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 27MODULE_LICENSE("GPL"); 28MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); 29 30static const char bcache_magic[] = { 31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca, 32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81 33}; 34 35static 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 */ 41const char * const bch_cache_modes[] = { 42 "default", 43 "writethrough", 44 "writeback", 45 "writearound", 46 "none", 47 NULL 48}; 49 50static struct kobject *bcache_kobj; 51struct mutex bch_register_lock; 52LIST_HEAD(bch_cache_sets); 53static LIST_HEAD(uncached_devices); 54 55static int bcache_major; 56static DEFINE_IDA(bcache_minor); 57static wait_queue_head_t unregister_wait; 58struct workqueue_struct *bcache_wq; 59 60#define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE) 61 62static 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 71static 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 87static 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; 219err: 220 put_bh(bh); 221 return err; 222} 223 224static 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 232static 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 267static 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 274void 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 293static 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 301static 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 308void 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 342static 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 352static 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 359static 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 399static 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 440static 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 460int 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 470static 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 482static 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 515static 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 524static 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 544void 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 610static 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 644static 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 654static 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 660static 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 667static 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 674void 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 680static 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 696static 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 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags); 713} 714 715static void bcache_device_detach(struct bcache_device *d) 716{ 717 lockdep_assert_held(&bch_register_lock); 718 719 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) { 720 struct uuid_entry *u = d->c->uuids + d->id; 721 722 SET_UUID_FLASH_ONLY(u, 0); 723 memcpy(u->uuid, invalid_uuid, 16); 724 u->invalidated = cpu_to_le32(get_seconds()); 725 bch_uuid_write(d->c); 726 } 727 728 bcache_device_unlink(d); 729 730 d->c->devices[d->id] = NULL; 731 closure_put(&d->c->caching); 732 d->c = NULL; 733} 734 735static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, 736 unsigned id) 737{ 738 d->id = id; 739 d->c = c; 740 c->devices[id] = d; 741 742 closure_get(&c->caching); 743} 744 745static void bcache_device_free(struct bcache_device *d) 746{ 747 lockdep_assert_held(&bch_register_lock); 748 749 pr_info("%s stopped", d->disk->disk_name); 750 751 if (d->c) 752 bcache_device_detach(d); 753 if (d->disk && d->disk->flags & GENHD_FL_UP) 754 del_gendisk(d->disk); 755 if (d->disk && d->disk->queue) 756 blk_cleanup_queue(d->disk->queue); 757 if (d->disk) { 758 ida_simple_remove(&bcache_minor, d->disk->first_minor); 759 put_disk(d->disk); 760 } 761 762 bio_split_pool_free(&d->bio_split_hook); 763 if (d->bio_split) 764 bioset_free(d->bio_split); 765 if (is_vmalloc_addr(d->full_dirty_stripes)) 766 vfree(d->full_dirty_stripes); 767 else 768 kfree(d->full_dirty_stripes); 769 if (is_vmalloc_addr(d->stripe_sectors_dirty)) 770 vfree(d->stripe_sectors_dirty); 771 else 772 kfree(d->stripe_sectors_dirty); 773 774 closure_debug_destroy(&d->cl); 775} 776 777static int bcache_device_init(struct bcache_device *d, unsigned block_size, 778 sector_t sectors) 779{ 780 struct request_queue *q; 781 size_t n; 782 int minor; 783 784 if (!d->stripe_size) 785 d->stripe_size = 1 << 31; 786 787 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size); 788 789 if (!d->nr_stripes || 790 d->nr_stripes > INT_MAX || 791 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) { 792 pr_err("nr_stripes too large"); 793 return -ENOMEM; 794 } 795 796 n = d->nr_stripes * sizeof(atomic_t); 797 d->stripe_sectors_dirty = n < PAGE_SIZE << 6 798 ? kzalloc(n, GFP_KERNEL) 799 : vzalloc(n); 800 if (!d->stripe_sectors_dirty) 801 return -ENOMEM; 802 803 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long); 804 d->full_dirty_stripes = n < PAGE_SIZE << 6 805 ? kzalloc(n, GFP_KERNEL) 806 : vzalloc(n); 807 if (!d->full_dirty_stripes) 808 return -ENOMEM; 809 810 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL); 811 if (minor < 0) 812 return minor; 813 814 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || 815 bio_split_pool_init(&d->bio_split_hook) || 816 !(d->disk = alloc_disk(1))) { 817 ida_simple_remove(&bcache_minor, minor); 818 return -ENOMEM; 819 } 820 821 set_capacity(d->disk, sectors); 822 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor); 823 824 d->disk->major = bcache_major; 825 d->disk->first_minor = minor; 826 d->disk->fops = &bcache_ops; 827 d->disk->private_data = d; 828 829 q = blk_alloc_queue(GFP_KERNEL); 830 if (!q) 831 return -ENOMEM; 832 833 blk_queue_make_request(q, NULL); 834 d->disk->queue = q; 835 q->queuedata = d; 836 q->backing_dev_info.congested_data = d; 837 q->limits.max_hw_sectors = UINT_MAX; 838 q->limits.max_sectors = UINT_MAX; 839 q->limits.max_segment_size = UINT_MAX; 840 q->limits.max_segments = BIO_MAX_PAGES; 841 q->limits.max_discard_sectors = UINT_MAX; 842 q->limits.discard_granularity = 512; 843 q->limits.io_min = block_size; 844 q->limits.logical_block_size = block_size; 845 q->limits.physical_block_size = block_size; 846 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags); 847 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags); 848 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags); 849 850 blk_queue_flush(q, REQ_FLUSH|REQ_FUA); 851 852 return 0; 853} 854 855/* Cached device */ 856 857static void calc_cached_dev_sectors(struct cache_set *c) 858{ 859 uint64_t sectors = 0; 860 struct cached_dev *dc; 861 862 list_for_each_entry(dc, &c->cached_devs, list) 863 sectors += bdev_sectors(dc->bdev); 864 865 c->cached_dev_sectors = sectors; 866} 867 868void bch_cached_dev_run(struct cached_dev *dc) 869{ 870 struct bcache_device *d = &dc->disk; 871 char buf[SB_LABEL_SIZE + 1]; 872 char *env[] = { 873 "DRIVER=bcache", 874 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), 875 NULL, 876 NULL, 877 }; 878 879 memcpy(buf, dc->sb.label, SB_LABEL_SIZE); 880 buf[SB_LABEL_SIZE] = '\0'; 881 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf); 882 883 if (atomic_xchg(&dc->running, 1)) { 884 kfree(env[1]); 885 kfree(env[2]); 886 return; 887 } 888 889 if (!d->c && 890 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { 891 struct closure cl; 892 closure_init_stack(&cl); 893 894 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); 895 bch_write_bdev_super(dc, &cl); 896 closure_sync(&cl); 897 } 898 899 add_disk(d->disk); 900 bd_link_disk_holder(dc->bdev, dc->disk.disk); 901 /* won't show up in the uevent file, use udevadm monitor -e instead 902 * only class / kset properties are persistent */ 903 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); 904 kfree(env[1]); 905 kfree(env[2]); 906 907 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || 908 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache")) 909 pr_debug("error creating sysfs link"); 910} 911 912static void cached_dev_detach_finish(struct work_struct *w) 913{ 914 struct cached_dev *dc = container_of(w, struct cached_dev, detach); 915 char buf[BDEVNAME_SIZE]; 916 struct closure cl; 917 closure_init_stack(&cl); 918 919 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)); 920 BUG_ON(atomic_read(&dc->count)); 921 922 mutex_lock(&bch_register_lock); 923 924 memset(&dc->sb.set_uuid, 0, 16); 925 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE); 926 927 bch_write_bdev_super(dc, &cl); 928 closure_sync(&cl); 929 930 bcache_device_detach(&dc->disk); 931 list_move(&dc->list, &uncached_devices); 932 933 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags); 934 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags); 935 936 mutex_unlock(&bch_register_lock); 937 938 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf)); 939 940 /* Drop ref we took in cached_dev_detach() */ 941 closure_put(&dc->disk.cl); 942} 943 944void bch_cached_dev_detach(struct cached_dev *dc) 945{ 946 lockdep_assert_held(&bch_register_lock); 947 948 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) 949 return; 950 951 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) 952 return; 953 954 /* 955 * Block the device from being closed and freed until we're finished 956 * detaching 957 */ 958 closure_get(&dc->disk.cl); 959 960 bch_writeback_queue(dc); 961 cached_dev_put(dc); 962} 963 964int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c) 965{ 966 uint32_t rtime = cpu_to_le32(get_seconds()); 967 struct uuid_entry *u; 968 char buf[BDEVNAME_SIZE]; 969 970 bdevname(dc->bdev, buf); 971 972 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)) 973 return -ENOENT; 974 975 if (dc->disk.c) { 976 pr_err("Can't attach %s: already attached", buf); 977 return -EINVAL; 978 } 979 980 if (test_bit(CACHE_SET_STOPPING, &c->flags)) { 981 pr_err("Can't attach %s: shutting down", buf); 982 return -EINVAL; 983 } 984 985 if (dc->sb.block_size < c->sb.block_size) { 986 /* Will die */ 987 pr_err("Couldn't attach %s: block size less than set's block size", 988 buf); 989 return -EINVAL; 990 } 991 992 u = uuid_find(c, dc->sb.uuid); 993 994 if (u && 995 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || 996 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { 997 memcpy(u->uuid, invalid_uuid, 16); 998 u->invalidated = cpu_to_le32(get_seconds()); 999 u = NULL; 1000 } 1001 1002 if (!u) { 1003 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1004 pr_err("Couldn't find uuid for %s in set", buf); 1005 return -ENOENT; 1006 } 1007 1008 u = uuid_find_empty(c); 1009 if (!u) { 1010 pr_err("Not caching %s, no room for UUID", buf); 1011 return -EINVAL; 1012 } 1013 } 1014 1015 /* Deadlocks since we're called via sysfs... 1016 sysfs_remove_file(&dc->kobj, &sysfs_attach); 1017 */ 1018 1019 if (bch_is_zero(u->uuid, 16)) { 1020 struct closure cl; 1021 closure_init_stack(&cl); 1022 1023 memcpy(u->uuid, dc->sb.uuid, 16); 1024 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); 1025 u->first_reg = u->last_reg = rtime; 1026 bch_uuid_write(c); 1027 1028 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16); 1029 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); 1030 1031 bch_write_bdev_super(dc, &cl); 1032 closure_sync(&cl); 1033 } else { 1034 u->last_reg = rtime; 1035 bch_uuid_write(c); 1036 } 1037 1038 bcache_device_attach(&dc->disk, c, u - c->uuids); 1039 list_move(&dc->list, &c->cached_devs); 1040 calc_cached_dev_sectors(c); 1041 1042 smp_wmb(); 1043 /* 1044 * dc->c must be set before dc->count != 0 - paired with the mb in 1045 * cached_dev_get() 1046 */ 1047 atomic_set(&dc->count, 1); 1048 1049 /* Block writeback thread, but spawn it */ 1050 down_write(&dc->writeback_lock); 1051 if (bch_cached_dev_writeback_start(dc)) { 1052 up_write(&dc->writeback_lock); 1053 return -ENOMEM; 1054 } 1055 1056 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1057 bch_sectors_dirty_init(dc); 1058 atomic_set(&dc->has_dirty, 1); 1059 atomic_inc(&dc->count); 1060 bch_writeback_queue(dc); 1061 } 1062 1063 bch_cached_dev_run(dc); 1064 bcache_device_link(&dc->disk, c, "bdev"); 1065 1066 /* Allow the writeback thread to proceed */ 1067 up_write(&dc->writeback_lock); 1068 1069 pr_info("Caching %s as %s on set %pU", 1070 bdevname(dc->bdev, buf), dc->disk.disk->disk_name, 1071 dc->disk.c->sb.set_uuid); 1072 return 0; 1073} 1074 1075void bch_cached_dev_release(struct kobject *kobj) 1076{ 1077 struct cached_dev *dc = container_of(kobj, struct cached_dev, 1078 disk.kobj); 1079 kfree(dc); 1080 module_put(THIS_MODULE); 1081} 1082 1083static void cached_dev_free(struct closure *cl) 1084{ 1085 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1086 1087 cancel_delayed_work_sync(&dc->writeback_rate_update); 1088 if (!IS_ERR_OR_NULL(dc->writeback_thread)) 1089 kthread_stop(dc->writeback_thread); 1090 1091 mutex_lock(&bch_register_lock); 1092 1093 if (atomic_read(&dc->running)) 1094 bd_unlink_disk_holder(dc->bdev, dc->disk.disk); 1095 bcache_device_free(&dc->disk); 1096 list_del(&dc->list); 1097 1098 mutex_unlock(&bch_register_lock); 1099 1100 if (!IS_ERR_OR_NULL(dc->bdev)) 1101 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1102 1103 wake_up(&unregister_wait); 1104 1105 kobject_put(&dc->disk.kobj); 1106} 1107 1108static void cached_dev_flush(struct closure *cl) 1109{ 1110 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1111 struct bcache_device *d = &dc->disk; 1112 1113 mutex_lock(&bch_register_lock); 1114 bcache_device_unlink(d); 1115 mutex_unlock(&bch_register_lock); 1116 1117 bch_cache_accounting_destroy(&dc->accounting); 1118 kobject_del(&d->kobj); 1119 1120 continue_at(cl, cached_dev_free, system_wq); 1121} 1122 1123static int cached_dev_init(struct cached_dev *dc, unsigned block_size) 1124{ 1125 int ret; 1126 struct io *io; 1127 struct request_queue *q = bdev_get_queue(dc->bdev); 1128 1129 __module_get(THIS_MODULE); 1130 INIT_LIST_HEAD(&dc->list); 1131 closure_init(&dc->disk.cl, NULL); 1132 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); 1133 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); 1134 INIT_WORK(&dc->detach, cached_dev_detach_finish); 1135 sema_init(&dc->sb_write_mutex, 1); 1136 INIT_LIST_HEAD(&dc->io_lru); 1137 spin_lock_init(&dc->io_lock); 1138 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); 1139 1140 dc->sequential_cutoff = 4 << 20; 1141 1142 for (io = dc->io; io < dc->io + RECENT_IO; io++) { 1143 list_add(&io->lru, &dc->io_lru); 1144 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); 1145 } 1146 1147 dc->disk.stripe_size = q->limits.io_opt >> 9; 1148 1149 if (dc->disk.stripe_size) 1150 dc->partial_stripes_expensive = 1151 q->limits.raid_partial_stripes_expensive; 1152 1153 ret = bcache_device_init(&dc->disk, block_size, 1154 dc->bdev->bd_part->nr_sects - dc->sb.data_offset); 1155 if (ret) 1156 return ret; 1157 1158 set_capacity(dc->disk.disk, 1159 dc->bdev->bd_part->nr_sects - dc->sb.data_offset); 1160 1161 dc->disk.disk->queue->backing_dev_info.ra_pages = 1162 max(dc->disk.disk->queue->backing_dev_info.ra_pages, 1163 q->backing_dev_info.ra_pages); 1164 1165 bch_cached_dev_request_init(dc); 1166 bch_cached_dev_writeback_init(dc); 1167 return 0; 1168} 1169 1170/* Cached device - bcache superblock */ 1171 1172static void register_bdev(struct cache_sb *sb, struct page *sb_page, 1173 struct block_device *bdev, 1174 struct cached_dev *dc) 1175{ 1176 char name[BDEVNAME_SIZE]; 1177 const char *err = "cannot allocate memory"; 1178 struct cache_set *c; 1179 1180 memcpy(&dc->sb, sb, sizeof(struct cache_sb)); 1181 dc->bdev = bdev; 1182 dc->bdev->bd_holder = dc; 1183 1184 bio_init(&dc->sb_bio); 1185 dc->sb_bio.bi_max_vecs = 1; 1186 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs; 1187 dc->sb_bio.bi_io_vec[0].bv_page = sb_page; 1188 get_page(sb_page); 1189 1190 if (cached_dev_init(dc, sb->block_size << 9)) 1191 goto err; 1192 1193 err = "error creating kobject"; 1194 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj, 1195 "bcache")) 1196 goto err; 1197 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) 1198 goto err; 1199 1200 pr_info("registered backing device %s", bdevname(bdev, name)); 1201 1202 list_add(&dc->list, &uncached_devices); 1203 list_for_each_entry(c, &bch_cache_sets, list) 1204 bch_cached_dev_attach(dc, c); 1205 1206 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || 1207 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) 1208 bch_cached_dev_run(dc); 1209 1210 return; 1211err: 1212 pr_notice("error opening %s: %s", bdevname(bdev, name), err); 1213 bcache_device_stop(&dc->disk); 1214} 1215 1216/* Flash only volumes */ 1217 1218void bch_flash_dev_release(struct kobject *kobj) 1219{ 1220 struct bcache_device *d = container_of(kobj, struct bcache_device, 1221 kobj); 1222 kfree(d); 1223} 1224 1225static void flash_dev_free(struct closure *cl) 1226{ 1227 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1228 mutex_lock(&bch_register_lock); 1229 bcache_device_free(d); 1230 mutex_unlock(&bch_register_lock); 1231 kobject_put(&d->kobj); 1232} 1233 1234static void flash_dev_flush(struct closure *cl) 1235{ 1236 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1237 1238 mutex_lock(&bch_register_lock); 1239 bcache_device_unlink(d); 1240 mutex_unlock(&bch_register_lock); 1241 kobject_del(&d->kobj); 1242 continue_at(cl, flash_dev_free, system_wq); 1243} 1244 1245static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) 1246{ 1247 struct bcache_device *d = kzalloc(sizeof(struct bcache_device), 1248 GFP_KERNEL); 1249 if (!d) 1250 return -ENOMEM; 1251 1252 closure_init(&d->cl, NULL); 1253 set_closure_fn(&d->cl, flash_dev_flush, system_wq); 1254 1255 kobject_init(&d->kobj, &bch_flash_dev_ktype); 1256 1257 if (bcache_device_init(d, block_bytes(c), u->sectors)) 1258 goto err; 1259 1260 bcache_device_attach(d, c, u - c->uuids); 1261 bch_flash_dev_request_init(d); 1262 add_disk(d->disk); 1263 1264 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache")) 1265 goto err; 1266 1267 bcache_device_link(d, c, "volume"); 1268 1269 return 0; 1270err: 1271 kobject_put(&d->kobj); 1272 return -ENOMEM; 1273} 1274 1275static int flash_devs_run(struct cache_set *c) 1276{ 1277 int ret = 0; 1278 struct uuid_entry *u; 1279 1280 for (u = c->uuids; 1281 u < c->uuids + c->nr_uuids && !ret; 1282 u++) 1283 if (UUID_FLASH_ONLY(u)) 1284 ret = flash_dev_run(c, u); 1285 1286 return ret; 1287} 1288 1289int bch_flash_dev_create(struct cache_set *c, uint64_t size) 1290{ 1291 struct uuid_entry *u; 1292 1293 if (test_bit(CACHE_SET_STOPPING, &c->flags)) 1294 return -EINTR; 1295 1296 if (!test_bit(CACHE_SET_RUNNING, &c->flags)) 1297 return -EPERM; 1298 1299 u = uuid_find_empty(c); 1300 if (!u) { 1301 pr_err("Can't create volume, no room for UUID"); 1302 return -EINVAL; 1303 } 1304 1305 get_random_bytes(u->uuid, 16); 1306 memset(u->label, 0, 32); 1307 u->first_reg = u->last_reg = cpu_to_le32(get_seconds()); 1308 1309 SET_UUID_FLASH_ONLY(u, 1); 1310 u->sectors = size >> 9; 1311 1312 bch_uuid_write(c); 1313 1314 return flash_dev_run(c, u); 1315} 1316 1317/* Cache set */ 1318 1319__printf(2, 3) 1320bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) 1321{ 1322 va_list args; 1323 1324 if (c->on_error != ON_ERROR_PANIC && 1325 test_bit(CACHE_SET_STOPPING, &c->flags)) 1326 return false; 1327 1328 /* XXX: we can be called from atomic context 1329 acquire_console_sem(); 1330 */ 1331 1332 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid); 1333 1334 va_start(args, fmt); 1335 vprintk(fmt, args); 1336 va_end(args); 1337 1338 printk(", disabling caching\n"); 1339 1340 if (c->on_error == ON_ERROR_PANIC) 1341 panic("panic forced after error\n"); 1342 1343 bch_cache_set_unregister(c); 1344 return true; 1345} 1346 1347void bch_cache_set_release(struct kobject *kobj) 1348{ 1349 struct cache_set *c = container_of(kobj, struct cache_set, kobj); 1350 kfree(c); 1351 module_put(THIS_MODULE); 1352} 1353 1354static void cache_set_free(struct closure *cl) 1355{ 1356 struct cache_set *c = container_of(cl, struct cache_set, cl); 1357 struct cache *ca; 1358 unsigned i; 1359 1360 if (!IS_ERR_OR_NULL(c->debug)) 1361 debugfs_remove(c->debug); 1362 1363 bch_open_buckets_free(c); 1364 bch_btree_cache_free(c); 1365 bch_journal_free(c); 1366 1367 for_each_cache(ca, c, i) 1368 if (ca) { 1369 ca->set = NULL; 1370 c->cache[ca->sb.nr_this_dev] = NULL; 1371 kobject_put(&ca->kobj); 1372 } 1373 1374 bch_bset_sort_state_free(&c->sort); 1375 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c))); 1376 1377 if (c->moving_gc_wq) 1378 destroy_workqueue(c->moving_gc_wq); 1379 if (c->bio_split) 1380 bioset_free(c->bio_split); 1381 if (c->fill_iter) 1382 mempool_destroy(c->fill_iter); 1383 if (c->bio_meta) 1384 mempool_destroy(c->bio_meta); 1385 if (c->search) 1386 mempool_destroy(c->search); 1387 kfree(c->devices); 1388 1389 mutex_lock(&bch_register_lock); 1390 list_del(&c->list); 1391 mutex_unlock(&bch_register_lock); 1392 1393 pr_info("Cache set %pU unregistered", c->sb.set_uuid); 1394 wake_up(&unregister_wait); 1395 1396 closure_debug_destroy(&c->cl); 1397 kobject_put(&c->kobj); 1398} 1399 1400static void cache_set_flush(struct closure *cl) 1401{ 1402 struct cache_set *c = container_of(cl, struct cache_set, caching); 1403 struct cache *ca; 1404 struct btree *b; 1405 unsigned i; 1406 1407 if (!c) 1408 closure_return(cl); 1409 1410 bch_cache_accounting_destroy(&c->accounting); 1411 1412 kobject_put(&c->internal); 1413 kobject_del(&c->kobj); 1414 1415 if (c->gc_thread) 1416 kthread_stop(c->gc_thread); 1417 1418 if (!IS_ERR_OR_NULL(c->root)) 1419 list_add(&c->root->list, &c->btree_cache); 1420 1421 /* Should skip this if we're unregistering because of an error */ 1422 list_for_each_entry(b, &c->btree_cache, list) { 1423 mutex_lock(&b->write_lock); 1424 if (btree_node_dirty(b)) 1425 __bch_btree_node_write(b, NULL); 1426 mutex_unlock(&b->write_lock); 1427 } 1428 1429 for_each_cache(ca, c, i) 1430 if (ca->alloc_thread) 1431 kthread_stop(ca->alloc_thread); 1432 1433 if (c->journal.cur) { 1434 cancel_delayed_work_sync(&c->journal.work); 1435 /* flush last journal entry if needed */ 1436 c->journal.work.work.func(&c->journal.work.work); 1437 } 1438 1439 closure_return(cl); 1440} 1441 1442static void __cache_set_unregister(struct closure *cl) 1443{ 1444 struct cache_set *c = container_of(cl, struct cache_set, caching); 1445 struct cached_dev *dc; 1446 size_t i; 1447 1448 mutex_lock(&bch_register_lock); 1449 1450 for (i = 0; i < c->nr_uuids; i++) 1451 if (c->devices[i]) { 1452 if (!UUID_FLASH_ONLY(&c->uuids[i]) && 1453 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { 1454 dc = container_of(c->devices[i], 1455 struct cached_dev, disk); 1456 bch_cached_dev_detach(dc); 1457 } else { 1458 bcache_device_stop(c->devices[i]); 1459 } 1460 } 1461 1462 mutex_unlock(&bch_register_lock); 1463 1464 continue_at(cl, cache_set_flush, system_wq); 1465} 1466 1467void bch_cache_set_stop(struct cache_set *c) 1468{ 1469 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) 1470 closure_queue(&c->caching); 1471} 1472 1473void bch_cache_set_unregister(struct cache_set *c) 1474{ 1475 set_bit(CACHE_SET_UNREGISTERING, &c->flags); 1476 bch_cache_set_stop(c); 1477} 1478 1479#define alloc_bucket_pages(gfp, c) \ 1480 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c)))) 1481 1482struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) 1483{ 1484 int iter_size; 1485 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); 1486 if (!c) 1487 return NULL; 1488 1489 __module_get(THIS_MODULE); 1490 closure_init(&c->cl, NULL); 1491 set_closure_fn(&c->cl, cache_set_free, system_wq); 1492 1493 closure_init(&c->caching, &c->cl); 1494 set_closure_fn(&c->caching, __cache_set_unregister, system_wq); 1495 1496 /* Maybe create continue_at_noreturn() and use it here? */ 1497 closure_set_stopped(&c->cl); 1498 closure_put(&c->cl); 1499 1500 kobject_init(&c->kobj, &bch_cache_set_ktype); 1501 kobject_init(&c->internal, &bch_cache_set_internal_ktype); 1502 1503 bch_cache_accounting_init(&c->accounting, &c->cl); 1504 1505 memcpy(c->sb.set_uuid, sb->set_uuid, 16); 1506 c->sb.block_size = sb->block_size; 1507 c->sb.bucket_size = sb->bucket_size; 1508 c->sb.nr_in_set = sb->nr_in_set; 1509 c->sb.last_mount = sb->last_mount; 1510 c->bucket_bits = ilog2(sb->bucket_size); 1511 c->block_bits = ilog2(sb->block_size); 1512 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry); 1513 1514 c->btree_pages = bucket_pages(c); 1515 if (c->btree_pages > BTREE_MAX_PAGES) 1516 c->btree_pages = max_t(int, c->btree_pages / 4, 1517 BTREE_MAX_PAGES); 1518 1519 sema_init(&c->sb_write_mutex, 1); 1520 mutex_init(&c->bucket_lock); 1521 init_waitqueue_head(&c->btree_cache_wait); 1522 init_waitqueue_head(&c->bucket_wait); 1523 sema_init(&c->uuid_write_mutex, 1); 1524 1525 spin_lock_init(&c->btree_gc_time.lock); 1526 spin_lock_init(&c->btree_split_time.lock); 1527 spin_lock_init(&c->btree_read_time.lock); 1528 1529 bch_moving_init_cache_set(c); 1530 1531 INIT_LIST_HEAD(&c->list); 1532 INIT_LIST_HEAD(&c->cached_devs); 1533 INIT_LIST_HEAD(&c->btree_cache); 1534 INIT_LIST_HEAD(&c->btree_cache_freeable); 1535 INIT_LIST_HEAD(&c->btree_cache_freed); 1536 INIT_LIST_HEAD(&c->data_buckets); 1537 1538 c->search = mempool_create_slab_pool(32, bch_search_cache); 1539 if (!c->search) 1540 goto err; 1541 1542 iter_size = (sb->bucket_size / sb->block_size + 1) * 1543 sizeof(struct btree_iter_set); 1544 1545 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) || 1546 !(c->bio_meta = mempool_create_kmalloc_pool(2, 1547 sizeof(struct bbio) + sizeof(struct bio_vec) * 1548 bucket_pages(c))) || 1549 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) || 1550 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) || 1551 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) || 1552 !(c->moving_gc_wq = create_workqueue("bcache_gc")) || 1553 bch_journal_alloc(c) || 1554 bch_btree_cache_alloc(c) || 1555 bch_open_buckets_alloc(c) || 1556 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages))) 1557 goto err; 1558 1559 c->congested_read_threshold_us = 2000; 1560 c->congested_write_threshold_us = 20000; 1561 c->error_limit = 8 << IO_ERROR_SHIFT; 1562 1563 return c; 1564err: 1565 bch_cache_set_unregister(c); 1566 return NULL; 1567} 1568 1569static void run_cache_set(struct cache_set *c) 1570{ 1571 const char *err = "cannot allocate memory"; 1572 struct cached_dev *dc, *t; 1573 struct cache *ca; 1574 struct closure cl; 1575 unsigned i; 1576 1577 closure_init_stack(&cl); 1578 1579 for_each_cache(ca, c, i) 1580 c->nbuckets += ca->sb.nbuckets; 1581 1582 if (CACHE_SYNC(&c->sb)) { 1583 LIST_HEAD(journal); 1584 struct bkey *k; 1585 struct jset *j; 1586 1587 err = "cannot allocate memory for journal"; 1588 if (bch_journal_read(c, &journal)) 1589 goto err; 1590 1591 pr_debug("btree_journal_read() done"); 1592 1593 err = "no journal entries found"; 1594 if (list_empty(&journal)) 1595 goto err; 1596 1597 j = &list_entry(journal.prev, struct journal_replay, list)->j; 1598 1599 err = "IO error reading priorities"; 1600 for_each_cache(ca, c, i) 1601 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]); 1602 1603 /* 1604 * If prio_read() fails it'll call cache_set_error and we'll 1605 * tear everything down right away, but if we perhaps checked 1606 * sooner we could avoid journal replay. 1607 */ 1608 1609 k = &j->btree_root; 1610 1611 err = "bad btree root"; 1612 if (__bch_btree_ptr_invalid(c, k)) 1613 goto err; 1614 1615 err = "error reading btree root"; 1616 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL); 1617 if (IS_ERR_OR_NULL(c->root)) 1618 goto err; 1619 1620 list_del_init(&c->root->list); 1621 rw_unlock(true, c->root); 1622 1623 err = uuid_read(c, j, &cl); 1624 if (err) 1625 goto err; 1626 1627 err = "error in recovery"; 1628 if (bch_btree_check(c)) 1629 goto err; 1630 1631 bch_journal_mark(c, &journal); 1632 bch_initial_gc_finish(c); 1633 pr_debug("btree_check() done"); 1634 1635 /* 1636 * bcache_journal_next() can't happen sooner, or 1637 * btree_gc_finish() will give spurious errors about last_gc > 1638 * gc_gen - this is a hack but oh well. 1639 */ 1640 bch_journal_next(&c->journal); 1641 1642 err = "error starting allocator thread"; 1643 for_each_cache(ca, c, i) 1644 if (bch_cache_allocator_start(ca)) 1645 goto err; 1646 1647 /* 1648 * First place it's safe to allocate: btree_check() and 1649 * btree_gc_finish() have to run before we have buckets to 1650 * allocate, and bch_bucket_alloc_set() might cause a journal 1651 * entry to be written so bcache_journal_next() has to be called 1652 * first. 1653 * 1654 * If the uuids were in the old format we have to rewrite them 1655 * before the next journal entry is written: 1656 */ 1657 if (j->version < BCACHE_JSET_VERSION_UUID) 1658 __uuid_write(c); 1659 1660 bch_journal_replay(c, &journal); 1661 } else { 1662 pr_notice("invalidating existing data"); 1663 1664 for_each_cache(ca, c, i) { 1665 unsigned j; 1666 1667 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, 1668 2, SB_JOURNAL_BUCKETS); 1669 1670 for (j = 0; j < ca->sb.keys; j++) 1671 ca->sb.d[j] = ca->sb.first_bucket + j; 1672 } 1673 1674 bch_initial_gc_finish(c); 1675 1676 err = "error starting allocator thread"; 1677 for_each_cache(ca, c, i) 1678 if (bch_cache_allocator_start(ca)) 1679 goto err; 1680 1681 mutex_lock(&c->bucket_lock); 1682 for_each_cache(ca, c, i) 1683 bch_prio_write(ca); 1684 mutex_unlock(&c->bucket_lock); 1685 1686 err = "cannot allocate new UUID bucket"; 1687 if (__uuid_write(c)) 1688 goto err; 1689 1690 err = "cannot allocate new btree root"; 1691 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL); 1692 if (IS_ERR_OR_NULL(c->root)) 1693 goto err; 1694 1695 mutex_lock(&c->root->write_lock); 1696 bkey_copy_key(&c->root->key, &MAX_KEY); 1697 bch_btree_node_write(c->root, &cl); 1698 mutex_unlock(&c->root->write_lock); 1699 1700 bch_btree_set_root(c->root); 1701 rw_unlock(true, c->root); 1702 1703 /* 1704 * We don't want to write the first journal entry until 1705 * everything is set up - fortunately journal entries won't be 1706 * written until the SET_CACHE_SYNC() here: 1707 */ 1708 SET_CACHE_SYNC(&c->sb, true); 1709 1710 bch_journal_next(&c->journal); 1711 bch_journal_meta(c, &cl); 1712 } 1713 1714 err = "error starting gc thread"; 1715 if (bch_gc_thread_start(c)) 1716 goto err; 1717 1718 closure_sync(&cl); 1719 c->sb.last_mount = get_seconds(); 1720 bcache_write_super(c); 1721 1722 list_for_each_entry_safe(dc, t, &uncached_devices, list) 1723 bch_cached_dev_attach(dc, c); 1724 1725 flash_devs_run(c); 1726 1727 set_bit(CACHE_SET_RUNNING, &c->flags); 1728 return; 1729err: 1730 closure_sync(&cl); 1731 /* XXX: test this, it's broken */ 1732 bch_cache_set_error(c, "%s", err); 1733} 1734 1735static bool can_attach_cache(struct cache *ca, struct cache_set *c) 1736{ 1737 return ca->sb.block_size == c->sb.block_size && 1738 ca->sb.bucket_size == c->sb.bucket_size && 1739 ca->sb.nr_in_set == c->sb.nr_in_set; 1740} 1741 1742static const char *register_cache_set(struct cache *ca) 1743{ 1744 char buf[12]; 1745 const char *err = "cannot allocate memory"; 1746 struct cache_set *c; 1747 1748 list_for_each_entry(c, &bch_cache_sets, list) 1749 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) { 1750 if (c->cache[ca->sb.nr_this_dev]) 1751 return "duplicate cache set member"; 1752 1753 if (!can_attach_cache(ca, c)) 1754 return "cache sb does not match set"; 1755 1756 if (!CACHE_SYNC(&ca->sb)) 1757 SET_CACHE_SYNC(&c->sb, false); 1758 1759 goto found; 1760 } 1761 1762 c = bch_cache_set_alloc(&ca->sb); 1763 if (!c) 1764 return err; 1765 1766 err = "error creating kobject"; 1767 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) || 1768 kobject_add(&c->internal, &c->kobj, "internal")) 1769 goto err; 1770 1771 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) 1772 goto err; 1773 1774 bch_debug_init_cache_set(c); 1775 1776 list_add(&c->list, &bch_cache_sets); 1777found: 1778 sprintf(buf, "cache%i", ca->sb.nr_this_dev); 1779 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || 1780 sysfs_create_link(&c->kobj, &ca->kobj, buf)) 1781 goto err; 1782 1783 if (ca->sb.seq > c->sb.seq) { 1784 c->sb.version = ca->sb.version; 1785 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16); 1786 c->sb.flags = ca->sb.flags; 1787 c->sb.seq = ca->sb.seq; 1788 pr_debug("set version = %llu", c->sb.version); 1789 } 1790 1791 kobject_get(&ca->kobj); 1792 ca->set = c; 1793 ca->set->cache[ca->sb.nr_this_dev] = ca; 1794 c->cache_by_alloc[c->caches_loaded++] = ca; 1795 1796 if (c->caches_loaded == c->sb.nr_in_set) 1797 run_cache_set(c); 1798 1799 return NULL; 1800err: 1801 bch_cache_set_unregister(c); 1802 return err; 1803} 1804 1805/* Cache device */ 1806 1807void bch_cache_release(struct kobject *kobj) 1808{ 1809 struct cache *ca = container_of(kobj, struct cache, kobj); 1810 unsigned i; 1811 1812 if (ca->set) { 1813 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca); 1814 ca->set->cache[ca->sb.nr_this_dev] = NULL; 1815 } 1816 1817 bio_split_pool_free(&ca->bio_split_hook); 1818 1819 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca))); 1820 kfree(ca->prio_buckets); 1821 vfree(ca->buckets); 1822 1823 free_heap(&ca->heap); 1824 free_fifo(&ca->free_inc); 1825 1826 for (i = 0; i < RESERVE_NR; i++) 1827 free_fifo(&ca->free[i]); 1828 1829 if (ca->sb_bio.bi_inline_vecs[0].bv_page) 1830 put_page(ca->sb_bio.bi_io_vec[0].bv_page); 1831 1832 if (!IS_ERR_OR_NULL(ca->bdev)) 1833 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1834 1835 kfree(ca); 1836 module_put(THIS_MODULE); 1837} 1838 1839static int cache_alloc(struct cache_sb *sb, struct cache *ca) 1840{ 1841 size_t free; 1842 struct bucket *b; 1843 1844 __module_get(THIS_MODULE); 1845 kobject_init(&ca->kobj, &bch_cache_ktype); 1846 1847 bio_init(&ca->journal.bio); 1848 ca->journal.bio.bi_max_vecs = 8; 1849 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs; 1850 1851 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; 1852 1853 if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) || 1854 !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) || 1855 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) || 1856 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) || 1857 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) || 1858 !init_heap(&ca->heap, free << 3, GFP_KERNEL) || 1859 !(ca->buckets = vzalloc(sizeof(struct bucket) * 1860 ca->sb.nbuckets)) || 1861 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) * 1862 2, GFP_KERNEL)) || 1863 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) || 1864 bio_split_pool_init(&ca->bio_split_hook)) 1865 return -ENOMEM; 1866 1867 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); 1868 1869 for_each_bucket(b, ca) 1870 atomic_set(&b->pin, 0); 1871 1872 return 0; 1873} 1874 1875static int register_cache(struct cache_sb *sb, struct page *sb_page, 1876 struct block_device *bdev, struct cache *ca) 1877{ 1878 char name[BDEVNAME_SIZE]; 1879 const char *err = NULL; 1880 int ret = 0; 1881 1882 memcpy(&ca->sb, sb, sizeof(struct cache_sb)); 1883 ca->bdev = bdev; 1884 ca->bdev->bd_holder = ca; 1885 1886 bio_init(&ca->sb_bio); 1887 ca->sb_bio.bi_max_vecs = 1; 1888 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs; 1889 ca->sb_bio.bi_io_vec[0].bv_page = sb_page; 1890 get_page(sb_page); 1891 1892 if (blk_queue_discard(bdev_get_queue(ca->bdev))) 1893 ca->discard = CACHE_DISCARD(&ca->sb); 1894 1895 ret = cache_alloc(sb, ca); 1896 if (ret != 0) 1897 goto err; 1898 1899 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) { 1900 err = "error calling kobject_add"; 1901 ret = -ENOMEM; 1902 goto out; 1903 } 1904 1905 mutex_lock(&bch_register_lock); 1906 err = register_cache_set(ca); 1907 mutex_unlock(&bch_register_lock); 1908 1909 if (err) { 1910 ret = -ENODEV; 1911 goto out; 1912 } 1913 1914 pr_info("registered cache device %s", bdevname(bdev, name)); 1915 1916out: 1917 kobject_put(&ca->kobj); 1918 1919err: 1920 if (err) 1921 pr_notice("error opening %s: %s", bdevname(bdev, name), err); 1922 1923 return ret; 1924} 1925 1926/* Global interfaces/init */ 1927 1928static ssize_t register_bcache(struct kobject *, struct kobj_attribute *, 1929 const char *, size_t); 1930 1931kobj_attribute_write(register, register_bcache); 1932kobj_attribute_write(register_quiet, register_bcache); 1933 1934static bool bch_is_open_backing(struct block_device *bdev) { 1935 struct cache_set *c, *tc; 1936 struct cached_dev *dc, *t; 1937 1938 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 1939 list_for_each_entry_safe(dc, t, &c->cached_devs, list) 1940 if (dc->bdev == bdev) 1941 return true; 1942 list_for_each_entry_safe(dc, t, &uncached_devices, list) 1943 if (dc->bdev == bdev) 1944 return true; 1945 return false; 1946} 1947 1948static bool bch_is_open_cache(struct block_device *bdev) { 1949 struct cache_set *c, *tc; 1950 struct cache *ca; 1951 unsigned i; 1952 1953 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 1954 for_each_cache(ca, c, i) 1955 if (ca->bdev == bdev) 1956 return true; 1957 return false; 1958} 1959 1960static bool bch_is_open(struct block_device *bdev) { 1961 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev); 1962} 1963 1964static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, 1965 const char *buffer, size_t size) 1966{ 1967 ssize_t ret = size; 1968 const char *err = "cannot allocate memory"; 1969 char *path = NULL; 1970 struct cache_sb *sb = NULL; 1971 struct block_device *bdev = NULL; 1972 struct page *sb_page = NULL; 1973 1974 if (!try_module_get(THIS_MODULE)) 1975 return -EBUSY; 1976 1977 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) || 1978 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL))) 1979 goto err; 1980 1981 err = "failed to open device"; 1982 bdev = blkdev_get_by_path(strim(path), 1983 FMODE_READ|FMODE_WRITE|FMODE_EXCL, 1984 sb); 1985 if (IS_ERR(bdev)) { 1986 if (bdev == ERR_PTR(-EBUSY)) { 1987 bdev = lookup_bdev(strim(path)); 1988 mutex_lock(&bch_register_lock); 1989 if (!IS_ERR(bdev) && bch_is_open(bdev)) 1990 err = "device already registered"; 1991 else 1992 err = "device busy"; 1993 mutex_unlock(&bch_register_lock); 1994 if (attr == &ksysfs_register_quiet) 1995 goto out; 1996 } 1997 goto err; 1998 } 1999 2000 err = "failed to set blocksize"; 2001 if (set_blocksize(bdev, 4096)) 2002 goto err_close; 2003 2004 err = read_super(sb, bdev, &sb_page); 2005 if (err) 2006 goto err_close; 2007 2008 if (SB_IS_BDEV(sb)) { 2009 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL); 2010 if (!dc) 2011 goto err_close; 2012 2013 mutex_lock(&bch_register_lock); 2014 register_bdev(sb, sb_page, bdev, dc); 2015 mutex_unlock(&bch_register_lock); 2016 } else { 2017 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL); 2018 if (!ca) 2019 goto err_close; 2020 2021 if (register_cache(sb, sb_page, bdev, ca) != 0) 2022 goto err_close; 2023 } 2024out: 2025 if (sb_page) 2026 put_page(sb_page); 2027 kfree(sb); 2028 kfree(path); 2029 module_put(THIS_MODULE); 2030 return ret; 2031 2032err_close: 2033 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2034err: 2035 pr_info("error opening %s: %s", path, err); 2036 ret = -EINVAL; 2037 goto out; 2038} 2039 2040static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) 2041{ 2042 if (code == SYS_DOWN || 2043 code == SYS_HALT || 2044 code == SYS_POWER_OFF) { 2045 DEFINE_WAIT(wait); 2046 unsigned long start = jiffies; 2047 bool stopped = false; 2048 2049 struct cache_set *c, *tc; 2050 struct cached_dev *dc, *tdc; 2051 2052 mutex_lock(&bch_register_lock); 2053 2054 if (list_empty(&bch_cache_sets) && 2055 list_empty(&uncached_devices)) 2056 goto out; 2057 2058 pr_info("Stopping all devices:"); 2059 2060 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2061 bch_cache_set_stop(c); 2062 2063 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) 2064 bcache_device_stop(&dc->disk); 2065 2066 /* What's a condition variable? */ 2067 while (1) { 2068 long timeout = start + 2 * HZ - jiffies; 2069 2070 stopped = list_empty(&bch_cache_sets) && 2071 list_empty(&uncached_devices); 2072 2073 if (timeout < 0 || stopped) 2074 break; 2075 2076 prepare_to_wait(&unregister_wait, &wait, 2077 TASK_UNINTERRUPTIBLE); 2078 2079 mutex_unlock(&bch_register_lock); 2080 schedule_timeout(timeout); 2081 mutex_lock(&bch_register_lock); 2082 } 2083 2084 finish_wait(&unregister_wait, &wait); 2085 2086 if (stopped) 2087 pr_info("All devices stopped"); 2088 else 2089 pr_notice("Timeout waiting for devices to be closed"); 2090out: 2091 mutex_unlock(&bch_register_lock); 2092 } 2093 2094 return NOTIFY_DONE; 2095} 2096 2097static struct notifier_block reboot = { 2098 .notifier_call = bcache_reboot, 2099 .priority = INT_MAX, /* before any real devices */ 2100}; 2101 2102static void bcache_exit(void) 2103{ 2104 bch_debug_exit(); 2105 bch_request_exit(); 2106 if (bcache_kobj) 2107 kobject_put(bcache_kobj); 2108 if (bcache_wq) 2109 destroy_workqueue(bcache_wq); 2110 if (bcache_major) 2111 unregister_blkdev(bcache_major, "bcache"); 2112 unregister_reboot_notifier(&reboot); 2113} 2114 2115static int __init bcache_init(void) 2116{ 2117 static const struct attribute *files[] = { 2118 &ksysfs_register.attr, 2119 &ksysfs_register_quiet.attr, 2120 NULL 2121 }; 2122 2123 mutex_init(&bch_register_lock); 2124 init_waitqueue_head(&unregister_wait); 2125 register_reboot_notifier(&reboot); 2126 closure_debug_init(); 2127 2128 bcache_major = register_blkdev(0, "bcache"); 2129 if (bcache_major < 0) { 2130 unregister_reboot_notifier(&reboot); 2131 return bcache_major; 2132 } 2133 2134 if (!(bcache_wq = create_workqueue("bcache")) || 2135 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) || 2136 sysfs_create_files(bcache_kobj, files) || 2137 bch_request_init() || 2138 bch_debug_init(bcache_kobj)) 2139 goto err; 2140 2141 return 0; 2142err: 2143 bcache_exit(); 2144 return -ENOMEM; 2145} 2146 2147module_exit(bcache_exit); 2148module_init(bcache_init); 2149