]>
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 | ||
28 | /* | |
29 | * Glossary: | |
30 | * | |
31 | * oblock: index of an origin block | |
32 | * cblock: index of a cache block | |
33 | * promotion: movement of a block from origin to cache | |
34 | * demotion: movement of a block from cache to origin | |
35 | * migration: movement of a block between the origin and cache device, | |
36 | * either direction | |
37 | */ | |
38 | ||
39 | /*----------------------------------------------------------------*/ | |
40 | ||
41 | static size_t bitset_size_in_bytes(unsigned nr_entries) | |
42 | { | |
43 | return sizeof(unsigned long) * dm_div_up(nr_entries, BITS_PER_LONG); | |
44 | } | |
45 | ||
46 | static unsigned long *alloc_bitset(unsigned nr_entries) | |
47 | { | |
48 | size_t s = bitset_size_in_bytes(nr_entries); | |
49 | return vzalloc(s); | |
50 | } | |
51 | ||
52 | static void clear_bitset(void *bitset, unsigned nr_entries) | |
53 | { | |
54 | size_t s = bitset_size_in_bytes(nr_entries); | |
55 | memset(bitset, 0, s); | |
56 | } | |
57 | ||
58 | static void free_bitset(unsigned long *bits) | |
59 | { | |
60 | vfree(bits); | |
61 | } | |
62 | ||
63 | /*----------------------------------------------------------------*/ | |
64 | ||
c9d28d5d JT |
65 | /* |
66 | * There are a couple of places where we let a bio run, but want to do some | |
67 | * work before calling its endio function. We do this by temporarily | |
68 | * changing the endio fn. | |
69 | */ | |
70 | struct dm_hook_info { | |
71 | bio_end_io_t *bi_end_io; | |
72 | void *bi_private; | |
73 | }; | |
74 | ||
75 | static void dm_hook_bio(struct dm_hook_info *h, struct bio *bio, | |
76 | bio_end_io_t *bi_end_io, void *bi_private) | |
77 | { | |
78 | h->bi_end_io = bio->bi_end_io; | |
79 | h->bi_private = bio->bi_private; | |
80 | ||
81 | bio->bi_end_io = bi_end_io; | |
82 | bio->bi_private = bi_private; | |
83 | } | |
84 | ||
85 | static void dm_unhook_bio(struct dm_hook_info *h, struct bio *bio) | |
86 | { | |
87 | bio->bi_end_io = h->bi_end_io; | |
88 | bio->bi_private = h->bi_private; | |
8d307269 MS |
89 | |
90 | /* | |
91 | * Must bump bi_remaining to allow bio to complete with | |
92 | * restored bi_end_io. | |
93 | */ | |
94 | atomic_inc(&bio->bi_remaining); | |
c9d28d5d JT |
95 | } |
96 | ||
97 | /*----------------------------------------------------------------*/ | |
98 | ||
c6b4fcba JT |
99 | #define MIGRATION_POOL_SIZE 128 |
100 | #define COMMIT_PERIOD HZ | |
101 | #define MIGRATION_COUNT_WINDOW 10 | |
102 | ||
103 | /* | |
05473044 MS |
104 | * The block size of the device holding cache data must be |
105 | * between 32KB and 1GB. | |
c6b4fcba JT |
106 | */ |
107 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT) | |
05473044 | 108 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
c6b4fcba JT |
109 | |
110 | /* | |
111 | * FIXME: the cache is read/write for the time being. | |
112 | */ | |
2ee57d58 | 113 | enum cache_metadata_mode { |
c6b4fcba JT |
114 | CM_WRITE, /* metadata may be changed */ |
115 | CM_READ_ONLY, /* metadata may not be changed */ | |
116 | }; | |
117 | ||
2ee57d58 JT |
118 | enum cache_io_mode { |
119 | /* | |
120 | * Data is written to cached blocks only. These blocks are marked | |
121 | * dirty. If you lose the cache device you will lose data. | |
122 | * Potential performance increase for both reads and writes. | |
123 | */ | |
124 | CM_IO_WRITEBACK, | |
125 | ||
126 | /* | |
127 | * Data is written to both cache and origin. Blocks are never | |
128 | * dirty. Potential performance benfit for reads only. | |
129 | */ | |
130 | CM_IO_WRITETHROUGH, | |
131 | ||
132 | /* | |
133 | * A degraded mode useful for various cache coherency situations | |
134 | * (eg, rolling back snapshots). Reads and writes always go to the | |
135 | * origin. If a write goes to a cached oblock, then the cache | |
136 | * block is invalidated. | |
137 | */ | |
138 | CM_IO_PASSTHROUGH | |
139 | }; | |
140 | ||
c6b4fcba | 141 | struct cache_features { |
2ee57d58 JT |
142 | enum cache_metadata_mode mode; |
143 | enum cache_io_mode io_mode; | |
c6b4fcba JT |
144 | }; |
145 | ||
146 | struct cache_stats { | |
147 | atomic_t read_hit; | |
148 | atomic_t read_miss; | |
149 | atomic_t write_hit; | |
150 | atomic_t write_miss; | |
151 | atomic_t demotion; | |
152 | atomic_t promotion; | |
153 | atomic_t copies_avoided; | |
154 | atomic_t cache_cell_clash; | |
155 | atomic_t commit_count; | |
156 | atomic_t discard_count; | |
157 | }; | |
158 | ||
65790ff9 JT |
159 | /* |
160 | * Defines a range of cblocks, begin to (end - 1) are in the range. end is | |
161 | * the one-past-the-end value. | |
162 | */ | |
163 | struct cblock_range { | |
164 | dm_cblock_t begin; | |
165 | dm_cblock_t end; | |
166 | }; | |
167 | ||
168 | struct invalidation_request { | |
169 | struct list_head list; | |
170 | struct cblock_range *cblocks; | |
171 | ||
172 | atomic_t complete; | |
173 | int err; | |
174 | ||
175 | wait_queue_head_t result_wait; | |
176 | }; | |
177 | ||
c6b4fcba JT |
178 | struct cache { |
179 | struct dm_target *ti; | |
180 | struct dm_target_callbacks callbacks; | |
181 | ||
c9ec5d7c MS |
182 | struct dm_cache_metadata *cmd; |
183 | ||
c6b4fcba JT |
184 | /* |
185 | * Metadata is written to this device. | |
186 | */ | |
187 | struct dm_dev *metadata_dev; | |
188 | ||
189 | /* | |
190 | * The slower of the two data devices. Typically a spindle. | |
191 | */ | |
192 | struct dm_dev *origin_dev; | |
193 | ||
194 | /* | |
195 | * The faster of the two data devices. Typically an SSD. | |
196 | */ | |
197 | struct dm_dev *cache_dev; | |
198 | ||
c6b4fcba JT |
199 | /* |
200 | * Size of the origin device in _complete_ blocks and native sectors. | |
201 | */ | |
202 | dm_oblock_t origin_blocks; | |
203 | sector_t origin_sectors; | |
204 | ||
205 | /* | |
206 | * Size of the cache device in blocks. | |
207 | */ | |
208 | dm_cblock_t cache_size; | |
209 | ||
210 | /* | |
211 | * Fields for converting from sectors to blocks. | |
212 | */ | |
213 | uint32_t sectors_per_block; | |
214 | int sectors_per_block_shift; | |
215 | ||
c6b4fcba JT |
216 | spinlock_t lock; |
217 | struct bio_list deferred_bios; | |
218 | struct bio_list deferred_flush_bios; | |
e2e74d61 | 219 | struct bio_list deferred_writethrough_bios; |
c6b4fcba JT |
220 | struct list_head quiesced_migrations; |
221 | struct list_head completed_migrations; | |
222 | struct list_head need_commit_migrations; | |
223 | sector_t migration_threshold; | |
c6b4fcba | 224 | wait_queue_head_t migration_wait; |
c9ec5d7c | 225 | atomic_t nr_migrations; |
c6b4fcba | 226 | |
66cb1910 | 227 | wait_queue_head_t quiescing_wait; |
238f8363 | 228 | atomic_t quiescing; |
66cb1910 JT |
229 | atomic_t quiescing_ack; |
230 | ||
c6b4fcba JT |
231 | /* |
232 | * cache_size entries, dirty if set | |
233 | */ | |
44fa816b | 234 | atomic_t nr_dirty; |
c6b4fcba JT |
235 | unsigned long *dirty_bitset; |
236 | ||
237 | /* | |
238 | * origin_blocks entries, discarded if set. | |
239 | */ | |
1bad9bc4 | 240 | dm_dblock_t discard_nr_blocks; |
c6b4fcba | 241 | unsigned long *discard_bitset; |
08b18451 | 242 | uint32_t discard_block_size; /* a power of 2 times sectors per block */ |
c9ec5d7c MS |
243 | |
244 | /* | |
245 | * Rather than reconstructing the table line for the status we just | |
246 | * save it and regurgitate. | |
247 | */ | |
248 | unsigned nr_ctr_args; | |
249 | const char **ctr_args; | |
c6b4fcba JT |
250 | |
251 | struct dm_kcopyd_client *copier; | |
252 | struct workqueue_struct *wq; | |
253 | struct work_struct worker; | |
254 | ||
255 | struct delayed_work waker; | |
256 | unsigned long last_commit_jiffies; | |
257 | ||
258 | struct dm_bio_prison *prison; | |
259 | struct dm_deferred_set *all_io_ds; | |
260 | ||
261 | mempool_t *migration_pool; | |
262 | struct dm_cache_migration *next_migration; | |
263 | ||
264 | struct dm_cache_policy *policy; | |
265 | unsigned policy_nr_args; | |
266 | ||
267 | bool need_tick_bio:1; | |
268 | bool sized:1; | |
65790ff9 | 269 | bool invalidate:1; |
c6b4fcba JT |
270 | bool commit_requested:1; |
271 | bool loaded_mappings:1; | |
272 | bool loaded_discards:1; | |
273 | ||
c6b4fcba | 274 | /* |
c9ec5d7c | 275 | * Cache features such as write-through. |
c6b4fcba | 276 | */ |
c9ec5d7c MS |
277 | struct cache_features features; |
278 | ||
279 | struct cache_stats stats; | |
65790ff9 JT |
280 | |
281 | /* | |
282 | * Invalidation fields. | |
283 | */ | |
284 | spinlock_t invalidation_lock; | |
285 | struct list_head invalidation_requests; | |
c6b4fcba JT |
286 | }; |
287 | ||
288 | struct per_bio_data { | |
289 | bool tick:1; | |
290 | unsigned req_nr:2; | |
291 | struct dm_deferred_entry *all_io_entry; | |
c6eda5e8 | 292 | struct dm_hook_info hook_info; |
e2e74d61 | 293 | |
19b0092e MS |
294 | /* |
295 | * writethrough fields. These MUST remain at the end of this | |
296 | * structure and the 'cache' member must be the first as it | |
aeed1420 | 297 | * is used to determine the offset of the writethrough fields. |
19b0092e | 298 | */ |
e2e74d61 JT |
299 | struct cache *cache; |
300 | dm_cblock_t cblock; | |
b844fe69 | 301 | struct dm_bio_details bio_details; |
c6b4fcba JT |
302 | }; |
303 | ||
304 | struct dm_cache_migration { | |
305 | struct list_head list; | |
306 | struct cache *cache; | |
307 | ||
308 | unsigned long start_jiffies; | |
309 | dm_oblock_t old_oblock; | |
310 | dm_oblock_t new_oblock; | |
311 | dm_cblock_t cblock; | |
312 | ||
313 | bool err:1; | |
7ae34e77 | 314 | bool discard:1; |
c6b4fcba JT |
315 | bool writeback:1; |
316 | bool demote:1; | |
317 | bool promote:1; | |
c9d28d5d | 318 | bool requeue_holder:1; |
65790ff9 | 319 | bool invalidate:1; |
c6b4fcba JT |
320 | |
321 | struct dm_bio_prison_cell *old_ocell; | |
322 | struct dm_bio_prison_cell *new_ocell; | |
323 | }; | |
324 | ||
325 | /* | |
326 | * Processing a bio in the worker thread may require these memory | |
327 | * allocations. We prealloc to avoid deadlocks (the same worker thread | |
328 | * frees them back to the mempool). | |
329 | */ | |
330 | struct prealloc { | |
331 | struct dm_cache_migration *mg; | |
332 | struct dm_bio_prison_cell *cell1; | |
333 | struct dm_bio_prison_cell *cell2; | |
334 | }; | |
335 | ||
336 | static void wake_worker(struct cache *cache) | |
337 | { | |
338 | queue_work(cache->wq, &cache->worker); | |
339 | } | |
340 | ||
341 | /*----------------------------------------------------------------*/ | |
342 | ||
343 | static struct dm_bio_prison_cell *alloc_prison_cell(struct cache *cache) | |
344 | { | |
345 | /* FIXME: change to use a local slab. */ | |
346 | return dm_bio_prison_alloc_cell(cache->prison, GFP_NOWAIT); | |
347 | } | |
348 | ||
349 | static void free_prison_cell(struct cache *cache, struct dm_bio_prison_cell *cell) | |
350 | { | |
351 | dm_bio_prison_free_cell(cache->prison, cell); | |
352 | } | |
353 | ||
354 | static int prealloc_data_structs(struct cache *cache, struct prealloc *p) | |
355 | { | |
356 | if (!p->mg) { | |
357 | p->mg = mempool_alloc(cache->migration_pool, GFP_NOWAIT); | |
358 | if (!p->mg) | |
359 | return -ENOMEM; | |
360 | } | |
361 | ||
362 | if (!p->cell1) { | |
363 | p->cell1 = alloc_prison_cell(cache); | |
364 | if (!p->cell1) | |
365 | return -ENOMEM; | |
366 | } | |
367 | ||
368 | if (!p->cell2) { | |
369 | p->cell2 = alloc_prison_cell(cache); | |
370 | if (!p->cell2) | |
371 | return -ENOMEM; | |
372 | } | |
373 | ||
374 | return 0; | |
375 | } | |
376 | ||
377 | static void prealloc_free_structs(struct cache *cache, struct prealloc *p) | |
378 | { | |
379 | if (p->cell2) | |
380 | free_prison_cell(cache, p->cell2); | |
381 | ||
382 | if (p->cell1) | |
383 | free_prison_cell(cache, p->cell1); | |
384 | ||
385 | if (p->mg) | |
386 | mempool_free(p->mg, cache->migration_pool); | |
387 | } | |
388 | ||
389 | static struct dm_cache_migration *prealloc_get_migration(struct prealloc *p) | |
390 | { | |
391 | struct dm_cache_migration *mg = p->mg; | |
392 | ||
393 | BUG_ON(!mg); | |
394 | p->mg = NULL; | |
395 | ||
396 | return mg; | |
397 | } | |
398 | ||
399 | /* | |
400 | * You must have a cell within the prealloc struct to return. If not this | |
401 | * function will BUG() rather than returning NULL. | |
402 | */ | |
403 | static struct dm_bio_prison_cell *prealloc_get_cell(struct prealloc *p) | |
404 | { | |
405 | struct dm_bio_prison_cell *r = NULL; | |
406 | ||
407 | if (p->cell1) { | |
408 | r = p->cell1; | |
409 | p->cell1 = NULL; | |
410 | ||
411 | } else if (p->cell2) { | |
412 | r = p->cell2; | |
413 | p->cell2 = NULL; | |
414 | } else | |
415 | BUG(); | |
416 | ||
417 | return r; | |
418 | } | |
419 | ||
420 | /* | |
421 | * You can't have more than two cells in a prealloc struct. BUG() will be | |
422 | * called if you try and overfill. | |
423 | */ | |
424 | static void prealloc_put_cell(struct prealloc *p, struct dm_bio_prison_cell *cell) | |
425 | { | |
426 | if (!p->cell2) | |
427 | p->cell2 = cell; | |
428 | ||
429 | else if (!p->cell1) | |
430 | p->cell1 = cell; | |
431 | ||
432 | else | |
433 | BUG(); | |
434 | } | |
435 | ||
436 | /*----------------------------------------------------------------*/ | |
437 | ||
7ae34e77 | 438 | static void build_key(dm_oblock_t begin, dm_oblock_t end, struct dm_cell_key *key) |
c6b4fcba JT |
439 | { |
440 | key->virtual = 0; | |
441 | key->dev = 0; | |
7ae34e77 JT |
442 | key->block_begin = from_oblock(begin); |
443 | key->block_end = from_oblock(end); | |
c6b4fcba JT |
444 | } |
445 | ||
446 | /* | |
447 | * The caller hands in a preallocated cell, and a free function for it. | |
448 | * The cell will be freed if there's an error, or if it wasn't used because | |
449 | * a cell with that key already exists. | |
450 | */ | |
451 | typedef void (*cell_free_fn)(void *context, struct dm_bio_prison_cell *cell); | |
452 | ||
7ae34e77 JT |
453 | static int bio_detain_range(struct cache *cache, dm_oblock_t oblock_begin, dm_oblock_t oblock_end, |
454 | struct bio *bio, struct dm_bio_prison_cell *cell_prealloc, | |
455 | cell_free_fn free_fn, void *free_context, | |
456 | struct dm_bio_prison_cell **cell_result) | |
c6b4fcba JT |
457 | { |
458 | int r; | |
459 | struct dm_cell_key key; | |
460 | ||
7ae34e77 | 461 | build_key(oblock_begin, oblock_end, &key); |
c6b4fcba JT |
462 | r = dm_bio_detain(cache->prison, &key, bio, cell_prealloc, cell_result); |
463 | if (r) | |
464 | free_fn(free_context, cell_prealloc); | |
465 | ||
466 | return r; | |
467 | } | |
468 | ||
7ae34e77 JT |
469 | static int bio_detain(struct cache *cache, dm_oblock_t oblock, |
470 | struct bio *bio, struct dm_bio_prison_cell *cell_prealloc, | |
471 | cell_free_fn free_fn, void *free_context, | |
472 | struct dm_bio_prison_cell **cell_result) | |
473 | { | |
474 | dm_oblock_t end = to_oblock(from_oblock(oblock) + 1ULL); | |
475 | return bio_detain_range(cache, oblock, end, bio, | |
476 | cell_prealloc, free_fn, free_context, cell_result); | |
477 | } | |
478 | ||
c6b4fcba JT |
479 | static int get_cell(struct cache *cache, |
480 | dm_oblock_t oblock, | |
481 | struct prealloc *structs, | |
482 | struct dm_bio_prison_cell **cell_result) | |
483 | { | |
484 | int r; | |
485 | struct dm_cell_key key; | |
486 | struct dm_bio_prison_cell *cell_prealloc; | |
487 | ||
488 | cell_prealloc = prealloc_get_cell(structs); | |
489 | ||
7ae34e77 | 490 | build_key(oblock, to_oblock(from_oblock(oblock) + 1ULL), &key); |
c6b4fcba JT |
491 | r = dm_get_cell(cache->prison, &key, cell_prealloc, cell_result); |
492 | if (r) | |
493 | prealloc_put_cell(structs, cell_prealloc); | |
494 | ||
495 | return r; | |
496 | } | |
497 | ||
aeed1420 | 498 | /*----------------------------------------------------------------*/ |
c6b4fcba JT |
499 | |
500 | static bool is_dirty(struct cache *cache, dm_cblock_t b) | |
501 | { | |
502 | return test_bit(from_cblock(b), cache->dirty_bitset); | |
503 | } | |
504 | ||
505 | static void set_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock) | |
506 | { | |
507 | if (!test_and_set_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
44fa816b | 508 | atomic_inc(&cache->nr_dirty); |
c6b4fcba JT |
509 | policy_set_dirty(cache->policy, oblock); |
510 | } | |
511 | } | |
512 | ||
513 | static void clear_dirty(struct cache *cache, dm_oblock_t oblock, dm_cblock_t cblock) | |
514 | { | |
515 | if (test_and_clear_bit(from_cblock(cblock), cache->dirty_bitset)) { | |
516 | policy_clear_dirty(cache->policy, oblock); | |
44fa816b | 517 | if (atomic_dec_return(&cache->nr_dirty) == 0) |
c6b4fcba JT |
518 | dm_table_event(cache->ti->table); |
519 | } | |
520 | } | |
521 | ||
522 | /*----------------------------------------------------------------*/ | |
aeed1420 | 523 | |
c6b4fcba JT |
524 | static bool block_size_is_power_of_two(struct cache *cache) |
525 | { | |
526 | return cache->sectors_per_block_shift >= 0; | |
527 | } | |
528 | ||
43aeaa29 MP |
529 | /* gcc on ARM generates spurious references to __udivdi3 and __umoddi3 */ |
530 | #if defined(CONFIG_ARM) && __GNUC__ == 4 && __GNUC_MINOR__ <= 6 | |
531 | __always_inline | |
532 | #endif | |
414dd67d JT |
533 | static dm_block_t block_div(dm_block_t b, uint32_t n) |
534 | { | |
535 | do_div(b, n); | |
536 | ||
537 | return b; | |
538 | } | |
539 | ||
7ae34e77 | 540 | static dm_block_t oblocks_per_dblock(struct cache *cache) |
1bad9bc4 | 541 | { |
7ae34e77 | 542 | dm_block_t oblocks = cache->discard_block_size; |
1bad9bc4 | 543 | |
7ae34e77 JT |
544 | if (block_size_is_power_of_two(cache)) |
545 | oblocks >>= cache->sectors_per_block_shift; | |
1bad9bc4 | 546 | else |
7ae34e77 | 547 | oblocks = block_div(oblocks, cache->sectors_per_block); |
1bad9bc4 | 548 | |
7ae34e77 JT |
549 | return oblocks; |
550 | } | |
551 | ||
552 | static dm_dblock_t oblock_to_dblock(struct cache *cache, dm_oblock_t oblock) | |
553 | { | |
554 | return to_dblock(block_div(from_oblock(oblock), | |
555 | oblocks_per_dblock(cache))); | |
556 | } | |
1bad9bc4 | 557 | |
7ae34e77 JT |
558 | static dm_oblock_t dblock_to_oblock(struct cache *cache, dm_dblock_t dblock) |
559 | { | |
560 | return to_oblock(from_dblock(dblock) * oblocks_per_dblock(cache)); | |
1bad9bc4 JT |
561 | } |
562 | ||
563 | static void set_discard(struct cache *cache, dm_dblock_t b) | |
c6b4fcba JT |
564 | { |
565 | unsigned long flags; | |
566 | ||
7ae34e77 | 567 | BUG_ON(from_dblock(b) >= from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
568 | atomic_inc(&cache->stats.discard_count); |
569 | ||
570 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 571 | set_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
572 | spin_unlock_irqrestore(&cache->lock, flags); |
573 | } | |
574 | ||
1bad9bc4 | 575 | static void clear_discard(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
576 | { |
577 | unsigned long flags; | |
578 | ||
579 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 580 | clear_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
581 | spin_unlock_irqrestore(&cache->lock, flags); |
582 | } | |
583 | ||
1bad9bc4 | 584 | static bool is_discarded(struct cache *cache, dm_dblock_t b) |
c6b4fcba JT |
585 | { |
586 | int r; | |
587 | unsigned long flags; | |
588 | ||
589 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 | 590 | r = test_bit(from_dblock(b), cache->discard_bitset); |
c6b4fcba JT |
591 | spin_unlock_irqrestore(&cache->lock, flags); |
592 | ||
593 | return r; | |
594 | } | |
595 | ||
596 | static bool is_discarded_oblock(struct cache *cache, dm_oblock_t b) | |
597 | { | |
598 | int r; | |
599 | unsigned long flags; | |
600 | ||
601 | spin_lock_irqsave(&cache->lock, flags); | |
1bad9bc4 JT |
602 | r = test_bit(from_dblock(oblock_to_dblock(cache, b)), |
603 | cache->discard_bitset); | |
c6b4fcba JT |
604 | spin_unlock_irqrestore(&cache->lock, flags); |
605 | ||
606 | return r; | |
607 | } | |
608 | ||
609 | /*----------------------------------------------------------------*/ | |
610 | ||
611 | static void load_stats(struct cache *cache) | |
612 | { | |
613 | struct dm_cache_statistics stats; | |
614 | ||
615 | dm_cache_metadata_get_stats(cache->cmd, &stats); | |
616 | atomic_set(&cache->stats.read_hit, stats.read_hits); | |
617 | atomic_set(&cache->stats.read_miss, stats.read_misses); | |
618 | atomic_set(&cache->stats.write_hit, stats.write_hits); | |
619 | atomic_set(&cache->stats.write_miss, stats.write_misses); | |
620 | } | |
621 | ||
622 | static void save_stats(struct cache *cache) | |
623 | { | |
624 | struct dm_cache_statistics stats; | |
625 | ||
626 | stats.read_hits = atomic_read(&cache->stats.read_hit); | |
627 | stats.read_misses = atomic_read(&cache->stats.read_miss); | |
628 | stats.write_hits = atomic_read(&cache->stats.write_hit); | |
629 | stats.write_misses = atomic_read(&cache->stats.write_miss); | |
630 | ||
631 | dm_cache_metadata_set_stats(cache->cmd, &stats); | |
632 | } | |
633 | ||
634 | /*---------------------------------------------------------------- | |
635 | * Per bio data | |
636 | *--------------------------------------------------------------*/ | |
19b0092e MS |
637 | |
638 | /* | |
639 | * If using writeback, leave out struct per_bio_data's writethrough fields. | |
640 | */ | |
641 | #define PB_DATA_SIZE_WB (offsetof(struct per_bio_data, cache)) | |
642 | #define PB_DATA_SIZE_WT (sizeof(struct per_bio_data)) | |
643 | ||
2ee57d58 JT |
644 | static bool writethrough_mode(struct cache_features *f) |
645 | { | |
646 | return f->io_mode == CM_IO_WRITETHROUGH; | |
647 | } | |
648 | ||
649 | static bool writeback_mode(struct cache_features *f) | |
650 | { | |
651 | return f->io_mode == CM_IO_WRITEBACK; | |
652 | } | |
653 | ||
654 | static bool passthrough_mode(struct cache_features *f) | |
655 | { | |
656 | return f->io_mode == CM_IO_PASSTHROUGH; | |
657 | } | |
658 | ||
19b0092e MS |
659 | static size_t get_per_bio_data_size(struct cache *cache) |
660 | { | |
2ee57d58 | 661 | return writethrough_mode(&cache->features) ? PB_DATA_SIZE_WT : PB_DATA_SIZE_WB; |
19b0092e MS |
662 | } |
663 | ||
664 | static struct per_bio_data *get_per_bio_data(struct bio *bio, size_t data_size) | |
c6b4fcba | 665 | { |
19b0092e | 666 | struct per_bio_data *pb = dm_per_bio_data(bio, data_size); |
c6b4fcba JT |
667 | BUG_ON(!pb); |
668 | return pb; | |
669 | } | |
670 | ||
19b0092e | 671 | static struct per_bio_data *init_per_bio_data(struct bio *bio, size_t data_size) |
c6b4fcba | 672 | { |
19b0092e | 673 | struct per_bio_data *pb = get_per_bio_data(bio, data_size); |
c6b4fcba JT |
674 | |
675 | pb->tick = false; | |
676 | pb->req_nr = dm_bio_get_target_bio_nr(bio); | |
677 | pb->all_io_entry = NULL; | |
678 | ||
679 | return pb; | |
680 | } | |
681 | ||
682 | /*---------------------------------------------------------------- | |
683 | * Remapping | |
684 | *--------------------------------------------------------------*/ | |
685 | static void remap_to_origin(struct cache *cache, struct bio *bio) | |
686 | { | |
687 | bio->bi_bdev = cache->origin_dev->bdev; | |
688 | } | |
689 | ||
690 | static void remap_to_cache(struct cache *cache, struct bio *bio, | |
691 | dm_cblock_t cblock) | |
692 | { | |
4f024f37 | 693 | sector_t bi_sector = bio->bi_iter.bi_sector; |
e0d849fa | 694 | sector_t block = from_cblock(cblock); |
c6b4fcba JT |
695 | |
696 | bio->bi_bdev = cache->cache_dev->bdev; | |
697 | if (!block_size_is_power_of_two(cache)) | |
4f024f37 | 698 | bio->bi_iter.bi_sector = |
e0d849fa | 699 | (block * cache->sectors_per_block) + |
4f024f37 | 700 | sector_div(bi_sector, cache->sectors_per_block); |
c6b4fcba | 701 | else |
4f024f37 | 702 | bio->bi_iter.bi_sector = |
e0d849fa | 703 | (block << cache->sectors_per_block_shift) | |
4f024f37 | 704 | (bi_sector & (cache->sectors_per_block - 1)); |
c6b4fcba JT |
705 | } |
706 | ||
707 | static void check_if_tick_bio_needed(struct cache *cache, struct bio *bio) | |
708 | { | |
709 | unsigned long flags; | |
19b0092e MS |
710 | size_t pb_data_size = get_per_bio_data_size(cache); |
711 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
712 | |
713 | spin_lock_irqsave(&cache->lock, flags); | |
714 | if (cache->need_tick_bio && | |
715 | !(bio->bi_rw & (REQ_FUA | REQ_FLUSH | REQ_DISCARD))) { | |
716 | pb->tick = true; | |
717 | cache->need_tick_bio = false; | |
718 | } | |
719 | spin_unlock_irqrestore(&cache->lock, flags); | |
720 | } | |
721 | ||
722 | static void remap_to_origin_clear_discard(struct cache *cache, struct bio *bio, | |
723 | dm_oblock_t oblock) | |
724 | { | |
725 | check_if_tick_bio_needed(cache, bio); | |
726 | remap_to_origin(cache, bio); | |
727 | if (bio_data_dir(bio) == WRITE) | |
1bad9bc4 | 728 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
729 | } |
730 | ||
731 | static void remap_to_cache_dirty(struct cache *cache, struct bio *bio, | |
732 | dm_oblock_t oblock, dm_cblock_t cblock) | |
733 | { | |
f8e5f01a | 734 | check_if_tick_bio_needed(cache, bio); |
c6b4fcba JT |
735 | remap_to_cache(cache, bio, cblock); |
736 | if (bio_data_dir(bio) == WRITE) { | |
737 | set_dirty(cache, oblock, cblock); | |
1bad9bc4 | 738 | clear_discard(cache, oblock_to_dblock(cache, oblock)); |
c6b4fcba JT |
739 | } |
740 | } | |
741 | ||
742 | static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio) | |
743 | { | |
4f024f37 | 744 | sector_t block_nr = bio->bi_iter.bi_sector; |
c6b4fcba JT |
745 | |
746 | if (!block_size_is_power_of_two(cache)) | |
747 | (void) sector_div(block_nr, cache->sectors_per_block); | |
748 | else | |
749 | block_nr >>= cache->sectors_per_block_shift; | |
750 | ||
751 | return to_oblock(block_nr); | |
752 | } | |
753 | ||
754 | static int bio_triggers_commit(struct cache *cache, struct bio *bio) | |
755 | { | |
756 | return bio->bi_rw & (REQ_FLUSH | REQ_FUA); | |
757 | } | |
758 | ||
8c081b52 JT |
759 | /* |
760 | * You must increment the deferred set whilst the prison cell is held. To | |
761 | * encourage this, we ask for 'cell' to be passed in. | |
762 | */ | |
763 | static void inc_ds(struct cache *cache, struct bio *bio, | |
764 | struct dm_bio_prison_cell *cell) | |
765 | { | |
766 | size_t pb_data_size = get_per_bio_data_size(cache); | |
767 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
768 | ||
769 | BUG_ON(!cell); | |
770 | BUG_ON(pb->all_io_entry); | |
771 | ||
772 | pb->all_io_entry = dm_deferred_entry_inc(cache->all_io_ds); | |
773 | } | |
774 | ||
c6b4fcba JT |
775 | static void issue(struct cache *cache, struct bio *bio) |
776 | { | |
777 | unsigned long flags; | |
778 | ||
779 | if (!bio_triggers_commit(cache, bio)) { | |
780 | generic_make_request(bio); | |
781 | return; | |
782 | } | |
783 | ||
784 | /* | |
785 | * Batch together any bios that trigger commits and then issue a | |
786 | * single commit for them in do_worker(). | |
787 | */ | |
788 | spin_lock_irqsave(&cache->lock, flags); | |
789 | cache->commit_requested = true; | |
790 | bio_list_add(&cache->deferred_flush_bios, bio); | |
791 | spin_unlock_irqrestore(&cache->lock, flags); | |
792 | } | |
793 | ||
8c081b52 JT |
794 | static void inc_and_issue(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell *cell) |
795 | { | |
796 | inc_ds(cache, bio, cell); | |
797 | issue(cache, bio); | |
798 | } | |
799 | ||
e2e74d61 JT |
800 | static void defer_writethrough_bio(struct cache *cache, struct bio *bio) |
801 | { | |
802 | unsigned long flags; | |
803 | ||
804 | spin_lock_irqsave(&cache->lock, flags); | |
805 | bio_list_add(&cache->deferred_writethrough_bios, bio); | |
806 | spin_unlock_irqrestore(&cache->lock, flags); | |
807 | ||
808 | wake_worker(cache); | |
809 | } | |
810 | ||
811 | static void writethrough_endio(struct bio *bio, int err) | |
812 | { | |
19b0092e | 813 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
c9d28d5d JT |
814 | |
815 | dm_unhook_bio(&pb->hook_info, bio); | |
e2e74d61 JT |
816 | |
817 | if (err) { | |
818 | bio_endio(bio, err); | |
819 | return; | |
820 | } | |
821 | ||
b844fe69 | 822 | dm_bio_restore(&pb->bio_details, bio); |
e2e74d61 JT |
823 | remap_to_cache(pb->cache, bio, pb->cblock); |
824 | ||
825 | /* | |
826 | * We can't issue this bio directly, since we're in interrupt | |
aeed1420 | 827 | * context. So it gets put on a bio list for processing by the |
e2e74d61 JT |
828 | * worker thread. |
829 | */ | |
830 | defer_writethrough_bio(pb->cache, bio); | |
831 | } | |
832 | ||
833 | /* | |
834 | * When running in writethrough mode we need to send writes to clean blocks | |
835 | * to both the cache and origin devices. In future we'd like to clone the | |
836 | * bio and send them in parallel, but for now we're doing them in | |
837 | * series as this is easier. | |
838 | */ | |
839 | static void remap_to_origin_then_cache(struct cache *cache, struct bio *bio, | |
840 | dm_oblock_t oblock, dm_cblock_t cblock) | |
841 | { | |
19b0092e | 842 | struct per_bio_data *pb = get_per_bio_data(bio, PB_DATA_SIZE_WT); |
e2e74d61 JT |
843 | |
844 | pb->cache = cache; | |
845 | pb->cblock = cblock; | |
c9d28d5d | 846 | dm_hook_bio(&pb->hook_info, bio, writethrough_endio, NULL); |
b844fe69 | 847 | dm_bio_record(&pb->bio_details, bio); |
e2e74d61 JT |
848 | |
849 | remap_to_origin_clear_discard(pb->cache, bio, oblock); | |
850 | } | |
851 | ||
c6b4fcba JT |
852 | /*---------------------------------------------------------------- |
853 | * Migration processing | |
854 | * | |
855 | * Migration covers moving data from the origin device to the cache, or | |
856 | * vice versa. | |
857 | *--------------------------------------------------------------*/ | |
858 | static void free_migration(struct dm_cache_migration *mg) | |
859 | { | |
860 | mempool_free(mg, mg->cache->migration_pool); | |
861 | } | |
862 | ||
863 | static void inc_nr_migrations(struct cache *cache) | |
864 | { | |
865 | atomic_inc(&cache->nr_migrations); | |
866 | } | |
867 | ||
868 | static void dec_nr_migrations(struct cache *cache) | |
869 | { | |
870 | atomic_dec(&cache->nr_migrations); | |
871 | ||
872 | /* | |
873 | * Wake the worker in case we're suspending the target. | |
874 | */ | |
875 | wake_up(&cache->migration_wait); | |
876 | } | |
877 | ||
878 | static void __cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, | |
879 | bool holder) | |
880 | { | |
881 | (holder ? dm_cell_release : dm_cell_release_no_holder) | |
882 | (cache->prison, cell, &cache->deferred_bios); | |
883 | free_prison_cell(cache, cell); | |
884 | } | |
885 | ||
886 | static void cell_defer(struct cache *cache, struct dm_bio_prison_cell *cell, | |
887 | bool holder) | |
888 | { | |
889 | unsigned long flags; | |
890 | ||
891 | spin_lock_irqsave(&cache->lock, flags); | |
892 | __cell_defer(cache, cell, holder); | |
893 | spin_unlock_irqrestore(&cache->lock, flags); | |
894 | ||
895 | wake_worker(cache); | |
896 | } | |
897 | ||
898 | static void cleanup_migration(struct dm_cache_migration *mg) | |
899 | { | |
66cb1910 | 900 | struct cache *cache = mg->cache; |
c6b4fcba | 901 | free_migration(mg); |
66cb1910 | 902 | dec_nr_migrations(cache); |
c6b4fcba JT |
903 | } |
904 | ||
905 | static void migration_failure(struct dm_cache_migration *mg) | |
906 | { | |
907 | struct cache *cache = mg->cache; | |
908 | ||
909 | if (mg->writeback) { | |
910 | DMWARN_LIMIT("writeback failed; couldn't copy block"); | |
911 | set_dirty(cache, mg->old_oblock, mg->cblock); | |
912 | cell_defer(cache, mg->old_ocell, false); | |
913 | ||
914 | } else if (mg->demote) { | |
915 | DMWARN_LIMIT("demotion failed; couldn't copy block"); | |
916 | policy_force_mapping(cache->policy, mg->new_oblock, mg->old_oblock); | |
917 | ||
80f659f3 | 918 | cell_defer(cache, mg->old_ocell, mg->promote ? false : true); |
c6b4fcba | 919 | if (mg->promote) |
80f659f3 | 920 | cell_defer(cache, mg->new_ocell, true); |
c6b4fcba JT |
921 | } else { |
922 | DMWARN_LIMIT("promotion failed; couldn't copy block"); | |
923 | policy_remove_mapping(cache->policy, mg->new_oblock); | |
80f659f3 | 924 | cell_defer(cache, mg->new_ocell, true); |
c6b4fcba JT |
925 | } |
926 | ||
927 | cleanup_migration(mg); | |
928 | } | |
929 | ||
930 | static void migration_success_pre_commit(struct dm_cache_migration *mg) | |
931 | { | |
932 | unsigned long flags; | |
933 | struct cache *cache = mg->cache; | |
934 | ||
935 | if (mg->writeback) { | |
c6b4fcba | 936 | clear_dirty(cache, mg->old_oblock, mg->cblock); |
40aa978e | 937 | cell_defer(cache, mg->old_ocell, false); |
c6b4fcba JT |
938 | cleanup_migration(mg); |
939 | return; | |
940 | ||
941 | } else if (mg->demote) { | |
942 | if (dm_cache_remove_mapping(cache->cmd, mg->cblock)) { | |
943 | DMWARN_LIMIT("demotion failed; couldn't update on disk metadata"); | |
944 | policy_force_mapping(cache->policy, mg->new_oblock, | |
945 | mg->old_oblock); | |
946 | if (mg->promote) | |
947 | cell_defer(cache, mg->new_ocell, true); | |
948 | cleanup_migration(mg); | |
949 | return; | |
950 | } | |
951 | } else { | |
952 | if (dm_cache_insert_mapping(cache->cmd, mg->cblock, mg->new_oblock)) { | |
953 | DMWARN_LIMIT("promotion failed; couldn't update on disk metadata"); | |
954 | policy_remove_mapping(cache->policy, mg->new_oblock); | |
955 | cleanup_migration(mg); | |
956 | return; | |
957 | } | |
958 | } | |
959 | ||
960 | spin_lock_irqsave(&cache->lock, flags); | |
961 | list_add_tail(&mg->list, &cache->need_commit_migrations); | |
962 | cache->commit_requested = true; | |
963 | spin_unlock_irqrestore(&cache->lock, flags); | |
964 | } | |
965 | ||
966 | static void migration_success_post_commit(struct dm_cache_migration *mg) | |
967 | { | |
968 | unsigned long flags; | |
969 | struct cache *cache = mg->cache; | |
970 | ||
971 | if (mg->writeback) { | |
972 | DMWARN("writeback unexpectedly triggered commit"); | |
973 | return; | |
974 | ||
975 | } else if (mg->demote) { | |
80f659f3 | 976 | cell_defer(cache, mg->old_ocell, mg->promote ? false : true); |
c6b4fcba JT |
977 | |
978 | if (mg->promote) { | |
979 | mg->demote = false; | |
980 | ||
981 | spin_lock_irqsave(&cache->lock, flags); | |
982 | list_add_tail(&mg->list, &cache->quiesced_migrations); | |
983 | spin_unlock_irqrestore(&cache->lock, flags); | |
984 | ||
65790ff9 JT |
985 | } else { |
986 | if (mg->invalidate) | |
987 | policy_remove_mapping(cache->policy, mg->old_oblock); | |
c6b4fcba | 988 | cleanup_migration(mg); |
65790ff9 | 989 | } |
c6b4fcba JT |
990 | |
991 | } else { | |
1e32134a JT |
992 | if (mg->requeue_holder) { |
993 | clear_dirty(cache, mg->new_oblock, mg->cblock); | |
c9d28d5d | 994 | cell_defer(cache, mg->new_ocell, true); |
1e32134a JT |
995 | } else { |
996 | /* | |
997 | * The block was promoted via an overwrite, so it's dirty. | |
998 | */ | |
999 | set_dirty(cache, mg->new_oblock, mg->cblock); | |
c9d28d5d JT |
1000 | bio_endio(mg->new_ocell->holder, 0); |
1001 | cell_defer(cache, mg->new_ocell, false); | |
1002 | } | |
c6b4fcba JT |
1003 | cleanup_migration(mg); |
1004 | } | |
1005 | } | |
1006 | ||
1007 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
1008 | { | |
1009 | unsigned long flags; | |
1010 | struct dm_cache_migration *mg = (struct dm_cache_migration *) context; | |
1011 | struct cache *cache = mg->cache; | |
1012 | ||
1013 | if (read_err || write_err) | |
1014 | mg->err = true; | |
1015 | ||
1016 | spin_lock_irqsave(&cache->lock, flags); | |
1017 | list_add_tail(&mg->list, &cache->completed_migrations); | |
1018 | spin_unlock_irqrestore(&cache->lock, flags); | |
1019 | ||
1020 | wake_worker(cache); | |
1021 | } | |
1022 | ||
7ae34e77 | 1023 | static void issue_copy(struct dm_cache_migration *mg) |
c6b4fcba JT |
1024 | { |
1025 | int r; | |
1026 | struct dm_io_region o_region, c_region; | |
1027 | struct cache *cache = mg->cache; | |
8b9d9666 | 1028 | sector_t cblock = from_cblock(mg->cblock); |
c6b4fcba JT |
1029 | |
1030 | o_region.bdev = cache->origin_dev->bdev; | |
1031 | o_region.count = cache->sectors_per_block; | |
1032 | ||
1033 | c_region.bdev = cache->cache_dev->bdev; | |
8b9d9666 | 1034 | c_region.sector = cblock * cache->sectors_per_block; |
c6b4fcba JT |
1035 | c_region.count = cache->sectors_per_block; |
1036 | ||
1037 | if (mg->writeback || mg->demote) { | |
1038 | /* demote */ | |
1039 | o_region.sector = from_oblock(mg->old_oblock) * cache->sectors_per_block; | |
1040 | r = dm_kcopyd_copy(cache->copier, &c_region, 1, &o_region, 0, copy_complete, mg); | |
1041 | } else { | |
1042 | /* promote */ | |
1043 | o_region.sector = from_oblock(mg->new_oblock) * cache->sectors_per_block; | |
1044 | r = dm_kcopyd_copy(cache->copier, &o_region, 1, &c_region, 0, copy_complete, mg); | |
1045 | } | |
1046 | ||
2c2263c9 HM |
1047 | if (r < 0) { |
1048 | DMERR_LIMIT("issuing migration failed"); | |
c6b4fcba | 1049 | migration_failure(mg); |
2c2263c9 | 1050 | } |
c6b4fcba JT |
1051 | } |
1052 | ||
c9d28d5d JT |
1053 | static void overwrite_endio(struct bio *bio, int err) |
1054 | { | |
1055 | struct dm_cache_migration *mg = bio->bi_private; | |
1056 | struct cache *cache = mg->cache; | |
1057 | size_t pb_data_size = get_per_bio_data_size(cache); | |
1058 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1059 | unsigned long flags; | |
1060 | ||
80ae49aa MS |
1061 | dm_unhook_bio(&pb->hook_info, bio); |
1062 | ||
c9d28d5d JT |
1063 | if (err) |
1064 | mg->err = true; | |
1065 | ||
80ae49aa MS |
1066 | mg->requeue_holder = false; |
1067 | ||
c9d28d5d JT |
1068 | spin_lock_irqsave(&cache->lock, flags); |
1069 | list_add_tail(&mg->list, &cache->completed_migrations); | |
c9d28d5d JT |
1070 | spin_unlock_irqrestore(&cache->lock, flags); |
1071 | ||
1072 | wake_worker(cache); | |
1073 | } | |
1074 | ||
1075 | static void issue_overwrite(struct dm_cache_migration *mg, struct bio *bio) | |
1076 | { | |
1077 | size_t pb_data_size = get_per_bio_data_size(mg->cache); | |
1078 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
1079 | ||
1080 | dm_hook_bio(&pb->hook_info, bio, overwrite_endio, mg); | |
1081 | remap_to_cache_dirty(mg->cache, bio, mg->new_oblock, mg->cblock); | |
8c081b52 JT |
1082 | |
1083 | /* | |
1084 | * No need to inc_ds() here, since the cell will be held for the | |
1085 | * duration of the io. | |
1086 | */ | |
c9d28d5d JT |
1087 | generic_make_request(bio); |
1088 | } | |
1089 | ||
1090 | static bool bio_writes_complete_block(struct cache *cache, struct bio *bio) | |
1091 | { | |
1092 | return (bio_data_dir(bio) == WRITE) && | |
4f024f37 | 1093 | (bio->bi_iter.bi_size == (cache->sectors_per_block << SECTOR_SHIFT)); |
c9d28d5d JT |
1094 | } |
1095 | ||
c6b4fcba JT |
1096 | static void avoid_copy(struct dm_cache_migration *mg) |
1097 | { | |
1098 | atomic_inc(&mg->cache->stats.copies_avoided); | |
1099 | migration_success_pre_commit(mg); | |
1100 | } | |
1101 | ||
7ae34e77 JT |
1102 | static void calc_discard_block_range(struct cache *cache, struct bio *bio, |
1103 | dm_dblock_t *b, dm_dblock_t *e) | |
1104 | { | |
1105 | sector_t sb = bio->bi_iter.bi_sector; | |
1106 | sector_t se = bio_end_sector(bio); | |
1107 | ||
1108 | *b = to_dblock(dm_sector_div_up(sb, cache->discard_block_size)); | |
1109 | ||
1110 | if (se - sb < cache->discard_block_size) | |
1111 | *e = *b; | |
1112 | else | |
1113 | *e = to_dblock(block_div(se, cache->discard_block_size)); | |
1114 | } | |
1115 | ||
1116 | static void issue_discard(struct dm_cache_migration *mg) | |
1117 | { | |
1118 | dm_dblock_t b, e; | |
1119 | struct bio *bio = mg->new_ocell->holder; | |
1120 | ||
1121 | calc_discard_block_range(mg->cache, bio, &b, &e); | |
1122 | while (b != e) { | |
1123 | set_discard(mg->cache, b); | |
1124 | b = to_dblock(from_dblock(b) + 1); | |
1125 | } | |
1126 | ||
1127 | bio_endio(bio, 0); | |
1128 | cell_defer(mg->cache, mg->new_ocell, false); | |
1129 | free_migration(mg); | |
1130 | } | |
1131 | ||
1132 | static void issue_copy_or_discard(struct dm_cache_migration *mg) | |
c6b4fcba JT |
1133 | { |
1134 | bool avoid; | |
1135 | struct cache *cache = mg->cache; | |
1136 | ||
7ae34e77 JT |
1137 | if (mg->discard) { |
1138 | issue_discard(mg); | |
1139 | return; | |
1140 | } | |
1141 | ||
c6b4fcba JT |
1142 | if (mg->writeback || mg->demote) |
1143 | avoid = !is_dirty(cache, mg->cblock) || | |
1144 | is_discarded_oblock(cache, mg->old_oblock); | |
c9d28d5d JT |
1145 | else { |
1146 | struct bio *bio = mg->new_ocell->holder; | |
1147 | ||
c6b4fcba JT |
1148 | avoid = is_discarded_oblock(cache, mg->new_oblock); |
1149 | ||
f29a3147 JT |
1150 | if (writeback_mode(&cache->features) && |
1151 | !avoid && bio_writes_complete_block(cache, bio)) { | |
c9d28d5d JT |
1152 | issue_overwrite(mg, bio); |
1153 | return; | |
1154 | } | |
1155 | } | |
1156 | ||
7ae34e77 | 1157 | avoid ? avoid_copy(mg) : issue_copy(mg); |
c6b4fcba JT |
1158 | } |
1159 | ||
1160 | static void complete_migration(struct dm_cache_migration *mg) | |
1161 | { | |
1162 | if (mg->err) | |
1163 | migration_failure(mg); | |
1164 | else | |
1165 | migration_success_pre_commit(mg); | |
1166 | } | |
1167 | ||
1168 | static void process_migrations(struct cache *cache, struct list_head *head, | |
1169 | void (*fn)(struct dm_cache_migration *)) | |
1170 | { | |
1171 | unsigned long flags; | |
1172 | struct list_head list; | |
1173 | struct dm_cache_migration *mg, *tmp; | |
1174 | ||
1175 | INIT_LIST_HEAD(&list); | |
1176 | spin_lock_irqsave(&cache->lock, flags); | |
1177 | list_splice_init(head, &list); | |
1178 | spin_unlock_irqrestore(&cache->lock, flags); | |
1179 | ||
1180 | list_for_each_entry_safe(mg, tmp, &list, list) | |
1181 | fn(mg); | |
1182 | } | |
1183 | ||
1184 | static void __queue_quiesced_migration(struct dm_cache_migration *mg) | |
1185 | { | |
1186 | list_add_tail(&mg->list, &mg->cache->quiesced_migrations); | |
1187 | } | |
1188 | ||
1189 | static void queue_quiesced_migration(struct dm_cache_migration *mg) | |
1190 | { | |
1191 | unsigned long flags; | |
1192 | struct cache *cache = mg->cache; | |
1193 | ||
1194 | spin_lock_irqsave(&cache->lock, flags); | |
1195 | __queue_quiesced_migration(mg); | |
1196 | spin_unlock_irqrestore(&cache->lock, flags); | |
1197 | ||
1198 | wake_worker(cache); | |
1199 | } | |
1200 | ||
1201 | static void queue_quiesced_migrations(struct cache *cache, struct list_head *work) | |
1202 | { | |
1203 | unsigned long flags; | |
1204 | struct dm_cache_migration *mg, *tmp; | |
1205 | ||
1206 | spin_lock_irqsave(&cache->lock, flags); | |
1207 | list_for_each_entry_safe(mg, tmp, work, list) | |
1208 | __queue_quiesced_migration(mg); | |
1209 | spin_unlock_irqrestore(&cache->lock, flags); | |
1210 | ||
1211 | wake_worker(cache); | |
1212 | } | |
1213 | ||
1214 | static void check_for_quiesced_migrations(struct cache *cache, | |
1215 | struct per_bio_data *pb) | |
1216 | { | |
1217 | struct list_head work; | |
1218 | ||
1219 | if (!pb->all_io_entry) | |
1220 | return; | |
1221 | ||
1222 | INIT_LIST_HEAD(&work); | |
8c081b52 | 1223 | dm_deferred_entry_dec(pb->all_io_entry, &work); |
c6b4fcba JT |
1224 | |
1225 | if (!list_empty(&work)) | |
1226 | queue_quiesced_migrations(cache, &work); | |
1227 | } | |
1228 | ||
1229 | static void quiesce_migration(struct dm_cache_migration *mg) | |
1230 | { | |
1231 | if (!dm_deferred_set_add_work(mg->cache->all_io_ds, &mg->list)) | |
1232 | queue_quiesced_migration(mg); | |
1233 | } | |
1234 | ||
1235 | static void promote(struct cache *cache, struct prealloc *structs, | |
1236 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1237 | struct dm_bio_prison_cell *cell) | |
1238 | { | |
1239 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1240 | ||
1241 | mg->err = false; | |
7ae34e77 | 1242 | mg->discard = false; |
c6b4fcba JT |
1243 | mg->writeback = false; |
1244 | mg->demote = false; | |
1245 | mg->promote = true; | |
c9d28d5d | 1246 | mg->requeue_holder = true; |
65790ff9 | 1247 | mg->invalidate = false; |
c6b4fcba JT |
1248 | mg->cache = cache; |
1249 | mg->new_oblock = oblock; | |
1250 | mg->cblock = cblock; | |
1251 | mg->old_ocell = NULL; | |
1252 | mg->new_ocell = cell; | |
1253 | mg->start_jiffies = jiffies; | |
1254 | ||
1255 | inc_nr_migrations(cache); | |
1256 | quiesce_migration(mg); | |
1257 | } | |
1258 | ||
1259 | static void writeback(struct cache *cache, struct prealloc *structs, | |
1260 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1261 | struct dm_bio_prison_cell *cell) | |
1262 | { | |
1263 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1264 | ||
1265 | mg->err = false; | |
7ae34e77 | 1266 | mg->discard = false; |
c6b4fcba JT |
1267 | mg->writeback = true; |
1268 | mg->demote = false; | |
1269 | mg->promote = false; | |
c9d28d5d | 1270 | mg->requeue_holder = true; |
65790ff9 | 1271 | mg->invalidate = false; |
c6b4fcba JT |
1272 | mg->cache = cache; |
1273 | mg->old_oblock = oblock; | |
1274 | mg->cblock = cblock; | |
1275 | mg->old_ocell = cell; | |
1276 | mg->new_ocell = NULL; | |
1277 | mg->start_jiffies = jiffies; | |
1278 | ||
1279 | inc_nr_migrations(cache); | |
1280 | quiesce_migration(mg); | |
1281 | } | |
1282 | ||
1283 | static void demote_then_promote(struct cache *cache, struct prealloc *structs, | |
1284 | dm_oblock_t old_oblock, dm_oblock_t new_oblock, | |
1285 | dm_cblock_t cblock, | |
1286 | struct dm_bio_prison_cell *old_ocell, | |
1287 | struct dm_bio_prison_cell *new_ocell) | |
1288 | { | |
1289 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1290 | ||
1291 | mg->err = false; | |
7ae34e77 | 1292 | mg->discard = false; |
c6b4fcba JT |
1293 | mg->writeback = false; |
1294 | mg->demote = true; | |
1295 | mg->promote = true; | |
c9d28d5d | 1296 | mg->requeue_holder = true; |
65790ff9 | 1297 | mg->invalidate = false; |
c6b4fcba JT |
1298 | mg->cache = cache; |
1299 | mg->old_oblock = old_oblock; | |
1300 | mg->new_oblock = new_oblock; | |
1301 | mg->cblock = cblock; | |
1302 | mg->old_ocell = old_ocell; | |
1303 | mg->new_ocell = new_ocell; | |
1304 | mg->start_jiffies = jiffies; | |
1305 | ||
1306 | inc_nr_migrations(cache); | |
1307 | quiesce_migration(mg); | |
1308 | } | |
1309 | ||
2ee57d58 JT |
1310 | /* |
1311 | * Invalidate a cache entry. No writeback occurs; any changes in the cache | |
1312 | * block are thrown away. | |
1313 | */ | |
1314 | static void invalidate(struct cache *cache, struct prealloc *structs, | |
1315 | dm_oblock_t oblock, dm_cblock_t cblock, | |
1316 | struct dm_bio_prison_cell *cell) | |
1317 | { | |
1318 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1319 | ||
1320 | mg->err = false; | |
7ae34e77 | 1321 | mg->discard = false; |
2ee57d58 JT |
1322 | mg->writeback = false; |
1323 | mg->demote = true; | |
1324 | mg->promote = false; | |
1325 | mg->requeue_holder = true; | |
65790ff9 | 1326 | mg->invalidate = true; |
2ee57d58 JT |
1327 | mg->cache = cache; |
1328 | mg->old_oblock = oblock; | |
1329 | mg->cblock = cblock; | |
1330 | mg->old_ocell = cell; | |
1331 | mg->new_ocell = NULL; | |
1332 | mg->start_jiffies = jiffies; | |
1333 | ||
1334 | inc_nr_migrations(cache); | |
1335 | quiesce_migration(mg); | |
1336 | } | |
1337 | ||
7ae34e77 JT |
1338 | static void discard(struct cache *cache, struct prealloc *structs, |
1339 | struct dm_bio_prison_cell *cell) | |
1340 | { | |
1341 | struct dm_cache_migration *mg = prealloc_get_migration(structs); | |
1342 | ||
1343 | mg->err = false; | |
1344 | mg->discard = true; | |
1345 | mg->writeback = false; | |
1346 | mg->demote = false; | |
1347 | mg->promote = false; | |
1348 | mg->requeue_holder = false; | |
1349 | mg->invalidate = false; | |
1350 | mg->cache = cache; | |
1351 | mg->old_ocell = NULL; | |
1352 | mg->new_ocell = cell; | |
1353 | mg->start_jiffies = jiffies; | |
1354 | ||
1355 | quiesce_migration(mg); | |
1356 | } | |
1357 | ||
c6b4fcba JT |
1358 | /*---------------------------------------------------------------- |
1359 | * bio processing | |
1360 | *--------------------------------------------------------------*/ | |
1361 | static void defer_bio(struct cache *cache, struct bio *bio) | |
1362 | { | |
1363 | unsigned long flags; | |
1364 | ||
1365 | spin_lock_irqsave(&cache->lock, flags); | |
1366 | bio_list_add(&cache->deferred_bios, bio); | |
1367 | spin_unlock_irqrestore(&cache->lock, flags); | |
1368 | ||
1369 | wake_worker(cache); | |
1370 | } | |
1371 | ||
1372 | static void process_flush_bio(struct cache *cache, struct bio *bio) | |
1373 | { | |
19b0092e MS |
1374 | size_t pb_data_size = get_per_bio_data_size(cache); |
1375 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba | 1376 | |
4f024f37 | 1377 | BUG_ON(bio->bi_iter.bi_size); |
c6b4fcba JT |
1378 | if (!pb->req_nr) |
1379 | remap_to_origin(cache, bio); | |
1380 | else | |
1381 | remap_to_cache(cache, bio, 0); | |
1382 | ||
8c081b52 JT |
1383 | /* |
1384 | * REQ_FLUSH is not directed at any particular block so we don't | |
1385 | * need to inc_ds(). REQ_FUA's are split into a write + REQ_FLUSH | |
1386 | * by dm-core. | |
1387 | */ | |
c6b4fcba JT |
1388 | issue(cache, bio); |
1389 | } | |
1390 | ||
7ae34e77 JT |
1391 | static void process_discard_bio(struct cache *cache, struct prealloc *structs, |
1392 | struct bio *bio) | |
c6b4fcba | 1393 | { |
7ae34e77 JT |
1394 | int r; |
1395 | dm_dblock_t b, e; | |
1396 | struct dm_bio_prison_cell *cell_prealloc, *new_ocell; | |
c6b4fcba | 1397 | |
7ae34e77 JT |
1398 | calc_discard_block_range(cache, bio, &b, &e); |
1399 | if (b == e) { | |
1400 | bio_endio(bio, 0); | |
1401 | return; | |
1402 | } | |
c6b4fcba | 1403 | |
7ae34e77 JT |
1404 | cell_prealloc = prealloc_get_cell(structs); |
1405 | r = bio_detain_range(cache, dblock_to_oblock(cache, b), dblock_to_oblock(cache, e), bio, cell_prealloc, | |
1406 | (cell_free_fn) prealloc_put_cell, | |
1407 | structs, &new_ocell); | |
1408 | if (r > 0) | |
1409 | return; | |
c6b4fcba | 1410 | |
7ae34e77 | 1411 | discard(cache, structs, new_ocell); |
c6b4fcba JT |
1412 | } |
1413 | ||
1414 | static bool spare_migration_bandwidth(struct cache *cache) | |
1415 | { | |
1416 | sector_t current_volume = (atomic_read(&cache->nr_migrations) + 1) * | |
1417 | cache->sectors_per_block; | |
1418 | return current_volume < cache->migration_threshold; | |
1419 | } | |
1420 | ||
c6b4fcba JT |
1421 | static void inc_hit_counter(struct cache *cache, struct bio *bio) |
1422 | { | |
1423 | atomic_inc(bio_data_dir(bio) == READ ? | |
1424 | &cache->stats.read_hit : &cache->stats.write_hit); | |
1425 | } | |
1426 | ||
1427 | static void inc_miss_counter(struct cache *cache, struct bio *bio) | |
1428 | { | |
1429 | atomic_inc(bio_data_dir(bio) == READ ? | |
1430 | &cache->stats.read_miss : &cache->stats.write_miss); | |
1431 | } | |
1432 | ||
1433 | static void process_bio(struct cache *cache, struct prealloc *structs, | |
1434 | struct bio *bio) | |
1435 | { | |
1436 | int r; | |
1437 | bool release_cell = true; | |
1438 | dm_oblock_t block = get_bio_block(cache, bio); | |
1439 | struct dm_bio_prison_cell *cell_prealloc, *old_ocell, *new_ocell; | |
1440 | struct policy_result lookup_result; | |
2ee57d58 | 1441 | bool passthrough = passthrough_mode(&cache->features); |
43c32bf2 | 1442 | bool discarded_block, can_migrate; |
c6b4fcba JT |
1443 | |
1444 | /* | |
1445 | * Check to see if that block is currently migrating. | |
1446 | */ | |
1447 | cell_prealloc = prealloc_get_cell(structs); | |
1448 | r = bio_detain(cache, block, bio, cell_prealloc, | |
1449 | (cell_free_fn) prealloc_put_cell, | |
1450 | structs, &new_ocell); | |
1451 | if (r > 0) | |
1452 | return; | |
1453 | ||
43c32bf2 JT |
1454 | discarded_block = is_discarded_oblock(cache, block); |
1455 | can_migrate = !passthrough && (discarded_block || spare_migration_bandwidth(cache)); | |
1456 | ||
c6b4fcba JT |
1457 | r = policy_map(cache->policy, block, true, can_migrate, discarded_block, |
1458 | bio, &lookup_result); | |
1459 | ||
1460 | if (r == -EWOULDBLOCK) | |
1461 | /* migration has been denied */ | |
1462 | lookup_result.op = POLICY_MISS; | |
1463 | ||
1464 | switch (lookup_result.op) { | |
1465 | case POLICY_HIT: | |
2ee57d58 JT |
1466 | if (passthrough) { |
1467 | inc_miss_counter(cache, bio); | |
c6b4fcba | 1468 | |
2ee57d58 JT |
1469 | /* |
1470 | * Passthrough always maps to the origin, | |
1471 | * invalidating any cache blocks that are written | |
1472 | * to. | |
1473 | */ | |
1474 | ||
1475 | if (bio_data_dir(bio) == WRITE) { | |
1476 | atomic_inc(&cache->stats.demotion); | |
1477 | invalidate(cache, structs, block, lookup_result.cblock, new_ocell); | |
1478 | release_cell = false; | |
1479 | ||
1480 | } else { | |
1481 | /* FIXME: factor out issue_origin() */ | |
2ee57d58 | 1482 | remap_to_origin_clear_discard(cache, bio, block); |
8c081b52 | 1483 | inc_and_issue(cache, bio, new_ocell); |
2ee57d58 JT |
1484 | } |
1485 | } else { | |
1486 | inc_hit_counter(cache, bio); | |
1487 | ||
1488 | if (bio_data_dir(bio) == WRITE && | |
1489 | writethrough_mode(&cache->features) && | |
1490 | !is_dirty(cache, lookup_result.cblock)) { | |
2ee57d58 | 1491 | remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock); |
8c081b52 JT |
1492 | inc_and_issue(cache, bio, new_ocell); |
1493 | ||
1494 | } else { | |
1495 | remap_to_cache_dirty(cache, bio, block, lookup_result.cblock); | |
1496 | inc_and_issue(cache, bio, new_ocell); | |
1497 | } | |
2ee57d58 | 1498 | } |
c6b4fcba | 1499 | |
c6b4fcba JT |
1500 | break; |
1501 | ||
1502 | case POLICY_MISS: | |
1503 | inc_miss_counter(cache, bio); | |
e2e74d61 | 1504 | remap_to_origin_clear_discard(cache, bio, block); |
8c081b52 | 1505 | inc_and_issue(cache, bio, new_ocell); |
c6b4fcba JT |
1506 | break; |
1507 | ||
1508 | case POLICY_NEW: | |
1509 | atomic_inc(&cache->stats.promotion); | |
1510 | promote(cache, structs, block, lookup_result.cblock, new_ocell); | |
1511 | release_cell = false; | |
1512 | break; | |
1513 | ||
1514 | case POLICY_REPLACE: | |
1515 | cell_prealloc = prealloc_get_cell(structs); | |
1516 | r = bio_detain(cache, lookup_result.old_oblock, bio, cell_prealloc, | |
1517 | (cell_free_fn) prealloc_put_cell, | |
1518 | structs, &old_ocell); | |
1519 | if (r > 0) { | |
1520 | /* | |
1521 | * We have to be careful to avoid lock inversion of | |
1522 | * the cells. So we back off, and wait for the | |
1523 | * old_ocell to become free. | |
1524 | */ | |
1525 | policy_force_mapping(cache->policy, block, | |
1526 | lookup_result.old_oblock); | |
1527 | atomic_inc(&cache->stats.cache_cell_clash); | |
1528 | break; | |
1529 | } | |
1530 | atomic_inc(&cache->stats.demotion); | |
1531 | atomic_inc(&cache->stats.promotion); | |
1532 | ||
1533 | demote_then_promote(cache, structs, lookup_result.old_oblock, | |
1534 | block, lookup_result.cblock, | |
1535 | old_ocell, new_ocell); | |
1536 | release_cell = false; | |
1537 | break; | |
1538 | ||
1539 | default: | |
1540 | DMERR_LIMIT("%s: erroring bio, unknown policy op: %u", __func__, | |
1541 | (unsigned) lookup_result.op); | |
1542 | bio_io_error(bio); | |
1543 | } | |
1544 | ||
1545 | if (release_cell) | |
1546 | cell_defer(cache, new_ocell, false); | |
1547 | } | |
1548 | ||
1549 | static int need_commit_due_to_time(struct cache *cache) | |
1550 | { | |
0f30af98 MS |
1551 | return !time_in_range(jiffies, cache->last_commit_jiffies, |
1552 | cache->last_commit_jiffies + COMMIT_PERIOD); | |
c6b4fcba JT |
1553 | } |
1554 | ||
1555 | static int commit_if_needed(struct cache *cache) | |
1556 | { | |
ffcbcb67 HM |
1557 | int r = 0; |
1558 | ||
1559 | if ((cache->commit_requested || need_commit_due_to_time(cache)) && | |
1560 | dm_cache_changed_this_transaction(cache->cmd)) { | |
c6b4fcba | 1561 | atomic_inc(&cache->stats.commit_count); |
c6b4fcba | 1562 | cache->commit_requested = false; |
ffcbcb67 HM |
1563 | r = dm_cache_commit(cache->cmd, false); |
1564 | cache->last_commit_jiffies = jiffies; | |
c6b4fcba JT |
1565 | } |
1566 | ||
ffcbcb67 | 1567 | return r; |
c6b4fcba JT |
1568 | } |
1569 | ||
1570 | static void process_deferred_bios(struct cache *cache) | |
1571 | { | |
1572 | unsigned long flags; | |
1573 | struct bio_list bios; | |
1574 | struct bio *bio; | |
1575 | struct prealloc structs; | |
1576 | ||
1577 | memset(&structs, 0, sizeof(structs)); | |
1578 | bio_list_init(&bios); | |
1579 | ||
1580 | spin_lock_irqsave(&cache->lock, flags); | |
1581 | bio_list_merge(&bios, &cache->deferred_bios); | |
1582 | bio_list_init(&cache->deferred_bios); | |
1583 | spin_unlock_irqrestore(&cache->lock, flags); | |
1584 | ||
1585 | while (!bio_list_empty(&bios)) { | |
1586 | /* | |
1587 | * If we've got no free migration structs, and processing | |
1588 | * this bio might require one, we pause until there are some | |
1589 | * prepared mappings to process. | |
1590 | */ | |
1591 | if (prealloc_data_structs(cache, &structs)) { | |
1592 | spin_lock_irqsave(&cache->lock, flags); | |
1593 | bio_list_merge(&cache->deferred_bios, &bios); | |
1594 | spin_unlock_irqrestore(&cache->lock, flags); | |
1595 | break; | |
1596 | } | |
1597 | ||
1598 | bio = bio_list_pop(&bios); | |
1599 | ||
1600 | if (bio->bi_rw & REQ_FLUSH) | |
1601 | process_flush_bio(cache, bio); | |
1602 | else if (bio->bi_rw & REQ_DISCARD) | |
7ae34e77 | 1603 | process_discard_bio(cache, &structs, bio); |
c6b4fcba JT |
1604 | else |
1605 | process_bio(cache, &structs, bio); | |
1606 | } | |
1607 | ||
1608 | prealloc_free_structs(cache, &structs); | |
1609 | } | |
1610 | ||
1611 | static void process_deferred_flush_bios(struct cache *cache, bool submit_bios) | |
1612 | { | |
1613 | unsigned long flags; | |
1614 | struct bio_list bios; | |
1615 | struct bio *bio; | |
1616 | ||
1617 | bio_list_init(&bios); | |
1618 | ||
1619 | spin_lock_irqsave(&cache->lock, flags); | |
1620 | bio_list_merge(&bios, &cache->deferred_flush_bios); | |
1621 | bio_list_init(&cache->deferred_flush_bios); | |
1622 | spin_unlock_irqrestore(&cache->lock, flags); | |
1623 | ||
8c081b52 JT |
1624 | /* |
1625 | * These bios have already been through inc_ds() | |
1626 | */ | |
c6b4fcba JT |
1627 | while ((bio = bio_list_pop(&bios))) |
1628 | submit_bios ? generic_make_request(bio) : bio_io_error(bio); | |
1629 | } | |
1630 | ||
e2e74d61 JT |
1631 | static void process_deferred_writethrough_bios(struct cache *cache) |
1632 | { | |
1633 | unsigned long flags; | |
1634 | struct bio_list bios; | |
1635 | struct bio *bio; | |
1636 | ||
1637 | bio_list_init(&bios); | |
1638 | ||
1639 | spin_lock_irqsave(&cache->lock, flags); | |
1640 | bio_list_merge(&bios, &cache->deferred_writethrough_bios); | |
1641 | bio_list_init(&cache->deferred_writethrough_bios); | |
1642 | spin_unlock_irqrestore(&cache->lock, flags); | |
1643 | ||
8c081b52 JT |
1644 | /* |
1645 | * These bios have already been through inc_ds() | |
1646 | */ | |
e2e74d61 JT |
1647 | while ((bio = bio_list_pop(&bios))) |
1648 | generic_make_request(bio); | |
1649 | } | |
1650 | ||
c6b4fcba JT |
1651 | static void writeback_some_dirty_blocks(struct cache *cache) |
1652 | { | |
1653 | int r = 0; | |
1654 | dm_oblock_t oblock; | |
1655 | dm_cblock_t cblock; | |
1656 | struct prealloc structs; | |
1657 | struct dm_bio_prison_cell *old_ocell; | |
1658 | ||
1659 | memset(&structs, 0, sizeof(structs)); | |
1660 | ||
1661 | while (spare_migration_bandwidth(cache)) { | |
1662 | if (prealloc_data_structs(cache, &structs)) | |
1663 | break; | |
1664 | ||
1665 | r = policy_writeback_work(cache->policy, &oblock, &cblock); | |
1666 | if (r) | |
1667 | break; | |
1668 | ||
1669 | r = get_cell(cache, oblock, &structs, &old_ocell); | |
1670 | if (r) { | |
1671 | policy_set_dirty(cache->policy, oblock); | |
1672 | break; | |
1673 | } | |
1674 | ||
1675 | writeback(cache, &structs, oblock, cblock, old_ocell); | |
1676 | } | |
1677 | ||
1678 | prealloc_free_structs(cache, &structs); | |
1679 | } | |
1680 | ||
65790ff9 JT |
1681 | /*---------------------------------------------------------------- |
1682 | * Invalidations. | |
1683 | * Dropping something from the cache *without* writing back. | |
1684 | *--------------------------------------------------------------*/ | |
1685 | ||
1686 | static void process_invalidation_request(struct cache *cache, struct invalidation_request *req) | |
1687 | { | |
1688 | int r = 0; | |
1689 | uint64_t begin = from_cblock(req->cblocks->begin); | |
1690 | uint64_t end = from_cblock(req->cblocks->end); | |
1691 | ||
1692 | while (begin != end) { | |
1693 | r = policy_remove_cblock(cache->policy, to_cblock(begin)); | |
1694 | if (!r) { | |
1695 | r = dm_cache_remove_mapping(cache->cmd, to_cblock(begin)); | |
1696 | if (r) | |
1697 | break; | |
1698 | ||
1699 | } else if (r == -ENODATA) { | |
1700 | /* harmless, already unmapped */ | |
1701 | r = 0; | |
1702 | ||
1703 | } else { | |
1704 | DMERR("policy_remove_cblock failed"); | |
1705 | break; | |
1706 | } | |
1707 | ||
1708 | begin++; | |
1709 | } | |
1710 | ||
1711 | cache->commit_requested = true; | |
1712 | ||
1713 | req->err = r; | |
1714 | atomic_set(&req->complete, 1); | |
1715 | ||
1716 | wake_up(&req->result_wait); | |
1717 | } | |
1718 | ||
1719 | static void process_invalidation_requests(struct cache *cache) | |
1720 | { | |
1721 | struct list_head list; | |
1722 | struct invalidation_request *req, *tmp; | |
1723 | ||
1724 | INIT_LIST_HEAD(&list); | |
1725 | spin_lock(&cache->invalidation_lock); | |
1726 | list_splice_init(&cache->invalidation_requests, &list); | |
1727 | spin_unlock(&cache->invalidation_lock); | |
1728 | ||
1729 | list_for_each_entry_safe (req, tmp, &list, list) | |
1730 | process_invalidation_request(cache, req); | |
1731 | } | |
1732 | ||
c6b4fcba JT |
1733 | /*---------------------------------------------------------------- |
1734 | * Main worker loop | |
1735 | *--------------------------------------------------------------*/ | |
66cb1910 | 1736 | static bool is_quiescing(struct cache *cache) |
c6b4fcba | 1737 | { |
238f8363 | 1738 | return atomic_read(&cache->quiescing); |
c6b4fcba JT |
1739 | } |
1740 | ||
66cb1910 JT |
1741 | static void ack_quiescing(struct cache *cache) |
1742 | { | |
1743 | if (is_quiescing(cache)) { | |
1744 | atomic_inc(&cache->quiescing_ack); | |
1745 | wake_up(&cache->quiescing_wait); | |
1746 | } | |
1747 | } | |
1748 | ||
1749 | static void wait_for_quiescing_ack(struct cache *cache) | |
1750 | { | |
1751 | wait_event(cache->quiescing_wait, atomic_read(&cache->quiescing_ack)); | |
1752 | } | |
1753 | ||
1754 | static void start_quiescing(struct cache *cache) | |
c6b4fcba | 1755 | { |
238f8363 | 1756 | atomic_inc(&cache->quiescing); |
66cb1910 | 1757 | wait_for_quiescing_ack(cache); |
c6b4fcba JT |
1758 | } |
1759 | ||
66cb1910 | 1760 | static void stop_quiescing(struct cache *cache) |
c6b4fcba | 1761 | { |
238f8363 | 1762 | atomic_set(&cache->quiescing, 0); |
66cb1910 | 1763 | atomic_set(&cache->quiescing_ack, 0); |
c6b4fcba JT |
1764 | } |
1765 | ||
1766 | static void wait_for_migrations(struct cache *cache) | |
1767 | { | |
1768 | wait_event(cache->migration_wait, !atomic_read(&cache->nr_migrations)); | |
1769 | } | |
1770 | ||
1771 | static void stop_worker(struct cache *cache) | |
1772 | { | |
1773 | cancel_delayed_work(&cache->waker); | |
1774 | flush_workqueue(cache->wq); | |
1775 | } | |
1776 | ||
1777 | static void requeue_deferred_io(struct cache *cache) | |
1778 | { | |
1779 | struct bio *bio; | |
1780 | struct bio_list bios; | |
1781 | ||
1782 | bio_list_init(&bios); | |
1783 | bio_list_merge(&bios, &cache->deferred_bios); | |
1784 | bio_list_init(&cache->deferred_bios); | |
1785 | ||
1786 | while ((bio = bio_list_pop(&bios))) | |
1787 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1788 | } | |
1789 | ||
1790 | static int more_work(struct cache *cache) | |
1791 | { | |
1792 | if (is_quiescing(cache)) | |
1793 | return !list_empty(&cache->quiesced_migrations) || | |
1794 | !list_empty(&cache->completed_migrations) || | |
1795 | !list_empty(&cache->need_commit_migrations); | |
1796 | else | |
1797 | return !bio_list_empty(&cache->deferred_bios) || | |
1798 | !bio_list_empty(&cache->deferred_flush_bios) || | |
e2e74d61 | 1799 | !bio_list_empty(&cache->deferred_writethrough_bios) || |
c6b4fcba JT |
1800 | !list_empty(&cache->quiesced_migrations) || |
1801 | !list_empty(&cache->completed_migrations) || | |
65790ff9 JT |
1802 | !list_empty(&cache->need_commit_migrations) || |
1803 | cache->invalidate; | |
c6b4fcba JT |
1804 | } |
1805 | ||
1806 | static void do_worker(struct work_struct *ws) | |
1807 | { | |
1808 | struct cache *cache = container_of(ws, struct cache, worker); | |
1809 | ||
1810 | do { | |
66cb1910 JT |
1811 | if (!is_quiescing(cache)) { |
1812 | writeback_some_dirty_blocks(cache); | |
1813 | process_deferred_writethrough_bios(cache); | |
c6b4fcba | 1814 | process_deferred_bios(cache); |
65790ff9 | 1815 | process_invalidation_requests(cache); |
66cb1910 | 1816 | } |
c6b4fcba | 1817 | |
7ae34e77 | 1818 | process_migrations(cache, &cache->quiesced_migrations, issue_copy_or_discard); |
c6b4fcba JT |
1819 | process_migrations(cache, &cache->completed_migrations, complete_migration); |
1820 | ||
c6b4fcba JT |
1821 | if (commit_if_needed(cache)) { |
1822 | process_deferred_flush_bios(cache, false); | |
304affaa | 1823 | process_migrations(cache, &cache->need_commit_migrations, migration_failure); |
c6b4fcba JT |
1824 | |
1825 | /* | |
1826 | * FIXME: rollback metadata or just go into a | |
1827 | * failure mode and error everything | |
1828 | */ | |
1829 | } else { | |
1830 | process_deferred_flush_bios(cache, true); | |
1831 | process_migrations(cache, &cache->need_commit_migrations, | |
1832 | migration_success_post_commit); | |
1833 | } | |
66cb1910 JT |
1834 | |
1835 | ack_quiescing(cache); | |
1836 | ||
c6b4fcba JT |
1837 | } while (more_work(cache)); |
1838 | } | |
1839 | ||
1840 | /* | |
1841 | * We want to commit periodically so that not too much | |
1842 | * unwritten metadata builds up. | |
1843 | */ | |
1844 | static void do_waker(struct work_struct *ws) | |
1845 | { | |
1846 | struct cache *cache = container_of(to_delayed_work(ws), struct cache, waker); | |
f8350daf | 1847 | policy_tick(cache->policy); |
c6b4fcba JT |
1848 | wake_worker(cache); |
1849 | queue_delayed_work(cache->wq, &cache->waker, COMMIT_PERIOD); | |
1850 | } | |
1851 | ||
1852 | /*----------------------------------------------------------------*/ | |
1853 | ||
1854 | static int is_congested(struct dm_dev *dev, int bdi_bits) | |
1855 | { | |
1856 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1857 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
1858 | } | |
1859 | ||
1860 | static int cache_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1861 | { | |
1862 | struct cache *cache = container_of(cb, struct cache, callbacks); | |
1863 | ||
1864 | return is_congested(cache->origin_dev, bdi_bits) || | |
1865 | is_congested(cache->cache_dev, bdi_bits); | |
1866 | } | |
1867 | ||
1868 | /*---------------------------------------------------------------- | |
1869 | * Target methods | |
1870 | *--------------------------------------------------------------*/ | |
1871 | ||
1872 | /* | |
1873 | * This function gets called on the error paths of the constructor, so we | |
1874 | * have to cope with a partially initialised struct. | |
1875 | */ | |
1876 | static void destroy(struct cache *cache) | |
1877 | { | |
1878 | unsigned i; | |
1879 | ||
1880 | if (cache->next_migration) | |
1881 | mempool_free(cache->next_migration, cache->migration_pool); | |
1882 | ||
1883 | if (cache->migration_pool) | |
1884 | mempool_destroy(cache->migration_pool); | |
1885 | ||
1886 | if (cache->all_io_ds) | |
1887 | dm_deferred_set_destroy(cache->all_io_ds); | |
1888 | ||
1889 | if (cache->prison) | |
1890 | dm_bio_prison_destroy(cache->prison); | |
1891 | ||
1892 | if (cache->wq) | |
1893 | destroy_workqueue(cache->wq); | |
1894 | ||
1895 | if (cache->dirty_bitset) | |
1896 | free_bitset(cache->dirty_bitset); | |
1897 | ||
1898 | if (cache->discard_bitset) | |
1899 | free_bitset(cache->discard_bitset); | |
1900 | ||
1901 | if (cache->copier) | |
1902 | dm_kcopyd_client_destroy(cache->copier); | |
1903 | ||
1904 | if (cache->cmd) | |
1905 | dm_cache_metadata_close(cache->cmd); | |
1906 | ||
1907 | if (cache->metadata_dev) | |
1908 | dm_put_device(cache->ti, cache->metadata_dev); | |
1909 | ||
1910 | if (cache->origin_dev) | |
1911 | dm_put_device(cache->ti, cache->origin_dev); | |
1912 | ||
1913 | if (cache->cache_dev) | |
1914 | dm_put_device(cache->ti, cache->cache_dev); | |
1915 | ||
1916 | if (cache->policy) | |
1917 | dm_cache_policy_destroy(cache->policy); | |
1918 | ||
1919 | for (i = 0; i < cache->nr_ctr_args ; i++) | |
1920 | kfree(cache->ctr_args[i]); | |
1921 | kfree(cache->ctr_args); | |
1922 | ||
1923 | kfree(cache); | |
1924 | } | |
1925 | ||
1926 | static void cache_dtr(struct dm_target *ti) | |
1927 | { | |
1928 | struct cache *cache = ti->private; | |
1929 | ||
1930 | destroy(cache); | |
1931 | } | |
1932 | ||
1933 | static sector_t get_dev_size(struct dm_dev *dev) | |
1934 | { | |
1935 | return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
1936 | } | |
1937 | ||
1938 | /*----------------------------------------------------------------*/ | |
1939 | ||
1940 | /* | |
1941 | * Construct a cache device mapping. | |
1942 | * | |
1943 | * cache <metadata dev> <cache dev> <origin dev> <block size> | |
1944 | * <#feature args> [<feature arg>]* | |
1945 | * <policy> <#policy args> [<policy arg>]* | |
1946 | * | |
1947 | * metadata dev : fast device holding the persistent metadata | |
1948 | * cache dev : fast device holding cached data blocks | |
1949 | * origin dev : slow device holding original data blocks | |
1950 | * block size : cache unit size in sectors | |
1951 | * | |
1952 | * #feature args : number of feature arguments passed | |
1953 | * feature args : writethrough. (The default is writeback.) | |
1954 | * | |
1955 | * policy : the replacement policy to use | |
1956 | * #policy args : an even number of policy arguments corresponding | |
1957 | * to key/value pairs passed to the policy | |
1958 | * policy args : key/value pairs passed to the policy | |
1959 | * E.g. 'sequential_threshold 1024' | |
1960 | * See cache-policies.txt for details. | |
1961 | * | |
1962 | * Optional feature arguments are: | |
1963 | * writethrough : write through caching that prohibits cache block | |
1964 | * content from being different from origin block content. | |
1965 | * Without this argument, the default behaviour is to write | |
1966 | * back cache block contents later for performance reasons, | |
1967 | * so they may differ from the corresponding origin blocks. | |
1968 | */ | |
1969 | struct cache_args { | |
1970 | struct dm_target *ti; | |
1971 | ||
1972 | struct dm_dev *metadata_dev; | |
1973 | ||
1974 | struct dm_dev *cache_dev; | |
1975 | sector_t cache_sectors; | |
1976 | ||
1977 | struct dm_dev *origin_dev; | |
1978 | sector_t origin_sectors; | |
1979 | ||
1980 | uint32_t block_size; | |
1981 | ||
1982 | const char *policy_name; | |
1983 | int policy_argc; | |
1984 | const char **policy_argv; | |
1985 | ||
1986 | struct cache_features features; | |
1987 | }; | |
1988 | ||
1989 | static void destroy_cache_args(struct cache_args *ca) | |
1990 | { | |
1991 | if (ca->metadata_dev) | |
1992 | dm_put_device(ca->ti, ca->metadata_dev); | |
1993 | ||
1994 | if (ca->cache_dev) | |
1995 | dm_put_device(ca->ti, ca->cache_dev); | |
1996 | ||
1997 | if (ca->origin_dev) | |
1998 | dm_put_device(ca->ti, ca->origin_dev); | |
1999 | ||
2000 | kfree(ca); | |
2001 | } | |
2002 | ||
2003 | static bool at_least_one_arg(struct dm_arg_set *as, char **error) | |
2004 | { | |
2005 | if (!as->argc) { | |
2006 | *error = "Insufficient args"; | |
2007 | return false; | |
2008 | } | |
2009 | ||
2010 | return true; | |
2011 | } | |
2012 | ||
2013 | static int parse_metadata_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2014 | char **error) | |
2015 | { | |
2016 | int r; | |
2017 | sector_t metadata_dev_size; | |
2018 | char b[BDEVNAME_SIZE]; | |
2019 | ||
2020 | if (!at_least_one_arg(as, error)) | |
2021 | return -EINVAL; | |
2022 | ||
2023 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2024 | &ca->metadata_dev); | |
2025 | if (r) { | |
2026 | *error = "Error opening metadata device"; | |
2027 | return r; | |
2028 | } | |
2029 | ||
2030 | metadata_dev_size = get_dev_size(ca->metadata_dev); | |
2031 | if (metadata_dev_size > DM_CACHE_METADATA_MAX_SECTORS_WARNING) | |
2032 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
2033 | bdevname(ca->metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
2034 | ||
2035 | return 0; | |
2036 | } | |
2037 | ||
2038 | static int parse_cache_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2039 | char **error) | |
2040 | { | |
2041 | int r; | |
2042 | ||
2043 | if (!at_least_one_arg(as, error)) | |
2044 | return -EINVAL; | |
2045 | ||
2046 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2047 | &ca->cache_dev); | |
2048 | if (r) { | |
2049 | *error = "Error opening cache device"; | |
2050 | return r; | |
2051 | } | |
2052 | ca->cache_sectors = get_dev_size(ca->cache_dev); | |
2053 | ||
2054 | return 0; | |
2055 | } | |
2056 | ||
2057 | static int parse_origin_dev(struct cache_args *ca, struct dm_arg_set *as, | |
2058 | char **error) | |
2059 | { | |
2060 | int r; | |
2061 | ||
2062 | if (!at_least_one_arg(as, error)) | |
2063 | return -EINVAL; | |
2064 | ||
2065 | r = dm_get_device(ca->ti, dm_shift_arg(as), FMODE_READ | FMODE_WRITE, | |
2066 | &ca->origin_dev); | |
2067 | if (r) { | |
2068 | *error = "Error opening origin device"; | |
2069 | return r; | |
2070 | } | |
2071 | ||
2072 | ca->origin_sectors = get_dev_size(ca->origin_dev); | |
2073 | if (ca->ti->len > ca->origin_sectors) { | |
2074 | *error = "Device size larger than cached device"; | |
2075 | return -EINVAL; | |
2076 | } | |
2077 | ||
2078 | return 0; | |
2079 | } | |
2080 | ||
2081 | static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as, | |
2082 | char **error) | |
2083 | { | |
05473044 | 2084 | unsigned long block_size; |
c6b4fcba JT |
2085 | |
2086 | if (!at_least_one_arg(as, error)) | |
2087 | return -EINVAL; | |
2088 | ||
05473044 MS |
2089 | if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size || |
2090 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2091 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
2092 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { | |
c6b4fcba JT |
2093 | *error = "Invalid data block size"; |
2094 | return -EINVAL; | |
2095 | } | |
2096 | ||
05473044 | 2097 | if (block_size > ca->cache_sectors) { |
c6b4fcba JT |
2098 | *error = "Data block size is larger than the cache device"; |
2099 | return -EINVAL; | |
2100 | } | |
2101 | ||
05473044 | 2102 | ca->block_size = block_size; |
c6b4fcba JT |
2103 | |
2104 | return 0; | |
2105 | } | |
2106 | ||
2107 | static void init_features(struct cache_features *cf) | |
2108 | { | |
2109 | cf->mode = CM_WRITE; | |
2ee57d58 | 2110 | cf->io_mode = CM_IO_WRITEBACK; |
c6b4fcba JT |
2111 | } |
2112 | ||
2113 | static int parse_features(struct cache_args *ca, struct dm_arg_set *as, | |
2114 | char **error) | |
2115 | { | |
2116 | static struct dm_arg _args[] = { | |
2117 | {0, 1, "Invalid number of cache feature arguments"}, | |
2118 | }; | |
2119 | ||
2120 | int r; | |
2121 | unsigned argc; | |
2122 | const char *arg; | |
2123 | struct cache_features *cf = &ca->features; | |
2124 | ||
2125 | init_features(cf); | |
2126 | ||
2127 | r = dm_read_arg_group(_args, as, &argc, error); | |
2128 | if (r) | |
2129 | return -EINVAL; | |
2130 | ||
2131 | while (argc--) { | |
2132 | arg = dm_shift_arg(as); | |
2133 | ||
2134 | if (!strcasecmp(arg, "writeback")) | |
2ee57d58 | 2135 | cf->io_mode = CM_IO_WRITEBACK; |
c6b4fcba JT |
2136 | |
2137 | else if (!strcasecmp(arg, "writethrough")) | |
2ee57d58 JT |
2138 | cf->io_mode = CM_IO_WRITETHROUGH; |
2139 | ||
2140 | else if (!strcasecmp(arg, "passthrough")) | |
2141 | cf->io_mode = CM_IO_PASSTHROUGH; | |
c6b4fcba JT |
2142 | |
2143 | else { | |
2144 | *error = "Unrecognised cache feature requested"; | |
2145 | return -EINVAL; | |
2146 | } | |
2147 | } | |
2148 | ||
2149 | return 0; | |
2150 | } | |
2151 | ||
2152 | static int parse_policy(struct cache_args *ca, struct dm_arg_set *as, | |
2153 | char **error) | |
2154 | { | |
2155 | static struct dm_arg _args[] = { | |
2156 | {0, 1024, "Invalid number of policy arguments"}, | |
2157 | }; | |
2158 | ||
2159 | int r; | |
2160 | ||
2161 | if (!at_least_one_arg(as, error)) | |
2162 | return -EINVAL; | |
2163 | ||
2164 | ca->policy_name = dm_shift_arg(as); | |
2165 | ||
2166 | r = dm_read_arg_group(_args, as, &ca->policy_argc, error); | |
2167 | if (r) | |
2168 | return -EINVAL; | |
2169 | ||
2170 | ca->policy_argv = (const char **)as->argv; | |
2171 | dm_consume_args(as, ca->policy_argc); | |
2172 | ||
2173 | return 0; | |
2174 | } | |
2175 | ||
2176 | static int parse_cache_args(struct cache_args *ca, int argc, char **argv, | |
2177 | char **error) | |
2178 | { | |
2179 | int r; | |
2180 | struct dm_arg_set as; | |
2181 | ||
2182 | as.argc = argc; | |
2183 | as.argv = argv; | |
2184 | ||
2185 | r = parse_metadata_dev(ca, &as, error); | |
2186 | if (r) | |
2187 | return r; | |
2188 | ||
2189 | r = parse_cache_dev(ca, &as, error); | |
2190 | if (r) | |
2191 | return r; | |
2192 | ||
2193 | r = parse_origin_dev(ca, &as, error); | |
2194 | if (r) | |
2195 | return r; | |
2196 | ||
2197 | r = parse_block_size(ca, &as, error); | |
2198 | if (r) | |
2199 | return r; | |
2200 | ||
2201 | r = parse_features(ca, &as, error); | |
2202 | if (r) | |
2203 | return r; | |
2204 | ||
2205 | r = parse_policy(ca, &as, error); | |
2206 | if (r) | |
2207 | return r; | |
2208 | ||
2209 | return 0; | |
2210 | } | |
2211 | ||
2212 | /*----------------------------------------------------------------*/ | |
2213 | ||
2214 | static struct kmem_cache *migration_cache; | |
2215 | ||
2c73c471 AK |
2216 | #define NOT_CORE_OPTION 1 |
2217 | ||
2f14f4b5 | 2218 | static int process_config_option(struct cache *cache, const char *key, const char *value) |
2c73c471 AK |
2219 | { |
2220 | unsigned long tmp; | |
2221 | ||
2f14f4b5 JT |
2222 | if (!strcasecmp(key, "migration_threshold")) { |
2223 | if (kstrtoul(value, 10, &tmp)) | |
2c73c471 AK |
2224 | return -EINVAL; |
2225 | ||
2226 | cache->migration_threshold = tmp; | |
2227 | return 0; | |
2228 | } | |
2229 | ||
2230 | return NOT_CORE_OPTION; | |
2231 | } | |
2232 | ||
2f14f4b5 JT |
2233 | static int set_config_value(struct cache *cache, const char *key, const char *value) |
2234 | { | |
2235 | int r = process_config_option(cache, key, value); | |
2236 | ||
2237 | if (r == NOT_CORE_OPTION) | |
2238 | r = policy_set_config_value(cache->policy, key, value); | |
2239 | ||
2240 | if (r) | |
2241 | DMWARN("bad config value for %s: %s", key, value); | |
2242 | ||
2243 | return r; | |
2244 | } | |
2245 | ||
2246 | static int set_config_values(struct cache *cache, int argc, const char **argv) | |
c6b4fcba JT |
2247 | { |
2248 | int r = 0; | |
2249 | ||
2250 | if (argc & 1) { | |
2251 | DMWARN("Odd number of policy arguments given but they should be <key> <value> pairs."); | |
2252 | return -EINVAL; | |
2253 | } | |
2254 | ||
2255 | while (argc) { | |
2f14f4b5 JT |
2256 | r = set_config_value(cache, argv[0], argv[1]); |
2257 | if (r) | |
2258 | break; | |
c6b4fcba JT |
2259 | |
2260 | argc -= 2; | |
2261 | argv += 2; | |
2262 | } | |
2263 | ||
2264 | return r; | |
2265 | } | |
2266 | ||
2267 | static int create_cache_policy(struct cache *cache, struct cache_args *ca, | |
2268 | char **error) | |
2269 | { | |
4cb3e1db MP |
2270 | struct dm_cache_policy *p = dm_cache_policy_create(ca->policy_name, |
2271 | cache->cache_size, | |
2272 | cache->origin_sectors, | |
2273 | cache->sectors_per_block); | |
2274 | if (IS_ERR(p)) { | |
c6b4fcba | 2275 | *error = "Error creating cache's policy"; |
4cb3e1db | 2276 | return PTR_ERR(p); |
c6b4fcba | 2277 | } |
4cb3e1db | 2278 | cache->policy = p; |
c6b4fcba | 2279 | |
2f14f4b5 | 2280 | return 0; |
c6b4fcba JT |
2281 | } |
2282 | ||
08b18451 | 2283 | /* |
2bb812df JT |
2284 | * We want the discard block size to be at least the size of the cache |
2285 | * block size and have no more than 2^14 discard blocks across the origin. | |
08b18451 JT |
2286 | */ |
2287 | #define MAX_DISCARD_BLOCKS (1 << 14) | |
2288 | ||
2289 | static bool too_many_discard_blocks(sector_t discard_block_size, | |
2290 | sector_t origin_size) | |
2291 | { | |
2292 | (void) sector_div(origin_size, discard_block_size); | |
2293 | ||
2294 | return origin_size > MAX_DISCARD_BLOCKS; | |
2295 | } | |
2296 | ||
2297 | static sector_t calculate_discard_block_size(sector_t cache_block_size, | |
2298 | sector_t origin_size) | |
2299 | { | |
2bb812df | 2300 | sector_t discard_block_size = cache_block_size; |
08b18451 JT |
2301 | |
2302 | if (origin_size) | |
2303 | while (too_many_discard_blocks(discard_block_size, origin_size)) | |
2304 | discard_block_size *= 2; | |
2305 | ||
2306 | return discard_block_size; | |
2307 | } | |
2308 | ||
d1d9220c JT |
2309 | static void set_cache_size(struct cache *cache, dm_cblock_t size) |
2310 | { | |
2311 | dm_block_t nr_blocks = from_cblock(size); | |
2312 | ||
2313 | if (nr_blocks > (1 << 20) && cache->cache_size != size) | |
2314 | DMWARN_LIMIT("You have created a cache device with a lot of individual cache blocks (%llu)\n" | |
2315 | "All these mappings can consume a lot of kernel memory, and take some time to read/write.\n" | |
2316 | "Please consider increasing the cache block size to reduce the overall cache block count.", | |
2317 | (unsigned long long) nr_blocks); | |
2318 | ||
2319 | cache->cache_size = size; | |
2320 | } | |
2321 | ||
f8350daf | 2322 | #define DEFAULT_MIGRATION_THRESHOLD 2048 |
c6b4fcba | 2323 | |
c6b4fcba JT |
2324 | static int cache_create(struct cache_args *ca, struct cache **result) |
2325 | { | |
2326 | int r = 0; | |
2327 | char **error = &ca->ti->error; | |
2328 | struct cache *cache; | |
2329 | struct dm_target *ti = ca->ti; | |
2330 | dm_block_t origin_blocks; | |
2331 | struct dm_cache_metadata *cmd; | |
2332 | bool may_format = ca->features.mode == CM_WRITE; | |
2333 | ||
2334 | cache = kzalloc(sizeof(*cache), GFP_KERNEL); | |
2335 | if (!cache) | |
2336 | return -ENOMEM; | |
2337 | ||
2338 | cache->ti = ca->ti; | |
2339 | ti->private = cache; | |
c6b4fcba JT |
2340 | ti->num_flush_bios = 2; |
2341 | ti->flush_supported = true; | |
2342 | ||
2343 | ti->num_discard_bios = 1; | |
2344 | ti->discards_supported = true; | |
2345 | ti->discard_zeroes_data_unsupported = true; | |
2572629a | 2346 | ti->split_discard_bios = false; |
c6b4fcba | 2347 | |
8c5008fa | 2348 | cache->features = ca->features; |
19b0092e | 2349 | ti->per_bio_data_size = get_per_bio_data_size(cache); |
c6b4fcba | 2350 | |
c6b4fcba JT |
2351 | cache->callbacks.congested_fn = cache_is_congested; |
2352 | dm_table_add_target_callbacks(ti->table, &cache->callbacks); | |
2353 | ||
2354 | cache->metadata_dev = ca->metadata_dev; | |
2355 | cache->origin_dev = ca->origin_dev; | |
2356 | cache->cache_dev = ca->cache_dev; | |
2357 | ||
2358 | ca->metadata_dev = ca->origin_dev = ca->cache_dev = NULL; | |
2359 | ||
2360 | /* FIXME: factor out this whole section */ | |
2361 | origin_blocks = cache->origin_sectors = ca->origin_sectors; | |
414dd67d | 2362 | origin_blocks = block_div(origin_blocks, ca->block_size); |
c6b4fcba JT |
2363 | cache->origin_blocks = to_oblock(origin_blocks); |
2364 | ||
2365 | cache->sectors_per_block = ca->block_size; | |
2366 | if (dm_set_target_max_io_len(ti, cache->sectors_per_block)) { | |
2367 | r = -EINVAL; | |
2368 | goto bad; | |
2369 | } | |
2370 | ||
2371 | if (ca->block_size & (ca->block_size - 1)) { | |
2372 | dm_block_t cache_size = ca->cache_sectors; | |
2373 | ||
2374 | cache->sectors_per_block_shift = -1; | |
414dd67d | 2375 | cache_size = block_div(cache_size, ca->block_size); |
d1d9220c | 2376 | set_cache_size(cache, to_cblock(cache_size)); |
c6b4fcba JT |
2377 | } else { |
2378 | cache->sectors_per_block_shift = __ffs(ca->block_size); | |
d1d9220c | 2379 | set_cache_size(cache, to_cblock(ca->cache_sectors >> cache->sectors_per_block_shift)); |
c6b4fcba JT |
2380 | } |
2381 | ||
2382 | r = create_cache_policy(cache, ca, error); | |
2383 | if (r) | |
2384 | goto bad; | |
2f14f4b5 | 2385 | |
c6b4fcba | 2386 | cache->policy_nr_args = ca->policy_argc; |
2f14f4b5 JT |
2387 | cache->migration_threshold = DEFAULT_MIGRATION_THRESHOLD; |
2388 | ||
2389 | r = set_config_values(cache, ca->policy_argc, ca->policy_argv); | |
2390 | if (r) { | |
2391 | *error = "Error setting cache policy's config values"; | |
2392 | goto bad; | |
2393 | } | |
c6b4fcba JT |
2394 | |
2395 | cmd = dm_cache_metadata_open(cache->metadata_dev->bdev, | |
2396 | ca->block_size, may_format, | |
2397 | dm_cache_policy_get_hint_size(cache->policy)); | |
2398 | if (IS_ERR(cmd)) { | |
2399 | *error = "Error creating metadata object"; | |
2400 | r = PTR_ERR(cmd); | |
2401 | goto bad; | |
2402 | } | |
2403 | cache->cmd = cmd; | |
2404 | ||
2ee57d58 JT |
2405 | if (passthrough_mode(&cache->features)) { |
2406 | bool all_clean; | |
2407 | ||
2408 | r = dm_cache_metadata_all_clean(cache->cmd, &all_clean); | |
2409 | if (r) { | |
2410 | *error = "dm_cache_metadata_all_clean() failed"; | |
2411 | goto bad; | |
2412 | } | |
2413 | ||
2414 | if (!all_clean) { | |
2415 | *error = "Cannot enter passthrough mode unless all blocks are clean"; | |
2416 | r = -EINVAL; | |
2417 | goto bad; | |
2418 | } | |
2419 | } | |
2420 | ||
c6b4fcba JT |
2421 | spin_lock_init(&cache->lock); |
2422 | bio_list_init(&cache->deferred_bios); | |
2423 | bio_list_init(&cache->deferred_flush_bios); | |
e2e74d61 | 2424 | bio_list_init(&cache->deferred_writethrough_bios); |
c6b4fcba JT |
2425 | INIT_LIST_HEAD(&cache->quiesced_migrations); |
2426 | INIT_LIST_HEAD(&cache->completed_migrations); | |
2427 | INIT_LIST_HEAD(&cache->need_commit_migrations); | |
c6b4fcba JT |
2428 | atomic_set(&cache->nr_migrations, 0); |
2429 | init_waitqueue_head(&cache->migration_wait); | |
2430 | ||
66cb1910 | 2431 | init_waitqueue_head(&cache->quiescing_wait); |
238f8363 | 2432 | atomic_set(&cache->quiescing, 0); |
66cb1910 JT |
2433 | atomic_set(&cache->quiescing_ack, 0); |
2434 | ||
fa4d683a | 2435 | r = -ENOMEM; |
44fa816b | 2436 | atomic_set(&cache->nr_dirty, 0); |
c6b4fcba JT |
2437 | cache->dirty_bitset = alloc_bitset(from_cblock(cache->cache_size)); |
2438 | if (!cache->dirty_bitset) { | |
2439 | *error = "could not allocate dirty bitset"; | |
2440 | goto bad; | |
2441 | } | |
2442 | clear_bitset(cache->dirty_bitset, from_cblock(cache->cache_size)); | |
2443 | ||
08b18451 JT |
2444 | cache->discard_block_size = |
2445 | calculate_discard_block_size(cache->sectors_per_block, | |
2446 | cache->origin_sectors); | |
2572629a JT |
2447 | cache->discard_nr_blocks = to_dblock(dm_sector_div_up(cache->origin_sectors, |
2448 | cache->discard_block_size)); | |
1bad9bc4 | 2449 | cache->discard_bitset = alloc_bitset(from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2450 | if (!cache->discard_bitset) { |
2451 | *error = "could not allocate discard bitset"; | |
2452 | goto bad; | |
2453 | } | |
1bad9bc4 | 2454 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); |
c6b4fcba JT |
2455 | |
2456 | cache->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); | |
2457 | if (IS_ERR(cache->copier)) { | |
2458 | *error = "could not create kcopyd client"; | |
2459 | r = PTR_ERR(cache->copier); | |
2460 | goto bad; | |
2461 | } | |
2462 | ||
2463 | cache->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2464 | if (!cache->wq) { | |
2465 | *error = "could not create workqueue for metadata object"; | |
2466 | goto bad; | |
2467 | } | |
2468 | INIT_WORK(&cache->worker, do_worker); | |
2469 | INIT_DELAYED_WORK(&cache->waker, do_waker); | |
2470 | cache->last_commit_jiffies = jiffies; | |
2471 | ||
a195db2d | 2472 | cache->prison = dm_bio_prison_create(); |
c6b4fcba JT |
2473 | if (!cache->prison) { |
2474 | *error = "could not create bio prison"; | |
2475 | goto bad; | |
2476 | } | |
2477 | ||
2478 | cache->all_io_ds = dm_deferred_set_create(); | |
2479 | if (!cache->all_io_ds) { | |
2480 | *error = "could not create all_io deferred set"; | |
2481 | goto bad; | |
2482 | } | |
2483 | ||
2484 | cache->migration_pool = mempool_create_slab_pool(MIGRATION_POOL_SIZE, | |
2485 | migration_cache); | |
2486 | if (!cache->migration_pool) { | |
2487 | *error = "Error creating cache's migration mempool"; | |
2488 | goto bad; | |
2489 | } | |
2490 | ||
2491 | cache->next_migration = NULL; | |
2492 | ||
2493 | cache->need_tick_bio = true; | |
2494 | cache->sized = false; | |
65790ff9 | 2495 | cache->invalidate = false; |
c6b4fcba JT |
2496 | cache->commit_requested = false; |
2497 | cache->loaded_mappings = false; | |
2498 | cache->loaded_discards = false; | |
2499 | ||
2500 | load_stats(cache); | |
2501 | ||
2502 | atomic_set(&cache->stats.demotion, 0); | |
2503 | atomic_set(&cache->stats.promotion, 0); | |
2504 | atomic_set(&cache->stats.copies_avoided, 0); | |
2505 | atomic_set(&cache->stats.cache_cell_clash, 0); | |
2506 | atomic_set(&cache->stats.commit_count, 0); | |
2507 | atomic_set(&cache->stats.discard_count, 0); | |
2508 | ||
65790ff9 JT |
2509 | spin_lock_init(&cache->invalidation_lock); |
2510 | INIT_LIST_HEAD(&cache->invalidation_requests); | |
2511 | ||
c6b4fcba JT |
2512 | *result = cache; |
2513 | return 0; | |
2514 | ||
2515 | bad: | |
2516 | destroy(cache); | |
2517 | return r; | |
2518 | } | |
2519 | ||
2520 | static int copy_ctr_args(struct cache *cache, int argc, const char **argv) | |
2521 | { | |
2522 | unsigned i; | |
2523 | const char **copy; | |
2524 | ||
2525 | copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL); | |
2526 | if (!copy) | |
2527 | return -ENOMEM; | |
2528 | for (i = 0; i < argc; i++) { | |
2529 | copy[i] = kstrdup(argv[i], GFP_KERNEL); | |
2530 | if (!copy[i]) { | |
2531 | while (i--) | |
2532 | kfree(copy[i]); | |
2533 | kfree(copy); | |
2534 | return -ENOMEM; | |
2535 | } | |
2536 | } | |
2537 | ||
2538 | cache->nr_ctr_args = argc; | |
2539 | cache->ctr_args = copy; | |
2540 | ||
2541 | return 0; | |
2542 | } | |
2543 | ||
2544 | static int cache_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2545 | { | |
2546 | int r = -EINVAL; | |
2547 | struct cache_args *ca; | |
2548 | struct cache *cache = NULL; | |
2549 | ||
2550 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
2551 | if (!ca) { | |
2552 | ti->error = "Error allocating memory for cache"; | |
2553 | return -ENOMEM; | |
2554 | } | |
2555 | ca->ti = ti; | |
2556 | ||
2557 | r = parse_cache_args(ca, argc, argv, &ti->error); | |
2558 | if (r) | |
2559 | goto out; | |
2560 | ||
2561 | r = cache_create(ca, &cache); | |
617a0b89 HM |
2562 | if (r) |
2563 | goto out; | |
c6b4fcba JT |
2564 | |
2565 | r = copy_ctr_args(cache, argc - 3, (const char **)argv + 3); | |
2566 | if (r) { | |
2567 | destroy(cache); | |
2568 | goto out; | |
2569 | } | |
2570 | ||
2571 | ti->private = cache; | |
2572 | ||
2573 | out: | |
2574 | destroy_cache_args(ca); | |
2575 | return r; | |
2576 | } | |
2577 | ||
8c081b52 | 2578 | static int __cache_map(struct cache *cache, struct bio *bio, struct dm_bio_prison_cell **cell) |
c6b4fcba | 2579 | { |
c6b4fcba JT |
2580 | int r; |
2581 | dm_oblock_t block = get_bio_block(cache, bio); | |
19b0092e | 2582 | size_t pb_data_size = get_per_bio_data_size(cache); |
c6b4fcba JT |
2583 | bool can_migrate = false; |
2584 | bool discarded_block; | |
c6b4fcba | 2585 | struct policy_result lookup_result; |
e893fba9 | 2586 | struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size); |
c6b4fcba | 2587 | |
e893fba9 | 2588 | if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) { |
c6b4fcba JT |
2589 | /* |
2590 | * This can only occur if the io goes to a partial block at | |
2591 | * the end of the origin device. We don't cache these. | |
2592 | * Just remap to the origin and carry on. | |
2593 | */ | |
e893fba9 | 2594 | remap_to_origin(cache, bio); |
c6b4fcba JT |
2595 | return DM_MAPIO_REMAPPED; |
2596 | } | |
2597 | ||
c6b4fcba JT |
2598 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_DISCARD)) { |
2599 | defer_bio(cache, bio); | |
2600 | return DM_MAPIO_SUBMITTED; | |
2601 | } | |
2602 | ||
2603 | /* | |
2604 | * Check to see if that block is currently migrating. | |
2605 | */ | |
8c081b52 JT |
2606 | *cell = alloc_prison_cell(cache); |
2607 | if (!*cell) { | |
c6b4fcba JT |
2608 | defer_bio(cache, bio); |
2609 | return DM_MAPIO_SUBMITTED; | |
2610 | } | |
2611 | ||
8c081b52 | 2612 | r = bio_detain(cache, block, bio, *cell, |
c6b4fcba | 2613 | (cell_free_fn) free_prison_cell, |
8c081b52 | 2614 | cache, cell); |
c6b4fcba JT |
2615 | if (r) { |
2616 | if (r < 0) | |
2617 | defer_bio(cache, bio); | |
2618 | ||
2619 | return DM_MAPIO_SUBMITTED; | |
2620 | } | |
2621 | ||
2622 | discarded_block = is_discarded_oblock(cache, block); | |
2623 | ||
2624 | r = policy_map(cache->policy, block, false, can_migrate, discarded_block, | |
2625 | bio, &lookup_result); | |
2626 | if (r == -EWOULDBLOCK) { | |
8c081b52 | 2627 | cell_defer(cache, *cell, true); |
c6b4fcba JT |
2628 | return DM_MAPIO_SUBMITTED; |
2629 | ||
2630 | } else if (r) { | |
2631 | DMERR_LIMIT("Unexpected return from cache replacement policy: %d", r); | |
8c081b52 | 2632 | cell_defer(cache, *cell, false); |
c6b4fcba JT |
2633 | bio_io_error(bio); |
2634 | return DM_MAPIO_SUBMITTED; | |
2635 | } | |
2636 | ||
2ee57d58 | 2637 | r = DM_MAPIO_REMAPPED; |
c6b4fcba JT |
2638 | switch (lookup_result.op) { |
2639 | case POLICY_HIT: | |
2ee57d58 JT |
2640 | if (passthrough_mode(&cache->features)) { |
2641 | if (bio_data_dir(bio) == WRITE) { | |
2642 | /* | |
2643 | * We need to invalidate this block, so | |
2644 | * defer for the worker thread. | |
2645 | */ | |
8c081b52 | 2646 | cell_defer(cache, *cell, true); |
2ee57d58 JT |
2647 | r = DM_MAPIO_SUBMITTED; |
2648 | ||
2649 | } else { | |
2ee57d58 JT |
2650 | inc_miss_counter(cache, bio); |
2651 | remap_to_origin_clear_discard(cache, bio, block); | |
2ee57d58 | 2652 | } |
c6b4fcba | 2653 | |
2ee57d58 JT |
2654 | } else { |
2655 | inc_hit_counter(cache, bio); | |
2ee57d58 JT |
2656 | if (bio_data_dir(bio) == WRITE && writethrough_mode(&cache->features) && |
2657 | !is_dirty(cache, lookup_result.cblock)) | |
2658 | remap_to_origin_then_cache(cache, bio, block, lookup_result.cblock); | |
2659 | else | |
2660 | remap_to_cache_dirty(cache, bio, block, lookup_result.cblock); | |
2ee57d58 | 2661 | } |
c6b4fcba JT |
2662 | break; |
2663 | ||
2664 | case POLICY_MISS: | |
2665 | inc_miss_counter(cache, bio); | |
c6b4fcba JT |
2666 | if (pb->req_nr != 0) { |
2667 | /* | |
2668 | * This is a duplicate writethrough io that is no | |
2669 | * longer needed because the block has been demoted. | |
2670 | */ | |
2671 | bio_endio(bio, 0); | |
8c081b52 JT |
2672 | cell_defer(cache, *cell, false); |
2673 | r = DM_MAPIO_SUBMITTED; | |
2674 | ||
2675 | } else | |
c6b4fcba | 2676 | remap_to_origin_clear_discard(cache, bio, block); |
8c081b52 | 2677 | |
c6b4fcba JT |
2678 | break; |
2679 | ||
2680 | default: | |
2681 | DMERR_LIMIT("%s: erroring bio: unknown policy op: %u", __func__, | |
2682 | (unsigned) lookup_result.op); | |
8c081b52 | 2683 | cell_defer(cache, *cell, false); |
c6b4fcba | 2684 | bio_io_error(bio); |
2ee57d58 | 2685 | r = DM_MAPIO_SUBMITTED; |
c6b4fcba JT |
2686 | } |
2687 | ||
2ee57d58 | 2688 | return r; |
c6b4fcba JT |
2689 | } |
2690 | ||
8c081b52 JT |
2691 | static int cache_map(struct dm_target *ti, struct bio *bio) |
2692 | { | |
2693 | int r; | |
f824a2af | 2694 | struct dm_bio_prison_cell *cell = NULL; |
8c081b52 JT |
2695 | struct cache *cache = ti->private; |
2696 | ||
2697 | r = __cache_map(cache, bio, &cell); | |
f824a2af | 2698 | if (r == DM_MAPIO_REMAPPED && cell) { |
8c081b52 JT |
2699 | inc_ds(cache, bio, cell); |
2700 | cell_defer(cache, cell, false); | |
2701 | } | |
2702 | ||
2703 | return r; | |
2704 | } | |
2705 | ||
c6b4fcba JT |
2706 | static int cache_end_io(struct dm_target *ti, struct bio *bio, int error) |
2707 | { | |
2708 | struct cache *cache = ti->private; | |
2709 | unsigned long flags; | |
19b0092e MS |
2710 | size_t pb_data_size = get_per_bio_data_size(cache); |
2711 | struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size); | |
c6b4fcba JT |
2712 | |
2713 | if (pb->tick) { | |
2714 | policy_tick(cache->policy); | |
2715 | ||
2716 | spin_lock_irqsave(&cache->lock, flags); | |
2717 | cache->need_tick_bio = true; | |
2718 | spin_unlock_irqrestore(&cache->lock, flags); | |
2719 | } | |
2720 | ||
2721 | check_for_quiesced_migrations(cache, pb); | |
2722 | ||
2723 | return 0; | |
2724 | } | |
2725 | ||
2726 | static int write_dirty_bitset(struct cache *cache) | |
2727 | { | |
2728 | unsigned i, r; | |
2729 | ||
2730 | for (i = 0; i < from_cblock(cache->cache_size); i++) { | |
2731 | r = dm_cache_set_dirty(cache->cmd, to_cblock(i), | |
2732 | is_dirty(cache, to_cblock(i))); | |
2733 | if (r) | |
2734 | return r; | |
2735 | } | |
2736 | ||
2737 | return 0; | |
2738 | } | |
2739 | ||
2740 | static int write_discard_bitset(struct cache *cache) | |
2741 | { | |
2742 | unsigned i, r; | |
2743 | ||
1bad9bc4 JT |
2744 | r = dm_cache_discard_bitset_resize(cache->cmd, cache->discard_block_size, |
2745 | cache->discard_nr_blocks); | |
c6b4fcba JT |
2746 | if (r) { |
2747 | DMERR("could not resize on-disk discard bitset"); | |
2748 | return r; | |
2749 | } | |
2750 | ||
1bad9bc4 JT |
2751 | for (i = 0; i < from_dblock(cache->discard_nr_blocks); i++) { |
2752 | r = dm_cache_set_discard(cache->cmd, to_dblock(i), | |
2753 | is_discarded(cache, to_dblock(i))); | |
c6b4fcba JT |
2754 | if (r) |
2755 | return r; | |
2756 | } | |
2757 | ||
2758 | return 0; | |
2759 | } | |
2760 | ||
c6b4fcba JT |
2761 | /* |
2762 | * returns true on success | |
2763 | */ | |
2764 | static bool sync_metadata(struct cache *cache) | |
2765 | { | |
2766 | int r1, r2, r3, r4; | |
2767 | ||
2768 | r1 = write_dirty_bitset(cache); | |
2769 | if (r1) | |
2770 | DMERR("could not write dirty bitset"); | |
2771 | ||
2772 | r2 = write_discard_bitset(cache); | |
2773 | if (r2) | |
2774 | DMERR("could not write discard bitset"); | |
2775 | ||
2776 | save_stats(cache); | |
2777 | ||
0596661f | 2778 | r3 = dm_cache_write_hints(cache->cmd, cache->policy); |
c6b4fcba JT |
2779 | if (r3) |
2780 | DMERR("could not write hints"); | |
2781 | ||
2782 | /* | |
2783 | * If writing the above metadata failed, we still commit, but don't | |
2784 | * set the clean shutdown flag. This will effectively force every | |
2785 | * dirty bit to be set on reload. | |
2786 | */ | |
2787 | r4 = dm_cache_commit(cache->cmd, !r1 && !r2 && !r3); | |
2788 | if (r4) | |
2789 | DMERR("could not write cache metadata. Data loss may occur."); | |
2790 | ||
2791 | return !r1 && !r2 && !r3 && !r4; | |
2792 | } | |
2793 | ||
2794 | static void cache_postsuspend(struct dm_target *ti) | |
2795 | { | |
2796 | struct cache *cache = ti->private; | |
2797 | ||
2798 | start_quiescing(cache); | |
2799 | wait_for_migrations(cache); | |
2800 | stop_worker(cache); | |
2801 | requeue_deferred_io(cache); | |
2802 | stop_quiescing(cache); | |
2803 | ||
2804 | (void) sync_metadata(cache); | |
2805 | } | |
2806 | ||
2807 | static int load_mapping(void *context, dm_oblock_t oblock, dm_cblock_t cblock, | |
2808 | bool dirty, uint32_t hint, bool hint_valid) | |
2809 | { | |
2810 | int r; | |
2811 | struct cache *cache = context; | |
2812 | ||
2813 | r = policy_load_mapping(cache->policy, oblock, cblock, hint, hint_valid); | |
2814 | if (r) | |
2815 | return r; | |
2816 | ||
2817 | if (dirty) | |
2818 | set_dirty(cache, oblock, cblock); | |
2819 | else | |
2820 | clear_dirty(cache, oblock, cblock); | |
2821 | ||
2822 | return 0; | |
2823 | } | |
2824 | ||
3e2e1c30 JT |
2825 | /* |
2826 | * The discard block size in the on disk metadata is not | |
2827 | * neccessarily the same as we're currently using. So we have to | |
2828 | * be careful to only set the discarded attribute if we know it | |
2829 | * covers a complete block of the new size. | |
2830 | */ | |
2831 | struct discard_load_info { | |
2832 | struct cache *cache; | |
2833 | ||
2834 | /* | |
2835 | * These blocks are sized using the on disk dblock size, rather | |
2836 | * than the current one. | |
2837 | */ | |
2838 | dm_block_t block_size; | |
2839 | dm_block_t discard_begin, discard_end; | |
2840 | }; | |
2841 | ||
2842 | static void discard_load_info_init(struct cache *cache, | |
2843 | struct discard_load_info *li) | |
2844 | { | |
2845 | li->cache = cache; | |
2846 | li->discard_begin = li->discard_end = 0; | |
2847 | } | |
2848 | ||
2849 | static void set_discard_range(struct discard_load_info *li) | |
2850 | { | |
2851 | sector_t b, e; | |
2852 | ||
2853 | if (li->discard_begin == li->discard_end) | |
2854 | return; | |
2855 | ||
2856 | /* | |
2857 | * Convert to sectors. | |
2858 | */ | |
2859 | b = li->discard_begin * li->block_size; | |
2860 | e = li->discard_end * li->block_size; | |
2861 | ||
2862 | /* | |
2863 | * Then convert back to the current dblock size. | |
2864 | */ | |
2865 | b = dm_sector_div_up(b, li->cache->discard_block_size); | |
2866 | sector_div(e, li->cache->discard_block_size); | |
2867 | ||
2868 | /* | |
2869 | * The origin may have shrunk, so we need to check we're still in | |
2870 | * bounds. | |
2871 | */ | |
2872 | if (e > from_dblock(li->cache->discard_nr_blocks)) | |
2873 | e = from_dblock(li->cache->discard_nr_blocks); | |
2874 | ||
2875 | for (; b < e; b++) | |
2876 | set_discard(li->cache, to_dblock(b)); | |
2877 | } | |
2878 | ||
c6b4fcba | 2879 | static int load_discard(void *context, sector_t discard_block_size, |
1bad9bc4 | 2880 | dm_dblock_t dblock, bool discard) |
c6b4fcba | 2881 | { |
3e2e1c30 | 2882 | struct discard_load_info *li = context; |
c6b4fcba | 2883 | |
3e2e1c30 | 2884 | li->block_size = discard_block_size; |
1bad9bc4 | 2885 | |
3e2e1c30 JT |
2886 | if (discard) { |
2887 | if (from_dblock(dblock) == li->discard_end) | |
2888 | /* | |
2889 | * We're already in a discard range, just extend it. | |
2890 | */ | |
2891 | li->discard_end = li->discard_end + 1ULL; | |
2892 | ||
2893 | else { | |
2894 | /* | |
2895 | * Emit the old range and start a new one. | |
2896 | */ | |
2897 | set_discard_range(li); | |
2898 | li->discard_begin = from_dblock(dblock); | |
2899 | li->discard_end = li->discard_begin + 1ULL; | |
2900 | } | |
2901 | } else { | |
2902 | set_discard_range(li); | |
2903 | li->discard_begin = li->discard_end = 0; | |
2904 | } | |
c6b4fcba JT |
2905 | |
2906 | return 0; | |
2907 | } | |
2908 | ||
f494a9c6 JT |
2909 | static dm_cblock_t get_cache_dev_size(struct cache *cache) |
2910 | { | |
2911 | sector_t size = get_dev_size(cache->cache_dev); | |
2912 | (void) sector_div(size, cache->sectors_per_block); | |
2913 | return to_cblock(size); | |
2914 | } | |
2915 | ||
2916 | static bool can_resize(struct cache *cache, dm_cblock_t new_size) | |
2917 | { | |
2918 | if (from_cblock(new_size) > from_cblock(cache->cache_size)) | |
2919 | return true; | |
2920 | ||
2921 | /* | |
2922 | * We can't drop a dirty block when shrinking the cache. | |
2923 | */ | |
2924 | while (from_cblock(new_size) < from_cblock(cache->cache_size)) { | |
2925 | new_size = to_cblock(from_cblock(new_size) + 1); | |
2926 | if (is_dirty(cache, new_size)) { | |
2927 | DMERR("unable to shrink cache; cache block %llu is dirty", | |
2928 | (unsigned long long) from_cblock(new_size)); | |
2929 | return false; | |
2930 | } | |
2931 | } | |
2932 | ||
2933 | return true; | |
2934 | } | |
2935 | ||
2936 | static int resize_cache_dev(struct cache *cache, dm_cblock_t new_size) | |
2937 | { | |
2938 | int r; | |
2939 | ||
08844800 | 2940 | r = dm_cache_resize(cache->cmd, new_size); |
f494a9c6 JT |
2941 | if (r) { |
2942 | DMERR("could not resize cache metadata"); | |
2943 | return r; | |
2944 | } | |
2945 | ||
d1d9220c | 2946 | set_cache_size(cache, new_size); |
f494a9c6 JT |
2947 | |
2948 | return 0; | |
2949 | } | |
2950 | ||
c6b4fcba JT |
2951 | static int cache_preresume(struct dm_target *ti) |
2952 | { | |
2953 | int r = 0; | |
2954 | struct cache *cache = ti->private; | |
f494a9c6 | 2955 | dm_cblock_t csize = get_cache_dev_size(cache); |
c6b4fcba JT |
2956 | |
2957 | /* | |
2958 | * Check to see if the cache has resized. | |
2959 | */ | |
f494a9c6 JT |
2960 | if (!cache->sized) { |
2961 | r = resize_cache_dev(cache, csize); | |
2962 | if (r) | |
c6b4fcba | 2963 | return r; |
c6b4fcba JT |
2964 | |
2965 | cache->sized = true; | |
f494a9c6 JT |
2966 | |
2967 | } else if (csize != cache->cache_size) { | |
2968 | if (!can_resize(cache, csize)) | |
2969 | return -EINVAL; | |
2970 | ||
2971 | r = resize_cache_dev(cache, csize); | |
2972 | if (r) | |
2973 | return r; | |
c6b4fcba JT |
2974 | } |
2975 | ||
2976 | if (!cache->loaded_mappings) { | |
ea2dd8c1 | 2977 | r = dm_cache_load_mappings(cache->cmd, cache->policy, |
c6b4fcba JT |
2978 | load_mapping, cache); |
2979 | if (r) { | |
2980 | DMERR("could not load cache mappings"); | |
2981 | return r; | |
2982 | } | |
2983 | ||
2984 | cache->loaded_mappings = true; | |
2985 | } | |
2986 | ||
2987 | if (!cache->loaded_discards) { | |
3e2e1c30 JT |
2988 | struct discard_load_info li; |
2989 | ||
2990 | /* | |
2991 | * The discard bitset could have been resized, or the | |
2992 | * discard block size changed. To be safe we start by | |
2993 | * setting every dblock to not discarded. | |
2994 | */ | |
2995 | clear_bitset(cache->discard_bitset, from_dblock(cache->discard_nr_blocks)); | |
2996 | ||
2997 | discard_load_info_init(cache, &li); | |
2998 | r = dm_cache_load_discards(cache->cmd, load_discard, &li); | |
c6b4fcba JT |
2999 | if (r) { |
3000 | DMERR("could not load origin discards"); | |
3001 | return r; | |
3002 | } | |
3e2e1c30 | 3003 | set_discard_range(&li); |
c6b4fcba JT |
3004 | |
3005 | cache->loaded_discards = true; | |
3006 | } | |
3007 | ||
3008 | return r; | |
3009 | } | |
3010 | ||
3011 | static void cache_resume(struct dm_target *ti) | |
3012 | { | |
3013 | struct cache *cache = ti->private; | |
3014 | ||
3015 | cache->need_tick_bio = true; | |
3016 | do_waker(&cache->waker.work); | |
3017 | } | |
3018 | ||
3019 | /* | |
3020 | * Status format: | |
3021 | * | |
6a388618 MS |
3022 | * <metadata block size> <#used metadata blocks>/<#total metadata blocks> |
3023 | * <cache block size> <#used cache blocks>/<#total cache blocks> | |
c6b4fcba | 3024 | * <#read hits> <#read misses> <#write hits> <#write misses> |
6a388618 | 3025 | * <#demotions> <#promotions> <#dirty> |
c6b4fcba JT |
3026 | * <#features> <features>* |
3027 | * <#core args> <core args> | |
2e68c4e6 | 3028 | * <policy name> <#policy args> <policy args>* |
c6b4fcba JT |
3029 | */ |
3030 | static void cache_status(struct dm_target *ti, status_type_t type, | |
3031 | unsigned status_flags, char *result, unsigned maxlen) | |
3032 | { | |
3033 | int r = 0; | |
3034 | unsigned i; | |
3035 | ssize_t sz = 0; | |
3036 | dm_block_t nr_free_blocks_metadata = 0; | |
3037 | dm_block_t nr_blocks_metadata = 0; | |
3038 | char buf[BDEVNAME_SIZE]; | |
3039 | struct cache *cache = ti->private; | |
3040 | dm_cblock_t residency; | |
3041 | ||
3042 | switch (type) { | |
3043 | case STATUSTYPE_INFO: | |
3044 | /* Commit to ensure statistics aren't out-of-date */ | |
3045 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) { | |
3046 | r = dm_cache_commit(cache->cmd, false); | |
3047 | if (r) | |
3048 | DMERR("could not commit metadata for accurate status"); | |
3049 | } | |
3050 | ||
3051 | r = dm_cache_get_free_metadata_block_count(cache->cmd, | |
3052 | &nr_free_blocks_metadata); | |
3053 | if (r) { | |
3054 | DMERR("could not get metadata free block count"); | |
3055 | goto err; | |
3056 | } | |
3057 | ||
3058 | r = dm_cache_get_metadata_dev_size(cache->cmd, &nr_blocks_metadata); | |
3059 | if (r) { | |
3060 | DMERR("could not get metadata device size"); | |
3061 | goto err; | |
3062 | } | |
3063 | ||
3064 | residency = policy_residency(cache->policy); | |
3065 | ||
44fa816b | 3066 | DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ", |
895b47d7 | 3067 | (unsigned)DM_CACHE_METADATA_BLOCK_SIZE, |
c6b4fcba JT |
3068 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
3069 | (unsigned long long)nr_blocks_metadata, | |
6a388618 MS |
3070 | cache->sectors_per_block, |
3071 | (unsigned long long) from_cblock(residency), | |
3072 | (unsigned long long) from_cblock(cache->cache_size), | |
c6b4fcba JT |
3073 | (unsigned) atomic_read(&cache->stats.read_hit), |
3074 | (unsigned) atomic_read(&cache->stats.read_miss), | |
3075 | (unsigned) atomic_read(&cache->stats.write_hit), | |
3076 | (unsigned) atomic_read(&cache->stats.write_miss), | |
3077 | (unsigned) atomic_read(&cache->stats.demotion), | |
3078 | (unsigned) atomic_read(&cache->stats.promotion), | |
44fa816b | 3079 | (unsigned long) atomic_read(&cache->nr_dirty)); |
c6b4fcba | 3080 | |
2ee57d58 | 3081 | if (writethrough_mode(&cache->features)) |
c6b4fcba | 3082 | DMEMIT("1 writethrough "); |
2ee57d58 JT |
3083 | |
3084 | else if (passthrough_mode(&cache->features)) | |
3085 | DMEMIT("1 passthrough "); | |
3086 | ||
3087 | else if (writeback_mode(&cache->features)) | |
3088 | DMEMIT("1 writeback "); | |
3089 | ||
3090 | else { | |
3091 | DMERR("internal error: unknown io mode: %d", (int) cache->features.io_mode); | |
3092 | goto err; | |
3093 | } | |
c6b4fcba JT |
3094 | |
3095 | DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold); | |
2e68c4e6 MS |
3096 | |
3097 | DMEMIT("%s ", dm_cache_policy_get_name(cache->policy)); | |
c6b4fcba JT |
3098 | if (sz < maxlen) { |
3099 | r = policy_emit_config_values(cache->policy, result + sz, maxlen - sz); | |
3100 | if (r) | |
3101 | DMERR("policy_emit_config_values returned %d", r); | |
3102 | } | |
3103 | ||
3104 | break; | |
3105 | ||
3106 | case STATUSTYPE_TABLE: | |
3107 | format_dev_t(buf, cache->metadata_dev->bdev->bd_dev); | |
3108 | DMEMIT("%s ", buf); | |
3109 | format_dev_t(buf, cache->cache_dev->bdev->bd_dev); | |
3110 | DMEMIT("%s ", buf); | |
3111 | format_dev_t(buf, cache->origin_dev->bdev->bd_dev); | |
3112 | DMEMIT("%s", buf); | |
3113 | ||
3114 | for (i = 0; i < cache->nr_ctr_args - 1; i++) | |
3115 | DMEMIT(" %s", cache->ctr_args[i]); | |
3116 | if (cache->nr_ctr_args) | |
3117 | DMEMIT(" %s", cache->ctr_args[cache->nr_ctr_args - 1]); | |
3118 | } | |
3119 | ||
3120 | return; | |
3121 | ||
3122 | err: | |
3123 | DMEMIT("Error"); | |
3124 | } | |
3125 | ||
c6b4fcba | 3126 | /* |
65790ff9 JT |
3127 | * A cache block range can take two forms: |
3128 | * | |
3129 | * i) A single cblock, eg. '3456' | |
3130 | * ii) A begin and end cblock with dots between, eg. 123-234 | |
3131 | */ | |
3132 | static int parse_cblock_range(struct cache *cache, const char *str, | |
3133 | struct cblock_range *result) | |
3134 | { | |
3135 | char dummy; | |
3136 | uint64_t b, e; | |
3137 | int r; | |
3138 | ||
3139 | /* | |
3140 | * Try and parse form (ii) first. | |
3141 | */ | |
3142 | r = sscanf(str, "%llu-%llu%c", &b, &e, &dummy); | |
3143 | if (r < 0) | |
3144 | return r; | |
3145 | ||
3146 | if (r == 2) { | |
3147 | result->begin = to_cblock(b); | |
3148 | result->end = to_cblock(e); | |
3149 | return 0; | |
3150 | } | |
3151 | ||
3152 | /* | |
3153 | * That didn't work, try form (i). | |
3154 | */ | |
3155 | r = sscanf(str, "%llu%c", &b, &dummy); | |
3156 | if (r < 0) | |
3157 | return r; | |
3158 | ||
3159 | if (r == 1) { | |
3160 | result->begin = to_cblock(b); | |
3161 | result->end = to_cblock(from_cblock(result->begin) + 1u); | |
3162 | return 0; | |
3163 | } | |
3164 | ||
3165 | DMERR("invalid cblock range '%s'", str); | |
3166 | return -EINVAL; | |
3167 | } | |
3168 | ||
3169 | static int validate_cblock_range(struct cache *cache, struct cblock_range *range) | |
3170 | { | |
3171 | uint64_t b = from_cblock(range->begin); | |
3172 | uint64_t e = from_cblock(range->end); | |
3173 | uint64_t n = from_cblock(cache->cache_size); | |
3174 | ||
3175 | if (b >= n) { | |
3176 | DMERR("begin cblock out of range: %llu >= %llu", b, n); | |
3177 | return -EINVAL; | |
3178 | } | |
3179 | ||
3180 | if (e > n) { | |
3181 | DMERR("end cblock out of range: %llu > %llu", e, n); | |
3182 | return -EINVAL; | |
3183 | } | |
3184 | ||
3185 | if (b >= e) { | |
3186 | DMERR("invalid cblock range: %llu >= %llu", b, e); | |
3187 | return -EINVAL; | |
3188 | } | |
3189 | ||
3190 | return 0; | |
3191 | } | |
3192 | ||
3193 | static int request_invalidation(struct cache *cache, struct cblock_range *range) | |
3194 | { | |
3195 | struct invalidation_request req; | |
3196 | ||
3197 | INIT_LIST_HEAD(&req.list); | |
3198 | req.cblocks = range; | |
3199 | atomic_set(&req.complete, 0); | |
3200 | req.err = 0; | |
3201 | init_waitqueue_head(&req.result_wait); | |
3202 | ||
3203 | spin_lock(&cache->invalidation_lock); | |
3204 | list_add(&req.list, &cache->invalidation_requests); | |
3205 | spin_unlock(&cache->invalidation_lock); | |
3206 | wake_worker(cache); | |
3207 | ||
3208 | wait_event(req.result_wait, atomic_read(&req.complete)); | |
3209 | return req.err; | |
3210 | } | |
3211 | ||
3212 | static int process_invalidate_cblocks_message(struct cache *cache, unsigned count, | |
3213 | const char **cblock_ranges) | |
3214 | { | |
3215 | int r = 0; | |
3216 | unsigned i; | |
3217 | struct cblock_range range; | |
3218 | ||
3219 | if (!passthrough_mode(&cache->features)) { | |
3220 | DMERR("cache has to be in passthrough mode for invalidation"); | |
3221 | return -EPERM; | |
3222 | } | |
3223 | ||
3224 | for (i = 0; i < count; i++) { | |
3225 | r = parse_cblock_range(cache, cblock_ranges[i], &range); | |
3226 | if (r) | |
3227 | break; | |
3228 | ||
3229 | r = validate_cblock_range(cache, &range); | |
3230 | if (r) | |
3231 | break; | |
3232 | ||
3233 | /* | |
3234 | * Pass begin and end origin blocks to the worker and wake it. | |
3235 | */ | |
3236 | r = request_invalidation(cache, &range); | |
3237 | if (r) | |
3238 | break; | |
3239 | } | |
3240 | ||
3241 | return r; | |
3242 | } | |
3243 | ||
3244 | /* | |
3245 | * Supports | |
3246 | * "<key> <value>" | |
3247 | * and | |
3248 | * "invalidate_cblocks [(<begin>)|(<begin>-<end>)]* | |
c6b4fcba JT |
3249 | * |
3250 | * The key migration_threshold is supported by the cache target core. | |
3251 | */ | |
3252 | static int cache_message(struct dm_target *ti, unsigned argc, char **argv) | |
3253 | { | |
c6b4fcba JT |
3254 | struct cache *cache = ti->private; |
3255 | ||
65790ff9 JT |
3256 | if (!argc) |
3257 | return -EINVAL; | |
3258 | ||
7b6b2bc9 | 3259 | if (!strcasecmp(argv[0], "invalidate_cblocks")) |
65790ff9 JT |
3260 | return process_invalidate_cblocks_message(cache, argc - 1, (const char **) argv + 1); |
3261 | ||
c6b4fcba JT |
3262 | if (argc != 2) |
3263 | return -EINVAL; | |
3264 | ||
2f14f4b5 | 3265 | return set_config_value(cache, argv[0], argv[1]); |
c6b4fcba JT |
3266 | } |
3267 | ||
3268 | static int cache_iterate_devices(struct dm_target *ti, | |
3269 | iterate_devices_callout_fn fn, void *data) | |
3270 | { | |
3271 | int r = 0; | |
3272 | struct cache *cache = ti->private; | |
3273 | ||
3274 | r = fn(ti, cache->cache_dev, 0, get_dev_size(cache->cache_dev), data); | |
3275 | if (!r) | |
3276 | r = fn(ti, cache->origin_dev, 0, ti->len, data); | |
3277 | ||
3278 | return r; | |
3279 | } | |
3280 | ||
3281 | /* | |
3282 | * We assume I/O is going to the origin (which is the volume | |
3283 | * more likely to have restrictions e.g. by being striped). | |
3284 | * (Looking up the exact location of the data would be expensive | |
3285 | * and could always be out of date by the time the bio is submitted.) | |
3286 | */ | |
3287 | static int cache_bvec_merge(struct dm_target *ti, | |
3288 | struct bvec_merge_data *bvm, | |
3289 | struct bio_vec *biovec, int max_size) | |
3290 | { | |
3291 | struct cache *cache = ti->private; | |
3292 | struct request_queue *q = bdev_get_queue(cache->origin_dev->bdev); | |
3293 | ||
3294 | if (!q->merge_bvec_fn) | |
3295 | return max_size; | |
3296 | ||
3297 | bvm->bi_bdev = cache->origin_dev->bdev; | |
3298 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3299 | } | |
3300 | ||
3301 | static void set_discard_limits(struct cache *cache, struct queue_limits *limits) | |
3302 | { | |
3303 | /* | |
3304 | * FIXME: these limits may be incompatible with the cache device | |
3305 | */ | |
7ae34e77 JT |
3306 | limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024, |
3307 | cache->origin_sectors); | |
1bad9bc4 | 3308 | limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT; |
c6b4fcba JT |
3309 | } |
3310 | ||
3311 | static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3312 | { | |
3313 | struct cache *cache = ti->private; | |
f6109372 | 3314 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
c6b4fcba | 3315 | |
f6109372 MS |
3316 | /* |
3317 | * If the system-determined stacked limits are compatible with the | |
3318 | * cache's blocksize (io_opt is a factor) do not override them. | |
3319 | */ | |
3320 | if (io_opt_sectors < cache->sectors_per_block || | |
3321 | do_div(io_opt_sectors, cache->sectors_per_block)) { | |
b0246530 | 3322 | blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT); |
f6109372 MS |
3323 | blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT); |
3324 | } | |
c6b4fcba JT |
3325 | set_discard_limits(cache, limits); |
3326 | } | |
3327 | ||
3328 | /*----------------------------------------------------------------*/ | |
3329 | ||
3330 | static struct target_type cache_target = { | |
3331 | .name = "cache", | |
7ae34e77 | 3332 | .version = {1, 6, 0}, |
c6b4fcba JT |
3333 | .module = THIS_MODULE, |
3334 | .ctr = cache_ctr, | |
3335 | .dtr = cache_dtr, | |
3336 | .map = cache_map, | |
3337 | .end_io = cache_end_io, | |
3338 | .postsuspend = cache_postsuspend, | |
3339 | .preresume = cache_preresume, | |
3340 | .resume = cache_resume, | |
3341 | .status = cache_status, | |
3342 | .message = cache_message, | |
3343 | .iterate_devices = cache_iterate_devices, | |
3344 | .merge = cache_bvec_merge, | |
3345 | .io_hints = cache_io_hints, | |
3346 | }; | |
3347 | ||
3348 | static int __init dm_cache_init(void) | |
3349 | { | |
3350 | int r; | |
3351 | ||
3352 | r = dm_register_target(&cache_target); | |
3353 | if (r) { | |
3354 | DMERR("cache target registration failed: %d", r); | |
3355 | return r; | |
3356 | } | |
3357 | ||
3358 | migration_cache = KMEM_CACHE(dm_cache_migration, 0); | |
3359 | if (!migration_cache) { | |
3360 | dm_unregister_target(&cache_target); | |
3361 | return -ENOMEM; | |
3362 | } | |
3363 | ||
3364 | return 0; | |
3365 | } | |
3366 | ||
3367 | static void __exit dm_cache_exit(void) | |
3368 | { | |
3369 | dm_unregister_target(&cache_target); | |
3370 | kmem_cache_destroy(migration_cache); | |
3371 | } | |
3372 | ||
3373 | module_init(dm_cache_init); | |
3374 | module_exit(dm_cache_exit); | |
3375 | ||
3376 | MODULE_DESCRIPTION(DM_NAME " cache target"); | |
3377 | MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>"); | |
3378 | MODULE_LICENSE("GPL"); |