]>
Commit | Line | Data |
---|---|---|
cafe5635 KO |
1 | /* |
2 | * bcache setup/teardown code, and some metadata io - read a superblock and | |
3 | * figure out what to do with it. | |
4 | * | |
5 | * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> | |
6 | * Copyright 2012 Google, Inc. | |
7 | */ | |
8 | ||
9 | #include "bcache.h" | |
10 | #include "btree.h" | |
11 | #include "debug.h" | |
12 | #include "request.h" | |
279afbad | 13 | #include "writeback.h" |
cafe5635 | 14 | |
c37511b8 | 15 | #include <linux/blkdev.h> |
cafe5635 KO |
16 | #include <linux/buffer_head.h> |
17 | #include <linux/debugfs.h> | |
18 | #include <linux/genhd.h> | |
79826c35 | 19 | #include <linux/kthread.h> |
cafe5635 KO |
20 | #include <linux/module.h> |
21 | #include <linux/random.h> | |
22 | #include <linux/reboot.h> | |
23 | #include <linux/sysfs.h> | |
24 | ||
25 | MODULE_LICENSE("GPL"); | |
26 | MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); | |
27 | ||
28 | static const char bcache_magic[] = { | |
29 | 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca, | |
30 | 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81 | |
31 | }; | |
32 | ||
33 | static const char invalid_uuid[] = { | |
34 | 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78, | |
35 | 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99 | |
36 | }; | |
37 | ||
38 | /* Default is -1; we skip past it for struct cached_dev's cache mode */ | |
39 | const char * const bch_cache_modes[] = { | |
40 | "default", | |
41 | "writethrough", | |
42 | "writeback", | |
43 | "writearound", | |
44 | "none", | |
45 | NULL | |
46 | }; | |
47 | ||
48 | struct uuid_entry_v0 { | |
49 | uint8_t uuid[16]; | |
50 | uint8_t label[32]; | |
51 | uint32_t first_reg; | |
52 | uint32_t last_reg; | |
53 | uint32_t invalidated; | |
54 | uint32_t pad; | |
55 | }; | |
56 | ||
57 | static struct kobject *bcache_kobj; | |
58 | struct mutex bch_register_lock; | |
59 | LIST_HEAD(bch_cache_sets); | |
60 | static LIST_HEAD(uncached_devices); | |
61 | ||
62 | static int bcache_major, bcache_minor; | |
63 | static wait_queue_head_t unregister_wait; | |
64 | struct workqueue_struct *bcache_wq; | |
65 | ||
66 | #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE) | |
67 | ||
68 | static void bio_split_pool_free(struct bio_split_pool *p) | |
69 | { | |
8ef74790 KO |
70 | if (p->bio_split_hook) |
71 | mempool_destroy(p->bio_split_hook); | |
72 | ||
cafe5635 KO |
73 | if (p->bio_split) |
74 | bioset_free(p->bio_split); | |
cafe5635 KO |
75 | } |
76 | ||
77 | static int bio_split_pool_init(struct bio_split_pool *p) | |
78 | { | |
79 | p->bio_split = bioset_create(4, 0); | |
80 | if (!p->bio_split) | |
81 | return -ENOMEM; | |
82 | ||
83 | p->bio_split_hook = mempool_create_kmalloc_pool(4, | |
84 | sizeof(struct bio_split_hook)); | |
85 | if (!p->bio_split_hook) | |
86 | return -ENOMEM; | |
87 | ||
88 | return 0; | |
89 | } | |
90 | ||
91 | /* Superblock */ | |
92 | ||
93 | static const char *read_super(struct cache_sb *sb, struct block_device *bdev, | |
94 | struct page **res) | |
95 | { | |
96 | const char *err; | |
97 | struct cache_sb *s; | |
98 | struct buffer_head *bh = __bread(bdev, 1, SB_SIZE); | |
99 | unsigned i; | |
100 | ||
101 | if (!bh) | |
102 | return "IO error"; | |
103 | ||
104 | s = (struct cache_sb *) bh->b_data; | |
105 | ||
106 | sb->offset = le64_to_cpu(s->offset); | |
107 | sb->version = le64_to_cpu(s->version); | |
108 | ||
109 | memcpy(sb->magic, s->magic, 16); | |
110 | memcpy(sb->uuid, s->uuid, 16); | |
111 | memcpy(sb->set_uuid, s->set_uuid, 16); | |
112 | memcpy(sb->label, s->label, SB_LABEL_SIZE); | |
113 | ||
114 | sb->flags = le64_to_cpu(s->flags); | |
115 | sb->seq = le64_to_cpu(s->seq); | |
cafe5635 | 116 | sb->last_mount = le32_to_cpu(s->last_mount); |
cafe5635 KO |
117 | sb->first_bucket = le16_to_cpu(s->first_bucket); |
118 | sb->keys = le16_to_cpu(s->keys); | |
119 | ||
120 | for (i = 0; i < SB_JOURNAL_BUCKETS; i++) | |
121 | sb->d[i] = le64_to_cpu(s->d[i]); | |
122 | ||
123 | pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u", | |
124 | sb->version, sb->flags, sb->seq, sb->keys); | |
125 | ||
126 | err = "Not a bcache superblock"; | |
127 | if (sb->offset != SB_SECTOR) | |
128 | goto err; | |
129 | ||
130 | if (memcmp(sb->magic, bcache_magic, 16)) | |
131 | goto err; | |
132 | ||
133 | err = "Too many journal buckets"; | |
134 | if (sb->keys > SB_JOURNAL_BUCKETS) | |
135 | goto err; | |
136 | ||
137 | err = "Bad checksum"; | |
138 | if (s->csum != csum_set(s)) | |
139 | goto err; | |
140 | ||
141 | err = "Bad UUID"; | |
169ef1cf | 142 | if (bch_is_zero(sb->uuid, 16)) |
cafe5635 KO |
143 | goto err; |
144 | ||
8abb2a5d KO |
145 | sb->block_size = le16_to_cpu(s->block_size); |
146 | ||
147 | err = "Superblock block size smaller than device block size"; | |
148 | if (sb->block_size << 9 < bdev_logical_block_size(bdev)) | |
149 | goto err; | |
150 | ||
2903381f KO |
151 | switch (sb->version) { |
152 | case BCACHE_SB_VERSION_BDEV: | |
2903381f KO |
153 | sb->data_offset = BDEV_DATA_START_DEFAULT; |
154 | break; | |
155 | case BCACHE_SB_VERSION_BDEV_WITH_OFFSET: | |
2903381f KO |
156 | sb->data_offset = le64_to_cpu(s->data_offset); |
157 | ||
158 | err = "Bad data offset"; | |
159 | if (sb->data_offset < BDEV_DATA_START_DEFAULT) | |
160 | goto err; | |
cafe5635 | 161 | |
2903381f KO |
162 | break; |
163 | case BCACHE_SB_VERSION_CDEV: | |
164 | case BCACHE_SB_VERSION_CDEV_WITH_UUID: | |
165 | sb->nbuckets = le64_to_cpu(s->nbuckets); | |
166 | sb->block_size = le16_to_cpu(s->block_size); | |
167 | sb->bucket_size = le16_to_cpu(s->bucket_size); | |
cafe5635 | 168 | |
2903381f KO |
169 | sb->nr_in_set = le16_to_cpu(s->nr_in_set); |
170 | sb->nr_this_dev = le16_to_cpu(s->nr_this_dev); | |
cafe5635 | 171 | |
2903381f KO |
172 | err = "Too many buckets"; |
173 | if (sb->nbuckets > LONG_MAX) | |
174 | goto err; | |
cafe5635 | 175 | |
2903381f KO |
176 | err = "Not enough buckets"; |
177 | if (sb->nbuckets < 1 << 7) | |
178 | goto err; | |
cafe5635 | 179 | |
2903381f KO |
180 | err = "Bad block/bucket size"; |
181 | if (!is_power_of_2(sb->block_size) || | |
182 | sb->block_size > PAGE_SECTORS || | |
183 | !is_power_of_2(sb->bucket_size) || | |
184 | sb->bucket_size < PAGE_SECTORS) | |
185 | goto err; | |
cafe5635 | 186 | |
2903381f KO |
187 | err = "Invalid superblock: device too small"; |
188 | if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets) | |
189 | goto err; | |
cafe5635 | 190 | |
2903381f KO |
191 | err = "Bad UUID"; |
192 | if (bch_is_zero(sb->set_uuid, 16)) | |
193 | goto err; | |
cafe5635 | 194 | |
2903381f KO |
195 | err = "Bad cache device number in set"; |
196 | if (!sb->nr_in_set || | |
197 | sb->nr_in_set <= sb->nr_this_dev || | |
198 | sb->nr_in_set > MAX_CACHES_PER_SET) | |
cafe5635 KO |
199 | goto err; |
200 | ||
2903381f KO |
201 | err = "Journal buckets not sequential"; |
202 | for (i = 0; i < sb->keys; i++) | |
203 | if (sb->d[i] != sb->first_bucket + i) | |
204 | goto err; | |
cafe5635 | 205 | |
2903381f KO |
206 | err = "Too many journal buckets"; |
207 | if (sb->first_bucket + sb->keys > sb->nbuckets) | |
208 | goto err; | |
209 | ||
210 | err = "Invalid superblock: first bucket comes before end of super"; | |
211 | if (sb->first_bucket * sb->bucket_size < 16) | |
212 | goto err; | |
213 | ||
214 | break; | |
215 | default: | |
216 | err = "Unsupported superblock version"; | |
cafe5635 | 217 | goto err; |
2903381f KO |
218 | } |
219 | ||
cafe5635 KO |
220 | sb->last_mount = get_seconds(); |
221 | err = NULL; | |
222 | ||
223 | get_page(bh->b_page); | |
224 | *res = bh->b_page; | |
225 | err: | |
226 | put_bh(bh); | |
227 | return err; | |
228 | } | |
229 | ||
230 | static void write_bdev_super_endio(struct bio *bio, int error) | |
231 | { | |
232 | struct cached_dev *dc = bio->bi_private; | |
233 | /* XXX: error checking */ | |
234 | ||
235 | closure_put(&dc->sb_write.cl); | |
236 | } | |
237 | ||
238 | static void __write_super(struct cache_sb *sb, struct bio *bio) | |
239 | { | |
240 | struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page); | |
241 | unsigned i; | |
242 | ||
243 | bio->bi_sector = SB_SECTOR; | |
244 | bio->bi_rw = REQ_SYNC|REQ_META; | |
245 | bio->bi_size = SB_SIZE; | |
169ef1cf | 246 | bch_bio_map(bio, NULL); |
cafe5635 KO |
247 | |
248 | out->offset = cpu_to_le64(sb->offset); | |
249 | out->version = cpu_to_le64(sb->version); | |
250 | ||
251 | memcpy(out->uuid, sb->uuid, 16); | |
252 | memcpy(out->set_uuid, sb->set_uuid, 16); | |
253 | memcpy(out->label, sb->label, SB_LABEL_SIZE); | |
254 | ||
255 | out->flags = cpu_to_le64(sb->flags); | |
256 | out->seq = cpu_to_le64(sb->seq); | |
257 | ||
258 | out->last_mount = cpu_to_le32(sb->last_mount); | |
259 | out->first_bucket = cpu_to_le16(sb->first_bucket); | |
260 | out->keys = cpu_to_le16(sb->keys); | |
261 | ||
262 | for (i = 0; i < sb->keys; i++) | |
263 | out->d[i] = cpu_to_le64(sb->d[i]); | |
264 | ||
265 | out->csum = csum_set(out); | |
266 | ||
267 | pr_debug("ver %llu, flags %llu, seq %llu", | |
268 | sb->version, sb->flags, sb->seq); | |
269 | ||
270 | submit_bio(REQ_WRITE, bio); | |
271 | } | |
272 | ||
273 | void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent) | |
274 | { | |
275 | struct closure *cl = &dc->sb_write.cl; | |
276 | struct bio *bio = &dc->sb_bio; | |
277 | ||
278 | closure_lock(&dc->sb_write, parent); | |
279 | ||
280 | bio_reset(bio); | |
281 | bio->bi_bdev = dc->bdev; | |
282 | bio->bi_end_io = write_bdev_super_endio; | |
283 | bio->bi_private = dc; | |
284 | ||
285 | closure_get(cl); | |
286 | __write_super(&dc->sb, bio); | |
287 | ||
288 | closure_return(cl); | |
289 | } | |
290 | ||
291 | static void write_super_endio(struct bio *bio, int error) | |
292 | { | |
293 | struct cache *ca = bio->bi_private; | |
294 | ||
295 | bch_count_io_errors(ca, error, "writing superblock"); | |
296 | closure_put(&ca->set->sb_write.cl); | |
297 | } | |
298 | ||
299 | void bcache_write_super(struct cache_set *c) | |
300 | { | |
301 | struct closure *cl = &c->sb_write.cl; | |
302 | struct cache *ca; | |
303 | unsigned i; | |
304 | ||
305 | closure_lock(&c->sb_write, &c->cl); | |
306 | ||
307 | c->sb.seq++; | |
308 | ||
309 | for_each_cache(ca, c, i) { | |
310 | struct bio *bio = &ca->sb_bio; | |
311 | ||
2903381f | 312 | ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID; |
cafe5635 KO |
313 | ca->sb.seq = c->sb.seq; |
314 | ca->sb.last_mount = c->sb.last_mount; | |
315 | ||
316 | SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb)); | |
317 | ||
318 | bio_reset(bio); | |
319 | bio->bi_bdev = ca->bdev; | |
320 | bio->bi_end_io = write_super_endio; | |
321 | bio->bi_private = ca; | |
322 | ||
323 | closure_get(cl); | |
324 | __write_super(&ca->sb, bio); | |
325 | } | |
326 | ||
327 | closure_return(cl); | |
328 | } | |
329 | ||
330 | /* UUID io */ | |
331 | ||
332 | static void uuid_endio(struct bio *bio, int error) | |
333 | { | |
334 | struct closure *cl = bio->bi_private; | |
335 | struct cache_set *c = container_of(cl, struct cache_set, uuid_write.cl); | |
336 | ||
337 | cache_set_err_on(error, c, "accessing uuids"); | |
338 | bch_bbio_free(bio, c); | |
339 | closure_put(cl); | |
340 | } | |
341 | ||
342 | static void uuid_io(struct cache_set *c, unsigned long rw, | |
343 | struct bkey *k, struct closure *parent) | |
344 | { | |
345 | struct closure *cl = &c->uuid_write.cl; | |
346 | struct uuid_entry *u; | |
347 | unsigned i; | |
85b1492e | 348 | char buf[80]; |
cafe5635 KO |
349 | |
350 | BUG_ON(!parent); | |
351 | closure_lock(&c->uuid_write, parent); | |
352 | ||
353 | for (i = 0; i < KEY_PTRS(k); i++) { | |
354 | struct bio *bio = bch_bbio_alloc(c); | |
355 | ||
356 | bio->bi_rw = REQ_SYNC|REQ_META|rw; | |
357 | bio->bi_size = KEY_SIZE(k) << 9; | |
358 | ||
359 | bio->bi_end_io = uuid_endio; | |
360 | bio->bi_private = cl; | |
169ef1cf | 361 | bch_bio_map(bio, c->uuids); |
cafe5635 KO |
362 | |
363 | bch_submit_bbio(bio, c, k, i); | |
364 | ||
365 | if (!(rw & WRITE)) | |
366 | break; | |
367 | } | |
368 | ||
85b1492e KO |
369 | bch_bkey_to_text(buf, sizeof(buf), k); |
370 | pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read", buf); | |
cafe5635 KO |
371 | |
372 | for (u = c->uuids; u < c->uuids + c->nr_uuids; u++) | |
169ef1cf | 373 | if (!bch_is_zero(u->uuid, 16)) |
cafe5635 KO |
374 | pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u", |
375 | u - c->uuids, u->uuid, u->label, | |
376 | u->first_reg, u->last_reg, u->invalidated); | |
377 | ||
378 | closure_return(cl); | |
379 | } | |
380 | ||
381 | static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl) | |
382 | { | |
383 | struct bkey *k = &j->uuid_bucket; | |
384 | ||
385 | if (__bch_ptr_invalid(c, 1, k)) | |
386 | return "bad uuid pointer"; | |
387 | ||
388 | bkey_copy(&c->uuid_bucket, k); | |
389 | uuid_io(c, READ_SYNC, k, cl); | |
390 | ||
391 | if (j->version < BCACHE_JSET_VERSION_UUIDv1) { | |
392 | struct uuid_entry_v0 *u0 = (void *) c->uuids; | |
393 | struct uuid_entry *u1 = (void *) c->uuids; | |
394 | int i; | |
395 | ||
396 | closure_sync(cl); | |
397 | ||
398 | /* | |
399 | * Since the new uuid entry is bigger than the old, we have to | |
400 | * convert starting at the highest memory address and work down | |
401 | * in order to do it in place | |
402 | */ | |
403 | ||
404 | for (i = c->nr_uuids - 1; | |
405 | i >= 0; | |
406 | --i) { | |
407 | memcpy(u1[i].uuid, u0[i].uuid, 16); | |
408 | memcpy(u1[i].label, u0[i].label, 32); | |
409 | ||
410 | u1[i].first_reg = u0[i].first_reg; | |
411 | u1[i].last_reg = u0[i].last_reg; | |
412 | u1[i].invalidated = u0[i].invalidated; | |
413 | ||
414 | u1[i].flags = 0; | |
415 | u1[i].sectors = 0; | |
416 | } | |
417 | } | |
418 | ||
419 | return NULL; | |
420 | } | |
421 | ||
422 | static int __uuid_write(struct cache_set *c) | |
423 | { | |
424 | BKEY_PADDED(key) k; | |
425 | struct closure cl; | |
426 | closure_init_stack(&cl); | |
427 | ||
428 | lockdep_assert_held(&bch_register_lock); | |
429 | ||
430 | if (bch_bucket_alloc_set(c, WATERMARK_METADATA, &k.key, 1, &cl)) | |
431 | return 1; | |
432 | ||
433 | SET_KEY_SIZE(&k.key, c->sb.bucket_size); | |
434 | uuid_io(c, REQ_WRITE, &k.key, &cl); | |
435 | closure_sync(&cl); | |
436 | ||
437 | bkey_copy(&c->uuid_bucket, &k.key); | |
438 | __bkey_put(c, &k.key); | |
439 | return 0; | |
440 | } | |
441 | ||
442 | int bch_uuid_write(struct cache_set *c) | |
443 | { | |
444 | int ret = __uuid_write(c); | |
445 | ||
446 | if (!ret) | |
447 | bch_journal_meta(c, NULL); | |
448 | ||
449 | return ret; | |
450 | } | |
451 | ||
452 | static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid) | |
453 | { | |
454 | struct uuid_entry *u; | |
455 | ||
456 | for (u = c->uuids; | |
457 | u < c->uuids + c->nr_uuids; u++) | |
458 | if (!memcmp(u->uuid, uuid, 16)) | |
459 | return u; | |
460 | ||
461 | return NULL; | |
462 | } | |
463 | ||
464 | static struct uuid_entry *uuid_find_empty(struct cache_set *c) | |
465 | { | |
466 | static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"; | |
467 | return uuid_find(c, zero_uuid); | |
468 | } | |
469 | ||
470 | /* | |
471 | * Bucket priorities/gens: | |
472 | * | |
473 | * For each bucket, we store on disk its | |
474 | * 8 bit gen | |
475 | * 16 bit priority | |
476 | * | |
477 | * See alloc.c for an explanation of the gen. The priority is used to implement | |
478 | * lru (and in the future other) cache replacement policies; for most purposes | |
479 | * it's just an opaque integer. | |
480 | * | |
481 | * The gens and the priorities don't have a whole lot to do with each other, and | |
482 | * it's actually the gens that must be written out at specific times - it's no | |
483 | * big deal if the priorities don't get written, if we lose them we just reuse | |
484 | * buckets in suboptimal order. | |
485 | * | |
486 | * On disk they're stored in a packed array, and in as many buckets are required | |
487 | * to fit them all. The buckets we use to store them form a list; the journal | |
488 | * header points to the first bucket, the first bucket points to the second | |
489 | * bucket, et cetera. | |
490 | * | |
491 | * This code is used by the allocation code; periodically (whenever it runs out | |
492 | * of buckets to allocate from) the allocation code will invalidate some | |
493 | * buckets, but it can't use those buckets until their new gens are safely on | |
494 | * disk. | |
495 | */ | |
496 | ||
497 | static void prio_endio(struct bio *bio, int error) | |
498 | { | |
499 | struct cache *ca = bio->bi_private; | |
500 | ||
501 | cache_set_err_on(error, ca->set, "accessing priorities"); | |
502 | bch_bbio_free(bio, ca->set); | |
503 | closure_put(&ca->prio); | |
504 | } | |
505 | ||
506 | static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw) | |
507 | { | |
508 | struct closure *cl = &ca->prio; | |
509 | struct bio *bio = bch_bbio_alloc(ca->set); | |
510 | ||
511 | closure_init_stack(cl); | |
512 | ||
513 | bio->bi_sector = bucket * ca->sb.bucket_size; | |
514 | bio->bi_bdev = ca->bdev; | |
515 | bio->bi_rw = REQ_SYNC|REQ_META|rw; | |
516 | bio->bi_size = bucket_bytes(ca); | |
517 | ||
518 | bio->bi_end_io = prio_endio; | |
519 | bio->bi_private = ca; | |
169ef1cf | 520 | bch_bio_map(bio, ca->disk_buckets); |
cafe5635 KO |
521 | |
522 | closure_bio_submit(bio, &ca->prio, ca); | |
523 | closure_sync(cl); | |
524 | } | |
525 | ||
526 | #define buckets_free(c) "free %zu, free_inc %zu, unused %zu", \ | |
527 | fifo_used(&c->free), fifo_used(&c->free_inc), fifo_used(&c->unused) | |
528 | ||
529 | void bch_prio_write(struct cache *ca) | |
530 | { | |
531 | int i; | |
532 | struct bucket *b; | |
533 | struct closure cl; | |
534 | ||
535 | closure_init_stack(&cl); | |
536 | ||
537 | lockdep_assert_held(&ca->set->bucket_lock); | |
538 | ||
539 | for (b = ca->buckets; | |
540 | b < ca->buckets + ca->sb.nbuckets; b++) | |
541 | b->disk_gen = b->gen; | |
542 | ||
543 | ca->disk_buckets->seq++; | |
544 | ||
545 | atomic_long_add(ca->sb.bucket_size * prio_buckets(ca), | |
546 | &ca->meta_sectors_written); | |
547 | ||
548 | pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free), | |
549 | fifo_used(&ca->free_inc), fifo_used(&ca->unused)); | |
cafe5635 KO |
550 | |
551 | for (i = prio_buckets(ca) - 1; i >= 0; --i) { | |
552 | long bucket; | |
553 | struct prio_set *p = ca->disk_buckets; | |
b1a67b0f KO |
554 | struct bucket_disk *d = p->data; |
555 | struct bucket_disk *end = d + prios_per_bucket(ca); | |
cafe5635 KO |
556 | |
557 | for (b = ca->buckets + i * prios_per_bucket(ca); | |
558 | b < ca->buckets + ca->sb.nbuckets && d < end; | |
559 | b++, d++) { | |
560 | d->prio = cpu_to_le16(b->prio); | |
561 | d->gen = b->gen; | |
562 | } | |
563 | ||
564 | p->next_bucket = ca->prio_buckets[i + 1]; | |
565 | p->magic = pset_magic(ca); | |
169ef1cf | 566 | p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8); |
cafe5635 KO |
567 | |
568 | bucket = bch_bucket_alloc(ca, WATERMARK_PRIO, &cl); | |
569 | BUG_ON(bucket == -1); | |
570 | ||
571 | mutex_unlock(&ca->set->bucket_lock); | |
572 | prio_io(ca, bucket, REQ_WRITE); | |
573 | mutex_lock(&ca->set->bucket_lock); | |
574 | ||
575 | ca->prio_buckets[i] = bucket; | |
576 | atomic_dec_bug(&ca->buckets[bucket].pin); | |
577 | } | |
578 | ||
579 | mutex_unlock(&ca->set->bucket_lock); | |
580 | ||
581 | bch_journal_meta(ca->set, &cl); | |
582 | closure_sync(&cl); | |
583 | ||
584 | mutex_lock(&ca->set->bucket_lock); | |
585 | ||
586 | ca->need_save_prio = 0; | |
587 | ||
588 | /* | |
589 | * Don't want the old priorities to get garbage collected until after we | |
590 | * finish writing the new ones, and they're journalled | |
591 | */ | |
592 | for (i = 0; i < prio_buckets(ca); i++) | |
593 | ca->prio_last_buckets[i] = ca->prio_buckets[i]; | |
594 | } | |
595 | ||
596 | static void prio_read(struct cache *ca, uint64_t bucket) | |
597 | { | |
598 | struct prio_set *p = ca->disk_buckets; | |
599 | struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d; | |
600 | struct bucket *b; | |
601 | unsigned bucket_nr = 0; | |
602 | ||
603 | for (b = ca->buckets; | |
604 | b < ca->buckets + ca->sb.nbuckets; | |
605 | b++, d++) { | |
606 | if (d == end) { | |
607 | ca->prio_buckets[bucket_nr] = bucket; | |
608 | ca->prio_last_buckets[bucket_nr] = bucket; | |
609 | bucket_nr++; | |
610 | ||
611 | prio_io(ca, bucket, READ_SYNC); | |
612 | ||
169ef1cf | 613 | if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8)) |
cafe5635 KO |
614 | pr_warn("bad csum reading priorities"); |
615 | ||
616 | if (p->magic != pset_magic(ca)) | |
617 | pr_warn("bad magic reading priorities"); | |
618 | ||
619 | bucket = p->next_bucket; | |
620 | d = p->data; | |
621 | } | |
622 | ||
623 | b->prio = le16_to_cpu(d->prio); | |
624 | b->gen = b->disk_gen = b->last_gc = b->gc_gen = d->gen; | |
625 | } | |
626 | } | |
627 | ||
628 | /* Bcache device */ | |
629 | ||
630 | static int open_dev(struct block_device *b, fmode_t mode) | |
631 | { | |
632 | struct bcache_device *d = b->bd_disk->private_data; | |
633 | if (atomic_read(&d->closing)) | |
634 | return -ENXIO; | |
635 | ||
636 | closure_get(&d->cl); | |
637 | return 0; | |
638 | } | |
639 | ||
867e1162 | 640 | static void release_dev(struct gendisk *b, fmode_t mode) |
cafe5635 KO |
641 | { |
642 | struct bcache_device *d = b->private_data; | |
643 | closure_put(&d->cl); | |
cafe5635 KO |
644 | } |
645 | ||
646 | static int ioctl_dev(struct block_device *b, fmode_t mode, | |
647 | unsigned int cmd, unsigned long arg) | |
648 | { | |
649 | struct bcache_device *d = b->bd_disk->private_data; | |
650 | return d->ioctl(d, mode, cmd, arg); | |
651 | } | |
652 | ||
653 | static const struct block_device_operations bcache_ops = { | |
654 | .open = open_dev, | |
655 | .release = release_dev, | |
656 | .ioctl = ioctl_dev, | |
657 | .owner = THIS_MODULE, | |
658 | }; | |
659 | ||
660 | void bcache_device_stop(struct bcache_device *d) | |
661 | { | |
662 | if (!atomic_xchg(&d->closing, 1)) | |
663 | closure_queue(&d->cl); | |
664 | } | |
665 | ||
ee668506 KO |
666 | static void bcache_device_unlink(struct bcache_device *d) |
667 | { | |
668 | unsigned i; | |
669 | struct cache *ca; | |
670 | ||
671 | sysfs_remove_link(&d->c->kobj, d->name); | |
672 | sysfs_remove_link(&d->kobj, "cache"); | |
673 | ||
674 | for_each_cache(ca, d->c, i) | |
675 | bd_unlink_disk_holder(ca->bdev, d->disk); | |
676 | } | |
677 | ||
678 | static void bcache_device_link(struct bcache_device *d, struct cache_set *c, | |
679 | const char *name) | |
680 | { | |
681 | unsigned i; | |
682 | struct cache *ca; | |
683 | ||
684 | for_each_cache(ca, d->c, i) | |
685 | bd_link_disk_holder(ca->bdev, d->disk); | |
686 | ||
687 | snprintf(d->name, BCACHEDEVNAME_SIZE, | |
688 | "%s%u", name, d->id); | |
689 | ||
690 | WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") || | |
691 | sysfs_create_link(&c->kobj, &d->kobj, d->name), | |
692 | "Couldn't create device <-> cache set symlinks"); | |
693 | } | |
694 | ||
cafe5635 KO |
695 | static void bcache_device_detach(struct bcache_device *d) |
696 | { | |
697 | lockdep_assert_held(&bch_register_lock); | |
698 | ||
699 | if (atomic_read(&d->detaching)) { | |
700 | struct uuid_entry *u = d->c->uuids + d->id; | |
701 | ||
702 | SET_UUID_FLASH_ONLY(u, 0); | |
703 | memcpy(u->uuid, invalid_uuid, 16); | |
704 | u->invalidated = cpu_to_le32(get_seconds()); | |
705 | bch_uuid_write(d->c); | |
706 | ||
707 | atomic_set(&d->detaching, 0); | |
708 | } | |
709 | ||
c9502ea4 KO |
710 | if (!d->flush_done) |
711 | bcache_device_unlink(d); | |
ee668506 | 712 | |
cafe5635 KO |
713 | d->c->devices[d->id] = NULL; |
714 | closure_put(&d->c->caching); | |
715 | d->c = NULL; | |
716 | } | |
717 | ||
718 | static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, | |
719 | unsigned id) | |
720 | { | |
721 | BUG_ON(test_bit(CACHE_SET_STOPPING, &c->flags)); | |
722 | ||
723 | d->id = id; | |
724 | d->c = c; | |
725 | c->devices[id] = d; | |
726 | ||
727 | closure_get(&c->caching); | |
728 | } | |
729 | ||
cafe5635 KO |
730 | static void bcache_device_free(struct bcache_device *d) |
731 | { | |
732 | lockdep_assert_held(&bch_register_lock); | |
733 | ||
734 | pr_info("%s stopped", d->disk->disk_name); | |
735 | ||
736 | if (d->c) | |
737 | bcache_device_detach(d); | |
f59fce84 | 738 | if (d->disk && d->disk->flags & GENHD_FL_UP) |
cafe5635 KO |
739 | del_gendisk(d->disk); |
740 | if (d->disk && d->disk->queue) | |
741 | blk_cleanup_queue(d->disk->queue); | |
742 | if (d->disk) | |
743 | put_disk(d->disk); | |
744 | ||
745 | bio_split_pool_free(&d->bio_split_hook); | |
746 | if (d->unaligned_bvec) | |
747 | mempool_destroy(d->unaligned_bvec); | |
748 | if (d->bio_split) | |
749 | bioset_free(d->bio_split); | |
279afbad KO |
750 | if (is_vmalloc_addr(d->stripe_sectors_dirty)) |
751 | vfree(d->stripe_sectors_dirty); | |
752 | else | |
753 | kfree(d->stripe_sectors_dirty); | |
cafe5635 KO |
754 | |
755 | closure_debug_destroy(&d->cl); | |
756 | } | |
757 | ||
279afbad KO |
758 | static int bcache_device_init(struct bcache_device *d, unsigned block_size, |
759 | sector_t sectors) | |
cafe5635 KO |
760 | { |
761 | struct request_queue *q; | |
279afbad KO |
762 | size_t n; |
763 | ||
764 | if (!d->stripe_size_bits) | |
765 | d->stripe_size_bits = 31; | |
766 | ||
767 | d->nr_stripes = round_up(sectors, 1 << d->stripe_size_bits) >> | |
768 | d->stripe_size_bits; | |
769 | ||
770 | if (!d->nr_stripes || d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) | |
771 | return -ENOMEM; | |
772 | ||
773 | n = d->nr_stripes * sizeof(atomic_t); | |
774 | d->stripe_sectors_dirty = n < PAGE_SIZE << 6 | |
775 | ? kzalloc(n, GFP_KERNEL) | |
776 | : vzalloc(n); | |
777 | if (!d->stripe_sectors_dirty) | |
778 | return -ENOMEM; | |
cafe5635 KO |
779 | |
780 | if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || | |
781 | !(d->unaligned_bvec = mempool_create_kmalloc_pool(1, | |
782 | sizeof(struct bio_vec) * BIO_MAX_PAGES)) || | |
f59fce84 KO |
783 | bio_split_pool_init(&d->bio_split_hook) || |
784 | !(d->disk = alloc_disk(1)) || | |
785 | !(q = blk_alloc_queue(GFP_KERNEL))) | |
cafe5635 KO |
786 | return -ENOMEM; |
787 | ||
279afbad | 788 | set_capacity(d->disk, sectors); |
cafe5635 KO |
789 | snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", bcache_minor); |
790 | ||
791 | d->disk->major = bcache_major; | |
792 | d->disk->first_minor = bcache_minor++; | |
793 | d->disk->fops = &bcache_ops; | |
794 | d->disk->private_data = d; | |
795 | ||
cafe5635 KO |
796 | blk_queue_make_request(q, NULL); |
797 | d->disk->queue = q; | |
798 | q->queuedata = d; | |
799 | q->backing_dev_info.congested_data = d; | |
800 | q->limits.max_hw_sectors = UINT_MAX; | |
801 | q->limits.max_sectors = UINT_MAX; | |
802 | q->limits.max_segment_size = UINT_MAX; | |
803 | q->limits.max_segments = BIO_MAX_PAGES; | |
804 | q->limits.max_discard_sectors = UINT_MAX; | |
805 | q->limits.io_min = block_size; | |
806 | q->limits.logical_block_size = block_size; | |
807 | q->limits.physical_block_size = block_size; | |
808 | set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags); | |
809 | set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags); | |
810 | ||
54d12f2b KO |
811 | blk_queue_flush(q, REQ_FLUSH|REQ_FUA); |
812 | ||
cafe5635 KO |
813 | return 0; |
814 | } | |
815 | ||
816 | /* Cached device */ | |
817 | ||
818 | static void calc_cached_dev_sectors(struct cache_set *c) | |
819 | { | |
820 | uint64_t sectors = 0; | |
821 | struct cached_dev *dc; | |
822 | ||
823 | list_for_each_entry(dc, &c->cached_devs, list) | |
824 | sectors += bdev_sectors(dc->bdev); | |
825 | ||
826 | c->cached_dev_sectors = sectors; | |
827 | } | |
828 | ||
829 | void bch_cached_dev_run(struct cached_dev *dc) | |
830 | { | |
831 | struct bcache_device *d = &dc->disk; | |
ab9e1400 | 832 | char buf[SB_LABEL_SIZE + 1]; |
a25c32be GP |
833 | char *env[] = { |
834 | "DRIVER=bcache", | |
835 | kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), | |
ab9e1400 GP |
836 | NULL, |
837 | NULL, | |
a25c32be | 838 | }; |
cafe5635 | 839 | |
ab9e1400 GP |
840 | memcpy(buf, dc->sb.label, SB_LABEL_SIZE); |
841 | buf[SB_LABEL_SIZE] = '\0'; | |
842 | env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf); | |
843 | ||
cafe5635 KO |
844 | if (atomic_xchg(&dc->running, 1)) |
845 | return; | |
846 | ||
847 | if (!d->c && | |
848 | BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { | |
849 | struct closure cl; | |
850 | closure_init_stack(&cl); | |
851 | ||
852 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); | |
853 | bch_write_bdev_super(dc, &cl); | |
854 | closure_sync(&cl); | |
855 | } | |
856 | ||
857 | add_disk(d->disk); | |
ee668506 | 858 | bd_link_disk_holder(dc->bdev, dc->disk.disk); |
a25c32be GP |
859 | /* won't show up in the uevent file, use udevadm monitor -e instead |
860 | * only class / kset properties are persistent */ | |
cafe5635 | 861 | kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); |
a25c32be | 862 | kfree(env[1]); |
ab9e1400 | 863 | kfree(env[2]); |
a25c32be | 864 | |
cafe5635 KO |
865 | if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || |
866 | sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache")) | |
867 | pr_debug("error creating sysfs link"); | |
868 | } | |
869 | ||
870 | static void cached_dev_detach_finish(struct work_struct *w) | |
871 | { | |
872 | struct cached_dev *dc = container_of(w, struct cached_dev, detach); | |
873 | char buf[BDEVNAME_SIZE]; | |
874 | struct closure cl; | |
875 | closure_init_stack(&cl); | |
876 | ||
877 | BUG_ON(!atomic_read(&dc->disk.detaching)); | |
878 | BUG_ON(atomic_read(&dc->count)); | |
879 | ||
cafe5635 KO |
880 | mutex_lock(&bch_register_lock); |
881 | ||
882 | memset(&dc->sb.set_uuid, 0, 16); | |
883 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE); | |
884 | ||
885 | bch_write_bdev_super(dc, &cl); | |
886 | closure_sync(&cl); | |
887 | ||
888 | bcache_device_detach(&dc->disk); | |
889 | list_move(&dc->list, &uncached_devices); | |
890 | ||
891 | mutex_unlock(&bch_register_lock); | |
892 | ||
893 | pr_info("Caching disabled for %s", bdevname(dc->bdev, buf)); | |
894 | ||
895 | /* Drop ref we took in cached_dev_detach() */ | |
896 | closure_put(&dc->disk.cl); | |
897 | } | |
898 | ||
899 | void bch_cached_dev_detach(struct cached_dev *dc) | |
900 | { | |
901 | lockdep_assert_held(&bch_register_lock); | |
902 | ||
903 | if (atomic_read(&dc->disk.closing)) | |
904 | return; | |
905 | ||
906 | if (atomic_xchg(&dc->disk.detaching, 1)) | |
907 | return; | |
908 | ||
909 | /* | |
910 | * Block the device from being closed and freed until we're finished | |
911 | * detaching | |
912 | */ | |
913 | closure_get(&dc->disk.cl); | |
914 | ||
915 | bch_writeback_queue(dc); | |
916 | cached_dev_put(dc); | |
917 | } | |
918 | ||
919 | int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c) | |
920 | { | |
921 | uint32_t rtime = cpu_to_le32(get_seconds()); | |
922 | struct uuid_entry *u; | |
923 | char buf[BDEVNAME_SIZE]; | |
924 | ||
925 | bdevname(dc->bdev, buf); | |
926 | ||
927 | if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)) | |
928 | return -ENOENT; | |
929 | ||
930 | if (dc->disk.c) { | |
931 | pr_err("Can't attach %s: already attached", buf); | |
932 | return -EINVAL; | |
933 | } | |
934 | ||
935 | if (test_bit(CACHE_SET_STOPPING, &c->flags)) { | |
936 | pr_err("Can't attach %s: shutting down", buf); | |
937 | return -EINVAL; | |
938 | } | |
939 | ||
940 | if (dc->sb.block_size < c->sb.block_size) { | |
941 | /* Will die */ | |
b1a67b0f KO |
942 | pr_err("Couldn't attach %s: block size less than set's block size", |
943 | buf); | |
cafe5635 KO |
944 | return -EINVAL; |
945 | } | |
946 | ||
947 | u = uuid_find(c, dc->sb.uuid); | |
948 | ||
949 | if (u && | |
950 | (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || | |
951 | BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { | |
952 | memcpy(u->uuid, invalid_uuid, 16); | |
953 | u->invalidated = cpu_to_le32(get_seconds()); | |
954 | u = NULL; | |
955 | } | |
956 | ||
957 | if (!u) { | |
958 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { | |
959 | pr_err("Couldn't find uuid for %s in set", buf); | |
960 | return -ENOENT; | |
961 | } | |
962 | ||
963 | u = uuid_find_empty(c); | |
964 | if (!u) { | |
965 | pr_err("Not caching %s, no room for UUID", buf); | |
966 | return -EINVAL; | |
967 | } | |
968 | } | |
969 | ||
970 | /* Deadlocks since we're called via sysfs... | |
971 | sysfs_remove_file(&dc->kobj, &sysfs_attach); | |
972 | */ | |
973 | ||
169ef1cf | 974 | if (bch_is_zero(u->uuid, 16)) { |
cafe5635 KO |
975 | struct closure cl; |
976 | closure_init_stack(&cl); | |
977 | ||
978 | memcpy(u->uuid, dc->sb.uuid, 16); | |
979 | memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); | |
980 | u->first_reg = u->last_reg = rtime; | |
981 | bch_uuid_write(c); | |
982 | ||
983 | memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16); | |
984 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); | |
985 | ||
986 | bch_write_bdev_super(dc, &cl); | |
987 | closure_sync(&cl); | |
988 | } else { | |
989 | u->last_reg = rtime; | |
990 | bch_uuid_write(c); | |
991 | } | |
992 | ||
993 | bcache_device_attach(&dc->disk, c, u - c->uuids); | |
cafe5635 KO |
994 | list_move(&dc->list, &c->cached_devs); |
995 | calc_cached_dev_sectors(c); | |
996 | ||
997 | smp_wmb(); | |
998 | /* | |
999 | * dc->c must be set before dc->count != 0 - paired with the mb in | |
1000 | * cached_dev_get() | |
1001 | */ | |
1002 | atomic_set(&dc->count, 1); | |
1003 | ||
1004 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { | |
444fc0b6 | 1005 | bch_sectors_dirty_init(dc); |
cafe5635 KO |
1006 | atomic_set(&dc->has_dirty, 1); |
1007 | atomic_inc(&dc->count); | |
1008 | bch_writeback_queue(dc); | |
1009 | } | |
1010 | ||
1011 | bch_cached_dev_run(dc); | |
ee668506 | 1012 | bcache_device_link(&dc->disk, c, "bdev"); |
cafe5635 KO |
1013 | |
1014 | pr_info("Caching %s as %s on set %pU", | |
1015 | bdevname(dc->bdev, buf), dc->disk.disk->disk_name, | |
1016 | dc->disk.c->sb.set_uuid); | |
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | void bch_cached_dev_release(struct kobject *kobj) | |
1021 | { | |
1022 | struct cached_dev *dc = container_of(kobj, struct cached_dev, | |
1023 | disk.kobj); | |
1024 | kfree(dc); | |
1025 | module_put(THIS_MODULE); | |
1026 | } | |
1027 | ||
1028 | static void cached_dev_free(struct closure *cl) | |
1029 | { | |
1030 | struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); | |
1031 | ||
1032 | cancel_delayed_work_sync(&dc->writeback_rate_update); | |
1033 | ||
1034 | mutex_lock(&bch_register_lock); | |
1035 | ||
f59fce84 KO |
1036 | if (atomic_read(&dc->running)) |
1037 | bd_unlink_disk_holder(dc->bdev, dc->disk.disk); | |
cafe5635 KO |
1038 | bcache_device_free(&dc->disk); |
1039 | list_del(&dc->list); | |
1040 | ||
1041 | mutex_unlock(&bch_register_lock); | |
1042 | ||
1043 | if (!IS_ERR_OR_NULL(dc->bdev)) { | |
f59fce84 KO |
1044 | if (dc->bdev->bd_disk) |
1045 | blk_sync_queue(bdev_get_queue(dc->bdev)); | |
1046 | ||
cafe5635 KO |
1047 | blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1048 | } | |
1049 | ||
1050 | wake_up(&unregister_wait); | |
1051 | ||
1052 | kobject_put(&dc->disk.kobj); | |
1053 | } | |
1054 | ||
1055 | static void cached_dev_flush(struct closure *cl) | |
1056 | { | |
1057 | struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); | |
1058 | struct bcache_device *d = &dc->disk; | |
1059 | ||
c9502ea4 KO |
1060 | mutex_lock(&bch_register_lock); |
1061 | d->flush_done = 1; | |
1062 | ||
1063 | if (d->c) | |
1064 | bcache_device_unlink(d); | |
1065 | ||
1066 | mutex_unlock(&bch_register_lock); | |
1067 | ||
cafe5635 KO |
1068 | bch_cache_accounting_destroy(&dc->accounting); |
1069 | kobject_del(&d->kobj); | |
1070 | ||
1071 | continue_at(cl, cached_dev_free, system_wq); | |
1072 | } | |
1073 | ||
1074 | static int cached_dev_init(struct cached_dev *dc, unsigned block_size) | |
1075 | { | |
f59fce84 | 1076 | int ret; |
cafe5635 | 1077 | struct io *io; |
f59fce84 | 1078 | struct request_queue *q = bdev_get_queue(dc->bdev); |
cafe5635 KO |
1079 | |
1080 | __module_get(THIS_MODULE); | |
1081 | INIT_LIST_HEAD(&dc->list); | |
f59fce84 KO |
1082 | closure_init(&dc->disk.cl, NULL); |
1083 | set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); | |
cafe5635 | 1084 | kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); |
cafe5635 | 1085 | INIT_WORK(&dc->detach, cached_dev_detach_finish); |
f59fce84 KO |
1086 | closure_init_unlocked(&dc->sb_write); |
1087 | INIT_LIST_HEAD(&dc->io_lru); | |
1088 | spin_lock_init(&dc->io_lock); | |
1089 | bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); | |
cafe5635 KO |
1090 | |
1091 | dc->sequential_merge = true; | |
1092 | dc->sequential_cutoff = 4 << 20; | |
1093 | ||
cafe5635 KO |
1094 | for (io = dc->io; io < dc->io + RECENT_IO; io++) { |
1095 | list_add(&io->lru, &dc->io_lru); | |
1096 | hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); | |
1097 | } | |
1098 | ||
279afbad KO |
1099 | ret = bcache_device_init(&dc->disk, block_size, |
1100 | dc->bdev->bd_part->nr_sects - dc->sb.data_offset); | |
f59fce84 KO |
1101 | if (ret) |
1102 | return ret; | |
1103 | ||
1104 | set_capacity(dc->disk.disk, | |
1105 | dc->bdev->bd_part->nr_sects - dc->sb.data_offset); | |
1106 | ||
1107 | dc->disk.disk->queue->backing_dev_info.ra_pages = | |
1108 | max(dc->disk.disk->queue->backing_dev_info.ra_pages, | |
1109 | q->backing_dev_info.ra_pages); | |
1110 | ||
1111 | bch_cached_dev_request_init(dc); | |
1112 | bch_cached_dev_writeback_init(dc); | |
cafe5635 | 1113 | return 0; |
cafe5635 KO |
1114 | } |
1115 | ||
1116 | /* Cached device - bcache superblock */ | |
1117 | ||
f59fce84 | 1118 | static void register_bdev(struct cache_sb *sb, struct page *sb_page, |
cafe5635 KO |
1119 | struct block_device *bdev, |
1120 | struct cached_dev *dc) | |
1121 | { | |
1122 | char name[BDEVNAME_SIZE]; | |
1123 | const char *err = "cannot allocate memory"; | |
cafe5635 KO |
1124 | struct cache_set *c; |
1125 | ||
cafe5635 | 1126 | memcpy(&dc->sb, sb, sizeof(struct cache_sb)); |
cafe5635 KO |
1127 | dc->bdev = bdev; |
1128 | dc->bdev->bd_holder = dc; | |
1129 | ||
f59fce84 KO |
1130 | bio_init(&dc->sb_bio); |
1131 | dc->sb_bio.bi_max_vecs = 1; | |
1132 | dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs; | |
1133 | dc->sb_bio.bi_io_vec[0].bv_page = sb_page; | |
1134 | get_page(sb_page); | |
4f0fd955 | 1135 | |
f59fce84 KO |
1136 | if (cached_dev_init(dc, sb->block_size << 9)) |
1137 | goto err; | |
cafe5635 KO |
1138 | |
1139 | err = "error creating kobject"; | |
1140 | if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj, | |
1141 | "bcache")) | |
1142 | goto err; | |
1143 | if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) | |
1144 | goto err; | |
1145 | ||
f59fce84 KO |
1146 | pr_info("registered backing device %s", bdevname(bdev, name)); |
1147 | ||
cafe5635 KO |
1148 | list_add(&dc->list, &uncached_devices); |
1149 | list_for_each_entry(c, &bch_cache_sets, list) | |
1150 | bch_cached_dev_attach(dc, c); | |
1151 | ||
1152 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || | |
1153 | BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) | |
1154 | bch_cached_dev_run(dc); | |
1155 | ||
f59fce84 | 1156 | return; |
cafe5635 | 1157 | err: |
cafe5635 | 1158 | pr_notice("error opening %s: %s", bdevname(bdev, name), err); |
f59fce84 | 1159 | bcache_device_stop(&dc->disk); |
cafe5635 KO |
1160 | } |
1161 | ||
1162 | /* Flash only volumes */ | |
1163 | ||
1164 | void bch_flash_dev_release(struct kobject *kobj) | |
1165 | { | |
1166 | struct bcache_device *d = container_of(kobj, struct bcache_device, | |
1167 | kobj); | |
1168 | kfree(d); | |
1169 | } | |
1170 | ||
1171 | static void flash_dev_free(struct closure *cl) | |
1172 | { | |
1173 | struct bcache_device *d = container_of(cl, struct bcache_device, cl); | |
1174 | bcache_device_free(d); | |
1175 | kobject_put(&d->kobj); | |
1176 | } | |
1177 | ||
1178 | static void flash_dev_flush(struct closure *cl) | |
1179 | { | |
1180 | struct bcache_device *d = container_of(cl, struct bcache_device, cl); | |
1181 | ||
ee668506 | 1182 | bcache_device_unlink(d); |
cafe5635 KO |
1183 | kobject_del(&d->kobj); |
1184 | continue_at(cl, flash_dev_free, system_wq); | |
1185 | } | |
1186 | ||
1187 | static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) | |
1188 | { | |
1189 | struct bcache_device *d = kzalloc(sizeof(struct bcache_device), | |
1190 | GFP_KERNEL); | |
1191 | if (!d) | |
1192 | return -ENOMEM; | |
1193 | ||
1194 | closure_init(&d->cl, NULL); | |
1195 | set_closure_fn(&d->cl, flash_dev_flush, system_wq); | |
1196 | ||
1197 | kobject_init(&d->kobj, &bch_flash_dev_ktype); | |
1198 | ||
279afbad | 1199 | if (bcache_device_init(d, block_bytes(c), u->sectors)) |
cafe5635 KO |
1200 | goto err; |
1201 | ||
1202 | bcache_device_attach(d, c, u - c->uuids); | |
cafe5635 KO |
1203 | bch_flash_dev_request_init(d); |
1204 | add_disk(d->disk); | |
1205 | ||
1206 | if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache")) | |
1207 | goto err; | |
1208 | ||
1209 | bcache_device_link(d, c, "volume"); | |
1210 | ||
1211 | return 0; | |
1212 | err: | |
1213 | kobject_put(&d->kobj); | |
1214 | return -ENOMEM; | |
1215 | } | |
1216 | ||
1217 | static int flash_devs_run(struct cache_set *c) | |
1218 | { | |
1219 | int ret = 0; | |
1220 | struct uuid_entry *u; | |
1221 | ||
1222 | for (u = c->uuids; | |
1223 | u < c->uuids + c->nr_uuids && !ret; | |
1224 | u++) | |
1225 | if (UUID_FLASH_ONLY(u)) | |
1226 | ret = flash_dev_run(c, u); | |
1227 | ||
1228 | return ret; | |
1229 | } | |
1230 | ||
1231 | int bch_flash_dev_create(struct cache_set *c, uint64_t size) | |
1232 | { | |
1233 | struct uuid_entry *u; | |
1234 | ||
1235 | if (test_bit(CACHE_SET_STOPPING, &c->flags)) | |
1236 | return -EINTR; | |
1237 | ||
1238 | u = uuid_find_empty(c); | |
1239 | if (!u) { | |
1240 | pr_err("Can't create volume, no room for UUID"); | |
1241 | return -EINVAL; | |
1242 | } | |
1243 | ||
1244 | get_random_bytes(u->uuid, 16); | |
1245 | memset(u->label, 0, 32); | |
1246 | u->first_reg = u->last_reg = cpu_to_le32(get_seconds()); | |
1247 | ||
1248 | SET_UUID_FLASH_ONLY(u, 1); | |
1249 | u->sectors = size >> 9; | |
1250 | ||
1251 | bch_uuid_write(c); | |
1252 | ||
1253 | return flash_dev_run(c, u); | |
1254 | } | |
1255 | ||
1256 | /* Cache set */ | |
1257 | ||
1258 | __printf(2, 3) | |
1259 | bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) | |
1260 | { | |
1261 | va_list args; | |
1262 | ||
77c320eb KO |
1263 | if (c->on_error != ON_ERROR_PANIC && |
1264 | test_bit(CACHE_SET_STOPPING, &c->flags)) | |
cafe5635 KO |
1265 | return false; |
1266 | ||
1267 | /* XXX: we can be called from atomic context | |
1268 | acquire_console_sem(); | |
1269 | */ | |
1270 | ||
1271 | printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid); | |
1272 | ||
1273 | va_start(args, fmt); | |
1274 | vprintk(fmt, args); | |
1275 | va_end(args); | |
1276 | ||
1277 | printk(", disabling caching\n"); | |
1278 | ||
77c320eb KO |
1279 | if (c->on_error == ON_ERROR_PANIC) |
1280 | panic("panic forced after error\n"); | |
1281 | ||
cafe5635 KO |
1282 | bch_cache_set_unregister(c); |
1283 | return true; | |
1284 | } | |
1285 | ||
1286 | void bch_cache_set_release(struct kobject *kobj) | |
1287 | { | |
1288 | struct cache_set *c = container_of(kobj, struct cache_set, kobj); | |
1289 | kfree(c); | |
1290 | module_put(THIS_MODULE); | |
1291 | } | |
1292 | ||
1293 | static void cache_set_free(struct closure *cl) | |
1294 | { | |
1295 | struct cache_set *c = container_of(cl, struct cache_set, cl); | |
1296 | struct cache *ca; | |
1297 | unsigned i; | |
1298 | ||
1299 | if (!IS_ERR_OR_NULL(c->debug)) | |
1300 | debugfs_remove(c->debug); | |
1301 | ||
1302 | bch_open_buckets_free(c); | |
1303 | bch_btree_cache_free(c); | |
1304 | bch_journal_free(c); | |
1305 | ||
1306 | for_each_cache(ca, c, i) | |
1307 | if (ca) | |
1308 | kobject_put(&ca->kobj); | |
1309 | ||
1310 | free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c))); | |
1311 | free_pages((unsigned long) c->sort, ilog2(bucket_pages(c))); | |
1312 | ||
cafe5635 KO |
1313 | if (c->bio_split) |
1314 | bioset_free(c->bio_split); | |
57943511 KO |
1315 | if (c->fill_iter) |
1316 | mempool_destroy(c->fill_iter); | |
cafe5635 KO |
1317 | if (c->bio_meta) |
1318 | mempool_destroy(c->bio_meta); | |
1319 | if (c->search) | |
1320 | mempool_destroy(c->search); | |
1321 | kfree(c->devices); | |
1322 | ||
1323 | mutex_lock(&bch_register_lock); | |
1324 | list_del(&c->list); | |
1325 | mutex_unlock(&bch_register_lock); | |
1326 | ||
1327 | pr_info("Cache set %pU unregistered", c->sb.set_uuid); | |
1328 | wake_up(&unregister_wait); | |
1329 | ||
1330 | closure_debug_destroy(&c->cl); | |
1331 | kobject_put(&c->kobj); | |
1332 | } | |
1333 | ||
1334 | static void cache_set_flush(struct closure *cl) | |
1335 | { | |
1336 | struct cache_set *c = container_of(cl, struct cache_set, caching); | |
79826c35 | 1337 | struct cache *ca; |
cafe5635 | 1338 | struct btree *b; |
79826c35 | 1339 | unsigned i; |
cafe5635 KO |
1340 | |
1341 | bch_cache_accounting_destroy(&c->accounting); | |
1342 | ||
1343 | kobject_put(&c->internal); | |
1344 | kobject_del(&c->kobj); | |
1345 | ||
1346 | if (!IS_ERR_OR_NULL(c->root)) | |
1347 | list_add(&c->root->list, &c->btree_cache); | |
1348 | ||
1349 | /* Should skip this if we're unregistering because of an error */ | |
1350 | list_for_each_entry(b, &c->btree_cache, list) | |
1351 | if (btree_node_dirty(b)) | |
57943511 | 1352 | bch_btree_node_write(b, NULL); |
cafe5635 | 1353 | |
79826c35 KO |
1354 | for_each_cache(ca, c, i) |
1355 | if (ca->alloc_thread) | |
1356 | kthread_stop(ca->alloc_thread); | |
1357 | ||
cafe5635 KO |
1358 | closure_return(cl); |
1359 | } | |
1360 | ||
1361 | static void __cache_set_unregister(struct closure *cl) | |
1362 | { | |
1363 | struct cache_set *c = container_of(cl, struct cache_set, caching); | |
5caa52af | 1364 | struct cached_dev *dc; |
cafe5635 KO |
1365 | size_t i; |
1366 | ||
1367 | mutex_lock(&bch_register_lock); | |
1368 | ||
cafe5635 | 1369 | for (i = 0; i < c->nr_uuids; i++) |
5caa52af KO |
1370 | if (c->devices[i]) { |
1371 | if (!UUID_FLASH_ONLY(&c->uuids[i]) && | |
1372 | test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { | |
1373 | dc = container_of(c->devices[i], | |
1374 | struct cached_dev, disk); | |
1375 | bch_cached_dev_detach(dc); | |
1376 | } else { | |
1377 | bcache_device_stop(c->devices[i]); | |
1378 | } | |
1379 | } | |
cafe5635 KO |
1380 | |
1381 | mutex_unlock(&bch_register_lock); | |
1382 | ||
1383 | continue_at(cl, cache_set_flush, system_wq); | |
1384 | } | |
1385 | ||
1386 | void bch_cache_set_stop(struct cache_set *c) | |
1387 | { | |
1388 | if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) | |
1389 | closure_queue(&c->caching); | |
1390 | } | |
1391 | ||
1392 | void bch_cache_set_unregister(struct cache_set *c) | |
1393 | { | |
1394 | set_bit(CACHE_SET_UNREGISTERING, &c->flags); | |
1395 | bch_cache_set_stop(c); | |
1396 | } | |
1397 | ||
1398 | #define alloc_bucket_pages(gfp, c) \ | |
1399 | ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c)))) | |
1400 | ||
1401 | struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) | |
1402 | { | |
1403 | int iter_size; | |
1404 | struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); | |
1405 | if (!c) | |
1406 | return NULL; | |
1407 | ||
1408 | __module_get(THIS_MODULE); | |
1409 | closure_init(&c->cl, NULL); | |
1410 | set_closure_fn(&c->cl, cache_set_free, system_wq); | |
1411 | ||
1412 | closure_init(&c->caching, &c->cl); | |
1413 | set_closure_fn(&c->caching, __cache_set_unregister, system_wq); | |
1414 | ||
1415 | /* Maybe create continue_at_noreturn() and use it here? */ | |
1416 | closure_set_stopped(&c->cl); | |
1417 | closure_put(&c->cl); | |
1418 | ||
1419 | kobject_init(&c->kobj, &bch_cache_set_ktype); | |
1420 | kobject_init(&c->internal, &bch_cache_set_internal_ktype); | |
1421 | ||
1422 | bch_cache_accounting_init(&c->accounting, &c->cl); | |
1423 | ||
1424 | memcpy(c->sb.set_uuid, sb->set_uuid, 16); | |
1425 | c->sb.block_size = sb->block_size; | |
1426 | c->sb.bucket_size = sb->bucket_size; | |
1427 | c->sb.nr_in_set = sb->nr_in_set; | |
1428 | c->sb.last_mount = sb->last_mount; | |
1429 | c->bucket_bits = ilog2(sb->bucket_size); | |
1430 | c->block_bits = ilog2(sb->block_size); | |
1431 | c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry); | |
1432 | ||
1433 | c->btree_pages = c->sb.bucket_size / PAGE_SECTORS; | |
1434 | if (c->btree_pages > BTREE_MAX_PAGES) | |
1435 | c->btree_pages = max_t(int, c->btree_pages / 4, | |
1436 | BTREE_MAX_PAGES); | |
1437 | ||
6ded34d1 KO |
1438 | c->sort_crit_factor = int_sqrt(c->btree_pages); |
1439 | ||
cafe5635 | 1440 | mutex_init(&c->bucket_lock); |
cafe5635 KO |
1441 | mutex_init(&c->sort_lock); |
1442 | spin_lock_init(&c->sort_time_lock); | |
1443 | closure_init_unlocked(&c->sb_write); | |
1444 | closure_init_unlocked(&c->uuid_write); | |
1445 | spin_lock_init(&c->btree_read_time_lock); | |
1446 | bch_moving_init_cache_set(c); | |
1447 | ||
1448 | INIT_LIST_HEAD(&c->list); | |
1449 | INIT_LIST_HEAD(&c->cached_devs); | |
1450 | INIT_LIST_HEAD(&c->btree_cache); | |
1451 | INIT_LIST_HEAD(&c->btree_cache_freeable); | |
1452 | INIT_LIST_HEAD(&c->btree_cache_freed); | |
1453 | INIT_LIST_HEAD(&c->data_buckets); | |
1454 | ||
1455 | c->search = mempool_create_slab_pool(32, bch_search_cache); | |
1456 | if (!c->search) | |
1457 | goto err; | |
1458 | ||
1459 | iter_size = (sb->bucket_size / sb->block_size + 1) * | |
1460 | sizeof(struct btree_iter_set); | |
1461 | ||
1462 | if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) || | |
1463 | !(c->bio_meta = mempool_create_kmalloc_pool(2, | |
1464 | sizeof(struct bbio) + sizeof(struct bio_vec) * | |
1465 | bucket_pages(c))) || | |
57943511 | 1466 | !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) || |
cafe5635 | 1467 | !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || |
cafe5635 KO |
1468 | !(c->sort = alloc_bucket_pages(GFP_KERNEL, c)) || |
1469 | !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) || | |
1470 | bch_journal_alloc(c) || | |
1471 | bch_btree_cache_alloc(c) || | |
1472 | bch_open_buckets_alloc(c)) | |
1473 | goto err; | |
1474 | ||
cafe5635 KO |
1475 | c->congested_read_threshold_us = 2000; |
1476 | c->congested_write_threshold_us = 20000; | |
1477 | c->error_limit = 8 << IO_ERROR_SHIFT; | |
1478 | ||
1479 | return c; | |
1480 | err: | |
1481 | bch_cache_set_unregister(c); | |
1482 | return NULL; | |
1483 | } | |
1484 | ||
1485 | static void run_cache_set(struct cache_set *c) | |
1486 | { | |
1487 | const char *err = "cannot allocate memory"; | |
1488 | struct cached_dev *dc, *t; | |
1489 | struct cache *ca; | |
1490 | unsigned i; | |
1491 | ||
1492 | struct btree_op op; | |
1493 | bch_btree_op_init_stack(&op); | |
1494 | op.lock = SHRT_MAX; | |
1495 | ||
1496 | for_each_cache(ca, c, i) | |
1497 | c->nbuckets += ca->sb.nbuckets; | |
1498 | ||
1499 | if (CACHE_SYNC(&c->sb)) { | |
1500 | LIST_HEAD(journal); | |
1501 | struct bkey *k; | |
1502 | struct jset *j; | |
1503 | ||
1504 | err = "cannot allocate memory for journal"; | |
1505 | if (bch_journal_read(c, &journal, &op)) | |
1506 | goto err; | |
1507 | ||
1508 | pr_debug("btree_journal_read() done"); | |
1509 | ||
1510 | err = "no journal entries found"; | |
1511 | if (list_empty(&journal)) | |
1512 | goto err; | |
1513 | ||
1514 | j = &list_entry(journal.prev, struct journal_replay, list)->j; | |
1515 | ||
1516 | err = "IO error reading priorities"; | |
1517 | for_each_cache(ca, c, i) | |
1518 | prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]); | |
1519 | ||
1520 | /* | |
1521 | * If prio_read() fails it'll call cache_set_error and we'll | |
1522 | * tear everything down right away, but if we perhaps checked | |
1523 | * sooner we could avoid journal replay. | |
1524 | */ | |
1525 | ||
1526 | k = &j->btree_root; | |
1527 | ||
1528 | err = "bad btree root"; | |
1529 | if (__bch_ptr_invalid(c, j->btree_level + 1, k)) | |
1530 | goto err; | |
1531 | ||
1532 | err = "error reading btree root"; | |
1533 | c->root = bch_btree_node_get(c, k, j->btree_level, &op); | |
1534 | if (IS_ERR_OR_NULL(c->root)) | |
1535 | goto err; | |
1536 | ||
1537 | list_del_init(&c->root->list); | |
1538 | rw_unlock(true, c->root); | |
1539 | ||
1540 | err = uuid_read(c, j, &op.cl); | |
1541 | if (err) | |
1542 | goto err; | |
1543 | ||
1544 | err = "error in recovery"; | |
1545 | if (bch_btree_check(c, &op)) | |
1546 | goto err; | |
1547 | ||
1548 | bch_journal_mark(c, &journal); | |
1549 | bch_btree_gc_finish(c); | |
1550 | pr_debug("btree_check() done"); | |
1551 | ||
1552 | /* | |
1553 | * bcache_journal_next() can't happen sooner, or | |
1554 | * btree_gc_finish() will give spurious errors about last_gc > | |
1555 | * gc_gen - this is a hack but oh well. | |
1556 | */ | |
1557 | bch_journal_next(&c->journal); | |
1558 | ||
119ba0f8 | 1559 | err = "error starting allocator thread"; |
cafe5635 | 1560 | for_each_cache(ca, c, i) |
119ba0f8 KO |
1561 | if (bch_cache_allocator_start(ca)) |
1562 | goto err; | |
cafe5635 KO |
1563 | |
1564 | /* | |
1565 | * First place it's safe to allocate: btree_check() and | |
1566 | * btree_gc_finish() have to run before we have buckets to | |
1567 | * allocate, and bch_bucket_alloc_set() might cause a journal | |
1568 | * entry to be written so bcache_journal_next() has to be called | |
1569 | * first. | |
1570 | * | |
1571 | * If the uuids were in the old format we have to rewrite them | |
1572 | * before the next journal entry is written: | |
1573 | */ | |
1574 | if (j->version < BCACHE_JSET_VERSION_UUID) | |
1575 | __uuid_write(c); | |
1576 | ||
1577 | bch_journal_replay(c, &journal, &op); | |
1578 | } else { | |
1579 | pr_notice("invalidating existing data"); | |
1580 | /* Don't want invalidate_buckets() to queue a gc yet */ | |
1581 | closure_lock(&c->gc, NULL); | |
1582 | ||
1583 | for_each_cache(ca, c, i) { | |
1584 | unsigned j; | |
1585 | ||
1586 | ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, | |
1587 | 2, SB_JOURNAL_BUCKETS); | |
1588 | ||
1589 | for (j = 0; j < ca->sb.keys; j++) | |
1590 | ca->sb.d[j] = ca->sb.first_bucket + j; | |
1591 | } | |
1592 | ||
1593 | bch_btree_gc_finish(c); | |
1594 | ||
119ba0f8 | 1595 | err = "error starting allocator thread"; |
cafe5635 | 1596 | for_each_cache(ca, c, i) |
119ba0f8 KO |
1597 | if (bch_cache_allocator_start(ca)) |
1598 | goto err; | |
cafe5635 KO |
1599 | |
1600 | mutex_lock(&c->bucket_lock); | |
1601 | for_each_cache(ca, c, i) | |
1602 | bch_prio_write(ca); | |
1603 | mutex_unlock(&c->bucket_lock); | |
1604 | ||
cafe5635 KO |
1605 | err = "cannot allocate new UUID bucket"; |
1606 | if (__uuid_write(c)) | |
1607 | goto err_unlock_gc; | |
1608 | ||
1609 | err = "cannot allocate new btree root"; | |
1610 | c->root = bch_btree_node_alloc(c, 0, &op.cl); | |
1611 | if (IS_ERR_OR_NULL(c->root)) | |
1612 | goto err_unlock_gc; | |
1613 | ||
1614 | bkey_copy_key(&c->root->key, &MAX_KEY); | |
57943511 | 1615 | bch_btree_node_write(c->root, &op.cl); |
cafe5635 KO |
1616 | |
1617 | bch_btree_set_root(c->root); | |
1618 | rw_unlock(true, c->root); | |
1619 | ||
1620 | /* | |
1621 | * We don't want to write the first journal entry until | |
1622 | * everything is set up - fortunately journal entries won't be | |
1623 | * written until the SET_CACHE_SYNC() here: | |
1624 | */ | |
1625 | SET_CACHE_SYNC(&c->sb, true); | |
1626 | ||
1627 | bch_journal_next(&c->journal); | |
1628 | bch_journal_meta(c, &op.cl); | |
1629 | ||
1630 | /* Unlock */ | |
1631 | closure_set_stopped(&c->gc.cl); | |
1632 | closure_put(&c->gc.cl); | |
1633 | } | |
1634 | ||
1635 | closure_sync(&op.cl); | |
1636 | c->sb.last_mount = get_seconds(); | |
1637 | bcache_write_super(c); | |
1638 | ||
1639 | list_for_each_entry_safe(dc, t, &uncached_devices, list) | |
1640 | bch_cached_dev_attach(dc, c); | |
1641 | ||
1642 | flash_devs_run(c); | |
1643 | ||
1644 | return; | |
1645 | err_unlock_gc: | |
1646 | closure_set_stopped(&c->gc.cl); | |
1647 | closure_put(&c->gc.cl); | |
1648 | err: | |
1649 | closure_sync(&op.cl); | |
1650 | /* XXX: test this, it's broken */ | |
1651 | bch_cache_set_error(c, err); | |
1652 | } | |
1653 | ||
1654 | static bool can_attach_cache(struct cache *ca, struct cache_set *c) | |
1655 | { | |
1656 | return ca->sb.block_size == c->sb.block_size && | |
1657 | ca->sb.bucket_size == c->sb.block_size && | |
1658 | ca->sb.nr_in_set == c->sb.nr_in_set; | |
1659 | } | |
1660 | ||
1661 | static const char *register_cache_set(struct cache *ca) | |
1662 | { | |
1663 | char buf[12]; | |
1664 | const char *err = "cannot allocate memory"; | |
1665 | struct cache_set *c; | |
1666 | ||
1667 | list_for_each_entry(c, &bch_cache_sets, list) | |
1668 | if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) { | |
1669 | if (c->cache[ca->sb.nr_this_dev]) | |
1670 | return "duplicate cache set member"; | |
1671 | ||
1672 | if (!can_attach_cache(ca, c)) | |
1673 | return "cache sb does not match set"; | |
1674 | ||
1675 | if (!CACHE_SYNC(&ca->sb)) | |
1676 | SET_CACHE_SYNC(&c->sb, false); | |
1677 | ||
1678 | goto found; | |
1679 | } | |
1680 | ||
1681 | c = bch_cache_set_alloc(&ca->sb); | |
1682 | if (!c) | |
1683 | return err; | |
1684 | ||
1685 | err = "error creating kobject"; | |
1686 | if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) || | |
1687 | kobject_add(&c->internal, &c->kobj, "internal")) | |
1688 | goto err; | |
1689 | ||
1690 | if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) | |
1691 | goto err; | |
1692 | ||
1693 | bch_debug_init_cache_set(c); | |
1694 | ||
1695 | list_add(&c->list, &bch_cache_sets); | |
1696 | found: | |
1697 | sprintf(buf, "cache%i", ca->sb.nr_this_dev); | |
1698 | if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || | |
1699 | sysfs_create_link(&c->kobj, &ca->kobj, buf)) | |
1700 | goto err; | |
1701 | ||
1702 | if (ca->sb.seq > c->sb.seq) { | |
1703 | c->sb.version = ca->sb.version; | |
1704 | memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16); | |
1705 | c->sb.flags = ca->sb.flags; | |
1706 | c->sb.seq = ca->sb.seq; | |
1707 | pr_debug("set version = %llu", c->sb.version); | |
1708 | } | |
1709 | ||
1710 | ca->set = c; | |
1711 | ca->set->cache[ca->sb.nr_this_dev] = ca; | |
1712 | c->cache_by_alloc[c->caches_loaded++] = ca; | |
1713 | ||
1714 | if (c->caches_loaded == c->sb.nr_in_set) | |
1715 | run_cache_set(c); | |
1716 | ||
1717 | return NULL; | |
1718 | err: | |
1719 | bch_cache_set_unregister(c); | |
1720 | return err; | |
1721 | } | |
1722 | ||
1723 | /* Cache device */ | |
1724 | ||
1725 | void bch_cache_release(struct kobject *kobj) | |
1726 | { | |
1727 | struct cache *ca = container_of(kobj, struct cache, kobj); | |
1728 | ||
1729 | if (ca->set) | |
1730 | ca->set->cache[ca->sb.nr_this_dev] = NULL; | |
1731 | ||
cafe5635 KO |
1732 | bio_split_pool_free(&ca->bio_split_hook); |
1733 | ||
cafe5635 KO |
1734 | free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca))); |
1735 | kfree(ca->prio_buckets); | |
1736 | vfree(ca->buckets); | |
1737 | ||
1738 | free_heap(&ca->heap); | |
1739 | free_fifo(&ca->unused); | |
1740 | free_fifo(&ca->free_inc); | |
1741 | free_fifo(&ca->free); | |
1742 | ||
1743 | if (ca->sb_bio.bi_inline_vecs[0].bv_page) | |
1744 | put_page(ca->sb_bio.bi_io_vec[0].bv_page); | |
1745 | ||
1746 | if (!IS_ERR_OR_NULL(ca->bdev)) { | |
1747 | blk_sync_queue(bdev_get_queue(ca->bdev)); | |
1748 | blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); | |
1749 | } | |
1750 | ||
1751 | kfree(ca); | |
1752 | module_put(THIS_MODULE); | |
1753 | } | |
1754 | ||
1755 | static int cache_alloc(struct cache_sb *sb, struct cache *ca) | |
1756 | { | |
1757 | size_t free; | |
1758 | struct bucket *b; | |
1759 | ||
cafe5635 KO |
1760 | __module_get(THIS_MODULE); |
1761 | kobject_init(&ca->kobj, &bch_cache_ktype); | |
1762 | ||
cafe5635 KO |
1763 | bio_init(&ca->journal.bio); |
1764 | ca->journal.bio.bi_max_vecs = 8; | |
1765 | ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs; | |
1766 | ||
1767 | free = roundup_pow_of_two(ca->sb.nbuckets) >> 9; | |
1768 | free = max_t(size_t, free, (prio_buckets(ca) + 8) * 2); | |
1769 | ||
1770 | if (!init_fifo(&ca->free, free, GFP_KERNEL) || | |
1771 | !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) || | |
1772 | !init_fifo(&ca->unused, free << 2, GFP_KERNEL) || | |
1773 | !init_heap(&ca->heap, free << 3, GFP_KERNEL) || | |
f59fce84 | 1774 | !(ca->buckets = vzalloc(sizeof(struct bucket) * |
cafe5635 KO |
1775 | ca->sb.nbuckets)) || |
1776 | !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) * | |
1777 | 2, GFP_KERNEL)) || | |
1778 | !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) || | |
cafe5635 | 1779 | bio_split_pool_init(&ca->bio_split_hook)) |
f59fce84 | 1780 | return -ENOMEM; |
cafe5635 KO |
1781 | |
1782 | ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); | |
1783 | ||
cafe5635 KO |
1784 | for_each_bucket(b, ca) |
1785 | atomic_set(&b->pin, 0); | |
1786 | ||
1787 | if (bch_cache_allocator_init(ca)) | |
1788 | goto err; | |
1789 | ||
1790 | return 0; | |
1791 | err: | |
1792 | kobject_put(&ca->kobj); | |
1793 | return -ENOMEM; | |
1794 | } | |
1795 | ||
f59fce84 | 1796 | static void register_cache(struct cache_sb *sb, struct page *sb_page, |
cafe5635 KO |
1797 | struct block_device *bdev, struct cache *ca) |
1798 | { | |
1799 | char name[BDEVNAME_SIZE]; | |
1800 | const char *err = "cannot allocate memory"; | |
1801 | ||
f59fce84 | 1802 | memcpy(&ca->sb, sb, sizeof(struct cache_sb)); |
cafe5635 KO |
1803 | ca->bdev = bdev; |
1804 | ca->bdev->bd_holder = ca; | |
1805 | ||
f59fce84 KO |
1806 | bio_init(&ca->sb_bio); |
1807 | ca->sb_bio.bi_max_vecs = 1; | |
1808 | ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs; | |
1809 | ca->sb_bio.bi_io_vec[0].bv_page = sb_page; | |
1810 | get_page(sb_page); | |
1811 | ||
cafe5635 KO |
1812 | if (blk_queue_discard(bdev_get_queue(ca->bdev))) |
1813 | ca->discard = CACHE_DISCARD(&ca->sb); | |
1814 | ||
f59fce84 KO |
1815 | if (cache_alloc(sb, ca) != 0) |
1816 | goto err; | |
1817 | ||
cafe5635 KO |
1818 | err = "error creating kobject"; |
1819 | if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) | |
1820 | goto err; | |
1821 | ||
1822 | err = register_cache_set(ca); | |
1823 | if (err) | |
1824 | goto err; | |
1825 | ||
1826 | pr_info("registered cache device %s", bdevname(bdev, name)); | |
f59fce84 | 1827 | return; |
cafe5635 | 1828 | err: |
f59fce84 | 1829 | pr_notice("error opening %s: %s", bdevname(bdev, name), err); |
cafe5635 | 1830 | kobject_put(&ca->kobj); |
cafe5635 KO |
1831 | } |
1832 | ||
1833 | /* Global interfaces/init */ | |
1834 | ||
1835 | static ssize_t register_bcache(struct kobject *, struct kobj_attribute *, | |
1836 | const char *, size_t); | |
1837 | ||
1838 | kobj_attribute_write(register, register_bcache); | |
1839 | kobj_attribute_write(register_quiet, register_bcache); | |
1840 | ||
a9dd53ad GP |
1841 | static bool bch_is_open_backing(struct block_device *bdev) { |
1842 | struct cache_set *c, *tc; | |
1843 | struct cached_dev *dc, *t; | |
1844 | ||
1845 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) | |
1846 | list_for_each_entry_safe(dc, t, &c->cached_devs, list) | |
1847 | if (dc->bdev == bdev) | |
1848 | return true; | |
1849 | list_for_each_entry_safe(dc, t, &uncached_devices, list) | |
1850 | if (dc->bdev == bdev) | |
1851 | return true; | |
1852 | return false; | |
1853 | } | |
1854 | ||
1855 | static bool bch_is_open_cache(struct block_device *bdev) { | |
1856 | struct cache_set *c, *tc; | |
1857 | struct cache *ca; | |
1858 | unsigned i; | |
1859 | ||
1860 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) | |
1861 | for_each_cache(ca, c, i) | |
1862 | if (ca->bdev == bdev) | |
1863 | return true; | |
1864 | return false; | |
1865 | } | |
1866 | ||
1867 | static bool bch_is_open(struct block_device *bdev) { | |
1868 | return bch_is_open_cache(bdev) || bch_is_open_backing(bdev); | |
1869 | } | |
1870 | ||
cafe5635 KO |
1871 | static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, |
1872 | const char *buffer, size_t size) | |
1873 | { | |
1874 | ssize_t ret = size; | |
1875 | const char *err = "cannot allocate memory"; | |
1876 | char *path = NULL; | |
1877 | struct cache_sb *sb = NULL; | |
1878 | struct block_device *bdev = NULL; | |
1879 | struct page *sb_page = NULL; | |
1880 | ||
1881 | if (!try_module_get(THIS_MODULE)) | |
1882 | return -EBUSY; | |
1883 | ||
1884 | mutex_lock(&bch_register_lock); | |
1885 | ||
1886 | if (!(path = kstrndup(buffer, size, GFP_KERNEL)) || | |
1887 | !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL))) | |
1888 | goto err; | |
1889 | ||
1890 | err = "failed to open device"; | |
1891 | bdev = blkdev_get_by_path(strim(path), | |
1892 | FMODE_READ|FMODE_WRITE|FMODE_EXCL, | |
1893 | sb); | |
f59fce84 | 1894 | if (IS_ERR(bdev)) { |
a9dd53ad GP |
1895 | if (bdev == ERR_PTR(-EBUSY)) { |
1896 | bdev = lookup_bdev(strim(path)); | |
1897 | if (!IS_ERR(bdev) && bch_is_open(bdev)) | |
1898 | err = "device already registered"; | |
1899 | else | |
1900 | err = "device busy"; | |
1901 | } | |
cafe5635 | 1902 | goto err; |
f59fce84 KO |
1903 | } |
1904 | ||
1905 | err = "failed to set blocksize"; | |
1906 | if (set_blocksize(bdev, 4096)) | |
1907 | goto err_close; | |
cafe5635 KO |
1908 | |
1909 | err = read_super(sb, bdev, &sb_page); | |
1910 | if (err) | |
1911 | goto err_close; | |
1912 | ||
2903381f | 1913 | if (SB_IS_BDEV(sb)) { |
cafe5635 | 1914 | struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL); |
f59fce84 KO |
1915 | if (!dc) |
1916 | goto err_close; | |
cafe5635 | 1917 | |
f59fce84 | 1918 | register_bdev(sb, sb_page, bdev, dc); |
cafe5635 KO |
1919 | } else { |
1920 | struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
f59fce84 KO |
1921 | if (!ca) |
1922 | goto err_close; | |
cafe5635 | 1923 | |
f59fce84 | 1924 | register_cache(sb, sb_page, bdev, ca); |
cafe5635 | 1925 | } |
f59fce84 KO |
1926 | out: |
1927 | if (sb_page) | |
cafe5635 | 1928 | put_page(sb_page); |
cafe5635 KO |
1929 | kfree(sb); |
1930 | kfree(path); | |
1931 | mutex_unlock(&bch_register_lock); | |
1932 | module_put(THIS_MODULE); | |
1933 | return ret; | |
f59fce84 KO |
1934 | |
1935 | err_close: | |
1936 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); | |
1937 | err: | |
1938 | if (attr != &ksysfs_register_quiet) | |
1939 | pr_info("error opening %s: %s", path, err); | |
1940 | ret = -EINVAL; | |
1941 | goto out; | |
cafe5635 KO |
1942 | } |
1943 | ||
1944 | static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) | |
1945 | { | |
1946 | if (code == SYS_DOWN || | |
1947 | code == SYS_HALT || | |
1948 | code == SYS_POWER_OFF) { | |
1949 | DEFINE_WAIT(wait); | |
1950 | unsigned long start = jiffies; | |
1951 | bool stopped = false; | |
1952 | ||
1953 | struct cache_set *c, *tc; | |
1954 | struct cached_dev *dc, *tdc; | |
1955 | ||
1956 | mutex_lock(&bch_register_lock); | |
1957 | ||
1958 | if (list_empty(&bch_cache_sets) && | |
1959 | list_empty(&uncached_devices)) | |
1960 | goto out; | |
1961 | ||
1962 | pr_info("Stopping all devices:"); | |
1963 | ||
1964 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) | |
1965 | bch_cache_set_stop(c); | |
1966 | ||
1967 | list_for_each_entry_safe(dc, tdc, &uncached_devices, list) | |
1968 | bcache_device_stop(&dc->disk); | |
1969 | ||
1970 | /* What's a condition variable? */ | |
1971 | while (1) { | |
1972 | long timeout = start + 2 * HZ - jiffies; | |
1973 | ||
1974 | stopped = list_empty(&bch_cache_sets) && | |
1975 | list_empty(&uncached_devices); | |
1976 | ||
1977 | if (timeout < 0 || stopped) | |
1978 | break; | |
1979 | ||
1980 | prepare_to_wait(&unregister_wait, &wait, | |
1981 | TASK_UNINTERRUPTIBLE); | |
1982 | ||
1983 | mutex_unlock(&bch_register_lock); | |
1984 | schedule_timeout(timeout); | |
1985 | mutex_lock(&bch_register_lock); | |
1986 | } | |
1987 | ||
1988 | finish_wait(&unregister_wait, &wait); | |
1989 | ||
1990 | if (stopped) | |
1991 | pr_info("All devices stopped"); | |
1992 | else | |
1993 | pr_notice("Timeout waiting for devices to be closed"); | |
1994 | out: | |
1995 | mutex_unlock(&bch_register_lock); | |
1996 | } | |
1997 | ||
1998 | return NOTIFY_DONE; | |
1999 | } | |
2000 | ||
2001 | static struct notifier_block reboot = { | |
2002 | .notifier_call = bcache_reboot, | |
2003 | .priority = INT_MAX, /* before any real devices */ | |
2004 | }; | |
2005 | ||
2006 | static void bcache_exit(void) | |
2007 | { | |
2008 | bch_debug_exit(); | |
2009 | bch_writeback_exit(); | |
2010 | bch_request_exit(); | |
2011 | bch_btree_exit(); | |
2012 | if (bcache_kobj) | |
2013 | kobject_put(bcache_kobj); | |
2014 | if (bcache_wq) | |
2015 | destroy_workqueue(bcache_wq); | |
2016 | unregister_blkdev(bcache_major, "bcache"); | |
2017 | unregister_reboot_notifier(&reboot); | |
2018 | } | |
2019 | ||
2020 | static int __init bcache_init(void) | |
2021 | { | |
2022 | static const struct attribute *files[] = { | |
2023 | &ksysfs_register.attr, | |
2024 | &ksysfs_register_quiet.attr, | |
2025 | NULL | |
2026 | }; | |
2027 | ||
2028 | mutex_init(&bch_register_lock); | |
2029 | init_waitqueue_head(&unregister_wait); | |
2030 | register_reboot_notifier(&reboot); | |
07e86ccb | 2031 | closure_debug_init(); |
cafe5635 KO |
2032 | |
2033 | bcache_major = register_blkdev(0, "bcache"); | |
2034 | if (bcache_major < 0) | |
2035 | return bcache_major; | |
2036 | ||
2037 | if (!(bcache_wq = create_workqueue("bcache")) || | |
2038 | !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) || | |
2039 | sysfs_create_files(bcache_kobj, files) || | |
2040 | bch_btree_init() || | |
2041 | bch_request_init() || | |
2042 | bch_writeback_init() || | |
2043 | bch_debug_init(bcache_kobj)) | |
2044 | goto err; | |
2045 | ||
2046 | return 0; | |
2047 | err: | |
2048 | bcache_exit(); | |
2049 | return -ENOMEM; | |
2050 | } | |
2051 | ||
2052 | module_exit(bcache_exit); | |
2053 | module_init(bcache_init); |