_flush_bios); | |
1912 | bio_list_init(&cache->deferred_flush_bios); | |
1913 | spin_unlock_irqrestore(&cache->lock, flags); | |
1914 | ||
8c081b52 JT |
1915 | /* |
1916 | * These bios have already been through inc_ds() | |
1917 | */ | |
c6b4fcba | 1918 | while ((bio = bio_list_pop(&bios))) |
066dbaa3 | 1919 | submit_bios ? accounted_request(cache, bio) : bio_io_error(bio); |
c6b4fcba JT |
1920 | } |
1921 | ||
e2e74d61 JT |
1922 | static void process_deferred_writethrough_bios(struct cache *cache) |
1923 | { | |
1924 | unsigned long flags; | |
1925 | struct bio_list bios; | |
1926 | struct bio *bio; | |
1927 | ||
1928 | bio_list_init(&bios); | |
1929 | ||
1930 | spin_lock_irqsave(&cache->lock, flags); | |
1931 | bio_list_merge(&bios, &cache->deferred_writethrough_bios); | |
1932 | bio_list_init(&cache->deferred_writethrough_bios); | |
1933 | spin_unlock_irqrestore(&cache->lock, flags); | |
1934 | ||
8c081b52 JT |
1935 | /* |
1936 | * These bios have already been through inc_ds() | |
1937 | */ | |
e2e74d61 | 1938 | while ((bio = bio_list_pop(&bios))) |
066dbaa3 | 1939 | accounted_request(cache, bio); |
e2e74d61 JT |
1940 | } |
1941 | ||
c6b4fcba JT |
1942 | static void writeback_some_dirty_blocks(struct cache *cache) |
1943 | { | |
1944 | int r = 0; | |
1945 | dm_oblock_t oblock; | |
1946 | dm_cblock_t cblock; | |
1947 | struct prealloc structs; | |
1948 | struct dm_bio_prison_cell *old_ocell; | |
20f6814b | 1949 | bool busy = !iot_idle_for(&cache->origin_tracker, HZ); |
c6b4fcba JT |
1950 | |
1951 | memset(&structs, 0, sizeof(structs)); | |
1952 | ||
1953 | while (spare_migration_bandwidth(cache)) { | |
1954 | if (prealloc_data_structs(cache, &structs)) | |
1955 | break; | |
1956 | ||
20f6814b | 1957 | r = policy_writeback_work(cache->policy, &oblock, &cblock, busy); |
c6b4fcba JT |
1958 | if (r) |
1959 | break; | |
1960 | ||
1961 | r = get_cell(cache, oblock, &structs, &old_ocell); | |
1962 | if (r) { | |
1963 | policy_set_dirty(cache->policy, oblock); | |
1964 | break; | |
1965 | } | |
1966 | ||
1967 | writeback(cache, &structs, oblock, cblock, old_ocell); | |
1968 | } | |
1969 | ||
1970 | prealloc_free_structs(cache, &structs); | |
1971 | } | |
1972 | ||
65790ff9 JT |
1973 | /*---------------------------------------------------------------- |
1974 | * Invalidations. | |
1975 | * Dropping something from the cache *without* writing back. | |
1976 | *--------------------------------------------------------------*/ | |
1977 | ||
1978 | static void process_invalidation_request(struct cache *cache, struct invalidation_request *req) | |
1979 | { | |
1980 | int r = 0; | |
1981 | uint64_t begin = from_cblock(req->cblocks->begin); | |
1982 | uint64_t end = from_cblock(req->cblocks->end); | |
1983 | ||
1984 | while (begin != end) { | |
1985 | r = policy_remove_cblock(cache->policy, to_cblock(begin)); | |
1986 | if (!r) { | |
1987 | r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin)); | |
1988 | if (r) | |
1989 | break; | |
1990 | ||
1991 | } else if (r == -ENODATA) { | |
1992 | /* harmless, already unmapped */ | |
1993 | r = 0; | |
1994 | ||
1995 | } else { | |
1996 | DMERR("policy_remove_cblock failed"); | |
1997 | break; | |
1998 | } | |
1999 | ||
2000 | begin++; | |
2001 | } | |
2002 | ||
2003 | cache->commit_requested = true; | |
2004 | ||
2005 | req->err = r; | |
2006 | atomic_set(&req->complete, 1); | |
2007 | ||
2008 | wake_up(&req->result_wait); | |
2009 | } | |
2010 | ||
2011 | static void process_invalidation_requests(struct cache *cache) | |
2012 | { | |
2013 | struct list_head list; | |
2014 | struct invalidation_request *req, *tmp; | |
2015 | ||
2016 | INIT_LIST_HEAD(&list); | |
2017 | spin_lock(&cache->invalidation_lock); | |
2018 | list_splice_init(&cache->invalidation_requests, &list); | |
2019 | spin_unlock(&cache->invalidation_lock); | |
2020 | ||
2021 | list_for_each_entry_safe (req, tmp, &list, list) | |
2022 | process_invalidation_request(cache, req); | |
2023 | } | |
2024 | ||
c6b4fcba JT |
2025 | /*---------------------------------------------------------------- |
2026 | * Main worker loop | |
2027 | *--------------------------------------------------------------*/ | |
66cb1910 | 2028 | static bool is_quiescing(struct cache *cache) |
c6b4fcba | 2029 | { |
238f8363 | 2030 | return atomic_read(&cache->quiescing); |
c6b4fcba JT |
2031 | } |
2032 | ||
66cb1910 JT |
2033 | static void ack_quiescing(struct cache *cache) |
2034 | { | |
2035 | if (is_quiescing(cache)) { | |
2036 | atomic_inc(&cache->quiescing_ack); | |
2037 | wake_up(&cache->quiescing_wait); | |
2038 | } | |
2039 | } | |
2040 | ||
2041 | static void wait_for_quiescing_ack(struct cache *cache) | |
2042 | { | |
2043 | wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack)); | |
2044 | } | |
2045 | ||
2046 | static void start_quiescing(struct cache *cache) | |
c6b4fcba | 2047 | { |
238f8363 | 2048 | atomic_inc(&cache->quiescing); |
66cb1910 | 2049 | wait_for_quiescing_ack(cache); |
c6b4fcba JT |
2050 | } |
2051 | ||
66cb1910 | 2052 | static void stop_quiescing(struct cache *cache) |
c6b4fcba | 2053 | { |
238f8363 | 2054 | atomic_set(&cache->quiescing, 0); |
66cb1910 | 2055 | atomic_set(&cache->quiescing_ack, 0); |
c6b4fcba JT |
2056 | } |
2057 | ||
2058 | static void wait_for_migrations(struct cache *cache) | |
2059 | { | |
a59db676 | 2060 | wait_event(cache->migration_wait, !atomic_read(&cache->nr_allocated_migrations)); |
c6b4fcba JT |
2061 | } |
2062 | ||
2063 | static void stop_worker(struct cache *cache) | |
2064 | { | |
2065 | cancel_delayed_work(&cache->waker); | |
2066 | flush_workqueue(cache->wq); | |
2067 | } | |
2068 | ||
651f5fa2 JT |
2069 | static void requeue_deferred_cells(struct cache *cache) |
2070 | { | |
2071 | unsigned long flags; | |
2072 | struct list_head cells; | |
2073 | struct dm_bio_prison_cell *cell, *tmp; | |
2074 | ||
2075 | INIT_LIST_HEAD(&cells); | |
2076 | spin_lock_irqsave(&cache->lock, flags); | |
2077 | list_splice_init(&cache->deferred_cells, &cells); | |
2078 | spin_unlock_irqrestore(&cache->lock, flags); | |
2079 | ||
2080 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
2081 | cell_requeue(cache, cell); | |
2082 | } | |
2083 | ||
2084 | static void requeue_deferred_bios(struct cache *cache) | |
c6b4fcba JT |
2085 | { |
2086 | struct bio *bio; | |
2087 | struct bio_list bios; | |
2088 | ||
2089 | bio_list_init(&bios); | |
2090 | bio_list_merge(&bios, &cache->deferred_bios); | |
2091 | bio_list_init(&cache->deferred_bios); | |
2092 | ||
2093 | while ((bio = bio_list_pop(&bios))) | |
2094 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
2095 | } | |
2096 | ||
2097 | static int more_work(struct cache *cache) | |
2098 | { | |
2099 | if (is_quiescing(cache)) | |
2100 | return !list_empty(&cache->quiesced_migrations) || | |
2101 | !list_empty(&cache->completed_migrations) || | |
2102 | !list_empty(&cache->need_commit_migrations); | |
2103 | else | |
2104 | return !bio_list_empty(&cache->deferred_bios) || | |
651f5fa2 | 2105 | !list_empty(&cache->deferred_cells) || |
c6b4fcba | 2106 | !bio_list_empty(&cache->deferred_flush_bios) || |
e2e74d61 | 2107 | !bio_list_empty(&cache->deferred_writethrough_bios) || |
c6b4fcba JT |
2108 | !list_empty(&cache->quiesced_migrations) || |
2109 | !list_empty(&cache->completed_migrations) || | |
65790ff9 JT |
2110 | !list_empty(&cache->need_commit_migrations) || |
2111 | cache->invalidate; | |
c6b4fcba JT |
2112 | } |
2113 | ||
2114 | static void do_worker(struct work_struct *ws) | |
2115 | { | |
2116 | struct cache *cache = container_of(ws, struct cache, worker); | |
2117 | ||
2118 | do { | |
66cb1910 JT |
2119 | if (!is_quiescing(cache)) { |
2120 | writeback_some_dirty_blocks(cache); | |
2121 | process_deferred_writethrough_bios(cache); | |
c6b4fcba | 2122 | process_deferred_bios(cache); |
651f5fa2 | 2123 | process_deferred_cells(cache); |
65790ff9 | 2124 | process_invalidation_requests(cache); |
66cb1910 | 2125 | } |
c6b4fcba | 2126 | |
7ae34e77 | 2127 | process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard); |
c6b4fcba JT |
2128 | process_migrations(cache, &cache->completed_migrations, complete_migration); |
2129 | ||
c6b4fcba JT |
2130 | if (commit_if_needed(cache)) { |
2131 | process_deferred_flush_bios(cache, false); | |
304affaa | 2132 | process_migrations(cache, &cache->need_commit_migrations, migration_failure); |
c6b4fcba JT |
2133 | |
2134 | /* | |
2135 | * FIXME: rollback metadata or just go into a | |
2136 | * failure mode and error everything | |
2137 | */ | |
651f5fa2 | 2138 | |
c6b4fcba JT |
2139 | } else { |
2140 | process_deferred_flush_bios(cache, true); | |
2141 | process_migrations(cache, &cache->need_commit_migrations, | |
2142 | migration_success_post_commit); | |
2143 | } | |
66cb1910 JT |
2144 | |
2145 | ack_quiescing(cache); | |
2146 | ||
c6b4fcba JT |
2147 | } while (more_work(cache)); |
2148 | } | |
2149 | ||
2150 | /* | |
2151 | * We want to commit periodically so that not too much | |
2152 | * unwritten metadata builds up. | |
2153 | */ | |
2154 | static void do_waker(struct work_struct *ws) | |
2155 | { | |
2156 | struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker); | |
f8350daf | 2157 | policy_tick(cache->policy); |
c6b4fcba JT |
2158 | wake_worker(cache); |
2159 | queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD); | |
2160 | } | |
2161 | ||
2162 | /*----------------------------------------------------------------*/ | |
2163 | ||
2164 | static int is_congested(struct dm_dev *dev, int bdi_bits) | |
2165 | { | |
2166 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
2167 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
2168 | } | |
2169 | ||
2170 | static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2171 | { | |
2172 | struct cache *cache = container_of(cb, struct cache, callbacks); | |
2173 | ||
2174 | return is_congested(cache->origin_dev, bdi_bits) || | |
2175 | is_congested(cache->cache_dev, bdi_bits); | |
2176 | } | |
2177 | ||
2178 | /*---------------------------------------------------------------- | |
2179 | * Target methods | |
2180 | *--------------------------------------------------------------*/ | |
2181 | ||
2182 | /* | |
2183 | * This function gets called on the error paths of the constructor, so we | |
2184 | * have to cope with a partially initialised struct. | |
2185 | */ | |
2186 | static void destroy(struct cache *cache) | |
2187 | { | |
2188 | unsigned i; | |
2189 | ||
c6b4fcba JT |
2190 | if (cache->migration_pool) |
2191 | mempool_destroy(cache->migration_pool); | |
2192 | ||
2193 | if (cache->all_io_ds) | |
2194 | dm_deferred_set_destroy(cache->all_io_ds); | |
2195 | ||
2196 | if (cache->prison) | |
2197 | dm_bio_prison_destroy(cache->prison); | |
2198 | ||
2199 | if (cache->wq) | |
2200 | destroy_workqueue(cache->wq); | |
2201 | ||
2202 | if (cache->dirty_bitset) | |
2203 | free_bitset(cache->dirty_bitset); | |
2204 | ||
2205 | if (cache->discard_bitset) | |
2206 | free_bitset(cache->discard_bitset); | |
2207 | ||
2208 | if (cache->copier) | |
2209 | dm_kcopyd_client_destroy(cache->copier); | |
2210 | ||
2211 | if (cache->cmd) | |
2212 | dm_cache_metadata_close(cache->cmd); | |
2213 | ||
2214 | if (cache->metadata_dev) | |
2215 | dm_put_device(cache->ti, cache->metadata_dev); | |
2216 | ||
2217 | if (cache->origin_dev) | |
2218 | dm_put_device(cache->ti, cache->origin_dev); | |
2219 | ||
2220 | if (cache->cache_dev) | |
2221 | dm_put_device(cache->ti, cache->cache_dev); | |
2222 | ||
2223 | if (cache->policy) | |
2224 | dm_cache_policy_destroy(cache->policy); | |
2225 | ||
2226 | for (i = 0; i < cache->nr_ctr_args ; i++) | |
2227 | kfree(cache->ctr_args[i]); | |
2228 | kfree(cache->ctr_args); | |
2229 | ||
2230 | kfree(cache); | |
2231 | } | |
2232 | ||
2233 | static void cache_dtr(struct dm_target *ti) | |
2234 | { | |
2235 | struct cache *cache = ti->private; | |
2236 | ||
2237 | destroy(cache); | |
2238 | } | |
2239 | ||
2240 | static sector_t get_dev_size(struct dm_dev *dev) | |
2241 | { | |
2242 | return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
2243 | } | |
2244 | ||
2245 | /*----------------------------------------------------------------*/ | |
2246 | ||
2247 | /* | |
2248 | * Construct a cache device mapping. | |
2249 | * | |
2250 | * cache <metadata dev> <cache dev> <origin dev> <block size> | |
2251 | * <#feature args> [<feature arg>]* | |
2252 | * <policy> <#policy args> [<policy arg>]* | |
2253 | * | |
2254 | * metadata dev : fast device holding the persistent metadata | |
2255 | * cache dev : fast device holding cached data blocks | |
2256 | * origin dev : slow device holding original data blocks | |
2257 | * block size : cache unit size in sectors | |
2258 | * | |
2259 | * #feature args : number of feature arguments passed | |
2260 | * feature args : writethrough. (The default is writeback.) | |
2261 | * | |
2262 | * policy : the replacement policy to use | |
2263 | * #policy args : an even number of policy arguments corresponding | |
2264 | * to key/value pairs passed to the policy | |
2265 | * policy args : key/value pairs passed to the policy | |
2266 | * E.g. 'sequential_threshold 1024' | |
2267 | * See cache-policies.txt for details. | |
2268 | * | |
2269 | * Optional feature arguments are: | |
2270 | * writethrough : write through caching that prohibits cache block | |
2271 | * content from being different from origin block content. | |
2272 | * Without this argument, the default behaviour is to write | |
2273 | * back cache block contents later for performance reasons, | |
2274 | * so they may differ from the corresponding origin blocks. | |
2275 | */ | |
2276 | struct cache_args { | |
2277 | struct dm_target *ti; | |
2278 | ||
2279 | struct dm_dev *metadata_dev; | |
2280 | ||
2281 | struct dm_dev *cache_dev; | |
2282 | sector_t cache_sectors; | |
2283 | ||
2284 | struct dm_dev *origin_dev; | |
2285 | sector_t origin_sectors; | |
2286 | ||
2287 | uint32_t block_size; | |
2288 | ||
2289 | const char *policy_name; | |
2290 | int policy_argc; | |
2291 | const char **policy_argv; | |
2292 | ||
2293 | struct cache_features features; | |
2294 | }; | |
2295 | ||
2296 | static void destroy_cache_args(struct cache_args *ca) | |
2297 | { | |
2298 | if (ca->metadata_dev) | |
2299 | dm_put_device(ca->ti, ca->metadata_dev); | |
2300 | ||
2301 | if (ca->cache_dev) | |
2302 | dm_put_device(ca->ti, ca->cache_dev); | |
2303 | ||
2304 | if (ca->origin_dev) | |
2305 | dm_put_device(ca->ti, ca->origin_dev); | |
2306 | ||
2307 | kfree(ca); | |
2308 | } | |
2309 | ||
2310 | static bool at_least_one_arg(struct dm_arg_set *as, char **error) | |
2311 | { | |
2312 | if (!as->argc) { | |
2313 | *error = "Insufficient args"; | |
2314 | return false; | |
2315 | } | |
2316 | ||
2317 | return true; | |
2318 | } | |
2319 | ||
2320 | static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2321 | char **error) | |
2322 | { | |
2323 | int r; | |
2324 | sector_t metadata_dev_size; | |
2325 | char b[BDEVNAME_SIZE]; | |
2326 | ||
2327 | if (!at_least_one_arg(as, error)) | |
2328 | return -EINVAL; | |
2329 | ||
2330 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2331 | &ca->metadata_dev); | |
2332 | if (r) { | |
2333 | *error = "Error opening metadata device"; | |
2334 | return r; | |
2335 | } | |
2336 | ||
2337 | metadata_dev_size = get_dev_size(ca->metadata_dev); | |
2338 | if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING) | |
2339 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
2340 | bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
2341 | ||
2342 | return 0; | |
2343 | } | |
2344 | ||
2345 | static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2346 | char **error) | |
2347 | { | |
2348 | int r; | |
2349 | ||
2350 | if (!at_least_one_arg(as, error)) | |
2351 | return -EINVAL; | |
2352 | ||
2353 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2354 | &ca->cache_dev); | |
2355 | if (r) { | |
2356 | *error = "Error opening cache device"; | |
2357 | return r; | |
2358 | } | |
2359 | ca->cache_sectors = get_dev_size(ca->cache_dev); | |
2360 | ||
2361 | return 0; | |
2362 | } | |
2363 | ||
2364 | static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2365 | char **error) | |
2366 | { | |
2367 | int r; | |
2368 | ||
2369 | if (!at_least_one_arg(as, error)) | |
2370 | return -EINVAL; | |
2371 | ||
2372 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2373 | &ca->origin_dev); | |
2374 | if (r) { | |
2375 | *error = "Error opening origin device"; | |
2376 | return r; | |
2377 | } | |
2378 | ||
2379 | ca->origin_sectors = get_dev_size(ca->origin_dev); | |
2380 | if (ca->ti->len > ca->origin_sectors) { | |
2381 | *error = "Device size larger than cached device"; | |
2382 | return -EINVAL; | |
2383 | } | |
2384 | ||
2385 | return 0; | |
2386 | } | |
2387 | ||
2388 | static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as, | |
2389 | char **error) | |
2390 | { | |
05473044 | 2391 | unsigned long block_size; |
c6b4fcba JT |
2392 | |
2393 | if (!at_least_one_arg(as, error)) | |
2394 | return -EINVAL; | |
2395 | ||
05473044 MS |
2396 | if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size || |
2397 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2398 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
2399 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { | |
c6b4fcba JT |
2400 | *error = "Invalid data block size"; |
2401 | return -EINVAL; | |
2402 | } | |
2403 | ||
05473044 | 2404 | if (block_size > ca->cache_sectors) { |
c6b4fcba JT |
2405 | *error = "Data block size is larger than the cache device"; |
2406 | return -EINVAL; | |
2407 | } | |
2408 | ||
05473044 | 2409 | ca->block_size = block_size; |
c6b4fcba JT |
2410 | |
2411 | return 0; | |
2412 | } | |
2413 | ||
2414 | static void init_features(struct cache_features *cf) | |
2415 | { | |
2416 | cf->mode = CM_WRITE; | |
2ee57d58 | 2417 | cf->io_mode = CM_IO_WRITEBACK; |
c6b4fcba JT |
2418 | } |
2419 | ||
2420 | static int parse_features(struct cache_args *ca, struct dm_arg_set *as, | |
2421 | char **error) | |
2422 | { | |
2423 | static struct dm_arg _args[] = { | |
2424 | {0, 1, "Invalid number of cache feature arguments"}, | |
2425 | }; | |
2426 | ||
2427 | int r; | |
2428 | unsigned argc; | |
2429 | const char *arg; | |
2430 | struct cache_features *cf = &ca->features; | |
2431 | ||
2432 | init_features(cf); | |
2433 | ||
2434 | r = dm_read_arg_group(_args, as, &argc, error); | |
2435 | if (r) | |
2436 | return -EINVAL; | |
2437 | ||
2438 | while (argc--) { | |
2439 | arg = dm_shift_arg(as); | |
2440 | ||
2441 | if (!strcasecmp(arg, "writeback")) | |
2ee57d58 | 2442 | cf->io_mode = CM_IO_WRITEBACK; |
c6b4fcba JT |
2443 | |
2444 | else if (!strcasecmp(arg, "writethrough")) | |
2ee57d58 JT |
2445 | cf->io_mode = CM_IO_WRITETHROUGH; |
2446 | ||
2447 | else if (!strcasecmp(arg, "passthrough")) | |
2448 | cf->io_mode = CM_IO_PASSTHROUGH; | |
c6b4fcba JT |
2449 | |
2450 | else { | |
2451 | *error = "Unrecognised cache feature requested"; | |
2452 | return -EINVAL; | |
2453 | } | |
2454 | } | |
2455 | ||
2456 | return 0; | |
2457 | } | |
2458 | ||
2459 | static int parse_policy(struct cache_args *ca, struct dm_arg_set *as, | |
2460 | char **error) | |
2461 | { | |
2462 | static struct dm_arg _args[] = { | |
2463 | {0, 1024, "Invalid number of policy arguments"}, | |
2464 | }; | |
2465 | ||
2466 | int r; | |
2467 | ||
2468 | if (!at_least_one_arg(as, error)) | |
2469 | return -EINVAL; | |
2470 | ||
2471 | ca->policy_name = dm_shift_arg(as); | |
2472 | ||
2473 | r = dm_read_arg_group(_args, as, &ca->policy_argc, error); | |
2474 | if (r) | |
2475 | return -EINVAL; | |
2476 | ||
2477 | ca->policy_argv = (const char **)as->argv; | |
2478 | dm_consume_args(as, ca->policy_argc); | |
2479 | ||
2480 | return 0; | |
2481 | } | |
2482 | ||
2483 | static int parse_cache_args(struct cache_args *ca, int argc, char **argv, | |
2484 | char **error) | |
2485 | { | |
2486 | int r; | |
2487 | struct dm_arg_set as; | |
2488 | ||
2489 | as.argc = argc; | |
2490 | as.argv = argv; | |
2491 | ||
2492 | r = parse_metadata_dev(ca, &as, error); | |
2493 | if (r) | |
2494 | return r; | |
2495 | ||
2496 | r = parse_cache_dev(ca, &as, error); | |
2497 | if (r) | |
2498 | return r; | |
2499 | ||
2500 | r = parse_origin_dev(ca, &as, error); | |
2501 | if (r) | |
2502 | return r; | |
2503 | ||
2504 | r = parse_block_size(ca, &as, error); | |
2505 | if (r) | |
2506 | return r; | |
2507 | ||
2508 | r = parse_features(ca, &as, error); | |
2509 | if (r) | |
2510 | return r; | |
2511 | ||
2512 | r = parse_policy(ca, &as, error); | |
2513 | if (r) | |
2514 | return r; | |
2515 | ||
2516 | return 0; | |
2517 | } | |
2518 | ||
2519 | /*----------------------------------------------------------------*/ | |
2520 | ||
2521 | static struct kmem_cache *migration_cache; | |
2522 | ||
2c73c471 AK |
2523 | #define NOT_CORE_OPTION 1 |
2524 | ||
2f14f4b5 | 2525 | static int process_config_option(struct cache *cache, const char *key, const char *value) |
2c73c471 AK |
2526 | { |
2527 | unsigned long tmp; | |
2528 | ||
2f14f4b5 JT |
2529 | if (!strcasecmp(key, "migration_threshold")) { |
2530 | if (kstrtoul(value, 10, &tmp)) | |
2c73c471 AK |
2531 | return -EINVAL; |
2532 | ||
2533 | cache->migration_threshold = tmp; | |
2534 | return 0; | |
2535 | } | |
2536 | ||
2537 | return NOT_CORE_OPTION; | |
2538 | } | |
2539 | ||
2f14f4b5 JT |
2540 | static int set_config_value(struct cache *cache, const char *key, const char *value) |
2541 | { | |
2542 | int r = process_config_option(cache, key, value); | |
2543 | ||
2544 | if (r == NOT_CORE_OPTION) | |
2545 | r = policy_set_config_value(cache->policy, key, value); | |
2546 | ||
2547 | if (r) | |
2548 | DMWARN("bad config value for %s: %s", key, value); | |
2549 | ||
2550 | return r; | |
2551 | } | |
2552 | ||
2553 | static int set_config_values(struct cache *cache, int argc, const char **argv) | |
c6b4fcba JT |
2554 | { |
2555 | int r = 0; | |
2556 | ||
2557 | if (argc & 1) { | |
2558 | DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs."); | |
2559 | return -EINVAL; | |
2560 | } | |
2561 | ||
2562 | while (argc) { | |
2f14f4b5 JT |
2563 | r = set_config_value(cache, argv[0], argv[1]); |
2564 | if (r) | |
2565 | break; | |
c6b4fcba JT |
2566 | |
2567 | argc -= 2; | |
2568 | argv += 2; | |
2569 | } | |
2570 | ||
2571 | return r; | |
2572 | } | |
2573 | ||
2574 | static int create_cache_policy(struct cache *cache, struct cache_args *ca, | |
2575 | char **error) | |
2576 | { | |
4cb3e1db MP |
2577 | struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name, |
2578 | cache->cache_size, | |
2579 | cache->origin_sectors, | |
2580 | cache->sectors_per_block); | |
2581 | if (IS_ERR(p)) { | |
c6b4fcba | 2582 | *error = "Error creating cache's policy"; |
4cb3e1db | 2583 | return PTR_ERR(p); |
c6b4fcba | 2584 | } |
4cb3e1db | 2585 | cache->policy = p; |
c6b4fcba | 2586 | |
2f14f4b5 | 2587 | return 0; |
c6b4fcba JT |
2588 | } |
2589 | ||
08b18451 | 2590 | /* |
2bb812df JT |
2591 | * We want the discard block size to be at least the size of the cache |
2592 | * block size and have no more than 2^14 discard blocks across the origin. | |
08b18451 JT |
2593 | */ |
2594 | #define MAX_DISCARD_BLOCKS (1 << 14) | |
2595 | ||
2596 | static bool too_many_discard_blocks(sector_t discard_block_size, | |
2597 | sector_t origin_size) | |
2598 | { | |
2599 | (void) sector_div(origin_size, discard_block_size); | |
2600 | ||
2601 | return origin_size > MAX_DISCARD_BLOCKS; | |
2602 | } | |
2603 | ||
2604 | static sector_t calculate_discard_block_size(sector_t cache_block_size, | |
2605 | sector_t origin_size) | |
2606 | { | |
2bb812df | 2607 | sector_t discard_block_size = cache_block_size; |
08b18451 JT |
2608 | |
2609 | if (origin_size) | |
2610 | while (too_many_discard_blocks(discard_block_size, origin_size)) | |
2611 | discard_block_size *= 2; | |
2612 | ||
2613 | return discard_block_size; | |
2614 | } | |
2615 | ||
d1d9220c JT |
2616 | static void set_cache_size(struct cache *cache, dm_cblock_t size) |
2617 | { | |
2618 | dm_block_t nr_blocks = from_cblock(size); | |
2619 | ||
2620 | if (nr_blocks > (1 << 20) && cache->cache_size != size) | |
2621 | DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n" | |
2622 | "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n" | |
2623 | "Please consider increasing the cache block size to reduce the overall cache block count.", | |
2624 | (unsigned long long) nr_blocks); | |
2625 | ||
2626 | cache->cache_size = size; | |
2627 | } | |
2628 | ||
f8350daf | 2629 | #define DEFAULT_MIGRATION_THRESHOLD 2048 |
c6b4fcba | 2630 | |
c6b4fcba JT |
2631 | static int cache_create(struct cache_args *ca, struct cache **result) |
2632 | { | |
2633 | int r = 0; | |
2634 | char **error = &ca->ti->error; | |
2635 | struct cache *cache; | |
2636 | struct dm_target *ti = ca->ti; | |
2637 | dm_block_t origin_blocks; | |
2638 | struct dm_cache_metadata *cmd; | |
2639 | bool may_format = ca->features.mode == CM_WRITE; | |
2640 | ||
2641 | cache = kzalloc(sizeof(*cache), GFP_KERNEL); | |
2642 | if (!cache) | |
2643 | return -ENOMEM; | |
2644 | ||
2645 | cache->ti = ca->ti; | |
2646 | ti->private = cache; | |
c6b4fcba JT |
2647 | ti->num_flush_bios = 2; |
2648 | ti->flush_supported = true; | |
2649 | ||
2650 | ti->num_discard_bios = 1; | |
2651 | ti->discards_supported = true; | |
2652 | ti->discard_zeroes_data_unsupported = true; | |
2572629a | 2653 | ti->split_discard_bios = false; |
c6b4fcba | 2654 | |
8c5008fa | 2655 | cache->features = ca->features; |
19b0092e | 2656 | ti->per_bio_data_size = get_per_bio_data_size(cache); |
c6b4fcba | 2657 | |
c6b4fcba JT |
2658 | cache->callbacks.congested_fn = cache_is_congested; |
2659 | dm_table_add_target_callbacks(ti->table, &cache->callbacks); | |
2660 | ||
2661 | cache->metadata_dev = ca->metadata_dev; | |
2662 | cache->origin_dev = ca->origin_dev; | |
2663 | cache->cache_dev = ca->cache_dev; | |
2664 | ||
2665 | ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL; | |
2666 | ||
2667 | /* FIXME: factor out this whole section */ | |
2668 | origin_blocks = cache->origin_sectors = ca->origin_sectors; | |
414dd67d | 2669 | origin_blocks = block_div(origin_blocks, ca->block_size); |
c6b4fcba JT |
2670 | cache->origin_blocks = to_oblock(origin_blocks); |
2671 | ||
2672 | cache->sectors_per_block = ca->block_size; | |
2673 | if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) { | |
2674 | r = -EINVAL; | |
2675 | goto bad; | |
2676 | } | |
2677 | ||
2678 | if (ca->block_size & (ca->block_size - 1)) { | |
2679 | dm_block_t cache_size = ca->cache_sectors; | |
2680 | ||
2681 | cache->sectors_per_block_shift = -1; | |
414dd67d | 2682 | cache_size = block_div(cache_size, ca->block_size); |
d1d9220c | 2683 | set_cache_size(cache, to_cblock(cache_size)); |
c6b4fcba JT |
2684 | } else { |
2685 | cache->sectors_per_block_shift = __ffs(ca->block_size); | |
d1d9220c | 2686 | set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift)); |
c6b4fcba JT |
2687 | } |
2688 | ||
2689 | r = create_cache_policy(cache, ca, error); | |
2690 | if (r) | |
2691 | goto bad; | |
2f14f4b5 | 2692 | |
c6b4fcba | 2693 | cache->policy_nr_args = ca->policy_argc; |
2f14f4b5 JT |
2694 | cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD; |
2695 | ||
2696 | r = set_config_values(cache, ca->policy_argc, ca->policy_argv); | |
2697 | if (r) { | |
2698 | *error = "Error setting cache policy's config values"; | |
2699 | goto bad; | |
2700 | } | |
c6b4fcba JT |
2701 | |
2702 | cmd = dm_cache_metadata_open(cache->metadata_dev->bdev, | |
2703 | ca->block_size, may_format, | |
2704 | dm_cache_policy_get_hint_size(cache->policy)); | |
2705 | if (IS_ERR(cmd)) { | |
2706 | *error = "Error creating metadata object"; | |
2707 | r = PTR_ERR(cmd); | |
2708 | goto bad; | |
2709 | } | |
2710 | cache->cmd = cmd; | |
2711 | ||
2ee57d58 JT |
2712 | if (passthrough_mode(&cache->features)) { |
2713 | bool all_clean; | |
2714 | ||
2715 | r = dm_cache_metadata_all_clean(cache->cmd, &all_clean); | |
2716 | if (r) { | |
2717 | *error = "dm_cache_metadata_all_clean() failed"; | |
2718 | goto bad; | |
2719 | } | |
2720 | ||
2721 | if (!all_clean) { | |
2722 | *error = "Cannot enter passthrough mode unless all blocks are clean"; | |
2723 | r = -EINVAL; | |
2724 | goto bad; | |
2725 | } | |
2726 | } | |
2727 | ||
c6b4fcba | 2728 | spin_lock_init(&cache->lock); |
651f5fa2 | 2729 | INIT_LIST_HEAD(&cache->deferred_cells); |
c6b4fcba JT |
2730 | bio_list_init(&cache->deferred_bios); |
2731 | bio_list_init(&cache->deferred_flush_bios); | |
e2e74d61 | 2732 | bio_list_init(&cache->deferred_writethrough_bios); |
c6b4fcba JT |
2733 | INIT_LIST_HEAD(&cache->quiesced_migrations); |
2734 | INIT_LIST_HEAD(&cache->completed_migrations); | |
2735 | INIT_LIST_HEAD(&cache->need_commit_migrations); | |
a59db676 JT |
2736 | atomic_set(&cache->nr_allocated_migrations, 0); |
2737 | atomic_set(&cache->nr_io_migrations, 0); | |
c6b4fcba JT |
2738 | init_waitqueue_head(&cache->migration_wait); |
2739 | ||
66cb1910 | 2740 | init_waitqueue_head(&cache->quiescing_wait); |
238f8363 | 2741 | atomic_set(&cache->quiescing, 0); |
66cb1910 JT |
2742 | atomic_set(&cache->quiescing_ack, 0); |
2743 | ||
fa4d683a | 2744 | r = -ENOMEM; |
44fa816b | 2745 | atomic_set(&cache->nr_dirty, 0); |
c6b4fcba JT |
2746 | cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size)); |
2747 | if (!cache->dirty_bitset) { | |
2748 | *error = "could not allocate dirty bitset"; | |
2749 | goto bad; | |
2750 | } | |
2751 | clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size)); | |
2752 | ||
08b18451 JT |
2753 | cache->discard_block_size = |
2754 | calculate_discard_block_size(cache->sectors_per_block, | |
2755 | cache->origin_sectors); | |
2572629a JT |
2756 | cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors, |
2757 | cache->discard_block_size)); | |
1bad9bc4 | 2758 | cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2759 | if (!cache->discard_bitset) { |
2760 | *error = "could not allocate discard bitset"; | |
2761 | goto bad; | |
2762 | } | |
1bad9bc4 | 2763 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2764 | |
2765 | cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); | |
2766 | if (IS_ERR(cache->copier)) { | |
2767 | *error = "could not create kcopyd client"; | |
2768 | r = PTR_ERR(cache->copier); | |
2769 | goto bad; | |
2770 | } | |
2771 | ||
2772 | cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2773 | if (!cache->wq) { | |
2774 | *error = "could not create workqueue for metadata object"; | |
2775 | goto bad; | |
2776 | } | |
2777 | INIT_WORK(&cache->worker, do_worker); | |
2778 | INIT_DELAYED_WORK(&cache->waker, do_waker); | |
2779 | cache->last_commit_jiffies = jiffies; | |
2780 | ||
a195db2d | 2781 | cache->prison = dm_bio_prison_create(); |
c6b4fcba JT |
2782 | if (!cache->prison) { |
2783 | *error = "could not create bio prison"; | |
2784 | goto bad; | |
2785 | } | |
2786 | ||
2787 | cache->all_io_ds = dm_deferred_set_create(); | |
2788 | if (!cache->all_io_ds) { | |
2789 | *error = "could not create all_io deferred set"; | |
2790 | goto bad; | |
2791 | } | |
2792 | ||
2793 | cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE, | |
2794 | migration_cache); | |
2795 | if (!cache->migration_pool) { | |
2796 | *error = "Error creating cache's migration mempool"; | |
2797 | goto bad; | |
2798 | } | |
2799 | ||
c6b4fcba JT |
2800 | cache->need_tick_bio = true; |
2801 | cache->sized = false; | |
65790ff9 | 2802 | cache->invalidate = false; |
c6b4fcba JT |
2803 | cache->commit_requested = false; |
2804 | cache->loaded_mappings = false; | |
2805 | cache->loaded_discards = false; | |
2806 | ||
2807 | load_stats(cache); | |
2808 | ||
2809 | atomic_set(&cache->stats.demotion, 0); | |
2810 | atomic_set(&cache->stats.promotion, 0); | |
2811 | atomic_set(&cache->stats.copies_avoided, 0); | |
2812 | atomic_set(&cache->stats.cache_cell_clash, 0); | |
2813 | atomic_set(&cache->stats.commit_count, 0); | |
2814 | atomic_set(&cache->stats.discard_count, 0); | |
2815 | ||
65790ff9 JT |
2816 | spin_lock_init(&cache->invalidation_lock); |
2817 | INIT_LIST_HEAD(&cache->invalidation_requests); | |
2818 | ||
066dbaa3 JT |
2819 | iot_init(&cache->origin_tracker); |
2820 | ||
c6b4fcba JT |
2821 | *result = cache; |
2822 | return 0; | |
2823 | ||
2824 | bad: | |
2825 | destroy(cache); | |
2826 | return r; | |
2827 | } | |
2828 | ||
2829 | static int copy_ctr_args(struct cache *cache, int argc, const char **argv) | |
2830 | { | |
2831 | unsigned i; | |
2832 | const char **copy; | |
2833 | ||
2834 | copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL); | |
2835 | if (!copy) | |
2836 | return -ENOMEM; | |
2837 | for (i = 0; i < argc; i++) { | |
2838 | copy[i] = kstrdup(argv[i], GFP_KERNEL); | |
2839 | if (!copy[i]) { | |
2840 | while (i--) | |
2841 | kfree(copy[i]); | |
2842 | kfree(copy); | |
2843 | return -ENOMEM; | |
2844 | } | |
2845 | } | |
2846 | ||
2847 | cache->nr_ctr_args = argc; | |
2848 | cache->ctr_args = copy; | |
2849 | ||
2850 | return 0; | |
2851 | } | |
2852 | ||
2853 | static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2854 | { | |
2855 | int r = -EINVAL; | |
2856 | struct cache_args *ca; | |
2857 | struct cache *cache = NULL; | |
2858 | ||
2859 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
2860 | if (!ca) { | |
2861 | ti->error = "Error allocating memory for cache"; | |
2862 | return -ENOMEM; | |
2863 | } | |
2864 | ca->ti = ti; | |
2865 | ||
2866 | r = parse_cache_args(ca, argc, argv, &ti->error); | |
2867 | if (r) | |
2868 | goto out; | |
2869 | ||
2870 | r = cache_create(ca, &cache); | |
617a0b89 HM |
2871 | if (r) |
2872 | goto out; | |
c6b4fcba JT |
2873 | |
2874 | r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3); | |
2875 | if (r) { | |
2876 | destroy(cache); | |
2877 | goto out; | |
2878 | } | |
2879 | ||
2880 | ti->private = cache; | |
2881 | ||
2882 | out: | |
2883 | destroy_cache_args(ca); | |
2884 | return r; | |
2885 | } | |
2886 | ||
651f5fa2 JT |
2887 | /*----------------------------------------------------------------*/ |
2888 | ||
2889 | static int cache_map(struct dm_target *ti, struct bio *bio) | |
c6b4fcba | 2890 | { |
651f5fa2 JT |
2891 | struct cache *cache = ti->private; |
2892 | ||
c6b4fcba | 2893 | int r; |
651f5fa2 | 2894 | struct dm_bio_prison_cell *cell = NULL; |
c6b4fcba | 2895 | dm_oblock_t block = get_bio_block(cache, bio); |
19b0092e | 2896 | size_t pb_data_size = get_per_bio_data_size(cache); |
c6b4fcba JT |
2897 | bool can_migrate = false; |
2898 | bool discarded_block; | |
c6b4fcba | 2899 | struct policy_result lookup_result; |
e893fba9 | 2900 | struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size); |
fb4100ae JT |
2901 | struct old_oblock_lock ool; |
2902 | ||
2903 | ool.locker.fn = null_locker; | |
c6b4fcba | 2904 | |
e893fba9 | 2905 | if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) { |
c6b4fcba JT |
2906 | /* |
2907 | * This can only occur if the io goes to a partial block at | |
2908 | * the end of the origin device. We don't cache these. | |
2909 | * Just remap to the origin and carry on. | |
2910 | */ | |
e893fba9 | 2911 | remap_to_origin(cache, bio); |
651f5fa2 | 2912 | accounted_begin(cache, bio); |
c6b4fcba JT |
2913 | return DM_MAPIO_REMAPPED; |
2914 | } | |
2915 | ||
651f5fa2 | 2916 | if (discard_or_flush(bio)) { |
c6b4fcba JT |
2917 | defer_bio(cache, bio); |
2918 | return DM_MAPIO_SUBMITTED; | |
2919 | } | |
2920 | ||
2921 | /* | |
2922 | * Check to see if that block is currently migrating. | |
2923 | */ | |
651f5fa2 JT |
2924 | cell = alloc_prison_cell(cache); |
2925 | if (!cell) { | |
c6b4fcba JT |
2926 | defer_bio(cache, bio); |
2927 | return DM_MAPIO_SUBMITTED; | |
2928 | } | |
2929 | ||
651f5fa2 | 2930 | r = bio_detain(cache, block, bio, cell, |
c6b4fcba | 2931 | (cell_free_fn) free_prison_cell, |
651f5fa2 | 2932 | cache, &cell); |
c6b4fcba JT |
2933 | if (r) { |
2934 | if (r < 0) | |
2935 | defer_bio(cache, bio); | |
2936 | ||
2937 | return DM_MAPIO_SUBMITTED; | |
2938 | } | |
2939 | ||
2940 | discarded_block = is_discarded_oblock(cache, block); | |
2941 | ||
2942 | r = policy_map(cache->policy, block, false, can_migrate, discarded_block, | |
fb4100ae | 2943 | bio, &ool.locker, &lookup_result); |
c6b4fcba | 2944 | if (r == -EWOULDBLOCK) { |
651f5fa2 | 2945 | cell_defer(cache, cell, true); |
c6b4fcba JT |
2946 | return DM_MAPIO_SUBMITTED; |
2947 | ||
2948 | } else if (r) { | |
2949 | DMERR_LIMIT("Unexpected return from cache replacement policy: %d", r); | |
651f5fa2 | 2950 | cell_defer(cache, cell, false); |
c6b4fcba JT |
2951 | bio_io_error(bio); |
2952 | return DM_MAPIO_SUBMITTED; | |
2953 | } | |
2954 | ||
2ee57d58 | 2955 | r = DM_MAPIO_REMAPPED; |
c6b4fcba JT |
2956 | switch (lookup_result.op) { |
2957 | case POLICY_HIT: | |
2ee57d58 JT |
2958 | if (passthrough_mode(&cache->features)) { |
2959 | if (bio_data_dir(bio) == WRITE) { | |
2960 | /* | |
2961 | * We need to invalidate this block, so | |
2962 | * defer for the worker thread. | |
2963 | */ | |
651f5fa2 | 2964 | cell_defer(cache, cell, true); |
2ee57d58 JT |
2965 | r = DM_MAPIO_SUBMITTED; |
2966 | ||
2967 | } else { | |
2ee57d58 JT |
2968 | inc_miss_counter(cache, bio); |
2969 | remap_to_origin_clear_discard(cache, bio, block); | |
651f5fa2 JT |
2970 | accounted_begin(cache, bio); |
2971 | inc_ds(cache, bio, cell); | |
2972 | // FIXME: we want to remap hits or misses straight | |
2973 | // away rather than passing over to the worker. | |
2974 | cell_defer(cache, cell, false); | |
2ee57d58 | 2975 | } |
c6b4fcba | 2976 | |
2ee57d58 JT |
2977 | } else { |
2978 | inc_hit_counter(cache, bio); | |
2ee57d58 | 2979 | if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) && |
651f5fa2 | 2980 | !is_dirty(cache, lookup_result.cblock)) { |
2ee57d58 | 2981 | remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock); |
651f5fa2 JT |
2982 | accounted_begin(cache, bio); |
2983 | inc_ds(cache, bio, cell); | |
2984 | cell_defer(cache, cell, false); | |
2985 | ||
2986 | } else | |
2987 | remap_cell_to_cache_dirty(cache, cell, block, lookup_result.cblock, false); | |
2ee57d58 | 2988 | } |
c6b4fcba JT |
2989 | break; |
2990 | ||
2991 | case POLICY_MISS: | |
2992 | inc_miss_counter(cache, bio); | |
c6b4fcba JT |
2993 | if (pb->req_nr != 0) { |
2994 | /* | |
2995 | * This is a duplicate writethrough io that is no | |
2996 | * longer needed because the block has been demoted. | |
2997 | */ | |
2998 | bio_endio(bio, 0); | |
651f5fa2 JT |
2999 | // FIXME: remap everything as a miss |
3000 | cell_defer(cache, cell, false); | |
8c081b52 JT |
3001 | r = DM_MAPIO_SUBMITTED; |
3002 | ||
3003 | } else | |
651f5fa2 | 3004 | remap_cell_to_origin_clear_discard(cache, cell, block, false); |
c6b4fcba JT |
3005 | break; |
3006 | ||
3007 | default: | |
3008 | DMERR_LIMIT("%s: erroring bio: unknown policy op: %u", __func__, | |
3009 | (unsigned) lookup_result.op); | |
651f5fa2 | 3010 | cell_defer(cache, cell, false); |
c6b4fcba | 3011 | bio_io_error(bio); |
2ee57d58 | 3012 | r = DM_MAPIO_SUBMITTED; |
c6b4fcba JT |
3013 | } |
3014 | ||
2ee57d58 | 3015 | return r; |
c6b4fcba JT |
3016 | } |
3017 | ||
3018 | static int cache_end_io(struct dm_target *ti, struct bio *bio, int error) | |
3019 | { | |
3020 | struct cache *cache = ti->private; | |
3021 | unsigned long flags; | |
19b0092e MS |
3022 | size_t pb_data_size = get_per_bio_data_size(cache); |
3023 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
3024 | |
3025 | if (pb->tick) { | |
3026 | policy_tick(cache->policy); | |
3027 | ||
3028 | spin_lock_irqsave(&cache->lock, flags); | |
3029 | cache->need_tick_bio = true; | |
3030 | spin_unlock_irqrestore(&cache->lock, flags); | |
3031 | } | |
3032 | ||
3033 | check_for_quiesced_migrations(cache, pb); | |
066dbaa3 | 3034 | accounted_complete(cache, bio); |
c6b4fcba JT |
3035 | |
3036 | return 0; | |
3037 | } | |
3038 | ||
3039 | static int write_dirty_bitset(struct cache *cache) | |
3040 | { | |
3041 | unsigned i, r; | |
3042 | ||
3043 | for (i = 0; i < from_cblock(cache->cache_size); i++) { | |
3044 | r = dm_cache_set_dirty(cache->cmd, to_cblock(i), | |
3045 | is_dirty(cache, to_cblock(i))); | |
3046 | if (r) | |
3047 | return r; | |
3048 | } | |
3049 | ||
3050 | return 0; | |
3051 | } | |
3052 | ||
3053 | static int write_discard_bitset(struct cache *cache) | |
3054 | { | |
3055 | unsigned i, r; | |
3056 | ||
1bad9bc4 JT |
3057 | r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size, |
3058 | cache->discard_nr_blocks); | |
c6b4fcba JT |
3059 | if (r) { |
3060 | DMERR("could not resize on-disk discard bitset"); | |
3061 | return r; | |
3062 | } | |
3063 | ||
1bad9bc4 JT |
3064 | for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) { |
3065 | r = dm_cache_set_discard(cache->cmd, to_dblock(i), | |
3066 | is_discarded(cache, to_dblock(i))); | |
c6b4fcba JT |
3067 | if (r) |
3068 | return r; | |
3069 | } | |
3070 | ||
3071 | return 0; | |
3072 | } | |
3073 | ||
c6b4fcba JT |
3074 | /* |
3075 | * returns true on success | |
3076 | */ | |
3077 | static bool sync_metadata(struct cache *cache) | |
3078 | { | |
3079 | int r1, r2, r3, r4; | |
3080 | ||
3081 | r1 = write_dirty_bitset(cache); | |
3082 | if (r1) | |
3083 | DMERR("could not write dirty bitset"); | |
3084 | ||
3085 | r2 = write_discard_bitset(cache); | |
3086 | if (r2) | |
3087 | DMERR("could not write discard bitset"); | |
3088 | ||
3089 | save_stats(cache); | |
3090 | ||
0596661f | 3091 | r3 = dm_cache_write_hints(cache->cmd, cache->policy); |
c6b4fcba JT |
3092 | if (r3) |
3093 | DMERR("could not write hints"); | |
3094 | ||
3095 | /* | |
3096 | * If writing the above metadata failed, we still commit, but don't | |
3097 | * set the clean shutdown flag. This will effectively force every | |
3098 | * dirty bit to be set on reload. | |
3099 | */ | |
3100 | r4 = dm_cache_commit(cache->cmd, !r1 && !r2 && !r3); | |
3101 | if (r4) | |
3102 | DMERR("could not write cache metadata. Data loss may occur."); | |
3103 | ||
3104 | return !r1 && !r2 && !r3 && !r4; | |
3105 | } | |
3106 | ||
3107 | static void cache_postsuspend(struct dm_target *ti) | |
3108 | { | |
3109 | struct cache *cache = ti->private; | |
3110 | ||
3111 | start_quiescing(cache); | |
3112 | wait_for_migrations(cache); | |
3113 | stop_worker(cache); | |
651f5fa2 JT |
3114 | requeue_deferred_bios(cache); |
3115 | requeue_deferred_cells(cache); | |
c6b4fcba JT |
3116 | stop_quiescing(cache); |
3117 | ||
3118 | (void) sync_metadata(cache); | |
3119 | } | |
3120 | ||
3121 | static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock, | |
3122 | bool dirty, uint32_t hint, bool hint_valid) | |
3123 | { | |
3124 | int r; | |
3125 | struct cache *cache = context; | |
3126 | ||
3127 | r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid); | |
3128 | if (r) | |
3129 | return r; | |
3130 | ||
3131 | if (dirty) | |
3132 | set_dirty(cache, oblock, cblock); | |
3133 | else | |
3134 | clear_dirty(cache, oblock, cblock); | |
3135 | ||
3136 | return 0; | |
3137 | } | |
3138 | ||
3e2e1c30 JT |
3139 | /* |
3140 | * The discard block size in the on disk metadata is not | |
3141 | * neccessarily the same as we're currently using. So we have to | |
3142 | * be careful to only set the discarded attribute if we know it | |
3143 | * covers a complete block of the new size. | |
3144 | */ | |
3145 | struct discard_load_info { | |
3146 | struct cache *cache; | |
3147 | ||
3148 | /* | |
3149 | * These blocks are sized using the on disk dblock size, rather | |
3150 | * than the current one. | |
3151 | */ | |
3152 | dm_block_t block_size; | |
3153 | dm_block_t discard_begin, discard_end; | |
3154 | }; | |
3155 | ||
3156 | static void discard_load_info_init(struct cache *cache, | |
3157 | struct discard_load_info *li) | |
3158 | { | |
3159 | li->cache = cache; | |
3160 | li->discard_begin = li->discard_end = 0; | |
3161 | } | |
3162 | ||
3163 | static void set_discard_range(struct discard_load_info *li) | |
3164 | { | |
3165 | sector_t b, e; | |
3166 | ||
3167 | if (li->discard_begin == li->discard_end) | |
3168 | return; | |
3169 | ||
3170 | /* | |
3171 | * Convert to sectors. | |
3172 | */ | |
3173 | b = li->discard_begin * li->block_size; | |
3174 | e = li->discard_end * li->block_size; | |
3175 | ||
3176 | /* | |
3177 | * Then convert back to the current dblock size. | |
3178 | */ | |
3179 | b = dm_sector_div_up(b, li->cache->discard_block_size); | |
3180 | sector_div(e, li->cache->discard_block_size); | |
3181 | ||
3182 | /* | |
3183 | * The origin may have shrunk, so we need to check we're still in | |
3184 | * bounds. | |
3185 | */ | |
3186 | if (e > from_dblock(li->cache->discard_nr_blocks)) | |
3187 | e = from_dblock(li->cache->discard_nr_blocks); | |
3188 | ||
3189 | for (; b < e; b++) | |
3190 | set_discard(li->cache, to_dblock(b)); | |
3191 | } | |
3192 | ||
c6b4fcba | 3193 | static int load_discard(void *context, sector_t discard_block_size, |
1bad9bc4 | 3194 | dm_dblock_t dblock, bool discard) |
c6b4fcba | 3195 | { |
3e2e1c30 | 3196 | struct discard_load_info *li = context; |
c6b4fcba | 3197 | |
3e2e1c30 | 3198 | li->block_size = discard_block_size; |
1bad9bc4 | 3199 | |
3e2e1c30 JT |
3200 | if (discard) { |
3201 | if (from_dblock(dblock) == li->discard_end) | |
3202 | /* | |
3203 | * We're already in a discard range, just extend it. | |
3204 | */ | |
3205 | li->discard_end = li->discard_end + 1ULL; | |
3206 | ||
3207 | else { | |
3208 | /* | |
3209 | * Emit the old range and start a new one. | |
3210 | */ | |
3211 | set_discard_range(li); | |
3212 | li->discard_begin = from_dblock(dblock); | |
3213 | li->discard_end = li->discard_begin + 1ULL; | |
3214 | } | |
3215 | } else { | |
3216 | set_discard_range(li); | |
3217 | li->discard_begin = li->discard_end = 0; | |
3218 | } | |
c6b4fcba JT |
3219 | |
3220 | return 0; | |
3221 | } | |
3222 | ||
f494a9c6 JT |
3223 | static dm_cblock_t get_cache_dev_size(struct cache *cache) |
3224 | { | |
3225 | sector_t size = get_dev_size(cache->cache_dev); | |
3226 | (void) sector_div(size, cache->sectors_per_block); | |
3227 | return to_cblock(size); | |
3228 | } | |
3229 | ||
3230 | static bool can_resize(struct cache *cache, dm_cblock_t new_size) | |
3231 | { | |
3232 | if (from_cblock(new_size) > from_cblock(cache->cache_size)) | |
3233 | return true; | |
3234 | ||
3235 | /* | |
3236 | * We can't drop a dirty block when shrinking the cache. | |
3237 | */ | |
3238 | while (from_cblock(new_size) < from_cblock(cache->cache_size)) { | |
3239 | new_size = to_cblock(from_cblock(new_size) + 1); | |
3240 | if (is_dirty(cache, new_size)) { | |
3241 | DMERR("unable to shrink cache; cache block %llu is dirty", | |
3242 | (unsigned long long) from_cblock(new_size)); | |
3243 | return false; | |
3244 | } | |
3245 | } | |
3246 | ||
3247 | return true; | |
3248 | } | |
3249 | ||
3250 | static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size) | |
3251 | { | |
3252 | int r; | |
3253 | ||
08844800 | 3254 | r = dm_cache_resize(cache->cmd, new_size); |
f494a9c6 JT |
3255 | if (r) { |
3256 | DMERR("could not resize cache metadata"); | |
3257 | return r; | |
3258 | } | |
3259 | ||
d1d9220c | 3260 | set_cache_size(cache, new_size); |
f494a9c6 JT |
3261 | |
3262 | return 0; | |
3263 | } | |
3264 | ||
c6b4fcba JT |
3265 | static int cache_preresume(struct dm_target *ti) |
3266 | { | |
3267 | int r = 0; | |
3268 | struct cache *cache = ti->private; | |
f494a9c6 | 3269 | dm_cblock_t csize = get_cache_dev_size(cache); |
c6b4fcba JT |
3270 | |
3271 | /* | |
3272 | * Check to see if the cache has resized. | |
3273 | */ | |
f494a9c6 JT |
3274 | if (!cache->sized) { |
3275 | r = resize_cache_dev(cache, csize); | |
3276 | if (r) | |
c6b4fcba | 3277 | return r; |
c6b4fcba JT |
3278 | |
3279 | cache->sized = true; | |
f494a9c6 JT |
3280 | |
3281 | } else if (csize != cache->cache_size) { | |
3282 | if (!can_resize(cache, csize)) | |
3283 | return -EINVAL; | |
3284 | ||
3285 | r = resize_cache_dev(cache, csize); | |
3286 | if (r) | |
3287 | return r; | |
c6b4fcba JT |
3288 | } |
3289 | ||
3290 | if (!cache->loaded_mappings) { | |
ea2dd8c1 | 3291 | r = dm_cache_load_mappings(cache->cmd, cache->policy, |
c6b4fcba JT |
3292 | load_mapping, cache); |
3293 | if (r) { | |
3294 | DMERR("could not load cache mappings"); | |
3295 | return r; | |
3296 | } | |
3297 | ||
3298 | cache->loaded_mappings = true; | |
3299 | } | |
3300 | ||
3301 | if (!cache->loaded_discards) { | |
3e2e1c30 JT |
3302 | struct discard_load_info li; |
3303 | ||
3304 | /* | |
3305 | * The discard bitset could have been resized, or the | |
3306 | * discard block size changed. To be safe we start by | |
3307 | * setting every dblock to not discarded. | |
3308 | */ | |
3309 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); | |
3310 | ||
3311 | discard_load_info_init(cache, &li); | |
3312 | r = dm_cache_load_discards(cache->cmd, load_discard, &li); | |
c6b4fcba JT |
3313 | if (r) { |
3314 | DMERR("could not load origin discards"); | |
3315 | return r; | |
3316 | } | |
3e2e1c30 | 3317 | set_discard_range(&li); |
c6b4fcba JT |
3318 | |
3319 | cache->loaded_discards = true; | |
3320 | } | |
3321 | ||
3322 | return r; | |
3323 | } | |
3324 | ||
3325 | static void cache_resume(struct dm_target *ti) | |
3326 | { | |
3327 | struct cache *cache = ti->private; | |
3328 | ||
3329 | cache->need_tick_bio = true; | |
3330 | do_waker(&cache->waker.work); | |
3331 | } | |
3332 | ||
3333 | /* | |
3334 | * Status format: | |
3335 | * | |
6a388618 MS |
3336 | * <metadata block size> <#used metadata blocks>/<#total metadata blocks> |
3337 | * <cache block size> <#used cache blocks>/<#total cache blocks> | |
c6b4fcba | 3338 | * <#read hits> <#read misses> <#write hits> <#write misses> |
6a388618 | 3339 | * <#demotions> <#promotions> <#dirty> |
c6b4fcba JT |
3340 | * <#features> <features>* |
3341 | * <#core args> <core args> | |
2e68c4e6 | 3342 | * <policy name> <#policy args> <policy args>* |
c6b4fcba JT |
3343 | */ |
3344 | static void cache_status(struct dm_target *ti, status_type_t type, | |
3345 | unsigned status_flags, char *result, unsigned maxlen) | |
3346 | { | |
3347 | int r = 0; | |
3348 | unsigned i; | |
3349 | ssize_t sz = 0; | |
3350 | dm_block_t nr_free_blocks_metadata = 0; | |
3351 | dm_block_t nr_blocks_metadata = 0; | |
3352 | char buf[BDEVNAME_SIZE]; | |
3353 | struct cache *cache = ti->private; | |
3354 | dm_cblock_t residency; | |
3355 | ||
3356 | switch (type) { | |
3357 | case STATUSTYPE_INFO: | |
3358 | /* Commit to ensure statistics aren't out-of-date */ | |
3359 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) { | |
3360 | r = dm_cache_commit(cache->cmd, false); | |
3361 | if (r) | |
3362 | DMERR("could not commit metadata for accurate status"); | |
3363 | } | |
3364 | ||
3365 | r = dm_cache_get_free_metadata_block_count(cache->cmd, | |
3366 | &nr_free_blocks_metadata); | |
3367 | if (r) { | |
3368 | DMERR("could not get metadata free block count"); | |
3369 | goto err; | |
3370 | } | |
3371 | ||
3372 | r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata); | |
3373 | if (r) { | |
3374 | DMERR("could not get metadata device size"); | |
3375 | goto err; | |
3376 | } | |
3377 | ||
3378 | residency = policy_residency(cache->policy); | |
3379 | ||
44fa816b | 3380 | DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ", |
895b47d7 | 3381 | (unsigned)DM_CACHE_METADATA_BLOCK_SIZE, |
c6b4fcba JT |
3382 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
3383 | (unsigned long long)nr_blocks_metadata, | |
6a388618 MS |
3384 | cache->sectors_per_block, |
3385 | (unsigned long long) from_cblock(residency), | |
3386 | (unsigned long long) from_cblock(cache->cache_size), | |
c6b4fcba JT |
3387 | (unsigned) atomic_read(&cache->stats.read_hit), |
3388 | (unsigned) atomic_read(&cache->stats.read_miss), | |
3389 | (unsigned) atomic_read(&cache->stats.write_hit), | |
3390 | (unsigned) atomic_read(&cache->stats.write_miss), | |
3391 | (unsigned) atomic_read(&cache->stats.demotion), | |
3392 | (unsigned) atomic_read(&cache->stats.promotion), | |
44fa816b | 3393 | (unsigned long) atomic_read(&cache->nr_dirty)); |
c6b4fcba | 3394 | |
2ee57d58 | 3395 | if (writethrough_mode(&cache->features)) |
c6b4fcba | 3396 | DMEMIT("1 writethrough "); |
2ee57d58 JT |
3397 | |
3398 | else if (passthrough_mode(&cache->features)) | |
3399 | DMEMIT("1 passthrough "); | |
3400 | ||
3401 | else if (writeback_mode(&cache->features)) | |
3402 | DMEMIT("1 writeback "); | |
3403 | ||
3404 | else { | |
3405 | DMERR("internal error: unknown io mode: %d", (int) cache->features.io_mode); | |
3406 | goto err; | |
3407 | } | |
c6b4fcba JT |
3408 | |
3409 | DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold); | |
2e68c4e6 MS |
3410 | |
3411 | DMEMIT("%s ", dm_cache_policy_get_name(cache->policy)); | |
c6b4fcba JT |
3412 | if (sz < maxlen) { |
3413 | r = policy_emit_config_values(cache->policy, result + sz, maxlen - sz); | |
3414 | if (r) | |
3415 | DMERR("policy_emit_config_values returned %d", r); | |
3416 | } | |
3417 | ||
3418 | break; | |
3419 | ||
3420 | case STATUSTYPE_TABLE: | |
3421 | format_dev_t(buf, cache->metadata_dev->bdev->bd_dev); | |
3422 | DMEMIT("%s ", buf); | |
3423 | format_dev_t(buf, cache->cache_dev->bdev->bd_dev); | |
3424 | DMEMIT("%s ", buf); | |
3425 | format_dev_t(buf, cache->origin_dev->bdev->bd_dev); | |
3426 | DMEMIT("%s", buf); | |
3427 | ||
3428 | for (i = 0; i < cache->nr_ctr_args - 1; i++) | |
3429 | DMEMIT(" %s", cache->ctr_args[i]); | |
3430 | if (cache->nr_ctr_args) | |
3431 | DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]); | |
3432 | } | |
3433 | ||
3434 | return; | |
3435 | ||
3436 | err: | |
3437 | DMEMIT("Error"); | |
3438 | } | |
3439 | ||
c6b4fcba | 3440 | /* |
65790ff9 JT |
3441 | * A cache block range can take two forms: |
3442 | * | |
3443 | * i) A single cblock, eg. '3456' | |
3444 | * ii) A begin and end cblock with dots between, eg. 123-234 | |
3445 | */ | |
3446 | static int parse_cblock_range(struct cache *cache, const char *str, | |
3447 | struct cblock_range *result) | |
3448 | { | |
3449 | char dummy; | |
3450 | uint64_t b, e; | |
3451 | int r; | |
3452 | ||
3453 | /* | |
3454 | * Try and parse form (ii) first. | |
3455 | */ | |
3456 | r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy); | |
3457 | if (r < 0) | |
3458 | return r; | |
3459 | ||
3460 | if (r == 2) { | |
3461 | result->begin = to_cblock(b); | |
3462 | result->end = to_cblock(e); | |
3463 | return 0; | |
3464 | } | |
3465 | ||
3466 | /* | |
3467 | * That didn't work, try form (i). | |
3468 | */ | |
3469 | r = sscanf(str, "%llu%c", &b, &dummy); | |
3470 | if (r < 0) | |
3471 | return r; | |
3472 | ||
3473 | if (r == 1) { | |
3474 | result->begin = to_cblock(b); | |
3475 | result->end = to_cblock(from_cblock(result->begin) + 1u); | |
3476 | return 0; | |
3477 | } | |
3478 | ||
3479 | DMERR("invalid cblock range '%s'", str); | |
3480 | return -EINVAL; | |
3481 | } | |
3482 | ||
3483 | static int validate_cblock_range(struct cache *cache, struct cblock_range *range) | |
3484 | { | |
3485 | uint64_t b = from_cblock(range->begin); | |
3486 | uint64_t e = from_cblock(range->end); | |
3487 | uint64_t n = from_cblock(cache->cache_size); | |
3488 | ||
3489 | if (b >= n) { | |
3490 | DMERR("begin cblock out of range: %llu >= %llu", b, n); | |
3491 | return -EINVAL; | |
3492 | } | |
3493 | ||
3494 | if (e > n) { | |
3495 | DMERR("end cblock out of range: %llu > %llu", e, n); | |
3496 | return -EINVAL; | |
3497 | } | |
3498 | ||
3499 | if (b >= e) { | |
3500 | DMERR("invalid cblock range: %llu >= %llu", b, e); | |
3501 | return -EINVAL; | |
3502 | } | |
3503 | ||
3504 | return 0; | |
3505 | } | |
3506 | ||
3507 | static int request_invalidation(struct cache *cache, struct cblock_range *range) | |
3508 | { | |
3509 | struct invalidation_request req; | |
3510 | ||
3511 | INIT_LIST_HEAD(&req.list); | |
3512 | req.cblocks = range; | |
3513 | atomic_set(&req.complete, 0); | |
3514 | req.err = 0; | |
3515 | init_waitqueue_head(&req.result_wait); | |
3516 | ||
3517 | spin_lock(&cache->invalidation_lock); | |
3518 | list_add(&req.list, &cache->invalidation_requests); | |
3519 | spin_unlock(&cache->invalidation_lock); | |
3520 | wake_worker(cache); | |
3521 | ||
3522 | wait_event(req.result_wait, atomic_read(&req.complete)); | |
3523 | return req.err; | |
3524 | } | |
3525 | ||
3526 | static int process_invalidate_cblocks_message(struct cache *cache, unsigned count, | |
3527 | const char **cblock_ranges) | |
3528 | { | |
3529 | int r = 0; | |
3530 | unsigned i; | |
3531 | struct cblock_range range; | |
3532 | ||
3533 | if (!passthrough_mode(&cache->features)) { | |
3534 | DMERR("cache has to be in passthrough mode for invalidation"); | |
3535 | return -EPERM; | |
3536 | } | |
3537 | ||
3538 | for (i = 0; i < count; i++) { | |
3539 | r = parse_cblock_range(cache, cblock_ranges[i], &range); | |
3540 | if (r) | |
3541 | break; | |
3542 | ||
3543 | r = validate_cblock_range(cache, &range); | |
3544 | if (r) | |
3545 | break; | |
3546 | ||
3547 | /* | |
3548 | * Pass begin and end origin blocks to the worker and wake it. | |
3549 | */ | |
3550 | r = request_invalidation(cache, &range); | |
3551 | if (r) | |
3552 | break; | |
3553 | } | |
3554 | ||
3555 | return r; | |
3556 | } | |
3557 | ||
3558 | /* | |
3559 | * Supports | |
3560 | * "<key> <value>" | |
3561 | * and | |
3562 | * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]* | |
c6b4fcba JT |
3563 | * |
3564 | * The key migration_threshold is supported by the cache target core. | |
3565 | */ | |
3566 | static int cache_message(struct dm_target *ti, unsigned argc, char **argv) | |
3567 | { | |
c6b4fcba JT |
3568 | struct cache *cache = ti->private; |
3569 | ||
65790ff9 JT |
3570 | if (!argc) |
3571 | return -EINVAL; | |
3572 | ||
7b6b2bc9 | 3573 | if (!strcasecmp(argv[0], "invalidate_cblocks")) |
65790ff9 JT |
3574 | return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1); |
3575 | ||
c6b4fcba JT |
3576 | if (argc != 2) |
3577 | return -EINVAL; | |
3578 | ||
2f14f4b5 | 3579 | return set_config_value(cache, argv[0], argv[1]); |
c6b4fcba JT |
3580 | } |
3581 | ||
3582 | static int cache_iterate_devices(struct dm_target *ti, | |
3583 | iterate_devices_callout_fn fn, void *data) | |
3584 | { | |
3585 | int r = 0; | |
3586 | struct cache *cache = ti->private; | |
3587 | ||
3588 | r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data); | |
3589 | if (!r) | |
3590 | r = fn(ti, cache->origin_dev, 0, ti->len, data); | |
3591 | ||
3592 | return r; | |
3593 | } | |
3594 | ||
3595 | /* | |
3596 | * We assume I/O is going to the origin (which is the volume | |
3597 | * more likely to have restrictions e.g. by being striped). | |
3598 | * (Looking up the exact location of the data would be expensive | |
3599 | * and could always be out of date by the time the bio is submitted.) | |
3600 | */ | |
3601 | static int cache_bvec_merge(struct dm_target *ti, | |
3602 | struct bvec_merge_data *bvm, | |
3603 | struct bio_vec *biovec, int max_size) | |
3604 | { | |
3605 | struct cache *cache = ti->private; | |
3606 | struct request_queue *q = bdev_get_queue(cache->origin_dev->bdev); | |
3607 | ||
3608 | if (!q->merge_bvec_fn) | |
3609 | return max_size; | |
3610 | ||
3611 | bvm->bi_bdev = cache->origin_dev->bdev; | |
3612 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3613 | } | |
3614 | ||
3615 | static void set_discard_limits(struct cache *cache, struct queue_limits *limits) | |
3616 | { | |
3617 | /* | |
3618 | * FIXME: these limits may be incompatible with the cache device | |
3619 | */ | |
7ae34e77 JT |
3620 | limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024, |
3621 | cache->origin_sectors); | |
1bad9bc4 | 3622 | limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT; |
c6b4fcba JT |
3623 | } |
3624 | ||
3625 | static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3626 | { | |
3627 | struct cache *cache = ti->private; | |
f6109372 | 3628 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
c6b4fcba | 3629 | |
f6109372 MS |
3630 | /* |
3631 | * If the system-determined stacked limits are compatible with the | |
3632 | * cache's blocksize (io_opt is a factor) do not override them. | |
3633 | */ | |
3634 | if (io_opt_sectors < cache->sectors_per_block || | |
3635 | do_div(io_opt_sectors, cache->sectors_per_block)) { | |
b0246530 | 3636 | blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT); |
f6109372 MS |
3637 | blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT); |
3638 | } | |
c6b4fcba JT |
3639 | set_discard_limits(cache, limits); |
3640 | } | |
3641 | ||
3642 | /*----------------------------------------------------------------*/ | |
3643 | ||
3644 | static struct target_type cache_target = { | |
3645 | .name = "cache", | |
7ae34e77 | 3646 | .version = {1, 6, 0}, |
c6b4fcba JT |
3647 | .module = THIS_MODULE, |
3648 | .ctr = cache_ctr, | |
3649 | .dtr = cache_dtr, | |
3650 | .map = cache_map, | |
3651 | .end_io = cache_end_io, | |
3652 | .postsuspend = cache_postsuspend, | |
3653 | .preresume = cache_preresume, | |
3654 | .resume = cache_resume, | |
3655 | .status = cache_status, | |
3656 | .message = cache_message, | |
3657 | .iterate_devices = cache_iterate_devices, | |
3658 | .merge = cache_bvec_merge, | |
3659 | .io_hints = cache_io_hints, | |
3660 | }; | |
3661 | ||
3662 | static int __init dm_cache_init(void) | |
3663 | { | |
3664 | int r; | |
3665 | ||
3666 | r = dm_register_target(&cache_target); | |
3667 | if (r) { | |
3668 | DMERR("cache target registration failed: %d", r); | |
3669 | return r; | |
3670 | } | |
3671 | ||
3672 | migration_cache = KMEM_CACHE(dm_cache_migration, 0); | |
3673 | if (!migration_cache) { | |
3674 | dm_unregister_target(&cache_target); | |
3675 | return -ENOMEM; | |
3676 | } | |
3677 | ||
3678 | return 0; | |
3679 | } | |
3680 | ||
3681 | static void __exit dm_cache_exit(void) | |
3682 | { | |
3683 | dm_unregister_target(&cache_target); | |
3684 | kmem_cache_destroy(migration_cache); | |
3685 | } | |
3686 | ||
3687 | module_init(dm_cache_init); | |
3688 | module_exit(dm_cache_exit); | |
3689 | ||
3690 | MODULE_DESCRIPTION(DM_NAME " cache target"); | |
3691 | MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>"); | |
3692 | MODULE_LICENSE("GPL"); |