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 /* Default is -1; we skip past it for stop_when_cache_set_failed */ 51 const char * const bch_stop_on_failure_modes[] = { 52 "default", 53 "auto", 54 "always", 55 NULL 56 }; 57 58 static struct kobject *bcache_kobj; 59 struct mutex bch_register_lock; 60 LIST_HEAD(bch_cache_sets); 61 static LIST_HEAD(uncached_devices); 62 63 static int bcache_major; 64 static DEFINE_IDA(bcache_device_idx); 65 static wait_queue_head_t unregister_wait; 66 struct workqueue_struct *bcache_wq; 67 68 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE) 69 /* limitation of partitions number on single bcache device */ 70 #define BCACHE_MINORS 128 71 /* limitation of bcache devices number on single system */ 72 #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS) 73 74 /* Superblock */ 75 76 static const char *read_super(struct cache_sb *sb, struct block_device *bdev, 77 struct page **res) 78 { 79 const char *err; 80 struct cache_sb *s; 81 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE); 82 unsigned i; 83 84 if (!bh) 85 return "IO error"; 86 87 s = (struct cache_sb *) bh->b_data; 88 89 sb->offset = le64_to_cpu(s->offset); 90 sb->version = le64_to_cpu(s->version); 91 92 memcpy(sb->magic, s->magic, 16); 93 memcpy(sb->uuid, s->uuid, 16); 94 memcpy(sb->set_uuid, s->set_uuid, 16); 95 memcpy(sb->label, s->label, SB_LABEL_SIZE); 96 97 sb->flags = le64_to_cpu(s->flags); 98 sb->seq = le64_to_cpu(s->seq); 99 sb->last_mount = le32_to_cpu(s->last_mount); 100 sb->first_bucket = le16_to_cpu(s->first_bucket); 101 sb->keys = le16_to_cpu(s->keys); 102 103 for (i = 0; i < SB_JOURNAL_BUCKETS; i++) 104 sb->d[i] = le64_to_cpu(s->d[i]); 105 106 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u", 107 sb->version, sb->flags, sb->seq, sb->keys); 108 109 err = "Not a bcache superblock"; 110 if (sb->offset != SB_SECTOR) 111 goto err; 112 113 if (memcmp(sb->magic, bcache_magic, 16)) 114 goto err; 115 116 err = "Too many journal buckets"; 117 if (sb->keys > SB_JOURNAL_BUCKETS) 118 goto err; 119 120 err = "Bad checksum"; 121 if (s->csum != csum_set(s)) 122 goto err; 123 124 err = "Bad UUID"; 125 if (bch_is_zero(sb->uuid, 16)) 126 goto err; 127 128 sb->block_size = le16_to_cpu(s->block_size); 129 130 err = "Superblock block size smaller than device block size"; 131 if (sb->block_size << 9 < bdev_logical_block_size(bdev)) 132 goto err; 133 134 switch (sb->version) { 135 case BCACHE_SB_VERSION_BDEV: 136 sb->data_offset = BDEV_DATA_START_DEFAULT; 137 break; 138 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET: 139 sb->data_offset = le64_to_cpu(s->data_offset); 140 141 err = "Bad data offset"; 142 if (sb->data_offset < BDEV_DATA_START_DEFAULT) 143 goto err; 144 145 break; 146 case BCACHE_SB_VERSION_CDEV: 147 case BCACHE_SB_VERSION_CDEV_WITH_UUID: 148 sb->nbuckets = le64_to_cpu(s->nbuckets); 149 sb->bucket_size = le16_to_cpu(s->bucket_size); 150 151 sb->nr_in_set = le16_to_cpu(s->nr_in_set); 152 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev); 153 154 err = "Too many buckets"; 155 if (sb->nbuckets > LONG_MAX) 156 goto err; 157 158 err = "Not enough buckets"; 159 if (sb->nbuckets < 1 << 7) 160 goto err; 161 162 err = "Bad block/bucket size"; 163 if (!is_power_of_2(sb->block_size) || 164 sb->block_size > PAGE_SECTORS || 165 !is_power_of_2(sb->bucket_size) || 166 sb->bucket_size < PAGE_SECTORS) 167 goto err; 168 169 err = "Invalid superblock: device too small"; 170 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets) 171 goto err; 172 173 err = "Bad UUID"; 174 if (bch_is_zero(sb->set_uuid, 16)) 175 goto err; 176 177 err = "Bad cache device number in set"; 178 if (!sb->nr_in_set || 179 sb->nr_in_set <= sb->nr_this_dev || 180 sb->nr_in_set > MAX_CACHES_PER_SET) 181 goto err; 182 183 err = "Journal buckets not sequential"; 184 for (i = 0; i < sb->keys; i++) 185 if (sb->d[i] != sb->first_bucket + i) 186 goto err; 187 188 err = "Too many journal buckets"; 189 if (sb->first_bucket + sb->keys > sb->nbuckets) 190 goto err; 191 192 err = "Invalid superblock: first bucket comes before end of super"; 193 if (sb->first_bucket * sb->bucket_size < 16) 194 goto err; 195 196 break; 197 default: 198 err = "Unsupported superblock version"; 199 goto err; 200 } 201 202 sb->last_mount = get_seconds(); 203 err = NULL; 204 205 get_page(bh->b_page); 206 *res = bh->b_page; 207 err: 208 put_bh(bh); 209 return err; 210 } 211 212 static void write_bdev_super_endio(struct bio *bio) 213 { 214 struct cached_dev *dc = bio->bi_private; 215 /* XXX: error checking */ 216 217 closure_put(&dc->sb_write); 218 } 219 220 static void __write_super(struct cache_sb *sb, struct bio *bio) 221 { 222 struct cache_sb *out = page_address(bio_first_page_all(bio)); 223 unsigned i; 224 225 bio->bi_iter.bi_sector = SB_SECTOR; 226 bio->bi_iter.bi_size = SB_SIZE; 227 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META); 228 bch_bio_map(bio, NULL); 229 230 out->offset = cpu_to_le64(sb->offset); 231 out->version = cpu_to_le64(sb->version); 232 233 memcpy(out->uuid, sb->uuid, 16); 234 memcpy(out->set_uuid, sb->set_uuid, 16); 235 memcpy(out->label, sb->label, SB_LABEL_SIZE); 236 237 out->flags = cpu_to_le64(sb->flags); 238 out->seq = cpu_to_le64(sb->seq); 239 240 out->last_mount = cpu_to_le32(sb->last_mount); 241 out->first_bucket = cpu_to_le16(sb->first_bucket); 242 out->keys = cpu_to_le16(sb->keys); 243 244 for (i = 0; i < sb->keys; i++) 245 out->d[i] = cpu_to_le64(sb->d[i]); 246 247 out->csum = csum_set(out); 248 249 pr_debug("ver %llu, flags %llu, seq %llu", 250 sb->version, sb->flags, sb->seq); 251 252 submit_bio(bio); 253 } 254 255 static void bch_write_bdev_super_unlock(struct closure *cl) 256 { 257 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write); 258 259 up(&dc->sb_write_mutex); 260 } 261 262 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent) 263 { 264 struct closure *cl = &dc->sb_write; 265 struct bio *bio = &dc->sb_bio; 266 267 down(&dc->sb_write_mutex); 268 closure_init(cl, parent); 269 270 bio_reset(bio); 271 bio_set_dev(bio, dc->bdev); 272 bio->bi_end_io = write_bdev_super_endio; 273 bio->bi_private = dc; 274 275 closure_get(cl); 276 /* I/O request sent to backing device */ 277 __write_super(&dc->sb, bio); 278 279 closure_return_with_destructor(cl, bch_write_bdev_super_unlock); 280 } 281 282 static void write_super_endio(struct bio *bio) 283 { 284 struct cache *ca = bio->bi_private; 285 286 /* is_read = 0 */ 287 bch_count_io_errors(ca, bio->bi_status, 0, 288 "writing superblock"); 289 closure_put(&ca->set->sb_write); 290 } 291 292 static void bcache_write_super_unlock(struct closure *cl) 293 { 294 struct cache_set *c = container_of(cl, struct cache_set, sb_write); 295 296 up(&c->sb_write_mutex); 297 } 298 299 void bcache_write_super(struct cache_set *c) 300 { 301 struct closure *cl = &c->sb_write; 302 struct cache *ca; 303 unsigned i; 304 305 down(&c->sb_write_mutex); 306 closure_init(cl, &c->cl); 307 308 c->sb.seq++; 309 310 for_each_cache(ca, c, i) { 311 struct bio *bio = &ca->sb_bio; 312 313 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID; 314 ca->sb.seq = c->sb.seq; 315 ca->sb.last_mount = c->sb.last_mount; 316 317 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb)); 318 319 bio_reset(bio); 320 bio_set_dev(bio, ca->bdev); 321 bio->bi_end_io = write_super_endio; 322 bio->bi_private = ca; 323 324 closure_get(cl); 325 __write_super(&ca->sb, bio); 326 } 327 328 closure_return_with_destructor(cl, bcache_write_super_unlock); 329 } 330 331 /* UUID io */ 332 333 static void uuid_endio(struct bio *bio) 334 { 335 struct closure *cl = bio->bi_private; 336 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 337 338 cache_set_err_on(bio->bi_status, c, "accessing uuids"); 339 bch_bbio_free(bio, c); 340 closure_put(cl); 341 } 342 343 static void uuid_io_unlock(struct closure *cl) 344 { 345 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 346 347 up(&c->uuid_write_mutex); 348 } 349 350 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags, 351 struct bkey *k, struct closure *parent) 352 { 353 struct closure *cl = &c->uuid_write; 354 struct uuid_entry *u; 355 unsigned i; 356 char buf[80]; 357 358 BUG_ON(!parent); 359 down(&c->uuid_write_mutex); 360 closure_init(cl, parent); 361 362 for (i = 0; i < KEY_PTRS(k); i++) { 363 struct bio *bio = bch_bbio_alloc(c); 364 365 bio->bi_opf = REQ_SYNC | REQ_META | op_flags; 366 bio->bi_iter.bi_size = KEY_SIZE(k) << 9; 367 368 bio->bi_end_io = uuid_endio; 369 bio->bi_private = cl; 370 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags); 371 bch_bio_map(bio, c->uuids); 372 373 bch_submit_bbio(bio, c, k, i); 374 375 if (op != REQ_OP_WRITE) 376 break; 377 } 378 379 bch_extent_to_text(buf, sizeof(buf), k); 380 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf); 381 382 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++) 383 if (!bch_is_zero(u->uuid, 16)) 384 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u", 385 u - c->uuids, u->uuid, u->label, 386 u->first_reg, u->last_reg, u->invalidated); 387 388 closure_return_with_destructor(cl, uuid_io_unlock); 389 } 390 391 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl) 392 { 393 struct bkey *k = &j->uuid_bucket; 394 395 if (__bch_btree_ptr_invalid(c, k)) 396 return "bad uuid pointer"; 397 398 bkey_copy(&c->uuid_bucket, k); 399 uuid_io(c, REQ_OP_READ, 0, k, cl); 400 401 if (j->version < BCACHE_JSET_VERSION_UUIDv1) { 402 struct uuid_entry_v0 *u0 = (void *) c->uuids; 403 struct uuid_entry *u1 = (void *) c->uuids; 404 int i; 405 406 closure_sync(cl); 407 408 /* 409 * Since the new uuid entry is bigger than the old, we have to 410 * convert starting at the highest memory address and work down 411 * in order to do it in place 412 */ 413 414 for (i = c->nr_uuids - 1; 415 i >= 0; 416 --i) { 417 memcpy(u1[i].uuid, u0[i].uuid, 16); 418 memcpy(u1[i].label, u0[i].label, 32); 419 420 u1[i].first_reg = u0[i].first_reg; 421 u1[i].last_reg = u0[i].last_reg; 422 u1[i].invalidated = u0[i].invalidated; 423 424 u1[i].flags = 0; 425 u1[i].sectors = 0; 426 } 427 } 428 429 return NULL; 430 } 431 432 static int __uuid_write(struct cache_set *c) 433 { 434 BKEY_PADDED(key) k; 435 struct closure cl; 436 closure_init_stack(&cl); 437 438 lockdep_assert_held(&bch_register_lock); 439 440 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true)) 441 return 1; 442 443 SET_KEY_SIZE(&k.key, c->sb.bucket_size); 444 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl); 445 closure_sync(&cl); 446 447 bkey_copy(&c->uuid_bucket, &k.key); 448 bkey_put(c, &k.key); 449 return 0; 450 } 451 452 int bch_uuid_write(struct cache_set *c) 453 { 454 int ret = __uuid_write(c); 455 456 if (!ret) 457 bch_journal_meta(c, NULL); 458 459 return ret; 460 } 461 462 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid) 463 { 464 struct uuid_entry *u; 465 466 for (u = c->uuids; 467 u < c->uuids + c->nr_uuids; u++) 468 if (!memcmp(u->uuid, uuid, 16)) 469 return u; 470 471 return NULL; 472 } 473 474 static struct uuid_entry *uuid_find_empty(struct cache_set *c) 475 { 476 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"; 477 return uuid_find(c, zero_uuid); 478 } 479 480 /* 481 * Bucket priorities/gens: 482 * 483 * For each bucket, we store on disk its 484 * 8 bit gen 485 * 16 bit priority 486 * 487 * See alloc.c for an explanation of the gen. The priority is used to implement 488 * lru (and in the future other) cache replacement policies; for most purposes 489 * it's just an opaque integer. 490 * 491 * The gens and the priorities don't have a whole lot to do with each other, and 492 * it's actually the gens that must be written out at specific times - it's no 493 * big deal if the priorities don't get written, if we lose them we just reuse 494 * buckets in suboptimal order. 495 * 496 * On disk they're stored in a packed array, and in as many buckets are required 497 * to fit them all. The buckets we use to store them form a list; the journal 498 * header points to the first bucket, the first bucket points to the second 499 * bucket, et cetera. 500 * 501 * This code is used by the allocation code; periodically (whenever it runs out 502 * of buckets to allocate from) the allocation code will invalidate some 503 * buckets, but it can't use those buckets until their new gens are safely on 504 * disk. 505 */ 506 507 static void prio_endio(struct bio *bio) 508 { 509 struct cache *ca = bio->bi_private; 510 511 cache_set_err_on(bio->bi_status, ca->set, "accessing priorities"); 512 bch_bbio_free(bio, ca->set); 513 closure_put(&ca->prio); 514 } 515 516 static void prio_io(struct cache *ca, uint64_t bucket, int op, 517 unsigned long op_flags) 518 { 519 struct closure *cl = &ca->prio; 520 struct bio *bio = bch_bbio_alloc(ca->set); 521 522 closure_init_stack(cl); 523 524 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size; 525 bio_set_dev(bio, ca->bdev); 526 bio->bi_iter.bi_size = bucket_bytes(ca); 527 528 bio->bi_end_io = prio_endio; 529 bio->bi_private = ca; 530 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags); 531 bch_bio_map(bio, ca->disk_buckets); 532 533 closure_bio_submit(ca->set, bio, &ca->prio); 534 closure_sync(cl); 535 } 536 537 void bch_prio_write(struct cache *ca) 538 { 539 int i; 540 struct bucket *b; 541 struct closure cl; 542 543 closure_init_stack(&cl); 544 545 lockdep_assert_held(&ca->set->bucket_lock); 546 547 ca->disk_buckets->seq++; 548 549 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca), 550 &ca->meta_sectors_written); 551 552 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free), 553 // fifo_used(&ca->free_inc), fifo_used(&ca->unused)); 554 555 for (i = prio_buckets(ca) - 1; i >= 0; --i) { 556 long bucket; 557 struct prio_set *p = ca->disk_buckets; 558 struct bucket_disk *d = p->data; 559 struct bucket_disk *end = d + prios_per_bucket(ca); 560 561 for (b = ca->buckets + i * prios_per_bucket(ca); 562 b < ca->buckets + ca->sb.nbuckets && d < end; 563 b++, d++) { 564 d->prio = cpu_to_le16(b->prio); 565 d->gen = b->gen; 566 } 567 568 p->next_bucket = ca->prio_buckets[i + 1]; 569 p->magic = pset_magic(&ca->sb); 570 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8); 571 572 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true); 573 BUG_ON(bucket == -1); 574 575 mutex_unlock(&ca->set->bucket_lock); 576 prio_io(ca, bucket, REQ_OP_WRITE, 0); 577 mutex_lock(&ca->set->bucket_lock); 578 579 ca->prio_buckets[i] = bucket; 580 atomic_dec_bug(&ca->buckets[bucket].pin); 581 } 582 583 mutex_unlock(&ca->set->bucket_lock); 584 585 bch_journal_meta(ca->set, &cl); 586 closure_sync(&cl); 587 588 mutex_lock(&ca->set->bucket_lock); 589 590 /* 591 * Don't want the old priorities to get garbage collected until after we 592 * finish writing the new ones, and they're journalled 593 */ 594 for (i = 0; i < prio_buckets(ca); i++) { 595 if (ca->prio_last_buckets[i]) 596 __bch_bucket_free(ca, 597 &ca->buckets[ca->prio_last_buckets[i]]); 598 599 ca->prio_last_buckets[i] = ca->prio_buckets[i]; 600 } 601 } 602 603 static void prio_read(struct cache *ca, uint64_t bucket) 604 { 605 struct prio_set *p = ca->disk_buckets; 606 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d; 607 struct bucket *b; 608 unsigned bucket_nr = 0; 609 610 for (b = ca->buckets; 611 b < ca->buckets + ca->sb.nbuckets; 612 b++, d++) { 613 if (d == end) { 614 ca->prio_buckets[bucket_nr] = bucket; 615 ca->prio_last_buckets[bucket_nr] = bucket; 616 bucket_nr++; 617 618 prio_io(ca, bucket, REQ_OP_READ, 0); 619 620 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8)) 621 pr_warn("bad csum reading priorities"); 622 623 if (p->magic != pset_magic(&ca->sb)) 624 pr_warn("bad magic reading priorities"); 625 626 bucket = p->next_bucket; 627 d = p->data; 628 } 629 630 b->prio = le16_to_cpu(d->prio); 631 b->gen = b->last_gc = d->gen; 632 } 633 } 634 635 /* Bcache device */ 636 637 static int open_dev(struct block_device *b, fmode_t mode) 638 { 639 struct bcache_device *d = b->bd_disk->private_data; 640 if (test_bit(BCACHE_DEV_CLOSING, &d->flags)) 641 return -ENXIO; 642 643 closure_get(&d->cl); 644 return 0; 645 } 646 647 static void release_dev(struct gendisk *b, fmode_t mode) 648 { 649 struct bcache_device *d = b->private_data; 650 closure_put(&d->cl); 651 } 652 653 static int ioctl_dev(struct block_device *b, fmode_t mode, 654 unsigned int cmd, unsigned long arg) 655 { 656 struct bcache_device *d = b->bd_disk->private_data; 657 return d->ioctl(d, mode, cmd, arg); 658 } 659 660 static const struct block_device_operations bcache_ops = { 661 .open = open_dev, 662 .release = release_dev, 663 .ioctl = ioctl_dev, 664 .owner = THIS_MODULE, 665 }; 666 667 void bcache_device_stop(struct bcache_device *d) 668 { 669 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags)) 670 closure_queue(&d->cl); 671 } 672 673 static void bcache_device_unlink(struct bcache_device *d) 674 { 675 lockdep_assert_held(&bch_register_lock); 676 677 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) { 678 unsigned i; 679 struct cache *ca; 680 681 sysfs_remove_link(&d->c->kobj, d->name); 682 sysfs_remove_link(&d->kobj, "cache"); 683 684 for_each_cache(ca, d->c, i) 685 bd_unlink_disk_holder(ca->bdev, d->disk); 686 } 687 } 688 689 static void bcache_device_link(struct bcache_device *d, struct cache_set *c, 690 const char *name) 691 { 692 unsigned i; 693 struct cache *ca; 694 695 for_each_cache(ca, d->c, i) 696 bd_link_disk_holder(ca->bdev, d->disk); 697 698 snprintf(d->name, BCACHEDEVNAME_SIZE, 699 "%s%u", name, d->id); 700 701 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") || 702 sysfs_create_link(&c->kobj, &d->kobj, d->name), 703 "Couldn't create device <-> cache set symlinks"); 704 705 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags); 706 } 707 708 static void bcache_device_detach(struct bcache_device *d) 709 { 710 lockdep_assert_held(&bch_register_lock); 711 712 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) { 713 struct uuid_entry *u = d->c->uuids + d->id; 714 715 SET_UUID_FLASH_ONLY(u, 0); 716 memcpy(u->uuid, invalid_uuid, 16); 717 u->invalidated = cpu_to_le32(get_seconds()); 718 bch_uuid_write(d->c); 719 } 720 721 bcache_device_unlink(d); 722 723 d->c->devices[d->id] = NULL; 724 closure_put(&d->c->caching); 725 d->c = NULL; 726 } 727 728 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, 729 unsigned id) 730 { 731 d->id = id; 732 d->c = c; 733 c->devices[id] = d; 734 735 if (id >= c->devices_max_used) 736 c->devices_max_used = id + 1; 737 738 closure_get(&c->caching); 739 } 740 741 static inline int first_minor_to_idx(int first_minor) 742 { 743 return (first_minor/BCACHE_MINORS); 744 } 745 746 static inline int idx_to_first_minor(int idx) 747 { 748 return (idx * BCACHE_MINORS); 749 } 750 751 static void bcache_device_free(struct bcache_device *d) 752 { 753 lockdep_assert_held(&bch_register_lock); 754 755 pr_info("%s stopped", d->disk->disk_name); 756 757 if (d->c) 758 bcache_device_detach(d); 759 if (d->disk && d->disk->flags & GENHD_FL_UP) 760 del_gendisk(d->disk); 761 if (d->disk && d->disk->queue) 762 blk_cleanup_queue(d->disk->queue); 763 if (d->disk) { 764 ida_simple_remove(&bcache_device_idx, 765 first_minor_to_idx(d->disk->first_minor)); 766 put_disk(d->disk); 767 } 768 769 if (d->bio_split) 770 bioset_free(d->bio_split); 771 kvfree(d->full_dirty_stripes); 772 kvfree(d->stripe_sectors_dirty); 773 774 closure_debug_destroy(&d->cl); 775 } 776 777 static int bcache_device_init(struct bcache_device *d, unsigned block_size, 778 sector_t sectors) 779 { 780 struct request_queue *q; 781 const size_t max_stripes = min_t(size_t, INT_MAX, 782 SIZE_MAX / sizeof(atomic_t)); 783 size_t n; 784 int idx; 785 786 if (!d->stripe_size) 787 d->stripe_size = 1 << 31; 788 789 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size); 790 791 if (!d->nr_stripes || d->nr_stripes > max_stripes) { 792 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)", 793 (unsigned)d->nr_stripes); 794 return -ENOMEM; 795 } 796 797 n = d->nr_stripes * sizeof(atomic_t); 798 d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL); 799 if (!d->stripe_sectors_dirty) 800 return -ENOMEM; 801 802 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long); 803 d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL); 804 if (!d->full_dirty_stripes) 805 return -ENOMEM; 806 807 idx = ida_simple_get(&bcache_device_idx, 0, 808 BCACHE_DEVICE_IDX_MAX, GFP_KERNEL); 809 if (idx < 0) 810 return idx; 811 812 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio), 813 BIOSET_NEED_BVECS | 814 BIOSET_NEED_RESCUER)) || 815 !(d->disk = alloc_disk(BCACHE_MINORS))) { 816 ida_simple_remove(&bcache_device_idx, idx); 817 return -ENOMEM; 818 } 819 820 set_capacity(d->disk, sectors); 821 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx); 822 823 d->disk->major = bcache_major; 824 d->disk->first_minor = idx_to_first_minor(idx); 825 d->disk->fops = &bcache_ops; 826 d->disk->private_data = d; 827 828 q = blk_alloc_queue(GFP_KERNEL); 829 if (!q) 830 return -ENOMEM; 831 832 blk_queue_make_request(q, NULL); 833 d->disk->queue = q; 834 q->queuedata = d; 835 q->backing_dev_info->congested_data = d; 836 q->limits.max_hw_sectors = UINT_MAX; 837 q->limits.max_sectors = UINT_MAX; 838 q->limits.max_segment_size = UINT_MAX; 839 q->limits.max_segments = BIO_MAX_PAGES; 840 blk_queue_max_discard_sectors(q, UINT_MAX); 841 q->limits.discard_granularity = 512; 842 q->limits.io_min = block_size; 843 q->limits.logical_block_size = block_size; 844 q->limits.physical_block_size = block_size; 845 blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue); 846 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, d->disk->queue); 847 blk_queue_flag_set(QUEUE_FLAG_DISCARD, d->disk->queue); 848 849 blk_queue_write_cache(q, true, true); 850 851 return 0; 852 } 853 854 /* Cached device */ 855 856 static void calc_cached_dev_sectors(struct cache_set *c) 857 { 858 uint64_t sectors = 0; 859 struct cached_dev *dc; 860 861 list_for_each_entry(dc, &c->cached_devs, list) 862 sectors += bdev_sectors(dc->bdev); 863 864 c->cached_dev_sectors = sectors; 865 } 866 867 void bch_cached_dev_run(struct cached_dev *dc) 868 { 869 struct bcache_device *d = &dc->disk; 870 char buf[SB_LABEL_SIZE + 1]; 871 char *env[] = { 872 "DRIVER=bcache", 873 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), 874 NULL, 875 NULL, 876 }; 877 878 memcpy(buf, dc->sb.label, SB_LABEL_SIZE); 879 buf[SB_LABEL_SIZE] = '\0'; 880 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf); 881 882 if (atomic_xchg(&dc->running, 1)) { 883 kfree(env[1]); 884 kfree(env[2]); 885 return; 886 } 887 888 if (!d->c && 889 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { 890 struct closure cl; 891 closure_init_stack(&cl); 892 893 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); 894 bch_write_bdev_super(dc, &cl); 895 closure_sync(&cl); 896 } 897 898 add_disk(d->disk); 899 bd_link_disk_holder(dc->bdev, dc->disk.disk); 900 /* won't show up in the uevent file, use udevadm monitor -e instead 901 * only class / kset properties are persistent */ 902 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); 903 kfree(env[1]); 904 kfree(env[2]); 905 906 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || 907 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache")) 908 pr_debug("error creating sysfs link"); 909 } 910 911 /* 912 * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed 913 * work dc->writeback_rate_update is running. Wait until the routine 914 * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to 915 * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out 916 * seconds, give up waiting here and continue to cancel it too. 917 */ 918 static void cancel_writeback_rate_update_dwork(struct cached_dev *dc) 919 { 920 int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ; 921 922 do { 923 if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING, 924 &dc->disk.flags)) 925 break; 926 time_out--; 927 schedule_timeout_interruptible(1); 928 } while (time_out > 0); 929 930 if (time_out == 0) 931 pr_warn("give up waiting for dc->writeback_write_update to quit"); 932 933 cancel_delayed_work_sync(&dc->writeback_rate_update); 934 } 935 936 static void cached_dev_detach_finish(struct work_struct *w) 937 { 938 struct cached_dev *dc = container_of(w, struct cached_dev, detach); 939 char buf[BDEVNAME_SIZE]; 940 struct closure cl; 941 closure_init_stack(&cl); 942 943 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)); 944 BUG_ON(refcount_read(&dc->count)); 945 946 mutex_lock(&bch_register_lock); 947 948 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) 949 cancel_writeback_rate_update_dwork(dc); 950 951 if (!IS_ERR_OR_NULL(dc->writeback_thread)) { 952 kthread_stop(dc->writeback_thread); 953 dc->writeback_thread = NULL; 954 } 955 956 memset(&dc->sb.set_uuid, 0, 16); 957 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE); 958 959 bch_write_bdev_super(dc, &cl); 960 closure_sync(&cl); 961 962 bcache_device_detach(&dc->disk); 963 list_move(&dc->list, &uncached_devices); 964 965 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags); 966 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags); 967 968 mutex_unlock(&bch_register_lock); 969 970 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf)); 971 972 /* Drop ref we took in cached_dev_detach() */ 973 closure_put(&dc->disk.cl); 974 } 975 976 void bch_cached_dev_detach(struct cached_dev *dc) 977 { 978 lockdep_assert_held(&bch_register_lock); 979 980 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) 981 return; 982 983 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) 984 return; 985 986 /* 987 * Block the device from being closed and freed until we're finished 988 * detaching 989 */ 990 closure_get(&dc->disk.cl); 991 992 bch_writeback_queue(dc); 993 994 cached_dev_put(dc); 995 } 996 997 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c, 998 uint8_t *set_uuid) 999 { 1000 uint32_t rtime = cpu_to_le32(get_seconds()); 1001 struct uuid_entry *u; 1002 char buf[BDEVNAME_SIZE]; 1003 struct cached_dev *exist_dc, *t; 1004 1005 bdevname(dc->bdev, buf); 1006 1007 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) || 1008 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))) 1009 return -ENOENT; 1010 1011 if (dc->disk.c) { 1012 pr_err("Can't attach %s: already attached", buf); 1013 return -EINVAL; 1014 } 1015 1016 if (test_bit(CACHE_SET_STOPPING, &c->flags)) { 1017 pr_err("Can't attach %s: shutting down", buf); 1018 return -EINVAL; 1019 } 1020 1021 if (dc->sb.block_size < c->sb.block_size) { 1022 /* Will die */ 1023 pr_err("Couldn't attach %s: block size less than set's block size", 1024 buf); 1025 return -EINVAL; 1026 } 1027 1028 /* Check whether already attached */ 1029 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) { 1030 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) { 1031 pr_err("Tried to attach %s but duplicate UUID already attached", 1032 buf); 1033 1034 return -EINVAL; 1035 } 1036 } 1037 1038 u = uuid_find(c, dc->sb.uuid); 1039 1040 if (u && 1041 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || 1042 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { 1043 memcpy(u->uuid, invalid_uuid, 16); 1044 u->invalidated = cpu_to_le32(get_seconds()); 1045 u = NULL; 1046 } 1047 1048 if (!u) { 1049 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1050 pr_err("Couldn't find uuid for %s in set", buf); 1051 return -ENOENT; 1052 } 1053 1054 u = uuid_find_empty(c); 1055 if (!u) { 1056 pr_err("Not caching %s, no room for UUID", buf); 1057 return -EINVAL; 1058 } 1059 } 1060 1061 /* Deadlocks since we're called via sysfs... 1062 sysfs_remove_file(&dc->kobj, &sysfs_attach); 1063 */ 1064 1065 if (bch_is_zero(u->uuid, 16)) { 1066 struct closure cl; 1067 closure_init_stack(&cl); 1068 1069 memcpy(u->uuid, dc->sb.uuid, 16); 1070 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); 1071 u->first_reg = u->last_reg = rtime; 1072 bch_uuid_write(c); 1073 1074 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16); 1075 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); 1076 1077 bch_write_bdev_super(dc, &cl); 1078 closure_sync(&cl); 1079 } else { 1080 u->last_reg = rtime; 1081 bch_uuid_write(c); 1082 } 1083 1084 bcache_device_attach(&dc->disk, c, u - c->uuids); 1085 list_move(&dc->list, &c->cached_devs); 1086 calc_cached_dev_sectors(c); 1087 1088 smp_wmb(); 1089 /* 1090 * dc->c must be set before dc->count != 0 - paired with the mb in 1091 * cached_dev_get() 1092 */ 1093 refcount_set(&dc->count, 1); 1094 1095 /* Block writeback thread, but spawn it */ 1096 down_write(&dc->writeback_lock); 1097 if (bch_cached_dev_writeback_start(dc)) { 1098 up_write(&dc->writeback_lock); 1099 return -ENOMEM; 1100 } 1101 1102 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1103 bch_sectors_dirty_init(&dc->disk); 1104 atomic_set(&dc->has_dirty, 1); 1105 bch_writeback_queue(dc); 1106 } 1107 1108 bch_cached_dev_run(dc); 1109 bcache_device_link(&dc->disk, c, "bdev"); 1110 1111 /* Allow the writeback thread to proceed */ 1112 up_write(&dc->writeback_lock); 1113 1114 pr_info("Caching %s as %s on set %pU", 1115 bdevname(dc->bdev, buf), dc->disk.disk->disk_name, 1116 dc->disk.c->sb.set_uuid); 1117 return 0; 1118 } 1119 1120 void bch_cached_dev_release(struct kobject *kobj) 1121 { 1122 struct cached_dev *dc = container_of(kobj, struct cached_dev, 1123 disk.kobj); 1124 kfree(dc); 1125 module_put(THIS_MODULE); 1126 } 1127 1128 static void cached_dev_free(struct closure *cl) 1129 { 1130 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1131 1132 mutex_lock(&bch_register_lock); 1133 1134 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) 1135 cancel_writeback_rate_update_dwork(dc); 1136 1137 if (!IS_ERR_OR_NULL(dc->writeback_thread)) 1138 kthread_stop(dc->writeback_thread); 1139 if (dc->writeback_write_wq) 1140 destroy_workqueue(dc->writeback_write_wq); 1141 1142 if (atomic_read(&dc->running)) 1143 bd_unlink_disk_holder(dc->bdev, dc->disk.disk); 1144 bcache_device_free(&dc->disk); 1145 list_del(&dc->list); 1146 1147 mutex_unlock(&bch_register_lock); 1148 1149 if (!IS_ERR_OR_NULL(dc->bdev)) 1150 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1151 1152 wake_up(&unregister_wait); 1153 1154 kobject_put(&dc->disk.kobj); 1155 } 1156 1157 static void cached_dev_flush(struct closure *cl) 1158 { 1159 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1160 struct bcache_device *d = &dc->disk; 1161 1162 mutex_lock(&bch_register_lock); 1163 bcache_device_unlink(d); 1164 mutex_unlock(&bch_register_lock); 1165 1166 bch_cache_accounting_destroy(&dc->accounting); 1167 kobject_del(&d->kobj); 1168 1169 continue_at(cl, cached_dev_free, system_wq); 1170 } 1171 1172 static int cached_dev_init(struct cached_dev *dc, unsigned block_size) 1173 { 1174 int ret; 1175 struct io *io; 1176 struct request_queue *q = bdev_get_queue(dc->bdev); 1177 1178 __module_get(THIS_MODULE); 1179 INIT_LIST_HEAD(&dc->list); 1180 closure_init(&dc->disk.cl, NULL); 1181 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); 1182 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); 1183 INIT_WORK(&dc->detach, cached_dev_detach_finish); 1184 sema_init(&dc->sb_write_mutex, 1); 1185 INIT_LIST_HEAD(&dc->io_lru); 1186 spin_lock_init(&dc->io_lock); 1187 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); 1188 1189 dc->sequential_cutoff = 4 << 20; 1190 1191 for (io = dc->io; io < dc->io + RECENT_IO; io++) { 1192 list_add(&io->lru, &dc->io_lru); 1193 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); 1194 } 1195 1196 dc->disk.stripe_size = q->limits.io_opt >> 9; 1197 1198 if (dc->disk.stripe_size) 1199 dc->partial_stripes_expensive = 1200 q->limits.raid_partial_stripes_expensive; 1201 1202 ret = bcache_device_init(&dc->disk, block_size, 1203 dc->bdev->bd_part->nr_sects - dc->sb.data_offset); 1204 if (ret) 1205 return ret; 1206 1207 dc->disk.disk->queue->backing_dev_info->ra_pages = 1208 max(dc->disk.disk->queue->backing_dev_info->ra_pages, 1209 q->backing_dev_info->ra_pages); 1210 1211 atomic_set(&dc->io_errors, 0); 1212 dc->io_disable = false; 1213 dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT; 1214 /* default to auto */ 1215 dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO; 1216 1217 bch_cached_dev_request_init(dc); 1218 bch_cached_dev_writeback_init(dc); 1219 return 0; 1220 } 1221 1222 /* Cached device - bcache superblock */ 1223 1224 static void register_bdev(struct cache_sb *sb, struct page *sb_page, 1225 struct block_device *bdev, 1226 struct cached_dev *dc) 1227 { 1228 char name[BDEVNAME_SIZE]; 1229 const char *err = "cannot allocate memory"; 1230 struct cache_set *c; 1231 1232 memcpy(&dc->sb, sb, sizeof(struct cache_sb)); 1233 dc->bdev = bdev; 1234 dc->bdev->bd_holder = dc; 1235 1236 bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1); 1237 bio_first_bvec_all(&dc->sb_bio)->bv_page = sb_page; 1238 get_page(sb_page); 1239 1240 if (cached_dev_init(dc, sb->block_size << 9)) 1241 goto err; 1242 1243 err = "error creating kobject"; 1244 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj, 1245 "bcache")) 1246 goto err; 1247 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) 1248 goto err; 1249 1250 pr_info("registered backing device %s", bdevname(bdev, name)); 1251 1252 list_add(&dc->list, &uncached_devices); 1253 list_for_each_entry(c, &bch_cache_sets, list) 1254 bch_cached_dev_attach(dc, c, NULL); 1255 1256 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || 1257 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) 1258 bch_cached_dev_run(dc); 1259 1260 return; 1261 err: 1262 pr_notice("error %s: %s", bdevname(bdev, name), err); 1263 bcache_device_stop(&dc->disk); 1264 } 1265 1266 /* Flash only volumes */ 1267 1268 void bch_flash_dev_release(struct kobject *kobj) 1269 { 1270 struct bcache_device *d = container_of(kobj, struct bcache_device, 1271 kobj); 1272 kfree(d); 1273 } 1274 1275 static void flash_dev_free(struct closure *cl) 1276 { 1277 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1278 mutex_lock(&bch_register_lock); 1279 bcache_device_free(d); 1280 mutex_unlock(&bch_register_lock); 1281 kobject_put(&d->kobj); 1282 } 1283 1284 static void flash_dev_flush(struct closure *cl) 1285 { 1286 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1287 1288 mutex_lock(&bch_register_lock); 1289 bcache_device_unlink(d); 1290 mutex_unlock(&bch_register_lock); 1291 kobject_del(&d->kobj); 1292 continue_at(cl, flash_dev_free, system_wq); 1293 } 1294 1295 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) 1296 { 1297 struct bcache_device *d = kzalloc(sizeof(struct bcache_device), 1298 GFP_KERNEL); 1299 if (!d) 1300 return -ENOMEM; 1301 1302 closure_init(&d->cl, NULL); 1303 set_closure_fn(&d->cl, flash_dev_flush, system_wq); 1304 1305 kobject_init(&d->kobj, &bch_flash_dev_ktype); 1306 1307 if (bcache_device_init(d, block_bytes(c), u->sectors)) 1308 goto err; 1309 1310 bcache_device_attach(d, c, u - c->uuids); 1311 bch_sectors_dirty_init(d); 1312 bch_flash_dev_request_init(d); 1313 add_disk(d->disk); 1314 1315 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache")) 1316 goto err; 1317 1318 bcache_device_link(d, c, "volume"); 1319 1320 return 0; 1321 err: 1322 kobject_put(&d->kobj); 1323 return -ENOMEM; 1324 } 1325 1326 static int flash_devs_run(struct cache_set *c) 1327 { 1328 int ret = 0; 1329 struct uuid_entry *u; 1330 1331 for (u = c->uuids; 1332 u < c->uuids + c->nr_uuids && !ret; 1333 u++) 1334 if (UUID_FLASH_ONLY(u)) 1335 ret = flash_dev_run(c, u); 1336 1337 return ret; 1338 } 1339 1340 int bch_flash_dev_create(struct cache_set *c, uint64_t size) 1341 { 1342 struct uuid_entry *u; 1343 1344 if (test_bit(CACHE_SET_STOPPING, &c->flags)) 1345 return -EINTR; 1346 1347 if (!test_bit(CACHE_SET_RUNNING, &c->flags)) 1348 return -EPERM; 1349 1350 u = uuid_find_empty(c); 1351 if (!u) { 1352 pr_err("Can't create volume, no room for UUID"); 1353 return -EINVAL; 1354 } 1355 1356 get_random_bytes(u->uuid, 16); 1357 memset(u->label, 0, 32); 1358 u->first_reg = u->last_reg = cpu_to_le32(get_seconds()); 1359 1360 SET_UUID_FLASH_ONLY(u, 1); 1361 u->sectors = size >> 9; 1362 1363 bch_uuid_write(c); 1364 1365 return flash_dev_run(c, u); 1366 } 1367 1368 bool bch_cached_dev_error(struct cached_dev *dc) 1369 { 1370 char name[BDEVNAME_SIZE]; 1371 1372 if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) 1373 return false; 1374 1375 dc->io_disable = true; 1376 /* make others know io_disable is true earlier */ 1377 smp_mb(); 1378 1379 pr_err("stop %s: too many IO errors on backing device %s\n", 1380 dc->disk.disk->disk_name, bdevname(dc->bdev, name)); 1381 1382 bcache_device_stop(&dc->disk); 1383 return true; 1384 } 1385 1386 /* Cache set */ 1387 1388 __printf(2, 3) 1389 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) 1390 { 1391 va_list args; 1392 1393 if (c->on_error != ON_ERROR_PANIC && 1394 test_bit(CACHE_SET_STOPPING, &c->flags)) 1395 return false; 1396 1397 if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags)) 1398 pr_warn("CACHE_SET_IO_DISABLE already set"); 1399 1400 /* XXX: we can be called from atomic context 1401 acquire_console_sem(); 1402 */ 1403 1404 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid); 1405 1406 va_start(args, fmt); 1407 vprintk(fmt, args); 1408 va_end(args); 1409 1410 printk(", disabling caching\n"); 1411 1412 if (c->on_error == ON_ERROR_PANIC) 1413 panic("panic forced after error\n"); 1414 1415 bch_cache_set_unregister(c); 1416 return true; 1417 } 1418 1419 void bch_cache_set_release(struct kobject *kobj) 1420 { 1421 struct cache_set *c = container_of(kobj, struct cache_set, kobj); 1422 kfree(c); 1423 module_put(THIS_MODULE); 1424 } 1425 1426 static void cache_set_free(struct closure *cl) 1427 { 1428 struct cache_set *c = container_of(cl, struct cache_set, cl); 1429 struct cache *ca; 1430 unsigned i; 1431 1432 if (!IS_ERR_OR_NULL(c->debug)) 1433 debugfs_remove(c->debug); 1434 1435 bch_open_buckets_free(c); 1436 bch_btree_cache_free(c); 1437 bch_journal_free(c); 1438 1439 for_each_cache(ca, c, i) 1440 if (ca) { 1441 ca->set = NULL; 1442 c->cache[ca->sb.nr_this_dev] = NULL; 1443 kobject_put(&ca->kobj); 1444 } 1445 1446 bch_bset_sort_state_free(&c->sort); 1447 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c))); 1448 1449 if (c->moving_gc_wq) 1450 destroy_workqueue(c->moving_gc_wq); 1451 if (c->bio_split) 1452 bioset_free(c->bio_split); 1453 if (c->fill_iter) 1454 mempool_destroy(c->fill_iter); 1455 if (c->bio_meta) 1456 mempool_destroy(c->bio_meta); 1457 if (c->search) 1458 mempool_destroy(c->search); 1459 kfree(c->devices); 1460 1461 mutex_lock(&bch_register_lock); 1462 list_del(&c->list); 1463 mutex_unlock(&bch_register_lock); 1464 1465 pr_info("Cache set %pU unregistered", c->sb.set_uuid); 1466 wake_up(&unregister_wait); 1467 1468 closure_debug_destroy(&c->cl); 1469 kobject_put(&c->kobj); 1470 } 1471 1472 static void cache_set_flush(struct closure *cl) 1473 { 1474 struct cache_set *c = container_of(cl, struct cache_set, caching); 1475 struct cache *ca; 1476 struct btree *b; 1477 unsigned i; 1478 1479 bch_cache_accounting_destroy(&c->accounting); 1480 1481 kobject_put(&c->internal); 1482 kobject_del(&c->kobj); 1483 1484 if (c->gc_thread) 1485 kthread_stop(c->gc_thread); 1486 1487 if (!IS_ERR_OR_NULL(c->root)) 1488 list_add(&c->root->list, &c->btree_cache); 1489 1490 /* Should skip this if we're unregistering because of an error */ 1491 list_for_each_entry(b, &c->btree_cache, list) { 1492 mutex_lock(&b->write_lock); 1493 if (btree_node_dirty(b)) 1494 __bch_btree_node_write(b, NULL); 1495 mutex_unlock(&b->write_lock); 1496 } 1497 1498 for_each_cache(ca, c, i) 1499 if (ca->alloc_thread) 1500 kthread_stop(ca->alloc_thread); 1501 1502 if (c->journal.cur) { 1503 cancel_delayed_work_sync(&c->journal.work); 1504 /* flush last journal entry if needed */ 1505 c->journal.work.work.func(&c->journal.work.work); 1506 } 1507 1508 closure_return(cl); 1509 } 1510 1511 /* 1512 * This function is only called when CACHE_SET_IO_DISABLE is set, which means 1513 * cache set is unregistering due to too many I/O errors. In this condition, 1514 * the bcache device might be stopped, it depends on stop_when_cache_set_failed 1515 * value and whether the broken cache has dirty data: 1516 * 1517 * dc->stop_when_cache_set_failed dc->has_dirty stop bcache device 1518 * BCH_CACHED_STOP_AUTO 0 NO 1519 * BCH_CACHED_STOP_AUTO 1 YES 1520 * BCH_CACHED_DEV_STOP_ALWAYS 0 YES 1521 * BCH_CACHED_DEV_STOP_ALWAYS 1 YES 1522 * 1523 * The expected behavior is, if stop_when_cache_set_failed is configured to 1524 * "auto" via sysfs interface, the bcache device will not be stopped if the 1525 * backing device is clean on the broken cache device. 1526 */ 1527 static void conditional_stop_bcache_device(struct cache_set *c, 1528 struct bcache_device *d, 1529 struct cached_dev *dc) 1530 { 1531 if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) { 1532 pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.", 1533 d->disk->disk_name, c->sb.set_uuid); 1534 bcache_device_stop(d); 1535 } else if (atomic_read(&dc->has_dirty)) { 1536 /* 1537 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO 1538 * and dc->has_dirty == 1 1539 */ 1540 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.", 1541 d->disk->disk_name); 1542 bcache_device_stop(d); 1543 } else { 1544 /* 1545 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO 1546 * and dc->has_dirty == 0 1547 */ 1548 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.", 1549 d->disk->disk_name); 1550 } 1551 } 1552 1553 static void __cache_set_unregister(struct closure *cl) 1554 { 1555 struct cache_set *c = container_of(cl, struct cache_set, caching); 1556 struct cached_dev *dc; 1557 struct bcache_device *d; 1558 size_t i; 1559 1560 mutex_lock(&bch_register_lock); 1561 1562 for (i = 0; i < c->devices_max_used; i++) { 1563 d = c->devices[i]; 1564 if (!d) 1565 continue; 1566 1567 if (!UUID_FLASH_ONLY(&c->uuids[i]) && 1568 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { 1569 dc = container_of(d, struct cached_dev, disk); 1570 bch_cached_dev_detach(dc); 1571 if (test_bit(CACHE_SET_IO_DISABLE, &c->flags)) 1572 conditional_stop_bcache_device(c, d, dc); 1573 } else { 1574 bcache_device_stop(d); 1575 } 1576 } 1577 1578 mutex_unlock(&bch_register_lock); 1579 1580 continue_at(cl, cache_set_flush, system_wq); 1581 } 1582 1583 void bch_cache_set_stop(struct cache_set *c) 1584 { 1585 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) 1586 closure_queue(&c->caching); 1587 } 1588 1589 void bch_cache_set_unregister(struct cache_set *c) 1590 { 1591 set_bit(CACHE_SET_UNREGISTERING, &c->flags); 1592 bch_cache_set_stop(c); 1593 } 1594 1595 #define alloc_bucket_pages(gfp, c) \ 1596 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c)))) 1597 1598 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) 1599 { 1600 int iter_size; 1601 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); 1602 if (!c) 1603 return NULL; 1604 1605 __module_get(THIS_MODULE); 1606 closure_init(&c->cl, NULL); 1607 set_closure_fn(&c->cl, cache_set_free, system_wq); 1608 1609 closure_init(&c->caching, &c->cl); 1610 set_closure_fn(&c->caching, __cache_set_unregister, system_wq); 1611 1612 /* Maybe create continue_at_noreturn() and use it here? */ 1613 closure_set_stopped(&c->cl); 1614 closure_put(&c->cl); 1615 1616 kobject_init(&c->kobj, &bch_cache_set_ktype); 1617 kobject_init(&c->internal, &bch_cache_set_internal_ktype); 1618 1619 bch_cache_accounting_init(&c->accounting, &c->cl); 1620 1621 memcpy(c->sb.set_uuid, sb->set_uuid, 16); 1622 c->sb.block_size = sb->block_size; 1623 c->sb.bucket_size = sb->bucket_size; 1624 c->sb.nr_in_set = sb->nr_in_set; 1625 c->sb.last_mount = sb->last_mount; 1626 c->bucket_bits = ilog2(sb->bucket_size); 1627 c->block_bits = ilog2(sb->block_size); 1628 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry); 1629 c->devices_max_used = 0; 1630 c->btree_pages = bucket_pages(c); 1631 if (c->btree_pages > BTREE_MAX_PAGES) 1632 c->btree_pages = max_t(int, c->btree_pages / 4, 1633 BTREE_MAX_PAGES); 1634 1635 sema_init(&c->sb_write_mutex, 1); 1636 mutex_init(&c->bucket_lock); 1637 init_waitqueue_head(&c->btree_cache_wait); 1638 init_waitqueue_head(&c->bucket_wait); 1639 init_waitqueue_head(&c->gc_wait); 1640 sema_init(&c->uuid_write_mutex, 1); 1641 1642 spin_lock_init(&c->btree_gc_time.lock); 1643 spin_lock_init(&c->btree_split_time.lock); 1644 spin_lock_init(&c->btree_read_time.lock); 1645 1646 bch_moving_init_cache_set(c); 1647 1648 INIT_LIST_HEAD(&c->list); 1649 INIT_LIST_HEAD(&c->cached_devs); 1650 INIT_LIST_HEAD(&c->btree_cache); 1651 INIT_LIST_HEAD(&c->btree_cache_freeable); 1652 INIT_LIST_HEAD(&c->btree_cache_freed); 1653 INIT_LIST_HEAD(&c->data_buckets); 1654 1655 c->search = mempool_create_slab_pool(32, bch_search_cache); 1656 if (!c->search) 1657 goto err; 1658 1659 iter_size = (sb->bucket_size / sb->block_size + 1) * 1660 sizeof(struct btree_iter_set); 1661 1662 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) || 1663 !(c->bio_meta = mempool_create_kmalloc_pool(2, 1664 sizeof(struct bbio) + sizeof(struct bio_vec) * 1665 bucket_pages(c))) || 1666 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) || 1667 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio), 1668 BIOSET_NEED_BVECS | 1669 BIOSET_NEED_RESCUER)) || 1670 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) || 1671 !(c->moving_gc_wq = alloc_workqueue("bcache_gc", 1672 WQ_MEM_RECLAIM, 0)) || 1673 bch_journal_alloc(c) || 1674 bch_btree_cache_alloc(c) || 1675 bch_open_buckets_alloc(c) || 1676 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages))) 1677 goto err; 1678 1679 c->congested_read_threshold_us = 2000; 1680 c->congested_write_threshold_us = 20000; 1681 c->error_limit = DEFAULT_IO_ERROR_LIMIT; 1682 WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags)); 1683 1684 return c; 1685 err: 1686 bch_cache_set_unregister(c); 1687 return NULL; 1688 } 1689 1690 static void run_cache_set(struct cache_set *c) 1691 { 1692 const char *err = "cannot allocate memory"; 1693 struct cached_dev *dc, *t; 1694 struct cache *ca; 1695 struct closure cl; 1696 unsigned i; 1697 1698 closure_init_stack(&cl); 1699 1700 for_each_cache(ca, c, i) 1701 c->nbuckets += ca->sb.nbuckets; 1702 set_gc_sectors(c); 1703 1704 if (CACHE_SYNC(&c->sb)) { 1705 LIST_HEAD(journal); 1706 struct bkey *k; 1707 struct jset *j; 1708 1709 err = "cannot allocate memory for journal"; 1710 if (bch_journal_read(c, &journal)) 1711 goto err; 1712 1713 pr_debug("btree_journal_read() done"); 1714 1715 err = "no journal entries found"; 1716 if (list_empty(&journal)) 1717 goto err; 1718 1719 j = &list_entry(journal.prev, struct journal_replay, list)->j; 1720 1721 err = "IO error reading priorities"; 1722 for_each_cache(ca, c, i) 1723 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]); 1724 1725 /* 1726 * If prio_read() fails it'll call cache_set_error and we'll 1727 * tear everything down right away, but if we perhaps checked 1728 * sooner we could avoid journal replay. 1729 */ 1730 1731 k = &j->btree_root; 1732 1733 err = "bad btree root"; 1734 if (__bch_btree_ptr_invalid(c, k)) 1735 goto err; 1736 1737 err = "error reading btree root"; 1738 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL); 1739 if (IS_ERR_OR_NULL(c->root)) 1740 goto err; 1741 1742 list_del_init(&c->root->list); 1743 rw_unlock(true, c->root); 1744 1745 err = uuid_read(c, j, &cl); 1746 if (err) 1747 goto err; 1748 1749 err = "error in recovery"; 1750 if (bch_btree_check(c)) 1751 goto err; 1752 1753 bch_journal_mark(c, &journal); 1754 bch_initial_gc_finish(c); 1755 pr_debug("btree_check() done"); 1756 1757 /* 1758 * bcache_journal_next() can't happen sooner, or 1759 * btree_gc_finish() will give spurious errors about last_gc > 1760 * gc_gen - this is a hack but oh well. 1761 */ 1762 bch_journal_next(&c->journal); 1763 1764 err = "error starting allocator thread"; 1765 for_each_cache(ca, c, i) 1766 if (bch_cache_allocator_start(ca)) 1767 goto err; 1768 1769 /* 1770 * First place it's safe to allocate: btree_check() and 1771 * btree_gc_finish() have to run before we have buckets to 1772 * allocate, and bch_bucket_alloc_set() might cause a journal 1773 * entry to be written so bcache_journal_next() has to be called 1774 * first. 1775 * 1776 * If the uuids were in the old format we have to rewrite them 1777 * before the next journal entry is written: 1778 */ 1779 if (j->version < BCACHE_JSET_VERSION_UUID) 1780 __uuid_write(c); 1781 1782 bch_journal_replay(c, &journal); 1783 } else { 1784 pr_notice("invalidating existing data"); 1785 1786 for_each_cache(ca, c, i) { 1787 unsigned j; 1788 1789 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, 1790 2, SB_JOURNAL_BUCKETS); 1791 1792 for (j = 0; j < ca->sb.keys; j++) 1793 ca->sb.d[j] = ca->sb.first_bucket + j; 1794 } 1795 1796 bch_initial_gc_finish(c); 1797 1798 err = "error starting allocator thread"; 1799 for_each_cache(ca, c, i) 1800 if (bch_cache_allocator_start(ca)) 1801 goto err; 1802 1803 mutex_lock(&c->bucket_lock); 1804 for_each_cache(ca, c, i) 1805 bch_prio_write(ca); 1806 mutex_unlock(&c->bucket_lock); 1807 1808 err = "cannot allocate new UUID bucket"; 1809 if (__uuid_write(c)) 1810 goto err; 1811 1812 err = "cannot allocate new btree root"; 1813 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL); 1814 if (IS_ERR_OR_NULL(c->root)) 1815 goto err; 1816 1817 mutex_lock(&c->root->write_lock); 1818 bkey_copy_key(&c->root->key, &MAX_KEY); 1819 bch_btree_node_write(c->root, &cl); 1820 mutex_unlock(&c->root->write_lock); 1821 1822 bch_btree_set_root(c->root); 1823 rw_unlock(true, c->root); 1824 1825 /* 1826 * We don't want to write the first journal entry until 1827 * everything is set up - fortunately journal entries won't be 1828 * written until the SET_CACHE_SYNC() here: 1829 */ 1830 SET_CACHE_SYNC(&c->sb, true); 1831 1832 bch_journal_next(&c->journal); 1833 bch_journal_meta(c, &cl); 1834 } 1835 1836 err = "error starting gc thread"; 1837 if (bch_gc_thread_start(c)) 1838 goto err; 1839 1840 closure_sync(&cl); 1841 c->sb.last_mount = get_seconds(); 1842 bcache_write_super(c); 1843 1844 list_for_each_entry_safe(dc, t, &uncached_devices, list) 1845 bch_cached_dev_attach(dc, c, NULL); 1846 1847 flash_devs_run(c); 1848 1849 set_bit(CACHE_SET_RUNNING, &c->flags); 1850 return; 1851 err: 1852 closure_sync(&cl); 1853 /* XXX: test this, it's broken */ 1854 bch_cache_set_error(c, "%s", err); 1855 } 1856 1857 static bool can_attach_cache(struct cache *ca, struct cache_set *c) 1858 { 1859 return ca->sb.block_size == c->sb.block_size && 1860 ca->sb.bucket_size == c->sb.bucket_size && 1861 ca->sb.nr_in_set == c->sb.nr_in_set; 1862 } 1863 1864 static const char *register_cache_set(struct cache *ca) 1865 { 1866 char buf[12]; 1867 const char *err = "cannot allocate memory"; 1868 struct cache_set *c; 1869 1870 list_for_each_entry(c, &bch_cache_sets, list) 1871 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) { 1872 if (c->cache[ca->sb.nr_this_dev]) 1873 return "duplicate cache set member"; 1874 1875 if (!can_attach_cache(ca, c)) 1876 return "cache sb does not match set"; 1877 1878 if (!CACHE_SYNC(&ca->sb)) 1879 SET_CACHE_SYNC(&c->sb, false); 1880 1881 goto found; 1882 } 1883 1884 c = bch_cache_set_alloc(&ca->sb); 1885 if (!c) 1886 return err; 1887 1888 err = "error creating kobject"; 1889 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) || 1890 kobject_add(&c->internal, &c->kobj, "internal")) 1891 goto err; 1892 1893 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) 1894 goto err; 1895 1896 bch_debug_init_cache_set(c); 1897 1898 list_add(&c->list, &bch_cache_sets); 1899 found: 1900 sprintf(buf, "cache%i", ca->sb.nr_this_dev); 1901 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || 1902 sysfs_create_link(&c->kobj, &ca->kobj, buf)) 1903 goto err; 1904 1905 if (ca->sb.seq > c->sb.seq) { 1906 c->sb.version = ca->sb.version; 1907 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16); 1908 c->sb.flags = ca->sb.flags; 1909 c->sb.seq = ca->sb.seq; 1910 pr_debug("set version = %llu", c->sb.version); 1911 } 1912 1913 kobject_get(&ca->kobj); 1914 ca->set = c; 1915 ca->set->cache[ca->sb.nr_this_dev] = ca; 1916 c->cache_by_alloc[c->caches_loaded++] = ca; 1917 1918 if (c->caches_loaded == c->sb.nr_in_set) 1919 run_cache_set(c); 1920 1921 return NULL; 1922 err: 1923 bch_cache_set_unregister(c); 1924 return err; 1925 } 1926 1927 /* Cache device */ 1928 1929 void bch_cache_release(struct kobject *kobj) 1930 { 1931 struct cache *ca = container_of(kobj, struct cache, kobj); 1932 unsigned i; 1933 1934 if (ca->set) { 1935 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca); 1936 ca->set->cache[ca->sb.nr_this_dev] = NULL; 1937 } 1938 1939 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca))); 1940 kfree(ca->prio_buckets); 1941 vfree(ca->buckets); 1942 1943 free_heap(&ca->heap); 1944 free_fifo(&ca->free_inc); 1945 1946 for (i = 0; i < RESERVE_NR; i++) 1947 free_fifo(&ca->free[i]); 1948 1949 if (ca->sb_bio.bi_inline_vecs[0].bv_page) 1950 put_page(bio_first_page_all(&ca->sb_bio)); 1951 1952 if (!IS_ERR_OR_NULL(ca->bdev)) 1953 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1954 1955 kfree(ca); 1956 module_put(THIS_MODULE); 1957 } 1958 1959 static int cache_alloc(struct cache *ca) 1960 { 1961 size_t free; 1962 size_t btree_buckets; 1963 struct bucket *b; 1964 1965 __module_get(THIS_MODULE); 1966 kobject_init(&ca->kobj, &bch_cache_ktype); 1967 1968 bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8); 1969 1970 /* 1971 * when ca->sb.njournal_buckets is not zero, journal exists, 1972 * and in bch_journal_replay(), tree node may split, 1973 * so bucket of RESERVE_BTREE type is needed, 1974 * the worst situation is all journal buckets are valid journal, 1975 * and all the keys need to replay, 1976 * so the number of RESERVE_BTREE type buckets should be as much 1977 * as journal buckets 1978 */ 1979 btree_buckets = ca->sb.njournal_buckets ?: 8; 1980 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; 1981 1982 if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, GFP_KERNEL) || 1983 !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) || 1984 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) || 1985 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) || 1986 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) || 1987 !init_heap(&ca->heap, free << 3, GFP_KERNEL) || 1988 !(ca->buckets = vzalloc(sizeof(struct bucket) * 1989 ca->sb.nbuckets)) || 1990 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) * 1991 2, GFP_KERNEL)) || 1992 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca))) 1993 return -ENOMEM; 1994 1995 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); 1996 1997 for_each_bucket(b, ca) 1998 atomic_set(&b->pin, 0); 1999 2000 return 0; 2001 } 2002 2003 static int register_cache(struct cache_sb *sb, struct page *sb_page, 2004 struct block_device *bdev, struct cache *ca) 2005 { 2006 char name[BDEVNAME_SIZE]; 2007 const char *err = NULL; /* must be set for any error case */ 2008 int ret = 0; 2009 2010 bdevname(bdev, name); 2011 2012 memcpy(&ca->sb, sb, sizeof(struct cache_sb)); 2013 ca->bdev = bdev; 2014 ca->bdev->bd_holder = ca; 2015 2016 bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1); 2017 bio_first_bvec_all(&ca->sb_bio)->bv_page = sb_page; 2018 get_page(sb_page); 2019 2020 if (blk_queue_discard(bdev_get_queue(bdev))) 2021 ca->discard = CACHE_DISCARD(&ca->sb); 2022 2023 ret = cache_alloc(ca); 2024 if (ret != 0) { 2025 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2026 if (ret == -ENOMEM) 2027 err = "cache_alloc(): -ENOMEM"; 2028 else 2029 err = "cache_alloc(): unknown error"; 2030 goto err; 2031 } 2032 2033 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) { 2034 err = "error calling kobject_add"; 2035 ret = -ENOMEM; 2036 goto out; 2037 } 2038 2039 mutex_lock(&bch_register_lock); 2040 err = register_cache_set(ca); 2041 mutex_unlock(&bch_register_lock); 2042 2043 if (err) { 2044 ret = -ENODEV; 2045 goto out; 2046 } 2047 2048 pr_info("registered cache device %s", name); 2049 2050 out: 2051 kobject_put(&ca->kobj); 2052 2053 err: 2054 if (err) 2055 pr_notice("error %s: %s", name, err); 2056 2057 return ret; 2058 } 2059 2060 /* Global interfaces/init */ 2061 2062 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *, 2063 const char *, size_t); 2064 2065 kobj_attribute_write(register, register_bcache); 2066 kobj_attribute_write(register_quiet, register_bcache); 2067 2068 static bool bch_is_open_backing(struct block_device *bdev) { 2069 struct cache_set *c, *tc; 2070 struct cached_dev *dc, *t; 2071 2072 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2073 list_for_each_entry_safe(dc, t, &c->cached_devs, list) 2074 if (dc->bdev == bdev) 2075 return true; 2076 list_for_each_entry_safe(dc, t, &uncached_devices, list) 2077 if (dc->bdev == bdev) 2078 return true; 2079 return false; 2080 } 2081 2082 static bool bch_is_open_cache(struct block_device *bdev) { 2083 struct cache_set *c, *tc; 2084 struct cache *ca; 2085 unsigned i; 2086 2087 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2088 for_each_cache(ca, c, i) 2089 if (ca->bdev == bdev) 2090 return true; 2091 return false; 2092 } 2093 2094 static bool bch_is_open(struct block_device *bdev) { 2095 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev); 2096 } 2097 2098 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, 2099 const char *buffer, size_t size) 2100 { 2101 ssize_t ret = size; 2102 const char *err = "cannot allocate memory"; 2103 char *path = NULL; 2104 struct cache_sb *sb = NULL; 2105 struct block_device *bdev = NULL; 2106 struct page *sb_page = NULL; 2107 2108 if (!try_module_get(THIS_MODULE)) 2109 return -EBUSY; 2110 2111 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) || 2112 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL))) 2113 goto err; 2114 2115 err = "failed to open device"; 2116 bdev = blkdev_get_by_path(strim(path), 2117 FMODE_READ|FMODE_WRITE|FMODE_EXCL, 2118 sb); 2119 if (IS_ERR(bdev)) { 2120 if (bdev == ERR_PTR(-EBUSY)) { 2121 bdev = lookup_bdev(strim(path)); 2122 mutex_lock(&bch_register_lock); 2123 if (!IS_ERR(bdev) && bch_is_open(bdev)) 2124 err = "device already registered"; 2125 else 2126 err = "device busy"; 2127 mutex_unlock(&bch_register_lock); 2128 if (!IS_ERR(bdev)) 2129 bdput(bdev); 2130 if (attr == &ksysfs_register_quiet) 2131 goto out; 2132 } 2133 goto err; 2134 } 2135 2136 err = "failed to set blocksize"; 2137 if (set_blocksize(bdev, 4096)) 2138 goto err_close; 2139 2140 err = read_super(sb, bdev, &sb_page); 2141 if (err) 2142 goto err_close; 2143 2144 err = "failed to register device"; 2145 if (SB_IS_BDEV(sb)) { 2146 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL); 2147 if (!dc) 2148 goto err_close; 2149 2150 mutex_lock(&bch_register_lock); 2151 register_bdev(sb, sb_page, bdev, dc); 2152 mutex_unlock(&bch_register_lock); 2153 } else { 2154 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL); 2155 if (!ca) 2156 goto err_close; 2157 2158 if (register_cache(sb, sb_page, bdev, ca) != 0) 2159 goto err; 2160 } 2161 out: 2162 if (sb_page) 2163 put_page(sb_page); 2164 kfree(sb); 2165 kfree(path); 2166 module_put(THIS_MODULE); 2167 return ret; 2168 2169 err_close: 2170 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2171 err: 2172 pr_info("error %s: %s", path, err); 2173 ret = -EINVAL; 2174 goto out; 2175 } 2176 2177 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) 2178 { 2179 if (code == SYS_DOWN || 2180 code == SYS_HALT || 2181 code == SYS_POWER_OFF) { 2182 DEFINE_WAIT(wait); 2183 unsigned long start = jiffies; 2184 bool stopped = false; 2185 2186 struct cache_set *c, *tc; 2187 struct cached_dev *dc, *tdc; 2188 2189 mutex_lock(&bch_register_lock); 2190 2191 if (list_empty(&bch_cache_sets) && 2192 list_empty(&uncached_devices)) 2193 goto out; 2194 2195 pr_info("Stopping all devices:"); 2196 2197 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2198 bch_cache_set_stop(c); 2199 2200 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) 2201 bcache_device_stop(&dc->disk); 2202 2203 /* What's a condition variable? */ 2204 while (1) { 2205 long timeout = start + 2 * HZ - jiffies; 2206 2207 stopped = list_empty(&bch_cache_sets) && 2208 list_empty(&uncached_devices); 2209 2210 if (timeout < 0 || stopped) 2211 break; 2212 2213 prepare_to_wait(&unregister_wait, &wait, 2214 TASK_UNINTERRUPTIBLE); 2215 2216 mutex_unlock(&bch_register_lock); 2217 schedule_timeout(timeout); 2218 mutex_lock(&bch_register_lock); 2219 } 2220 2221 finish_wait(&unregister_wait, &wait); 2222 2223 if (stopped) 2224 pr_info("All devices stopped"); 2225 else 2226 pr_notice("Timeout waiting for devices to be closed"); 2227 out: 2228 mutex_unlock(&bch_register_lock); 2229 } 2230 2231 return NOTIFY_DONE; 2232 } 2233 2234 static struct notifier_block reboot = { 2235 .notifier_call = bcache_reboot, 2236 .priority = INT_MAX, /* before any real devices */ 2237 }; 2238 2239 static void bcache_exit(void) 2240 { 2241 bch_debug_exit(); 2242 bch_request_exit(); 2243 if (bcache_kobj) 2244 kobject_put(bcache_kobj); 2245 if (bcache_wq) 2246 destroy_workqueue(bcache_wq); 2247 if (bcache_major) 2248 unregister_blkdev(bcache_major, "bcache"); 2249 unregister_reboot_notifier(&reboot); 2250 mutex_destroy(&bch_register_lock); 2251 } 2252 2253 static int __init bcache_init(void) 2254 { 2255 static const struct attribute *files[] = { 2256 &ksysfs_register.attr, 2257 &ksysfs_register_quiet.attr, 2258 NULL 2259 }; 2260 2261 mutex_init(&bch_register_lock); 2262 init_waitqueue_head(&unregister_wait); 2263 register_reboot_notifier(&reboot); 2264 2265 bcache_major = register_blkdev(0, "bcache"); 2266 if (bcache_major < 0) { 2267 unregister_reboot_notifier(&reboot); 2268 mutex_destroy(&bch_register_lock); 2269 return bcache_major; 2270 } 2271 2272 if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) || 2273 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) || 2274 bch_request_init() || 2275 bch_debug_init(bcache_kobj) || closure_debug_init() || 2276 sysfs_create_files(bcache_kobj, files)) 2277 goto err; 2278 2279 return 0; 2280 err: 2281 bcache_exit(); 2282 return -ENOMEM; 2283 } 2284 2285 module_exit(bcache_exit); 2286 module_init(bcache_init); 2287