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 static void bio_split_pool_free(struct bio_split_pool *p) 63 { 64 if (p->bio_split_hook) 65 mempool_destroy(p->bio_split_hook); 66 67 if (p->bio_split) 68 bioset_free(p->bio_split); 69 } 70 71 static int bio_split_pool_init(struct bio_split_pool *p) 72 { 73 p->bio_split = bioset_create(4, 0); 74 if (!p->bio_split) 75 return -ENOMEM; 76 77 p->bio_split_hook = mempool_create_kmalloc_pool(4, 78 sizeof(struct bio_split_hook)); 79 if (!p->bio_split_hook) 80 return -ENOMEM; 81 82 return 0; 83 } 84 85 /* Superblock */ 86 87 static const char *read_super(struct cache_sb *sb, struct block_device *bdev, 88 struct page **res) 89 { 90 const char *err; 91 struct cache_sb *s; 92 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE); 93 unsigned i; 94 95 if (!bh) 96 return "IO error"; 97 98 s = (struct cache_sb *) bh->b_data; 99 100 sb->offset = le64_to_cpu(s->offset); 101 sb->version = le64_to_cpu(s->version); 102 103 memcpy(sb->magic, s->magic, 16); 104 memcpy(sb->uuid, s->uuid, 16); 105 memcpy(sb->set_uuid, s->set_uuid, 16); 106 memcpy(sb->label, s->label, SB_LABEL_SIZE); 107 108 sb->flags = le64_to_cpu(s->flags); 109 sb->seq = le64_to_cpu(s->seq); 110 sb->last_mount = le32_to_cpu(s->last_mount); 111 sb->first_bucket = le16_to_cpu(s->first_bucket); 112 sb->keys = le16_to_cpu(s->keys); 113 114 for (i = 0; i < SB_JOURNAL_BUCKETS; i++) 115 sb->d[i] = le64_to_cpu(s->d[i]); 116 117 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u", 118 sb->version, sb->flags, sb->seq, sb->keys); 119 120 err = "Not a bcache superblock"; 121 if (sb->offset != SB_SECTOR) 122 goto err; 123 124 if (memcmp(sb->magic, bcache_magic, 16)) 125 goto err; 126 127 err = "Too many journal buckets"; 128 if (sb->keys > SB_JOURNAL_BUCKETS) 129 goto err; 130 131 err = "Bad checksum"; 132 if (s->csum != csum_set(s)) 133 goto err; 134 135 err = "Bad UUID"; 136 if (bch_is_zero(sb->uuid, 16)) 137 goto err; 138 139 sb->block_size = le16_to_cpu(s->block_size); 140 141 err = "Superblock block size smaller than device block size"; 142 if (sb->block_size << 9 < bdev_logical_block_size(bdev)) 143 goto err; 144 145 switch (sb->version) { 146 case BCACHE_SB_VERSION_BDEV: 147 sb->data_offset = BDEV_DATA_START_DEFAULT; 148 break; 149 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET: 150 sb->data_offset = le64_to_cpu(s->data_offset); 151 152 err = "Bad data offset"; 153 if (sb->data_offset < BDEV_DATA_START_DEFAULT) 154 goto err; 155 156 break; 157 case BCACHE_SB_VERSION_CDEV: 158 case BCACHE_SB_VERSION_CDEV_WITH_UUID: 159 sb->nbuckets = le64_to_cpu(s->nbuckets); 160 sb->block_size = le16_to_cpu(s->block_size); 161 sb->bucket_size = le16_to_cpu(s->bucket_size); 162 163 sb->nr_in_set = le16_to_cpu(s->nr_in_set); 164 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev); 165 166 err = "Too many buckets"; 167 if (sb->nbuckets > LONG_MAX) 168 goto err; 169 170 err = "Not enough buckets"; 171 if (sb->nbuckets < 1 << 7) 172 goto err; 173 174 err = "Bad block/bucket size"; 175 if (!is_power_of_2(sb->block_size) || 176 sb->block_size > PAGE_SECTORS || 177 !is_power_of_2(sb->bucket_size) || 178 sb->bucket_size < PAGE_SECTORS) 179 goto err; 180 181 err = "Invalid superblock: device too small"; 182 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets) 183 goto err; 184 185 err = "Bad UUID"; 186 if (bch_is_zero(sb->set_uuid, 16)) 187 goto err; 188 189 err = "Bad cache device number in set"; 190 if (!sb->nr_in_set || 191 sb->nr_in_set <= sb->nr_this_dev || 192 sb->nr_in_set > MAX_CACHES_PER_SET) 193 goto err; 194 195 err = "Journal buckets not sequential"; 196 for (i = 0; i < sb->keys; i++) 197 if (sb->d[i] != sb->first_bucket + i) 198 goto err; 199 200 err = "Too many journal buckets"; 201 if (sb->first_bucket + sb->keys > sb->nbuckets) 202 goto err; 203 204 err = "Invalid superblock: first bucket comes before end of super"; 205 if (sb->first_bucket * sb->bucket_size < 16) 206 goto err; 207 208 break; 209 default: 210 err = "Unsupported superblock version"; 211 goto err; 212 } 213 214 sb->last_mount = get_seconds(); 215 err = NULL; 216 217 get_page(bh->b_page); 218 *res = bh->b_page; 219 err: 220 put_bh(bh); 221 return err; 222 } 223 224 static void write_bdev_super_endio(struct bio *bio, int error) 225 { 226 struct cached_dev *dc = bio->bi_private; 227 /* XXX: error checking */ 228 229 closure_put(&dc->sb_write); 230 } 231 232 static void __write_super(struct cache_sb *sb, struct bio *bio) 233 { 234 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page); 235 unsigned i; 236 237 bio->bi_iter.bi_sector = SB_SECTOR; 238 bio->bi_rw = REQ_SYNC|REQ_META; 239 bio->bi_iter.bi_size = SB_SIZE; 240 bch_bio_map(bio, NULL); 241 242 out->offset = cpu_to_le64(sb->offset); 243 out->version = cpu_to_le64(sb->version); 244 245 memcpy(out->uuid, sb->uuid, 16); 246 memcpy(out->set_uuid, sb->set_uuid, 16); 247 memcpy(out->label, sb->label, SB_LABEL_SIZE); 248 249 out->flags = cpu_to_le64(sb->flags); 250 out->seq = cpu_to_le64(sb->seq); 251 252 out->last_mount = cpu_to_le32(sb->last_mount); 253 out->first_bucket = cpu_to_le16(sb->first_bucket); 254 out->keys = cpu_to_le16(sb->keys); 255 256 for (i = 0; i < sb->keys; i++) 257 out->d[i] = cpu_to_le64(sb->d[i]); 258 259 out->csum = csum_set(out); 260 261 pr_debug("ver %llu, flags %llu, seq %llu", 262 sb->version, sb->flags, sb->seq); 263 264 submit_bio(REQ_WRITE, bio); 265 } 266 267 static void bch_write_bdev_super_unlock(struct closure *cl) 268 { 269 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write); 270 271 up(&dc->sb_write_mutex); 272 } 273 274 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent) 275 { 276 struct closure *cl = &dc->sb_write; 277 struct bio *bio = &dc->sb_bio; 278 279 down(&dc->sb_write_mutex); 280 closure_init(cl, parent); 281 282 bio_reset(bio); 283 bio->bi_bdev = dc->bdev; 284 bio->bi_end_io = write_bdev_super_endio; 285 bio->bi_private = dc; 286 287 closure_get(cl); 288 __write_super(&dc->sb, bio); 289 290 closure_return_with_destructor(cl, bch_write_bdev_super_unlock); 291 } 292 293 static void write_super_endio(struct bio *bio, int error) 294 { 295 struct cache *ca = bio->bi_private; 296 297 bch_count_io_errors(ca, error, "writing superblock"); 298 closure_put(&ca->set->sb_write); 299 } 300 301 static void bcache_write_super_unlock(struct closure *cl) 302 { 303 struct cache_set *c = container_of(cl, struct cache_set, sb_write); 304 305 up(&c->sb_write_mutex); 306 } 307 308 void bcache_write_super(struct cache_set *c) 309 { 310 struct closure *cl = &c->sb_write; 311 struct cache *ca; 312 unsigned i; 313 314 down(&c->sb_write_mutex); 315 closure_init(cl, &c->cl); 316 317 c->sb.seq++; 318 319 for_each_cache(ca, c, i) { 320 struct bio *bio = &ca->sb_bio; 321 322 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID; 323 ca->sb.seq = c->sb.seq; 324 ca->sb.last_mount = c->sb.last_mount; 325 326 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb)); 327 328 bio_reset(bio); 329 bio->bi_bdev = ca->bdev; 330 bio->bi_end_io = write_super_endio; 331 bio->bi_private = ca; 332 333 closure_get(cl); 334 __write_super(&ca->sb, bio); 335 } 336 337 closure_return_with_destructor(cl, bcache_write_super_unlock); 338 } 339 340 /* UUID io */ 341 342 static void uuid_endio(struct bio *bio, int error) 343 { 344 struct closure *cl = bio->bi_private; 345 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 346 347 cache_set_err_on(error, c, "accessing uuids"); 348 bch_bbio_free(bio, c); 349 closure_put(cl); 350 } 351 352 static void uuid_io_unlock(struct closure *cl) 353 { 354 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 355 356 up(&c->uuid_write_mutex); 357 } 358 359 static void uuid_io(struct cache_set *c, unsigned long rw, 360 struct bkey *k, struct closure *parent) 361 { 362 struct closure *cl = &c->uuid_write; 363 struct uuid_entry *u; 364 unsigned i; 365 char buf[80]; 366 367 BUG_ON(!parent); 368 down(&c->uuid_write_mutex); 369 closure_init(cl, parent); 370 371 for (i = 0; i < KEY_PTRS(k); i++) { 372 struct bio *bio = bch_bbio_alloc(c); 373 374 bio->bi_rw = REQ_SYNC|REQ_META|rw; 375 bio->bi_iter.bi_size = KEY_SIZE(k) << 9; 376 377 bio->bi_end_io = uuid_endio; 378 bio->bi_private = cl; 379 bch_bio_map(bio, c->uuids); 380 381 bch_submit_bbio(bio, c, k, i); 382 383 if (!(rw & WRITE)) 384 break; 385 } 386 387 bch_extent_to_text(buf, sizeof(buf), k); 388 pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read", buf); 389 390 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++) 391 if (!bch_is_zero(u->uuid, 16)) 392 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u", 393 u - c->uuids, u->uuid, u->label, 394 u->first_reg, u->last_reg, u->invalidated); 395 396 closure_return_with_destructor(cl, uuid_io_unlock); 397 } 398 399 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl) 400 { 401 struct bkey *k = &j->uuid_bucket; 402 403 if (__bch_btree_ptr_invalid(c, k)) 404 return "bad uuid pointer"; 405 406 bkey_copy(&c->uuid_bucket, k); 407 uuid_io(c, READ_SYNC, k, cl); 408 409 if (j->version < BCACHE_JSET_VERSION_UUIDv1) { 410 struct uuid_entry_v0 *u0 = (void *) c->uuids; 411 struct uuid_entry *u1 = (void *) c->uuids; 412 int i; 413 414 closure_sync(cl); 415 416 /* 417 * Since the new uuid entry is bigger than the old, we have to 418 * convert starting at the highest memory address and work down 419 * in order to do it in place 420 */ 421 422 for (i = c->nr_uuids - 1; 423 i >= 0; 424 --i) { 425 memcpy(u1[i].uuid, u0[i].uuid, 16); 426 memcpy(u1[i].label, u0[i].label, 32); 427 428 u1[i].first_reg = u0[i].first_reg; 429 u1[i].last_reg = u0[i].last_reg; 430 u1[i].invalidated = u0[i].invalidated; 431 432 u1[i].flags = 0; 433 u1[i].sectors = 0; 434 } 435 } 436 437 return NULL; 438 } 439 440 static int __uuid_write(struct cache_set *c) 441 { 442 BKEY_PADDED(key) k; 443 struct closure cl; 444 closure_init_stack(&cl); 445 446 lockdep_assert_held(&bch_register_lock); 447 448 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true)) 449 return 1; 450 451 SET_KEY_SIZE(&k.key, c->sb.bucket_size); 452 uuid_io(c, REQ_WRITE, &k.key, &cl); 453 closure_sync(&cl); 454 455 bkey_copy(&c->uuid_bucket, &k.key); 456 bkey_put(c, &k.key); 457 return 0; 458 } 459 460 int bch_uuid_write(struct cache_set *c) 461 { 462 int ret = __uuid_write(c); 463 464 if (!ret) 465 bch_journal_meta(c, NULL); 466 467 return ret; 468 } 469 470 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid) 471 { 472 struct uuid_entry *u; 473 474 for (u = c->uuids; 475 u < c->uuids + c->nr_uuids; u++) 476 if (!memcmp(u->uuid, uuid, 16)) 477 return u; 478 479 return NULL; 480 } 481 482 static struct uuid_entry *uuid_find_empty(struct cache_set *c) 483 { 484 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"; 485 return uuid_find(c, zero_uuid); 486 } 487 488 /* 489 * Bucket priorities/gens: 490 * 491 * For each bucket, we store on disk its 492 * 8 bit gen 493 * 16 bit priority 494 * 495 * See alloc.c for an explanation of the gen. The priority is used to implement 496 * lru (and in the future other) cache replacement policies; for most purposes 497 * it's just an opaque integer. 498 * 499 * The gens and the priorities don't have a whole lot to do with each other, and 500 * it's actually the gens that must be written out at specific times - it's no 501 * big deal if the priorities don't get written, if we lose them we just reuse 502 * buckets in suboptimal order. 503 * 504 * On disk they're stored in a packed array, and in as many buckets are required 505 * to fit them all. The buckets we use to store them form a list; the journal 506 * header points to the first bucket, the first bucket points to the second 507 * bucket, et cetera. 508 * 509 * This code is used by the allocation code; periodically (whenever it runs out 510 * of buckets to allocate from) the allocation code will invalidate some 511 * buckets, but it can't use those buckets until their new gens are safely on 512 * disk. 513 */ 514 515 static void prio_endio(struct bio *bio, int error) 516 { 517 struct cache *ca = bio->bi_private; 518 519 cache_set_err_on(error, ca->set, "accessing priorities"); 520 bch_bbio_free(bio, ca->set); 521 closure_put(&ca->prio); 522 } 523 524 static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw) 525 { 526 struct closure *cl = &ca->prio; 527 struct bio *bio = bch_bbio_alloc(ca->set); 528 529 closure_init_stack(cl); 530 531 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size; 532 bio->bi_bdev = ca->bdev; 533 bio->bi_rw = REQ_SYNC|REQ_META|rw; 534 bio->bi_iter.bi_size = bucket_bytes(ca); 535 536 bio->bi_end_io = prio_endio; 537 bio->bi_private = ca; 538 bch_bio_map(bio, ca->disk_buckets); 539 540 closure_bio_submit(bio, &ca->prio, ca); 541 closure_sync(cl); 542 } 543 544 void bch_prio_write(struct cache *ca) 545 { 546 int i; 547 struct bucket *b; 548 struct closure cl; 549 550 closure_init_stack(&cl); 551 552 lockdep_assert_held(&ca->set->bucket_lock); 553 554 ca->disk_buckets->seq++; 555 556 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca), 557 &ca->meta_sectors_written); 558 559 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free), 560 // fifo_used(&ca->free_inc), fifo_used(&ca->unused)); 561 562 for (i = prio_buckets(ca) - 1; i >= 0; --i) { 563 long bucket; 564 struct prio_set *p = ca->disk_buckets; 565 struct bucket_disk *d = p->data; 566 struct bucket_disk *end = d + prios_per_bucket(ca); 567 568 for (b = ca->buckets + i * prios_per_bucket(ca); 569 b < ca->buckets + ca->sb.nbuckets && d < end; 570 b++, d++) { 571 d->prio = cpu_to_le16(b->prio); 572 d->gen = b->gen; 573 } 574 575 p->next_bucket = ca->prio_buckets[i + 1]; 576 p->magic = pset_magic(&ca->sb); 577 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8); 578 579 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true); 580 BUG_ON(bucket == -1); 581 582 mutex_unlock(&ca->set->bucket_lock); 583 prio_io(ca, bucket, REQ_WRITE); 584 mutex_lock(&ca->set->bucket_lock); 585 586 ca->prio_buckets[i] = bucket; 587 atomic_dec_bug(&ca->buckets[bucket].pin); 588 } 589 590 mutex_unlock(&ca->set->bucket_lock); 591 592 bch_journal_meta(ca->set, &cl); 593 closure_sync(&cl); 594 595 mutex_lock(&ca->set->bucket_lock); 596 597 /* 598 * Don't want the old priorities to get garbage collected until after we 599 * finish writing the new ones, and they're journalled 600 */ 601 for (i = 0; i < prio_buckets(ca); i++) { 602 if (ca->prio_last_buckets[i]) 603 __bch_bucket_free(ca, 604 &ca->buckets[ca->prio_last_buckets[i]]); 605 606 ca->prio_last_buckets[i] = ca->prio_buckets[i]; 607 } 608 } 609 610 static void prio_read(struct cache *ca, uint64_t bucket) 611 { 612 struct prio_set *p = ca->disk_buckets; 613 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d; 614 struct bucket *b; 615 unsigned bucket_nr = 0; 616 617 for (b = ca->buckets; 618 b < ca->buckets + ca->sb.nbuckets; 619 b++, d++) { 620 if (d == end) { 621 ca->prio_buckets[bucket_nr] = bucket; 622 ca->prio_last_buckets[bucket_nr] = bucket; 623 bucket_nr++; 624 625 prio_io(ca, bucket, READ_SYNC); 626 627 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8)) 628 pr_warn("bad csum reading priorities"); 629 630 if (p->magic != pset_magic(&ca->sb)) 631 pr_warn("bad magic reading priorities"); 632 633 bucket = p->next_bucket; 634 d = p->data; 635 } 636 637 b->prio = le16_to_cpu(d->prio); 638 b->gen = b->last_gc = d->gen; 639 } 640 } 641 642 /* Bcache device */ 643 644 static int open_dev(struct block_device *b, fmode_t mode) 645 { 646 struct bcache_device *d = b->bd_disk->private_data; 647 if (test_bit(BCACHE_DEV_CLOSING, &d->flags)) 648 return -ENXIO; 649 650 closure_get(&d->cl); 651 return 0; 652 } 653 654 static void release_dev(struct gendisk *b, fmode_t mode) 655 { 656 struct bcache_device *d = b->private_data; 657 closure_put(&d->cl); 658 } 659 660 static int ioctl_dev(struct block_device *b, fmode_t mode, 661 unsigned int cmd, unsigned long arg) 662 { 663 struct bcache_device *d = b->bd_disk->private_data; 664 return d->ioctl(d, mode, cmd, arg); 665 } 666 667 static const struct block_device_operations bcache_ops = { 668 .open = open_dev, 669 .release = release_dev, 670 .ioctl = ioctl_dev, 671 .owner = THIS_MODULE, 672 }; 673 674 void bcache_device_stop(struct bcache_device *d) 675 { 676 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags)) 677 closure_queue(&d->cl); 678 } 679 680 static void bcache_device_unlink(struct bcache_device *d) 681 { 682 lockdep_assert_held(&bch_register_lock); 683 684 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) { 685 unsigned i; 686 struct cache *ca; 687 688 sysfs_remove_link(&d->c->kobj, d->name); 689 sysfs_remove_link(&d->kobj, "cache"); 690 691 for_each_cache(ca, d->c, i) 692 bd_unlink_disk_holder(ca->bdev, d->disk); 693 } 694 } 695 696 static void bcache_device_link(struct bcache_device *d, struct cache_set *c, 697 const char *name) 698 { 699 unsigned i; 700 struct cache *ca; 701 702 for_each_cache(ca, d->c, i) 703 bd_link_disk_holder(ca->bdev, d->disk); 704 705 snprintf(d->name, BCACHEDEVNAME_SIZE, 706 "%s%u", name, d->id); 707 708 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") || 709 sysfs_create_link(&c->kobj, &d->kobj, d->name), 710 "Couldn't create device <-> cache set symlinks"); 711 } 712 713 static void bcache_device_detach(struct bcache_device *d) 714 { 715 lockdep_assert_held(&bch_register_lock); 716 717 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) { 718 struct uuid_entry *u = d->c->uuids + d->id; 719 720 SET_UUID_FLASH_ONLY(u, 0); 721 memcpy(u->uuid, invalid_uuid, 16); 722 u->invalidated = cpu_to_le32(get_seconds()); 723 bch_uuid_write(d->c); 724 } 725 726 bcache_device_unlink(d); 727 728 d->c->devices[d->id] = NULL; 729 closure_put(&d->c->caching); 730 d->c = NULL; 731 } 732 733 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, 734 unsigned id) 735 { 736 d->id = id; 737 d->c = c; 738 c->devices[id] = d; 739 740 closure_get(&c->caching); 741 } 742 743 static void bcache_device_free(struct bcache_device *d) 744 { 745 lockdep_assert_held(&bch_register_lock); 746 747 pr_info("%s stopped", d->disk->disk_name); 748 749 if (d->c) 750 bcache_device_detach(d); 751 if (d->disk && d->disk->flags & GENHD_FL_UP) 752 del_gendisk(d->disk); 753 if (d->disk && d->disk->queue) 754 blk_cleanup_queue(d->disk->queue); 755 if (d->disk) { 756 ida_simple_remove(&bcache_minor, d->disk->first_minor); 757 put_disk(d->disk); 758 } 759 760 bio_split_pool_free(&d->bio_split_hook); 761 if (d->bio_split) 762 bioset_free(d->bio_split); 763 if (is_vmalloc_addr(d->full_dirty_stripes)) 764 vfree(d->full_dirty_stripes); 765 else 766 kfree(d->full_dirty_stripes); 767 if (is_vmalloc_addr(d->stripe_sectors_dirty)) 768 vfree(d->stripe_sectors_dirty); 769 else 770 kfree(d->stripe_sectors_dirty); 771 772 closure_debug_destroy(&d->cl); 773 } 774 775 static int bcache_device_init(struct bcache_device *d, unsigned block_size, 776 sector_t sectors) 777 { 778 struct request_queue *q; 779 size_t n; 780 int minor; 781 782 if (!d->stripe_size) 783 d->stripe_size = 1 << 31; 784 785 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size); 786 787 if (!d->nr_stripes || 788 d->nr_stripes > INT_MAX || 789 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) { 790 pr_err("nr_stripes too large"); 791 return -ENOMEM; 792 } 793 794 n = d->nr_stripes * sizeof(atomic_t); 795 d->stripe_sectors_dirty = n < PAGE_SIZE << 6 796 ? kzalloc(n, GFP_KERNEL) 797 : vzalloc(n); 798 if (!d->stripe_sectors_dirty) 799 return -ENOMEM; 800 801 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long); 802 d->full_dirty_stripes = n < PAGE_SIZE << 6 803 ? kzalloc(n, GFP_KERNEL) 804 : vzalloc(n); 805 if (!d->full_dirty_stripes) 806 return -ENOMEM; 807 808 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL); 809 if (minor < 0) 810 return minor; 811 812 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || 813 bio_split_pool_init(&d->bio_split_hook) || 814 !(d->disk = alloc_disk(1))) { 815 ida_simple_remove(&bcache_minor, minor); 816 return -ENOMEM; 817 } 818 819 set_capacity(d->disk, sectors); 820 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor); 821 822 d->disk->major = bcache_major; 823 d->disk->first_minor = minor; 824 d->disk->fops = &bcache_ops; 825 d->disk->private_data = d; 826 827 q = blk_alloc_queue(GFP_KERNEL); 828 if (!q) 829 return -ENOMEM; 830 831 blk_queue_make_request(q, NULL); 832 d->disk->queue = q; 833 q->queuedata = d; 834 q->backing_dev_info.congested_data = d; 835 q->limits.max_hw_sectors = UINT_MAX; 836 q->limits.max_sectors = UINT_MAX; 837 q->limits.max_segment_size = UINT_MAX; 838 q->limits.max_segments = BIO_MAX_PAGES; 839 q->limits.max_discard_sectors = UINT_MAX; 840 q->limits.discard_granularity = 512; 841 q->limits.io_min = block_size; 842 q->limits.logical_block_size = block_size; 843 q->limits.physical_block_size = block_size; 844 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags); 845 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags); 846 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags); 847 848 blk_queue_flush(q, REQ_FLUSH|REQ_FUA); 849 850 return 0; 851 } 852 853 /* Cached device */ 854 855 static void calc_cached_dev_sectors(struct cache_set *c) 856 { 857 uint64_t sectors = 0; 858 struct cached_dev *dc; 859 860 list_for_each_entry(dc, &c->cached_devs, list) 861 sectors += bdev_sectors(dc->bdev); 862 863 c->cached_dev_sectors = sectors; 864 } 865 866 void bch_cached_dev_run(struct cached_dev *dc) 867 { 868 struct bcache_device *d = &dc->disk; 869 char buf[SB_LABEL_SIZE + 1]; 870 char *env[] = { 871 "DRIVER=bcache", 872 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), 873 NULL, 874 NULL, 875 }; 876 877 memcpy(buf, dc->sb.label, SB_LABEL_SIZE); 878 buf[SB_LABEL_SIZE] = '\0'; 879 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf); 880 881 if (atomic_xchg(&dc->running, 1)) 882 return; 883 884 if (!d->c && 885 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { 886 struct closure cl; 887 closure_init_stack(&cl); 888 889 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); 890 bch_write_bdev_super(dc, &cl); 891 closure_sync(&cl); 892 } 893 894 add_disk(d->disk); 895 bd_link_disk_holder(dc->bdev, dc->disk.disk); 896 /* won't show up in the uevent file, use udevadm monitor -e instead 897 * only class / kset properties are persistent */ 898 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); 899 kfree(env[1]); 900 kfree(env[2]); 901 902 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || 903 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache")) 904 pr_debug("error creating sysfs link"); 905 } 906 907 static void cached_dev_detach_finish(struct work_struct *w) 908 { 909 struct cached_dev *dc = container_of(w, struct cached_dev, detach); 910 char buf[BDEVNAME_SIZE]; 911 struct closure cl; 912 closure_init_stack(&cl); 913 914 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)); 915 BUG_ON(atomic_read(&dc->count)); 916 917 mutex_lock(&bch_register_lock); 918 919 memset(&dc->sb.set_uuid, 0, 16); 920 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE); 921 922 bch_write_bdev_super(dc, &cl); 923 closure_sync(&cl); 924 925 bcache_device_detach(&dc->disk); 926 list_move(&dc->list, &uncached_devices); 927 928 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags); 929 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags); 930 931 mutex_unlock(&bch_register_lock); 932 933 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf)); 934 935 /* Drop ref we took in cached_dev_detach() */ 936 closure_put(&dc->disk.cl); 937 } 938 939 void bch_cached_dev_detach(struct cached_dev *dc) 940 { 941 lockdep_assert_held(&bch_register_lock); 942 943 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) 944 return; 945 946 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) 947 return; 948 949 /* 950 * Block the device from being closed and freed until we're finished 951 * detaching 952 */ 953 closure_get(&dc->disk.cl); 954 955 bch_writeback_queue(dc); 956 cached_dev_put(dc); 957 } 958 959 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c) 960 { 961 uint32_t rtime = cpu_to_le32(get_seconds()); 962 struct uuid_entry *u; 963 char buf[BDEVNAME_SIZE]; 964 965 bdevname(dc->bdev, buf); 966 967 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)) 968 return -ENOENT; 969 970 if (dc->disk.c) { 971 pr_err("Can't attach %s: already attached", buf); 972 return -EINVAL; 973 } 974 975 if (test_bit(CACHE_SET_STOPPING, &c->flags)) { 976 pr_err("Can't attach %s: shutting down", buf); 977 return -EINVAL; 978 } 979 980 if (dc->sb.block_size < c->sb.block_size) { 981 /* Will die */ 982 pr_err("Couldn't attach %s: block size less than set's block size", 983 buf); 984 return -EINVAL; 985 } 986 987 u = uuid_find(c, dc->sb.uuid); 988 989 if (u && 990 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || 991 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { 992 memcpy(u->uuid, invalid_uuid, 16); 993 u->invalidated = cpu_to_le32(get_seconds()); 994 u = NULL; 995 } 996 997 if (!u) { 998 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 999 pr_err("Couldn't find uuid for %s in set", buf); 1000 return -ENOENT; 1001 } 1002 1003 u = uuid_find_empty(c); 1004 if (!u) { 1005 pr_err("Not caching %s, no room for UUID", buf); 1006 return -EINVAL; 1007 } 1008 } 1009 1010 /* Deadlocks since we're called via sysfs... 1011 sysfs_remove_file(&dc->kobj, &sysfs_attach); 1012 */ 1013 1014 if (bch_is_zero(u->uuid, 16)) { 1015 struct closure cl; 1016 closure_init_stack(&cl); 1017 1018 memcpy(u->uuid, dc->sb.uuid, 16); 1019 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); 1020 u->first_reg = u->last_reg = rtime; 1021 bch_uuid_write(c); 1022 1023 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16); 1024 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); 1025 1026 bch_write_bdev_super(dc, &cl); 1027 closure_sync(&cl); 1028 } else { 1029 u->last_reg = rtime; 1030 bch_uuid_write(c); 1031 } 1032 1033 bcache_device_attach(&dc->disk, c, u - c->uuids); 1034 list_move(&dc->list, &c->cached_devs); 1035 calc_cached_dev_sectors(c); 1036 1037 smp_wmb(); 1038 /* 1039 * dc->c must be set before dc->count != 0 - paired with the mb in 1040 * cached_dev_get() 1041 */ 1042 atomic_set(&dc->count, 1); 1043 1044 if (bch_cached_dev_writeback_start(dc)) 1045 return -ENOMEM; 1046 1047 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1048 bch_sectors_dirty_init(dc); 1049 atomic_set(&dc->has_dirty, 1); 1050 atomic_inc(&dc->count); 1051 bch_writeback_queue(dc); 1052 } 1053 1054 bch_cached_dev_run(dc); 1055 bcache_device_link(&dc->disk, c, "bdev"); 1056 1057 pr_info("Caching %s as %s on set %pU", 1058 bdevname(dc->bdev, buf), dc->disk.disk->disk_name, 1059 dc->disk.c->sb.set_uuid); 1060 return 0; 1061 } 1062 1063 void bch_cached_dev_release(struct kobject *kobj) 1064 { 1065 struct cached_dev *dc = container_of(kobj, struct cached_dev, 1066 disk.kobj); 1067 kfree(dc); 1068 module_put(THIS_MODULE); 1069 } 1070 1071 static void cached_dev_free(struct closure *cl) 1072 { 1073 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1074 1075 cancel_delayed_work_sync(&dc->writeback_rate_update); 1076 if (!IS_ERR_OR_NULL(dc->writeback_thread)) 1077 kthread_stop(dc->writeback_thread); 1078 1079 mutex_lock(&bch_register_lock); 1080 1081 if (atomic_read(&dc->running)) 1082 bd_unlink_disk_holder(dc->bdev, dc->disk.disk); 1083 bcache_device_free(&dc->disk); 1084 list_del(&dc->list); 1085 1086 mutex_unlock(&bch_register_lock); 1087 1088 if (!IS_ERR_OR_NULL(dc->bdev)) 1089 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1090 1091 wake_up(&unregister_wait); 1092 1093 kobject_put(&dc->disk.kobj); 1094 } 1095 1096 static void cached_dev_flush(struct closure *cl) 1097 { 1098 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1099 struct bcache_device *d = &dc->disk; 1100 1101 mutex_lock(&bch_register_lock); 1102 bcache_device_unlink(d); 1103 mutex_unlock(&bch_register_lock); 1104 1105 bch_cache_accounting_destroy(&dc->accounting); 1106 kobject_del(&d->kobj); 1107 1108 continue_at(cl, cached_dev_free, system_wq); 1109 } 1110 1111 static int cached_dev_init(struct cached_dev *dc, unsigned block_size) 1112 { 1113 int ret; 1114 struct io *io; 1115 struct request_queue *q = bdev_get_queue(dc->bdev); 1116 1117 __module_get(THIS_MODULE); 1118 INIT_LIST_HEAD(&dc->list); 1119 closure_init(&dc->disk.cl, NULL); 1120 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); 1121 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); 1122 INIT_WORK(&dc->detach, cached_dev_detach_finish); 1123 sema_init(&dc->sb_write_mutex, 1); 1124 INIT_LIST_HEAD(&dc->io_lru); 1125 spin_lock_init(&dc->io_lock); 1126 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); 1127 1128 dc->sequential_cutoff = 4 << 20; 1129 1130 for (io = dc->io; io < dc->io + RECENT_IO; io++) { 1131 list_add(&io->lru, &dc->io_lru); 1132 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); 1133 } 1134 1135 dc->disk.stripe_size = q->limits.io_opt >> 9; 1136 1137 if (dc->disk.stripe_size) 1138 dc->partial_stripes_expensive = 1139 q->limits.raid_partial_stripes_expensive; 1140 1141 ret = bcache_device_init(&dc->disk, block_size, 1142 dc->bdev->bd_part->nr_sects - dc->sb.data_offset); 1143 if (ret) 1144 return ret; 1145 1146 set_capacity(dc->disk.disk, 1147 dc->bdev->bd_part->nr_sects - dc->sb.data_offset); 1148 1149 dc->disk.disk->queue->backing_dev_info.ra_pages = 1150 max(dc->disk.disk->queue->backing_dev_info.ra_pages, 1151 q->backing_dev_info.ra_pages); 1152 1153 bch_cached_dev_request_init(dc); 1154 bch_cached_dev_writeback_init(dc); 1155 return 0; 1156 } 1157 1158 /* Cached device - bcache superblock */ 1159 1160 static void register_bdev(struct cache_sb *sb, struct page *sb_page, 1161 struct block_device *bdev, 1162 struct cached_dev *dc) 1163 { 1164 char name[BDEVNAME_SIZE]; 1165 const char *err = "cannot allocate memory"; 1166 struct cache_set *c; 1167 1168 memcpy(&dc->sb, sb, sizeof(struct cache_sb)); 1169 dc->bdev = bdev; 1170 dc->bdev->bd_holder = dc; 1171 1172 bio_init(&dc->sb_bio); 1173 dc->sb_bio.bi_max_vecs = 1; 1174 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs; 1175 dc->sb_bio.bi_io_vec[0].bv_page = sb_page; 1176 get_page(sb_page); 1177 1178 if (cached_dev_init(dc, sb->block_size << 9)) 1179 goto err; 1180 1181 err = "error creating kobject"; 1182 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj, 1183 "bcache")) 1184 goto err; 1185 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) 1186 goto err; 1187 1188 pr_info("registered backing device %s", bdevname(bdev, name)); 1189 1190 list_add(&dc->list, &uncached_devices); 1191 list_for_each_entry(c, &bch_cache_sets, list) 1192 bch_cached_dev_attach(dc, c); 1193 1194 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || 1195 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) 1196 bch_cached_dev_run(dc); 1197 1198 return; 1199 err: 1200 pr_notice("error opening %s: %s", bdevname(bdev, name), err); 1201 bcache_device_stop(&dc->disk); 1202 } 1203 1204 /* Flash only volumes */ 1205 1206 void bch_flash_dev_release(struct kobject *kobj) 1207 { 1208 struct bcache_device *d = container_of(kobj, struct bcache_device, 1209 kobj); 1210 kfree(d); 1211 } 1212 1213 static void flash_dev_free(struct closure *cl) 1214 { 1215 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1216 mutex_lock(&bch_register_lock); 1217 bcache_device_free(d); 1218 mutex_unlock(&bch_register_lock); 1219 kobject_put(&d->kobj); 1220 } 1221 1222 static void flash_dev_flush(struct closure *cl) 1223 { 1224 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1225 1226 mutex_lock(&bch_register_lock); 1227 bcache_device_unlink(d); 1228 mutex_unlock(&bch_register_lock); 1229 kobject_del(&d->kobj); 1230 continue_at(cl, flash_dev_free, system_wq); 1231 } 1232 1233 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) 1234 { 1235 struct bcache_device *d = kzalloc(sizeof(struct bcache_device), 1236 GFP_KERNEL); 1237 if (!d) 1238 return -ENOMEM; 1239 1240 closure_init(&d->cl, NULL); 1241 set_closure_fn(&d->cl, flash_dev_flush, system_wq); 1242 1243 kobject_init(&d->kobj, &bch_flash_dev_ktype); 1244 1245 if (bcache_device_init(d, block_bytes(c), u->sectors)) 1246 goto err; 1247 1248 bcache_device_attach(d, c, u - c->uuids); 1249 bch_flash_dev_request_init(d); 1250 add_disk(d->disk); 1251 1252 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache")) 1253 goto err; 1254 1255 bcache_device_link(d, c, "volume"); 1256 1257 return 0; 1258 err: 1259 kobject_put(&d->kobj); 1260 return -ENOMEM; 1261 } 1262 1263 static int flash_devs_run(struct cache_set *c) 1264 { 1265 int ret = 0; 1266 struct uuid_entry *u; 1267 1268 for (u = c->uuids; 1269 u < c->uuids + c->nr_uuids && !ret; 1270 u++) 1271 if (UUID_FLASH_ONLY(u)) 1272 ret = flash_dev_run(c, u); 1273 1274 return ret; 1275 } 1276 1277 int bch_flash_dev_create(struct cache_set *c, uint64_t size) 1278 { 1279 struct uuid_entry *u; 1280 1281 if (test_bit(CACHE_SET_STOPPING, &c->flags)) 1282 return -EINTR; 1283 1284 if (!test_bit(CACHE_SET_RUNNING, &c->flags)) 1285 return -EPERM; 1286 1287 u = uuid_find_empty(c); 1288 if (!u) { 1289 pr_err("Can't create volume, no room for UUID"); 1290 return -EINVAL; 1291 } 1292 1293 get_random_bytes(u->uuid, 16); 1294 memset(u->label, 0, 32); 1295 u->first_reg = u->last_reg = cpu_to_le32(get_seconds()); 1296 1297 SET_UUID_FLASH_ONLY(u, 1); 1298 u->sectors = size >> 9; 1299 1300 bch_uuid_write(c); 1301 1302 return flash_dev_run(c, u); 1303 } 1304 1305 /* Cache set */ 1306 1307 __printf(2, 3) 1308 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) 1309 { 1310 va_list args; 1311 1312 if (c->on_error != ON_ERROR_PANIC && 1313 test_bit(CACHE_SET_STOPPING, &c->flags)) 1314 return false; 1315 1316 /* XXX: we can be called from atomic context 1317 acquire_console_sem(); 1318 */ 1319 1320 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid); 1321 1322 va_start(args, fmt); 1323 vprintk(fmt, args); 1324 va_end(args); 1325 1326 printk(", disabling caching\n"); 1327 1328 if (c->on_error == ON_ERROR_PANIC) 1329 panic("panic forced after error\n"); 1330 1331 bch_cache_set_unregister(c); 1332 return true; 1333 } 1334 1335 void bch_cache_set_release(struct kobject *kobj) 1336 { 1337 struct cache_set *c = container_of(kobj, struct cache_set, kobj); 1338 kfree(c); 1339 module_put(THIS_MODULE); 1340 } 1341 1342 static void cache_set_free(struct closure *cl) 1343 { 1344 struct cache_set *c = container_of(cl, struct cache_set, cl); 1345 struct cache *ca; 1346 unsigned i; 1347 1348 if (!IS_ERR_OR_NULL(c->debug)) 1349 debugfs_remove(c->debug); 1350 1351 bch_open_buckets_free(c); 1352 bch_btree_cache_free(c); 1353 bch_journal_free(c); 1354 1355 for_each_cache(ca, c, i) 1356 if (ca) { 1357 ca->set = NULL; 1358 c->cache[ca->sb.nr_this_dev] = NULL; 1359 kobject_put(&ca->kobj); 1360 } 1361 1362 bch_bset_sort_state_free(&c->sort); 1363 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c))); 1364 1365 if (c->moving_gc_wq) 1366 destroy_workqueue(c->moving_gc_wq); 1367 if (c->bio_split) 1368 bioset_free(c->bio_split); 1369 if (c->fill_iter) 1370 mempool_destroy(c->fill_iter); 1371 if (c->bio_meta) 1372 mempool_destroy(c->bio_meta); 1373 if (c->search) 1374 mempool_destroy(c->search); 1375 kfree(c->devices); 1376 1377 mutex_lock(&bch_register_lock); 1378 list_del(&c->list); 1379 mutex_unlock(&bch_register_lock); 1380 1381 pr_info("Cache set %pU unregistered", c->sb.set_uuid); 1382 wake_up(&unregister_wait); 1383 1384 closure_debug_destroy(&c->cl); 1385 kobject_put(&c->kobj); 1386 } 1387 1388 static void cache_set_flush(struct closure *cl) 1389 { 1390 struct cache_set *c = container_of(cl, struct cache_set, caching); 1391 struct cache *ca; 1392 struct btree *b; 1393 unsigned i; 1394 1395 bch_cache_accounting_destroy(&c->accounting); 1396 1397 kobject_put(&c->internal); 1398 kobject_del(&c->kobj); 1399 1400 if (c->gc_thread) 1401 kthread_stop(c->gc_thread); 1402 1403 if (!IS_ERR_OR_NULL(c->root)) 1404 list_add(&c->root->list, &c->btree_cache); 1405 1406 /* Should skip this if we're unregistering because of an error */ 1407 list_for_each_entry(b, &c->btree_cache, list) { 1408 mutex_lock(&b->write_lock); 1409 if (btree_node_dirty(b)) 1410 __bch_btree_node_write(b, NULL); 1411 mutex_unlock(&b->write_lock); 1412 } 1413 1414 for_each_cache(ca, c, i) 1415 if (ca->alloc_thread) 1416 kthread_stop(ca->alloc_thread); 1417 1418 if (c->journal.cur) { 1419 cancel_delayed_work_sync(&c->journal.work); 1420 /* flush last journal entry if needed */ 1421 c->journal.work.work.func(&c->journal.work.work); 1422 } 1423 1424 closure_return(cl); 1425 } 1426 1427 static void __cache_set_unregister(struct closure *cl) 1428 { 1429 struct cache_set *c = container_of(cl, struct cache_set, caching); 1430 struct cached_dev *dc; 1431 size_t i; 1432 1433 mutex_lock(&bch_register_lock); 1434 1435 for (i = 0; i < c->nr_uuids; i++) 1436 if (c->devices[i]) { 1437 if (!UUID_FLASH_ONLY(&c->uuids[i]) && 1438 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { 1439 dc = container_of(c->devices[i], 1440 struct cached_dev, disk); 1441 bch_cached_dev_detach(dc); 1442 } else { 1443 bcache_device_stop(c->devices[i]); 1444 } 1445 } 1446 1447 mutex_unlock(&bch_register_lock); 1448 1449 continue_at(cl, cache_set_flush, system_wq); 1450 } 1451 1452 void bch_cache_set_stop(struct cache_set *c) 1453 { 1454 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) 1455 closure_queue(&c->caching); 1456 } 1457 1458 void bch_cache_set_unregister(struct cache_set *c) 1459 { 1460 set_bit(CACHE_SET_UNREGISTERING, &c->flags); 1461 bch_cache_set_stop(c); 1462 } 1463 1464 #define alloc_bucket_pages(gfp, c) \ 1465 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c)))) 1466 1467 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) 1468 { 1469 int iter_size; 1470 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); 1471 if (!c) 1472 return NULL; 1473 1474 __module_get(THIS_MODULE); 1475 closure_init(&c->cl, NULL); 1476 set_closure_fn(&c->cl, cache_set_free, system_wq); 1477 1478 closure_init(&c->caching, &c->cl); 1479 set_closure_fn(&c->caching, __cache_set_unregister, system_wq); 1480 1481 /* Maybe create continue_at_noreturn() and use it here? */ 1482 closure_set_stopped(&c->cl); 1483 closure_put(&c->cl); 1484 1485 kobject_init(&c->kobj, &bch_cache_set_ktype); 1486 kobject_init(&c->internal, &bch_cache_set_internal_ktype); 1487 1488 bch_cache_accounting_init(&c->accounting, &c->cl); 1489 1490 memcpy(c->sb.set_uuid, sb->set_uuid, 16); 1491 c->sb.block_size = sb->block_size; 1492 c->sb.bucket_size = sb->bucket_size; 1493 c->sb.nr_in_set = sb->nr_in_set; 1494 c->sb.last_mount = sb->last_mount; 1495 c->bucket_bits = ilog2(sb->bucket_size); 1496 c->block_bits = ilog2(sb->block_size); 1497 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry); 1498 1499 c->btree_pages = bucket_pages(c); 1500 if (c->btree_pages > BTREE_MAX_PAGES) 1501 c->btree_pages = max_t(int, c->btree_pages / 4, 1502 BTREE_MAX_PAGES); 1503 1504 sema_init(&c->sb_write_mutex, 1); 1505 mutex_init(&c->bucket_lock); 1506 init_waitqueue_head(&c->btree_cache_wait); 1507 init_waitqueue_head(&c->bucket_wait); 1508 sema_init(&c->uuid_write_mutex, 1); 1509 1510 spin_lock_init(&c->btree_gc_time.lock); 1511 spin_lock_init(&c->btree_split_time.lock); 1512 spin_lock_init(&c->btree_read_time.lock); 1513 1514 bch_moving_init_cache_set(c); 1515 1516 INIT_LIST_HEAD(&c->list); 1517 INIT_LIST_HEAD(&c->cached_devs); 1518 INIT_LIST_HEAD(&c->btree_cache); 1519 INIT_LIST_HEAD(&c->btree_cache_freeable); 1520 INIT_LIST_HEAD(&c->btree_cache_freed); 1521 INIT_LIST_HEAD(&c->data_buckets); 1522 1523 c->search = mempool_create_slab_pool(32, bch_search_cache); 1524 if (!c->search) 1525 goto err; 1526 1527 iter_size = (sb->bucket_size / sb->block_size + 1) * 1528 sizeof(struct btree_iter_set); 1529 1530 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) || 1531 !(c->bio_meta = mempool_create_kmalloc_pool(2, 1532 sizeof(struct bbio) + sizeof(struct bio_vec) * 1533 bucket_pages(c))) || 1534 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) || 1535 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || 1536 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) || 1537 !(c->moving_gc_wq = create_workqueue("bcache_gc")) || 1538 bch_journal_alloc(c) || 1539 bch_btree_cache_alloc(c) || 1540 bch_open_buckets_alloc(c) || 1541 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages))) 1542 goto err; 1543 1544 c->congested_read_threshold_us = 2000; 1545 c->congested_write_threshold_us = 20000; 1546 c->error_limit = 8 << IO_ERROR_SHIFT; 1547 1548 return c; 1549 err: 1550 bch_cache_set_unregister(c); 1551 return NULL; 1552 } 1553 1554 static void run_cache_set(struct cache_set *c) 1555 { 1556 const char *err = "cannot allocate memory"; 1557 struct cached_dev *dc, *t; 1558 struct cache *ca; 1559 struct closure cl; 1560 unsigned i; 1561 1562 closure_init_stack(&cl); 1563 1564 for_each_cache(ca, c, i) 1565 c->nbuckets += ca->sb.nbuckets; 1566 1567 if (CACHE_SYNC(&c->sb)) { 1568 LIST_HEAD(journal); 1569 struct bkey *k; 1570 struct jset *j; 1571 1572 err = "cannot allocate memory for journal"; 1573 if (bch_journal_read(c, &journal)) 1574 goto err; 1575 1576 pr_debug("btree_journal_read() done"); 1577 1578 err = "no journal entries found"; 1579 if (list_empty(&journal)) 1580 goto err; 1581 1582 j = &list_entry(journal.prev, struct journal_replay, list)->j; 1583 1584 err = "IO error reading priorities"; 1585 for_each_cache(ca, c, i) 1586 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]); 1587 1588 /* 1589 * If prio_read() fails it'll call cache_set_error and we'll 1590 * tear everything down right away, but if we perhaps checked 1591 * sooner we could avoid journal replay. 1592 */ 1593 1594 k = &j->btree_root; 1595 1596 err = "bad btree root"; 1597 if (__bch_btree_ptr_invalid(c, k)) 1598 goto err; 1599 1600 err = "error reading btree root"; 1601 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL); 1602 if (IS_ERR_OR_NULL(c->root)) 1603 goto err; 1604 1605 list_del_init(&c->root->list); 1606 rw_unlock(true, c->root); 1607 1608 err = uuid_read(c, j, &cl); 1609 if (err) 1610 goto err; 1611 1612 err = "error in recovery"; 1613 if (bch_btree_check(c)) 1614 goto err; 1615 1616 bch_journal_mark(c, &journal); 1617 bch_initial_gc_finish(c); 1618 pr_debug("btree_check() done"); 1619 1620 /* 1621 * bcache_journal_next() can't happen sooner, or 1622 * btree_gc_finish() will give spurious errors about last_gc > 1623 * gc_gen - this is a hack but oh well. 1624 */ 1625 bch_journal_next(&c->journal); 1626 1627 err = "error starting allocator thread"; 1628 for_each_cache(ca, c, i) 1629 if (bch_cache_allocator_start(ca)) 1630 goto err; 1631 1632 /* 1633 * First place it's safe to allocate: btree_check() and 1634 * btree_gc_finish() have to run before we have buckets to 1635 * allocate, and bch_bucket_alloc_set() might cause a journal 1636 * entry to be written so bcache_journal_next() has to be called 1637 * first. 1638 * 1639 * If the uuids were in the old format we have to rewrite them 1640 * before the next journal entry is written: 1641 */ 1642 if (j->version < BCACHE_JSET_VERSION_UUID) 1643 __uuid_write(c); 1644 1645 bch_journal_replay(c, &journal); 1646 } else { 1647 pr_notice("invalidating existing data"); 1648 1649 for_each_cache(ca, c, i) { 1650 unsigned j; 1651 1652 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, 1653 2, SB_JOURNAL_BUCKETS); 1654 1655 for (j = 0; j < ca->sb.keys; j++) 1656 ca->sb.d[j] = ca->sb.first_bucket + j; 1657 } 1658 1659 bch_initial_gc_finish(c); 1660 1661 err = "error starting allocator thread"; 1662 for_each_cache(ca, c, i) 1663 if (bch_cache_allocator_start(ca)) 1664 goto err; 1665 1666 mutex_lock(&c->bucket_lock); 1667 for_each_cache(ca, c, i) 1668 bch_prio_write(ca); 1669 mutex_unlock(&c->bucket_lock); 1670 1671 err = "cannot allocate new UUID bucket"; 1672 if (__uuid_write(c)) 1673 goto err; 1674 1675 err = "cannot allocate new btree root"; 1676 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL); 1677 if (IS_ERR_OR_NULL(c->root)) 1678 goto err; 1679 1680 mutex_lock(&c->root->write_lock); 1681 bkey_copy_key(&c->root->key, &MAX_KEY); 1682 bch_btree_node_write(c->root, &cl); 1683 mutex_unlock(&c->root->write_lock); 1684 1685 bch_btree_set_root(c->root); 1686 rw_unlock(true, c->root); 1687 1688 /* 1689 * We don't want to write the first journal entry until 1690 * everything is set up - fortunately journal entries won't be 1691 * written until the SET_CACHE_SYNC() here: 1692 */ 1693 SET_CACHE_SYNC(&c->sb, true); 1694 1695 bch_journal_next(&c->journal); 1696 bch_journal_meta(c, &cl); 1697 } 1698 1699 err = "error starting gc thread"; 1700 if (bch_gc_thread_start(c)) 1701 goto err; 1702 1703 closure_sync(&cl); 1704 c->sb.last_mount = get_seconds(); 1705 bcache_write_super(c); 1706 1707 list_for_each_entry_safe(dc, t, &uncached_devices, list) 1708 bch_cached_dev_attach(dc, c); 1709 1710 flash_devs_run(c); 1711 1712 set_bit(CACHE_SET_RUNNING, &c->flags); 1713 return; 1714 err: 1715 closure_sync(&cl); 1716 /* XXX: test this, it's broken */ 1717 bch_cache_set_error(c, "%s", err); 1718 } 1719 1720 static bool can_attach_cache(struct cache *ca, struct cache_set *c) 1721 { 1722 return ca->sb.block_size == c->sb.block_size && 1723 ca->sb.bucket_size == c->sb.bucket_size && 1724 ca->sb.nr_in_set == c->sb.nr_in_set; 1725 } 1726 1727 static const char *register_cache_set(struct cache *ca) 1728 { 1729 char buf[12]; 1730 const char *err = "cannot allocate memory"; 1731 struct cache_set *c; 1732 1733 list_for_each_entry(c, &bch_cache_sets, list) 1734 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) { 1735 if (c->cache[ca->sb.nr_this_dev]) 1736 return "duplicate cache set member"; 1737 1738 if (!can_attach_cache(ca, c)) 1739 return "cache sb does not match set"; 1740 1741 if (!CACHE_SYNC(&ca->sb)) 1742 SET_CACHE_SYNC(&c->sb, false); 1743 1744 goto found; 1745 } 1746 1747 c = bch_cache_set_alloc(&ca->sb); 1748 if (!c) 1749 return err; 1750 1751 err = "error creating kobject"; 1752 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) || 1753 kobject_add(&c->internal, &c->kobj, "internal")) 1754 goto err; 1755 1756 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) 1757 goto err; 1758 1759 bch_debug_init_cache_set(c); 1760 1761 list_add(&c->list, &bch_cache_sets); 1762 found: 1763 sprintf(buf, "cache%i", ca->sb.nr_this_dev); 1764 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || 1765 sysfs_create_link(&c->kobj, &ca->kobj, buf)) 1766 goto err; 1767 1768 if (ca->sb.seq > c->sb.seq) { 1769 c->sb.version = ca->sb.version; 1770 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16); 1771 c->sb.flags = ca->sb.flags; 1772 c->sb.seq = ca->sb.seq; 1773 pr_debug("set version = %llu", c->sb.version); 1774 } 1775 1776 kobject_get(&ca->kobj); 1777 ca->set = c; 1778 ca->set->cache[ca->sb.nr_this_dev] = ca; 1779 c->cache_by_alloc[c->caches_loaded++] = ca; 1780 1781 if (c->caches_loaded == c->sb.nr_in_set) 1782 run_cache_set(c); 1783 1784 return NULL; 1785 err: 1786 bch_cache_set_unregister(c); 1787 return err; 1788 } 1789 1790 /* Cache device */ 1791 1792 void bch_cache_release(struct kobject *kobj) 1793 { 1794 struct cache *ca = container_of(kobj, struct cache, kobj); 1795 unsigned i; 1796 1797 if (ca->set) { 1798 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca); 1799 ca->set->cache[ca->sb.nr_this_dev] = NULL; 1800 } 1801 1802 bio_split_pool_free(&ca->bio_split_hook); 1803 1804 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca))); 1805 kfree(ca->prio_buckets); 1806 vfree(ca->buckets); 1807 1808 free_heap(&ca->heap); 1809 free_fifo(&ca->free_inc); 1810 1811 for (i = 0; i < RESERVE_NR; i++) 1812 free_fifo(&ca->free[i]); 1813 1814 if (ca->sb_bio.bi_inline_vecs[0].bv_page) 1815 put_page(ca->sb_bio.bi_io_vec[0].bv_page); 1816 1817 if (!IS_ERR_OR_NULL(ca->bdev)) 1818 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1819 1820 kfree(ca); 1821 module_put(THIS_MODULE); 1822 } 1823 1824 static int cache_alloc(struct cache_sb *sb, struct cache *ca) 1825 { 1826 size_t free; 1827 struct bucket *b; 1828 1829 __module_get(THIS_MODULE); 1830 kobject_init(&ca->kobj, &bch_cache_ktype); 1831 1832 bio_init(&ca->journal.bio); 1833 ca->journal.bio.bi_max_vecs = 8; 1834 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs; 1835 1836 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; 1837 1838 if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) || 1839 !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) || 1840 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) || 1841 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) || 1842 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) || 1843 !init_heap(&ca->heap, free << 3, GFP_KERNEL) || 1844 !(ca->buckets = vzalloc(sizeof(struct bucket) * 1845 ca->sb.nbuckets)) || 1846 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) * 1847 2, GFP_KERNEL)) || 1848 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) || 1849 bio_split_pool_init(&ca->bio_split_hook)) 1850 return -ENOMEM; 1851 1852 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); 1853 1854 for_each_bucket(b, ca) 1855 atomic_set(&b->pin, 0); 1856 1857 return 0; 1858 } 1859 1860 static void register_cache(struct cache_sb *sb, struct page *sb_page, 1861 struct block_device *bdev, struct cache *ca) 1862 { 1863 char name[BDEVNAME_SIZE]; 1864 const char *err = "cannot allocate memory"; 1865 1866 memcpy(&ca->sb, sb, sizeof(struct cache_sb)); 1867 ca->bdev = bdev; 1868 ca->bdev->bd_holder = ca; 1869 1870 bio_init(&ca->sb_bio); 1871 ca->sb_bio.bi_max_vecs = 1; 1872 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs; 1873 ca->sb_bio.bi_io_vec[0].bv_page = sb_page; 1874 get_page(sb_page); 1875 1876 if (blk_queue_discard(bdev_get_queue(ca->bdev))) 1877 ca->discard = CACHE_DISCARD(&ca->sb); 1878 1879 if (cache_alloc(sb, ca) != 0) 1880 goto err; 1881 1882 err = "error creating kobject"; 1883 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) 1884 goto err; 1885 1886 mutex_lock(&bch_register_lock); 1887 err = register_cache_set(ca); 1888 mutex_unlock(&bch_register_lock); 1889 1890 if (err) 1891 goto err; 1892 1893 pr_info("registered cache device %s", bdevname(bdev, name)); 1894 out: 1895 kobject_put(&ca->kobj); 1896 return; 1897 err: 1898 pr_notice("error opening %s: %s", bdevname(bdev, name), err); 1899 goto out; 1900 } 1901 1902 /* Global interfaces/init */ 1903 1904 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *, 1905 const char *, size_t); 1906 1907 kobj_attribute_write(register, register_bcache); 1908 kobj_attribute_write(register_quiet, register_bcache); 1909 1910 static bool bch_is_open_backing(struct block_device *bdev) { 1911 struct cache_set *c, *tc; 1912 struct cached_dev *dc, *t; 1913 1914 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 1915 list_for_each_entry_safe(dc, t, &c->cached_devs, list) 1916 if (dc->bdev == bdev) 1917 return true; 1918 list_for_each_entry_safe(dc, t, &uncached_devices, list) 1919 if (dc->bdev == bdev) 1920 return true; 1921 return false; 1922 } 1923 1924 static bool bch_is_open_cache(struct block_device *bdev) { 1925 struct cache_set *c, *tc; 1926 struct cache *ca; 1927 unsigned i; 1928 1929 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 1930 for_each_cache(ca, c, i) 1931 if (ca->bdev == bdev) 1932 return true; 1933 return false; 1934 } 1935 1936 static bool bch_is_open(struct block_device *bdev) { 1937 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev); 1938 } 1939 1940 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, 1941 const char *buffer, size_t size) 1942 { 1943 ssize_t ret = size; 1944 const char *err = "cannot allocate memory"; 1945 char *path = NULL; 1946 struct cache_sb *sb = NULL; 1947 struct block_device *bdev = NULL; 1948 struct page *sb_page = NULL; 1949 1950 if (!try_module_get(THIS_MODULE)) 1951 return -EBUSY; 1952 1953 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) || 1954 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL))) 1955 goto err; 1956 1957 err = "failed to open device"; 1958 bdev = blkdev_get_by_path(strim(path), 1959 FMODE_READ|FMODE_WRITE|FMODE_EXCL, 1960 sb); 1961 if (IS_ERR(bdev)) { 1962 if (bdev == ERR_PTR(-EBUSY)) { 1963 bdev = lookup_bdev(strim(path)); 1964 mutex_lock(&bch_register_lock); 1965 if (!IS_ERR(bdev) && bch_is_open(bdev)) 1966 err = "device already registered"; 1967 else 1968 err = "device busy"; 1969 mutex_unlock(&bch_register_lock); 1970 } 1971 goto err; 1972 } 1973 1974 err = "failed to set blocksize"; 1975 if (set_blocksize(bdev, 4096)) 1976 goto err_close; 1977 1978 err = read_super(sb, bdev, &sb_page); 1979 if (err) 1980 goto err_close; 1981 1982 if (SB_IS_BDEV(sb)) { 1983 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL); 1984 if (!dc) 1985 goto err_close; 1986 1987 mutex_lock(&bch_register_lock); 1988 register_bdev(sb, sb_page, bdev, dc); 1989 mutex_unlock(&bch_register_lock); 1990 } else { 1991 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL); 1992 if (!ca) 1993 goto err_close; 1994 1995 register_cache(sb, sb_page, bdev, ca); 1996 } 1997 out: 1998 if (sb_page) 1999 put_page(sb_page); 2000 kfree(sb); 2001 kfree(path); 2002 module_put(THIS_MODULE); 2003 return ret; 2004 2005 err_close: 2006 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2007 err: 2008 if (attr != &ksysfs_register_quiet) 2009 pr_info("error opening %s: %s", path, err); 2010 ret = -EINVAL; 2011 goto out; 2012 } 2013 2014 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) 2015 { 2016 if (code == SYS_DOWN || 2017 code == SYS_HALT || 2018 code == SYS_POWER_OFF) { 2019 DEFINE_WAIT(wait); 2020 unsigned long start = jiffies; 2021 bool stopped = false; 2022 2023 struct cache_set *c, *tc; 2024 struct cached_dev *dc, *tdc; 2025 2026 mutex_lock(&bch_register_lock); 2027 2028 if (list_empty(&bch_cache_sets) && 2029 list_empty(&uncached_devices)) 2030 goto out; 2031 2032 pr_info("Stopping all devices:"); 2033 2034 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2035 bch_cache_set_stop(c); 2036 2037 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) 2038 bcache_device_stop(&dc->disk); 2039 2040 /* What's a condition variable? */ 2041 while (1) { 2042 long timeout = start + 2 * HZ - jiffies; 2043 2044 stopped = list_empty(&bch_cache_sets) && 2045 list_empty(&uncached_devices); 2046 2047 if (timeout < 0 || stopped) 2048 break; 2049 2050 prepare_to_wait(&unregister_wait, &wait, 2051 TASK_UNINTERRUPTIBLE); 2052 2053 mutex_unlock(&bch_register_lock); 2054 schedule_timeout(timeout); 2055 mutex_lock(&bch_register_lock); 2056 } 2057 2058 finish_wait(&unregister_wait, &wait); 2059 2060 if (stopped) 2061 pr_info("All devices stopped"); 2062 else 2063 pr_notice("Timeout waiting for devices to be closed"); 2064 out: 2065 mutex_unlock(&bch_register_lock); 2066 } 2067 2068 return NOTIFY_DONE; 2069 } 2070 2071 static struct notifier_block reboot = { 2072 .notifier_call = bcache_reboot, 2073 .priority = INT_MAX, /* before any real devices */ 2074 }; 2075 2076 static void bcache_exit(void) 2077 { 2078 bch_debug_exit(); 2079 bch_request_exit(); 2080 if (bcache_kobj) 2081 kobject_put(bcache_kobj); 2082 if (bcache_wq) 2083 destroy_workqueue(bcache_wq); 2084 if (bcache_major) 2085 unregister_blkdev(bcache_major, "bcache"); 2086 unregister_reboot_notifier(&reboot); 2087 } 2088 2089 static int __init bcache_init(void) 2090 { 2091 static const struct attribute *files[] = { 2092 &ksysfs_register.attr, 2093 &ksysfs_register_quiet.attr, 2094 NULL 2095 }; 2096 2097 mutex_init(&bch_register_lock); 2098 init_waitqueue_head(&unregister_wait); 2099 register_reboot_notifier(&reboot); 2100 closure_debug_init(); 2101 2102 bcache_major = register_blkdev(0, "bcache"); 2103 if (bcache_major < 0) 2104 return bcache_major; 2105 2106 if (!(bcache_wq = create_workqueue("bcache")) || 2107 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) || 2108 sysfs_create_files(bcache_kobj, files) || 2109 bch_request_init() || 2110 bch_debug_init(bcache_kobj)) 2111 goto err; 2112 2113 return 0; 2114 err: 2115 bcache_exit(); 2116 return -ENOMEM; 2117 } 2118 2119 module_exit(bcache_exit); 2120 module_init(bcache_init); 2121