Commit | Line | Data |
---|---|---|
87418ef9 | 1 | // SPDX-License-Identifier: GPL-2.0 |
cafe5635 KO |
2 | /* |
3 | * bcache setup/teardown code, and some metadata io - read a superblock and | |
4 | * figure out what to do with it. | |
5 | * | |
6 | * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> | |
7 | * Copyright 2012 Google, Inc. | |
8 | */ | |
9 | ||
10 | #include "bcache.h" | |
11 | #include "btree.h" | |
12 | #include "debug.h" | |
65d45231 | 13 | #include "extents.h" |
cafe5635 | 14 | #include "request.h" |
279afbad | 15 | #include "writeback.h" |
d721a43f | 16 | #include "features.h" |
cafe5635 | 17 | |
c37511b8 | 18 | #include <linux/blkdev.h> |
cafe5635 KO |
19 | #include <linux/debugfs.h> |
20 | #include <linux/genhd.h> | |
28935ab5 | 21 | #include <linux/idr.h> |
79826c35 | 22 | #include <linux/kthread.h> |
ee4a36f4 | 23 | #include <linux/workqueue.h> |
cafe5635 KO |
24 | #include <linux/module.h> |
25 | #include <linux/random.h> | |
26 | #include <linux/reboot.h> | |
27 | #include <linux/sysfs.h> | |
28 | ||
9aaf5165 CL |
29 | unsigned int bch_cutoff_writeback; |
30 | unsigned int bch_cutoff_writeback_sync; | |
31 | ||
cafe5635 KO |
32 | static const char bcache_magic[] = { |
33 | 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca, | |
34 | 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81 | |
35 | }; | |
36 | ||
37 | static const char invalid_uuid[] = { | |
38 | 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78, | |
39 | 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99 | |
40 | }; | |
41 | ||
cafe5635 KO |
42 | static struct kobject *bcache_kobj; |
43 | struct mutex bch_register_lock; | |
a59ff6cc | 44 | bool bcache_is_reboot; |
cafe5635 KO |
45 | LIST_HEAD(bch_cache_sets); |
46 | static LIST_HEAD(uncached_devices); | |
47 | ||
28935ab5 | 48 | static int bcache_major; |
1dbe32ad | 49 | static DEFINE_IDA(bcache_device_idx); |
cafe5635 KO |
50 | static wait_queue_head_t unregister_wait; |
51 | struct workqueue_struct *bcache_wq; | |
0f843e65 | 52 | struct workqueue_struct *bch_journal_wq; |
cafe5635 | 53 | |
a59ff6cc | 54 | |
cafe5635 | 55 | #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE) |
1dbe32ad CL |
56 | /* limitation of partitions number on single bcache device */ |
57 | #define BCACHE_MINORS 128 | |
58 | /* limitation of bcache devices number on single system */ | |
59 | #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS) | |
cafe5635 | 60 | |
cafe5635 KO |
61 | /* Superblock */ |
62 | ||
ffa47032 CL |
63 | static unsigned int get_bucket_size(struct cache_sb *sb, struct cache_sb_disk *s) |
64 | { | |
65 | unsigned int bucket_size = le16_to_cpu(s->bucket_size); | |
66 | ||
67 | if (sb->version >= BCACHE_SB_VERSION_CDEV_WITH_FEATURES && | |
68 | bch_has_feature_large_bucket(sb)) | |
69 | bucket_size |= le16_to_cpu(s->bucket_size_hi) << 16; | |
70 | ||
71 | return bucket_size; | |
72 | } | |
73 | ||
5b21403c CL |
74 | static const char *read_super_common(struct cache_sb *sb, struct block_device *bdev, |
75 | struct cache_sb_disk *s) | |
76 | { | |
77 | const char *err; | |
78 | unsigned int i; | |
79 | ||
198efa35 | 80 | sb->first_bucket= le16_to_cpu(s->first_bucket); |
5b21403c | 81 | sb->nbuckets = le64_to_cpu(s->nbuckets); |
ffa47032 | 82 | sb->bucket_size = get_bucket_size(sb, s); |
5b21403c CL |
83 | |
84 | sb->nr_in_set = le16_to_cpu(s->nr_in_set); | |
85 | sb->nr_this_dev = le16_to_cpu(s->nr_this_dev); | |
86 | ||
198efa35 CL |
87 | err = "Too many journal buckets"; |
88 | if (sb->keys > SB_JOURNAL_BUCKETS) | |
89 | goto err; | |
90 | ||
5b21403c CL |
91 | err = "Too many buckets"; |
92 | if (sb->nbuckets > LONG_MAX) | |
93 | goto err; | |
94 | ||
95 | err = "Not enough buckets"; | |
96 | if (sb->nbuckets < 1 << 7) | |
97 | goto err; | |
98 | ||
c557a5f7 CL |
99 | err = "Bad block size (not power of 2)"; |
100 | if (!is_power_of_2(sb->block_size)) | |
101 | goto err; | |
102 | ||
103 | err = "Bad block size (larger than page size)"; | |
104 | if (sb->block_size > PAGE_SECTORS) | |
105 | goto err; | |
106 | ||
107 | err = "Bad bucket size (not power of 2)"; | |
108 | if (!is_power_of_2(sb->bucket_size)) | |
109 | goto err; | |
110 | ||
111 | err = "Bad bucket size (smaller than page size)"; | |
112 | if (sb->bucket_size < PAGE_SECTORS) | |
5b21403c CL |
113 | goto err; |
114 | ||
115 | err = "Invalid superblock: device too small"; | |
116 | if (get_capacity(bdev->bd_disk) < | |
117 | sb->bucket_size * sb->nbuckets) | |
118 | goto err; | |
119 | ||
120 | err = "Bad UUID"; | |
121 | if (bch_is_zero(sb->set_uuid, 16)) | |
122 | goto err; | |
123 | ||
124 | err = "Bad cache device number in set"; | |
125 | if (!sb->nr_in_set || | |
126 | sb->nr_in_set <= sb->nr_this_dev || | |
127 | sb->nr_in_set > MAX_CACHES_PER_SET) | |
128 | goto err; | |
129 | ||
130 | err = "Journal buckets not sequential"; | |
131 | for (i = 0; i < sb->keys; i++) | |
132 | if (sb->d[i] != sb->first_bucket + i) | |
133 | goto err; | |
134 | ||
135 | err = "Too many journal buckets"; | |
136 | if (sb->first_bucket + sb->keys > sb->nbuckets) | |
137 | goto err; | |
138 | ||
139 | err = "Invalid superblock: first bucket comes before end of super"; | |
140 | if (sb->first_bucket * sb->bucket_size < 16) | |
141 | goto err; | |
142 | ||
143 | err = NULL; | |
144 | err: | |
145 | return err; | |
146 | } | |
147 | ||
148 | ||
cafe5635 | 149 | static const char *read_super(struct cache_sb *sb, struct block_device *bdev, |
cfa0c56d | 150 | struct cache_sb_disk **res) |
cafe5635 KO |
151 | { |
152 | const char *err; | |
a702a692 | 153 | struct cache_sb_disk *s; |
6321bef0 | 154 | struct page *page; |
6f10f7d1 | 155 | unsigned int i; |
cafe5635 | 156 | |
6321bef0 CH |
157 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, |
158 | SB_OFFSET >> PAGE_SHIFT, GFP_KERNEL); | |
159 | if (IS_ERR(page)) | |
cafe5635 | 160 | return "IO error"; |
6321bef0 | 161 | s = page_address(page) + offset_in_page(SB_OFFSET); |
cafe5635 KO |
162 | |
163 | sb->offset = le64_to_cpu(s->offset); | |
164 | sb->version = le64_to_cpu(s->version); | |
165 | ||
166 | memcpy(sb->magic, s->magic, 16); | |
167 | memcpy(sb->uuid, s->uuid, 16); | |
168 | memcpy(sb->set_uuid, s->set_uuid, 16); | |
169 | memcpy(sb->label, s->label, SB_LABEL_SIZE); | |
170 | ||
171 | sb->flags = le64_to_cpu(s->flags); | |
172 | sb->seq = le64_to_cpu(s->seq); | |
cafe5635 | 173 | sb->last_mount = le32_to_cpu(s->last_mount); |
cafe5635 KO |
174 | sb->keys = le16_to_cpu(s->keys); |
175 | ||
176 | for (i = 0; i < SB_JOURNAL_BUCKETS; i++) | |
177 | sb->d[i] = le64_to_cpu(s->d[i]); | |
178 | ||
46f5aa88 | 179 | pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u\n", |
cafe5635 KO |
180 | sb->version, sb->flags, sb->seq, sb->keys); |
181 | ||
aaf8dbea | 182 | err = "Not a bcache superblock (bad offset)"; |
cafe5635 KO |
183 | if (sb->offset != SB_SECTOR) |
184 | goto err; | |
185 | ||
aaf8dbea | 186 | err = "Not a bcache superblock (bad magic)"; |
cafe5635 KO |
187 | if (memcmp(sb->magic, bcache_magic, 16)) |
188 | goto err; | |
189 | ||
cafe5635 KO |
190 | err = "Bad checksum"; |
191 | if (s->csum != csum_set(s)) | |
192 | goto err; | |
193 | ||
194 | err = "Bad UUID"; | |
169ef1cf | 195 | if (bch_is_zero(sb->uuid, 16)) |
cafe5635 KO |
196 | goto err; |
197 | ||
8abb2a5d KO |
198 | sb->block_size = le16_to_cpu(s->block_size); |
199 | ||
200 | err = "Superblock block size smaller than device block size"; | |
201 | if (sb->block_size << 9 < bdev_logical_block_size(bdev)) | |
202 | goto err; | |
203 | ||
2903381f KO |
204 | switch (sb->version) { |
205 | case BCACHE_SB_VERSION_BDEV: | |
2903381f KO |
206 | sb->data_offset = BDEV_DATA_START_DEFAULT; |
207 | break; | |
208 | case BCACHE_SB_VERSION_BDEV_WITH_OFFSET: | |
d721a43f | 209 | case BCACHE_SB_VERSION_BDEV_WITH_FEATURES: |
2903381f KO |
210 | sb->data_offset = le64_to_cpu(s->data_offset); |
211 | ||
212 | err = "Bad data offset"; | |
213 | if (sb->data_offset < BDEV_DATA_START_DEFAULT) | |
214 | goto err; | |
cafe5635 | 215 | |
2903381f KO |
216 | break; |
217 | case BCACHE_SB_VERSION_CDEV: | |
218 | case BCACHE_SB_VERSION_CDEV_WITH_UUID: | |
5b21403c CL |
219 | err = read_super_common(sb, bdev, s); |
220 | if (err) | |
2903381f | 221 | goto err; |
2903381f | 222 | break; |
d721a43f | 223 | case BCACHE_SB_VERSION_CDEV_WITH_FEATURES: |
ffa47032 CL |
224 | /* |
225 | * Feature bits are needed in read_super_common(), | |
226 | * convert them firstly. | |
227 | */ | |
d721a43f CL |
228 | sb->feature_compat = le64_to_cpu(s->feature_compat); |
229 | sb->feature_incompat = le64_to_cpu(s->feature_incompat); | |
230 | sb->feature_ro_compat = le64_to_cpu(s->feature_ro_compat); | |
ffa47032 CL |
231 | err = read_super_common(sb, bdev, s); |
232 | if (err) | |
2903381f | 233 | goto err; |
2903381f KO |
234 | break; |
235 | default: | |
236 | err = "Unsupported superblock version"; | |
cafe5635 | 237 | goto err; |
2903381f KO |
238 | } |
239 | ||
75cbb3f1 | 240 | sb->last_mount = (u32)ktime_get_real_seconds(); |
cfa0c56d | 241 | *res = s; |
6321bef0 | 242 | return NULL; |
cafe5635 | 243 | err: |
6321bef0 | 244 | put_page(page); |
cafe5635 KO |
245 | return err; |
246 | } | |
247 | ||
4246a0b6 | 248 | static void write_bdev_super_endio(struct bio *bio) |
cafe5635 KO |
249 | { |
250 | struct cached_dev *dc = bio->bi_private; | |
08ec1e62 CL |
251 | |
252 | if (bio->bi_status) | |
253 | bch_count_backing_io_errors(dc, bio); | |
cafe5635 | 254 | |
cb7a583e | 255 | closure_put(&dc->sb_write); |
cafe5635 KO |
256 | } |
257 | ||
475389ae CH |
258 | static void __write_super(struct cache_sb *sb, struct cache_sb_disk *out, |
259 | struct bio *bio) | |
cafe5635 | 260 | { |
6f10f7d1 | 261 | unsigned int i; |
cafe5635 | 262 | |
475389ae | 263 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META; |
4f024f37 | 264 | bio->bi_iter.bi_sector = SB_SECTOR; |
475389ae CH |
265 | __bio_add_page(bio, virt_to_page(out), SB_SIZE, |
266 | offset_in_page(out)); | |
cafe5635 KO |
267 | |
268 | out->offset = cpu_to_le64(sb->offset); | |
cafe5635 KO |
269 | |
270 | memcpy(out->uuid, sb->uuid, 16); | |
271 | memcpy(out->set_uuid, sb->set_uuid, 16); | |
272 | memcpy(out->label, sb->label, SB_LABEL_SIZE); | |
273 | ||
274 | out->flags = cpu_to_le64(sb->flags); | |
275 | out->seq = cpu_to_le64(sb->seq); | |
276 | ||
277 | out->last_mount = cpu_to_le32(sb->last_mount); | |
278 | out->first_bucket = cpu_to_le16(sb->first_bucket); | |
279 | out->keys = cpu_to_le16(sb->keys); | |
280 | ||
281 | for (i = 0; i < sb->keys; i++) | |
282 | out->d[i] = cpu_to_le64(sb->d[i]); | |
283 | ||
d721a43f CL |
284 | if (sb->version >= BCACHE_SB_VERSION_CDEV_WITH_FEATURES) { |
285 | out->feature_compat = cpu_to_le64(sb->feature_compat); | |
286 | out->feature_incompat = cpu_to_le64(sb->feature_incompat); | |
287 | out->feature_ro_compat = cpu_to_le64(sb->feature_ro_compat); | |
288 | } | |
289 | ||
290 | out->version = cpu_to_le64(sb->version); | |
cafe5635 KO |
291 | out->csum = csum_set(out); |
292 | ||
46f5aa88 | 293 | pr_debug("ver %llu, flags %llu, seq %llu\n", |
cafe5635 KO |
294 | sb->version, sb->flags, sb->seq); |
295 | ||
4e49ea4a | 296 | submit_bio(bio); |
cafe5635 KO |
297 | } |
298 | ||
cb7a583e KO |
299 | static void bch_write_bdev_super_unlock(struct closure *cl) |
300 | { | |
301 | struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write); | |
302 | ||
303 | up(&dc->sb_write_mutex); | |
304 | } | |
305 | ||
cafe5635 KO |
306 | void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent) |
307 | { | |
cb7a583e | 308 | struct closure *cl = &dc->sb_write; |
cafe5635 KO |
309 | struct bio *bio = &dc->sb_bio; |
310 | ||
cb7a583e KO |
311 | down(&dc->sb_write_mutex); |
312 | closure_init(cl, parent); | |
cafe5635 | 313 | |
475389ae | 314 | bio_init(bio, dc->sb_bv, 1); |
74d46992 | 315 | bio_set_dev(bio, dc->bdev); |
cafe5635 KO |
316 | bio->bi_end_io = write_bdev_super_endio; |
317 | bio->bi_private = dc; | |
318 | ||
319 | closure_get(cl); | |
27a40ab9 | 320 | /* I/O request sent to backing device */ |
475389ae | 321 | __write_super(&dc->sb, dc->sb_disk, bio); |
cafe5635 | 322 | |
cb7a583e | 323 | closure_return_with_destructor(cl, bch_write_bdev_super_unlock); |
cafe5635 KO |
324 | } |
325 | ||
4246a0b6 | 326 | static void write_super_endio(struct bio *bio) |
cafe5635 KO |
327 | { |
328 | struct cache *ca = bio->bi_private; | |
329 | ||
5138ac67 CL |
330 | /* is_read = 0 */ |
331 | bch_count_io_errors(ca, bio->bi_status, 0, | |
332 | "writing superblock"); | |
cb7a583e KO |
333 | closure_put(&ca->set->sb_write); |
334 | } | |
335 | ||
336 | static void bcache_write_super_unlock(struct closure *cl) | |
337 | { | |
338 | struct cache_set *c = container_of(cl, struct cache_set, sb_write); | |
339 | ||
340 | up(&c->sb_write_mutex); | |
cafe5635 KO |
341 | } |
342 | ||
343 | void bcache_write_super(struct cache_set *c) | |
344 | { | |
cb7a583e | 345 | struct closure *cl = &c->sb_write; |
08fdb2cd CL |
346 | struct cache *ca = c->cache; |
347 | struct bio *bio = &ca->sb_bio; | |
348 | unsigned int version = BCACHE_SB_VERSION_CDEV_WITH_UUID; | |
cafe5635 | 349 | |
cb7a583e KO |
350 | down(&c->sb_write_mutex); |
351 | closure_init(cl, &c->cl); | |
cafe5635 | 352 | |
4a784266 | 353 | ca->sb.seq++; |
cafe5635 | 354 | |
4a784266 CL |
355 | if (ca->sb.version < version) |
356 | ca->sb.version = version; | |
cafe5635 | 357 | |
08fdb2cd CL |
358 | bio_init(bio, ca->sb_bv, 1); |
359 | bio_set_dev(bio, ca->bdev); | |
360 | bio->bi_end_io = write_super_endio; | |
361 | bio->bi_private = ca; | |
cafe5635 | 362 | |
08fdb2cd CL |
363 | closure_get(cl); |
364 | __write_super(&ca->sb, ca->sb_disk, bio); | |
cafe5635 | 365 | |
cb7a583e | 366 | closure_return_with_destructor(cl, bcache_write_super_unlock); |
cafe5635 KO |
367 | } |
368 | ||
369 | /* UUID io */ | |
370 | ||
4246a0b6 | 371 | static void uuid_endio(struct bio *bio) |
cafe5635 KO |
372 | { |
373 | struct closure *cl = bio->bi_private; | |
cb7a583e | 374 | struct cache_set *c = container_of(cl, struct cache_set, uuid_write); |
cafe5635 | 375 | |
4e4cbee9 | 376 | cache_set_err_on(bio->bi_status, c, "accessing uuids"); |
cafe5635 KO |
377 | bch_bbio_free(bio, c); |
378 | closure_put(cl); | |
379 | } | |
380 | ||
cb7a583e KO |
381 | static void uuid_io_unlock(struct closure *cl) |
382 | { | |
383 | struct cache_set *c = container_of(cl, struct cache_set, uuid_write); | |
384 | ||
385 | up(&c->uuid_write_mutex); | |
386 | } | |
387 | ||
ad0d9e76 | 388 | static void uuid_io(struct cache_set *c, int op, unsigned long op_flags, |
cafe5635 KO |
389 | struct bkey *k, struct closure *parent) |
390 | { | |
cb7a583e | 391 | struct closure *cl = &c->uuid_write; |
cafe5635 | 392 | struct uuid_entry *u; |
6f10f7d1 | 393 | unsigned int i; |
85b1492e | 394 | char buf[80]; |
cafe5635 KO |
395 | |
396 | BUG_ON(!parent); | |
cb7a583e KO |
397 | down(&c->uuid_write_mutex); |
398 | closure_init(cl, parent); | |
cafe5635 KO |
399 | |
400 | for (i = 0; i < KEY_PTRS(k); i++) { | |
401 | struct bio *bio = bch_bbio_alloc(c); | |
402 | ||
1eff9d32 | 403 | bio->bi_opf = REQ_SYNC | REQ_META | op_flags; |
4f024f37 | 404 | bio->bi_iter.bi_size = KEY_SIZE(k) << 9; |
cafe5635 KO |
405 | |
406 | bio->bi_end_io = uuid_endio; | |
407 | bio->bi_private = cl; | |
ad0d9e76 | 408 | bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags); |
169ef1cf | 409 | bch_bio_map(bio, c->uuids); |
cafe5635 KO |
410 | |
411 | bch_submit_bbio(bio, c, k, i); | |
412 | ||
ad0d9e76 | 413 | if (op != REQ_OP_WRITE) |
cafe5635 KO |
414 | break; |
415 | } | |
416 | ||
dc9d98d6 | 417 | bch_extent_to_text(buf, sizeof(buf), k); |
46f5aa88 | 418 | pr_debug("%s UUIDs at %s\n", op == REQ_OP_WRITE ? "wrote" : "read", buf); |
cafe5635 KO |
419 | |
420 | for (u = c->uuids; u < c->uuids + c->nr_uuids; u++) | |
169ef1cf | 421 | if (!bch_is_zero(u->uuid, 16)) |
46f5aa88 | 422 | pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u\n", |
cafe5635 KO |
423 | u - c->uuids, u->uuid, u->label, |
424 | u->first_reg, u->last_reg, u->invalidated); | |
425 | ||
cb7a583e | 426 | closure_return_with_destructor(cl, uuid_io_unlock); |
cafe5635 KO |
427 | } |
428 | ||
429 | static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl) | |
430 | { | |
431 | struct bkey *k = &j->uuid_bucket; | |
432 | ||
65d45231 | 433 | if (__bch_btree_ptr_invalid(c, k)) |
cafe5635 KO |
434 | return "bad uuid pointer"; |
435 | ||
436 | bkey_copy(&c->uuid_bucket, k); | |
70fd7614 | 437 | uuid_io(c, REQ_OP_READ, 0, k, cl); |
cafe5635 KO |
438 | |
439 | if (j->version < BCACHE_JSET_VERSION_UUIDv1) { | |
440 | struct uuid_entry_v0 *u0 = (void *) c->uuids; | |
441 | struct uuid_entry *u1 = (void *) c->uuids; | |
442 | int i; | |
443 | ||
444 | closure_sync(cl); | |
445 | ||
446 | /* | |
447 | * Since the new uuid entry is bigger than the old, we have to | |
448 | * convert starting at the highest memory address and work down | |
449 | * in order to do it in place | |
450 | */ | |
451 | ||
452 | for (i = c->nr_uuids - 1; | |
453 | i >= 0; | |
454 | --i) { | |
455 | memcpy(u1[i].uuid, u0[i].uuid, 16); | |
456 | memcpy(u1[i].label, u0[i].label, 32); | |
457 | ||
458 | u1[i].first_reg = u0[i].first_reg; | |
459 | u1[i].last_reg = u0[i].last_reg; | |
460 | u1[i].invalidated = u0[i].invalidated; | |
461 | ||
462 | u1[i].flags = 0; | |
463 | u1[i].sectors = 0; | |
464 | } | |
465 | } | |
466 | ||
467 | return NULL; | |
468 | } | |
469 | ||
470 | static int __uuid_write(struct cache_set *c) | |
471 | { | |
472 | BKEY_PADDED(key) k; | |
473 | struct closure cl; | |
4a784266 | 474 | struct cache *ca = c->cache; |
21e478dd | 475 | unsigned int size; |
cafe5635 | 476 | |
1fae7cf0 | 477 | closure_init_stack(&cl); |
cafe5635 KO |
478 | lockdep_assert_held(&bch_register_lock); |
479 | ||
17e4aed8 | 480 | if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, true)) |
cafe5635 KO |
481 | return 1; |
482 | ||
4a784266 | 483 | size = meta_bucket_pages(&ca->sb) * PAGE_SECTORS; |
21e478dd | 484 | SET_KEY_SIZE(&k.key, size); |
ad0d9e76 | 485 | uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl); |
cafe5635 KO |
486 | closure_sync(&cl); |
487 | ||
7a55948d | 488 | /* Only one bucket used for uuid write */ |
7a55948d SW |
489 | atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written); |
490 | ||
cafe5635 | 491 | bkey_copy(&c->uuid_bucket, &k.key); |
3a3b6a4e | 492 | bkey_put(c, &k.key); |
cafe5635 KO |
493 | return 0; |
494 | } | |
495 | ||
496 | int bch_uuid_write(struct cache_set *c) | |
497 | { | |
498 | int ret = __uuid_write(c); | |
499 | ||
500 | if (!ret) | |
501 | bch_journal_meta(c, NULL); | |
502 | ||
503 | return ret; | |
504 | } | |
505 | ||
506 | static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid) | |
507 | { | |
508 | struct uuid_entry *u; | |
509 | ||
510 | for (u = c->uuids; | |
511 | u < c->uuids + c->nr_uuids; u++) | |
512 | if (!memcmp(u->uuid, uuid, 16)) | |
513 | return u; | |
514 | ||
515 | return NULL; | |
516 | } | |
517 | ||
518 | static struct uuid_entry *uuid_find_empty(struct cache_set *c) | |
519 | { | |
520 | static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"; | |
1fae7cf0 | 521 | |
cafe5635 KO |
522 | return uuid_find(c, zero_uuid); |
523 | } | |
524 | ||
525 | /* | |
526 | * Bucket priorities/gens: | |
527 | * | |
528 | * For each bucket, we store on disk its | |
3be11dba CL |
529 | * 8 bit gen |
530 | * 16 bit priority | |
cafe5635 KO |
531 | * |
532 | * See alloc.c for an explanation of the gen. The priority is used to implement | |
533 | * lru (and in the future other) cache replacement policies; for most purposes | |
534 | * it's just an opaque integer. | |
535 | * | |
536 | * The gens and the priorities don't have a whole lot to do with each other, and | |
537 | * it's actually the gens that must be written out at specific times - it's no | |
538 | * big deal if the priorities don't get written, if we lose them we just reuse | |
539 | * buckets in suboptimal order. | |
540 | * | |
541 | * On disk they're stored in a packed array, and in as many buckets are required | |
542 | * to fit them all. The buckets we use to store them form a list; the journal | |
543 | * header points to the first bucket, the first bucket points to the second | |
544 | * bucket, et cetera. | |
545 | * | |
546 | * This code is used by the allocation code; periodically (whenever it runs out | |
547 | * of buckets to allocate from) the allocation code will invalidate some | |
548 | * buckets, but it can't use those buckets until their new gens are safely on | |
549 | * disk. | |
550 | */ | |
551 | ||
4246a0b6 | 552 | static void prio_endio(struct bio *bio) |
cafe5635 KO |
553 | { |
554 | struct cache *ca = bio->bi_private; | |
555 | ||
4e4cbee9 | 556 | cache_set_err_on(bio->bi_status, ca->set, "accessing priorities"); |
cafe5635 KO |
557 | bch_bbio_free(bio, ca->set); |
558 | closure_put(&ca->prio); | |
559 | } | |
560 | ||
ad0d9e76 MC |
561 | static void prio_io(struct cache *ca, uint64_t bucket, int op, |
562 | unsigned long op_flags) | |
cafe5635 KO |
563 | { |
564 | struct closure *cl = &ca->prio; | |
565 | struct bio *bio = bch_bbio_alloc(ca->set); | |
566 | ||
567 | closure_init_stack(cl); | |
568 | ||
4f024f37 | 569 | bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size; |
74d46992 | 570 | bio_set_dev(bio, ca->bdev); |
c954ac8d | 571 | bio->bi_iter.bi_size = meta_bucket_bytes(&ca->sb); |
cafe5635 KO |
572 | |
573 | bio->bi_end_io = prio_endio; | |
574 | bio->bi_private = ca; | |
ad0d9e76 | 575 | bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags); |
169ef1cf | 576 | bch_bio_map(bio, ca->disk_buckets); |
cafe5635 | 577 | |
771f393e | 578 | closure_bio_submit(ca->set, bio, &ca->prio); |
cafe5635 KO |
579 | closure_sync(cl); |
580 | } | |
581 | ||
84c529ae | 582 | int bch_prio_write(struct cache *ca, bool wait) |
cafe5635 KO |
583 | { |
584 | int i; | |
585 | struct bucket *b; | |
586 | struct closure cl; | |
587 | ||
46f5aa88 | 588 | pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu\n", |
84c529ae AR |
589 | fifo_used(&ca->free[RESERVE_PRIO]), |
590 | fifo_used(&ca->free[RESERVE_NONE]), | |
591 | fifo_used(&ca->free_inc)); | |
592 | ||
593 | /* | |
594 | * Pre-check if there are enough free buckets. In the non-blocking | |
595 | * scenario it's better to fail early rather than starting to allocate | |
596 | * buckets and do a cleanup later in case of failure. | |
597 | */ | |
598 | if (!wait) { | |
599 | size_t avail = fifo_used(&ca->free[RESERVE_PRIO]) + | |
600 | fifo_used(&ca->free[RESERVE_NONE]); | |
601 | if (prio_buckets(ca) > avail) | |
602 | return -ENOMEM; | |
603 | } | |
604 | ||
cafe5635 KO |
605 | closure_init_stack(&cl); |
606 | ||
607 | lockdep_assert_held(&ca->set->bucket_lock); | |
608 | ||
cafe5635 KO |
609 | ca->disk_buckets->seq++; |
610 | ||
611 | atomic_long_add(ca->sb.bucket_size * prio_buckets(ca), | |
612 | &ca->meta_sectors_written); | |
613 | ||
cafe5635 KO |
614 | for (i = prio_buckets(ca) - 1; i >= 0; --i) { |
615 | long bucket; | |
616 | struct prio_set *p = ca->disk_buckets; | |
b1a67b0f KO |
617 | struct bucket_disk *d = p->data; |
618 | struct bucket_disk *end = d + prios_per_bucket(ca); | |
cafe5635 KO |
619 | |
620 | for (b = ca->buckets + i * prios_per_bucket(ca); | |
621 | b < ca->buckets + ca->sb.nbuckets && d < end; | |
622 | b++, d++) { | |
623 | d->prio = cpu_to_le16(b->prio); | |
624 | d->gen = b->gen; | |
625 | } | |
626 | ||
627 | p->next_bucket = ca->prio_buckets[i + 1]; | |
81ab4190 | 628 | p->magic = pset_magic(&ca->sb); |
c954ac8d | 629 | p->csum = bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8); |
cafe5635 | 630 | |
84c529ae | 631 | bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait); |
cafe5635 KO |
632 | BUG_ON(bucket == -1); |
633 | ||
634 | mutex_unlock(&ca->set->bucket_lock); | |
ad0d9e76 | 635 | prio_io(ca, bucket, REQ_OP_WRITE, 0); |
cafe5635 KO |
636 | mutex_lock(&ca->set->bucket_lock); |
637 | ||
638 | ca->prio_buckets[i] = bucket; | |
639 | atomic_dec_bug(&ca->buckets[bucket].pin); | |
640 | } | |
641 | ||
642 | mutex_unlock(&ca->set->bucket_lock); | |
643 | ||
644 | bch_journal_meta(ca->set, &cl); | |
645 | closure_sync(&cl); | |
646 | ||
647 | mutex_lock(&ca->set->bucket_lock); | |
648 | ||
cafe5635 KO |
649 | /* |
650 | * Don't want the old priorities to get garbage collected until after we | |
651 | * finish writing the new ones, and they're journalled | |
652 | */ | |
2531d9ee KO |
653 | for (i = 0; i < prio_buckets(ca); i++) { |
654 | if (ca->prio_last_buckets[i]) | |
655 | __bch_bucket_free(ca, | |
656 | &ca->buckets[ca->prio_last_buckets[i]]); | |
657 | ||
cafe5635 | 658 | ca->prio_last_buckets[i] = ca->prio_buckets[i]; |
2531d9ee | 659 | } |
84c529ae | 660 | return 0; |
cafe5635 KO |
661 | } |
662 | ||
49d08d59 | 663 | static int prio_read(struct cache *ca, uint64_t bucket) |
cafe5635 KO |
664 | { |
665 | struct prio_set *p = ca->disk_buckets; | |
666 | struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d; | |
667 | struct bucket *b; | |
6f10f7d1 | 668 | unsigned int bucket_nr = 0; |
49d08d59 | 669 | int ret = -EIO; |
cafe5635 KO |
670 | |
671 | for (b = ca->buckets; | |
672 | b < ca->buckets + ca->sb.nbuckets; | |
673 | b++, d++) { | |
674 | if (d == end) { | |
675 | ca->prio_buckets[bucket_nr] = bucket; | |
676 | ca->prio_last_buckets[bucket_nr] = bucket; | |
677 | bucket_nr++; | |
678 | ||
70fd7614 | 679 | prio_io(ca, bucket, REQ_OP_READ, 0); |
cafe5635 | 680 | |
b0d30981 | 681 | if (p->csum != |
c954ac8d | 682 | bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8)) { |
46f5aa88 | 683 | pr_warn("bad csum reading priorities\n"); |
49d08d59 CL |
684 | goto out; |
685 | } | |
cafe5635 | 686 | |
49d08d59 | 687 | if (p->magic != pset_magic(&ca->sb)) { |
46f5aa88 | 688 | pr_warn("bad magic reading priorities\n"); |
49d08d59 CL |
689 | goto out; |
690 | } | |
cafe5635 KO |
691 | |
692 | bucket = p->next_bucket; | |
693 | d = p->data; | |
694 | } | |
695 | ||
696 | b->prio = le16_to_cpu(d->prio); | |
3a2fd9d5 | 697 | b->gen = b->last_gc = d->gen; |
cafe5635 | 698 | } |
49d08d59 CL |
699 | |
700 | ret = 0; | |
701 | out: | |
702 | return ret; | |
cafe5635 KO |
703 | } |
704 | ||
705 | /* Bcache device */ | |
706 | ||
707 | static int open_dev(struct block_device *b, fmode_t mode) | |
708 | { | |
709 | struct bcache_device *d = b->bd_disk->private_data; | |
1fae7cf0 | 710 | |
c4d951dd | 711 | if (test_bit(BCACHE_DEV_CLOSING, &d->flags)) |
cafe5635 KO |
712 | return -ENXIO; |
713 | ||
714 | closure_get(&d->cl); | |
715 | return 0; | |
716 | } | |
717 | ||
867e1162 | 718 | static void release_dev(struct gendisk *b, fmode_t mode) |
cafe5635 KO |
719 | { |
720 | struct bcache_device *d = b->private_data; | |
1fae7cf0 | 721 | |
cafe5635 | 722 | closure_put(&d->cl); |
cafe5635 KO |
723 | } |
724 | ||
725 | static int ioctl_dev(struct block_device *b, fmode_t mode, | |
726 | unsigned int cmd, unsigned long arg) | |
727 | { | |
728 | struct bcache_device *d = b->bd_disk->private_data; | |
0f0709e6 | 729 | |
cafe5635 KO |
730 | return d->ioctl(d, mode, cmd, arg); |
731 | } | |
732 | ||
c62b37d9 CH |
733 | static const struct block_device_operations bcache_cached_ops = { |
734 | .submit_bio = cached_dev_submit_bio, | |
735 | .open = open_dev, | |
736 | .release = release_dev, | |
737 | .ioctl = ioctl_dev, | |
738 | .owner = THIS_MODULE, | |
739 | }; | |
740 | ||
741 | static const struct block_device_operations bcache_flash_ops = { | |
742 | .submit_bio = flash_dev_submit_bio, | |
cafe5635 KO |
743 | .open = open_dev, |
744 | .release = release_dev, | |
745 | .ioctl = ioctl_dev, | |
746 | .owner = THIS_MODULE, | |
747 | }; | |
748 | ||
749 | void bcache_device_stop(struct bcache_device *d) | |
750 | { | |
c4d951dd | 751 | if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags)) |
63d63b51 CL |
752 | /* |
753 | * closure_fn set to | |
754 | * - cached device: cached_dev_flush() | |
755 | * - flash dev: flash_dev_flush() | |
756 | */ | |
cafe5635 KO |
757 | closure_queue(&d->cl); |
758 | } | |
759 | ||
ee668506 KO |
760 | static void bcache_device_unlink(struct bcache_device *d) |
761 | { | |
c4d951dd | 762 | lockdep_assert_held(&bch_register_lock); |
ee668506 | 763 | |
c4d951dd | 764 | if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) { |
08fdb2cd | 765 | struct cache *ca = d->c->cache; |
ee668506 | 766 | |
c4d951dd KO |
767 | sysfs_remove_link(&d->c->kobj, d->name); |
768 | sysfs_remove_link(&d->kobj, "cache"); | |
769 | ||
08fdb2cd | 770 | bd_unlink_disk_holder(ca->bdev, d->disk); |
c4d951dd | 771 | } |
ee668506 KO |
772 | } |
773 | ||
774 | static void bcache_device_link(struct bcache_device *d, struct cache_set *c, | |
775 | const char *name) | |
776 | { | |
08fdb2cd | 777 | struct cache *ca = c->cache; |
4b6efb4b | 778 | int ret; |
ee668506 | 779 | |
08fdb2cd | 780 | bd_link_disk_holder(ca->bdev, d->disk); |
ee668506 KO |
781 | |
782 | snprintf(d->name, BCACHEDEVNAME_SIZE, | |
783 | "%s%u", name, d->id); | |
784 | ||
4b6efb4b CL |
785 | ret = sysfs_create_link(&d->kobj, &c->kobj, "cache"); |
786 | if (ret < 0) | |
46f5aa88 | 787 | pr_err("Couldn't create device -> cache set symlink\n"); |
4b6efb4b CL |
788 | |
789 | ret = sysfs_create_link(&c->kobj, &d->kobj, d->name); | |
790 | if (ret < 0) | |
46f5aa88 | 791 | pr_err("Couldn't create cache set -> device symlink\n"); |
fecaee6f ZL |
792 | |
793 | clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags); | |
ee668506 KO |
794 | } |
795 | ||
cafe5635 KO |
796 | static void bcache_device_detach(struct bcache_device *d) |
797 | { | |
798 | lockdep_assert_held(&bch_register_lock); | |
799 | ||
ea8c5356 CL |
800 | atomic_dec(&d->c->attached_dev_nr); |
801 | ||
c4d951dd | 802 | if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) { |
cafe5635 KO |
803 | struct uuid_entry *u = d->c->uuids + d->id; |
804 | ||
805 | SET_UUID_FLASH_ONLY(u, 0); | |
806 | memcpy(u->uuid, invalid_uuid, 16); | |
75cbb3f1 | 807 | u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 | 808 | bch_uuid_write(d->c); |
cafe5635 KO |
809 | } |
810 | ||
c4d951dd | 811 | bcache_device_unlink(d); |
ee668506 | 812 | |
cafe5635 KO |
813 | d->c->devices[d->id] = NULL; |
814 | closure_put(&d->c->caching); | |
815 | d->c = NULL; | |
816 | } | |
817 | ||
818 | static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, | |
6f10f7d1 | 819 | unsigned int id) |
cafe5635 | 820 | { |
cafe5635 KO |
821 | d->id = id; |
822 | d->c = c; | |
823 | c->devices[id] = d; | |
824 | ||
2831231d CL |
825 | if (id >= c->devices_max_used) |
826 | c->devices_max_used = id + 1; | |
827 | ||
cafe5635 KO |
828 | closure_get(&c->caching); |
829 | } | |
830 | ||
1dbe32ad CL |
831 | static inline int first_minor_to_idx(int first_minor) |
832 | { | |
833 | return (first_minor/BCACHE_MINORS); | |
834 | } | |
835 | ||
836 | static inline int idx_to_first_minor(int idx) | |
837 | { | |
838 | return (idx * BCACHE_MINORS); | |
839 | } | |
840 | ||
cafe5635 KO |
841 | static void bcache_device_free(struct bcache_device *d) |
842 | { | |
2d886951 CL |
843 | struct gendisk *disk = d->disk; |
844 | ||
cafe5635 KO |
845 | lockdep_assert_held(&bch_register_lock); |
846 | ||
2d886951 | 847 | if (disk) |
46f5aa88 | 848 | pr_info("%s stopped\n", disk->disk_name); |
2d886951 | 849 | else |
46f5aa88 | 850 | pr_err("bcache device (NULL gendisk) stopped\n"); |
cafe5635 KO |
851 | |
852 | if (d->c) | |
853 | bcache_device_detach(d); | |
2d886951 CL |
854 | |
855 | if (disk) { | |
86da9f73 CL |
856 | bool disk_added = (disk->flags & GENHD_FL_UP) != 0; |
857 | ||
858 | if (disk_added) | |
2d886951 CL |
859 | del_gendisk(disk); |
860 | ||
861 | if (disk->queue) | |
862 | blk_cleanup_queue(disk->queue); | |
863 | ||
1dbe32ad | 864 | ida_simple_remove(&bcache_device_idx, |
2d886951 | 865 | first_minor_to_idx(disk->first_minor)); |
86da9f73 CL |
866 | if (disk_added) |
867 | put_disk(disk); | |
28935ab5 | 868 | } |
cafe5635 | 869 | |
d19936a2 | 870 | bioset_exit(&d->bio_split); |
958b4338 PE |
871 | kvfree(d->full_dirty_stripes); |
872 | kvfree(d->stripe_sectors_dirty); | |
cafe5635 KO |
873 | |
874 | closure_debug_destroy(&d->cl); | |
875 | } | |
876 | ||
6f10f7d1 | 877 | static int bcache_device_init(struct bcache_device *d, unsigned int block_size, |
c62b37d9 CH |
878 | sector_t sectors, struct block_device *cached_bdev, |
879 | const struct block_device_operations *ops) | |
cafe5635 KO |
880 | { |
881 | struct request_queue *q; | |
5f2b18ec BVA |
882 | const size_t max_stripes = min_t(size_t, INT_MAX, |
883 | SIZE_MAX / sizeof(atomic_t)); | |
65f0f017 | 884 | uint64_t n; |
1dbe32ad | 885 | int idx; |
279afbad | 886 | |
2d679fc7 KO |
887 | if (!d->stripe_size) |
888 | d->stripe_size = 1 << 31; | |
279afbad | 889 | |
65f0f017 CL |
890 | n = DIV_ROUND_UP_ULL(sectors, d->stripe_size); |
891 | if (!n || n > max_stripes) { | |
892 | pr_err("nr_stripes too large or invalid: %llu (start sector beyond end of disk?)\n", | |
893 | n); | |
279afbad | 894 | return -ENOMEM; |
48a915a8 | 895 | } |
65f0f017 | 896 | d->nr_stripes = n; |
279afbad KO |
897 | |
898 | n = d->nr_stripes * sizeof(atomic_t); | |
bc4e54f6 | 899 | d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL); |
279afbad KO |
900 | if (!d->stripe_sectors_dirty) |
901 | return -ENOMEM; | |
cafe5635 | 902 | |
48a915a8 | 903 | n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long); |
bc4e54f6 | 904 | d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL); |
48a915a8 KO |
905 | if (!d->full_dirty_stripes) |
906 | return -ENOMEM; | |
907 | ||
1dbe32ad CL |
908 | idx = ida_simple_get(&bcache_device_idx, 0, |
909 | BCACHE_DEVICE_IDX_MAX, GFP_KERNEL); | |
910 | if (idx < 0) | |
911 | return idx; | |
b8c0d911 | 912 | |
d19936a2 | 913 | if (bioset_init(&d->bio_split, 4, offsetof(struct bbio, bio), |
9b4e9f5a FS |
914 | BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER)) |
915 | goto err; | |
916 | ||
917 | d->disk = alloc_disk(BCACHE_MINORS); | |
918 | if (!d->disk) | |
919 | goto err; | |
cafe5635 | 920 | |
279afbad | 921 | set_capacity(d->disk, sectors); |
1dbe32ad | 922 | snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx); |
cafe5635 KO |
923 | |
924 | d->disk->major = bcache_major; | |
1dbe32ad | 925 | d->disk->first_minor = idx_to_first_minor(idx); |
c62b37d9 | 926 | d->disk->fops = ops; |
cafe5635 KO |
927 | d->disk->private_data = d; |
928 | ||
c62b37d9 | 929 | q = blk_alloc_queue(NUMA_NO_NODE); |
28935ab5 KO |
930 | if (!q) |
931 | return -ENOMEM; | |
932 | ||
cafe5635 | 933 | d->disk->queue = q; |
cafe5635 KO |
934 | q->limits.max_hw_sectors = UINT_MAX; |
935 | q->limits.max_sectors = UINT_MAX; | |
936 | q->limits.max_segment_size = UINT_MAX; | |
937 | q->limits.max_segments = BIO_MAX_PAGES; | |
2bb4cd5c | 938 | blk_queue_max_discard_sectors(q, UINT_MAX); |
90db6919 | 939 | q->limits.discard_granularity = 512; |
cafe5635 KO |
940 | q->limits.io_min = block_size; |
941 | q->limits.logical_block_size = block_size; | |
942 | q->limits.physical_block_size = block_size; | |
dcacbc12 MFO |
943 | |
944 | if (q->limits.logical_block_size > PAGE_SIZE && cached_bdev) { | |
945 | /* | |
946 | * This should only happen with BCACHE_SB_VERSION_BDEV. | |
947 | * Block/page size is checked for BCACHE_SB_VERSION_CDEV. | |
948 | */ | |
4b25bbf5 | 949 | pr_info("%s: sb/logical block size (%u) greater than page size (%lu) falling back to device logical block size (%u)\n", |
dcacbc12 MFO |
950 | d->disk->disk_name, q->limits.logical_block_size, |
951 | PAGE_SIZE, bdev_logical_block_size(cached_bdev)); | |
952 | ||
953 | /* This also adjusts physical block size/min io size if needed */ | |
954 | blk_queue_logical_block_size(q, bdev_logical_block_size(cached_bdev)); | |
955 | } | |
956 | ||
44e1ebe2 BVA |
957 | blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue); |
958 | blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, d->disk->queue); | |
959 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, d->disk->queue); | |
cafe5635 | 960 | |
84b4ff9e | 961 | blk_queue_write_cache(q, true, true); |
54d12f2b | 962 | |
cafe5635 | 963 | return 0; |
9b4e9f5a FS |
964 | |
965 | err: | |
966 | ida_simple_remove(&bcache_device_idx, idx); | |
967 | return -ENOMEM; | |
968 | ||
cafe5635 KO |
969 | } |
970 | ||
971 | /* Cached device */ | |
972 | ||
973 | static void calc_cached_dev_sectors(struct cache_set *c) | |
974 | { | |
975 | uint64_t sectors = 0; | |
976 | struct cached_dev *dc; | |
977 | ||
978 | list_for_each_entry(dc, &c->cached_devs, list) | |
979 | sectors += bdev_sectors(dc->bdev); | |
980 | ||
981 | c->cached_dev_sectors = sectors; | |
982 | } | |
983 | ||
0f0709e6 CL |
984 | #define BACKING_DEV_OFFLINE_TIMEOUT 5 |
985 | static int cached_dev_status_update(void *arg) | |
986 | { | |
987 | struct cached_dev *dc = arg; | |
988 | struct request_queue *q; | |
989 | ||
990 | /* | |
991 | * If this delayed worker is stopping outside, directly quit here. | |
992 | * dc->io_disable might be set via sysfs interface, so check it | |
993 | * here too. | |
994 | */ | |
995 | while (!kthread_should_stop() && !dc->io_disable) { | |
996 | q = bdev_get_queue(dc->bdev); | |
997 | if (blk_queue_dying(q)) | |
998 | dc->offline_seconds++; | |
999 | else | |
1000 | dc->offline_seconds = 0; | |
1001 | ||
1002 | if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) { | |
46f5aa88 | 1003 | pr_err("%s: device offline for %d seconds\n", |
0f0709e6 CL |
1004 | dc->backing_dev_name, |
1005 | BACKING_DEV_OFFLINE_TIMEOUT); | |
46f5aa88 JP |
1006 | pr_err("%s: disable I/O request due to backing device offline\n", |
1007 | dc->disk.name); | |
0f0709e6 CL |
1008 | dc->io_disable = true; |
1009 | /* let others know earlier that io_disable is true */ | |
1010 | smp_mb(); | |
1011 | bcache_device_stop(&dc->disk); | |
1012 | break; | |
1013 | } | |
1014 | schedule_timeout_interruptible(HZ); | |
1015 | } | |
1016 | ||
1017 | wait_for_kthread_stop(); | |
1018 | return 0; | |
1019 | } | |
1020 | ||
1021 | ||
0b13efec | 1022 | int bch_cached_dev_run(struct cached_dev *dc) |
cafe5635 KO |
1023 | { |
1024 | struct bcache_device *d = &dc->disk; | |
792732d9 | 1025 | char *buf = kmemdup_nul(dc->sb.label, SB_LABEL_SIZE, GFP_KERNEL); |
a25c32be GP |
1026 | char *env[] = { |
1027 | "DRIVER=bcache", | |
1028 | kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), | |
792732d9 | 1029 | kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf ? : ""), |
ab9e1400 | 1030 | NULL, |
a25c32be | 1031 | }; |
cafe5635 | 1032 | |
e0faa3d7 | 1033 | if (dc->io_disable) { |
46f5aa88 | 1034 | pr_err("I/O disabled on cached dev %s\n", |
e0faa3d7 | 1035 | dc->backing_dev_name); |
5d9e06d6 WY |
1036 | kfree(env[1]); |
1037 | kfree(env[2]); | |
1038 | kfree(buf); | |
0b13efec | 1039 | return -EIO; |
e0faa3d7 | 1040 | } |
0b13efec | 1041 | |
4d4d8573 AV |
1042 | if (atomic_xchg(&dc->running, 1)) { |
1043 | kfree(env[1]); | |
1044 | kfree(env[2]); | |
792732d9 | 1045 | kfree(buf); |
46f5aa88 | 1046 | pr_info("cached dev %s is running already\n", |
e0faa3d7 | 1047 | dc->backing_dev_name); |
0b13efec | 1048 | return -EBUSY; |
4d4d8573 | 1049 | } |
cafe5635 KO |
1050 | |
1051 | if (!d->c && | |
1052 | BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { | |
1053 | struct closure cl; | |
1fae7cf0 | 1054 | |
cafe5635 KO |
1055 | closure_init_stack(&cl); |
1056 | ||
1057 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); | |
1058 | bch_write_bdev_super(dc, &cl); | |
1059 | closure_sync(&cl); | |
1060 | } | |
1061 | ||
1062 | add_disk(d->disk); | |
ee668506 | 1063 | bd_link_disk_holder(dc->bdev, dc->disk.disk); |
3be11dba CL |
1064 | /* |
1065 | * won't show up in the uevent file, use udevadm monitor -e instead | |
1066 | * only class / kset properties are persistent | |
1067 | */ | |
cafe5635 | 1068 | kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); |
a25c32be | 1069 | kfree(env[1]); |
ab9e1400 | 1070 | kfree(env[2]); |
792732d9 | 1071 | kfree(buf); |
a25c32be | 1072 | |
cafe5635 | 1073 | if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || |
0b13efec CL |
1074 | sysfs_create_link(&disk_to_dev(d->disk)->kobj, |
1075 | &d->kobj, "bcache")) { | |
46f5aa88 | 1076 | pr_err("Couldn't create bcache dev <-> disk sysfs symlinks\n"); |
0b13efec CL |
1077 | return -ENOMEM; |
1078 | } | |
0f0709e6 CL |
1079 | |
1080 | dc->status_update_thread = kthread_run(cached_dev_status_update, | |
1081 | dc, "bcache_status_update"); | |
1082 | if (IS_ERR(dc->status_update_thread)) { | |
46f5aa88 | 1083 | pr_warn("failed to create bcache_status_update kthread, continue to run without monitoring backing device status\n"); |
0f0709e6 | 1084 | } |
0b13efec CL |
1085 | |
1086 | return 0; | |
cafe5635 KO |
1087 | } |
1088 | ||
3fd47bfe CL |
1089 | /* |
1090 | * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed | |
1091 | * work dc->writeback_rate_update is running. Wait until the routine | |
1092 | * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to | |
1093 | * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out | |
1094 | * seconds, give up waiting here and continue to cancel it too. | |
1095 | */ | |
1096 | static void cancel_writeback_rate_update_dwork(struct cached_dev *dc) | |
1097 | { | |
1098 | int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ; | |
1099 | ||
1100 | do { | |
1101 | if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING, | |
1102 | &dc->disk.flags)) | |
1103 | break; | |
1104 | time_out--; | |
1105 | schedule_timeout_interruptible(1); | |
1106 | } while (time_out > 0); | |
1107 | ||
1108 | if (time_out == 0) | |
46f5aa88 | 1109 | pr_warn("give up waiting for dc->writeback_write_update to quit\n"); |
3fd47bfe CL |
1110 | |
1111 | cancel_delayed_work_sync(&dc->writeback_rate_update); | |
1112 | } | |
1113 | ||
cafe5635 KO |
1114 | static void cached_dev_detach_finish(struct work_struct *w) |
1115 | { | |
1116 | struct cached_dev *dc = container_of(w, struct cached_dev, detach); | |
cafe5635 | 1117 | struct closure cl; |
1fae7cf0 | 1118 | |
cafe5635 KO |
1119 | closure_init_stack(&cl); |
1120 | ||
c4d951dd | 1121 | BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)); |
3b304d24 | 1122 | BUG_ON(refcount_read(&dc->count)); |
cafe5635 | 1123 | |
cafe5635 | 1124 | |
3fd47bfe CL |
1125 | if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) |
1126 | cancel_writeback_rate_update_dwork(dc); | |
1127 | ||
8d29c442 TJ |
1128 | if (!IS_ERR_OR_NULL(dc->writeback_thread)) { |
1129 | kthread_stop(dc->writeback_thread); | |
1130 | dc->writeback_thread = NULL; | |
1131 | } | |
1132 | ||
cafe5635 KO |
1133 | memset(&dc->sb.set_uuid, 0, 16); |
1134 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE); | |
1135 | ||
1136 | bch_write_bdev_super(dc, &cl); | |
1137 | closure_sync(&cl); | |
1138 | ||
97ba3b81 CL |
1139 | mutex_lock(&bch_register_lock); |
1140 | ||
46010141 | 1141 | calc_cached_dev_sectors(dc->disk.c); |
cafe5635 KO |
1142 | bcache_device_detach(&dc->disk); |
1143 | list_move(&dc->list, &uncached_devices); | |
1144 | ||
c4d951dd | 1145 | clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags); |
5b1016e6 | 1146 | clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags); |
c4d951dd | 1147 | |
cafe5635 KO |
1148 | mutex_unlock(&bch_register_lock); |
1149 | ||
46f5aa88 | 1150 | pr_info("Caching disabled for %s\n", dc->backing_dev_name); |
cafe5635 KO |
1151 | |
1152 | /* Drop ref we took in cached_dev_detach() */ | |
1153 | closure_put(&dc->disk.cl); | |
1154 | } | |
1155 | ||
1156 | void bch_cached_dev_detach(struct cached_dev *dc) | |
1157 | { | |
1158 | lockdep_assert_held(&bch_register_lock); | |
1159 | ||
c4d951dd | 1160 | if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) |
cafe5635 KO |
1161 | return; |
1162 | ||
c4d951dd | 1163 | if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) |
cafe5635 KO |
1164 | return; |
1165 | ||
1166 | /* | |
1167 | * Block the device from being closed and freed until we're finished | |
1168 | * detaching | |
1169 | */ | |
1170 | closure_get(&dc->disk.cl); | |
1171 | ||
1172 | bch_writeback_queue(dc); | |
3fd47bfe | 1173 | |
cafe5635 KO |
1174 | cached_dev_put(dc); |
1175 | } | |
1176 | ||
73ac105b TJ |
1177 | int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c, |
1178 | uint8_t *set_uuid) | |
cafe5635 | 1179 | { |
75cbb3f1 | 1180 | uint32_t rtime = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 | 1181 | struct uuid_entry *u; |
86755b7a | 1182 | struct cached_dev *exist_dc, *t; |
0b13efec | 1183 | int ret = 0; |
cafe5635 | 1184 | |
1132e56e CL |
1185 | if ((set_uuid && memcmp(set_uuid, c->set_uuid, 16)) || |
1186 | (!set_uuid && memcmp(dc->sb.set_uuid, c->set_uuid, 16))) | |
cafe5635 KO |
1187 | return -ENOENT; |
1188 | ||
1189 | if (dc->disk.c) { | |
46f5aa88 | 1190 | pr_err("Can't attach %s: already attached\n", |
6e916a7e | 1191 | dc->backing_dev_name); |
cafe5635 KO |
1192 | return -EINVAL; |
1193 | } | |
1194 | ||
1195 | if (test_bit(CACHE_SET_STOPPING, &c->flags)) { | |
46f5aa88 | 1196 | pr_err("Can't attach %s: shutting down\n", |
6e916a7e | 1197 | dc->backing_dev_name); |
cafe5635 KO |
1198 | return -EINVAL; |
1199 | } | |
1200 | ||
4a784266 | 1201 | if (dc->sb.block_size < c->cache->sb.block_size) { |
cafe5635 | 1202 | /* Will die */ |
46f5aa88 | 1203 | pr_err("Couldn't attach %s: block size less than set's block size\n", |
6e916a7e | 1204 | dc->backing_dev_name); |
cafe5635 KO |
1205 | return -EINVAL; |
1206 | } | |
1207 | ||
86755b7a ML |
1208 | /* Check whether already attached */ |
1209 | list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) { | |
1210 | if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) { | |
46f5aa88 | 1211 | pr_err("Tried to attach %s but duplicate UUID already attached\n", |
6e916a7e | 1212 | dc->backing_dev_name); |
86755b7a ML |
1213 | |
1214 | return -EINVAL; | |
1215 | } | |
1216 | } | |
1217 | ||
cafe5635 KO |
1218 | u = uuid_find(c, dc->sb.uuid); |
1219 | ||
1220 | if (u && | |
1221 | (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || | |
1222 | BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { | |
1223 | memcpy(u->uuid, invalid_uuid, 16); | |
75cbb3f1 | 1224 | u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 KO |
1225 | u = NULL; |
1226 | } | |
1227 | ||
1228 | if (!u) { | |
1229 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { | |
46f5aa88 | 1230 | pr_err("Couldn't find uuid for %s in set\n", |
6e916a7e | 1231 | dc->backing_dev_name); |
cafe5635 KO |
1232 | return -ENOENT; |
1233 | } | |
1234 | ||
1235 | u = uuid_find_empty(c); | |
1236 | if (!u) { | |
46f5aa88 | 1237 | pr_err("Not caching %s, no room for UUID\n", |
6e916a7e | 1238 | dc->backing_dev_name); |
cafe5635 KO |
1239 | return -EINVAL; |
1240 | } | |
1241 | } | |
1242 | ||
3be11dba CL |
1243 | /* |
1244 | * Deadlocks since we're called via sysfs... | |
1245 | * sysfs_remove_file(&dc->kobj, &sysfs_attach); | |
cafe5635 KO |
1246 | */ |
1247 | ||
169ef1cf | 1248 | if (bch_is_zero(u->uuid, 16)) { |
cafe5635 | 1249 | struct closure cl; |
1fae7cf0 | 1250 | |
cafe5635 KO |
1251 | closure_init_stack(&cl); |
1252 | ||
1253 | memcpy(u->uuid, dc->sb.uuid, 16); | |
1254 | memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); | |
1255 | u->first_reg = u->last_reg = rtime; | |
1256 | bch_uuid_write(c); | |
1257 | ||
1132e56e | 1258 | memcpy(dc->sb.set_uuid, c->set_uuid, 16); |
cafe5635 KO |
1259 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); |
1260 | ||
1261 | bch_write_bdev_super(dc, &cl); | |
1262 | closure_sync(&cl); | |
1263 | } else { | |
1264 | u->last_reg = rtime; | |
1265 | bch_uuid_write(c); | |
1266 | } | |
1267 | ||
1268 | bcache_device_attach(&dc->disk, c, u - c->uuids); | |
cafe5635 KO |
1269 | list_move(&dc->list, &c->cached_devs); |
1270 | calc_cached_dev_sectors(c); | |
1271 | ||
cafe5635 KO |
1272 | /* |
1273 | * dc->c must be set before dc->count != 0 - paired with the mb in | |
1274 | * cached_dev_get() | |
1275 | */ | |
eb2b3d03 | 1276 | smp_wmb(); |
3b304d24 | 1277 | refcount_set(&dc->count, 1); |
cafe5635 | 1278 | |
07cc6ef8 EW |
1279 | /* Block writeback thread, but spawn it */ |
1280 | down_write(&dc->writeback_lock); | |
1281 | if (bch_cached_dev_writeback_start(dc)) { | |
1282 | up_write(&dc->writeback_lock); | |
46f5aa88 | 1283 | pr_err("Couldn't start writeback facilities for %s\n", |
633bb2ce | 1284 | dc->disk.disk->disk_name); |
9e5c3535 | 1285 | return -ENOMEM; |
07cc6ef8 | 1286 | } |
9e5c3535 | 1287 | |
cafe5635 KO |
1288 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { |
1289 | atomic_set(&dc->has_dirty, 1); | |
cafe5635 KO |
1290 | bch_writeback_queue(dc); |
1291 | } | |
1292 | ||
2e17a262 TJ |
1293 | bch_sectors_dirty_init(&dc->disk); |
1294 | ||
0b13efec CL |
1295 | ret = bch_cached_dev_run(dc); |
1296 | if (ret && (ret != -EBUSY)) { | |
1297 | up_write(&dc->writeback_lock); | |
5c2a634c CL |
1298 | /* |
1299 | * bch_register_lock is held, bcache_device_stop() is not | |
1300 | * able to be directly called. The kthread and kworker | |
1301 | * created previously in bch_cached_dev_writeback_start() | |
1302 | * have to be stopped manually here. | |
1303 | */ | |
1304 | kthread_stop(dc->writeback_thread); | |
1305 | cancel_writeback_rate_update_dwork(dc); | |
46f5aa88 | 1306 | pr_err("Couldn't run cached device %s\n", |
633bb2ce | 1307 | dc->backing_dev_name); |
0b13efec CL |
1308 | return ret; |
1309 | } | |
1310 | ||
ee668506 | 1311 | bcache_device_link(&dc->disk, c, "bdev"); |
ea8c5356 | 1312 | atomic_inc(&c->attached_dev_nr); |
cafe5635 | 1313 | |
07cc6ef8 EW |
1314 | /* Allow the writeback thread to proceed */ |
1315 | up_write(&dc->writeback_lock); | |
1316 | ||
46f5aa88 | 1317 | pr_info("Caching %s as %s on set %pU\n", |
6e916a7e CL |
1318 | dc->backing_dev_name, |
1319 | dc->disk.disk->disk_name, | |
1132e56e | 1320 | dc->disk.c->set_uuid); |
cafe5635 KO |
1321 | return 0; |
1322 | } | |
1323 | ||
2d17456e | 1324 | /* when dc->disk.kobj released */ |
cafe5635 KO |
1325 | void bch_cached_dev_release(struct kobject *kobj) |
1326 | { | |
1327 | struct cached_dev *dc = container_of(kobj, struct cached_dev, | |
1328 | disk.kobj); | |
1329 | kfree(dc); | |
1330 | module_put(THIS_MODULE); | |
1331 | } | |
1332 | ||
1333 | static void cached_dev_free(struct closure *cl) | |
1334 | { | |
1335 | struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); | |
1336 | ||
3fd47bfe CL |
1337 | if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) |
1338 | cancel_writeback_rate_update_dwork(dc); | |
1339 | ||
a664d0f0 SP |
1340 | if (!IS_ERR_OR_NULL(dc->writeback_thread)) |
1341 | kthread_stop(dc->writeback_thread); | |
0f0709e6 CL |
1342 | if (!IS_ERR_OR_NULL(dc->status_update_thread)) |
1343 | kthread_stop(dc->status_update_thread); | |
cafe5635 | 1344 | |
80265d8d CL |
1345 | mutex_lock(&bch_register_lock); |
1346 | ||
f59fce84 KO |
1347 | if (atomic_read(&dc->running)) |
1348 | bd_unlink_disk_holder(dc->bdev, dc->disk.disk); | |
cafe5635 KO |
1349 | bcache_device_free(&dc->disk); |
1350 | list_del(&dc->list); | |
1351 | ||
1352 | mutex_unlock(&bch_register_lock); | |
1353 | ||
475389ae CH |
1354 | if (dc->sb_disk) |
1355 | put_page(virt_to_page(dc->sb_disk)); | |
e8547d42 | 1356 | |
0781c874 | 1357 | if (!IS_ERR_OR_NULL(dc->bdev)) |
cafe5635 | 1358 | blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
cafe5635 KO |
1359 | |
1360 | wake_up(&unregister_wait); | |
1361 | ||
1362 | kobject_put(&dc->disk.kobj); | |
1363 | } | |
1364 | ||
1365 | static void cached_dev_flush(struct closure *cl) | |
1366 | { | |
1367 | struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); | |
1368 | struct bcache_device *d = &dc->disk; | |
1369 | ||
c9502ea4 | 1370 | mutex_lock(&bch_register_lock); |
c4d951dd | 1371 | bcache_device_unlink(d); |
c9502ea4 KO |
1372 | mutex_unlock(&bch_register_lock); |
1373 | ||
cafe5635 KO |
1374 | bch_cache_accounting_destroy(&dc->accounting); |
1375 | kobject_del(&d->kobj); | |
1376 | ||
1377 | continue_at(cl, cached_dev_free, system_wq); | |
1378 | } | |
1379 | ||
6f10f7d1 | 1380 | static int cached_dev_init(struct cached_dev *dc, unsigned int block_size) |
cafe5635 | 1381 | { |
f59fce84 | 1382 | int ret; |
cafe5635 | 1383 | struct io *io; |
f59fce84 | 1384 | struct request_queue *q = bdev_get_queue(dc->bdev); |
cafe5635 KO |
1385 | |
1386 | __module_get(THIS_MODULE); | |
1387 | INIT_LIST_HEAD(&dc->list); | |
f59fce84 KO |
1388 | closure_init(&dc->disk.cl, NULL); |
1389 | set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); | |
cafe5635 | 1390 | kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); |
cafe5635 | 1391 | INIT_WORK(&dc->detach, cached_dev_detach_finish); |
cb7a583e | 1392 | sema_init(&dc->sb_write_mutex, 1); |
f59fce84 KO |
1393 | INIT_LIST_HEAD(&dc->io_lru); |
1394 | spin_lock_init(&dc->io_lock); | |
1395 | bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); | |
cafe5635 | 1396 | |
cafe5635 KO |
1397 | dc->sequential_cutoff = 4 << 20; |
1398 | ||
cafe5635 KO |
1399 | for (io = dc->io; io < dc->io + RECENT_IO; io++) { |
1400 | list_add(&io->lru, &dc->io_lru); | |
1401 | hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); | |
1402 | } | |
1403 | ||
c78afc62 KO |
1404 | dc->disk.stripe_size = q->limits.io_opt >> 9; |
1405 | ||
1406 | if (dc->disk.stripe_size) | |
1407 | dc->partial_stripes_expensive = | |
1408 | q->limits.raid_partial_stripes_expensive; | |
1409 | ||
279afbad | 1410 | ret = bcache_device_init(&dc->disk, block_size, |
a782483c | 1411 | bdev_nr_sectors(dc->bdev) - dc->sb.data_offset, |
c62b37d9 | 1412 | dc->bdev, &bcache_cached_ops); |
f59fce84 KO |
1413 | if (ret) |
1414 | return ret; | |
1415 | ||
5d4ce78b CH |
1416 | blk_queue_io_opt(dc->disk.disk->queue, |
1417 | max(queue_io_opt(dc->disk.disk->queue), queue_io_opt(q))); | |
f59fce84 | 1418 | |
c7b7bd07 CL |
1419 | atomic_set(&dc->io_errors, 0); |
1420 | dc->io_disable = false; | |
1421 | dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT; | |
7e027ca4 CL |
1422 | /* default to auto */ |
1423 | dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO; | |
1424 | ||
f59fce84 KO |
1425 | bch_cached_dev_request_init(dc); |
1426 | bch_cached_dev_writeback_init(dc); | |
cafe5635 | 1427 | return 0; |
cafe5635 KO |
1428 | } |
1429 | ||
1430 | /* Cached device - bcache superblock */ | |
1431 | ||
cfa0c56d | 1432 | static int register_bdev(struct cache_sb *sb, struct cache_sb_disk *sb_disk, |
cafe5635 KO |
1433 | struct block_device *bdev, |
1434 | struct cached_dev *dc) | |
1435 | { | |
cafe5635 | 1436 | const char *err = "cannot allocate memory"; |
cafe5635 | 1437 | struct cache_set *c; |
0b13efec | 1438 | int ret = -ENOMEM; |
cafe5635 | 1439 | |
6e916a7e | 1440 | bdevname(bdev, dc->backing_dev_name); |
cafe5635 | 1441 | memcpy(&dc->sb, sb, sizeof(struct cache_sb)); |
cafe5635 KO |
1442 | dc->bdev = bdev; |
1443 | dc->bdev->bd_holder = dc; | |
475389ae | 1444 | dc->sb_disk = sb_disk; |
6e916a7e | 1445 | |
f59fce84 KO |
1446 | if (cached_dev_init(dc, sb->block_size << 9)) |
1447 | goto err; | |
cafe5635 KO |
1448 | |
1449 | err = "error creating kobject"; | |
8d65269f | 1450 | if (kobject_add(&dc->disk.kobj, bdev_kobj(bdev), "bcache")) |
cafe5635 KO |
1451 | goto err; |
1452 | if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) | |
1453 | goto err; | |
1454 | ||
46f5aa88 | 1455 | pr_info("registered backing device %s\n", dc->backing_dev_name); |
f59fce84 | 1456 | |
cafe5635 | 1457 | list_add(&dc->list, &uncached_devices); |
e57fd746 | 1458 | /* attach to a matched cache set if it exists */ |
cafe5635 | 1459 | list_for_each_entry(c, &bch_cache_sets, list) |
73ac105b | 1460 | bch_cached_dev_attach(dc, c, NULL); |
cafe5635 KO |
1461 | |
1462 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || | |
0b13efec CL |
1463 | BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) { |
1464 | err = "failed to run cached device"; | |
1465 | ret = bch_cached_dev_run(dc); | |
1466 | if (ret) | |
1467 | goto err; | |
1468 | } | |
cafe5635 | 1469 | |
88c12d42 | 1470 | return 0; |
cafe5635 | 1471 | err: |
46f5aa88 | 1472 | pr_notice("error %s: %s\n", dc->backing_dev_name, err); |
f59fce84 | 1473 | bcache_device_stop(&dc->disk); |
0b13efec | 1474 | return ret; |
cafe5635 KO |
1475 | } |
1476 | ||
1477 | /* Flash only volumes */ | |
1478 | ||
2d17456e | 1479 | /* When d->kobj released */ |
cafe5635 KO |
1480 | void bch_flash_dev_release(struct kobject *kobj) |
1481 | { | |
1482 | struct bcache_device *d = container_of(kobj, struct bcache_device, | |
1483 | kobj); | |
1484 | kfree(d); | |
1485 | } | |
1486 | ||
1487 | static void flash_dev_free(struct closure *cl) | |
1488 | { | |
1489 | struct bcache_device *d = container_of(cl, struct bcache_device, cl); | |
1fae7cf0 | 1490 | |
e5112201 | 1491 | mutex_lock(&bch_register_lock); |
99a27d59 TJ |
1492 | atomic_long_sub(bcache_dev_sectors_dirty(d), |
1493 | &d->c->flash_dev_dirty_sectors); | |
cafe5635 | 1494 | bcache_device_free(d); |
e5112201 | 1495 | mutex_unlock(&bch_register_lock); |
cafe5635 KO |
1496 | kobject_put(&d->kobj); |
1497 | } | |
1498 | ||
1499 | static void flash_dev_flush(struct closure *cl) | |
1500 | { | |
1501 | struct bcache_device *d = container_of(cl, struct bcache_device, cl); | |
1502 | ||
e5112201 | 1503 | mutex_lock(&bch_register_lock); |
ee668506 | 1504 | bcache_device_unlink(d); |
e5112201 | 1505 | mutex_unlock(&bch_register_lock); |
cafe5635 KO |
1506 | kobject_del(&d->kobj); |
1507 | continue_at(cl, flash_dev_free, system_wq); | |
1508 | } | |
1509 | ||
1510 | static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) | |
1511 | { | |
1512 | struct bcache_device *d = kzalloc(sizeof(struct bcache_device), | |
1513 | GFP_KERNEL); | |
1514 | if (!d) | |
1515 | return -ENOMEM; | |
1516 | ||
1517 | closure_init(&d->cl, NULL); | |
1518 | set_closure_fn(&d->cl, flash_dev_flush, system_wq); | |
1519 | ||
1520 | kobject_init(&d->kobj, &bch_flash_dev_ktype); | |
1521 | ||
4e1ebae3 | 1522 | if (bcache_device_init(d, block_bytes(c->cache), u->sectors, |
c62b37d9 | 1523 | NULL, &bcache_flash_ops)) |
cafe5635 KO |
1524 | goto err; |
1525 | ||
1526 | bcache_device_attach(d, c, u - c->uuids); | |
175206cf | 1527 | bch_sectors_dirty_init(d); |
cafe5635 KO |
1528 | bch_flash_dev_request_init(d); |
1529 | add_disk(d->disk); | |
1530 | ||
1531 | if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache")) | |
1532 | goto err; | |
1533 | ||
1534 | bcache_device_link(d, c, "volume"); | |
1535 | ||
1536 | return 0; | |
1537 | err: | |
1538 | kobject_put(&d->kobj); | |
1539 | return -ENOMEM; | |
1540 | } | |
1541 | ||
1542 | static int flash_devs_run(struct cache_set *c) | |
1543 | { | |
1544 | int ret = 0; | |
1545 | struct uuid_entry *u; | |
1546 | ||
1547 | for (u = c->uuids; | |
02aa8a8b | 1548 | u < c->uuids + c->nr_uuids && !ret; |
cafe5635 KO |
1549 | u++) |
1550 | if (UUID_FLASH_ONLY(u)) | |
1551 | ret = flash_dev_run(c, u); | |
1552 | ||
1553 | return ret; | |
1554 | } | |
1555 | ||
1556 | int bch_flash_dev_create(struct cache_set *c, uint64_t size) | |
1557 | { | |
1558 | struct uuid_entry *u; | |
1559 | ||
1560 | if (test_bit(CACHE_SET_STOPPING, &c->flags)) | |
1561 | return -EINTR; | |
1562 | ||
bf0c55c9 SP |
1563 | if (!test_bit(CACHE_SET_RUNNING, &c->flags)) |
1564 | return -EPERM; | |
1565 | ||
cafe5635 KO |
1566 | u = uuid_find_empty(c); |
1567 | if (!u) { | |
46f5aa88 | 1568 | pr_err("Can't create volume, no room for UUID\n"); |
cafe5635 KO |
1569 | return -EINVAL; |
1570 | } | |
1571 | ||
1572 | get_random_bytes(u->uuid, 16); | |
1573 | memset(u->label, 0, 32); | |
75cbb3f1 | 1574 | u->first_reg = u->last_reg = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 KO |
1575 | |
1576 | SET_UUID_FLASH_ONLY(u, 1); | |
1577 | u->sectors = size >> 9; | |
1578 | ||
1579 | bch_uuid_write(c); | |
1580 | ||
1581 | return flash_dev_run(c, u); | |
1582 | } | |
1583 | ||
c7b7bd07 CL |
1584 | bool bch_cached_dev_error(struct cached_dev *dc) |
1585 | { | |
c7b7bd07 CL |
1586 | if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) |
1587 | return false; | |
1588 | ||
1589 | dc->io_disable = true; | |
1590 | /* make others know io_disable is true earlier */ | |
1591 | smp_mb(); | |
1592 | ||
1593 | pr_err("stop %s: too many IO errors on backing device %s\n", | |
46f5aa88 | 1594 | dc->disk.disk->disk_name, dc->backing_dev_name); |
c7b7bd07 CL |
1595 | |
1596 | bcache_device_stop(&dc->disk); | |
1597 | return true; | |
1598 | } | |
1599 | ||
cafe5635 KO |
1600 | /* Cache set */ |
1601 | ||
1602 | __printf(2, 3) | |
1603 | bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) | |
1604 | { | |
46f5aa88 | 1605 | struct va_format vaf; |
cafe5635 KO |
1606 | va_list args; |
1607 | ||
77c320eb KO |
1608 | if (c->on_error != ON_ERROR_PANIC && |
1609 | test_bit(CACHE_SET_STOPPING, &c->flags)) | |
cafe5635 KO |
1610 | return false; |
1611 | ||
771f393e | 1612 | if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags)) |
46f5aa88 | 1613 | pr_info("CACHE_SET_IO_DISABLE already set\n"); |
771f393e | 1614 | |
3be11dba CL |
1615 | /* |
1616 | * XXX: we can be called from atomic context | |
1617 | * acquire_console_sem(); | |
1618 | */ | |
cafe5635 | 1619 | |
cafe5635 | 1620 | va_start(args, fmt); |
cafe5635 | 1621 | |
46f5aa88 JP |
1622 | vaf.fmt = fmt; |
1623 | vaf.va = &args; | |
1624 | ||
1625 | pr_err("error on %pU: %pV, disabling caching\n", | |
1132e56e | 1626 | c->set_uuid, &vaf); |
46f5aa88 JP |
1627 | |
1628 | va_end(args); | |
cafe5635 | 1629 | |
77c320eb KO |
1630 | if (c->on_error == ON_ERROR_PANIC) |
1631 | panic("panic forced after error\n"); | |
1632 | ||
cafe5635 KO |
1633 | bch_cache_set_unregister(c); |
1634 | return true; | |
1635 | } | |
1636 | ||
2d17456e | 1637 | /* When c->kobj released */ |
cafe5635 KO |
1638 | void bch_cache_set_release(struct kobject *kobj) |
1639 | { | |
1640 | struct cache_set *c = container_of(kobj, struct cache_set, kobj); | |
1fae7cf0 | 1641 | |
cafe5635 KO |
1642 | kfree(c); |
1643 | module_put(THIS_MODULE); | |
1644 | } | |
1645 | ||
1646 | static void cache_set_free(struct closure *cl) | |
1647 | { | |
1648 | struct cache_set *c = container_of(cl, struct cache_set, cl); | |
1649 | struct cache *ca; | |
cafe5635 | 1650 | |
ae171023 | 1651 | debugfs_remove(c->debug); |
cafe5635 KO |
1652 | |
1653 | bch_open_buckets_free(c); | |
1654 | bch_btree_cache_free(c); | |
1655 | bch_journal_free(c); | |
1656 | ||
a4b732a2 | 1657 | mutex_lock(&bch_register_lock); |
4a784266 CL |
1658 | bch_bset_sort_state_free(&c->sort); |
1659 | free_pages((unsigned long) c->uuids, ilog2(meta_bucket_pages(&c->cache->sb))); | |
1660 | ||
08fdb2cd CL |
1661 | ca = c->cache; |
1662 | if (ca) { | |
1663 | ca->set = NULL; | |
1664 | c->cache = NULL; | |
1665 | kobject_put(&ca->kobj); | |
1666 | } | |
cafe5635 | 1667 | |
cafe5635 | 1668 | |
da415a09 NS |
1669 | if (c->moving_gc_wq) |
1670 | destroy_workqueue(c->moving_gc_wq); | |
d19936a2 KO |
1671 | bioset_exit(&c->bio_split); |
1672 | mempool_exit(&c->fill_iter); | |
1673 | mempool_exit(&c->bio_meta); | |
1674 | mempool_exit(&c->search); | |
cafe5635 KO |
1675 | kfree(c->devices); |
1676 | ||
cafe5635 KO |
1677 | list_del(&c->list); |
1678 | mutex_unlock(&bch_register_lock); | |
1679 | ||
1132e56e | 1680 | pr_info("Cache set %pU unregistered\n", c->set_uuid); |
cafe5635 KO |
1681 | wake_up(&unregister_wait); |
1682 | ||
1683 | closure_debug_destroy(&c->cl); | |
1684 | kobject_put(&c->kobj); | |
1685 | } | |
1686 | ||
1687 | static void cache_set_flush(struct closure *cl) | |
1688 | { | |
1689 | struct cache_set *c = container_of(cl, struct cache_set, caching); | |
08fdb2cd | 1690 | struct cache *ca = c->cache; |
cafe5635 | 1691 | struct btree *b; |
cafe5635 KO |
1692 | |
1693 | bch_cache_accounting_destroy(&c->accounting); | |
1694 | ||
1695 | kobject_put(&c->internal); | |
1696 | kobject_del(&c->kobj); | |
1697 | ||
b387e9b5 | 1698 | if (!IS_ERR_OR_NULL(c->gc_thread)) |
72a44517 KO |
1699 | kthread_stop(c->gc_thread); |
1700 | ||
cafe5635 KO |
1701 | if (!IS_ERR_OR_NULL(c->root)) |
1702 | list_add(&c->root->list, &c->btree_cache); | |
1703 | ||
e6dcbd3e CL |
1704 | /* |
1705 | * Avoid flushing cached nodes if cache set is retiring | |
1706 | * due to too many I/O errors detected. | |
1707 | */ | |
1708 | if (!test_bit(CACHE_SET_IO_DISABLE, &c->flags)) | |
1709 | list_for_each_entry(b, &c->btree_cache, list) { | |
1710 | mutex_lock(&b->write_lock); | |
1711 | if (btree_node_dirty(b)) | |
1712 | __bch_btree_node_write(b, NULL); | |
1713 | mutex_unlock(&b->write_lock); | |
1714 | } | |
cafe5635 | 1715 | |
08fdb2cd CL |
1716 | if (ca->alloc_thread) |
1717 | kthread_stop(ca->alloc_thread); | |
79826c35 | 1718 | |
5b1016e6 KO |
1719 | if (c->journal.cur) { |
1720 | cancel_delayed_work_sync(&c->journal.work); | |
1721 | /* flush last journal entry if needed */ | |
1722 | c->journal.work.work.func(&c->journal.work.work); | |
1723 | } | |
dabb4433 | 1724 | |
cafe5635 KO |
1725 | closure_return(cl); |
1726 | } | |
1727 | ||
7e027ca4 CL |
1728 | /* |
1729 | * This function is only called when CACHE_SET_IO_DISABLE is set, which means | |
1730 | * cache set is unregistering due to too many I/O errors. In this condition, | |
1731 | * the bcache device might be stopped, it depends on stop_when_cache_set_failed | |
1732 | * value and whether the broken cache has dirty data: | |
1733 | * | |
1734 | * dc->stop_when_cache_set_failed dc->has_dirty stop bcache device | |
1735 | * BCH_CACHED_STOP_AUTO 0 NO | |
1736 | * BCH_CACHED_STOP_AUTO 1 YES | |
1737 | * BCH_CACHED_DEV_STOP_ALWAYS 0 YES | |
1738 | * BCH_CACHED_DEV_STOP_ALWAYS 1 YES | |
1739 | * | |
1740 | * The expected behavior is, if stop_when_cache_set_failed is configured to | |
1741 | * "auto" via sysfs interface, the bcache device will not be stopped if the | |
1742 | * backing device is clean on the broken cache device. | |
1743 | */ | |
1744 | static void conditional_stop_bcache_device(struct cache_set *c, | |
1745 | struct bcache_device *d, | |
1746 | struct cached_dev *dc) | |
1747 | { | |
1748 | if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) { | |
46f5aa88 | 1749 | pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.\n", |
1132e56e | 1750 | d->disk->disk_name, c->set_uuid); |
7e027ca4 CL |
1751 | bcache_device_stop(d); |
1752 | } else if (atomic_read(&dc->has_dirty)) { | |
1753 | /* | |
1754 | * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO | |
1755 | * and dc->has_dirty == 1 | |
1756 | */ | |
46f5aa88 | 1757 | pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.\n", |
7e027ca4 | 1758 | d->disk->disk_name); |
e8cf978d CIK |
1759 | /* |
1760 | * There might be a small time gap that cache set is | |
1761 | * released but bcache device is not. Inside this time | |
1762 | * gap, regular I/O requests will directly go into | |
1763 | * backing device as no cache set attached to. This | |
1764 | * behavior may also introduce potential inconsistence | |
1765 | * data in writeback mode while cache is dirty. | |
1766 | * Therefore before calling bcache_device_stop() due | |
1767 | * to a broken cache device, dc->io_disable should be | |
1768 | * explicitly set to true. | |
1769 | */ | |
1770 | dc->io_disable = true; | |
1771 | /* make others know io_disable is true earlier */ | |
1772 | smp_mb(); | |
1773 | bcache_device_stop(d); | |
7e027ca4 CL |
1774 | } else { |
1775 | /* | |
1776 | * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO | |
1777 | * and dc->has_dirty == 0 | |
1778 | */ | |
46f5aa88 | 1779 | pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.\n", |
7e027ca4 CL |
1780 | d->disk->disk_name); |
1781 | } | |
1782 | } | |
1783 | ||
cafe5635 KO |
1784 | static void __cache_set_unregister(struct closure *cl) |
1785 | { | |
1786 | struct cache_set *c = container_of(cl, struct cache_set, caching); | |
5caa52af | 1787 | struct cached_dev *dc; |
7e027ca4 | 1788 | struct bcache_device *d; |
cafe5635 KO |
1789 | size_t i; |
1790 | ||
1791 | mutex_lock(&bch_register_lock); | |
1792 | ||
7e027ca4 CL |
1793 | for (i = 0; i < c->devices_max_used; i++) { |
1794 | d = c->devices[i]; | |
1795 | if (!d) | |
1796 | continue; | |
1797 | ||
1798 | if (!UUID_FLASH_ONLY(&c->uuids[i]) && | |
1799 | test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { | |
1800 | dc = container_of(d, struct cached_dev, disk); | |
1801 | bch_cached_dev_detach(dc); | |
1802 | if (test_bit(CACHE_SET_IO_DISABLE, &c->flags)) | |
1803 | conditional_stop_bcache_device(c, d, dc); | |
1804 | } else { | |
1805 | bcache_device_stop(d); | |
5caa52af | 1806 | } |
7e027ca4 | 1807 | } |
cafe5635 KO |
1808 | |
1809 | mutex_unlock(&bch_register_lock); | |
1810 | ||
1811 | continue_at(cl, cache_set_flush, system_wq); | |
1812 | } | |
1813 | ||
1814 | void bch_cache_set_stop(struct cache_set *c) | |
1815 | { | |
1816 | if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) | |
63d63b51 | 1817 | /* closure_fn set to __cache_set_unregister() */ |
cafe5635 KO |
1818 | closure_queue(&c->caching); |
1819 | } | |
1820 | ||
1821 | void bch_cache_set_unregister(struct cache_set *c) | |
1822 | { | |
1823 | set_bit(CACHE_SET_UNREGISTERING, &c->flags); | |
1824 | bch_cache_set_stop(c); | |
1825 | } | |
1826 | ||
de1fafab CL |
1827 | #define alloc_meta_bucket_pages(gfp, sb) \ |
1828 | ((void *) __get_free_pages(__GFP_ZERO|__GFP_COMP|gfp, ilog2(meta_bucket_pages(sb)))) | |
cafe5635 KO |
1829 | |
1830 | struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) | |
1831 | { | |
1832 | int iter_size; | |
4a784266 | 1833 | struct cache *ca = container_of(sb, struct cache, sb); |
cafe5635 | 1834 | struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); |
1fae7cf0 | 1835 | |
cafe5635 KO |
1836 | if (!c) |
1837 | return NULL; | |
1838 | ||
1839 | __module_get(THIS_MODULE); | |
1840 | closure_init(&c->cl, NULL); | |
1841 | set_closure_fn(&c->cl, cache_set_free, system_wq); | |
1842 | ||
1843 | closure_init(&c->caching, &c->cl); | |
1844 | set_closure_fn(&c->caching, __cache_set_unregister, system_wq); | |
1845 | ||
1846 | /* Maybe create continue_at_noreturn() and use it here? */ | |
1847 | closure_set_stopped(&c->cl); | |
1848 | closure_put(&c->cl); | |
1849 | ||
1850 | kobject_init(&c->kobj, &bch_cache_set_ktype); | |
1851 | kobject_init(&c->internal, &bch_cache_set_internal_ktype); | |
1852 | ||
1853 | bch_cache_accounting_init(&c->accounting, &c->cl); | |
1854 | ||
1132e56e | 1855 | memcpy(c->set_uuid, sb->set_uuid, 16); |
d721a43f | 1856 | |
4a784266 CL |
1857 | c->cache = ca; |
1858 | c->cache->set = c; | |
cafe5635 KO |
1859 | c->bucket_bits = ilog2(sb->bucket_size); |
1860 | c->block_bits = ilog2(sb->block_size); | |
4a784266 | 1861 | c->nr_uuids = meta_bucket_bytes(sb) / sizeof(struct uuid_entry); |
2831231d | 1862 | c->devices_max_used = 0; |
ea8c5356 | 1863 | atomic_set(&c->attached_dev_nr, 0); |
4a784266 | 1864 | c->btree_pages = meta_bucket_pages(sb); |
cafe5635 KO |
1865 | if (c->btree_pages > BTREE_MAX_PAGES) |
1866 | c->btree_pages = max_t(int, c->btree_pages / 4, | |
1867 | BTREE_MAX_PAGES); | |
1868 | ||
cb7a583e | 1869 | sema_init(&c->sb_write_mutex, 1); |
e8e1d468 | 1870 | mutex_init(&c->bucket_lock); |
0a63b66d | 1871 | init_waitqueue_head(&c->btree_cache_wait); |
34cf78bf | 1872 | spin_lock_init(&c->btree_cannibalize_lock); |
35fcd848 | 1873 | init_waitqueue_head(&c->bucket_wait); |
be628be0 | 1874 | init_waitqueue_head(&c->gc_wait); |
cb7a583e | 1875 | sema_init(&c->uuid_write_mutex, 1); |
65d22e91 | 1876 | |
65d22e91 KO |
1877 | spin_lock_init(&c->btree_gc_time.lock); |
1878 | spin_lock_init(&c->btree_split_time.lock); | |
1879 | spin_lock_init(&c->btree_read_time.lock); | |
e8e1d468 | 1880 | |
cafe5635 KO |
1881 | bch_moving_init_cache_set(c); |
1882 | ||
1883 | INIT_LIST_HEAD(&c->list); | |
1884 | INIT_LIST_HEAD(&c->cached_devs); | |
1885 | INIT_LIST_HEAD(&c->btree_cache); | |
1886 | INIT_LIST_HEAD(&c->btree_cache_freeable); | |
1887 | INIT_LIST_HEAD(&c->btree_cache_freed); | |
1888 | INIT_LIST_HEAD(&c->data_buckets); | |
1889 | ||
6907dc49 | 1890 | iter_size = ((meta_bucket_pages(sb) * PAGE_SECTORS) / sb->block_size + 1) * |
cafe5635 KO |
1891 | sizeof(struct btree_iter_set); |
1892 | ||
a42d3c64 CL |
1893 | c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL); |
1894 | if (!c->devices) | |
1895 | goto err; | |
1896 | ||
1897 | if (mempool_init_slab_pool(&c->search, 32, bch_search_cache)) | |
1898 | goto err; | |
1899 | ||
1900 | if (mempool_init_kmalloc_pool(&c->bio_meta, 2, | |
1901 | sizeof(struct bbio) + | |
4a784266 | 1902 | sizeof(struct bio_vec) * meta_bucket_pages(sb))) |
a42d3c64 CL |
1903 | goto err; |
1904 | ||
1905 | if (mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size)) | |
1906 | goto err; | |
1907 | ||
1908 | if (bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio), | |
1909 | BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER)) | |
1910 | goto err; | |
1911 | ||
4a784266 | 1912 | c->uuids = alloc_meta_bucket_pages(GFP_KERNEL, sb); |
a42d3c64 CL |
1913 | if (!c->uuids) |
1914 | goto err; | |
1915 | ||
1916 | c->moving_gc_wq = alloc_workqueue("bcache_gc", WQ_MEM_RECLAIM, 0); | |
1917 | if (!c->moving_gc_wq) | |
1918 | goto err; | |
1919 | ||
1920 | if (bch_journal_alloc(c)) | |
1921 | goto err; | |
1922 | ||
1923 | if (bch_btree_cache_alloc(c)) | |
1924 | goto err; | |
1925 | ||
1926 | if (bch_open_buckets_alloc(c)) | |
1927 | goto err; | |
1928 | ||
1929 | if (bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages))) | |
cafe5635 KO |
1930 | goto err; |
1931 | ||
cafe5635 KO |
1932 | c->congested_read_threshold_us = 2000; |
1933 | c->congested_write_threshold_us = 20000; | |
7ba0d830 | 1934 | c->error_limit = DEFAULT_IO_ERROR_LIMIT; |
c5fcdedc | 1935 | c->idle_max_writeback_rate_enabled = 1; |
771f393e | 1936 | WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags)); |
cafe5635 KO |
1937 | |
1938 | return c; | |
1939 | err: | |
1940 | bch_cache_set_unregister(c); | |
1941 | return NULL; | |
1942 | } | |
1943 | ||
ce3e4cfb | 1944 | static int run_cache_set(struct cache_set *c) |
cafe5635 KO |
1945 | { |
1946 | const char *err = "cannot allocate memory"; | |
1947 | struct cached_dev *dc, *t; | |
08fdb2cd | 1948 | struct cache *ca = c->cache; |
c18536a7 | 1949 | struct closure cl; |
95f18c9d SW |
1950 | LIST_HEAD(journal); |
1951 | struct journal_replay *l; | |
cafe5635 | 1952 | |
c18536a7 | 1953 | closure_init_stack(&cl); |
cafe5635 | 1954 | |
08fdb2cd | 1955 | c->nbuckets = ca->sb.nbuckets; |
be628be0 | 1956 | set_gc_sectors(c); |
cafe5635 | 1957 | |
6f9414e0 | 1958 | if (CACHE_SYNC(&c->cache->sb)) { |
cafe5635 KO |
1959 | struct bkey *k; |
1960 | struct jset *j; | |
1961 | ||
1962 | err = "cannot allocate memory for journal"; | |
c18536a7 | 1963 | if (bch_journal_read(c, &journal)) |
cafe5635 KO |
1964 | goto err; |
1965 | ||
46f5aa88 | 1966 | pr_debug("btree_journal_read() done\n"); |
cafe5635 KO |
1967 | |
1968 | err = "no journal entries found"; | |
1969 | if (list_empty(&journal)) | |
1970 | goto err; | |
1971 | ||
1972 | j = &list_entry(journal.prev, struct journal_replay, list)->j; | |
1973 | ||
1974 | err = "IO error reading priorities"; | |
08fdb2cd CL |
1975 | if (prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev])) |
1976 | goto err; | |
cafe5635 KO |
1977 | |
1978 | /* | |
1979 | * If prio_read() fails it'll call cache_set_error and we'll | |
1980 | * tear everything down right away, but if we perhaps checked | |
1981 | * sooner we could avoid journal replay. | |
1982 | */ | |
1983 | ||
1984 | k = &j->btree_root; | |
1985 | ||
1986 | err = "bad btree root"; | |
65d45231 | 1987 | if (__bch_btree_ptr_invalid(c, k)) |
cafe5635 KO |
1988 | goto err; |
1989 | ||
1990 | err = "error reading btree root"; | |
b0d30981 CL |
1991 | c->root = bch_btree_node_get(c, NULL, k, |
1992 | j->btree_level, | |
1993 | true, NULL); | |
cafe5635 KO |
1994 | if (IS_ERR_OR_NULL(c->root)) |
1995 | goto err; | |
1996 | ||
1997 | list_del_init(&c->root->list); | |
1998 | rw_unlock(true, c->root); | |
1999 | ||
c18536a7 | 2000 | err = uuid_read(c, j, &cl); |
cafe5635 KO |
2001 | if (err) |
2002 | goto err; | |
2003 | ||
2004 | err = "error in recovery"; | |
c18536a7 | 2005 | if (bch_btree_check(c)) |
cafe5635 KO |
2006 | goto err; |
2007 | ||
2008 | bch_journal_mark(c, &journal); | |
2531d9ee | 2009 | bch_initial_gc_finish(c); |
46f5aa88 | 2010 | pr_debug("btree_check() done\n"); |
cafe5635 KO |
2011 | |
2012 | /* | |
2013 | * bcache_journal_next() can't happen sooner, or | |
2014 | * btree_gc_finish() will give spurious errors about last_gc > | |
2015 | * gc_gen - this is a hack but oh well. | |
2016 | */ | |
2017 | bch_journal_next(&c->journal); | |
2018 | ||
119ba0f8 | 2019 | err = "error starting allocator thread"; |
08fdb2cd CL |
2020 | if (bch_cache_allocator_start(ca)) |
2021 | goto err; | |
cafe5635 KO |
2022 | |
2023 | /* | |
2024 | * First place it's safe to allocate: btree_check() and | |
2025 | * btree_gc_finish() have to run before we have buckets to | |
2026 | * allocate, and bch_bucket_alloc_set() might cause a journal | |
2027 | * entry to be written so bcache_journal_next() has to be called | |
2028 | * first. | |
2029 | * | |
2030 | * If the uuids were in the old format we have to rewrite them | |
2031 | * before the next journal entry is written: | |
2032 | */ | |
2033 | if (j->version < BCACHE_JSET_VERSION_UUID) | |
2034 | __uuid_write(c); | |
2035 | ||
ce3e4cfb CL |
2036 | err = "bcache: replay journal failed"; |
2037 | if (bch_journal_replay(c, &journal)) | |
2038 | goto err; | |
cafe5635 | 2039 | } else { |
08fdb2cd | 2040 | unsigned int j; |
cafe5635 | 2041 | |
08fdb2cd CL |
2042 | pr_notice("invalidating existing data\n"); |
2043 | ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, | |
2044 | 2, SB_JOURNAL_BUCKETS); | |
cafe5635 | 2045 | |
08fdb2cd CL |
2046 | for (j = 0; j < ca->sb.keys; j++) |
2047 | ca->sb.d[j] = ca->sb.first_bucket + j; | |
cafe5635 | 2048 | |
2531d9ee | 2049 | bch_initial_gc_finish(c); |
cafe5635 | 2050 | |
119ba0f8 | 2051 | err = "error starting allocator thread"; |
08fdb2cd CL |
2052 | if (bch_cache_allocator_start(ca)) |
2053 | goto err; | |
cafe5635 KO |
2054 | |
2055 | mutex_lock(&c->bucket_lock); | |
08fdb2cd | 2056 | bch_prio_write(ca, true); |
cafe5635 KO |
2057 | mutex_unlock(&c->bucket_lock); |
2058 | ||
cafe5635 KO |
2059 | err = "cannot allocate new UUID bucket"; |
2060 | if (__uuid_write(c)) | |
72a44517 | 2061 | goto err; |
cafe5635 KO |
2062 | |
2063 | err = "cannot allocate new btree root"; | |
2452cc89 | 2064 | c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL); |
cafe5635 | 2065 | if (IS_ERR_OR_NULL(c->root)) |
72a44517 | 2066 | goto err; |
cafe5635 | 2067 | |
2a285686 | 2068 | mutex_lock(&c->root->write_lock); |
cafe5635 | 2069 | bkey_copy_key(&c->root->key, &MAX_KEY); |
c18536a7 | 2070 | bch_btree_node_write(c->root, &cl); |
2a285686 | 2071 | mutex_unlock(&c->root->write_lock); |
cafe5635 KO |
2072 | |
2073 | bch_btree_set_root(c->root); | |
2074 | rw_unlock(true, c->root); | |
2075 | ||
2076 | /* | |
2077 | * We don't want to write the first journal entry until | |
2078 | * everything is set up - fortunately journal entries won't be | |
2079 | * written until the SET_CACHE_SYNC() here: | |
2080 | */ | |
6f9414e0 | 2081 | SET_CACHE_SYNC(&c->cache->sb, true); |
cafe5635 KO |
2082 | |
2083 | bch_journal_next(&c->journal); | |
c18536a7 | 2084 | bch_journal_meta(c, &cl); |
cafe5635 KO |
2085 | } |
2086 | ||
72a44517 KO |
2087 | err = "error starting gc thread"; |
2088 | if (bch_gc_thread_start(c)) | |
2089 | goto err; | |
2090 | ||
c18536a7 | 2091 | closure_sync(&cl); |
4a784266 | 2092 | c->cache->sb.last_mount = (u32)ktime_get_real_seconds(); |
cafe5635 KO |
2093 | bcache_write_super(c); |
2094 | ||
2095 | list_for_each_entry_safe(dc, t, &uncached_devices, list) | |
73ac105b | 2096 | bch_cached_dev_attach(dc, c, NULL); |
cafe5635 KO |
2097 | |
2098 | flash_devs_run(c); | |
2099 | ||
bf0c55c9 | 2100 | set_bit(CACHE_SET_RUNNING, &c->flags); |
ce3e4cfb | 2101 | return 0; |
cafe5635 | 2102 | err: |
95f18c9d SW |
2103 | while (!list_empty(&journal)) { |
2104 | l = list_first_entry(&journal, struct journal_replay, list); | |
2105 | list_del(&l->list); | |
2106 | kfree(l); | |
2107 | } | |
2108 | ||
c18536a7 | 2109 | closure_sync(&cl); |
68a53c95 | 2110 | |
c8694948 | 2111 | bch_cache_set_error(c, "%s", err); |
ce3e4cfb CL |
2112 | |
2113 | return -EIO; | |
cafe5635 KO |
2114 | } |
2115 | ||
cafe5635 KO |
2116 | static const char *register_cache_set(struct cache *ca) |
2117 | { | |
2118 | char buf[12]; | |
2119 | const char *err = "cannot allocate memory"; | |
2120 | struct cache_set *c; | |
2121 | ||
2122 | list_for_each_entry(c, &bch_cache_sets, list) | |
1132e56e | 2123 | if (!memcmp(c->set_uuid, ca->sb.set_uuid, 16)) { |
697e2349 | 2124 | if (c->cache) |
cafe5635 KO |
2125 | return "duplicate cache set member"; |
2126 | ||
cafe5635 KO |
2127 | goto found; |
2128 | } | |
2129 | ||
2130 | c = bch_cache_set_alloc(&ca->sb); | |
2131 | if (!c) | |
2132 | return err; | |
2133 | ||
2134 | err = "error creating kobject"; | |
1132e56e | 2135 | if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->set_uuid) || |
cafe5635 KO |
2136 | kobject_add(&c->internal, &c->kobj, "internal")) |
2137 | goto err; | |
2138 | ||
2139 | if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) | |
2140 | goto err; | |
2141 | ||
2142 | bch_debug_init_cache_set(c); | |
2143 | ||
2144 | list_add(&c->list, &bch_cache_sets); | |
2145 | found: | |
2146 | sprintf(buf, "cache%i", ca->sb.nr_this_dev); | |
2147 | if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || | |
2148 | sysfs_create_link(&c->kobj, &ca->kobj, buf)) | |
2149 | goto err; | |
2150 | ||
d83353b3 | 2151 | kobject_get(&ca->kobj); |
cafe5635 | 2152 | ca->set = c; |
697e2349 | 2153 | ca->set->cache = ca; |
cafe5635 | 2154 | |
697e2349 CL |
2155 | err = "failed to run cache set"; |
2156 | if (run_cache_set(c) < 0) | |
2157 | goto err; | |
cafe5635 KO |
2158 | |
2159 | return NULL; | |
2160 | err: | |
2161 | bch_cache_set_unregister(c); | |
2162 | return err; | |
2163 | } | |
2164 | ||
2165 | /* Cache device */ | |
2166 | ||
2d17456e | 2167 | /* When ca->kobj released */ |
cafe5635 KO |
2168 | void bch_cache_release(struct kobject *kobj) |
2169 | { | |
2170 | struct cache *ca = container_of(kobj, struct cache, kobj); | |
6f10f7d1 | 2171 | unsigned int i; |
cafe5635 | 2172 | |
c9a78332 | 2173 | if (ca->set) { |
697e2349 CL |
2174 | BUG_ON(ca->set->cache != ca); |
2175 | ca->set->cache = NULL; | |
c9a78332 | 2176 | } |
cafe5635 | 2177 | |
c954ac8d | 2178 | free_pages((unsigned long) ca->disk_buckets, ilog2(meta_bucket_pages(&ca->sb))); |
cafe5635 KO |
2179 | kfree(ca->prio_buckets); |
2180 | vfree(ca->buckets); | |
2181 | ||
2182 | free_heap(&ca->heap); | |
cafe5635 | 2183 | free_fifo(&ca->free_inc); |
78365411 KO |
2184 | |
2185 | for (i = 0; i < RESERVE_NR; i++) | |
2186 | free_fifo(&ca->free[i]); | |
cafe5635 | 2187 | |
475389ae CH |
2188 | if (ca->sb_disk) |
2189 | put_page(virt_to_page(ca->sb_disk)); | |
cafe5635 | 2190 | |
0781c874 | 2191 | if (!IS_ERR_OR_NULL(ca->bdev)) |
cafe5635 | 2192 | blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
cafe5635 KO |
2193 | |
2194 | kfree(ca); | |
2195 | module_put(THIS_MODULE); | |
2196 | } | |
2197 | ||
c50d4d5d | 2198 | static int cache_alloc(struct cache *ca) |
cafe5635 KO |
2199 | { |
2200 | size_t free; | |
682811b3 | 2201 | size_t btree_buckets; |
cafe5635 | 2202 | struct bucket *b; |
f6027bca DC |
2203 | int ret = -ENOMEM; |
2204 | const char *err = NULL; | |
cafe5635 | 2205 | |
cafe5635 KO |
2206 | __module_get(THIS_MODULE); |
2207 | kobject_init(&ca->kobj, &bch_cache_ktype); | |
2208 | ||
3a83f467 | 2209 | bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8); |
cafe5635 | 2210 | |
682811b3 TJ |
2211 | /* |
2212 | * when ca->sb.njournal_buckets is not zero, journal exists, | |
2213 | * and in bch_journal_replay(), tree node may split, | |
2214 | * so bucket of RESERVE_BTREE type is needed, | |
2215 | * the worst situation is all journal buckets are valid journal, | |
2216 | * and all the keys need to replay, | |
2217 | * so the number of RESERVE_BTREE type buckets should be as much | |
2218 | * as journal buckets | |
2219 | */ | |
2220 | btree_buckets = ca->sb.njournal_buckets ?: 8; | |
78365411 | 2221 | free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; |
3a646fd7 DC |
2222 | if (!free) { |
2223 | ret = -EPERM; | |
2224 | err = "ca->sb.nbuckets is too small"; | |
2225 | goto err_free; | |
2226 | } | |
cafe5635 | 2227 | |
f6027bca DC |
2228 | if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, |
2229 | GFP_KERNEL)) { | |
2230 | err = "ca->free[RESERVE_BTREE] alloc failed"; | |
2231 | goto err_btree_alloc; | |
2232 | } | |
2233 | ||
2234 | if (!init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), | |
2235 | GFP_KERNEL)) { | |
2236 | err = "ca->free[RESERVE_PRIO] alloc failed"; | |
2237 | goto err_prio_alloc; | |
2238 | } | |
2239 | ||
2240 | if (!init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL)) { | |
2241 | err = "ca->free[RESERVE_MOVINGGC] alloc failed"; | |
2242 | goto err_movinggc_alloc; | |
2243 | } | |
2244 | ||
2245 | if (!init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL)) { | |
2246 | err = "ca->free[RESERVE_NONE] alloc failed"; | |
2247 | goto err_none_alloc; | |
2248 | } | |
2249 | ||
2250 | if (!init_fifo(&ca->free_inc, free << 2, GFP_KERNEL)) { | |
2251 | err = "ca->free_inc alloc failed"; | |
2252 | goto err_free_inc_alloc; | |
2253 | } | |
2254 | ||
2255 | if (!init_heap(&ca->heap, free << 3, GFP_KERNEL)) { | |
2256 | err = "ca->heap alloc failed"; | |
2257 | goto err_heap_alloc; | |
2258 | } | |
2259 | ||
2260 | ca->buckets = vzalloc(array_size(sizeof(struct bucket), | |
2261 | ca->sb.nbuckets)); | |
2262 | if (!ca->buckets) { | |
2263 | err = "ca->buckets alloc failed"; | |
2264 | goto err_buckets_alloc; | |
2265 | } | |
2266 | ||
2267 | ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t), | |
2268 | prio_buckets(ca), 2), | |
2269 | GFP_KERNEL); | |
2270 | if (!ca->prio_buckets) { | |
2271 | err = "ca->prio_buckets alloc failed"; | |
2272 | goto err_prio_buckets_alloc; | |
2273 | } | |
2274 | ||
c954ac8d | 2275 | ca->disk_buckets = alloc_meta_bucket_pages(GFP_KERNEL, &ca->sb); |
f6027bca DC |
2276 | if (!ca->disk_buckets) { |
2277 | err = "ca->disk_buckets alloc failed"; | |
2278 | goto err_disk_buckets_alloc; | |
2279 | } | |
cafe5635 KO |
2280 | |
2281 | ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); | |
2282 | ||
cafe5635 KO |
2283 | for_each_bucket(b, ca) |
2284 | atomic_set(&b->pin, 0); | |
cafe5635 | 2285 | return 0; |
f6027bca DC |
2286 | |
2287 | err_disk_buckets_alloc: | |
2288 | kfree(ca->prio_buckets); | |
2289 | err_prio_buckets_alloc: | |
2290 | vfree(ca->buckets); | |
2291 | err_buckets_alloc: | |
2292 | free_heap(&ca->heap); | |
2293 | err_heap_alloc: | |
2294 | free_fifo(&ca->free_inc); | |
2295 | err_free_inc_alloc: | |
2296 | free_fifo(&ca->free[RESERVE_NONE]); | |
2297 | err_none_alloc: | |
2298 | free_fifo(&ca->free[RESERVE_MOVINGGC]); | |
2299 | err_movinggc_alloc: | |
2300 | free_fifo(&ca->free[RESERVE_PRIO]); | |
2301 | err_prio_alloc: | |
2302 | free_fifo(&ca->free[RESERVE_BTREE]); | |
2303 | err_btree_alloc: | |
3a646fd7 | 2304 | err_free: |
f6027bca DC |
2305 | module_put(THIS_MODULE); |
2306 | if (err) | |
46f5aa88 | 2307 | pr_notice("error %s: %s\n", ca->cache_dev_name, err); |
f6027bca | 2308 | return ret; |
cafe5635 KO |
2309 | } |
2310 | ||
cfa0c56d | 2311 | static int register_cache(struct cache_sb *sb, struct cache_sb_disk *sb_disk, |
c9a78332 | 2312 | struct block_device *bdev, struct cache *ca) |
cafe5635 | 2313 | { |
d9dc1702 | 2314 | const char *err = NULL; /* must be set for any error case */ |
9b299728 | 2315 | int ret = 0; |
cafe5635 | 2316 | |
6e916a7e | 2317 | bdevname(bdev, ca->cache_dev_name); |
f59fce84 | 2318 | memcpy(&ca->sb, sb, sizeof(struct cache_sb)); |
cafe5635 KO |
2319 | ca->bdev = bdev; |
2320 | ca->bdev->bd_holder = ca; | |
475389ae | 2321 | ca->sb_disk = sb_disk; |
f59fce84 | 2322 | |
cc40daf9 | 2323 | if (blk_queue_discard(bdev_get_queue(bdev))) |
cafe5635 KO |
2324 | ca->discard = CACHE_DISCARD(&ca->sb); |
2325 | ||
c50d4d5d | 2326 | ret = cache_alloc(ca); |
d9dc1702 | 2327 | if (ret != 0) { |
bb6d355c CL |
2328 | /* |
2329 | * If we failed here, it means ca->kobj is not initialized yet, | |
2330 | * kobject_put() won't be called and there is no chance to | |
2331 | * call blkdev_put() to bdev in bch_cache_release(). So we | |
2332 | * explicitly call blkdev_put() here. | |
2333 | */ | |
cc40daf9 | 2334 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
d9dc1702 EW |
2335 | if (ret == -ENOMEM) |
2336 | err = "cache_alloc(): -ENOMEM"; | |
3a646fd7 DC |
2337 | else if (ret == -EPERM) |
2338 | err = "cache_alloc(): cache device is too small"; | |
d9dc1702 EW |
2339 | else |
2340 | err = "cache_alloc(): unknown error"; | |
f59fce84 | 2341 | goto err; |
d9dc1702 | 2342 | } |
f59fce84 | 2343 | |
8d65269f | 2344 | if (kobject_add(&ca->kobj, bdev_kobj(bdev), "bcache")) { |
9b299728 EW |
2345 | err = "error calling kobject_add"; |
2346 | ret = -ENOMEM; | |
2347 | goto out; | |
2348 | } | |
cafe5635 | 2349 | |
4fa03402 | 2350 | mutex_lock(&bch_register_lock); |
cafe5635 | 2351 | err = register_cache_set(ca); |
4fa03402 KO |
2352 | mutex_unlock(&bch_register_lock); |
2353 | ||
9b299728 EW |
2354 | if (err) { |
2355 | ret = -ENODEV; | |
2356 | goto out; | |
2357 | } | |
cafe5635 | 2358 | |
46f5aa88 | 2359 | pr_info("registered cache device %s\n", ca->cache_dev_name); |
9b299728 | 2360 | |
d83353b3 KO |
2361 | out: |
2362 | kobject_put(&ca->kobj); | |
9b299728 | 2363 | |
cafe5635 | 2364 | err: |
9b299728 | 2365 | if (err) |
46f5aa88 | 2366 | pr_notice("error %s: %s\n", ca->cache_dev_name, err); |
9b299728 EW |
2367 | |
2368 | return ret; | |
cafe5635 KO |
2369 | } |
2370 | ||
2371 | /* Global interfaces/init */ | |
2372 | ||
fc2d5988 CL |
2373 | static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, |
2374 | const char *buffer, size_t size); | |
0c277e21 CL |
2375 | static ssize_t bch_pending_bdevs_cleanup(struct kobject *k, |
2376 | struct kobj_attribute *attr, | |
2377 | const char *buffer, size_t size); | |
cafe5635 KO |
2378 | |
2379 | kobj_attribute_write(register, register_bcache); | |
2380 | kobj_attribute_write(register_quiet, register_bcache); | |
0c277e21 | 2381 | kobj_attribute_write(pendings_cleanup, bch_pending_bdevs_cleanup); |
cafe5635 | 2382 | |
4e7b5671 | 2383 | static bool bch_is_open_backing(dev_t dev) |
b3cf37bf | 2384 | { |
a9dd53ad GP |
2385 | struct cache_set *c, *tc; |
2386 | struct cached_dev *dc, *t; | |
2387 | ||
2388 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) | |
2389 | list_for_each_entry_safe(dc, t, &c->cached_devs, list) | |
4e7b5671 | 2390 | if (dc->bdev->bd_dev == dev) |
a9dd53ad GP |
2391 | return true; |
2392 | list_for_each_entry_safe(dc, t, &uncached_devices, list) | |
4e7b5671 | 2393 | if (dc->bdev->bd_dev == dev) |
a9dd53ad GP |
2394 | return true; |
2395 | return false; | |
2396 | } | |
2397 | ||
4e7b5671 | 2398 | static bool bch_is_open_cache(dev_t dev) |
b3cf37bf | 2399 | { |
a9dd53ad | 2400 | struct cache_set *c, *tc; |
a9dd53ad | 2401 | |
08fdb2cd CL |
2402 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) { |
2403 | struct cache *ca = c->cache; | |
2404 | ||
4e7b5671 | 2405 | if (ca->bdev->bd_dev == dev) |
08fdb2cd CL |
2406 | return true; |
2407 | } | |
2408 | ||
a9dd53ad GP |
2409 | return false; |
2410 | } | |
2411 | ||
4e7b5671 | 2412 | static bool bch_is_open(dev_t dev) |
b3cf37bf | 2413 | { |
4e7b5671 | 2414 | return bch_is_open_cache(dev) || bch_is_open_backing(dev); |
a9dd53ad GP |
2415 | } |
2416 | ||
9e23ccf8 | 2417 | struct async_reg_args { |
ee4a36f4 | 2418 | struct delayed_work reg_work; |
9e23ccf8 CL |
2419 | char *path; |
2420 | struct cache_sb *sb; | |
2421 | struct cache_sb_disk *sb_disk; | |
2422 | struct block_device *bdev; | |
2423 | }; | |
2424 | ||
2425 | static void register_bdev_worker(struct work_struct *work) | |
2426 | { | |
2427 | int fail = false; | |
2428 | struct async_reg_args *args = | |
ee4a36f4 | 2429 | container_of(work, struct async_reg_args, reg_work.work); |
9e23ccf8 CL |
2430 | struct cached_dev *dc; |
2431 | ||
2432 | dc = kzalloc(sizeof(*dc), GFP_KERNEL); | |
2433 | if (!dc) { | |
2434 | fail = true; | |
2435 | put_page(virt_to_page(args->sb_disk)); | |
2436 | blkdev_put(args->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
2437 | goto out; | |
2438 | } | |
2439 | ||
2440 | mutex_lock(&bch_register_lock); | |
2441 | if (register_bdev(args->sb, args->sb_disk, args->bdev, dc) < 0) | |
2442 | fail = true; | |
2443 | mutex_unlock(&bch_register_lock); | |
2444 | ||
2445 | out: | |
2446 | if (fail) | |
2447 | pr_info("error %s: fail to register backing device\n", | |
2448 | args->path); | |
2449 | kfree(args->sb); | |
2450 | kfree(args->path); | |
2451 | kfree(args); | |
2452 | module_put(THIS_MODULE); | |
2453 | } | |
2454 | ||
2455 | static void register_cache_worker(struct work_struct *work) | |
2456 | { | |
2457 | int fail = false; | |
2458 | struct async_reg_args *args = | |
ee4a36f4 | 2459 | container_of(work, struct async_reg_args, reg_work.work); |
9e23ccf8 CL |
2460 | struct cache *ca; |
2461 | ||
2462 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
2463 | if (!ca) { | |
2464 | fail = true; | |
2465 | put_page(virt_to_page(args->sb_disk)); | |
2466 | blkdev_put(args->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
2467 | goto out; | |
2468 | } | |
2469 | ||
2470 | /* blkdev_put() will be called in bch_cache_release() */ | |
2471 | if (register_cache(args->sb, args->sb_disk, args->bdev, ca) != 0) | |
2472 | fail = true; | |
2473 | ||
2474 | out: | |
2475 | if (fail) | |
2476 | pr_info("error %s: fail to register cache device\n", | |
2477 | args->path); | |
2478 | kfree(args->sb); | |
2479 | kfree(args->path); | |
2480 | kfree(args); | |
2481 | module_put(THIS_MODULE); | |
2482 | } | |
2483 | ||
2484 | static void register_device_aync(struct async_reg_args *args) | |
2485 | { | |
2486 | if (SB_IS_BDEV(args->sb)) | |
ee4a36f4 | 2487 | INIT_DELAYED_WORK(&args->reg_work, register_bdev_worker); |
9e23ccf8 | 2488 | else |
ee4a36f4 | 2489 | INIT_DELAYED_WORK(&args->reg_work, register_cache_worker); |
9e23ccf8 | 2490 | |
ee4a36f4 CL |
2491 | /* 10 jiffies is enough for a delay */ |
2492 | queue_delayed_work(system_wq, &args->reg_work, 10); | |
9e23ccf8 CL |
2493 | } |
2494 | ||
cafe5635 KO |
2495 | static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, |
2496 | const char *buffer, size_t size) | |
2497 | { | |
50246693 | 2498 | const char *err; |
29cda393 | 2499 | char *path = NULL; |
50246693 | 2500 | struct cache_sb *sb; |
cfa0c56d | 2501 | struct cache_sb_disk *sb_disk; |
fc8f19cc | 2502 | struct block_device *bdev; |
50246693 | 2503 | ssize_t ret; |
a58e88bf CL |
2504 | bool async_registration = false; |
2505 | ||
2506 | #ifdef CONFIG_BCACHE_ASYNC_REGISTRATION | |
2507 | async_registration = true; | |
2508 | #endif | |
cafe5635 | 2509 | |
50246693 | 2510 | ret = -EBUSY; |
29cda393 | 2511 | err = "failed to reference bcache module"; |
cafe5635 | 2512 | if (!try_module_get(THIS_MODULE)) |
50246693 | 2513 | goto out; |
cafe5635 | 2514 | |
a59ff6cc CL |
2515 | /* For latest state of bcache_is_reboot */ |
2516 | smp_mb(); | |
29cda393 | 2517 | err = "bcache is in reboot"; |
a59ff6cc | 2518 | if (bcache_is_reboot) |
50246693 | 2519 | goto out_module_put; |
a59ff6cc | 2520 | |
50246693 CH |
2521 | ret = -ENOMEM; |
2522 | err = "cannot allocate memory"; | |
a56489d4 FS |
2523 | path = kstrndup(buffer, size, GFP_KERNEL); |
2524 | if (!path) | |
50246693 | 2525 | goto out_module_put; |
a56489d4 FS |
2526 | |
2527 | sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL); | |
2528 | if (!sb) | |
50246693 | 2529 | goto out_free_path; |
cafe5635 | 2530 | |
50246693 | 2531 | ret = -EINVAL; |
cafe5635 KO |
2532 | err = "failed to open device"; |
2533 | bdev = blkdev_get_by_path(strim(path), | |
2534 | FMODE_READ|FMODE_WRITE|FMODE_EXCL, | |
2535 | sb); | |
f59fce84 | 2536 | if (IS_ERR(bdev)) { |
a9dd53ad | 2537 | if (bdev == ERR_PTR(-EBUSY)) { |
4e7b5671 CH |
2538 | dev_t dev; |
2539 | ||
789d21db | 2540 | mutex_lock(&bch_register_lock); |
4e7b5671 CH |
2541 | if (lookup_bdev(strim(path), &dev) == 0 && |
2542 | bch_is_open(dev)) | |
a9dd53ad GP |
2543 | err = "device already registered"; |
2544 | else | |
2545 | err = "device busy"; | |
789d21db | 2546 | mutex_unlock(&bch_register_lock); |
4b758df2 JK |
2547 | if (!IS_ERR(bdev)) |
2548 | bdput(bdev); | |
d7076f21 | 2549 | if (attr == &ksysfs_register_quiet) |
50246693 | 2550 | goto done; |
a9dd53ad | 2551 | } |
50246693 | 2552 | goto out_free_sb; |
f59fce84 KO |
2553 | } |
2554 | ||
2555 | err = "failed to set blocksize"; | |
2556 | if (set_blocksize(bdev, 4096)) | |
50246693 | 2557 | goto out_blkdev_put; |
cafe5635 | 2558 | |
cfa0c56d | 2559 | err = read_super(sb, bdev, &sb_disk); |
cafe5635 | 2560 | if (err) |
50246693 | 2561 | goto out_blkdev_put; |
cafe5635 | 2562 | |
cc40daf9 | 2563 | err = "failed to register device"; |
a58e88bf CL |
2564 | |
2565 | if (async_registration) { | |
9e23ccf8 CL |
2566 | /* register in asynchronous way */ |
2567 | struct async_reg_args *args = | |
2568 | kzalloc(sizeof(struct async_reg_args), GFP_KERNEL); | |
2569 | ||
2570 | if (!args) { | |
2571 | ret = -ENOMEM; | |
2572 | err = "cannot allocate memory"; | |
2573 | goto out_put_sb_page; | |
2574 | } | |
2575 | ||
2576 | args->path = path; | |
2577 | args->sb = sb; | |
2578 | args->sb_disk = sb_disk; | |
2579 | args->bdev = bdev; | |
2580 | register_device_aync(args); | |
2581 | /* No wait and returns to user space */ | |
2582 | goto async_done; | |
2583 | } | |
2584 | ||
2903381f | 2585 | if (SB_IS_BDEV(sb)) { |
cafe5635 | 2586 | struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL); |
1fae7cf0 | 2587 | |
f59fce84 | 2588 | if (!dc) |
50246693 | 2589 | goto out_put_sb_page; |
cafe5635 | 2590 | |
4fa03402 | 2591 | mutex_lock(&bch_register_lock); |
cfa0c56d | 2592 | ret = register_bdev(sb, sb_disk, bdev, dc); |
4fa03402 | 2593 | mutex_unlock(&bch_register_lock); |
bb6d355c | 2594 | /* blkdev_put() will be called in cached_dev_free() */ |
fc8f19cc CH |
2595 | if (ret < 0) |
2596 | goto out_free_sb; | |
cafe5635 KO |
2597 | } else { |
2598 | struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
1fae7cf0 | 2599 | |
f59fce84 | 2600 | if (!ca) |
50246693 | 2601 | goto out_put_sb_page; |
cafe5635 | 2602 | |
bb6d355c | 2603 | /* blkdev_put() will be called in bch_cache_release() */ |
cfa0c56d | 2604 | if (register_cache(sb, sb_disk, bdev, ca) != 0) |
fc8f19cc | 2605 | goto out_free_sb; |
cafe5635 | 2606 | } |
50246693 | 2607 | |
50246693 | 2608 | done: |
cafe5635 KO |
2609 | kfree(sb); |
2610 | kfree(path); | |
cafe5635 | 2611 | module_put(THIS_MODULE); |
9e23ccf8 | 2612 | async_done: |
50246693 CH |
2613 | return size; |
2614 | ||
2615 | out_put_sb_page: | |
cfa0c56d | 2616 | put_page(virt_to_page(sb_disk)); |
50246693 | 2617 | out_blkdev_put: |
fc8f19cc | 2618 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
50246693 CH |
2619 | out_free_sb: |
2620 | kfree(sb); | |
2621 | out_free_path: | |
2622 | kfree(path); | |
ae3cd299 | 2623 | path = NULL; |
50246693 CH |
2624 | out_module_put: |
2625 | module_put(THIS_MODULE); | |
2626 | out: | |
46f5aa88 | 2627 | pr_info("error %s: %s\n", path?path:"", err); |
50246693 | 2628 | return ret; |
cafe5635 KO |
2629 | } |
2630 | ||
0c277e21 CL |
2631 | |
2632 | struct pdev { | |
2633 | struct list_head list; | |
2634 | struct cached_dev *dc; | |
2635 | }; | |
2636 | ||
2637 | static ssize_t bch_pending_bdevs_cleanup(struct kobject *k, | |
2638 | struct kobj_attribute *attr, | |
2639 | const char *buffer, | |
2640 | size_t size) | |
2641 | { | |
2642 | LIST_HEAD(pending_devs); | |
2643 | ssize_t ret = size; | |
2644 | struct cached_dev *dc, *tdc; | |
2645 | struct pdev *pdev, *tpdev; | |
2646 | struct cache_set *c, *tc; | |
2647 | ||
2648 | mutex_lock(&bch_register_lock); | |
2649 | list_for_each_entry_safe(dc, tdc, &uncached_devices, list) { | |
2650 | pdev = kmalloc(sizeof(struct pdev), GFP_KERNEL); | |
2651 | if (!pdev) | |
2652 | break; | |
2653 | pdev->dc = dc; | |
2654 | list_add(&pdev->list, &pending_devs); | |
2655 | } | |
2656 | ||
2657 | list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) { | |
2658 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) { | |
2659 | char *pdev_set_uuid = pdev->dc->sb.set_uuid; | |
1132e56e | 2660 | char *set_uuid = c->set_uuid; |
0c277e21 CL |
2661 | |
2662 | if (!memcmp(pdev_set_uuid, set_uuid, 16)) { | |
2663 | list_del(&pdev->list); | |
2664 | kfree(pdev); | |
2665 | break; | |
2666 | } | |
2667 | } | |
2668 | } | |
2669 | mutex_unlock(&bch_register_lock); | |
2670 | ||
2671 | list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) { | |
46f5aa88 | 2672 | pr_info("delete pdev %p\n", pdev); |
0c277e21 CL |
2673 | list_del(&pdev->list); |
2674 | bcache_device_stop(&pdev->dc->disk); | |
2675 | kfree(pdev); | |
2676 | } | |
2677 | ||
2678 | return ret; | |
2679 | } | |
2680 | ||
cafe5635 KO |
2681 | static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) |
2682 | { | |
a59ff6cc CL |
2683 | if (bcache_is_reboot) |
2684 | return NOTIFY_DONE; | |
2685 | ||
cafe5635 KO |
2686 | if (code == SYS_DOWN || |
2687 | code == SYS_HALT || | |
2688 | code == SYS_POWER_OFF) { | |
2689 | DEFINE_WAIT(wait); | |
2690 | unsigned long start = jiffies; | |
2691 | bool stopped = false; | |
2692 | ||
2693 | struct cache_set *c, *tc; | |
2694 | struct cached_dev *dc, *tdc; | |
2695 | ||
2696 | mutex_lock(&bch_register_lock); | |
2697 | ||
a59ff6cc CL |
2698 | if (bcache_is_reboot) |
2699 | goto out; | |
2700 | ||
2701 | /* New registration is rejected since now */ | |
2702 | bcache_is_reboot = true; | |
2703 | /* | |
2704 | * Make registering caller (if there is) on other CPU | |
2705 | * core know bcache_is_reboot set to true earlier | |
2706 | */ | |
2707 | smp_mb(); | |
2708 | ||
cafe5635 KO |
2709 | if (list_empty(&bch_cache_sets) && |
2710 | list_empty(&uncached_devices)) | |
2711 | goto out; | |
2712 | ||
a59ff6cc CL |
2713 | mutex_unlock(&bch_register_lock); |
2714 | ||
46f5aa88 | 2715 | pr_info("Stopping all devices:\n"); |
cafe5635 | 2716 | |
a59ff6cc CL |
2717 | /* |
2718 | * The reason bch_register_lock is not held to call | |
2719 | * bch_cache_set_stop() and bcache_device_stop() is to | |
2720 | * avoid potential deadlock during reboot, because cache | |
2721 | * set or bcache device stopping process will acqurie | |
2722 | * bch_register_lock too. | |
2723 | * | |
2724 | * We are safe here because bcache_is_reboot sets to | |
2725 | * true already, register_bcache() will reject new | |
2726 | * registration now. bcache_is_reboot also makes sure | |
2727 | * bcache_reboot() won't be re-entered on by other thread, | |
2728 | * so there is no race in following list iteration by | |
2729 | * list_for_each_entry_safe(). | |
2730 | */ | |
cafe5635 KO |
2731 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) |
2732 | bch_cache_set_stop(c); | |
2733 | ||
2734 | list_for_each_entry_safe(dc, tdc, &uncached_devices, list) | |
2735 | bcache_device_stop(&dc->disk); | |
2736 | ||
eb8cbb6d CL |
2737 | |
2738 | /* | |
2739 | * Give an early chance for other kthreads and | |
2740 | * kworkers to stop themselves | |
2741 | */ | |
2742 | schedule(); | |
2743 | ||
cafe5635 KO |
2744 | /* What's a condition variable? */ |
2745 | while (1) { | |
eb8cbb6d | 2746 | long timeout = start + 10 * HZ - jiffies; |
cafe5635 | 2747 | |
eb8cbb6d | 2748 | mutex_lock(&bch_register_lock); |
cafe5635 KO |
2749 | stopped = list_empty(&bch_cache_sets) && |
2750 | list_empty(&uncached_devices); | |
2751 | ||
2752 | if (timeout < 0 || stopped) | |
2753 | break; | |
2754 | ||
2755 | prepare_to_wait(&unregister_wait, &wait, | |
2756 | TASK_UNINTERRUPTIBLE); | |
2757 | ||
2758 | mutex_unlock(&bch_register_lock); | |
2759 | schedule_timeout(timeout); | |
cafe5635 KO |
2760 | } |
2761 | ||
2762 | finish_wait(&unregister_wait, &wait); | |
2763 | ||
2764 | if (stopped) | |
46f5aa88 | 2765 | pr_info("All devices stopped\n"); |
cafe5635 | 2766 | else |
46f5aa88 | 2767 | pr_notice("Timeout waiting for devices to be closed\n"); |
cafe5635 KO |
2768 | out: |
2769 | mutex_unlock(&bch_register_lock); | |
2770 | } | |
2771 | ||
2772 | return NOTIFY_DONE; | |
2773 | } | |
2774 | ||
2775 | static struct notifier_block reboot = { | |
2776 | .notifier_call = bcache_reboot, | |
2777 | .priority = INT_MAX, /* before any real devices */ | |
2778 | }; | |
2779 | ||
2780 | static void bcache_exit(void) | |
2781 | { | |
2782 | bch_debug_exit(); | |
cafe5635 | 2783 | bch_request_exit(); |
cafe5635 KO |
2784 | if (bcache_kobj) |
2785 | kobject_put(bcache_kobj); | |
2786 | if (bcache_wq) | |
2787 | destroy_workqueue(bcache_wq); | |
0f843e65 GF |
2788 | if (bch_journal_wq) |
2789 | destroy_workqueue(bch_journal_wq); | |
2790 | ||
5c41c8a7 KO |
2791 | if (bcache_major) |
2792 | unregister_blkdev(bcache_major, "bcache"); | |
cafe5635 | 2793 | unregister_reboot_notifier(&reboot); |
330a4db8 | 2794 | mutex_destroy(&bch_register_lock); |
cafe5635 KO |
2795 | } |
2796 | ||
9aaf5165 CL |
2797 | /* Check and fixup module parameters */ |
2798 | static void check_module_parameters(void) | |
2799 | { | |
2800 | if (bch_cutoff_writeback_sync == 0) | |
2801 | bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC; | |
2802 | else if (bch_cutoff_writeback_sync > CUTOFF_WRITEBACK_SYNC_MAX) { | |
46f5aa88 | 2803 | pr_warn("set bch_cutoff_writeback_sync (%u) to max value %u\n", |
9aaf5165 CL |
2804 | bch_cutoff_writeback_sync, CUTOFF_WRITEBACK_SYNC_MAX); |
2805 | bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC_MAX; | |
2806 | } | |
2807 | ||
2808 | if (bch_cutoff_writeback == 0) | |
2809 | bch_cutoff_writeback = CUTOFF_WRITEBACK; | |
2810 | else if (bch_cutoff_writeback > CUTOFF_WRITEBACK_MAX) { | |
46f5aa88 | 2811 | pr_warn("set bch_cutoff_writeback (%u) to max value %u\n", |
9aaf5165 CL |
2812 | bch_cutoff_writeback, CUTOFF_WRITEBACK_MAX); |
2813 | bch_cutoff_writeback = CUTOFF_WRITEBACK_MAX; | |
2814 | } | |
2815 | ||
2816 | if (bch_cutoff_writeback > bch_cutoff_writeback_sync) { | |
46f5aa88 | 2817 | pr_warn("set bch_cutoff_writeback (%u) to %u\n", |
9aaf5165 CL |
2818 | bch_cutoff_writeback, bch_cutoff_writeback_sync); |
2819 | bch_cutoff_writeback = bch_cutoff_writeback_sync; | |
2820 | } | |
2821 | } | |
2822 | ||
cafe5635 KO |
2823 | static int __init bcache_init(void) |
2824 | { | |
2825 | static const struct attribute *files[] = { | |
2826 | &ksysfs_register.attr, | |
2827 | &ksysfs_register_quiet.attr, | |
0c277e21 | 2828 | &ksysfs_pendings_cleanup.attr, |
cafe5635 KO |
2829 | NULL |
2830 | }; | |
2831 | ||
9aaf5165 CL |
2832 | check_module_parameters(); |
2833 | ||
cafe5635 KO |
2834 | mutex_init(&bch_register_lock); |
2835 | init_waitqueue_head(&unregister_wait); | |
2836 | register_reboot_notifier(&reboot); | |
2837 | ||
2838 | bcache_major = register_blkdev(0, "bcache"); | |
2ecf0cdb ZL |
2839 | if (bcache_major < 0) { |
2840 | unregister_reboot_notifier(&reboot); | |
330a4db8 | 2841 | mutex_destroy(&bch_register_lock); |
cafe5635 | 2842 | return bcache_major; |
2ecf0cdb | 2843 | } |
cafe5635 | 2844 | |
16c1fdf4 FS |
2845 | bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0); |
2846 | if (!bcache_wq) | |
2847 | goto err; | |
2848 | ||
0f843e65 GF |
2849 | bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0); |
2850 | if (!bch_journal_wq) | |
2851 | goto err; | |
2852 | ||
16c1fdf4 FS |
2853 | bcache_kobj = kobject_create_and_add("bcache", fs_kobj); |
2854 | if (!bcache_kobj) | |
2855 | goto err; | |
2856 | ||
2857 | if (bch_request_init() || | |
330a4db8 | 2858 | sysfs_create_files(bcache_kobj, files)) |
cafe5635 KO |
2859 | goto err; |
2860 | ||
91bafdf0 | 2861 | bch_debug_init(); |
78ac2107 CL |
2862 | closure_debug_init(); |
2863 | ||
a59ff6cc CL |
2864 | bcache_is_reboot = false; |
2865 | ||
cafe5635 KO |
2866 | return 0; |
2867 | err: | |
2868 | bcache_exit(); | |
2869 | return -ENOMEM; | |
2870 | } | |
2871 | ||
9aaf5165 CL |
2872 | /* |
2873 | * Module hooks | |
2874 | */ | |
cafe5635 KO |
2875 | module_exit(bcache_exit); |
2876 | module_init(bcache_init); | |
009673d0 | 2877 | |
9aaf5165 CL |
2878 | module_param(bch_cutoff_writeback, uint, 0); |
2879 | MODULE_PARM_DESC(bch_cutoff_writeback, "threshold to cutoff writeback"); | |
2880 | ||
2881 | module_param(bch_cutoff_writeback_sync, uint, 0); | |
2882 | MODULE_PARM_DESC(bch_cutoff_writeback_sync, "hard threshold to cutoff writeback"); | |
2883 | ||
009673d0 CL |
2884 | MODULE_DESCRIPTION("Bcache: a Linux block layer cache"); |
2885 | MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); | |
2886 | MODULE_LICENSE("GPL"); |