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> |
4ee60ec1 | 19 | #include <linux/pagemap.h> |
cafe5635 | 20 | #include <linux/debugfs.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; | |
afe78ab4 | 52 | struct workqueue_struct *bch_flush_wq; |
0f843e65 | 53 | struct workqueue_struct *bch_journal_wq; |
cafe5635 | 54 | |
a59ff6cc | 55 | |
cafe5635 | 56 | #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE) |
1dbe32ad CL |
57 | /* limitation of partitions number on single bcache device */ |
58 | #define BCACHE_MINORS 128 | |
59 | /* limitation of bcache devices number on single system */ | |
60 | #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS) | |
cafe5635 | 61 | |
cafe5635 KO |
62 | /* Superblock */ |
63 | ||
ffa47032 CL |
64 | static unsigned int get_bucket_size(struct cache_sb *sb, struct cache_sb_disk *s) |
65 | { | |
66 | unsigned int bucket_size = le16_to_cpu(s->bucket_size); | |
67 | ||
b16671e8 CL |
68 | if (sb->version >= BCACHE_SB_VERSION_CDEV_WITH_FEATURES) { |
69 | if (bch_has_feature_large_bucket(sb)) { | |
70 | unsigned int max, order; | |
71 | ||
72 | max = sizeof(unsigned int) * BITS_PER_BYTE - 1; | |
73 | order = le16_to_cpu(s->bucket_size); | |
74 | /* | |
75 | * bcache tool will make sure the overflow won't | |
76 | * happen, an error message here is enough. | |
77 | */ | |
78 | if (order > max) | |
79 | pr_err("Bucket size (1 << %u) overflows\n", | |
80 | order); | |
81 | bucket_size = 1 << order; | |
82 | } else if (bch_has_feature_obso_large_bucket(sb)) { | |
83 | bucket_size += | |
84 | le16_to_cpu(s->obso_bucket_size_hi) << 16; | |
85 | } | |
86 | } | |
ffa47032 CL |
87 | |
88 | return bucket_size; | |
89 | } | |
90 | ||
5b21403c CL |
91 | static const char *read_super_common(struct cache_sb *sb, struct block_device *bdev, |
92 | struct cache_sb_disk *s) | |
93 | { | |
94 | const char *err; | |
95 | unsigned int i; | |
96 | ||
198efa35 | 97 | sb->first_bucket= le16_to_cpu(s->first_bucket); |
5b21403c | 98 | sb->nbuckets = le64_to_cpu(s->nbuckets); |
ffa47032 | 99 | sb->bucket_size = get_bucket_size(sb, s); |
5b21403c CL |
100 | |
101 | sb->nr_in_set = le16_to_cpu(s->nr_in_set); | |
102 | sb->nr_this_dev = le16_to_cpu(s->nr_this_dev); | |
103 | ||
198efa35 CL |
104 | err = "Too many journal buckets"; |
105 | if (sb->keys > SB_JOURNAL_BUCKETS) | |
106 | goto err; | |
107 | ||
5b21403c CL |
108 | err = "Too many buckets"; |
109 | if (sb->nbuckets > LONG_MAX) | |
110 | goto err; | |
111 | ||
112 | err = "Not enough buckets"; | |
113 | if (sb->nbuckets < 1 << 7) | |
114 | goto err; | |
115 | ||
c557a5f7 CL |
116 | err = "Bad block size (not power of 2)"; |
117 | if (!is_power_of_2(sb->block_size)) | |
118 | goto err; | |
119 | ||
120 | err = "Bad block size (larger than page size)"; | |
121 | if (sb->block_size > PAGE_SECTORS) | |
122 | goto err; | |
123 | ||
124 | err = "Bad bucket size (not power of 2)"; | |
125 | if (!is_power_of_2(sb->bucket_size)) | |
126 | goto err; | |
127 | ||
128 | err = "Bad bucket size (smaller than page size)"; | |
129 | if (sb->bucket_size < PAGE_SECTORS) | |
5b21403c CL |
130 | goto err; |
131 | ||
132 | err = "Invalid superblock: device too small"; | |
133 | if (get_capacity(bdev->bd_disk) < | |
134 | sb->bucket_size * sb->nbuckets) | |
135 | goto err; | |
136 | ||
137 | err = "Bad UUID"; | |
138 | if (bch_is_zero(sb->set_uuid, 16)) | |
139 | goto err; | |
140 | ||
141 | err = "Bad cache device number in set"; | |
142 | if (!sb->nr_in_set || | |
143 | sb->nr_in_set <= sb->nr_this_dev || | |
144 | sb->nr_in_set > MAX_CACHES_PER_SET) | |
145 | goto err; | |
146 | ||
147 | err = "Journal buckets not sequential"; | |
148 | for (i = 0; i < sb->keys; i++) | |
149 | if (sb->d[i] != sb->first_bucket + i) | |
150 | goto err; | |
151 | ||
152 | err = "Too many journal buckets"; | |
153 | if (sb->first_bucket + sb->keys > sb->nbuckets) | |
154 | goto err; | |
155 | ||
156 | err = "Invalid superblock: first bucket comes before end of super"; | |
157 | if (sb->first_bucket * sb->bucket_size < 16) | |
158 | goto err; | |
159 | ||
160 | err = NULL; | |
161 | err: | |
162 | return err; | |
163 | } | |
164 | ||
165 | ||
cafe5635 | 166 | static const char *read_super(struct cache_sb *sb, struct block_device *bdev, |
cfa0c56d | 167 | struct cache_sb_disk **res) |
cafe5635 KO |
168 | { |
169 | const char *err; | |
a702a692 | 170 | struct cache_sb_disk *s; |
6321bef0 | 171 | struct page *page; |
6f10f7d1 | 172 | unsigned int i; |
cafe5635 | 173 | |
224941e8 | 174 | page = read_cache_page_gfp(bdev->bd_mapping, |
6321bef0 CH |
175 | SB_OFFSET >> PAGE_SHIFT, GFP_KERNEL); |
176 | if (IS_ERR(page)) | |
cafe5635 | 177 | return "IO error"; |
6321bef0 | 178 | s = page_address(page) + offset_in_page(SB_OFFSET); |
cafe5635 KO |
179 | |
180 | sb->offset = le64_to_cpu(s->offset); | |
181 | sb->version = le64_to_cpu(s->version); | |
182 | ||
183 | memcpy(sb->magic, s->magic, 16); | |
184 | memcpy(sb->uuid, s->uuid, 16); | |
185 | memcpy(sb->set_uuid, s->set_uuid, 16); | |
186 | memcpy(sb->label, s->label, SB_LABEL_SIZE); | |
187 | ||
188 | sb->flags = le64_to_cpu(s->flags); | |
189 | sb->seq = le64_to_cpu(s->seq); | |
cafe5635 | 190 | sb->last_mount = le32_to_cpu(s->last_mount); |
cafe5635 KO |
191 | sb->keys = le16_to_cpu(s->keys); |
192 | ||
193 | for (i = 0; i < SB_JOURNAL_BUCKETS; i++) | |
194 | sb->d[i] = le64_to_cpu(s->d[i]); | |
195 | ||
46f5aa88 | 196 | pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u\n", |
cafe5635 KO |
197 | sb->version, sb->flags, sb->seq, sb->keys); |
198 | ||
aaf8dbea | 199 | err = "Not a bcache superblock (bad offset)"; |
cafe5635 KO |
200 | if (sb->offset != SB_SECTOR) |
201 | goto err; | |
202 | ||
aaf8dbea | 203 | err = "Not a bcache superblock (bad magic)"; |
cafe5635 KO |
204 | if (memcmp(sb->magic, bcache_magic, 16)) |
205 | goto err; | |
206 | ||
cafe5635 KO |
207 | err = "Bad checksum"; |
208 | if (s->csum != csum_set(s)) | |
209 | goto err; | |
210 | ||
211 | err = "Bad UUID"; | |
169ef1cf | 212 | if (bch_is_zero(sb->uuid, 16)) |
cafe5635 KO |
213 | goto err; |
214 | ||
8abb2a5d KO |
215 | sb->block_size = le16_to_cpu(s->block_size); |
216 | ||
217 | err = "Superblock block size smaller than device block size"; | |
218 | if (sb->block_size << 9 < bdev_logical_block_size(bdev)) | |
219 | goto err; | |
220 | ||
2903381f KO |
221 | switch (sb->version) { |
222 | case BCACHE_SB_VERSION_BDEV: | |
2903381f KO |
223 | sb->data_offset = BDEV_DATA_START_DEFAULT; |
224 | break; | |
225 | case BCACHE_SB_VERSION_BDEV_WITH_OFFSET: | |
d721a43f | 226 | case BCACHE_SB_VERSION_BDEV_WITH_FEATURES: |
2903381f KO |
227 | sb->data_offset = le64_to_cpu(s->data_offset); |
228 | ||
229 | err = "Bad data offset"; | |
230 | if (sb->data_offset < BDEV_DATA_START_DEFAULT) | |
231 | goto err; | |
cafe5635 | 232 | |
2903381f KO |
233 | break; |
234 | case BCACHE_SB_VERSION_CDEV: | |
235 | case BCACHE_SB_VERSION_CDEV_WITH_UUID: | |
5b21403c CL |
236 | err = read_super_common(sb, bdev, s); |
237 | if (err) | |
2903381f | 238 | goto err; |
2903381f | 239 | break; |
d721a43f | 240 | case BCACHE_SB_VERSION_CDEV_WITH_FEATURES: |
ffa47032 CL |
241 | /* |
242 | * Feature bits are needed in read_super_common(), | |
243 | * convert them firstly. | |
244 | */ | |
d721a43f CL |
245 | sb->feature_compat = le64_to_cpu(s->feature_compat); |
246 | sb->feature_incompat = le64_to_cpu(s->feature_incompat); | |
247 | sb->feature_ro_compat = le64_to_cpu(s->feature_ro_compat); | |
1dfc0686 CL |
248 | |
249 | /* Check incompatible features */ | |
250 | err = "Unsupported compatible feature found"; | |
251 | if (bch_has_unknown_compat_features(sb)) | |
252 | goto err; | |
253 | ||
254 | err = "Unsupported read-only compatible feature found"; | |
255 | if (bch_has_unknown_ro_compat_features(sb)) | |
256 | goto err; | |
257 | ||
258 | err = "Unsupported incompatible feature found"; | |
259 | if (bch_has_unknown_incompat_features(sb)) | |
260 | goto err; | |
261 | ||
ffa47032 CL |
262 | err = read_super_common(sb, bdev, s); |
263 | if (err) | |
2903381f | 264 | goto err; |
2903381f KO |
265 | break; |
266 | default: | |
267 | err = "Unsupported superblock version"; | |
cafe5635 | 268 | goto err; |
2903381f KO |
269 | } |
270 | ||
75cbb3f1 | 271 | sb->last_mount = (u32)ktime_get_real_seconds(); |
cfa0c56d | 272 | *res = s; |
6321bef0 | 273 | return NULL; |
cafe5635 | 274 | err: |
6321bef0 | 275 | put_page(page); |
cafe5635 KO |
276 | return err; |
277 | } | |
278 | ||
4246a0b6 | 279 | static void write_bdev_super_endio(struct bio *bio) |
cafe5635 KO |
280 | { |
281 | struct cached_dev *dc = bio->bi_private; | |
08ec1e62 CL |
282 | |
283 | if (bio->bi_status) | |
284 | bch_count_backing_io_errors(dc, bio); | |
cafe5635 | 285 | |
cb7a583e | 286 | closure_put(&dc->sb_write); |
cafe5635 KO |
287 | } |
288 | ||
475389ae CH |
289 | static void __write_super(struct cache_sb *sb, struct cache_sb_disk *out, |
290 | struct bio *bio) | |
cafe5635 | 291 | { |
6f10f7d1 | 292 | unsigned int i; |
cafe5635 | 293 | |
475389ae | 294 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META; |
4f024f37 | 295 | bio->bi_iter.bi_sector = SB_SECTOR; |
475389ae CH |
296 | __bio_add_page(bio, virt_to_page(out), SB_SIZE, |
297 | offset_in_page(out)); | |
cafe5635 KO |
298 | |
299 | out->offset = cpu_to_le64(sb->offset); | |
cafe5635 KO |
300 | |
301 | memcpy(out->uuid, sb->uuid, 16); | |
302 | memcpy(out->set_uuid, sb->set_uuid, 16); | |
303 | memcpy(out->label, sb->label, SB_LABEL_SIZE); | |
304 | ||
305 | out->flags = cpu_to_le64(sb->flags); | |
306 | out->seq = cpu_to_le64(sb->seq); | |
307 | ||
308 | out->last_mount = cpu_to_le32(sb->last_mount); | |
309 | out->first_bucket = cpu_to_le16(sb->first_bucket); | |
310 | out->keys = cpu_to_le16(sb->keys); | |
311 | ||
312 | for (i = 0; i < sb->keys; i++) | |
313 | out->d[i] = cpu_to_le64(sb->d[i]); | |
314 | ||
d721a43f CL |
315 | if (sb->version >= BCACHE_SB_VERSION_CDEV_WITH_FEATURES) { |
316 | out->feature_compat = cpu_to_le64(sb->feature_compat); | |
317 | out->feature_incompat = cpu_to_le64(sb->feature_incompat); | |
318 | out->feature_ro_compat = cpu_to_le64(sb->feature_ro_compat); | |
319 | } | |
320 | ||
321 | out->version = cpu_to_le64(sb->version); | |
cafe5635 KO |
322 | out->csum = csum_set(out); |
323 | ||
46f5aa88 | 324 | pr_debug("ver %llu, flags %llu, seq %llu\n", |
cafe5635 KO |
325 | sb->version, sb->flags, sb->seq); |
326 | ||
4e49ea4a | 327 | submit_bio(bio); |
cafe5635 KO |
328 | } |
329 | ||
d4e3b928 | 330 | static CLOSURE_CALLBACK(bch_write_bdev_super_unlock) |
cb7a583e | 331 | { |
d4e3b928 | 332 | closure_type(dc, struct cached_dev, sb_write); |
cb7a583e KO |
333 | |
334 | up(&dc->sb_write_mutex); | |
335 | } | |
336 | ||
cafe5635 KO |
337 | void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent) |
338 | { | |
cb7a583e | 339 | struct closure *cl = &dc->sb_write; |
cafe5635 KO |
340 | struct bio *bio = &dc->sb_bio; |
341 | ||
cb7a583e KO |
342 | down(&dc->sb_write_mutex); |
343 | closure_init(cl, parent); | |
cafe5635 | 344 | |
49add496 | 345 | bio_init(bio, dc->bdev, dc->sb_bv, 1, 0); |
cafe5635 KO |
346 | bio->bi_end_io = write_bdev_super_endio; |
347 | bio->bi_private = dc; | |
348 | ||
349 | closure_get(cl); | |
27a40ab9 | 350 | /* I/O request sent to backing device */ |
475389ae | 351 | __write_super(&dc->sb, dc->sb_disk, bio); |
cafe5635 | 352 | |
cb7a583e | 353 | closure_return_with_destructor(cl, bch_write_bdev_super_unlock); |
cafe5635 KO |
354 | } |
355 | ||
4246a0b6 | 356 | static void write_super_endio(struct bio *bio) |
cafe5635 KO |
357 | { |
358 | struct cache *ca = bio->bi_private; | |
359 | ||
5138ac67 CL |
360 | /* is_read = 0 */ |
361 | bch_count_io_errors(ca, bio->bi_status, 0, | |
362 | "writing superblock"); | |
cb7a583e KO |
363 | closure_put(&ca->set->sb_write); |
364 | } | |
365 | ||
d4e3b928 | 366 | static CLOSURE_CALLBACK(bcache_write_super_unlock) |
cb7a583e | 367 | { |
d4e3b928 | 368 | closure_type(c, struct cache_set, sb_write); |
cb7a583e KO |
369 | |
370 | up(&c->sb_write_mutex); | |
cafe5635 KO |
371 | } |
372 | ||
373 | void bcache_write_super(struct cache_set *c) | |
374 | { | |
cb7a583e | 375 | struct closure *cl = &c->sb_write; |
08fdb2cd CL |
376 | struct cache *ca = c->cache; |
377 | struct bio *bio = &ca->sb_bio; | |
378 | unsigned int version = BCACHE_SB_VERSION_CDEV_WITH_UUID; | |
cafe5635 | 379 | |
cb7a583e KO |
380 | down(&c->sb_write_mutex); |
381 | closure_init(cl, &c->cl); | |
cafe5635 | 382 | |
4a784266 | 383 | ca->sb.seq++; |
cafe5635 | 384 | |
4a784266 CL |
385 | if (ca->sb.version < version) |
386 | ca->sb.version = version; | |
cafe5635 | 387 | |
49add496 | 388 | bio_init(bio, ca->bdev, ca->sb_bv, 1, 0); |
08fdb2cd CL |
389 | bio->bi_end_io = write_super_endio; |
390 | bio->bi_private = ca; | |
cafe5635 | 391 | |
08fdb2cd CL |
392 | closure_get(cl); |
393 | __write_super(&ca->sb, ca->sb_disk, bio); | |
cafe5635 | 394 | |
cb7a583e | 395 | closure_return_with_destructor(cl, bcache_write_super_unlock); |
cafe5635 KO |
396 | } |
397 | ||
398 | /* UUID io */ | |
399 | ||
4246a0b6 | 400 | static void uuid_endio(struct bio *bio) |
cafe5635 KO |
401 | { |
402 | struct closure *cl = bio->bi_private; | |
cb7a583e | 403 | struct cache_set *c = container_of(cl, struct cache_set, uuid_write); |
cafe5635 | 404 | |
4e4cbee9 | 405 | cache_set_err_on(bio->bi_status, c, "accessing uuids"); |
cafe5635 KO |
406 | bch_bbio_free(bio, c); |
407 | closure_put(cl); | |
408 | } | |
409 | ||
d4e3b928 | 410 | static CLOSURE_CALLBACK(uuid_io_unlock) |
cb7a583e | 411 | { |
d4e3b928 | 412 | closure_type(c, struct cache_set, uuid_write); |
cb7a583e KO |
413 | |
414 | up(&c->uuid_write_mutex); | |
415 | } | |
416 | ||
9a4fd6a2 BVA |
417 | static void uuid_io(struct cache_set *c, blk_opf_t opf, struct bkey *k, |
418 | struct closure *parent) | |
cafe5635 | 419 | { |
cb7a583e | 420 | struct closure *cl = &c->uuid_write; |
cafe5635 | 421 | struct uuid_entry *u; |
6f10f7d1 | 422 | unsigned int i; |
85b1492e | 423 | char buf[80]; |
cafe5635 KO |
424 | |
425 | BUG_ON(!parent); | |
cb7a583e KO |
426 | down(&c->uuid_write_mutex); |
427 | closure_init(cl, parent); | |
cafe5635 KO |
428 | |
429 | for (i = 0; i < KEY_PTRS(k); i++) { | |
430 | struct bio *bio = bch_bbio_alloc(c); | |
431 | ||
9a4fd6a2 | 432 | bio->bi_opf = opf | REQ_SYNC | REQ_META; |
4f024f37 | 433 | bio->bi_iter.bi_size = KEY_SIZE(k) << 9; |
cafe5635 KO |
434 | |
435 | bio->bi_end_io = uuid_endio; | |
436 | bio->bi_private = cl; | |
169ef1cf | 437 | bch_bio_map(bio, c->uuids); |
cafe5635 KO |
438 | |
439 | bch_submit_bbio(bio, c, k, i); | |
440 | ||
9a4fd6a2 | 441 | if ((opf & REQ_OP_MASK) != REQ_OP_WRITE) |
cafe5635 KO |
442 | break; |
443 | } | |
444 | ||
dc9d98d6 | 445 | bch_extent_to_text(buf, sizeof(buf), k); |
9a4fd6a2 BVA |
446 | pr_debug("%s UUIDs at %s\n", (opf & REQ_OP_MASK) == REQ_OP_WRITE ? |
447 | "wrote" : "read", buf); | |
cafe5635 KO |
448 | |
449 | for (u = c->uuids; u < c->uuids + c->nr_uuids; u++) | |
169ef1cf | 450 | if (!bch_is_zero(u->uuid, 16)) |
46f5aa88 | 451 | pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u\n", |
cafe5635 KO |
452 | u - c->uuids, u->uuid, u->label, |
453 | u->first_reg, u->last_reg, u->invalidated); | |
454 | ||
cb7a583e | 455 | closure_return_with_destructor(cl, uuid_io_unlock); |
cafe5635 KO |
456 | } |
457 | ||
458 | static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl) | |
459 | { | |
460 | struct bkey *k = &j->uuid_bucket; | |
461 | ||
65d45231 | 462 | if (__bch_btree_ptr_invalid(c, k)) |
cafe5635 KO |
463 | return "bad uuid pointer"; |
464 | ||
465 | bkey_copy(&c->uuid_bucket, k); | |
9a4fd6a2 | 466 | uuid_io(c, REQ_OP_READ, k, cl); |
cafe5635 KO |
467 | |
468 | if (j->version < BCACHE_JSET_VERSION_UUIDv1) { | |
469 | struct uuid_entry_v0 *u0 = (void *) c->uuids; | |
470 | struct uuid_entry *u1 = (void *) c->uuids; | |
471 | int i; | |
472 | ||
473 | closure_sync(cl); | |
474 | ||
475 | /* | |
476 | * Since the new uuid entry is bigger than the old, we have to | |
477 | * convert starting at the highest memory address and work down | |
478 | * in order to do it in place | |
479 | */ | |
480 | ||
481 | for (i = c->nr_uuids - 1; | |
482 | i >= 0; | |
483 | --i) { | |
484 | memcpy(u1[i].uuid, u0[i].uuid, 16); | |
485 | memcpy(u1[i].label, u0[i].label, 32); | |
486 | ||
487 | u1[i].first_reg = u0[i].first_reg; | |
488 | u1[i].last_reg = u0[i].last_reg; | |
489 | u1[i].invalidated = u0[i].invalidated; | |
490 | ||
491 | u1[i].flags = 0; | |
492 | u1[i].sectors = 0; | |
493 | } | |
494 | } | |
495 | ||
496 | return NULL; | |
497 | } | |
498 | ||
499 | static int __uuid_write(struct cache_set *c) | |
500 | { | |
501 | BKEY_PADDED(key) k; | |
502 | struct closure cl; | |
4a784266 | 503 | struct cache *ca = c->cache; |
21e478dd | 504 | unsigned int size; |
cafe5635 | 505 | |
1fae7cf0 | 506 | closure_init_stack(&cl); |
cafe5635 KO |
507 | lockdep_assert_held(&bch_register_lock); |
508 | ||
17e4aed8 | 509 | if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, true)) |
cafe5635 KO |
510 | return 1; |
511 | ||
4a784266 | 512 | size = meta_bucket_pages(&ca->sb) * PAGE_SECTORS; |
21e478dd | 513 | SET_KEY_SIZE(&k.key, size); |
9a4fd6a2 | 514 | uuid_io(c, REQ_OP_WRITE, &k.key, &cl); |
cafe5635 KO |
515 | closure_sync(&cl); |
516 | ||
7a55948d | 517 | /* Only one bucket used for uuid write */ |
7a55948d SW |
518 | atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written); |
519 | ||
cafe5635 | 520 | bkey_copy(&c->uuid_bucket, &k.key); |
3a3b6a4e | 521 | bkey_put(c, &k.key); |
cafe5635 KO |
522 | return 0; |
523 | } | |
524 | ||
525 | int bch_uuid_write(struct cache_set *c) | |
526 | { | |
527 | int ret = __uuid_write(c); | |
528 | ||
529 | if (!ret) | |
530 | bch_journal_meta(c, NULL); | |
531 | ||
532 | return ret; | |
533 | } | |
534 | ||
535 | static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid) | |
536 | { | |
537 | struct uuid_entry *u; | |
538 | ||
539 | for (u = c->uuids; | |
540 | u < c->uuids + c->nr_uuids; u++) | |
541 | if (!memcmp(u->uuid, uuid, 16)) | |
542 | return u; | |
543 | ||
544 | return NULL; | |
545 | } | |
546 | ||
547 | static struct uuid_entry *uuid_find_empty(struct cache_set *c) | |
548 | { | |
549 | static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"; | |
1fae7cf0 | 550 | |
cafe5635 KO |
551 | return uuid_find(c, zero_uuid); |
552 | } | |
553 | ||
554 | /* | |
555 | * Bucket priorities/gens: | |
556 | * | |
557 | * For each bucket, we store on disk its | |
3be11dba CL |
558 | * 8 bit gen |
559 | * 16 bit priority | |
cafe5635 KO |
560 | * |
561 | * See alloc.c for an explanation of the gen. The priority is used to implement | |
562 | * lru (and in the future other) cache replacement policies; for most purposes | |
563 | * it's just an opaque integer. | |
564 | * | |
565 | * The gens and the priorities don't have a whole lot to do with each other, and | |
566 | * it's actually the gens that must be written out at specific times - it's no | |
567 | * big deal if the priorities don't get written, if we lose them we just reuse | |
568 | * buckets in suboptimal order. | |
569 | * | |
570 | * On disk they're stored in a packed array, and in as many buckets are required | |
571 | * to fit them all. The buckets we use to store them form a list; the journal | |
572 | * header points to the first bucket, the first bucket points to the second | |
573 | * bucket, et cetera. | |
574 | * | |
575 | * This code is used by the allocation code; periodically (whenever it runs out | |
576 | * of buckets to allocate from) the allocation code will invalidate some | |
577 | * buckets, but it can't use those buckets until their new gens are safely on | |
578 | * disk. | |
579 | */ | |
580 | ||
4246a0b6 | 581 | static void prio_endio(struct bio *bio) |
cafe5635 KO |
582 | { |
583 | struct cache *ca = bio->bi_private; | |
584 | ||
4e4cbee9 | 585 | cache_set_err_on(bio->bi_status, ca->set, "accessing priorities"); |
cafe5635 KO |
586 | bch_bbio_free(bio, ca->set); |
587 | closure_put(&ca->prio); | |
588 | } | |
589 | ||
552eee3b | 590 | static void prio_io(struct cache *ca, uint64_t bucket, blk_opf_t opf) |
cafe5635 KO |
591 | { |
592 | struct closure *cl = &ca->prio; | |
593 | struct bio *bio = bch_bbio_alloc(ca->set); | |
594 | ||
595 | closure_init_stack(cl); | |
596 | ||
4f024f37 | 597 | bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size; |
74d46992 | 598 | bio_set_dev(bio, ca->bdev); |
c954ac8d | 599 | bio->bi_iter.bi_size = meta_bucket_bytes(&ca->sb); |
cafe5635 KO |
600 | |
601 | bio->bi_end_io = prio_endio; | |
602 | bio->bi_private = ca; | |
552eee3b | 603 | bio->bi_opf = opf | REQ_SYNC | REQ_META; |
169ef1cf | 604 | bch_bio_map(bio, ca->disk_buckets); |
cafe5635 | 605 | |
771f393e | 606 | closure_bio_submit(ca->set, bio, &ca->prio); |
cafe5635 KO |
607 | closure_sync(cl); |
608 | } | |
609 | ||
84c529ae | 610 | int bch_prio_write(struct cache *ca, bool wait) |
cafe5635 KO |
611 | { |
612 | int i; | |
613 | struct bucket *b; | |
614 | struct closure cl; | |
615 | ||
46f5aa88 | 616 | pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu\n", |
84c529ae AR |
617 | fifo_used(&ca->free[RESERVE_PRIO]), |
618 | fifo_used(&ca->free[RESERVE_NONE]), | |
619 | fifo_used(&ca->free_inc)); | |
620 | ||
621 | /* | |
622 | * Pre-check if there are enough free buckets. In the non-blocking | |
623 | * scenario it's better to fail early rather than starting to allocate | |
624 | * buckets and do a cleanup later in case of failure. | |
625 | */ | |
626 | if (!wait) { | |
627 | size_t avail = fifo_used(&ca->free[RESERVE_PRIO]) + | |
628 | fifo_used(&ca->free[RESERVE_NONE]); | |
629 | if (prio_buckets(ca) > avail) | |
630 | return -ENOMEM; | |
631 | } | |
632 | ||
cafe5635 KO |
633 | closure_init_stack(&cl); |
634 | ||
635 | lockdep_assert_held(&ca->set->bucket_lock); | |
636 | ||
cafe5635 KO |
637 | ca->disk_buckets->seq++; |
638 | ||
639 | atomic_long_add(ca->sb.bucket_size * prio_buckets(ca), | |
640 | &ca->meta_sectors_written); | |
641 | ||
cafe5635 KO |
642 | for (i = prio_buckets(ca) - 1; i >= 0; --i) { |
643 | long bucket; | |
644 | struct prio_set *p = ca->disk_buckets; | |
b1a67b0f KO |
645 | struct bucket_disk *d = p->data; |
646 | struct bucket_disk *end = d + prios_per_bucket(ca); | |
cafe5635 KO |
647 | |
648 | for (b = ca->buckets + i * prios_per_bucket(ca); | |
649 | b < ca->buckets + ca->sb.nbuckets && d < end; | |
650 | b++, d++) { | |
651 | d->prio = cpu_to_le16(b->prio); | |
652 | d->gen = b->gen; | |
653 | } | |
654 | ||
655 | p->next_bucket = ca->prio_buckets[i + 1]; | |
81ab4190 | 656 | p->magic = pset_magic(&ca->sb); |
c954ac8d | 657 | p->csum = bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8); |
cafe5635 | 658 | |
84c529ae | 659 | bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait); |
cafe5635 KO |
660 | BUG_ON(bucket == -1); |
661 | ||
662 | mutex_unlock(&ca->set->bucket_lock); | |
552eee3b | 663 | prio_io(ca, bucket, REQ_OP_WRITE); |
cafe5635 KO |
664 | mutex_lock(&ca->set->bucket_lock); |
665 | ||
666 | ca->prio_buckets[i] = bucket; | |
667 | atomic_dec_bug(&ca->buckets[bucket].pin); | |
668 | } | |
669 | ||
670 | mutex_unlock(&ca->set->bucket_lock); | |
671 | ||
672 | bch_journal_meta(ca->set, &cl); | |
673 | closure_sync(&cl); | |
674 | ||
675 | mutex_lock(&ca->set->bucket_lock); | |
676 | ||
cafe5635 KO |
677 | /* |
678 | * Don't want the old priorities to get garbage collected until after we | |
679 | * finish writing the new ones, and they're journalled | |
680 | */ | |
2531d9ee KO |
681 | for (i = 0; i < prio_buckets(ca); i++) { |
682 | if (ca->prio_last_buckets[i]) | |
683 | __bch_bucket_free(ca, | |
684 | &ca->buckets[ca->prio_last_buckets[i]]); | |
685 | ||
cafe5635 | 686 | ca->prio_last_buckets[i] = ca->prio_buckets[i]; |
2531d9ee | 687 | } |
84c529ae | 688 | return 0; |
cafe5635 KO |
689 | } |
690 | ||
49d08d59 | 691 | static int prio_read(struct cache *ca, uint64_t bucket) |
cafe5635 KO |
692 | { |
693 | struct prio_set *p = ca->disk_buckets; | |
694 | struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d; | |
695 | struct bucket *b; | |
6f10f7d1 | 696 | unsigned int bucket_nr = 0; |
49d08d59 | 697 | int ret = -EIO; |
cafe5635 KO |
698 | |
699 | for (b = ca->buckets; | |
700 | b < ca->buckets + ca->sb.nbuckets; | |
701 | b++, d++) { | |
702 | if (d == end) { | |
703 | ca->prio_buckets[bucket_nr] = bucket; | |
704 | ca->prio_last_buckets[bucket_nr] = bucket; | |
705 | bucket_nr++; | |
706 | ||
552eee3b | 707 | prio_io(ca, bucket, REQ_OP_READ); |
cafe5635 | 708 | |
b0d30981 | 709 | if (p->csum != |
c954ac8d | 710 | bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8)) { |
46f5aa88 | 711 | pr_warn("bad csum reading priorities\n"); |
49d08d59 CL |
712 | goto out; |
713 | } | |
cafe5635 | 714 | |
49d08d59 | 715 | if (p->magic != pset_magic(&ca->sb)) { |
46f5aa88 | 716 | pr_warn("bad magic reading priorities\n"); |
49d08d59 CL |
717 | goto out; |
718 | } | |
cafe5635 KO |
719 | |
720 | bucket = p->next_bucket; | |
721 | d = p->data; | |
722 | } | |
723 | ||
724 | b->prio = le16_to_cpu(d->prio); | |
3a2fd9d5 | 725 | b->gen = b->last_gc = d->gen; |
cafe5635 | 726 | } |
49d08d59 CL |
727 | |
728 | ret = 0; | |
729 | out: | |
730 | return ret; | |
cafe5635 KO |
731 | } |
732 | ||
733 | /* Bcache device */ | |
734 | ||
05bdb996 | 735 | static int open_dev(struct gendisk *disk, blk_mode_t mode) |
cafe5635 | 736 | { |
d32e2bf8 | 737 | struct bcache_device *d = disk->private_data; |
1fae7cf0 | 738 | |
c4d951dd | 739 | if (test_bit(BCACHE_DEV_CLOSING, &d->flags)) |
cafe5635 KO |
740 | return -ENXIO; |
741 | ||
742 | closure_get(&d->cl); | |
743 | return 0; | |
744 | } | |
745 | ||
ae220766 | 746 | static void release_dev(struct gendisk *b) |
cafe5635 KO |
747 | { |
748 | struct bcache_device *d = b->private_data; | |
1fae7cf0 | 749 | |
cafe5635 | 750 | closure_put(&d->cl); |
cafe5635 KO |
751 | } |
752 | ||
05bdb996 | 753 | static int ioctl_dev(struct block_device *b, blk_mode_t mode, |
cafe5635 KO |
754 | unsigned int cmd, unsigned long arg) |
755 | { | |
756 | struct bcache_device *d = b->bd_disk->private_data; | |
0f0709e6 | 757 | |
cafe5635 KO |
758 | return d->ioctl(d, mode, cmd, arg); |
759 | } | |
760 | ||
c62b37d9 CH |
761 | static const struct block_device_operations bcache_cached_ops = { |
762 | .submit_bio = cached_dev_submit_bio, | |
763 | .open = open_dev, | |
764 | .release = release_dev, | |
765 | .ioctl = ioctl_dev, | |
766 | .owner = THIS_MODULE, | |
767 | }; | |
768 | ||
769 | static const struct block_device_operations bcache_flash_ops = { | |
770 | .submit_bio = flash_dev_submit_bio, | |
cafe5635 KO |
771 | .open = open_dev, |
772 | .release = release_dev, | |
773 | .ioctl = ioctl_dev, | |
774 | .owner = THIS_MODULE, | |
775 | }; | |
776 | ||
777 | void bcache_device_stop(struct bcache_device *d) | |
778 | { | |
c4d951dd | 779 | if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags)) |
63d63b51 CL |
780 | /* |
781 | * closure_fn set to | |
782 | * - cached device: cached_dev_flush() | |
783 | * - flash dev: flash_dev_flush() | |
784 | */ | |
cafe5635 KO |
785 | closure_queue(&d->cl); |
786 | } | |
787 | ||
ee668506 KO |
788 | static void bcache_device_unlink(struct bcache_device *d) |
789 | { | |
c4d951dd | 790 | lockdep_assert_held(&bch_register_lock); |
ee668506 | 791 | |
c4d951dd | 792 | if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) { |
08fdb2cd | 793 | struct cache *ca = d->c->cache; |
ee668506 | 794 | |
c4d951dd KO |
795 | sysfs_remove_link(&d->c->kobj, d->name); |
796 | sysfs_remove_link(&d->kobj, "cache"); | |
797 | ||
08fdb2cd | 798 | bd_unlink_disk_holder(ca->bdev, d->disk); |
c4d951dd | 799 | } |
ee668506 KO |
800 | } |
801 | ||
802 | static void bcache_device_link(struct bcache_device *d, struct cache_set *c, | |
803 | const char *name) | |
804 | { | |
08fdb2cd | 805 | struct cache *ca = c->cache; |
4b6efb4b | 806 | int ret; |
ee668506 | 807 | |
08fdb2cd | 808 | bd_link_disk_holder(ca->bdev, d->disk); |
ee668506 KO |
809 | |
810 | snprintf(d->name, BCACHEDEVNAME_SIZE, | |
811 | "%s%u", name, d->id); | |
812 | ||
4b6efb4b CL |
813 | ret = sysfs_create_link(&d->kobj, &c->kobj, "cache"); |
814 | if (ret < 0) | |
46f5aa88 | 815 | pr_err("Couldn't create device -> cache set symlink\n"); |
4b6efb4b CL |
816 | |
817 | ret = sysfs_create_link(&c->kobj, &d->kobj, d->name); | |
818 | if (ret < 0) | |
46f5aa88 | 819 | pr_err("Couldn't create cache set -> device symlink\n"); |
fecaee6f ZL |
820 | |
821 | clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags); | |
ee668506 KO |
822 | } |
823 | ||
cafe5635 KO |
824 | static void bcache_device_detach(struct bcache_device *d) |
825 | { | |
826 | lockdep_assert_held(&bch_register_lock); | |
827 | ||
ea8c5356 CL |
828 | atomic_dec(&d->c->attached_dev_nr); |
829 | ||
c4d951dd | 830 | if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) { |
cafe5635 KO |
831 | struct uuid_entry *u = d->c->uuids + d->id; |
832 | ||
833 | SET_UUID_FLASH_ONLY(u, 0); | |
834 | memcpy(u->uuid, invalid_uuid, 16); | |
75cbb3f1 | 835 | u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 | 836 | bch_uuid_write(d->c); |
cafe5635 KO |
837 | } |
838 | ||
c4d951dd | 839 | bcache_device_unlink(d); |
ee668506 | 840 | |
cafe5635 KO |
841 | d->c->devices[d->id] = NULL; |
842 | closure_put(&d->c->caching); | |
843 | d->c = NULL; | |
844 | } | |
845 | ||
846 | static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, | |
6f10f7d1 | 847 | unsigned int id) |
cafe5635 | 848 | { |
cafe5635 KO |
849 | d->id = id; |
850 | d->c = c; | |
851 | c->devices[id] = d; | |
852 | ||
2831231d CL |
853 | if (id >= c->devices_max_used) |
854 | c->devices_max_used = id + 1; | |
855 | ||
cafe5635 KO |
856 | closure_get(&c->caching); |
857 | } | |
858 | ||
1dbe32ad CL |
859 | static inline int first_minor_to_idx(int first_minor) |
860 | { | |
861 | return (first_minor/BCACHE_MINORS); | |
862 | } | |
863 | ||
864 | static inline int idx_to_first_minor(int idx) | |
865 | { | |
866 | return (idx * BCACHE_MINORS); | |
867 | } | |
868 | ||
cafe5635 KO |
869 | static void bcache_device_free(struct bcache_device *d) |
870 | { | |
2d886951 CL |
871 | struct gendisk *disk = d->disk; |
872 | ||
cafe5635 KO |
873 | lockdep_assert_held(&bch_register_lock); |
874 | ||
2d886951 | 875 | if (disk) |
46f5aa88 | 876 | pr_info("%s stopped\n", disk->disk_name); |
2d886951 | 877 | else |
46f5aa88 | 878 | pr_err("bcache device (NULL gendisk) stopped\n"); |
cafe5635 KO |
879 | |
880 | if (d->c) | |
881 | bcache_device_detach(d); | |
2d886951 CL |
882 | |
883 | if (disk) { | |
2abd9a19 CJ |
884 | ida_free(&bcache_device_idx, |
885 | first_minor_to_idx(disk->first_minor)); | |
8b9ab626 | 886 | put_disk(disk); |
28935ab5 | 887 | } |
cafe5635 | 888 | |
d19936a2 | 889 | bioset_exit(&d->bio_split); |
958b4338 PE |
890 | kvfree(d->full_dirty_stripes); |
891 | kvfree(d->stripe_sectors_dirty); | |
cafe5635 KO |
892 | |
893 | closure_debug_destroy(&d->cl); | |
894 | } | |
895 | ||
6f10f7d1 | 896 | static int bcache_device_init(struct bcache_device *d, unsigned int block_size, |
c62b37d9 CH |
897 | sector_t sectors, struct block_device *cached_bdev, |
898 | const struct block_device_operations *ops) | |
cafe5635 KO |
899 | { |
900 | struct request_queue *q; | |
5f2b18ec BVA |
901 | const size_t max_stripes = min_t(size_t, INT_MAX, |
902 | SIZE_MAX / sizeof(atomic_t)); | |
b3f0846e CH |
903 | struct queue_limits lim = { |
904 | .max_hw_sectors = UINT_MAX, | |
905 | .max_sectors = UINT_MAX, | |
906 | .max_segment_size = UINT_MAX, | |
907 | .max_segments = BIO_MAX_VECS, | |
908 | .max_hw_discard_sectors = UINT_MAX, | |
909 | .io_min = block_size, | |
910 | .logical_block_size = block_size, | |
911 | .physical_block_size = block_size, | |
912 | }; | |
65f0f017 | 913 | uint64_t n; |
1dbe32ad | 914 | int idx; |
279afbad | 915 | |
34a2cf3f CH |
916 | if (cached_bdev) { |
917 | d->stripe_size = bdev_io_opt(cached_bdev) >> SECTOR_SHIFT; | |
918 | lim.io_opt = umax(block_size, bdev_io_opt(cached_bdev)); | |
919 | } | |
2d679fc7 KO |
920 | if (!d->stripe_size) |
921 | d->stripe_size = 1 << 31; | |
baf8fb7e CL |
922 | else if (d->stripe_size < BCH_MIN_STRIPE_SZ) |
923 | d->stripe_size = roundup(BCH_MIN_STRIPE_SZ, d->stripe_size); | |
279afbad | 924 | |
65f0f017 CL |
925 | n = DIV_ROUND_UP_ULL(sectors, d->stripe_size); |
926 | if (!n || n > max_stripes) { | |
927 | pr_err("nr_stripes too large or invalid: %llu (start sector beyond end of disk?)\n", | |
928 | n); | |
279afbad | 929 | return -ENOMEM; |
48a915a8 | 930 | } |
65f0f017 | 931 | d->nr_stripes = n; |
279afbad KO |
932 | |
933 | n = d->nr_stripes * sizeof(atomic_t); | |
bc4e54f6 | 934 | d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL); |
279afbad KO |
935 | if (!d->stripe_sectors_dirty) |
936 | return -ENOMEM; | |
cafe5635 | 937 | |
48a915a8 | 938 | n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long); |
bc4e54f6 | 939 | d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL); |
48a915a8 | 940 | if (!d->full_dirty_stripes) |
224b0683 | 941 | goto out_free_stripe_sectors_dirty; |
48a915a8 | 942 | |
2abd9a19 CJ |
943 | idx = ida_alloc_max(&bcache_device_idx, BCACHE_DEVICE_IDX_MAX - 1, |
944 | GFP_KERNEL); | |
1dbe32ad | 945 | if (idx < 0) |
224b0683 | 946 | goto out_free_full_dirty_stripes; |
b8c0d911 | 947 | |
d19936a2 | 948 | if (bioset_init(&d->bio_split, 4, offsetof(struct bbio, bio), |
9b4e9f5a | 949 | BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER)) |
224b0683 | 950 | goto out_ida_remove; |
9b4e9f5a | 951 | |
b3f0846e CH |
952 | if (lim.logical_block_size > PAGE_SIZE && cached_bdev) { |
953 | /* | |
954 | * This should only happen with BCACHE_SB_VERSION_BDEV. | |
955 | * Block/page size is checked for BCACHE_SB_VERSION_CDEV. | |
956 | */ | |
957 | pr_info("bcache%i: sb/logical block size (%u) greater than page size (%lu) falling back to device logical block size (%u)\n", | |
958 | idx, lim.logical_block_size, | |
959 | PAGE_SIZE, bdev_logical_block_size(cached_bdev)); | |
960 | ||
961 | /* This also adjusts physical block size/min io size if needed */ | |
962 | lim.logical_block_size = bdev_logical_block_size(cached_bdev); | |
963 | } | |
964 | ||
965 | d->disk = blk_alloc_disk(&lim, NUMA_NO_NODE); | |
74fa8f9c | 966 | if (IS_ERR(d->disk)) |
224b0683 | 967 | goto out_bioset_exit; |
cafe5635 | 968 | |
279afbad | 969 | set_capacity(d->disk, sectors); |
1dbe32ad | 970 | snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx); |
cafe5635 KO |
971 | |
972 | d->disk->major = bcache_major; | |
1dbe32ad | 973 | d->disk->first_minor = idx_to_first_minor(idx); |
bc70852f | 974 | d->disk->minors = BCACHE_MINORS; |
c62b37d9 | 975 | d->disk->fops = ops; |
cafe5635 KO |
976 | d->disk->private_data = d; |
977 | ||
bc70852f | 978 | q = d->disk->queue; |
dcacbc12 | 979 | |
44e1ebe2 | 980 | blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue); |
cafe5635 | 981 | |
84b4ff9e | 982 | blk_queue_write_cache(q, true, true); |
54d12f2b | 983 | |
cafe5635 | 984 | return 0; |
9b4e9f5a | 985 | |
224b0683 CH |
986 | out_bioset_exit: |
987 | bioset_exit(&d->bio_split); | |
988 | out_ida_remove: | |
2abd9a19 | 989 | ida_free(&bcache_device_idx, idx); |
224b0683 CH |
990 | out_free_full_dirty_stripes: |
991 | kvfree(d->full_dirty_stripes); | |
992 | out_free_stripe_sectors_dirty: | |
993 | kvfree(d->stripe_sectors_dirty); | |
9b4e9f5a FS |
994 | return -ENOMEM; |
995 | ||
cafe5635 KO |
996 | } |
997 | ||
998 | /* Cached device */ | |
999 | ||
1000 | static void calc_cached_dev_sectors(struct cache_set *c) | |
1001 | { | |
1002 | uint64_t sectors = 0; | |
1003 | struct cached_dev *dc; | |
1004 | ||
1005 | list_for_each_entry(dc, &c->cached_devs, list) | |
cda25b82 | 1006 | sectors += bdev_nr_sectors(dc->bdev); |
cafe5635 KO |
1007 | |
1008 | c->cached_dev_sectors = sectors; | |
1009 | } | |
1010 | ||
0f0709e6 CL |
1011 | #define BACKING_DEV_OFFLINE_TIMEOUT 5 |
1012 | static int cached_dev_status_update(void *arg) | |
1013 | { | |
1014 | struct cached_dev *dc = arg; | |
1015 | struct request_queue *q; | |
1016 | ||
1017 | /* | |
1018 | * If this delayed worker is stopping outside, directly quit here. | |
1019 | * dc->io_disable might be set via sysfs interface, so check it | |
1020 | * here too. | |
1021 | */ | |
1022 | while (!kthread_should_stop() && !dc->io_disable) { | |
1023 | q = bdev_get_queue(dc->bdev); | |
1024 | if (blk_queue_dying(q)) | |
1025 | dc->offline_seconds++; | |
1026 | else | |
1027 | dc->offline_seconds = 0; | |
1028 | ||
1029 | if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) { | |
0f5cd781 CH |
1030 | pr_err("%pg: device offline for %d seconds\n", |
1031 | dc->bdev, | |
0f0709e6 | 1032 | BACKING_DEV_OFFLINE_TIMEOUT); |
46f5aa88 JP |
1033 | pr_err("%s: disable I/O request due to backing device offline\n", |
1034 | dc->disk.name); | |
0f0709e6 CL |
1035 | dc->io_disable = true; |
1036 | /* let others know earlier that io_disable is true */ | |
1037 | smp_mb(); | |
1038 | bcache_device_stop(&dc->disk); | |
1039 | break; | |
1040 | } | |
1041 | schedule_timeout_interruptible(HZ); | |
1042 | } | |
1043 | ||
1044 | wait_for_kthread_stop(); | |
1045 | return 0; | |
1046 | } | |
1047 | ||
1048 | ||
0b13efec | 1049 | int bch_cached_dev_run(struct cached_dev *dc) |
cafe5635 | 1050 | { |
13e1db65 | 1051 | int ret = 0; |
cafe5635 | 1052 | struct bcache_device *d = &dc->disk; |
792732d9 | 1053 | char *buf = kmemdup_nul(dc->sb.label, SB_LABEL_SIZE, GFP_KERNEL); |
a25c32be GP |
1054 | char *env[] = { |
1055 | "DRIVER=bcache", | |
1056 | kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), | |
792732d9 | 1057 | kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf ? : ""), |
ab9e1400 | 1058 | NULL, |
a25c32be | 1059 | }; |
cafe5635 | 1060 | |
e0faa3d7 | 1061 | if (dc->io_disable) { |
0f5cd781 | 1062 | pr_err("I/O disabled on cached dev %pg\n", dc->bdev); |
13e1db65 ZL |
1063 | ret = -EIO; |
1064 | goto out; | |
e0faa3d7 | 1065 | } |
0b13efec | 1066 | |
4d4d8573 | 1067 | if (atomic_xchg(&dc->running, 1)) { |
0f5cd781 | 1068 | pr_info("cached dev %pg is running already\n", dc->bdev); |
13e1db65 ZL |
1069 | ret = -EBUSY; |
1070 | goto out; | |
4d4d8573 | 1071 | } |
cafe5635 KO |
1072 | |
1073 | if (!d->c && | |
1074 | BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { | |
1075 | struct closure cl; | |
1fae7cf0 | 1076 | |
cafe5635 KO |
1077 | closure_init_stack(&cl); |
1078 | ||
1079 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); | |
1080 | bch_write_bdev_super(dc, &cl); | |
1081 | closure_sync(&cl); | |
1082 | } | |
1083 | ||
2961c3bb LC |
1084 | ret = add_disk(d->disk); |
1085 | if (ret) | |
1086 | goto out; | |
ee668506 | 1087 | bd_link_disk_holder(dc->bdev, dc->disk.disk); |
3be11dba CL |
1088 | /* |
1089 | * won't show up in the uevent file, use udevadm monitor -e instead | |
1090 | * only class / kset properties are persistent | |
1091 | */ | |
cafe5635 | 1092 | kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); |
a25c32be | 1093 | |
cafe5635 | 1094 | if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || |
0b13efec CL |
1095 | sysfs_create_link(&disk_to_dev(d->disk)->kobj, |
1096 | &d->kobj, "bcache")) { | |
46f5aa88 | 1097 | pr_err("Couldn't create bcache dev <-> disk sysfs symlinks\n"); |
13e1db65 ZL |
1098 | ret = -ENOMEM; |
1099 | goto out; | |
0b13efec | 1100 | } |
0f0709e6 CL |
1101 | |
1102 | dc->status_update_thread = kthread_run(cached_dev_status_update, | |
1103 | dc, "bcache_status_update"); | |
1104 | if (IS_ERR(dc->status_update_thread)) { | |
46f5aa88 | 1105 | pr_warn("failed to create bcache_status_update kthread, continue to run without monitoring backing device status\n"); |
0f0709e6 | 1106 | } |
0b13efec | 1107 | |
13e1db65 ZL |
1108 | out: |
1109 | kfree(env[1]); | |
1110 | kfree(env[2]); | |
1111 | kfree(buf); | |
1112 | return ret; | |
cafe5635 KO |
1113 | } |
1114 | ||
3fd47bfe CL |
1115 | /* |
1116 | * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed | |
1117 | * work dc->writeback_rate_update is running. Wait until the routine | |
1118 | * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to | |
1119 | * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out | |
1120 | * seconds, give up waiting here and continue to cancel it too. | |
1121 | */ | |
1122 | static void cancel_writeback_rate_update_dwork(struct cached_dev *dc) | |
1123 | { | |
1124 | int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ; | |
1125 | ||
1126 | do { | |
1127 | if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING, | |
1128 | &dc->disk.flags)) | |
1129 | break; | |
1130 | time_out--; | |
1131 | schedule_timeout_interruptible(1); | |
1132 | } while (time_out > 0); | |
1133 | ||
1134 | if (time_out == 0) | |
46f5aa88 | 1135 | pr_warn("give up waiting for dc->writeback_write_update to quit\n"); |
3fd47bfe CL |
1136 | |
1137 | cancel_delayed_work_sync(&dc->writeback_rate_update); | |
1138 | } | |
1139 | ||
cafe5635 KO |
1140 | static void cached_dev_detach_finish(struct work_struct *w) |
1141 | { | |
1142 | struct cached_dev *dc = container_of(w, struct cached_dev, detach); | |
aa97f6cd | 1143 | struct cache_set *c = dc->disk.c; |
cafe5635 | 1144 | |
c4d951dd | 1145 | BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)); |
3b304d24 | 1146 | BUG_ON(refcount_read(&dc->count)); |
cafe5635 | 1147 | |
cafe5635 | 1148 | |
3fd47bfe CL |
1149 | if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) |
1150 | cancel_writeback_rate_update_dwork(dc); | |
1151 | ||
8d29c442 TJ |
1152 | if (!IS_ERR_OR_NULL(dc->writeback_thread)) { |
1153 | kthread_stop(dc->writeback_thread); | |
1154 | dc->writeback_thread = NULL; | |
1155 | } | |
1156 | ||
97ba3b81 CL |
1157 | mutex_lock(&bch_register_lock); |
1158 | ||
cafe5635 KO |
1159 | bcache_device_detach(&dc->disk); |
1160 | list_move(&dc->list, &uncached_devices); | |
aa97f6cd | 1161 | calc_cached_dev_sectors(c); |
cafe5635 | 1162 | |
c4d951dd | 1163 | clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags); |
5b1016e6 | 1164 | clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags); |
c4d951dd | 1165 | |
cafe5635 KO |
1166 | mutex_unlock(&bch_register_lock); |
1167 | ||
0f5cd781 | 1168 | pr_info("Caching disabled for %pg\n", dc->bdev); |
cafe5635 KO |
1169 | |
1170 | /* Drop ref we took in cached_dev_detach() */ | |
1171 | closure_put(&dc->disk.cl); | |
1172 | } | |
1173 | ||
1174 | void bch_cached_dev_detach(struct cached_dev *dc) | |
1175 | { | |
1176 | lockdep_assert_held(&bch_register_lock); | |
1177 | ||
c4d951dd | 1178 | if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) |
cafe5635 KO |
1179 | return; |
1180 | ||
c4d951dd | 1181 | if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) |
cafe5635 KO |
1182 | return; |
1183 | ||
1184 | /* | |
1185 | * Block the device from being closed and freed until we're finished | |
1186 | * detaching | |
1187 | */ | |
1188 | closure_get(&dc->disk.cl); | |
1189 | ||
1190 | bch_writeback_queue(dc); | |
3fd47bfe | 1191 | |
cafe5635 KO |
1192 | cached_dev_put(dc); |
1193 | } | |
1194 | ||
73ac105b TJ |
1195 | int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c, |
1196 | uint8_t *set_uuid) | |
cafe5635 | 1197 | { |
75cbb3f1 | 1198 | uint32_t rtime = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 | 1199 | struct uuid_entry *u; |
86755b7a | 1200 | struct cached_dev *exist_dc, *t; |
0b13efec | 1201 | int ret = 0; |
cafe5635 | 1202 | |
1132e56e CL |
1203 | if ((set_uuid && memcmp(set_uuid, c->set_uuid, 16)) || |
1204 | (!set_uuid && memcmp(dc->sb.set_uuid, c->set_uuid, 16))) | |
cafe5635 KO |
1205 | return -ENOENT; |
1206 | ||
1207 | if (dc->disk.c) { | |
0f5cd781 | 1208 | pr_err("Can't attach %pg: already attached\n", dc->bdev); |
cafe5635 KO |
1209 | return -EINVAL; |
1210 | } | |
1211 | ||
1212 | if (test_bit(CACHE_SET_STOPPING, &c->flags)) { | |
0f5cd781 | 1213 | pr_err("Can't attach %pg: shutting down\n", dc->bdev); |
cafe5635 KO |
1214 | return -EINVAL; |
1215 | } | |
1216 | ||
4a784266 | 1217 | if (dc->sb.block_size < c->cache->sb.block_size) { |
cafe5635 | 1218 | /* Will die */ |
0f5cd781 CH |
1219 | pr_err("Couldn't attach %pg: block size less than set's block size\n", |
1220 | dc->bdev); | |
cafe5635 KO |
1221 | return -EINVAL; |
1222 | } | |
1223 | ||
86755b7a ML |
1224 | /* Check whether already attached */ |
1225 | list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) { | |
1226 | if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) { | |
0f5cd781 CH |
1227 | pr_err("Tried to attach %pg but duplicate UUID already attached\n", |
1228 | dc->bdev); | |
86755b7a ML |
1229 | |
1230 | return -EINVAL; | |
1231 | } | |
1232 | } | |
1233 | ||
cafe5635 KO |
1234 | u = uuid_find(c, dc->sb.uuid); |
1235 | ||
1236 | if (u && | |
1237 | (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || | |
1238 | BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { | |
1239 | memcpy(u->uuid, invalid_uuid, 16); | |
75cbb3f1 | 1240 | u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 KO |
1241 | u = NULL; |
1242 | } | |
1243 | ||
1244 | if (!u) { | |
1245 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { | |
0f5cd781 | 1246 | pr_err("Couldn't find uuid for %pg in set\n", dc->bdev); |
cafe5635 KO |
1247 | return -ENOENT; |
1248 | } | |
1249 | ||
1250 | u = uuid_find_empty(c); | |
1251 | if (!u) { | |
0f5cd781 | 1252 | pr_err("Not caching %pg, no room for UUID\n", dc->bdev); |
cafe5635 KO |
1253 | return -EINVAL; |
1254 | } | |
1255 | } | |
1256 | ||
3be11dba CL |
1257 | /* |
1258 | * Deadlocks since we're called via sysfs... | |
1259 | * sysfs_remove_file(&dc->kobj, &sysfs_attach); | |
cafe5635 KO |
1260 | */ |
1261 | ||
169ef1cf | 1262 | if (bch_is_zero(u->uuid, 16)) { |
cafe5635 | 1263 | struct closure cl; |
1fae7cf0 | 1264 | |
cafe5635 KO |
1265 | closure_init_stack(&cl); |
1266 | ||
1267 | memcpy(u->uuid, dc->sb.uuid, 16); | |
1268 | memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); | |
1269 | u->first_reg = u->last_reg = rtime; | |
1270 | bch_uuid_write(c); | |
1271 | ||
1132e56e | 1272 | memcpy(dc->sb.set_uuid, c->set_uuid, 16); |
cafe5635 KO |
1273 | SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); |
1274 | ||
1275 | bch_write_bdev_super(dc, &cl); | |
1276 | closure_sync(&cl); | |
1277 | } else { | |
1278 | u->last_reg = rtime; | |
1279 | bch_uuid_write(c); | |
1280 | } | |
1281 | ||
1282 | bcache_device_attach(&dc->disk, c, u - c->uuids); | |
cafe5635 KO |
1283 | list_move(&dc->list, &c->cached_devs); |
1284 | calc_cached_dev_sectors(c); | |
1285 | ||
cafe5635 KO |
1286 | /* |
1287 | * dc->c must be set before dc->count != 0 - paired with the mb in | |
1288 | * cached_dev_get() | |
1289 | */ | |
eb2b3d03 | 1290 | smp_wmb(); |
3b304d24 | 1291 | refcount_set(&dc->count, 1); |
cafe5635 | 1292 | |
07cc6ef8 EW |
1293 | /* Block writeback thread, but spawn it */ |
1294 | down_write(&dc->writeback_lock); | |
1295 | if (bch_cached_dev_writeback_start(dc)) { | |
1296 | up_write(&dc->writeback_lock); | |
46f5aa88 | 1297 | pr_err("Couldn't start writeback facilities for %s\n", |
633bb2ce | 1298 | dc->disk.disk->disk_name); |
9e5c3535 | 1299 | return -ENOMEM; |
07cc6ef8 | 1300 | } |
9e5c3535 | 1301 | |
cafe5635 KO |
1302 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { |
1303 | atomic_set(&dc->has_dirty, 1); | |
cafe5635 KO |
1304 | bch_writeback_queue(dc); |
1305 | } | |
1306 | ||
2e17a262 TJ |
1307 | bch_sectors_dirty_init(&dc->disk); |
1308 | ||
0b13efec CL |
1309 | ret = bch_cached_dev_run(dc); |
1310 | if (ret && (ret != -EBUSY)) { | |
1311 | up_write(&dc->writeback_lock); | |
5c2a634c CL |
1312 | /* |
1313 | * bch_register_lock is held, bcache_device_stop() is not | |
1314 | * able to be directly called. The kthread and kworker | |
1315 | * created previously in bch_cached_dev_writeback_start() | |
1316 | * have to be stopped manually here. | |
1317 | */ | |
1318 | kthread_stop(dc->writeback_thread); | |
1319 | cancel_writeback_rate_update_dwork(dc); | |
0f5cd781 | 1320 | pr_err("Couldn't run cached device %pg\n", dc->bdev); |
0b13efec CL |
1321 | return ret; |
1322 | } | |
1323 | ||
ee668506 | 1324 | bcache_device_link(&dc->disk, c, "bdev"); |
ea8c5356 | 1325 | atomic_inc(&c->attached_dev_nr); |
cafe5635 | 1326 | |
5342fd42 CL |
1327 | if (bch_has_feature_obso_large_bucket(&(c->cache->sb))) { |
1328 | pr_err("The obsoleted large bucket layout is unsupported, set the bcache device into read-only\n"); | |
1329 | pr_err("Please update to the latest bcache-tools to create the cache device\n"); | |
1330 | set_disk_ro(dc->disk.disk, 1); | |
1331 | } | |
1332 | ||
07cc6ef8 EW |
1333 | /* Allow the writeback thread to proceed */ |
1334 | up_write(&dc->writeback_lock); | |
1335 | ||
0f5cd781 CH |
1336 | pr_info("Caching %pg as %s on set %pU\n", |
1337 | dc->bdev, | |
6e916a7e | 1338 | dc->disk.disk->disk_name, |
1132e56e | 1339 | dc->disk.c->set_uuid); |
cafe5635 KO |
1340 | return 0; |
1341 | } | |
1342 | ||
2d17456e | 1343 | /* when dc->disk.kobj released */ |
cafe5635 KO |
1344 | void bch_cached_dev_release(struct kobject *kobj) |
1345 | { | |
1346 | struct cached_dev *dc = container_of(kobj, struct cached_dev, | |
1347 | disk.kobj); | |
1348 | kfree(dc); | |
1349 | module_put(THIS_MODULE); | |
1350 | } | |
1351 | ||
d4e3b928 | 1352 | static CLOSURE_CALLBACK(cached_dev_free) |
cafe5635 | 1353 | { |
d4e3b928 | 1354 | closure_type(dc, struct cached_dev, disk.cl); |
cafe5635 | 1355 | |
3fd47bfe CL |
1356 | if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) |
1357 | cancel_writeback_rate_update_dwork(dc); | |
1358 | ||
a664d0f0 SP |
1359 | if (!IS_ERR_OR_NULL(dc->writeback_thread)) |
1360 | kthread_stop(dc->writeback_thread); | |
0f0709e6 CL |
1361 | if (!IS_ERR_OR_NULL(dc->status_update_thread)) |
1362 | kthread_stop(dc->status_update_thread); | |
cafe5635 | 1363 | |
80265d8d CL |
1364 | mutex_lock(&bch_register_lock); |
1365 | ||
b75f4aed | 1366 | if (atomic_read(&dc->running)) { |
f59fce84 | 1367 | bd_unlink_disk_holder(dc->bdev, dc->disk.disk); |
b75f4aed CH |
1368 | del_gendisk(dc->disk.disk); |
1369 | } | |
cafe5635 KO |
1370 | bcache_device_free(&dc->disk); |
1371 | list_del(&dc->list); | |
1372 | ||
1373 | mutex_unlock(&bch_register_lock); | |
1374 | ||
475389ae CH |
1375 | if (dc->sb_disk) |
1376 | put_page(virt_to_page(dc->sb_disk)); | |
e8547d42 | 1377 | |
3789fb87 CB |
1378 | if (dc->bdev_file) |
1379 | fput(dc->bdev_file); | |
cafe5635 KO |
1380 | |
1381 | wake_up(&unregister_wait); | |
1382 | ||
1383 | kobject_put(&dc->disk.kobj); | |
1384 | } | |
1385 | ||
d4e3b928 | 1386 | static CLOSURE_CALLBACK(cached_dev_flush) |
cafe5635 | 1387 | { |
d4e3b928 | 1388 | closure_type(dc, struct cached_dev, disk.cl); |
cafe5635 KO |
1389 | struct bcache_device *d = &dc->disk; |
1390 | ||
c9502ea4 | 1391 | mutex_lock(&bch_register_lock); |
c4d951dd | 1392 | bcache_device_unlink(d); |
c9502ea4 KO |
1393 | mutex_unlock(&bch_register_lock); |
1394 | ||
cafe5635 KO |
1395 | bch_cache_accounting_destroy(&dc->accounting); |
1396 | kobject_del(&d->kobj); | |
1397 | ||
1398 | continue_at(cl, cached_dev_free, system_wq); | |
1399 | } | |
1400 | ||
6f10f7d1 | 1401 | static int cached_dev_init(struct cached_dev *dc, unsigned int block_size) |
cafe5635 | 1402 | { |
f59fce84 | 1403 | int ret; |
cafe5635 | 1404 | struct io *io; |
f59fce84 | 1405 | struct request_queue *q = bdev_get_queue(dc->bdev); |
cafe5635 KO |
1406 | |
1407 | __module_get(THIS_MODULE); | |
1408 | INIT_LIST_HEAD(&dc->list); | |
f59fce84 KO |
1409 | closure_init(&dc->disk.cl, NULL); |
1410 | set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); | |
cafe5635 | 1411 | kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); |
cafe5635 | 1412 | INIT_WORK(&dc->detach, cached_dev_detach_finish); |
cb7a583e | 1413 | sema_init(&dc->sb_write_mutex, 1); |
f59fce84 KO |
1414 | INIT_LIST_HEAD(&dc->io_lru); |
1415 | spin_lock_init(&dc->io_lock); | |
1416 | bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); | |
cafe5635 | 1417 | |
cafe5635 KO |
1418 | dc->sequential_cutoff = 4 << 20; |
1419 | ||
cafe5635 KO |
1420 | for (io = dc->io; io < dc->io + RECENT_IO; io++) { |
1421 | list_add(&io->lru, &dc->io_lru); | |
1422 | hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); | |
1423 | } | |
1424 | ||
34a2cf3f | 1425 | if (bdev_io_opt(dc->bdev)) |
c78afc62 KO |
1426 | dc->partial_stripes_expensive = |
1427 | q->limits.raid_partial_stripes_expensive; | |
1428 | ||
279afbad | 1429 | ret = bcache_device_init(&dc->disk, block_size, |
a782483c | 1430 | bdev_nr_sectors(dc->bdev) - dc->sb.data_offset, |
c62b37d9 | 1431 | dc->bdev, &bcache_cached_ops); |
f59fce84 KO |
1432 | if (ret) |
1433 | return ret; | |
1434 | ||
c7b7bd07 CL |
1435 | atomic_set(&dc->io_errors, 0); |
1436 | dc->io_disable = false; | |
1437 | dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT; | |
7e027ca4 CL |
1438 | /* default to auto */ |
1439 | dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO; | |
1440 | ||
f59fce84 KO |
1441 | bch_cached_dev_request_init(dc); |
1442 | bch_cached_dev_writeback_init(dc); | |
cafe5635 | 1443 | return 0; |
cafe5635 KO |
1444 | } |
1445 | ||
1446 | /* Cached device - bcache superblock */ | |
1447 | ||
cfa0c56d | 1448 | static int register_bdev(struct cache_sb *sb, struct cache_sb_disk *sb_disk, |
3789fb87 | 1449 | struct file *bdev_file, |
cafe5635 KO |
1450 | struct cached_dev *dc) |
1451 | { | |
cafe5635 | 1452 | const char *err = "cannot allocate memory"; |
cafe5635 | 1453 | struct cache_set *c; |
0b13efec | 1454 | int ret = -ENOMEM; |
cafe5635 | 1455 | |
cafe5635 | 1456 | memcpy(&dc->sb, sb, sizeof(struct cache_sb)); |
3789fb87 CB |
1457 | dc->bdev_file = bdev_file; |
1458 | dc->bdev = file_bdev(bdev_file); | |
475389ae | 1459 | dc->sb_disk = sb_disk; |
6e916a7e | 1460 | |
f59fce84 KO |
1461 | if (cached_dev_init(dc, sb->block_size << 9)) |
1462 | goto err; | |
cafe5635 KO |
1463 | |
1464 | err = "error creating kobject"; | |
631b001f | 1465 | if (kobject_add(&dc->disk.kobj, bdev_kobj(dc->bdev), "bcache")) |
cafe5635 KO |
1466 | goto err; |
1467 | if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) | |
1468 | goto err; | |
1469 | ||
0f5cd781 | 1470 | pr_info("registered backing device %pg\n", dc->bdev); |
f59fce84 | 1471 | |
cafe5635 | 1472 | list_add(&dc->list, &uncached_devices); |
e57fd746 | 1473 | /* attach to a matched cache set if it exists */ |
cafe5635 | 1474 | list_for_each_entry(c, &bch_cache_sets, list) |
73ac105b | 1475 | bch_cached_dev_attach(dc, c, NULL); |
cafe5635 KO |
1476 | |
1477 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || | |
0b13efec CL |
1478 | BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) { |
1479 | err = "failed to run cached device"; | |
1480 | ret = bch_cached_dev_run(dc); | |
1481 | if (ret) | |
1482 | goto err; | |
1483 | } | |
cafe5635 | 1484 | |
88c12d42 | 1485 | return 0; |
cafe5635 | 1486 | err: |
0f5cd781 | 1487 | pr_notice("error %pg: %s\n", dc->bdev, err); |
f59fce84 | 1488 | bcache_device_stop(&dc->disk); |
0b13efec | 1489 | return ret; |
cafe5635 KO |
1490 | } |
1491 | ||
1492 | /* Flash only volumes */ | |
1493 | ||
2d17456e | 1494 | /* When d->kobj released */ |
cafe5635 KO |
1495 | void bch_flash_dev_release(struct kobject *kobj) |
1496 | { | |
1497 | struct bcache_device *d = container_of(kobj, struct bcache_device, | |
1498 | kobj); | |
1499 | kfree(d); | |
1500 | } | |
1501 | ||
d4e3b928 | 1502 | static CLOSURE_CALLBACK(flash_dev_free) |
cafe5635 | 1503 | { |
d4e3b928 | 1504 | closure_type(d, struct bcache_device, cl); |
1fae7cf0 | 1505 | |
e5112201 | 1506 | mutex_lock(&bch_register_lock); |
99a27d59 TJ |
1507 | atomic_long_sub(bcache_dev_sectors_dirty(d), |
1508 | &d->c->flash_dev_dirty_sectors); | |
b75f4aed | 1509 | del_gendisk(d->disk); |
cafe5635 | 1510 | bcache_device_free(d); |
e5112201 | 1511 | mutex_unlock(&bch_register_lock); |
cafe5635 KO |
1512 | kobject_put(&d->kobj); |
1513 | } | |
1514 | ||
d4e3b928 | 1515 | static CLOSURE_CALLBACK(flash_dev_flush) |
cafe5635 | 1516 | { |
d4e3b928 | 1517 | closure_type(d, struct bcache_device, cl); |
cafe5635 | 1518 | |
e5112201 | 1519 | mutex_lock(&bch_register_lock); |
ee668506 | 1520 | bcache_device_unlink(d); |
e5112201 | 1521 | mutex_unlock(&bch_register_lock); |
cafe5635 KO |
1522 | kobject_del(&d->kobj); |
1523 | continue_at(cl, flash_dev_free, system_wq); | |
1524 | } | |
1525 | ||
1526 | static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) | |
1527 | { | |
2961c3bb | 1528 | int err = -ENOMEM; |
cafe5635 KO |
1529 | struct bcache_device *d = kzalloc(sizeof(struct bcache_device), |
1530 | GFP_KERNEL); | |
1531 | if (!d) | |
2961c3bb | 1532 | goto err_ret; |
cafe5635 KO |
1533 | |
1534 | closure_init(&d->cl, NULL); | |
1535 | set_closure_fn(&d->cl, flash_dev_flush, system_wq); | |
1536 | ||
1537 | kobject_init(&d->kobj, &bch_flash_dev_ktype); | |
1538 | ||
4e1ebae3 | 1539 | if (bcache_device_init(d, block_bytes(c->cache), u->sectors, |
c62b37d9 | 1540 | NULL, &bcache_flash_ops)) |
cafe5635 KO |
1541 | goto err; |
1542 | ||
1543 | bcache_device_attach(d, c, u - c->uuids); | |
175206cf | 1544 | bch_sectors_dirty_init(d); |
cafe5635 | 1545 | bch_flash_dev_request_init(d); |
2961c3bb LC |
1546 | err = add_disk(d->disk); |
1547 | if (err) | |
1548 | goto err; | |
cafe5635 | 1549 | |
2961c3bb LC |
1550 | err = kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"); |
1551 | if (err) | |
cafe5635 KO |
1552 | goto err; |
1553 | ||
1554 | bcache_device_link(d, c, "volume"); | |
1555 | ||
5342fd42 CL |
1556 | if (bch_has_feature_obso_large_bucket(&c->cache->sb)) { |
1557 | pr_err("The obsoleted large bucket layout is unsupported, set the bcache device into read-only\n"); | |
1558 | pr_err("Please update to the latest bcache-tools to create the cache device\n"); | |
1559 | set_disk_ro(d->disk, 1); | |
1560 | } | |
1561 | ||
cafe5635 KO |
1562 | return 0; |
1563 | err: | |
1564 | kobject_put(&d->kobj); | |
2961c3bb LC |
1565 | err_ret: |
1566 | return err; | |
cafe5635 KO |
1567 | } |
1568 | ||
1569 | static int flash_devs_run(struct cache_set *c) | |
1570 | { | |
1571 | int ret = 0; | |
1572 | struct uuid_entry *u; | |
1573 | ||
1574 | for (u = c->uuids; | |
02aa8a8b | 1575 | u < c->uuids + c->nr_uuids && !ret; |
cafe5635 KO |
1576 | u++) |
1577 | if (UUID_FLASH_ONLY(u)) | |
1578 | ret = flash_dev_run(c, u); | |
1579 | ||
1580 | return ret; | |
1581 | } | |
1582 | ||
1583 | int bch_flash_dev_create(struct cache_set *c, uint64_t size) | |
1584 | { | |
1585 | struct uuid_entry *u; | |
1586 | ||
1587 | if (test_bit(CACHE_SET_STOPPING, &c->flags)) | |
1588 | return -EINTR; | |
1589 | ||
bf0c55c9 SP |
1590 | if (!test_bit(CACHE_SET_RUNNING, &c->flags)) |
1591 | return -EPERM; | |
1592 | ||
cafe5635 KO |
1593 | u = uuid_find_empty(c); |
1594 | if (!u) { | |
46f5aa88 | 1595 | pr_err("Can't create volume, no room for UUID\n"); |
cafe5635 KO |
1596 | return -EINVAL; |
1597 | } | |
1598 | ||
1599 | get_random_bytes(u->uuid, 16); | |
1600 | memset(u->label, 0, 32); | |
75cbb3f1 | 1601 | u->first_reg = u->last_reg = cpu_to_le32((u32)ktime_get_real_seconds()); |
cafe5635 KO |
1602 | |
1603 | SET_UUID_FLASH_ONLY(u, 1); | |
1604 | u->sectors = size >> 9; | |
1605 | ||
1606 | bch_uuid_write(c); | |
1607 | ||
1608 | return flash_dev_run(c, u); | |
1609 | } | |
1610 | ||
c7b7bd07 CL |
1611 | bool bch_cached_dev_error(struct cached_dev *dc) |
1612 | { | |
c7b7bd07 CL |
1613 | if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) |
1614 | return false; | |
1615 | ||
1616 | dc->io_disable = true; | |
1617 | /* make others know io_disable is true earlier */ | |
1618 | smp_mb(); | |
1619 | ||
0f5cd781 CH |
1620 | pr_err("stop %s: too many IO errors on backing device %pg\n", |
1621 | dc->disk.disk->disk_name, dc->bdev); | |
c7b7bd07 CL |
1622 | |
1623 | bcache_device_stop(&dc->disk); | |
1624 | return true; | |
1625 | } | |
1626 | ||
cafe5635 KO |
1627 | /* Cache set */ |
1628 | ||
1629 | __printf(2, 3) | |
1630 | bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) | |
1631 | { | |
46f5aa88 | 1632 | struct va_format vaf; |
cafe5635 KO |
1633 | va_list args; |
1634 | ||
77c320eb KO |
1635 | if (c->on_error != ON_ERROR_PANIC && |
1636 | test_bit(CACHE_SET_STOPPING, &c->flags)) | |
cafe5635 KO |
1637 | return false; |
1638 | ||
771f393e | 1639 | if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags)) |
46f5aa88 | 1640 | pr_info("CACHE_SET_IO_DISABLE already set\n"); |
771f393e | 1641 | |
3be11dba CL |
1642 | /* |
1643 | * XXX: we can be called from atomic context | |
1644 | * acquire_console_sem(); | |
1645 | */ | |
cafe5635 | 1646 | |
cafe5635 | 1647 | va_start(args, fmt); |
cafe5635 | 1648 | |
46f5aa88 JP |
1649 | vaf.fmt = fmt; |
1650 | vaf.va = &args; | |
1651 | ||
1652 | pr_err("error on %pU: %pV, disabling caching\n", | |
1132e56e | 1653 | c->set_uuid, &vaf); |
46f5aa88 JP |
1654 | |
1655 | va_end(args); | |
cafe5635 | 1656 | |
77c320eb KO |
1657 | if (c->on_error == ON_ERROR_PANIC) |
1658 | panic("panic forced after error\n"); | |
1659 | ||
cafe5635 KO |
1660 | bch_cache_set_unregister(c); |
1661 | return true; | |
1662 | } | |
1663 | ||
2d17456e | 1664 | /* When c->kobj released */ |
cafe5635 KO |
1665 | void bch_cache_set_release(struct kobject *kobj) |
1666 | { | |
1667 | struct cache_set *c = container_of(kobj, struct cache_set, kobj); | |
1fae7cf0 | 1668 | |
cafe5635 KO |
1669 | kfree(c); |
1670 | module_put(THIS_MODULE); | |
1671 | } | |
1672 | ||
d4e3b928 | 1673 | static CLOSURE_CALLBACK(cache_set_free) |
cafe5635 | 1674 | { |
d4e3b928 | 1675 | closure_type(c, struct cache_set, cl); |
cafe5635 | 1676 | struct cache *ca; |
cafe5635 | 1677 | |
ae171023 | 1678 | debugfs_remove(c->debug); |
cafe5635 KO |
1679 | |
1680 | bch_open_buckets_free(c); | |
1681 | bch_btree_cache_free(c); | |
1682 | bch_journal_free(c); | |
1683 | ||
a4b732a2 | 1684 | mutex_lock(&bch_register_lock); |
4a784266 CL |
1685 | bch_bset_sort_state_free(&c->sort); |
1686 | free_pages((unsigned long) c->uuids, ilog2(meta_bucket_pages(&c->cache->sb))); | |
1687 | ||
08fdb2cd CL |
1688 | ca = c->cache; |
1689 | if (ca) { | |
1690 | ca->set = NULL; | |
1691 | c->cache = NULL; | |
1692 | kobject_put(&ca->kobj); | |
1693 | } | |
cafe5635 | 1694 | |
cafe5635 | 1695 | |
da415a09 NS |
1696 | if (c->moving_gc_wq) |
1697 | destroy_workqueue(c->moving_gc_wq); | |
d19936a2 KO |
1698 | bioset_exit(&c->bio_split); |
1699 | mempool_exit(&c->fill_iter); | |
1700 | mempool_exit(&c->bio_meta); | |
1701 | mempool_exit(&c->search); | |
cafe5635 KO |
1702 | kfree(c->devices); |
1703 | ||
cafe5635 KO |
1704 | list_del(&c->list); |
1705 | mutex_unlock(&bch_register_lock); | |
1706 | ||
1132e56e | 1707 | pr_info("Cache set %pU unregistered\n", c->set_uuid); |
cafe5635 KO |
1708 | wake_up(&unregister_wait); |
1709 | ||
1710 | closure_debug_destroy(&c->cl); | |
1711 | kobject_put(&c->kobj); | |
1712 | } | |
1713 | ||
d4e3b928 | 1714 | static CLOSURE_CALLBACK(cache_set_flush) |
cafe5635 | 1715 | { |
d4e3b928 | 1716 | closure_type(c, struct cache_set, caching); |
08fdb2cd | 1717 | struct cache *ca = c->cache; |
cafe5635 | 1718 | struct btree *b; |
cafe5635 KO |
1719 | |
1720 | bch_cache_accounting_destroy(&c->accounting); | |
1721 | ||
1722 | kobject_put(&c->internal); | |
1723 | kobject_del(&c->kobj); | |
1724 | ||
b387e9b5 | 1725 | if (!IS_ERR_OR_NULL(c->gc_thread)) |
72a44517 KO |
1726 | kthread_stop(c->gc_thread); |
1727 | ||
028ddcac | 1728 | if (!IS_ERR(c->root)) |
cafe5635 KO |
1729 | list_add(&c->root->list, &c->btree_cache); |
1730 | ||
e6dcbd3e CL |
1731 | /* |
1732 | * Avoid flushing cached nodes if cache set is retiring | |
1733 | * due to too many I/O errors detected. | |
1734 | */ | |
1735 | if (!test_bit(CACHE_SET_IO_DISABLE, &c->flags)) | |
1736 | list_for_each_entry(b, &c->btree_cache, list) { | |
1737 | mutex_lock(&b->write_lock); | |
1738 | if (btree_node_dirty(b)) | |
1739 | __bch_btree_node_write(b, NULL); | |
1740 | mutex_unlock(&b->write_lock); | |
1741 | } | |
cafe5635 | 1742 | |
08fdb2cd CL |
1743 | if (ca->alloc_thread) |
1744 | kthread_stop(ca->alloc_thread); | |
79826c35 | 1745 | |
5b1016e6 KO |
1746 | if (c->journal.cur) { |
1747 | cancel_delayed_work_sync(&c->journal.work); | |
1748 | /* flush last journal entry if needed */ | |
1749 | c->journal.work.work.func(&c->journal.work.work); | |
1750 | } | |
dabb4433 | 1751 | |
cafe5635 KO |
1752 | closure_return(cl); |
1753 | } | |
1754 | ||
7e027ca4 CL |
1755 | /* |
1756 | * This function is only called when CACHE_SET_IO_DISABLE is set, which means | |
1757 | * cache set is unregistering due to too many I/O errors. In this condition, | |
1758 | * the bcache device might be stopped, it depends on stop_when_cache_set_failed | |
1759 | * value and whether the broken cache has dirty data: | |
1760 | * | |
1761 | * dc->stop_when_cache_set_failed dc->has_dirty stop bcache device | |
1762 | * BCH_CACHED_STOP_AUTO 0 NO | |
1763 | * BCH_CACHED_STOP_AUTO 1 YES | |
1764 | * BCH_CACHED_DEV_STOP_ALWAYS 0 YES | |
1765 | * BCH_CACHED_DEV_STOP_ALWAYS 1 YES | |
1766 | * | |
1767 | * The expected behavior is, if stop_when_cache_set_failed is configured to | |
1768 | * "auto" via sysfs interface, the bcache device will not be stopped if the | |
1769 | * backing device is clean on the broken cache device. | |
1770 | */ | |
1771 | static void conditional_stop_bcache_device(struct cache_set *c, | |
1772 | struct bcache_device *d, | |
1773 | struct cached_dev *dc) | |
1774 | { | |
1775 | if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) { | |
46f5aa88 | 1776 | pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.\n", |
1132e56e | 1777 | d->disk->disk_name, c->set_uuid); |
7e027ca4 CL |
1778 | bcache_device_stop(d); |
1779 | } else if (atomic_read(&dc->has_dirty)) { | |
1780 | /* | |
1781 | * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO | |
1782 | * and dc->has_dirty == 1 | |
1783 | */ | |
46f5aa88 | 1784 | pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.\n", |
7e027ca4 | 1785 | d->disk->disk_name); |
e8cf978d CIK |
1786 | /* |
1787 | * There might be a small time gap that cache set is | |
1788 | * released but bcache device is not. Inside this time | |
1789 | * gap, regular I/O requests will directly go into | |
1790 | * backing device as no cache set attached to. This | |
1791 | * behavior may also introduce potential inconsistence | |
1792 | * data in writeback mode while cache is dirty. | |
1793 | * Therefore before calling bcache_device_stop() due | |
1794 | * to a broken cache device, dc->io_disable should be | |
1795 | * explicitly set to true. | |
1796 | */ | |
1797 | dc->io_disable = true; | |
1798 | /* make others know io_disable is true earlier */ | |
1799 | smp_mb(); | |
1800 | bcache_device_stop(d); | |
7e027ca4 CL |
1801 | } else { |
1802 | /* | |
1803 | * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO | |
1804 | * and dc->has_dirty == 0 | |
1805 | */ | |
46f5aa88 | 1806 | pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.\n", |
7e027ca4 CL |
1807 | d->disk->disk_name); |
1808 | } | |
1809 | } | |
1810 | ||
d4e3b928 | 1811 | static CLOSURE_CALLBACK(__cache_set_unregister) |
cafe5635 | 1812 | { |
d4e3b928 | 1813 | closure_type(c, struct cache_set, caching); |
5caa52af | 1814 | struct cached_dev *dc; |
7e027ca4 | 1815 | struct bcache_device *d; |
cafe5635 KO |
1816 | size_t i; |
1817 | ||
1818 | mutex_lock(&bch_register_lock); | |
1819 | ||
7e027ca4 CL |
1820 | for (i = 0; i < c->devices_max_used; i++) { |
1821 | d = c->devices[i]; | |
1822 | if (!d) | |
1823 | continue; | |
1824 | ||
1825 | if (!UUID_FLASH_ONLY(&c->uuids[i]) && | |
1826 | test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { | |
1827 | dc = container_of(d, struct cached_dev, disk); | |
1828 | bch_cached_dev_detach(dc); | |
1829 | if (test_bit(CACHE_SET_IO_DISABLE, &c->flags)) | |
1830 | conditional_stop_bcache_device(c, d, dc); | |
1831 | } else { | |
1832 | bcache_device_stop(d); | |
5caa52af | 1833 | } |
7e027ca4 | 1834 | } |
cafe5635 KO |
1835 | |
1836 | mutex_unlock(&bch_register_lock); | |
1837 | ||
1838 | continue_at(cl, cache_set_flush, system_wq); | |
1839 | } | |
1840 | ||
1841 | void bch_cache_set_stop(struct cache_set *c) | |
1842 | { | |
1843 | if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) | |
63d63b51 | 1844 | /* closure_fn set to __cache_set_unregister() */ |
cafe5635 KO |
1845 | closure_queue(&c->caching); |
1846 | } | |
1847 | ||
1848 | void bch_cache_set_unregister(struct cache_set *c) | |
1849 | { | |
1850 | set_bit(CACHE_SET_UNREGISTERING, &c->flags); | |
1851 | bch_cache_set_stop(c); | |
1852 | } | |
1853 | ||
de1fafab CL |
1854 | #define alloc_meta_bucket_pages(gfp, sb) \ |
1855 | ((void *) __get_free_pages(__GFP_ZERO|__GFP_COMP|gfp, ilog2(meta_bucket_pages(sb)))) | |
cafe5635 KO |
1856 | |
1857 | struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) | |
1858 | { | |
1859 | int iter_size; | |
4a784266 | 1860 | struct cache *ca = container_of(sb, struct cache, sb); |
cafe5635 | 1861 | struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); |
1fae7cf0 | 1862 | |
cafe5635 KO |
1863 | if (!c) |
1864 | return NULL; | |
1865 | ||
1866 | __module_get(THIS_MODULE); | |
1867 | closure_init(&c->cl, NULL); | |
1868 | set_closure_fn(&c->cl, cache_set_free, system_wq); | |
1869 | ||
1870 | closure_init(&c->caching, &c->cl); | |
1871 | set_closure_fn(&c->caching, __cache_set_unregister, system_wq); | |
1872 | ||
1873 | /* Maybe create continue_at_noreturn() and use it here? */ | |
1874 | closure_set_stopped(&c->cl); | |
1875 | closure_put(&c->cl); | |
1876 | ||
1877 | kobject_init(&c->kobj, &bch_cache_set_ktype); | |
1878 | kobject_init(&c->internal, &bch_cache_set_internal_ktype); | |
1879 | ||
1880 | bch_cache_accounting_init(&c->accounting, &c->cl); | |
1881 | ||
1132e56e | 1882 | memcpy(c->set_uuid, sb->set_uuid, 16); |
d721a43f | 1883 | |
4a784266 CL |
1884 | c->cache = ca; |
1885 | c->cache->set = c; | |
cafe5635 KO |
1886 | c->bucket_bits = ilog2(sb->bucket_size); |
1887 | c->block_bits = ilog2(sb->block_size); | |
4a784266 | 1888 | c->nr_uuids = meta_bucket_bytes(sb) / sizeof(struct uuid_entry); |
2831231d | 1889 | c->devices_max_used = 0; |
ea8c5356 | 1890 | atomic_set(&c->attached_dev_nr, 0); |
4a784266 | 1891 | c->btree_pages = meta_bucket_pages(sb); |
cafe5635 KO |
1892 | if (c->btree_pages > BTREE_MAX_PAGES) |
1893 | c->btree_pages = max_t(int, c->btree_pages / 4, | |
1894 | BTREE_MAX_PAGES); | |
1895 | ||
cb7a583e | 1896 | sema_init(&c->sb_write_mutex, 1); |
e8e1d468 | 1897 | mutex_init(&c->bucket_lock); |
0a63b66d | 1898 | init_waitqueue_head(&c->btree_cache_wait); |
34cf78bf | 1899 | spin_lock_init(&c->btree_cannibalize_lock); |
35fcd848 | 1900 | init_waitqueue_head(&c->bucket_wait); |
be628be0 | 1901 | init_waitqueue_head(&c->gc_wait); |
cb7a583e | 1902 | sema_init(&c->uuid_write_mutex, 1); |
65d22e91 | 1903 | |
65d22e91 KO |
1904 | spin_lock_init(&c->btree_gc_time.lock); |
1905 | spin_lock_init(&c->btree_split_time.lock); | |
1906 | spin_lock_init(&c->btree_read_time.lock); | |
e8e1d468 | 1907 | |
cafe5635 KO |
1908 | bch_moving_init_cache_set(c); |
1909 | ||
1910 | INIT_LIST_HEAD(&c->list); | |
1911 | INIT_LIST_HEAD(&c->cached_devs); | |
1912 | INIT_LIST_HEAD(&c->btree_cache); | |
1913 | INIT_LIST_HEAD(&c->btree_cache_freeable); | |
1914 | INIT_LIST_HEAD(&c->btree_cache_freed); | |
1915 | INIT_LIST_HEAD(&c->data_buckets); | |
1916 | ||
3a861560 MM |
1917 | iter_size = sizeof(struct btree_iter) + |
1918 | ((meta_bucket_pages(sb) * PAGE_SECTORS) / sb->block_size) * | |
1919 | sizeof(struct btree_iter_set); | |
cafe5635 | 1920 | |
a42d3c64 CL |
1921 | c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL); |
1922 | if (!c->devices) | |
1923 | goto err; | |
1924 | ||
1925 | if (mempool_init_slab_pool(&c->search, 32, bch_search_cache)) | |
1926 | goto err; | |
1927 | ||
1928 | if (mempool_init_kmalloc_pool(&c->bio_meta, 2, | |
1929 | sizeof(struct bbio) + | |
4a784266 | 1930 | sizeof(struct bio_vec) * meta_bucket_pages(sb))) |
a42d3c64 CL |
1931 | goto err; |
1932 | ||
1933 | if (mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size)) | |
1934 | goto err; | |
1935 | ||
1936 | if (bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio), | |
faa8e2c4 | 1937 | BIOSET_NEED_RESCUER)) |
a42d3c64 CL |
1938 | goto err; |
1939 | ||
4a784266 | 1940 | c->uuids = alloc_meta_bucket_pages(GFP_KERNEL, sb); |
a42d3c64 CL |
1941 | if (!c->uuids) |
1942 | goto err; | |
1943 | ||
1944 | c->moving_gc_wq = alloc_workqueue("bcache_gc", WQ_MEM_RECLAIM, 0); | |
1945 | if (!c->moving_gc_wq) | |
1946 | goto err; | |
1947 | ||
1948 | if (bch_journal_alloc(c)) | |
1949 | goto err; | |
1950 | ||
1951 | if (bch_btree_cache_alloc(c)) | |
1952 | goto err; | |
1953 | ||
1954 | if (bch_open_buckets_alloc(c)) | |
1955 | goto err; | |
1956 | ||
1957 | if (bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages))) | |
cafe5635 KO |
1958 | goto err; |
1959 | ||
cafe5635 KO |
1960 | c->congested_read_threshold_us = 2000; |
1961 | c->congested_write_threshold_us = 20000; | |
7ba0d830 | 1962 | c->error_limit = DEFAULT_IO_ERROR_LIMIT; |
c5fcdedc | 1963 | c->idle_max_writeback_rate_enabled = 1; |
771f393e | 1964 | WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags)); |
cafe5635 KO |
1965 | |
1966 | return c; | |
1967 | err: | |
1968 | bch_cache_set_unregister(c); | |
1969 | return NULL; | |
1970 | } | |
1971 | ||
ce3e4cfb | 1972 | static int run_cache_set(struct cache_set *c) |
cafe5635 KO |
1973 | { |
1974 | const char *err = "cannot allocate memory"; | |
1975 | struct cached_dev *dc, *t; | |
08fdb2cd | 1976 | struct cache *ca = c->cache; |
c18536a7 | 1977 | struct closure cl; |
95f18c9d SW |
1978 | LIST_HEAD(journal); |
1979 | struct journal_replay *l; | |
cafe5635 | 1980 | |
c18536a7 | 1981 | closure_init_stack(&cl); |
cafe5635 | 1982 | |
08fdb2cd | 1983 | c->nbuckets = ca->sb.nbuckets; |
be628be0 | 1984 | set_gc_sectors(c); |
cafe5635 | 1985 | |
6f9414e0 | 1986 | if (CACHE_SYNC(&c->cache->sb)) { |
cafe5635 KO |
1987 | struct bkey *k; |
1988 | struct jset *j; | |
1989 | ||
1990 | err = "cannot allocate memory for journal"; | |
c18536a7 | 1991 | if (bch_journal_read(c, &journal)) |
cafe5635 KO |
1992 | goto err; |
1993 | ||
46f5aa88 | 1994 | pr_debug("btree_journal_read() done\n"); |
cafe5635 KO |
1995 | |
1996 | err = "no journal entries found"; | |
1997 | if (list_empty(&journal)) | |
1998 | goto err; | |
1999 | ||
2000 | j = &list_entry(journal.prev, struct journal_replay, list)->j; | |
2001 | ||
2002 | err = "IO error reading priorities"; | |
08fdb2cd CL |
2003 | if (prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev])) |
2004 | goto err; | |
cafe5635 KO |
2005 | |
2006 | /* | |
2007 | * If prio_read() fails it'll call cache_set_error and we'll | |
2008 | * tear everything down right away, but if we perhaps checked | |
2009 | * sooner we could avoid journal replay. | |
2010 | */ | |
2011 | ||
2012 | k = &j->btree_root; | |
2013 | ||
2014 | err = "bad btree root"; | |
65d45231 | 2015 | if (__bch_btree_ptr_invalid(c, k)) |
cafe5635 KO |
2016 | goto err; |
2017 | ||
2018 | err = "error reading btree root"; | |
b0d30981 CL |
2019 | c->root = bch_btree_node_get(c, NULL, k, |
2020 | j->btree_level, | |
2021 | true, NULL); | |
3eba5e0b | 2022 | if (IS_ERR(c->root)) |
cafe5635 KO |
2023 | goto err; |
2024 | ||
2025 | list_del_init(&c->root->list); | |
2026 | rw_unlock(true, c->root); | |
2027 | ||
c18536a7 | 2028 | err = uuid_read(c, j, &cl); |
cafe5635 KO |
2029 | if (err) |
2030 | goto err; | |
2031 | ||
2032 | err = "error in recovery"; | |
c18536a7 | 2033 | if (bch_btree_check(c)) |
cafe5635 KO |
2034 | goto err; |
2035 | ||
2036 | bch_journal_mark(c, &journal); | |
2531d9ee | 2037 | bch_initial_gc_finish(c); |
46f5aa88 | 2038 | pr_debug("btree_check() done\n"); |
cafe5635 KO |
2039 | |
2040 | /* | |
2041 | * bcache_journal_next() can't happen sooner, or | |
2042 | * btree_gc_finish() will give spurious errors about last_gc > | |
2043 | * gc_gen - this is a hack but oh well. | |
2044 | */ | |
2045 | bch_journal_next(&c->journal); | |
2046 | ||
119ba0f8 | 2047 | err = "error starting allocator thread"; |
08fdb2cd CL |
2048 | if (bch_cache_allocator_start(ca)) |
2049 | goto err; | |
cafe5635 KO |
2050 | |
2051 | /* | |
2052 | * First place it's safe to allocate: btree_check() and | |
2053 | * btree_gc_finish() have to run before we have buckets to | |
2054 | * allocate, and bch_bucket_alloc_set() might cause a journal | |
2055 | * entry to be written so bcache_journal_next() has to be called | |
2056 | * first. | |
2057 | * | |
2058 | * If the uuids were in the old format we have to rewrite them | |
2059 | * before the next journal entry is written: | |
2060 | */ | |
2061 | if (j->version < BCACHE_JSET_VERSION_UUID) | |
2062 | __uuid_write(c); | |
2063 | ||
ce3e4cfb CL |
2064 | err = "bcache: replay journal failed"; |
2065 | if (bch_journal_replay(c, &journal)) | |
2066 | goto err; | |
cafe5635 | 2067 | } else { |
08fdb2cd | 2068 | unsigned int j; |
cafe5635 | 2069 | |
08fdb2cd CL |
2070 | pr_notice("invalidating existing data\n"); |
2071 | ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, | |
2072 | 2, SB_JOURNAL_BUCKETS); | |
cafe5635 | 2073 | |
08fdb2cd CL |
2074 | for (j = 0; j < ca->sb.keys; j++) |
2075 | ca->sb.d[j] = ca->sb.first_bucket + j; | |
cafe5635 | 2076 | |
2531d9ee | 2077 | bch_initial_gc_finish(c); |
cafe5635 | 2078 | |
119ba0f8 | 2079 | err = "error starting allocator thread"; |
08fdb2cd CL |
2080 | if (bch_cache_allocator_start(ca)) |
2081 | goto err; | |
cafe5635 KO |
2082 | |
2083 | mutex_lock(&c->bucket_lock); | |
08fdb2cd | 2084 | bch_prio_write(ca, true); |
cafe5635 KO |
2085 | mutex_unlock(&c->bucket_lock); |
2086 | ||
cafe5635 KO |
2087 | err = "cannot allocate new UUID bucket"; |
2088 | if (__uuid_write(c)) | |
72a44517 | 2089 | goto err; |
cafe5635 KO |
2090 | |
2091 | err = "cannot allocate new btree root"; | |
2452cc89 | 2092 | c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL); |
028ddcac | 2093 | if (IS_ERR(c->root)) |
72a44517 | 2094 | goto err; |
cafe5635 | 2095 | |
2a285686 | 2096 | mutex_lock(&c->root->write_lock); |
cafe5635 | 2097 | bkey_copy_key(&c->root->key, &MAX_KEY); |
c18536a7 | 2098 | bch_btree_node_write(c->root, &cl); |
2a285686 | 2099 | mutex_unlock(&c->root->write_lock); |
cafe5635 KO |
2100 | |
2101 | bch_btree_set_root(c->root); | |
2102 | rw_unlock(true, c->root); | |
2103 | ||
2104 | /* | |
2105 | * We don't want to write the first journal entry until | |
2106 | * everything is set up - fortunately journal entries won't be | |
2107 | * written until the SET_CACHE_SYNC() here: | |
2108 | */ | |
6f9414e0 | 2109 | SET_CACHE_SYNC(&c->cache->sb, true); |
cafe5635 KO |
2110 | |
2111 | bch_journal_next(&c->journal); | |
c18536a7 | 2112 | bch_journal_meta(c, &cl); |
cafe5635 KO |
2113 | } |
2114 | ||
72a44517 KO |
2115 | err = "error starting gc thread"; |
2116 | if (bch_gc_thread_start(c)) | |
2117 | goto err; | |
2118 | ||
c18536a7 | 2119 | closure_sync(&cl); |
4a784266 | 2120 | c->cache->sb.last_mount = (u32)ktime_get_real_seconds(); |
cafe5635 KO |
2121 | bcache_write_super(c); |
2122 | ||
5342fd42 CL |
2123 | if (bch_has_feature_obso_large_bucket(&c->cache->sb)) |
2124 | pr_err("Detect obsoleted large bucket layout, all attached bcache device will be read-only\n"); | |
2125 | ||
cafe5635 | 2126 | list_for_each_entry_safe(dc, t, &uncached_devices, list) |
73ac105b | 2127 | bch_cached_dev_attach(dc, c, NULL); |
cafe5635 KO |
2128 | |
2129 | flash_devs_run(c); | |
2130 | ||
32feee36 | 2131 | bch_journal_space_reserve(&c->journal); |
bf0c55c9 | 2132 | set_bit(CACHE_SET_RUNNING, &c->flags); |
ce3e4cfb | 2133 | return 0; |
cafe5635 | 2134 | err: |
95f18c9d SW |
2135 | while (!list_empty(&journal)) { |
2136 | l = list_first_entry(&journal, struct journal_replay, list); | |
2137 | list_del(&l->list); | |
2138 | kfree(l); | |
2139 | } | |
2140 | ||
c18536a7 | 2141 | closure_sync(&cl); |
68a53c95 | 2142 | |
c8694948 | 2143 | bch_cache_set_error(c, "%s", err); |
ce3e4cfb CL |
2144 | |
2145 | return -EIO; | |
cafe5635 KO |
2146 | } |
2147 | ||
cafe5635 KO |
2148 | static const char *register_cache_set(struct cache *ca) |
2149 | { | |
2150 | char buf[12]; | |
2151 | const char *err = "cannot allocate memory"; | |
2152 | struct cache_set *c; | |
2153 | ||
2154 | list_for_each_entry(c, &bch_cache_sets, list) | |
1132e56e | 2155 | if (!memcmp(c->set_uuid, ca->sb.set_uuid, 16)) { |
697e2349 | 2156 | if (c->cache) |
cafe5635 KO |
2157 | return "duplicate cache set member"; |
2158 | ||
cafe5635 KO |
2159 | goto found; |
2160 | } | |
2161 | ||
2162 | c = bch_cache_set_alloc(&ca->sb); | |
2163 | if (!c) | |
2164 | return err; | |
2165 | ||
2166 | err = "error creating kobject"; | |
1132e56e | 2167 | if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->set_uuid) || |
cafe5635 KO |
2168 | kobject_add(&c->internal, &c->kobj, "internal")) |
2169 | goto err; | |
2170 | ||
2171 | if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) | |
2172 | goto err; | |
2173 | ||
2174 | bch_debug_init_cache_set(c); | |
2175 | ||
2176 | list_add(&c->list, &bch_cache_sets); | |
2177 | found: | |
2178 | sprintf(buf, "cache%i", ca->sb.nr_this_dev); | |
2179 | if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || | |
2180 | sysfs_create_link(&c->kobj, &ca->kobj, buf)) | |
2181 | goto err; | |
2182 | ||
d83353b3 | 2183 | kobject_get(&ca->kobj); |
cafe5635 | 2184 | ca->set = c; |
697e2349 | 2185 | ca->set->cache = ca; |
cafe5635 | 2186 | |
697e2349 CL |
2187 | err = "failed to run cache set"; |
2188 | if (run_cache_set(c) < 0) | |
2189 | goto err; | |
cafe5635 KO |
2190 | |
2191 | return NULL; | |
2192 | err: | |
2193 | bch_cache_set_unregister(c); | |
2194 | return err; | |
2195 | } | |
2196 | ||
2197 | /* Cache device */ | |
2198 | ||
2d17456e | 2199 | /* When ca->kobj released */ |
cafe5635 KO |
2200 | void bch_cache_release(struct kobject *kobj) |
2201 | { | |
2202 | struct cache *ca = container_of(kobj, struct cache, kobj); | |
6f10f7d1 | 2203 | unsigned int i; |
cafe5635 | 2204 | |
c9a78332 | 2205 | if (ca->set) { |
697e2349 CL |
2206 | BUG_ON(ca->set->cache != ca); |
2207 | ca->set->cache = NULL; | |
c9a78332 | 2208 | } |
cafe5635 | 2209 | |
c954ac8d | 2210 | free_pages((unsigned long) ca->disk_buckets, ilog2(meta_bucket_pages(&ca->sb))); |
cafe5635 KO |
2211 | kfree(ca->prio_buckets); |
2212 | vfree(ca->buckets); | |
2213 | ||
2214 | free_heap(&ca->heap); | |
cafe5635 | 2215 | free_fifo(&ca->free_inc); |
78365411 KO |
2216 | |
2217 | for (i = 0; i < RESERVE_NR; i++) | |
2218 | free_fifo(&ca->free[i]); | |
cafe5635 | 2219 | |
475389ae CH |
2220 | if (ca->sb_disk) |
2221 | put_page(virt_to_page(ca->sb_disk)); | |
cafe5635 | 2222 | |
3789fb87 CB |
2223 | if (ca->bdev_file) |
2224 | fput(ca->bdev_file); | |
cafe5635 KO |
2225 | |
2226 | kfree(ca); | |
2227 | module_put(THIS_MODULE); | |
2228 | } | |
2229 | ||
c50d4d5d | 2230 | static int cache_alloc(struct cache *ca) |
cafe5635 KO |
2231 | { |
2232 | size_t free; | |
682811b3 | 2233 | size_t btree_buckets; |
cafe5635 | 2234 | struct bucket *b; |
f6027bca DC |
2235 | int ret = -ENOMEM; |
2236 | const char *err = NULL; | |
cafe5635 | 2237 | |
cafe5635 KO |
2238 | __module_get(THIS_MODULE); |
2239 | kobject_init(&ca->kobj, &bch_cache_ktype); | |
2240 | ||
49add496 | 2241 | bio_init(&ca->journal.bio, NULL, ca->journal.bio.bi_inline_vecs, 8, 0); |
cafe5635 | 2242 | |
682811b3 TJ |
2243 | /* |
2244 | * when ca->sb.njournal_buckets is not zero, journal exists, | |
2245 | * and in bch_journal_replay(), tree node may split, | |
2246 | * so bucket of RESERVE_BTREE type is needed, | |
2247 | * the worst situation is all journal buckets are valid journal, | |
2248 | * and all the keys need to replay, | |
2249 | * so the number of RESERVE_BTREE type buckets should be as much | |
2250 | * as journal buckets | |
2251 | */ | |
2252 | btree_buckets = ca->sb.njournal_buckets ?: 8; | |
78365411 | 2253 | free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; |
3a646fd7 DC |
2254 | if (!free) { |
2255 | ret = -EPERM; | |
2256 | err = "ca->sb.nbuckets is too small"; | |
2257 | goto err_free; | |
2258 | } | |
cafe5635 | 2259 | |
f6027bca DC |
2260 | if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, |
2261 | GFP_KERNEL)) { | |
2262 | err = "ca->free[RESERVE_BTREE] alloc failed"; | |
2263 | goto err_btree_alloc; | |
2264 | } | |
2265 | ||
2266 | if (!init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), | |
2267 | GFP_KERNEL)) { | |
2268 | err = "ca->free[RESERVE_PRIO] alloc failed"; | |
2269 | goto err_prio_alloc; | |
2270 | } | |
2271 | ||
2272 | if (!init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL)) { | |
2273 | err = "ca->free[RESERVE_MOVINGGC] alloc failed"; | |
2274 | goto err_movinggc_alloc; | |
2275 | } | |
2276 | ||
2277 | if (!init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL)) { | |
2278 | err = "ca->free[RESERVE_NONE] alloc failed"; | |
2279 | goto err_none_alloc; | |
2280 | } | |
2281 | ||
2282 | if (!init_fifo(&ca->free_inc, free << 2, GFP_KERNEL)) { | |
2283 | err = "ca->free_inc alloc failed"; | |
2284 | goto err_free_inc_alloc; | |
2285 | } | |
2286 | ||
2287 | if (!init_heap(&ca->heap, free << 3, GFP_KERNEL)) { | |
2288 | err = "ca->heap alloc failed"; | |
2289 | goto err_heap_alloc; | |
2290 | } | |
2291 | ||
2292 | ca->buckets = vzalloc(array_size(sizeof(struct bucket), | |
2293 | ca->sb.nbuckets)); | |
2294 | if (!ca->buckets) { | |
2295 | err = "ca->buckets alloc failed"; | |
2296 | goto err_buckets_alloc; | |
2297 | } | |
2298 | ||
2299 | ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t), | |
2300 | prio_buckets(ca), 2), | |
2301 | GFP_KERNEL); | |
2302 | if (!ca->prio_buckets) { | |
2303 | err = "ca->prio_buckets alloc failed"; | |
2304 | goto err_prio_buckets_alloc; | |
2305 | } | |
2306 | ||
c954ac8d | 2307 | ca->disk_buckets = alloc_meta_bucket_pages(GFP_KERNEL, &ca->sb); |
f6027bca DC |
2308 | if (!ca->disk_buckets) { |
2309 | err = "ca->disk_buckets alloc failed"; | |
2310 | goto err_disk_buckets_alloc; | |
2311 | } | |
cafe5635 KO |
2312 | |
2313 | ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); | |
2314 | ||
cafe5635 KO |
2315 | for_each_bucket(b, ca) |
2316 | atomic_set(&b->pin, 0); | |
cafe5635 | 2317 | return 0; |
f6027bca DC |
2318 | |
2319 | err_disk_buckets_alloc: | |
2320 | kfree(ca->prio_buckets); | |
2321 | err_prio_buckets_alloc: | |
2322 | vfree(ca->buckets); | |
2323 | err_buckets_alloc: | |
2324 | free_heap(&ca->heap); | |
2325 | err_heap_alloc: | |
2326 | free_fifo(&ca->free_inc); | |
2327 | err_free_inc_alloc: | |
2328 | free_fifo(&ca->free[RESERVE_NONE]); | |
2329 | err_none_alloc: | |
2330 | free_fifo(&ca->free[RESERVE_MOVINGGC]); | |
2331 | err_movinggc_alloc: | |
2332 | free_fifo(&ca->free[RESERVE_PRIO]); | |
2333 | err_prio_alloc: | |
2334 | free_fifo(&ca->free[RESERVE_BTREE]); | |
2335 | err_btree_alloc: | |
3a646fd7 | 2336 | err_free: |
f6027bca DC |
2337 | module_put(THIS_MODULE); |
2338 | if (err) | |
7e84c215 | 2339 | pr_notice("error %pg: %s\n", ca->bdev, err); |
f6027bca | 2340 | return ret; |
cafe5635 KO |
2341 | } |
2342 | ||
cfa0c56d | 2343 | static int register_cache(struct cache_sb *sb, struct cache_sb_disk *sb_disk, |
3789fb87 | 2344 | struct file *bdev_file, |
631b001f | 2345 | struct cache *ca) |
cafe5635 | 2346 | { |
d9dc1702 | 2347 | const char *err = NULL; /* must be set for any error case */ |
9b299728 | 2348 | int ret = 0; |
cafe5635 | 2349 | |
f59fce84 | 2350 | memcpy(&ca->sb, sb, sizeof(struct cache_sb)); |
3789fb87 CB |
2351 | ca->bdev_file = bdev_file; |
2352 | ca->bdev = file_bdev(bdev_file); | |
475389ae | 2353 | ca->sb_disk = sb_disk; |
f59fce84 | 2354 | |
3789fb87 | 2355 | if (bdev_max_discard_sectors(file_bdev(bdev_file))) |
cafe5635 KO |
2356 | ca->discard = CACHE_DISCARD(&ca->sb); |
2357 | ||
c50d4d5d | 2358 | ret = cache_alloc(ca); |
d9dc1702 | 2359 | if (ret != 0) { |
b3856da7 JK |
2360 | if (ret == -ENOMEM) |
2361 | err = "cache_alloc(): -ENOMEM"; | |
2362 | else if (ret == -EPERM) | |
2363 | err = "cache_alloc(): cache device is too small"; | |
2364 | else | |
2365 | err = "cache_alloc(): unknown error"; | |
3789fb87 | 2366 | pr_notice("error %pg: %s\n", file_bdev(bdev_file), err); |
bb6d355c CL |
2367 | /* |
2368 | * If we failed here, it means ca->kobj is not initialized yet, | |
2369 | * kobject_put() won't be called and there is no chance to | |
3789fb87 CB |
2370 | * call fput() to bdev in bch_cache_release(). So |
2371 | * we explicitly call fput() on the block device here. | |
bb6d355c | 2372 | */ |
3789fb87 | 2373 | fput(bdev_file); |
b3856da7 | 2374 | return ret; |
d9dc1702 | 2375 | } |
f59fce84 | 2376 | |
3789fb87 | 2377 | if (kobject_add(&ca->kobj, bdev_kobj(file_bdev(bdev_file)), "bcache")) { |
b3856da7 | 2378 | pr_notice("error %pg: error calling kobject_add\n", |
3789fb87 | 2379 | file_bdev(bdev_file)); |
9b299728 EW |
2380 | ret = -ENOMEM; |
2381 | goto out; | |
2382 | } | |
cafe5635 | 2383 | |
4fa03402 | 2384 | mutex_lock(&bch_register_lock); |
cafe5635 | 2385 | err = register_cache_set(ca); |
4fa03402 KO |
2386 | mutex_unlock(&bch_register_lock); |
2387 | ||
9b299728 EW |
2388 | if (err) { |
2389 | ret = -ENODEV; | |
2390 | goto out; | |
2391 | } | |
cafe5635 | 2392 | |
3789fb87 | 2393 | pr_info("registered cache device %pg\n", file_bdev(ca->bdev_file)); |
9b299728 | 2394 | |
d83353b3 KO |
2395 | out: |
2396 | kobject_put(&ca->kobj); | |
9b299728 | 2397 | return ret; |
cafe5635 KO |
2398 | } |
2399 | ||
2400 | /* Global interfaces/init */ | |
2401 | ||
fc2d5988 CL |
2402 | static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, |
2403 | const char *buffer, size_t size); | |
0c277e21 CL |
2404 | static ssize_t bch_pending_bdevs_cleanup(struct kobject *k, |
2405 | struct kobj_attribute *attr, | |
2406 | const char *buffer, size_t size); | |
cafe5635 KO |
2407 | |
2408 | kobj_attribute_write(register, register_bcache); | |
2409 | kobj_attribute_write(register_quiet, register_bcache); | |
0c277e21 | 2410 | kobj_attribute_write(pendings_cleanup, bch_pending_bdevs_cleanup); |
cafe5635 | 2411 | |
4e7b5671 | 2412 | static bool bch_is_open_backing(dev_t dev) |
b3cf37bf | 2413 | { |
a9dd53ad GP |
2414 | struct cache_set *c, *tc; |
2415 | struct cached_dev *dc, *t; | |
2416 | ||
2417 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) | |
2418 | list_for_each_entry_safe(dc, t, &c->cached_devs, list) | |
4e7b5671 | 2419 | if (dc->bdev->bd_dev == dev) |
a9dd53ad GP |
2420 | return true; |
2421 | list_for_each_entry_safe(dc, t, &uncached_devices, list) | |
4e7b5671 | 2422 | if (dc->bdev->bd_dev == dev) |
a9dd53ad GP |
2423 | return true; |
2424 | return false; | |
2425 | } | |
2426 | ||
4e7b5671 | 2427 | static bool bch_is_open_cache(dev_t dev) |
b3cf37bf | 2428 | { |
a9dd53ad | 2429 | struct cache_set *c, *tc; |
a9dd53ad | 2430 | |
08fdb2cd CL |
2431 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) { |
2432 | struct cache *ca = c->cache; | |
2433 | ||
4e7b5671 | 2434 | if (ca->bdev->bd_dev == dev) |
08fdb2cd CL |
2435 | return true; |
2436 | } | |
2437 | ||
a9dd53ad GP |
2438 | return false; |
2439 | } | |
2440 | ||
4e7b5671 | 2441 | static bool bch_is_open(dev_t dev) |
b3cf37bf | 2442 | { |
4e7b5671 | 2443 | return bch_is_open_cache(dev) || bch_is_open_backing(dev); |
a9dd53ad GP |
2444 | } |
2445 | ||
9e23ccf8 | 2446 | struct async_reg_args { |
ee4a36f4 | 2447 | struct delayed_work reg_work; |
9e23ccf8 CL |
2448 | char *path; |
2449 | struct cache_sb *sb; | |
2450 | struct cache_sb_disk *sb_disk; | |
3789fb87 | 2451 | struct file *bdev_file; |
abcc0cbd | 2452 | void *holder; |
9e23ccf8 CL |
2453 | }; |
2454 | ||
2455 | static void register_bdev_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 | |
2461 | mutex_lock(&bch_register_lock); | |
3789fb87 | 2462 | if (register_bdev(args->sb, args->sb_disk, args->bdev_file, |
631b001f | 2463 | args->holder) < 0) |
9e23ccf8 CL |
2464 | fail = true; |
2465 | mutex_unlock(&bch_register_lock); | |
2466 | ||
9e23ccf8 CL |
2467 | if (fail) |
2468 | pr_info("error %s: fail to register backing device\n", | |
2469 | args->path); | |
2470 | kfree(args->sb); | |
2471 | kfree(args->path); | |
2472 | kfree(args); | |
2473 | module_put(THIS_MODULE); | |
2474 | } | |
2475 | ||
2476 | static void register_cache_worker(struct work_struct *work) | |
2477 | { | |
2478 | int fail = false; | |
2479 | struct async_reg_args *args = | |
ee4a36f4 | 2480 | container_of(work, struct async_reg_args, reg_work.work); |
9e23ccf8 CL |
2481 | |
2482 | /* blkdev_put() will be called in bch_cache_release() */ | |
3789fb87 | 2483 | if (register_cache(args->sb, args->sb_disk, args->bdev_file, |
631b001f | 2484 | args->holder)) |
9e23ccf8 CL |
2485 | fail = true; |
2486 | ||
9e23ccf8 CL |
2487 | if (fail) |
2488 | pr_info("error %s: fail to register cache device\n", | |
2489 | args->path); | |
2490 | kfree(args->sb); | |
2491 | kfree(args->path); | |
2492 | kfree(args); | |
2493 | module_put(THIS_MODULE); | |
2494 | } | |
2495 | ||
d7fae7b4 | 2496 | static void register_device_async(struct async_reg_args *args) |
9e23ccf8 CL |
2497 | { |
2498 | if (SB_IS_BDEV(args->sb)) | |
ee4a36f4 | 2499 | INIT_DELAYED_WORK(&args->reg_work, register_bdev_worker); |
9e23ccf8 | 2500 | else |
ee4a36f4 | 2501 | INIT_DELAYED_WORK(&args->reg_work, register_cache_worker); |
9e23ccf8 | 2502 | |
ee4a36f4 CL |
2503 | /* 10 jiffies is enough for a delay */ |
2504 | queue_delayed_work(system_wq, &args->reg_work, 10); | |
9e23ccf8 CL |
2505 | } |
2506 | ||
abcc0cbd JK |
2507 | static void *alloc_holder_object(struct cache_sb *sb) |
2508 | { | |
2509 | if (SB_IS_BDEV(sb)) | |
2510 | return kzalloc(sizeof(struct cached_dev), GFP_KERNEL); | |
2511 | return kzalloc(sizeof(struct cache), GFP_KERNEL); | |
2512 | } | |
2513 | ||
cafe5635 KO |
2514 | static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, |
2515 | const char *buffer, size_t size) | |
2516 | { | |
50246693 | 2517 | const char *err; |
29cda393 | 2518 | char *path = NULL; |
50246693 | 2519 | struct cache_sb *sb; |
cfa0c56d | 2520 | struct cache_sb_disk *sb_disk; |
3789fb87 | 2521 | struct file *bdev_file, *bdev_file2; |
2c555598 | 2522 | void *holder = NULL; |
50246693 | 2523 | ssize_t ret; |
a58e88bf | 2524 | bool async_registration = false; |
2c555598 | 2525 | bool quiet = false; |
a58e88bf CL |
2526 | |
2527 | #ifdef CONFIG_BCACHE_ASYNC_REGISTRATION | |
2528 | async_registration = true; | |
2529 | #endif | |
cafe5635 | 2530 | |
50246693 | 2531 | ret = -EBUSY; |
29cda393 | 2532 | err = "failed to reference bcache module"; |
cafe5635 | 2533 | if (!try_module_get(THIS_MODULE)) |
50246693 | 2534 | goto out; |
cafe5635 | 2535 | |
a59ff6cc CL |
2536 | /* For latest state of bcache_is_reboot */ |
2537 | smp_mb(); | |
29cda393 | 2538 | err = "bcache is in reboot"; |
a59ff6cc | 2539 | if (bcache_is_reboot) |
50246693 | 2540 | goto out_module_put; |
a59ff6cc | 2541 | |
50246693 CH |
2542 | ret = -ENOMEM; |
2543 | err = "cannot allocate memory"; | |
a56489d4 FS |
2544 | path = kstrndup(buffer, size, GFP_KERNEL); |
2545 | if (!path) | |
50246693 | 2546 | goto out_module_put; |
a56489d4 FS |
2547 | |
2548 | sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL); | |
2549 | if (!sb) | |
50246693 | 2550 | goto out_free_path; |
cafe5635 | 2551 | |
50246693 | 2552 | ret = -EINVAL; |
cafe5635 | 2553 | err = "failed to open device"; |
3789fb87 CB |
2554 | bdev_file = bdev_file_open_by_path(strim(path), BLK_OPEN_READ, NULL, NULL); |
2555 | if (IS_ERR(bdev_file)) | |
50246693 | 2556 | goto out_free_sb; |
f59fce84 | 2557 | |
3789fb87 | 2558 | err = read_super(sb, file_bdev(bdev_file), &sb_disk); |
cafe5635 | 2559 | if (err) |
50246693 | 2560 | goto out_blkdev_put; |
cafe5635 | 2561 | |
abcc0cbd JK |
2562 | holder = alloc_holder_object(sb); |
2563 | if (!holder) { | |
2564 | ret = -ENOMEM; | |
2565 | err = "cannot allocate memory"; | |
2566 | goto out_put_sb_page; | |
2567 | } | |
2568 | ||
2c555598 | 2569 | /* Now reopen in exclusive mode with proper holder */ |
3789fb87 | 2570 | bdev_file2 = bdev_file_open_by_dev(file_bdev(bdev_file)->bd_dev, |
631b001f | 2571 | BLK_OPEN_READ | BLK_OPEN_WRITE, holder, NULL); |
3789fb87 CB |
2572 | fput(bdev_file); |
2573 | bdev_file = bdev_file2; | |
2574 | if (IS_ERR(bdev_file)) { | |
2575 | ret = PTR_ERR(bdev_file); | |
2576 | bdev_file = NULL; | |
2c555598 JK |
2577 | if (ret == -EBUSY) { |
2578 | dev_t dev; | |
2579 | ||
2580 | mutex_lock(&bch_register_lock); | |
2581 | if (lookup_bdev(strim(path), &dev) == 0 && | |
2582 | bch_is_open(dev)) | |
2583 | err = "device already registered"; | |
2584 | else | |
2585 | err = "device busy"; | |
2586 | mutex_unlock(&bch_register_lock); | |
2587 | if (attr == &ksysfs_register_quiet) { | |
2588 | quiet = true; | |
2589 | ret = size; | |
2590 | } | |
2591 | } | |
2592 | goto out_free_holder; | |
2593 | } | |
2594 | ||
cc40daf9 | 2595 | err = "failed to register device"; |
a58e88bf CL |
2596 | |
2597 | if (async_registration) { | |
9e23ccf8 CL |
2598 | /* register in asynchronous way */ |
2599 | struct async_reg_args *args = | |
2600 | kzalloc(sizeof(struct async_reg_args), GFP_KERNEL); | |
2601 | ||
2602 | if (!args) { | |
2603 | ret = -ENOMEM; | |
2604 | err = "cannot allocate memory"; | |
abcc0cbd | 2605 | goto out_free_holder; |
9e23ccf8 CL |
2606 | } |
2607 | ||
2608 | args->path = path; | |
2609 | args->sb = sb; | |
2610 | args->sb_disk = sb_disk; | |
3789fb87 | 2611 | args->bdev_file = bdev_file; |
abcc0cbd | 2612 | args->holder = holder; |
d7fae7b4 | 2613 | register_device_async(args); |
9e23ccf8 CL |
2614 | /* No wait and returns to user space */ |
2615 | goto async_done; | |
2616 | } | |
2617 | ||
2903381f | 2618 | if (SB_IS_BDEV(sb)) { |
4fa03402 | 2619 | mutex_lock(&bch_register_lock); |
3789fb87 | 2620 | ret = register_bdev(sb, sb_disk, bdev_file, holder); |
4fa03402 | 2621 | mutex_unlock(&bch_register_lock); |
bb6d355c | 2622 | /* blkdev_put() will be called in cached_dev_free() */ |
fc8f19cc CH |
2623 | if (ret < 0) |
2624 | goto out_free_sb; | |
cafe5635 | 2625 | } else { |
bb6d355c | 2626 | /* blkdev_put() will be called in bch_cache_release() */ |
3789fb87 | 2627 | ret = register_cache(sb, sb_disk, bdev_file, holder); |
d55f7cb2 | 2628 | if (ret) |
fc8f19cc | 2629 | goto out_free_sb; |
cafe5635 | 2630 | } |
50246693 | 2631 | |
cafe5635 KO |
2632 | kfree(sb); |
2633 | kfree(path); | |
cafe5635 | 2634 | module_put(THIS_MODULE); |
9e23ccf8 | 2635 | async_done: |
50246693 CH |
2636 | return size; |
2637 | ||
abcc0cbd JK |
2638 | out_free_holder: |
2639 | kfree(holder); | |
50246693 | 2640 | out_put_sb_page: |
cfa0c56d | 2641 | put_page(virt_to_page(sb_disk)); |
50246693 | 2642 | out_blkdev_put: |
3789fb87 CB |
2643 | if (bdev_file) |
2644 | fput(bdev_file); | |
50246693 CH |
2645 | out_free_sb: |
2646 | kfree(sb); | |
2647 | out_free_path: | |
2648 | kfree(path); | |
ae3cd299 | 2649 | path = NULL; |
50246693 CH |
2650 | out_module_put: |
2651 | module_put(THIS_MODULE); | |
2652 | out: | |
2c555598 JK |
2653 | if (!quiet) |
2654 | pr_info("error %s: %s\n", path?path:"", err); | |
50246693 | 2655 | return ret; |
cafe5635 KO |
2656 | } |
2657 | ||
0c277e21 CL |
2658 | |
2659 | struct pdev { | |
2660 | struct list_head list; | |
2661 | struct cached_dev *dc; | |
2662 | }; | |
2663 | ||
2664 | static ssize_t bch_pending_bdevs_cleanup(struct kobject *k, | |
2665 | struct kobj_attribute *attr, | |
2666 | const char *buffer, | |
2667 | size_t size) | |
2668 | { | |
2669 | LIST_HEAD(pending_devs); | |
2670 | ssize_t ret = size; | |
2671 | struct cached_dev *dc, *tdc; | |
2672 | struct pdev *pdev, *tpdev; | |
2673 | struct cache_set *c, *tc; | |
2674 | ||
2675 | mutex_lock(&bch_register_lock); | |
2676 | list_for_each_entry_safe(dc, tdc, &uncached_devices, list) { | |
2677 | pdev = kmalloc(sizeof(struct pdev), GFP_KERNEL); | |
2678 | if (!pdev) | |
2679 | break; | |
2680 | pdev->dc = dc; | |
2681 | list_add(&pdev->list, &pending_devs); | |
2682 | } | |
2683 | ||
2684 | list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) { | |
e8092707 | 2685 | char *pdev_set_uuid = pdev->dc->sb.set_uuid; |
0c277e21 | 2686 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) { |
1132e56e | 2687 | char *set_uuid = c->set_uuid; |
0c277e21 CL |
2688 | |
2689 | if (!memcmp(pdev_set_uuid, set_uuid, 16)) { | |
2690 | list_del(&pdev->list); | |
2691 | kfree(pdev); | |
2692 | break; | |
2693 | } | |
2694 | } | |
2695 | } | |
2696 | mutex_unlock(&bch_register_lock); | |
2697 | ||
2698 | list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) { | |
46f5aa88 | 2699 | pr_info("delete pdev %p\n", pdev); |
0c277e21 CL |
2700 | list_del(&pdev->list); |
2701 | bcache_device_stop(&pdev->dc->disk); | |
2702 | kfree(pdev); | |
2703 | } | |
2704 | ||
2705 | return ret; | |
2706 | } | |
2707 | ||
cafe5635 KO |
2708 | static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) |
2709 | { | |
a59ff6cc CL |
2710 | if (bcache_is_reboot) |
2711 | return NOTIFY_DONE; | |
2712 | ||
cafe5635 KO |
2713 | if (code == SYS_DOWN || |
2714 | code == SYS_HALT || | |
2715 | code == SYS_POWER_OFF) { | |
2716 | DEFINE_WAIT(wait); | |
2717 | unsigned long start = jiffies; | |
2718 | bool stopped = false; | |
2719 | ||
2720 | struct cache_set *c, *tc; | |
2721 | struct cached_dev *dc, *tdc; | |
2722 | ||
2723 | mutex_lock(&bch_register_lock); | |
2724 | ||
a59ff6cc CL |
2725 | if (bcache_is_reboot) |
2726 | goto out; | |
2727 | ||
2728 | /* New registration is rejected since now */ | |
2729 | bcache_is_reboot = true; | |
2730 | /* | |
2731 | * Make registering caller (if there is) on other CPU | |
2732 | * core know bcache_is_reboot set to true earlier | |
2733 | */ | |
2734 | smp_mb(); | |
2735 | ||
cafe5635 KO |
2736 | if (list_empty(&bch_cache_sets) && |
2737 | list_empty(&uncached_devices)) | |
2738 | goto out; | |
2739 | ||
a59ff6cc CL |
2740 | mutex_unlock(&bch_register_lock); |
2741 | ||
46f5aa88 | 2742 | pr_info("Stopping all devices:\n"); |
cafe5635 | 2743 | |
a59ff6cc CL |
2744 | /* |
2745 | * The reason bch_register_lock is not held to call | |
2746 | * bch_cache_set_stop() and bcache_device_stop() is to | |
2747 | * avoid potential deadlock during reboot, because cache | |
a307e2ab | 2748 | * set or bcache device stopping process will acquire |
a59ff6cc CL |
2749 | * bch_register_lock too. |
2750 | * | |
2751 | * We are safe here because bcache_is_reboot sets to | |
2752 | * true already, register_bcache() will reject new | |
2753 | * registration now. bcache_is_reboot also makes sure | |
2754 | * bcache_reboot() won't be re-entered on by other thread, | |
2755 | * so there is no race in following list iteration by | |
2756 | * list_for_each_entry_safe(). | |
2757 | */ | |
cafe5635 KO |
2758 | list_for_each_entry_safe(c, tc, &bch_cache_sets, list) |
2759 | bch_cache_set_stop(c); | |
2760 | ||
2761 | list_for_each_entry_safe(dc, tdc, &uncached_devices, list) | |
2762 | bcache_device_stop(&dc->disk); | |
2763 | ||
eb8cbb6d CL |
2764 | |
2765 | /* | |
2766 | * Give an early chance for other kthreads and | |
2767 | * kworkers to stop themselves | |
2768 | */ | |
2769 | schedule(); | |
2770 | ||
cafe5635 KO |
2771 | /* What's a condition variable? */ |
2772 | while (1) { | |
eb8cbb6d | 2773 | long timeout = start + 10 * HZ - jiffies; |
cafe5635 | 2774 | |
eb8cbb6d | 2775 | mutex_lock(&bch_register_lock); |
cafe5635 KO |
2776 | stopped = list_empty(&bch_cache_sets) && |
2777 | list_empty(&uncached_devices); | |
2778 | ||
2779 | if (timeout < 0 || stopped) | |
2780 | break; | |
2781 | ||
2782 | prepare_to_wait(&unregister_wait, &wait, | |
2783 | TASK_UNINTERRUPTIBLE); | |
2784 | ||
2785 | mutex_unlock(&bch_register_lock); | |
2786 | schedule_timeout(timeout); | |
cafe5635 KO |
2787 | } |
2788 | ||
2789 | finish_wait(&unregister_wait, &wait); | |
2790 | ||
2791 | if (stopped) | |
46f5aa88 | 2792 | pr_info("All devices stopped\n"); |
cafe5635 | 2793 | else |
46f5aa88 | 2794 | pr_notice("Timeout waiting for devices to be closed\n"); |
cafe5635 KO |
2795 | out: |
2796 | mutex_unlock(&bch_register_lock); | |
2797 | } | |
2798 | ||
2799 | return NOTIFY_DONE; | |
2800 | } | |
2801 | ||
2802 | static struct notifier_block reboot = { | |
2803 | .notifier_call = bcache_reboot, | |
2804 | .priority = INT_MAX, /* before any real devices */ | |
2805 | }; | |
2806 | ||
2807 | static void bcache_exit(void) | |
2808 | { | |
2809 | bch_debug_exit(); | |
cafe5635 | 2810 | bch_request_exit(); |
cafe5635 KO |
2811 | if (bcache_kobj) |
2812 | kobject_put(bcache_kobj); | |
2813 | if (bcache_wq) | |
2814 | destroy_workqueue(bcache_wq); | |
0f843e65 GF |
2815 | if (bch_journal_wq) |
2816 | destroy_workqueue(bch_journal_wq); | |
afe78ab4 KK |
2817 | if (bch_flush_wq) |
2818 | destroy_workqueue(bch_flush_wq); | |
9f233ffe | 2819 | bch_btree_exit(); |
0f843e65 | 2820 | |
5c41c8a7 KO |
2821 | if (bcache_major) |
2822 | unregister_blkdev(bcache_major, "bcache"); | |
cafe5635 | 2823 | unregister_reboot_notifier(&reboot); |
330a4db8 | 2824 | mutex_destroy(&bch_register_lock); |
cafe5635 KO |
2825 | } |
2826 | ||
9aaf5165 CL |
2827 | /* Check and fixup module parameters */ |
2828 | static void check_module_parameters(void) | |
2829 | { | |
2830 | if (bch_cutoff_writeback_sync == 0) | |
2831 | bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC; | |
2832 | else if (bch_cutoff_writeback_sync > CUTOFF_WRITEBACK_SYNC_MAX) { | |
46f5aa88 | 2833 | pr_warn("set bch_cutoff_writeback_sync (%u) to max value %u\n", |
9aaf5165 CL |
2834 | bch_cutoff_writeback_sync, CUTOFF_WRITEBACK_SYNC_MAX); |
2835 | bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC_MAX; | |
2836 | } | |
2837 | ||
2838 | if (bch_cutoff_writeback == 0) | |
2839 | bch_cutoff_writeback = CUTOFF_WRITEBACK; | |
2840 | else if (bch_cutoff_writeback > CUTOFF_WRITEBACK_MAX) { | |
46f5aa88 | 2841 | pr_warn("set bch_cutoff_writeback (%u) to max value %u\n", |
9aaf5165 CL |
2842 | bch_cutoff_writeback, CUTOFF_WRITEBACK_MAX); |
2843 | bch_cutoff_writeback = CUTOFF_WRITEBACK_MAX; | |
2844 | } | |
2845 | ||
2846 | if (bch_cutoff_writeback > bch_cutoff_writeback_sync) { | |
46f5aa88 | 2847 | pr_warn("set bch_cutoff_writeback (%u) to %u\n", |
9aaf5165 CL |
2848 | bch_cutoff_writeback, bch_cutoff_writeback_sync); |
2849 | bch_cutoff_writeback = bch_cutoff_writeback_sync; | |
2850 | } | |
2851 | } | |
2852 | ||
cafe5635 KO |
2853 | static int __init bcache_init(void) |
2854 | { | |
2855 | static const struct attribute *files[] = { | |
2856 | &ksysfs_register.attr, | |
2857 | &ksysfs_register_quiet.attr, | |
0c277e21 | 2858 | &ksysfs_pendings_cleanup.attr, |
cafe5635 KO |
2859 | NULL |
2860 | }; | |
2861 | ||
9aaf5165 CL |
2862 | check_module_parameters(); |
2863 | ||
cafe5635 KO |
2864 | mutex_init(&bch_register_lock); |
2865 | init_waitqueue_head(&unregister_wait); | |
2866 | register_reboot_notifier(&reboot); | |
2867 | ||
2868 | bcache_major = register_blkdev(0, "bcache"); | |
2ecf0cdb ZL |
2869 | if (bcache_major < 0) { |
2870 | unregister_reboot_notifier(&reboot); | |
330a4db8 | 2871 | mutex_destroy(&bch_register_lock); |
cafe5635 | 2872 | return bcache_major; |
2ecf0cdb | 2873 | } |
cafe5635 | 2874 | |
9f233ffe KK |
2875 | if (bch_btree_init()) |
2876 | goto err; | |
2877 | ||
16c1fdf4 FS |
2878 | bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0); |
2879 | if (!bcache_wq) | |
2880 | goto err; | |
2881 | ||
afe78ab4 KK |
2882 | /* |
2883 | * Let's not make this `WQ_MEM_RECLAIM` for the following reasons: | |
2884 | * | |
2885 | * 1. It used `system_wq` before which also does no memory reclaim. | |
2886 | * 2. With `WQ_MEM_RECLAIM` desktop stalls, increased boot times, and | |
2887 | * reduced throughput can be observed. | |
2888 | * | |
2889 | * We still want to user our own queue to not congest the `system_wq`. | |
2890 | */ | |
2891 | bch_flush_wq = alloc_workqueue("bch_flush", 0, 0); | |
2892 | if (!bch_flush_wq) | |
2893 | goto err; | |
2894 | ||
0f843e65 GF |
2895 | bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0); |
2896 | if (!bch_journal_wq) | |
2897 | goto err; | |
2898 | ||
16c1fdf4 FS |
2899 | bcache_kobj = kobject_create_and_add("bcache", fs_kobj); |
2900 | if (!bcache_kobj) | |
2901 | goto err; | |
2902 | ||
2903 | if (bch_request_init() || | |
330a4db8 | 2904 | sysfs_create_files(bcache_kobj, files)) |
cafe5635 KO |
2905 | goto err; |
2906 | ||
91bafdf0 | 2907 | bch_debug_init(); |
78ac2107 | 2908 | |
a59ff6cc CL |
2909 | bcache_is_reboot = false; |
2910 | ||
cafe5635 KO |
2911 | return 0; |
2912 | err: | |
2913 | bcache_exit(); | |
2914 | return -ENOMEM; | |
2915 | } | |
2916 | ||
9aaf5165 CL |
2917 | /* |
2918 | * Module hooks | |
2919 | */ | |
cafe5635 KO |
2920 | module_exit(bcache_exit); |
2921 | module_init(bcache_init); | |
009673d0 | 2922 | |
9aaf5165 CL |
2923 | module_param(bch_cutoff_writeback, uint, 0); |
2924 | MODULE_PARM_DESC(bch_cutoff_writeback, "threshold to cutoff writeback"); | |
2925 | ||
2926 | module_param(bch_cutoff_writeback_sync, uint, 0); | |
2927 | MODULE_PARM_DESC(bch_cutoff_writeback_sync, "hard threshold to cutoff writeback"); | |
2928 | ||
009673d0 CL |
2929 | MODULE_DESCRIPTION("Bcache: a Linux block layer cache"); |
2930 | MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); | |
2931 | MODULE_LICENSE("GPL"); |