1 /* 2 * bcache setup/teardown code, and some metadata io - read a superblock and 3 * figure out what to do with it. 4 * 5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> 6 * Copyright 2012 Google, Inc. 7 */ 8 9 #include "bcache.h" 10 #include "btree.h" 11 #include "debug.h" 12 #include "extents.h" 13 #include "request.h" 14 #include "writeback.h" 15 16 #include <linux/blkdev.h> 17 #include <linux/buffer_head.h> 18 #include <linux/debugfs.h> 19 #include <linux/genhd.h> 20 #include <linux/idr.h> 21 #include <linux/kthread.h> 22 #include <linux/module.h> 23 #include <linux/random.h> 24 #include <linux/reboot.h> 25 #include <linux/sysfs.h> 26 27 MODULE_LICENSE("GPL"); 28 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); 29 30 static const char bcache_magic[] = { 31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca, 32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81 33 }; 34 35 static const char invalid_uuid[] = { 36 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78, 37 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99 38 }; 39 40 /* Default is -1; we skip past it for struct cached_dev's cache mode */ 41 const char * const bch_cache_modes[] = { 42 "default", 43 "writethrough", 44 "writeback", 45 "writearound", 46 "none", 47 NULL 48 }; 49 50 static struct kobject *bcache_kobj; 51 struct mutex bch_register_lock; 52 LIST_HEAD(bch_cache_sets); 53 static LIST_HEAD(uncached_devices); 54 55 static int bcache_major; 56 static DEFINE_IDA(bcache_minor); 57 static wait_queue_head_t unregister_wait; 58 struct workqueue_struct *bcache_wq; 59 60 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE) 61 62 /* Superblock */ 63 64 static const char *read_super(struct cache_sb *sb, struct block_device *bdev, 65 struct page **res) 66 { 67 const char *err; 68 struct cache_sb *s; 69 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE); 70 unsigned i; 71 72 if (!bh) 73 return "IO error"; 74 75 s = (struct cache_sb *) bh->b_data; 76 77 sb->offset = le64_to_cpu(s->offset); 78 sb->version = le64_to_cpu(s->version); 79 80 memcpy(sb->magic, s->magic, 16); 81 memcpy(sb->uuid, s->uuid, 16); 82 memcpy(sb->set_uuid, s->set_uuid, 16); 83 memcpy(sb->label, s->label, SB_LABEL_SIZE); 84 85 sb->flags = le64_to_cpu(s->flags); 86 sb->seq = le64_to_cpu(s->seq); 87 sb->last_mount = le32_to_cpu(s->last_mount); 88 sb->first_bucket = le16_to_cpu(s->first_bucket); 89 sb->keys = le16_to_cpu(s->keys); 90 91 for (i = 0; i < SB_JOURNAL_BUCKETS; i++) 92 sb->d[i] = le64_to_cpu(s->d[i]); 93 94 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u", 95 sb->version, sb->flags, sb->seq, sb->keys); 96 97 err = "Not a bcache superblock"; 98 if (sb->offset != SB_SECTOR) 99 goto err; 100 101 if (memcmp(sb->magic, bcache_magic, 16)) 102 goto err; 103 104 err = "Too many journal buckets"; 105 if (sb->keys > SB_JOURNAL_BUCKETS) 106 goto err; 107 108 err = "Bad checksum"; 109 if (s->csum != csum_set(s)) 110 goto err; 111 112 err = "Bad UUID"; 113 if (bch_is_zero(sb->uuid, 16)) 114 goto err; 115 116 sb->block_size = le16_to_cpu(s->block_size); 117 118 err = "Superblock block size smaller than device block size"; 119 if (sb->block_size << 9 < bdev_logical_block_size(bdev)) 120 goto err; 121 122 switch (sb->version) { 123 case BCACHE_SB_VERSION_BDEV: 124 sb->data_offset = BDEV_DATA_START_DEFAULT; 125 break; 126 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET: 127 sb->data_offset = le64_to_cpu(s->data_offset); 128 129 err = "Bad data offset"; 130 if (sb->data_offset < BDEV_DATA_START_DEFAULT) 131 goto err; 132 133 break; 134 case BCACHE_SB_VERSION_CDEV: 135 case BCACHE_SB_VERSION_CDEV_WITH_UUID: 136 sb->nbuckets = le64_to_cpu(s->nbuckets); 137 sb->block_size = le16_to_cpu(s->block_size); 138 sb->bucket_size = le16_to_cpu(s->bucket_size); 139 140 sb->nr_in_set = le16_to_cpu(s->nr_in_set); 141 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev); 142 143 err = "Too many buckets"; 144 if (sb->nbuckets > LONG_MAX) 145 goto err; 146 147 err = "Not enough buckets"; 148 if (sb->nbuckets < 1 << 7) 149 goto err; 150 151 err = "Bad block/bucket size"; 152 if (!is_power_of_2(sb->block_size) || 153 sb->block_size > PAGE_SECTORS || 154 !is_power_of_2(sb->bucket_size) || 155 sb->bucket_size < PAGE_SECTORS) 156 goto err; 157 158 err = "Invalid superblock: device too small"; 159 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets) 160 goto err; 161 162 err = "Bad UUID"; 163 if (bch_is_zero(sb->set_uuid, 16)) 164 goto err; 165 166 err = "Bad cache device number in set"; 167 if (!sb->nr_in_set || 168 sb->nr_in_set <= sb->nr_this_dev || 169 sb->nr_in_set > MAX_CACHES_PER_SET) 170 goto err; 171 172 err = "Journal buckets not sequential"; 173 for (i = 0; i < sb->keys; i++) 174 if (sb->d[i] != sb->first_bucket + i) 175 goto err; 176 177 err = "Too many journal buckets"; 178 if (sb->first_bucket + sb->keys > sb->nbuckets) 179 goto err; 180 181 err = "Invalid superblock: first bucket comes before end of super"; 182 if (sb->first_bucket * sb->bucket_size < 16) 183 goto err; 184 185 break; 186 default: 187 err = "Unsupported superblock version"; 188 goto err; 189 } 190 191 sb->last_mount = get_seconds(); 192 err = NULL; 193 194 get_page(bh->b_page); 195 *res = bh->b_page; 196 err: 197 put_bh(bh); 198 return err; 199 } 200 201 static void write_bdev_super_endio(struct bio *bio) 202 { 203 struct cached_dev *dc = bio->bi_private; 204 /* XXX: error checking */ 205 206 closure_put(&dc->sb_write); 207 } 208 209 static void __write_super(struct cache_sb *sb, struct bio *bio) 210 { 211 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page); 212 unsigned i; 213 214 bio->bi_iter.bi_sector = SB_SECTOR; 215 bio->bi_rw = REQ_SYNC|REQ_META; 216 bio->bi_iter.bi_size = SB_SIZE; 217 bch_bio_map(bio, NULL); 218 219 out->offset = cpu_to_le64(sb->offset); 220 out->version = cpu_to_le64(sb->version); 221 222 memcpy(out->uuid, sb->uuid, 16); 223 memcpy(out->set_uuid, sb->set_uuid, 16); 224 memcpy(out->label, sb->label, SB_LABEL_SIZE); 225 226 out->flags = cpu_to_le64(sb->flags); 227 out->seq = cpu_to_le64(sb->seq); 228 229 out->last_mount = cpu_to_le32(sb->last_mount); 230 out->first_bucket = cpu_to_le16(sb->first_bucket); 231 out->keys = cpu_to_le16(sb->keys); 232 233 for (i = 0; i < sb->keys; i++) 234 out->d[i] = cpu_to_le64(sb->d[i]); 235 236 out->csum = csum_set(out); 237 238 pr_debug("ver %llu, flags %llu, seq %llu", 239 sb->version, sb->flags, sb->seq); 240 241 submit_bio(REQ_WRITE, bio); 242 } 243 244 static void bch_write_bdev_super_unlock(struct closure *cl) 245 { 246 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write); 247 248 up(&dc->sb_write_mutex); 249 } 250 251 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent) 252 { 253 struct closure *cl = &dc->sb_write; 254 struct bio *bio = &dc->sb_bio; 255 256 down(&dc->sb_write_mutex); 257 closure_init(cl, parent); 258 259 bio_reset(bio); 260 bio->bi_bdev = dc->bdev; 261 bio->bi_end_io = write_bdev_super_endio; 262 bio->bi_private = dc; 263 264 closure_get(cl); 265 __write_super(&dc->sb, bio); 266 267 closure_return_with_destructor(cl, bch_write_bdev_super_unlock); 268 } 269 270 static void write_super_endio(struct bio *bio) 271 { 272 struct cache *ca = bio->bi_private; 273 274 bch_count_io_errors(ca, bio->bi_error, "writing superblock"); 275 closure_put(&ca->set->sb_write); 276 } 277 278 static void bcache_write_super_unlock(struct closure *cl) 279 { 280 struct cache_set *c = container_of(cl, struct cache_set, sb_write); 281 282 up(&c->sb_write_mutex); 283 } 284 285 void bcache_write_super(struct cache_set *c) 286 { 287 struct closure *cl = &c->sb_write; 288 struct cache *ca; 289 unsigned i; 290 291 down(&c->sb_write_mutex); 292 closure_init(cl, &c->cl); 293 294 c->sb.seq++; 295 296 for_each_cache(ca, c, i) { 297 struct bio *bio = &ca->sb_bio; 298 299 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID; 300 ca->sb.seq = c->sb.seq; 301 ca->sb.last_mount = c->sb.last_mount; 302 303 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb)); 304 305 bio_reset(bio); 306 bio->bi_bdev = ca->bdev; 307 bio->bi_end_io = write_super_endio; 308 bio->bi_private = ca; 309 310 closure_get(cl); 311 __write_super(&ca->sb, bio); 312 } 313 314 closure_return_with_destructor(cl, bcache_write_super_unlock); 315 } 316 317 /* UUID io */ 318 319 static void uuid_endio(struct bio *bio) 320 { 321 struct closure *cl = bio->bi_private; 322 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 323 324 cache_set_err_on(bio->bi_error, c, "accessing uuids"); 325 bch_bbio_free(bio, c); 326 closure_put(cl); 327 } 328 329 static void uuid_io_unlock(struct closure *cl) 330 { 331 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 332 333 up(&c->uuid_write_mutex); 334 } 335 336 static void uuid_io(struct cache_set *c, unsigned long rw, 337 struct bkey *k, struct closure *parent) 338 { 339 struct closure *cl = &c->uuid_write; 340 struct uuid_entry *u; 341 unsigned i; 342 char buf[80]; 343 344 BUG_ON(!parent); 345 down(&c->uuid_write_mutex); 346 closure_init(cl, parent); 347 348 for (i = 0; i < KEY_PTRS(k); i++) { 349 struct bio *bio = bch_bbio_alloc(c); 350 351 bio->bi_rw = REQ_SYNC|REQ_META|rw; 352 bio->bi_iter.bi_size = KEY_SIZE(k) << 9; 353 354 bio->bi_end_io = uuid_endio; 355 bio->bi_private = cl; 356 bch_bio_map(bio, c->uuids); 357 358 bch_submit_bbio(bio, c, k, i); 359 360 if (!(rw & WRITE)) 361 break; 362 } 363 364 bch_extent_to_text(buf, sizeof(buf), k); 365 pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read", buf); 366 367 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++) 368 if (!bch_is_zero(u->uuid, 16)) 369 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u", 370 u - c->uuids, u->uuid, u->label, 371 u->first_reg, u->last_reg, u->invalidated); 372 373 closure_return_with_destructor(cl, uuid_io_unlock); 374 } 375 376 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl) 377 { 378 struct bkey *k = &j->uuid_bucket; 379 380 if (__bch_btree_ptr_invalid(c, k)) 381 return "bad uuid pointer"; 382 383 bkey_copy(&c->uuid_bucket, k); 384 uuid_io(c, READ_SYNC, k, cl); 385 386 if (j->version < BCACHE_JSET_VERSION_UUIDv1) { 387 struct uuid_entry_v0 *u0 = (void *) c->uuids; 388 struct uuid_entry *u1 = (void *) c->uuids; 389 int i; 390 391 closure_sync(cl); 392 393 /* 394 * Since the new uuid entry is bigger than the old, we have to 395 * convert starting at the highest memory address and work down 396 * in order to do it in place 397 */ 398 399 for (i = c->nr_uuids - 1; 400 i >= 0; 401 --i) { 402 memcpy(u1[i].uuid, u0[i].uuid, 16); 403 memcpy(u1[i].label, u0[i].label, 32); 404 405 u1[i].first_reg = u0[i].first_reg; 406 u1[i].last_reg = u0[i].last_reg; 407 u1[i].invalidated = u0[i].invalidated; 408 409 u1[i].flags = 0; 410 u1[i].sectors = 0; 411 } 412 } 413 414 return NULL; 415 } 416 417 static int __uuid_write(struct cache_set *c) 418 { 419 BKEY_PADDED(key) k; 420 struct closure cl; 421 closure_init_stack(&cl); 422 423 lockdep_assert_held(&bch_register_lock); 424 425 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true)) 426 return 1; 427 428 SET_KEY_SIZE(&k.key, c->sb.bucket_size); 429 uuid_io(c, REQ_WRITE, &k.key, &cl); 430 closure_sync(&cl); 431 432 bkey_copy(&c->uuid_bucket, &k.key); 433 bkey_put(c, &k.key); 434 return 0; 435 } 436 437 int bch_uuid_write(struct cache_set *c) 438 { 439 int ret = __uuid_write(c); 440 441 if (!ret) 442 bch_journal_meta(c, NULL); 443 444 return ret; 445 } 446 447 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid) 448 { 449 struct uuid_entry *u; 450 451 for (u = c->uuids; 452 u < c->uuids + c->nr_uuids; u++) 453 if (!memcmp(u->uuid, uuid, 16)) 454 return u; 455 456 return NULL; 457 } 458 459 static struct uuid_entry *uuid_find_empty(struct cache_set *c) 460 { 461 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"; 462 return uuid_find(c, zero_uuid); 463 } 464 465 /* 466 * Bucket priorities/gens: 467 * 468 * For each bucket, we store on disk its 469 * 8 bit gen 470 * 16 bit priority 471 * 472 * See alloc.c for an explanation of the gen. The priority is used to implement 473 * lru (and in the future other) cache replacement policies; for most purposes 474 * it's just an opaque integer. 475 * 476 * The gens and the priorities don't have a whole lot to do with each other, and 477 * it's actually the gens that must be written out at specific times - it's no 478 * big deal if the priorities don't get written, if we lose them we just reuse 479 * buckets in suboptimal order. 480 * 481 * On disk they're stored in a packed array, and in as many buckets are required 482 * to fit them all. The buckets we use to store them form a list; the journal 483 * header points to the first bucket, the first bucket points to the second 484 * bucket, et cetera. 485 * 486 * This code is used by the allocation code; periodically (whenever it runs out 487 * of buckets to allocate from) the allocation code will invalidate some 488 * buckets, but it can't use those buckets until their new gens are safely on 489 * disk. 490 */ 491 492 static void prio_endio(struct bio *bio) 493 { 494 struct cache *ca = bio->bi_private; 495 496 cache_set_err_on(bio->bi_error, ca->set, "accessing priorities"); 497 bch_bbio_free(bio, ca->set); 498 closure_put(&ca->prio); 499 } 500 501 static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw) 502 { 503 struct closure *cl = &ca->prio; 504 struct bio *bio = bch_bbio_alloc(ca->set); 505 506 closure_init_stack(cl); 507 508 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size; 509 bio->bi_bdev = ca->bdev; 510 bio->bi_rw = REQ_SYNC|REQ_META|rw; 511 bio->bi_iter.bi_size = bucket_bytes(ca); 512 513 bio->bi_end_io = prio_endio; 514 bio->bi_private = ca; 515 bch_bio_map(bio, ca->disk_buckets); 516 517 closure_bio_submit(bio, &ca->prio); 518 closure_sync(cl); 519 } 520 521 void bch_prio_write(struct cache *ca) 522 { 523 int i; 524 struct bucket *b; 525 struct closure cl; 526 527 closure_init_stack(&cl); 528 529 lockdep_assert_held(&ca->set->bucket_lock); 530 531 ca->disk_buckets->seq++; 532 533 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca), 534 &ca->meta_sectors_written); 535 536 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free), 537 // fifo_used(&ca->free_inc), fifo_used(&ca->unused)); 538 539 for (i = prio_buckets(ca) - 1; i >= 0; --i) { 540 long bucket; 541 struct prio_set *p = ca->disk_buckets; 542 struct bucket_disk *d = p->data; 543 struct bucket_disk *end = d + prios_per_bucket(ca); 544 545 for (b = ca->buckets + i * prios_per_bucket(ca); 546 b < ca->buckets + ca->sb.nbuckets && d < end; 547 b++, d++) { 548 d->prio = cpu_to_le16(b->prio); 549 d->gen = b->gen; 550 } 551 552 p->next_bucket = ca->prio_buckets[i + 1]; 553 p->magic = pset_magic(&ca->sb); 554 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8); 555 556 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true); 557 BUG_ON(bucket == -1); 558 559 mutex_unlock(&ca->set->bucket_lock); 560 prio_io(ca, bucket, REQ_WRITE); 561 mutex_lock(&ca->set->bucket_lock); 562 563 ca->prio_buckets[i] = bucket; 564 atomic_dec_bug(&ca->buckets[bucket].pin); 565 } 566 567 mutex_unlock(&ca->set->bucket_lock); 568 569 bch_journal_meta(ca->set, &cl); 570 closure_sync(&cl); 571 572 mutex_lock(&ca->set->bucket_lock); 573 574 /* 575 * Don't want the old priorities to get garbage collected until after we 576 * finish writing the new ones, and they're journalled 577 */ 578 for (i = 0; i < prio_buckets(ca); i++) { 579 if (ca->prio_last_buckets[i]) 580 __bch_bucket_free(ca, 581 &ca->buckets[ca->prio_last_buckets[i]]); 582 583 ca->prio_last_buckets[i] = ca->prio_buckets[i]; 584 } 585 } 586 587 static void prio_read(struct cache *ca, uint64_t bucket) 588 { 589 struct prio_set *p = ca->disk_buckets; 590 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d; 591 struct bucket *b; 592 unsigned bucket_nr = 0; 593 594 for (b = ca->buckets; 595 b < ca->buckets + ca->sb.nbuckets; 596 b++, d++) { 597 if (d == end) { 598 ca->prio_buckets[bucket_nr] = bucket; 599 ca->prio_last_buckets[bucket_nr] = bucket; 600 bucket_nr++; 601 602 prio_io(ca, bucket, READ_SYNC); 603 604 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8)) 605 pr_warn("bad csum reading priorities"); 606 607 if (p->magic != pset_magic(&ca->sb)) 608 pr_warn("bad magic reading priorities"); 609 610 bucket = p->next_bucket; 611 d = p->data; 612 } 613 614 b->prio = le16_to_cpu(d->prio); 615 b->gen = b->last_gc = d->gen; 616 } 617 } 618 619 /* Bcache device */ 620 621 static int open_dev(struct block_device *b, fmode_t mode) 622 { 623 struct bcache_device *d = b->bd_disk->private_data; 624 if (test_bit(BCACHE_DEV_CLOSING, &d->flags)) 625 return -ENXIO; 626 627 closure_get(&d->cl); 628 return 0; 629 } 630 631 static void release_dev(struct gendisk *b, fmode_t mode) 632 { 633 struct bcache_device *d = b->private_data; 634 closure_put(&d->cl); 635 } 636 637 static int ioctl_dev(struct block_device *b, fmode_t mode, 638 unsigned int cmd, unsigned long arg) 639 { 640 struct bcache_device *d = b->bd_disk->private_data; 641 return d->ioctl(d, mode, cmd, arg); 642 } 643 644 static const struct block_device_operations bcache_ops = { 645 .open = open_dev, 646 .release = release_dev, 647 .ioctl = ioctl_dev, 648 .owner = THIS_MODULE, 649 }; 650 651 void bcache_device_stop(struct bcache_device *d) 652 { 653 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags)) 654 closure_queue(&d->cl); 655 } 656 657 static void bcache_device_unlink(struct bcache_device *d) 658 { 659 lockdep_assert_held(&bch_register_lock); 660 661 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) { 662 unsigned i; 663 struct cache *ca; 664 665 sysfs_remove_link(&d->c->kobj, d->name); 666 sysfs_remove_link(&d->kobj, "cache"); 667 668 for_each_cache(ca, d->c, i) 669 bd_unlink_disk_holder(ca->bdev, d->disk); 670 } 671 } 672 673 static void bcache_device_link(struct bcache_device *d, struct cache_set *c, 674 const char *name) 675 { 676 unsigned i; 677 struct cache *ca; 678 679 for_each_cache(ca, d->c, i) 680 bd_link_disk_holder(ca->bdev, d->disk); 681 682 snprintf(d->name, BCACHEDEVNAME_SIZE, 683 "%s%u", name, d->id); 684 685 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") || 686 sysfs_create_link(&c->kobj, &d->kobj, d->name), 687 "Couldn't create device <-> cache set symlinks"); 688 } 689 690 static void bcache_device_detach(struct bcache_device *d) 691 { 692 lockdep_assert_held(&bch_register_lock); 693 694 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) { 695 struct uuid_entry *u = d->c->uuids + d->id; 696 697 SET_UUID_FLASH_ONLY(u, 0); 698 memcpy(u->uuid, invalid_uuid, 16); 699 u->invalidated = cpu_to_le32(get_seconds()); 700 bch_uuid_write(d->c); 701 } 702 703 bcache_device_unlink(d); 704 705 d->c->devices[d->id] = NULL; 706 closure_put(&d->c->caching); 707 d->c = NULL; 708 } 709 710 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, 711 unsigned id) 712 { 713 d->id = id; 714 d->c = c; 715 c->devices[id] = d; 716 717 closure_get(&c->caching); 718 } 719 720 static void bcache_device_free(struct bcache_device *d) 721 { 722 lockdep_assert_held(&bch_register_lock); 723 724 pr_info("%s stopped", d->disk->disk_name); 725 726 if (d->c) 727 bcache_device_detach(d); 728 if (d->disk && d->disk->flags & GENHD_FL_UP) 729 del_gendisk(d->disk); 730 if (d->disk && d->disk->queue) 731 blk_cleanup_queue(d->disk->queue); 732 if (d->disk) { 733 ida_simple_remove(&bcache_minor, d->disk->first_minor); 734 put_disk(d->disk); 735 } 736 737 if (d->bio_split) 738 bioset_free(d->bio_split); 739 kvfree(d->full_dirty_stripes); 740 kvfree(d->stripe_sectors_dirty); 741 742 closure_debug_destroy(&d->cl); 743 } 744 745 static int bcache_device_init(struct bcache_device *d, unsigned block_size, 746 sector_t sectors) 747 { 748 struct request_queue *q; 749 size_t n; 750 int minor; 751 752 if (!d->stripe_size) 753 d->stripe_size = 1 << 31; 754 755 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size); 756 757 if (!d->nr_stripes || 758 d->nr_stripes > INT_MAX || 759 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) { 760 pr_err("nr_stripes too large"); 761 return -ENOMEM; 762 } 763 764 n = d->nr_stripes * sizeof(atomic_t); 765 d->stripe_sectors_dirty = n < PAGE_SIZE << 6 766 ? kzalloc(n, GFP_KERNEL) 767 : vzalloc(n); 768 if (!d->stripe_sectors_dirty) 769 return -ENOMEM; 770 771 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long); 772 d->full_dirty_stripes = n < PAGE_SIZE << 6 773 ? kzalloc(n, GFP_KERNEL) 774 : vzalloc(n); 775 if (!d->full_dirty_stripes) 776 return -ENOMEM; 777 778 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL); 779 if (minor < 0) 780 return minor; 781 782 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || 783 !(d->disk = alloc_disk(1))) { 784 ida_simple_remove(&bcache_minor, minor); 785 return -ENOMEM; 786 } 787 788 set_capacity(d->disk, sectors); 789 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor); 790 791 d->disk->major = bcache_major; 792 d->disk->first_minor = minor; 793 d->disk->fops = &bcache_ops; 794 d->disk->private_data = d; 795 796 q = blk_alloc_queue(GFP_KERNEL); 797 if (!q) 798 return -ENOMEM; 799 800 blk_queue_make_request(q, NULL); 801 d->disk->queue = q; 802 q->queuedata = d; 803 q->backing_dev_info.congested_data = d; 804 q->limits.max_hw_sectors = UINT_MAX; 805 q->limits.max_sectors = UINT_MAX; 806 q->limits.max_segment_size = UINT_MAX; 807 q->limits.max_segments = BIO_MAX_PAGES; 808 blk_queue_max_discard_sectors(q, UINT_MAX); 809 q->limits.discard_granularity = 512; 810 q->limits.io_min = block_size; 811 q->limits.logical_block_size = block_size; 812 q->limits.physical_block_size = block_size; 813 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags); 814 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags); 815 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags); 816 817 blk_queue_flush(q, REQ_FLUSH|REQ_FUA); 818 819 return 0; 820 } 821 822 /* Cached device */ 823 824 static void calc_cached_dev_sectors(struct cache_set *c) 825 { 826 uint64_t sectors = 0; 827 struct cached_dev *dc; 828 829 list_for_each_entry(dc, &c->cached_devs, list) 830 sectors += bdev_sectors(dc->bdev); 831 832 c->cached_dev_sectors = sectors; 833 } 834 835 void bch_cached_dev_run(struct cached_dev *dc) 836 { 837 struct bcache_device *d = &dc->disk; 838 char buf[SB_LABEL_SIZE + 1]; 839 char *env[] = { 840 "DRIVER=bcache", 841 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), 842 NULL, 843 NULL, 844 }; 845 846 memcpy(buf, dc->sb.label, SB_LABEL_SIZE); 847 buf[SB_LABEL_SIZE] = '\0'; 848 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf); 849 850 if (atomic_xchg(&dc->running, 1)) 851 return; 852 853 if (!d->c && 854 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { 855 struct closure cl; 856 closure_init_stack(&cl); 857 858 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); 859 bch_write_bdev_super(dc, &cl); 860 closure_sync(&cl); 861 } 862 863 add_disk(d->disk); 864 bd_link_disk_holder(dc->bdev, dc->disk.disk); 865 /* won't show up in the uevent file, use udevadm monitor -e instead 866 * only class / kset properties are persistent */ 867 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); 868 kfree(env[1]); 869 kfree(env[2]); 870 871 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || 872 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache")) 873 pr_debug("error creating sysfs link"); 874 } 875 876 static void cached_dev_detach_finish(struct work_struct *w) 877 { 878 struct cached_dev *dc = container_of(w, struct cached_dev, detach); 879 char buf[BDEVNAME_SIZE]; 880 struct closure cl; 881 closure_init_stack(&cl); 882 883 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)); 884 BUG_ON(atomic_read(&dc->count)); 885 886 mutex_lock(&bch_register_lock); 887 888 memset(&dc->sb.set_uuid, 0, 16); 889 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE); 890 891 bch_write_bdev_super(dc, &cl); 892 closure_sync(&cl); 893 894 bcache_device_detach(&dc->disk); 895 list_move(&dc->list, &uncached_devices); 896 897 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags); 898 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags); 899 900 mutex_unlock(&bch_register_lock); 901 902 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf)); 903 904 /* Drop ref we took in cached_dev_detach() */ 905 closure_put(&dc->disk.cl); 906 } 907 908 void bch_cached_dev_detach(struct cached_dev *dc) 909 { 910 lockdep_assert_held(&bch_register_lock); 911 912 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) 913 return; 914 915 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) 916 return; 917 918 /* 919 * Block the device from being closed and freed until we're finished 920 * detaching 921 */ 922 closure_get(&dc->disk.cl); 923 924 bch_writeback_queue(dc); 925 cached_dev_put(dc); 926 } 927 928 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c) 929 { 930 uint32_t rtime = cpu_to_le32(get_seconds()); 931 struct uuid_entry *u; 932 char buf[BDEVNAME_SIZE]; 933 934 bdevname(dc->bdev, buf); 935 936 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)) 937 return -ENOENT; 938 939 if (dc->disk.c) { 940 pr_err("Can't attach %s: already attached", buf); 941 return -EINVAL; 942 } 943 944 if (test_bit(CACHE_SET_STOPPING, &c->flags)) { 945 pr_err("Can't attach %s: shutting down", buf); 946 return -EINVAL; 947 } 948 949 if (dc->sb.block_size < c->sb.block_size) { 950 /* Will die */ 951 pr_err("Couldn't attach %s: block size less than set's block size", 952 buf); 953 return -EINVAL; 954 } 955 956 u = uuid_find(c, dc->sb.uuid); 957 958 if (u && 959 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || 960 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { 961 memcpy(u->uuid, invalid_uuid, 16); 962 u->invalidated = cpu_to_le32(get_seconds()); 963 u = NULL; 964 } 965 966 if (!u) { 967 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 968 pr_err("Couldn't find uuid for %s in set", buf); 969 return -ENOENT; 970 } 971 972 u = uuid_find_empty(c); 973 if (!u) { 974 pr_err("Not caching %s, no room for UUID", buf); 975 return -EINVAL; 976 } 977 } 978 979 /* Deadlocks since we're called via sysfs... 980 sysfs_remove_file(&dc->kobj, &sysfs_attach); 981 */ 982 983 if (bch_is_zero(u->uuid, 16)) { 984 struct closure cl; 985 closure_init_stack(&cl); 986 987 memcpy(u->uuid, dc->sb.uuid, 16); 988 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); 989 u->first_reg = u->last_reg = rtime; 990 bch_uuid_write(c); 991 992 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16); 993 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); 994 995 bch_write_bdev_super(dc, &cl); 996 closure_sync(&cl); 997 } else { 998 u->last_reg = rtime; 999 bch_uuid_write(c); 1000 } 1001 1002 bcache_device_attach(&dc->disk, c, u - c->uuids); 1003 list_move(&dc->list, &c->cached_devs); 1004 calc_cached_dev_sectors(c); 1005 1006 smp_wmb(); 1007 /* 1008 * dc->c must be set before dc->count != 0 - paired with the mb in 1009 * cached_dev_get() 1010 */ 1011 atomic_set(&dc->count, 1); 1012 1013 if (bch_cached_dev_writeback_start(dc)) 1014 return -ENOMEM; 1015 1016 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1017 bch_sectors_dirty_init(dc); 1018 atomic_set(&dc->has_dirty, 1); 1019 atomic_inc(&dc->count); 1020 bch_writeback_queue(dc); 1021 } 1022 1023 bch_cached_dev_run(dc); 1024 bcache_device_link(&dc->disk, c, "bdev"); 1025 1026 pr_info("Caching %s as %s on set %pU", 1027 bdevname(dc->bdev, buf), dc->disk.disk->disk_name, 1028 dc->disk.c->sb.set_uuid); 1029 return 0; 1030 } 1031 1032 void bch_cached_dev_release(struct kobject *kobj) 1033 { 1034 struct cached_dev *dc = container_of(kobj, struct cached_dev, 1035 disk.kobj); 1036 kfree(dc); 1037 module_put(THIS_MODULE); 1038 } 1039 1040 static void cached_dev_free(struct closure *cl) 1041 { 1042 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1043 1044 cancel_delayed_work_sync(&dc->writeback_rate_update); 1045 if (!IS_ERR_OR_NULL(dc->writeback_thread)) 1046 kthread_stop(dc->writeback_thread); 1047 1048 mutex_lock(&bch_register_lock); 1049 1050 if (atomic_read(&dc->running)) 1051 bd_unlink_disk_holder(dc->bdev, dc->disk.disk); 1052 bcache_device_free(&dc->disk); 1053 list_del(&dc->list); 1054 1055 mutex_unlock(&bch_register_lock); 1056 1057 if (!IS_ERR_OR_NULL(dc->bdev)) 1058 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1059 1060 wake_up(&unregister_wait); 1061 1062 kobject_put(&dc->disk.kobj); 1063 } 1064 1065 static void cached_dev_flush(struct closure *cl) 1066 { 1067 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1068 struct bcache_device *d = &dc->disk; 1069 1070 mutex_lock(&bch_register_lock); 1071 bcache_device_unlink(d); 1072 mutex_unlock(&bch_register_lock); 1073 1074 bch_cache_accounting_destroy(&dc->accounting); 1075 kobject_del(&d->kobj); 1076 1077 continue_at(cl, cached_dev_free, system_wq); 1078 } 1079 1080 static int cached_dev_init(struct cached_dev *dc, unsigned block_size) 1081 { 1082 int ret; 1083 struct io *io; 1084 struct request_queue *q = bdev_get_queue(dc->bdev); 1085 1086 __module_get(THIS_MODULE); 1087 INIT_LIST_HEAD(&dc->list); 1088 closure_init(&dc->disk.cl, NULL); 1089 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); 1090 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); 1091 INIT_WORK(&dc->detach, cached_dev_detach_finish); 1092 sema_init(&dc->sb_write_mutex, 1); 1093 INIT_LIST_HEAD(&dc->io_lru); 1094 spin_lock_init(&dc->io_lock); 1095 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); 1096 1097 dc->sequential_cutoff = 4 << 20; 1098 1099 for (io = dc->io; io < dc->io + RECENT_IO; io++) { 1100 list_add(&io->lru, &dc->io_lru); 1101 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); 1102 } 1103 1104 dc->disk.stripe_size = q->limits.io_opt >> 9; 1105 1106 if (dc->disk.stripe_size) 1107 dc->partial_stripes_expensive = 1108 q->limits.raid_partial_stripes_expensive; 1109 1110 ret = bcache_device_init(&dc->disk, block_size, 1111 dc->bdev->bd_part->nr_sects - dc->sb.data_offset); 1112 if (ret) 1113 return ret; 1114 1115 set_capacity(dc->disk.disk, 1116 dc->bdev->bd_part->nr_sects - dc->sb.data_offset); 1117 1118 dc->disk.disk->queue->backing_dev_info.ra_pages = 1119 max(dc->disk.disk->queue->backing_dev_info.ra_pages, 1120 q->backing_dev_info.ra_pages); 1121 1122 bch_cached_dev_request_init(dc); 1123 bch_cached_dev_writeback_init(dc); 1124 return 0; 1125 } 1126 1127 /* Cached device - bcache superblock */ 1128 1129 static void register_bdev(struct cache_sb *sb, struct page *sb_page, 1130 struct block_device *bdev, 1131 struct cached_dev *dc) 1132 { 1133 char name[BDEVNAME_SIZE]; 1134 const char *err = "cannot allocate memory"; 1135 struct cache_set *c; 1136 1137 memcpy(&dc->sb, sb, sizeof(struct cache_sb)); 1138 dc->bdev = bdev; 1139 dc->bdev->bd_holder = dc; 1140 1141 bio_init(&dc->sb_bio); 1142 dc->sb_bio.bi_max_vecs = 1; 1143 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs; 1144 dc->sb_bio.bi_io_vec[0].bv_page = sb_page; 1145 get_page(sb_page); 1146 1147 if (cached_dev_init(dc, sb->block_size << 9)) 1148 goto err; 1149 1150 err = "error creating kobject"; 1151 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj, 1152 "bcache")) 1153 goto err; 1154 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) 1155 goto err; 1156 1157 pr_info("registered backing device %s", bdevname(bdev, name)); 1158 1159 list_add(&dc->list, &uncached_devices); 1160 list_for_each_entry(c, &bch_cache_sets, list) 1161 bch_cached_dev_attach(dc, c); 1162 1163 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || 1164 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) 1165 bch_cached_dev_run(dc); 1166 1167 return; 1168 err: 1169 pr_notice("error opening %s: %s", bdevname(bdev, name), err); 1170 bcache_device_stop(&dc->disk); 1171 } 1172 1173 /* Flash only volumes */ 1174 1175 void bch_flash_dev_release(struct kobject *kobj) 1176 { 1177 struct bcache_device *d = container_of(kobj, struct bcache_device, 1178 kobj); 1179 kfree(d); 1180 } 1181 1182 static void flash_dev_free(struct closure *cl) 1183 { 1184 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1185 mutex_lock(&bch_register_lock); 1186 bcache_device_free(d); 1187 mutex_unlock(&bch_register_lock); 1188 kobject_put(&d->kobj); 1189 } 1190 1191 static void flash_dev_flush(struct closure *cl) 1192 { 1193 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1194 1195 mutex_lock(&bch_register_lock); 1196 bcache_device_unlink(d); 1197 mutex_unlock(&bch_register_lock); 1198 kobject_del(&d->kobj); 1199 continue_at(cl, flash_dev_free, system_wq); 1200 } 1201 1202 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) 1203 { 1204 struct bcache_device *d = kzalloc(sizeof(struct bcache_device), 1205 GFP_KERNEL); 1206 if (!d) 1207 return -ENOMEM; 1208 1209 closure_init(&d->cl, NULL); 1210 set_closure_fn(&d->cl, flash_dev_flush, system_wq); 1211 1212 kobject_init(&d->kobj, &bch_flash_dev_ktype); 1213 1214 if (bcache_device_init(d, block_bytes(c), u->sectors)) 1215 goto err; 1216 1217 bcache_device_attach(d, c, u - c->uuids); 1218 bch_flash_dev_request_init(d); 1219 add_disk(d->disk); 1220 1221 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache")) 1222 goto err; 1223 1224 bcache_device_link(d, c, "volume"); 1225 1226 return 0; 1227 err: 1228 kobject_put(&d->kobj); 1229 return -ENOMEM; 1230 } 1231 1232 static int flash_devs_run(struct cache_set *c) 1233 { 1234 int ret = 0; 1235 struct uuid_entry *u; 1236 1237 for (u = c->uuids; 1238 u < c->uuids + c->nr_uuids && !ret; 1239 u++) 1240 if (UUID_FLASH_ONLY(u)) 1241 ret = flash_dev_run(c, u); 1242 1243 return ret; 1244 } 1245 1246 int bch_flash_dev_create(struct cache_set *c, uint64_t size) 1247 { 1248 struct uuid_entry *u; 1249 1250 if (test_bit(CACHE_SET_STOPPING, &c->flags)) 1251 return -EINTR; 1252 1253 if (!test_bit(CACHE_SET_RUNNING, &c->flags)) 1254 return -EPERM; 1255 1256 u = uuid_find_empty(c); 1257 if (!u) { 1258 pr_err("Can't create volume, no room for UUID"); 1259 return -EINVAL; 1260 } 1261 1262 get_random_bytes(u->uuid, 16); 1263 memset(u->label, 0, 32); 1264 u->first_reg = u->last_reg = cpu_to_le32(get_seconds()); 1265 1266 SET_UUID_FLASH_ONLY(u, 1); 1267 u->sectors = size >> 9; 1268 1269 bch_uuid_write(c); 1270 1271 return flash_dev_run(c, u); 1272 } 1273 1274 /* Cache set */ 1275 1276 __printf(2, 3) 1277 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) 1278 { 1279 va_list args; 1280 1281 if (c->on_error != ON_ERROR_PANIC && 1282 test_bit(CACHE_SET_STOPPING, &c->flags)) 1283 return false; 1284 1285 /* XXX: we can be called from atomic context 1286 acquire_console_sem(); 1287 */ 1288 1289 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid); 1290 1291 va_start(args, fmt); 1292 vprintk(fmt, args); 1293 va_end(args); 1294 1295 printk(", disabling caching\n"); 1296 1297 if (c->on_error == ON_ERROR_PANIC) 1298 panic("panic forced after error\n"); 1299 1300 bch_cache_set_unregister(c); 1301 return true; 1302 } 1303 1304 void bch_cache_set_release(struct kobject *kobj) 1305 { 1306 struct cache_set *c = container_of(kobj, struct cache_set, kobj); 1307 kfree(c); 1308 module_put(THIS_MODULE); 1309 } 1310 1311 static void cache_set_free(struct closure *cl) 1312 { 1313 struct cache_set *c = container_of(cl, struct cache_set, cl); 1314 struct cache *ca; 1315 unsigned i; 1316 1317 if (!IS_ERR_OR_NULL(c->debug)) 1318 debugfs_remove(c->debug); 1319 1320 bch_open_buckets_free(c); 1321 bch_btree_cache_free(c); 1322 bch_journal_free(c); 1323 1324 for_each_cache(ca, c, i) 1325 if (ca) { 1326 ca->set = NULL; 1327 c->cache[ca->sb.nr_this_dev] = NULL; 1328 kobject_put(&ca->kobj); 1329 } 1330 1331 bch_bset_sort_state_free(&c->sort); 1332 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c))); 1333 1334 if (c->moving_gc_wq) 1335 destroy_workqueue(c->moving_gc_wq); 1336 if (c->bio_split) 1337 bioset_free(c->bio_split); 1338 if (c->fill_iter) 1339 mempool_destroy(c->fill_iter); 1340 if (c->bio_meta) 1341 mempool_destroy(c->bio_meta); 1342 if (c->search) 1343 mempool_destroy(c->search); 1344 kfree(c->devices); 1345 1346 mutex_lock(&bch_register_lock); 1347 list_del(&c->list); 1348 mutex_unlock(&bch_register_lock); 1349 1350 pr_info("Cache set %pU unregistered", c->sb.set_uuid); 1351 wake_up(&unregister_wait); 1352 1353 closure_debug_destroy(&c->cl); 1354 kobject_put(&c->kobj); 1355 } 1356 1357 static void cache_set_flush(struct closure *cl) 1358 { 1359 struct cache_set *c = container_of(cl, struct cache_set, caching); 1360 struct cache *ca; 1361 struct btree *b; 1362 unsigned i; 1363 1364 bch_cache_accounting_destroy(&c->accounting); 1365 1366 kobject_put(&c->internal); 1367 kobject_del(&c->kobj); 1368 1369 if (c->gc_thread) 1370 kthread_stop(c->gc_thread); 1371 1372 if (!IS_ERR_OR_NULL(c->root)) 1373 list_add(&c->root->list, &c->btree_cache); 1374 1375 /* Should skip this if we're unregistering because of an error */ 1376 list_for_each_entry(b, &c->btree_cache, list) { 1377 mutex_lock(&b->write_lock); 1378 if (btree_node_dirty(b)) 1379 __bch_btree_node_write(b, NULL); 1380 mutex_unlock(&b->write_lock); 1381 } 1382 1383 for_each_cache(ca, c, i) 1384 if (ca->alloc_thread) 1385 kthread_stop(ca->alloc_thread); 1386 1387 if (c->journal.cur) { 1388 cancel_delayed_work_sync(&c->journal.work); 1389 /* flush last journal entry if needed */ 1390 c->journal.work.work.func(&c->journal.work.work); 1391 } 1392 1393 closure_return(cl); 1394 } 1395 1396 static void __cache_set_unregister(struct closure *cl) 1397 { 1398 struct cache_set *c = container_of(cl, struct cache_set, caching); 1399 struct cached_dev *dc; 1400 size_t i; 1401 1402 mutex_lock(&bch_register_lock); 1403 1404 for (i = 0; i < c->nr_uuids; i++) 1405 if (c->devices[i]) { 1406 if (!UUID_FLASH_ONLY(&c->uuids[i]) && 1407 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { 1408 dc = container_of(c->devices[i], 1409 struct cached_dev, disk); 1410 bch_cached_dev_detach(dc); 1411 } else { 1412 bcache_device_stop(c->devices[i]); 1413 } 1414 } 1415 1416 mutex_unlock(&bch_register_lock); 1417 1418 continue_at(cl, cache_set_flush, system_wq); 1419 } 1420 1421 void bch_cache_set_stop(struct cache_set *c) 1422 { 1423 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) 1424 closure_queue(&c->caching); 1425 } 1426 1427 void bch_cache_set_unregister(struct cache_set *c) 1428 { 1429 set_bit(CACHE_SET_UNREGISTERING, &c->flags); 1430 bch_cache_set_stop(c); 1431 } 1432 1433 #define alloc_bucket_pages(gfp, c) \ 1434 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c)))) 1435 1436 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) 1437 { 1438 int iter_size; 1439 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); 1440 if (!c) 1441 return NULL; 1442 1443 __module_get(THIS_MODULE); 1444 closure_init(&c->cl, NULL); 1445 set_closure_fn(&c->cl, cache_set_free, system_wq); 1446 1447 closure_init(&c->caching, &c->cl); 1448 set_closure_fn(&c->caching, __cache_set_unregister, system_wq); 1449 1450 /* Maybe create continue_at_noreturn() and use it here? */ 1451 closure_set_stopped(&c->cl); 1452 closure_put(&c->cl); 1453 1454 kobject_init(&c->kobj, &bch_cache_set_ktype); 1455 kobject_init(&c->internal, &bch_cache_set_internal_ktype); 1456 1457 bch_cache_accounting_init(&c->accounting, &c->cl); 1458 1459 memcpy(c->sb.set_uuid, sb->set_uuid, 16); 1460 c->sb.block_size = sb->block_size; 1461 c->sb.bucket_size = sb->bucket_size; 1462 c->sb.nr_in_set = sb->nr_in_set; 1463 c->sb.last_mount = sb->last_mount; 1464 c->bucket_bits = ilog2(sb->bucket_size); 1465 c->block_bits = ilog2(sb->block_size); 1466 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry); 1467 1468 c->btree_pages = bucket_pages(c); 1469 if (c->btree_pages > BTREE_MAX_PAGES) 1470 c->btree_pages = max_t(int, c->btree_pages / 4, 1471 BTREE_MAX_PAGES); 1472 1473 sema_init(&c->sb_write_mutex, 1); 1474 mutex_init(&c->bucket_lock); 1475 init_waitqueue_head(&c->btree_cache_wait); 1476 init_waitqueue_head(&c->bucket_wait); 1477 sema_init(&c->uuid_write_mutex, 1); 1478 1479 spin_lock_init(&c->btree_gc_time.lock); 1480 spin_lock_init(&c->btree_split_time.lock); 1481 spin_lock_init(&c->btree_read_time.lock); 1482 1483 bch_moving_init_cache_set(c); 1484 1485 INIT_LIST_HEAD(&c->list); 1486 INIT_LIST_HEAD(&c->cached_devs); 1487 INIT_LIST_HEAD(&c->btree_cache); 1488 INIT_LIST_HEAD(&c->btree_cache_freeable); 1489 INIT_LIST_HEAD(&c->btree_cache_freed); 1490 INIT_LIST_HEAD(&c->data_buckets); 1491 1492 c->search = mempool_create_slab_pool(32, bch_search_cache); 1493 if (!c->search) 1494 goto err; 1495 1496 iter_size = (sb->bucket_size / sb->block_size + 1) * 1497 sizeof(struct btree_iter_set); 1498 1499 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) || 1500 !(c->bio_meta = mempool_create_kmalloc_pool(2, 1501 sizeof(struct bbio) + sizeof(struct bio_vec) * 1502 bucket_pages(c))) || 1503 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) || 1504 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || 1505 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) || 1506 !(c->moving_gc_wq = create_workqueue("bcache_gc")) || 1507 bch_journal_alloc(c) || 1508 bch_btree_cache_alloc(c) || 1509 bch_open_buckets_alloc(c) || 1510 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages))) 1511 goto err; 1512 1513 c->congested_read_threshold_us = 2000; 1514 c->congested_write_threshold_us = 20000; 1515 c->error_limit = 8 << IO_ERROR_SHIFT; 1516 1517 return c; 1518 err: 1519 bch_cache_set_unregister(c); 1520 return NULL; 1521 } 1522 1523 static void run_cache_set(struct cache_set *c) 1524 { 1525 const char *err = "cannot allocate memory"; 1526 struct cached_dev *dc, *t; 1527 struct cache *ca; 1528 struct closure cl; 1529 unsigned i; 1530 1531 closure_init_stack(&cl); 1532 1533 for_each_cache(ca, c, i) 1534 c->nbuckets += ca->sb.nbuckets; 1535 1536 if (CACHE_SYNC(&c->sb)) { 1537 LIST_HEAD(journal); 1538 struct bkey *k; 1539 struct jset *j; 1540 1541 err = "cannot allocate memory for journal"; 1542 if (bch_journal_read(c, &journal)) 1543 goto err; 1544 1545 pr_debug("btree_journal_read() done"); 1546 1547 err = "no journal entries found"; 1548 if (list_empty(&journal)) 1549 goto err; 1550 1551 j = &list_entry(journal.prev, struct journal_replay, list)->j; 1552 1553 err = "IO error reading priorities"; 1554 for_each_cache(ca, c, i) 1555 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]); 1556 1557 /* 1558 * If prio_read() fails it'll call cache_set_error and we'll 1559 * tear everything down right away, but if we perhaps checked 1560 * sooner we could avoid journal replay. 1561 */ 1562 1563 k = &j->btree_root; 1564 1565 err = "bad btree root"; 1566 if (__bch_btree_ptr_invalid(c, k)) 1567 goto err; 1568 1569 err = "error reading btree root"; 1570 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL); 1571 if (IS_ERR_OR_NULL(c->root)) 1572 goto err; 1573 1574 list_del_init(&c->root->list); 1575 rw_unlock(true, c->root); 1576 1577 err = uuid_read(c, j, &cl); 1578 if (err) 1579 goto err; 1580 1581 err = "error in recovery"; 1582 if (bch_btree_check(c)) 1583 goto err; 1584 1585 bch_journal_mark(c, &journal); 1586 bch_initial_gc_finish(c); 1587 pr_debug("btree_check() done"); 1588 1589 /* 1590 * bcache_journal_next() can't happen sooner, or 1591 * btree_gc_finish() will give spurious errors about last_gc > 1592 * gc_gen - this is a hack but oh well. 1593 */ 1594 bch_journal_next(&c->journal); 1595 1596 err = "error starting allocator thread"; 1597 for_each_cache(ca, c, i) 1598 if (bch_cache_allocator_start(ca)) 1599 goto err; 1600 1601 /* 1602 * First place it's safe to allocate: btree_check() and 1603 * btree_gc_finish() have to run before we have buckets to 1604 * allocate, and bch_bucket_alloc_set() might cause a journal 1605 * entry to be written so bcache_journal_next() has to be called 1606 * first. 1607 * 1608 * If the uuids were in the old format we have to rewrite them 1609 * before the next journal entry is written: 1610 */ 1611 if (j->version < BCACHE_JSET_VERSION_UUID) 1612 __uuid_write(c); 1613 1614 bch_journal_replay(c, &journal); 1615 } else { 1616 pr_notice("invalidating existing data"); 1617 1618 for_each_cache(ca, c, i) { 1619 unsigned j; 1620 1621 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, 1622 2, SB_JOURNAL_BUCKETS); 1623 1624 for (j = 0; j < ca->sb.keys; j++) 1625 ca->sb.d[j] = ca->sb.first_bucket + j; 1626 } 1627 1628 bch_initial_gc_finish(c); 1629 1630 err = "error starting allocator thread"; 1631 for_each_cache(ca, c, i) 1632 if (bch_cache_allocator_start(ca)) 1633 goto err; 1634 1635 mutex_lock(&c->bucket_lock); 1636 for_each_cache(ca, c, i) 1637 bch_prio_write(ca); 1638 mutex_unlock(&c->bucket_lock); 1639 1640 err = "cannot allocate new UUID bucket"; 1641 if (__uuid_write(c)) 1642 goto err; 1643 1644 err = "cannot allocate new btree root"; 1645 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL); 1646 if (IS_ERR_OR_NULL(c->root)) 1647 goto err; 1648 1649 mutex_lock(&c->root->write_lock); 1650 bkey_copy_key(&c->root->key, &MAX_KEY); 1651 bch_btree_node_write(c->root, &cl); 1652 mutex_unlock(&c->root->write_lock); 1653 1654 bch_btree_set_root(c->root); 1655 rw_unlock(true, c->root); 1656 1657 /* 1658 * We don't want to write the first journal entry until 1659 * everything is set up - fortunately journal entries won't be 1660 * written until the SET_CACHE_SYNC() here: 1661 */ 1662 SET_CACHE_SYNC(&c->sb, true); 1663 1664 bch_journal_next(&c->journal); 1665 bch_journal_meta(c, &cl); 1666 } 1667 1668 err = "error starting gc thread"; 1669 if (bch_gc_thread_start(c)) 1670 goto err; 1671 1672 closure_sync(&cl); 1673 c->sb.last_mount = get_seconds(); 1674 bcache_write_super(c); 1675 1676 list_for_each_entry_safe(dc, t, &uncached_devices, list) 1677 bch_cached_dev_attach(dc, c); 1678 1679 flash_devs_run(c); 1680 1681 set_bit(CACHE_SET_RUNNING, &c->flags); 1682 return; 1683 err: 1684 closure_sync(&cl); 1685 /* XXX: test this, it's broken */ 1686 bch_cache_set_error(c, "%s", err); 1687 } 1688 1689 static bool can_attach_cache(struct cache *ca, struct cache_set *c) 1690 { 1691 return ca->sb.block_size == c->sb.block_size && 1692 ca->sb.bucket_size == c->sb.bucket_size && 1693 ca->sb.nr_in_set == c->sb.nr_in_set; 1694 } 1695 1696 static const char *register_cache_set(struct cache *ca) 1697 { 1698 char buf[12]; 1699 const char *err = "cannot allocate memory"; 1700 struct cache_set *c; 1701 1702 list_for_each_entry(c, &bch_cache_sets, list) 1703 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) { 1704 if (c->cache[ca->sb.nr_this_dev]) 1705 return "duplicate cache set member"; 1706 1707 if (!can_attach_cache(ca, c)) 1708 return "cache sb does not match set"; 1709 1710 if (!CACHE_SYNC(&ca->sb)) 1711 SET_CACHE_SYNC(&c->sb, false); 1712 1713 goto found; 1714 } 1715 1716 c = bch_cache_set_alloc(&ca->sb); 1717 if (!c) 1718 return err; 1719 1720 err = "error creating kobject"; 1721 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) || 1722 kobject_add(&c->internal, &c->kobj, "internal")) 1723 goto err; 1724 1725 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) 1726 goto err; 1727 1728 bch_debug_init_cache_set(c); 1729 1730 list_add(&c->list, &bch_cache_sets); 1731 found: 1732 sprintf(buf, "cache%i", ca->sb.nr_this_dev); 1733 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || 1734 sysfs_create_link(&c->kobj, &ca->kobj, buf)) 1735 goto err; 1736 1737 if (ca->sb.seq > c->sb.seq) { 1738 c->sb.version = ca->sb.version; 1739 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16); 1740 c->sb.flags = ca->sb.flags; 1741 c->sb.seq = ca->sb.seq; 1742 pr_debug("set version = %llu", c->sb.version); 1743 } 1744 1745 kobject_get(&ca->kobj); 1746 ca->set = c; 1747 ca->set->cache[ca->sb.nr_this_dev] = ca; 1748 c->cache_by_alloc[c->caches_loaded++] = ca; 1749 1750 if (c->caches_loaded == c->sb.nr_in_set) 1751 run_cache_set(c); 1752 1753 return NULL; 1754 err: 1755 bch_cache_set_unregister(c); 1756 return err; 1757 } 1758 1759 /* Cache device */ 1760 1761 void bch_cache_release(struct kobject *kobj) 1762 { 1763 struct cache *ca = container_of(kobj, struct cache, kobj); 1764 unsigned i; 1765 1766 if (ca->set) { 1767 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca); 1768 ca->set->cache[ca->sb.nr_this_dev] = NULL; 1769 } 1770 1771 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca))); 1772 kfree(ca->prio_buckets); 1773 vfree(ca->buckets); 1774 1775 free_heap(&ca->heap); 1776 free_fifo(&ca->free_inc); 1777 1778 for (i = 0; i < RESERVE_NR; i++) 1779 free_fifo(&ca->free[i]); 1780 1781 if (ca->sb_bio.bi_inline_vecs[0].bv_page) 1782 put_page(ca->sb_bio.bi_io_vec[0].bv_page); 1783 1784 if (!IS_ERR_OR_NULL(ca->bdev)) 1785 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1786 1787 kfree(ca); 1788 module_put(THIS_MODULE); 1789 } 1790 1791 static int cache_alloc(struct cache_sb *sb, struct cache *ca) 1792 { 1793 size_t free; 1794 struct bucket *b; 1795 1796 __module_get(THIS_MODULE); 1797 kobject_init(&ca->kobj, &bch_cache_ktype); 1798 1799 bio_init(&ca->journal.bio); 1800 ca->journal.bio.bi_max_vecs = 8; 1801 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs; 1802 1803 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; 1804 1805 if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) || 1806 !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) || 1807 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) || 1808 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) || 1809 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) || 1810 !init_heap(&ca->heap, free << 3, GFP_KERNEL) || 1811 !(ca->buckets = vzalloc(sizeof(struct bucket) * 1812 ca->sb.nbuckets)) || 1813 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) * 1814 2, GFP_KERNEL)) || 1815 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca))) 1816 return -ENOMEM; 1817 1818 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); 1819 1820 for_each_bucket(b, ca) 1821 atomic_set(&b->pin, 0); 1822 1823 return 0; 1824 } 1825 1826 static void register_cache(struct cache_sb *sb, struct page *sb_page, 1827 struct block_device *bdev, struct cache *ca) 1828 { 1829 char name[BDEVNAME_SIZE]; 1830 const char *err = "cannot allocate memory"; 1831 1832 memcpy(&ca->sb, sb, sizeof(struct cache_sb)); 1833 ca->bdev = bdev; 1834 ca->bdev->bd_holder = ca; 1835 1836 bio_init(&ca->sb_bio); 1837 ca->sb_bio.bi_max_vecs = 1; 1838 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs; 1839 ca->sb_bio.bi_io_vec[0].bv_page = sb_page; 1840 get_page(sb_page); 1841 1842 if (blk_queue_discard(bdev_get_queue(ca->bdev))) 1843 ca->discard = CACHE_DISCARD(&ca->sb); 1844 1845 if (cache_alloc(sb, ca) != 0) 1846 goto err; 1847 1848 err = "error creating kobject"; 1849 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) 1850 goto err; 1851 1852 mutex_lock(&bch_register_lock); 1853 err = register_cache_set(ca); 1854 mutex_unlock(&bch_register_lock); 1855 1856 if (err) 1857 goto err; 1858 1859 pr_info("registered cache device %s", bdevname(bdev, name)); 1860 out: 1861 kobject_put(&ca->kobj); 1862 return; 1863 err: 1864 pr_notice("error opening %s: %s", bdevname(bdev, name), err); 1865 goto out; 1866 } 1867 1868 /* Global interfaces/init */ 1869 1870 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *, 1871 const char *, size_t); 1872 1873 kobj_attribute_write(register, register_bcache); 1874 kobj_attribute_write(register_quiet, register_bcache); 1875 1876 static bool bch_is_open_backing(struct block_device *bdev) { 1877 struct cache_set *c, *tc; 1878 struct cached_dev *dc, *t; 1879 1880 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 1881 list_for_each_entry_safe(dc, t, &c->cached_devs, list) 1882 if (dc->bdev == bdev) 1883 return true; 1884 list_for_each_entry_safe(dc, t, &uncached_devices, list) 1885 if (dc->bdev == bdev) 1886 return true; 1887 return false; 1888 } 1889 1890 static bool bch_is_open_cache(struct block_device *bdev) { 1891 struct cache_set *c, *tc; 1892 struct cache *ca; 1893 unsigned i; 1894 1895 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 1896 for_each_cache(ca, c, i) 1897 if (ca->bdev == bdev) 1898 return true; 1899 return false; 1900 } 1901 1902 static bool bch_is_open(struct block_device *bdev) { 1903 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev); 1904 } 1905 1906 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, 1907 const char *buffer, size_t size) 1908 { 1909 ssize_t ret = size; 1910 const char *err = "cannot allocate memory"; 1911 char *path = NULL; 1912 struct cache_sb *sb = NULL; 1913 struct block_device *bdev = NULL; 1914 struct page *sb_page = NULL; 1915 1916 if (!try_module_get(THIS_MODULE)) 1917 return -EBUSY; 1918 1919 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) || 1920 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL))) 1921 goto err; 1922 1923 err = "failed to open device"; 1924 bdev = blkdev_get_by_path(strim(path), 1925 FMODE_READ|FMODE_WRITE|FMODE_EXCL, 1926 sb); 1927 if (IS_ERR(bdev)) { 1928 if (bdev == ERR_PTR(-EBUSY)) { 1929 bdev = lookup_bdev(strim(path)); 1930 mutex_lock(&bch_register_lock); 1931 if (!IS_ERR(bdev) && bch_is_open(bdev)) 1932 err = "device already registered"; 1933 else 1934 err = "device busy"; 1935 mutex_unlock(&bch_register_lock); 1936 } 1937 goto err; 1938 } 1939 1940 err = "failed to set blocksize"; 1941 if (set_blocksize(bdev, 4096)) 1942 goto err_close; 1943 1944 err = read_super(sb, bdev, &sb_page); 1945 if (err) 1946 goto err_close; 1947 1948 if (SB_IS_BDEV(sb)) { 1949 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL); 1950 if (!dc) 1951 goto err_close; 1952 1953 mutex_lock(&bch_register_lock); 1954 register_bdev(sb, sb_page, bdev, dc); 1955 mutex_unlock(&bch_register_lock); 1956 } else { 1957 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL); 1958 if (!ca) 1959 goto err_close; 1960 1961 register_cache(sb, sb_page, bdev, ca); 1962 } 1963 out: 1964 if (sb_page) 1965 put_page(sb_page); 1966 kfree(sb); 1967 kfree(path); 1968 module_put(THIS_MODULE); 1969 return ret; 1970 1971 err_close: 1972 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1973 err: 1974 if (attr != &ksysfs_register_quiet) 1975 pr_info("error opening %s: %s", path, err); 1976 ret = -EINVAL; 1977 goto out; 1978 } 1979 1980 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) 1981 { 1982 if (code == SYS_DOWN || 1983 code == SYS_HALT || 1984 code == SYS_POWER_OFF) { 1985 DEFINE_WAIT(wait); 1986 unsigned long start = jiffies; 1987 bool stopped = false; 1988 1989 struct cache_set *c, *tc; 1990 struct cached_dev *dc, *tdc; 1991 1992 mutex_lock(&bch_register_lock); 1993 1994 if (list_empty(&bch_cache_sets) && 1995 list_empty(&uncached_devices)) 1996 goto out; 1997 1998 pr_info("Stopping all devices:"); 1999 2000 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2001 bch_cache_set_stop(c); 2002 2003 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) 2004 bcache_device_stop(&dc->disk); 2005 2006 /* What's a condition variable? */ 2007 while (1) { 2008 long timeout = start + 2 * HZ - jiffies; 2009 2010 stopped = list_empty(&bch_cache_sets) && 2011 list_empty(&uncached_devices); 2012 2013 if (timeout < 0 || stopped) 2014 break; 2015 2016 prepare_to_wait(&unregister_wait, &wait, 2017 TASK_UNINTERRUPTIBLE); 2018 2019 mutex_unlock(&bch_register_lock); 2020 schedule_timeout(timeout); 2021 mutex_lock(&bch_register_lock); 2022 } 2023 2024 finish_wait(&unregister_wait, &wait); 2025 2026 if (stopped) 2027 pr_info("All devices stopped"); 2028 else 2029 pr_notice("Timeout waiting for devices to be closed"); 2030 out: 2031 mutex_unlock(&bch_register_lock); 2032 } 2033 2034 return NOTIFY_DONE; 2035 } 2036 2037 static struct notifier_block reboot = { 2038 .notifier_call = bcache_reboot, 2039 .priority = INT_MAX, /* before any real devices */ 2040 }; 2041 2042 static void bcache_exit(void) 2043 { 2044 bch_debug_exit(); 2045 bch_request_exit(); 2046 if (bcache_kobj) 2047 kobject_put(bcache_kobj); 2048 if (bcache_wq) 2049 destroy_workqueue(bcache_wq); 2050 if (bcache_major) 2051 unregister_blkdev(bcache_major, "bcache"); 2052 unregister_reboot_notifier(&reboot); 2053 } 2054 2055 static int __init bcache_init(void) 2056 { 2057 static const struct attribute *files[] = { 2058 &ksysfs_register.attr, 2059 &ksysfs_register_quiet.attr, 2060 NULL 2061 }; 2062 2063 mutex_init(&bch_register_lock); 2064 init_waitqueue_head(&unregister_wait); 2065 register_reboot_notifier(&reboot); 2066 closure_debug_init(); 2067 2068 bcache_major = register_blkdev(0, "bcache"); 2069 if (bcache_major < 0) 2070 return bcache_major; 2071 2072 if (!(bcache_wq = create_workqueue("bcache")) || 2073 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) || 2074 sysfs_create_files(bcache_kobj, files) || 2075 bch_request_init() || 2076 bch_debug_init(bcache_kobj)) 2077 goto err; 2078 2079 return 0; 2080 err: 2081 bcache_exit(); 2082 return -ENOMEM; 2083 } 2084 2085 module_exit(bcache_exit); 2086 module_init(bcache_init); 2087