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