1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * bcache setup/teardown code, and some metadata io - read a superblock and 4 * figure out what to do with it. 5 * 6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> 7 * Copyright 2012 Google, Inc. 8 */ 9 10 #include "bcache.h" 11 #include "btree.h" 12 #include "debug.h" 13 #include "extents.h" 14 #include "request.h" 15 #include "writeback.h" 16 17 #include <linux/blkdev.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 unsigned int bch_cutoff_writeback; 28 unsigned int bch_cutoff_writeback_sync; 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 static struct kobject *bcache_kobj; 41 struct mutex bch_register_lock; 42 bool bcache_is_reboot; 43 LIST_HEAD(bch_cache_sets); 44 static LIST_HEAD(uncached_devices); 45 46 static int bcache_major; 47 static DEFINE_IDA(bcache_device_idx); 48 static wait_queue_head_t unregister_wait; 49 struct workqueue_struct *bcache_wq; 50 struct workqueue_struct *bch_journal_wq; 51 52 53 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE) 54 /* limitation of partitions number on single bcache device */ 55 #define BCACHE_MINORS 128 56 /* limitation of bcache devices number on single system */ 57 #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS) 58 59 /* Superblock */ 60 61 static const char *read_super(struct cache_sb *sb, struct block_device *bdev, 62 struct cache_sb_disk **res) 63 { 64 const char *err; 65 struct cache_sb_disk *s; 66 struct page *page; 67 unsigned int i; 68 69 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, 70 SB_OFFSET >> PAGE_SHIFT, GFP_KERNEL); 71 if (IS_ERR(page)) 72 return "IO error"; 73 s = page_address(page) + offset_in_page(SB_OFFSET); 74 75 sb->offset = le64_to_cpu(s->offset); 76 sb->version = le64_to_cpu(s->version); 77 78 memcpy(sb->magic, s->magic, 16); 79 memcpy(sb->uuid, s->uuid, 16); 80 memcpy(sb->set_uuid, s->set_uuid, 16); 81 memcpy(sb->label, s->label, SB_LABEL_SIZE); 82 83 sb->flags = le64_to_cpu(s->flags); 84 sb->seq = le64_to_cpu(s->seq); 85 sb->last_mount = le32_to_cpu(s->last_mount); 86 sb->first_bucket = le16_to_cpu(s->first_bucket); 87 sb->keys = le16_to_cpu(s->keys); 88 89 for (i = 0; i < SB_JOURNAL_BUCKETS; i++) 90 sb->d[i] = le64_to_cpu(s->d[i]); 91 92 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u\n", 93 sb->version, sb->flags, sb->seq, sb->keys); 94 95 err = "Not a bcache superblock (bad offset)"; 96 if (sb->offset != SB_SECTOR) 97 goto err; 98 99 err = "Not a bcache superblock (bad magic)"; 100 if (memcmp(sb->magic, bcache_magic, 16)) 101 goto err; 102 103 err = "Too many journal buckets"; 104 if (sb->keys > SB_JOURNAL_BUCKETS) 105 goto err; 106 107 err = "Bad checksum"; 108 if (s->csum != csum_set(s)) 109 goto err; 110 111 err = "Bad UUID"; 112 if (bch_is_zero(sb->uuid, 16)) 113 goto err; 114 115 sb->block_size = le16_to_cpu(s->block_size); 116 117 err = "Superblock block size smaller than device block size"; 118 if (sb->block_size << 9 < bdev_logical_block_size(bdev)) 119 goto err; 120 121 switch (sb->version) { 122 case BCACHE_SB_VERSION_BDEV: 123 sb->data_offset = BDEV_DATA_START_DEFAULT; 124 break; 125 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET: 126 sb->data_offset = le64_to_cpu(s->data_offset); 127 128 err = "Bad data offset"; 129 if (sb->data_offset < BDEV_DATA_START_DEFAULT) 130 goto err; 131 132 break; 133 case BCACHE_SB_VERSION_CDEV: 134 case BCACHE_SB_VERSION_CDEV_WITH_UUID: 135 sb->nbuckets = le64_to_cpu(s->nbuckets); 136 sb->bucket_size = le16_to_cpu(s->bucket_size); 137 138 sb->nr_in_set = le16_to_cpu(s->nr_in_set); 139 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev); 140 141 err = "Too many buckets"; 142 if (sb->nbuckets > LONG_MAX) 143 goto err; 144 145 err = "Not enough buckets"; 146 if (sb->nbuckets < 1 << 7) 147 goto err; 148 149 err = "Bad block/bucket size"; 150 if (!is_power_of_2(sb->block_size) || 151 sb->block_size > PAGE_SECTORS || 152 !is_power_of_2(sb->bucket_size) || 153 sb->bucket_size < PAGE_SECTORS) 154 goto err; 155 156 err = "Invalid superblock: device too small"; 157 if (get_capacity(bdev->bd_disk) < 158 sb->bucket_size * sb->nbuckets) 159 goto err; 160 161 err = "Bad UUID"; 162 if (bch_is_zero(sb->set_uuid, 16)) 163 goto err; 164 165 err = "Bad cache device number in set"; 166 if (!sb->nr_in_set || 167 sb->nr_in_set <= sb->nr_this_dev || 168 sb->nr_in_set > MAX_CACHES_PER_SET) 169 goto err; 170 171 err = "Journal buckets not sequential"; 172 for (i = 0; i < sb->keys; i++) 173 if (sb->d[i] != sb->first_bucket + i) 174 goto err; 175 176 err = "Too many journal buckets"; 177 if (sb->first_bucket + sb->keys > sb->nbuckets) 178 goto err; 179 180 err = "Invalid superblock: first bucket comes before end of super"; 181 if (sb->first_bucket * sb->bucket_size < 16) 182 goto err; 183 184 break; 185 default: 186 err = "Unsupported superblock version"; 187 goto err; 188 } 189 190 sb->last_mount = (u32)ktime_get_real_seconds(); 191 *res = s; 192 return NULL; 193 err: 194 put_page(page); 195 return err; 196 } 197 198 static void write_bdev_super_endio(struct bio *bio) 199 { 200 struct cached_dev *dc = bio->bi_private; 201 202 if (bio->bi_status) 203 bch_count_backing_io_errors(dc, bio); 204 205 closure_put(&dc->sb_write); 206 } 207 208 static void __write_super(struct cache_sb *sb, struct cache_sb_disk *out, 209 struct bio *bio) 210 { 211 unsigned int i; 212 213 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META; 214 bio->bi_iter.bi_sector = SB_SECTOR; 215 __bio_add_page(bio, virt_to_page(out), SB_SIZE, 216 offset_in_page(out)); 217 218 out->offset = cpu_to_le64(sb->offset); 219 out->version = cpu_to_le64(sb->version); 220 221 memcpy(out->uuid, sb->uuid, 16); 222 memcpy(out->set_uuid, sb->set_uuid, 16); 223 memcpy(out->label, sb->label, SB_LABEL_SIZE); 224 225 out->flags = cpu_to_le64(sb->flags); 226 out->seq = cpu_to_le64(sb->seq); 227 228 out->last_mount = cpu_to_le32(sb->last_mount); 229 out->first_bucket = cpu_to_le16(sb->first_bucket); 230 out->keys = cpu_to_le16(sb->keys); 231 232 for (i = 0; i < sb->keys; i++) 233 out->d[i] = cpu_to_le64(sb->d[i]); 234 235 out->csum = csum_set(out); 236 237 pr_debug("ver %llu, flags %llu, seq %llu\n", 238 sb->version, sb->flags, sb->seq); 239 240 submit_bio(bio); 241 } 242 243 static void bch_write_bdev_super_unlock(struct closure *cl) 244 { 245 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write); 246 247 up(&dc->sb_write_mutex); 248 } 249 250 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent) 251 { 252 struct closure *cl = &dc->sb_write; 253 struct bio *bio = &dc->sb_bio; 254 255 down(&dc->sb_write_mutex); 256 closure_init(cl, parent); 257 258 bio_init(bio, dc->sb_bv, 1); 259 bio_set_dev(bio, dc->bdev); 260 bio->bi_end_io = write_bdev_super_endio; 261 bio->bi_private = dc; 262 263 closure_get(cl); 264 /* I/O request sent to backing device */ 265 __write_super(&dc->sb, dc->sb_disk, bio); 266 267 closure_return_with_destructor(cl, bch_write_bdev_super_unlock); 268 } 269 270 static void write_super_endio(struct bio *bio) 271 { 272 struct cache *ca = bio->bi_private; 273 274 /* is_read = 0 */ 275 bch_count_io_errors(ca, bio->bi_status, 0, 276 "writing superblock"); 277 closure_put(&ca->set->sb_write); 278 } 279 280 static void bcache_write_super_unlock(struct closure *cl) 281 { 282 struct cache_set *c = container_of(cl, struct cache_set, sb_write); 283 284 up(&c->sb_write_mutex); 285 } 286 287 void bcache_write_super(struct cache_set *c) 288 { 289 struct closure *cl = &c->sb_write; 290 struct cache *ca; 291 unsigned int i; 292 293 down(&c->sb_write_mutex); 294 closure_init(cl, &c->cl); 295 296 c->sb.seq++; 297 298 for_each_cache(ca, c, i) { 299 struct bio *bio = &ca->sb_bio; 300 301 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID; 302 ca->sb.seq = c->sb.seq; 303 ca->sb.last_mount = c->sb.last_mount; 304 305 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb)); 306 307 bio_init(bio, ca->sb_bv, 1); 308 bio_set_dev(bio, ca->bdev); 309 bio->bi_end_io = write_super_endio; 310 bio->bi_private = ca; 311 312 closure_get(cl); 313 __write_super(&ca->sb, ca->sb_disk, bio); 314 } 315 316 closure_return_with_destructor(cl, bcache_write_super_unlock); 317 } 318 319 /* UUID io */ 320 321 static void uuid_endio(struct bio *bio) 322 { 323 struct closure *cl = bio->bi_private; 324 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 325 326 cache_set_err_on(bio->bi_status, c, "accessing uuids"); 327 bch_bbio_free(bio, c); 328 closure_put(cl); 329 } 330 331 static void uuid_io_unlock(struct closure *cl) 332 { 333 struct cache_set *c = container_of(cl, struct cache_set, uuid_write); 334 335 up(&c->uuid_write_mutex); 336 } 337 338 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags, 339 struct bkey *k, struct closure *parent) 340 { 341 struct closure *cl = &c->uuid_write; 342 struct uuid_entry *u; 343 unsigned int i; 344 char buf[80]; 345 346 BUG_ON(!parent); 347 down(&c->uuid_write_mutex); 348 closure_init(cl, parent); 349 350 for (i = 0; i < KEY_PTRS(k); i++) { 351 struct bio *bio = bch_bbio_alloc(c); 352 353 bio->bi_opf = REQ_SYNC | REQ_META | op_flags; 354 bio->bi_iter.bi_size = KEY_SIZE(k) << 9; 355 356 bio->bi_end_io = uuid_endio; 357 bio->bi_private = cl; 358 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags); 359 bch_bio_map(bio, c->uuids); 360 361 bch_submit_bbio(bio, c, k, i); 362 363 if (op != REQ_OP_WRITE) 364 break; 365 } 366 367 bch_extent_to_text(buf, sizeof(buf), k); 368 pr_debug("%s UUIDs at %s\n", op == REQ_OP_WRITE ? "wrote" : "read", buf); 369 370 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++) 371 if (!bch_is_zero(u->uuid, 16)) 372 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u\n", 373 u - c->uuids, u->uuid, u->label, 374 u->first_reg, u->last_reg, u->invalidated); 375 376 closure_return_with_destructor(cl, uuid_io_unlock); 377 } 378 379 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl) 380 { 381 struct bkey *k = &j->uuid_bucket; 382 383 if (__bch_btree_ptr_invalid(c, k)) 384 return "bad uuid pointer"; 385 386 bkey_copy(&c->uuid_bucket, k); 387 uuid_io(c, REQ_OP_READ, 0, k, cl); 388 389 if (j->version < BCACHE_JSET_VERSION_UUIDv1) { 390 struct uuid_entry_v0 *u0 = (void *) c->uuids; 391 struct uuid_entry *u1 = (void *) c->uuids; 392 int i; 393 394 closure_sync(cl); 395 396 /* 397 * Since the new uuid entry is bigger than the old, we have to 398 * convert starting at the highest memory address and work down 399 * in order to do it in place 400 */ 401 402 for (i = c->nr_uuids - 1; 403 i >= 0; 404 --i) { 405 memcpy(u1[i].uuid, u0[i].uuid, 16); 406 memcpy(u1[i].label, u0[i].label, 32); 407 408 u1[i].first_reg = u0[i].first_reg; 409 u1[i].last_reg = u0[i].last_reg; 410 u1[i].invalidated = u0[i].invalidated; 411 412 u1[i].flags = 0; 413 u1[i].sectors = 0; 414 } 415 } 416 417 return NULL; 418 } 419 420 static int __uuid_write(struct cache_set *c) 421 { 422 BKEY_PADDED(key) k; 423 struct closure cl; 424 struct cache *ca; 425 426 closure_init_stack(&cl); 427 lockdep_assert_held(&bch_register_lock); 428 429 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true)) 430 return 1; 431 432 SET_KEY_SIZE(&k.key, c->sb.bucket_size); 433 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl); 434 closure_sync(&cl); 435 436 /* Only one bucket used for uuid write */ 437 ca = PTR_CACHE(c, &k.key, 0); 438 atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written); 439 440 bkey_copy(&c->uuid_bucket, &k.key); 441 bkey_put(c, &k.key); 442 return 0; 443 } 444 445 int bch_uuid_write(struct cache_set *c) 446 { 447 int ret = __uuid_write(c); 448 449 if (!ret) 450 bch_journal_meta(c, NULL); 451 452 return ret; 453 } 454 455 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid) 456 { 457 struct uuid_entry *u; 458 459 for (u = c->uuids; 460 u < c->uuids + c->nr_uuids; u++) 461 if (!memcmp(u->uuid, uuid, 16)) 462 return u; 463 464 return NULL; 465 } 466 467 static struct uuid_entry *uuid_find_empty(struct cache_set *c) 468 { 469 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"; 470 471 return uuid_find(c, zero_uuid); 472 } 473 474 /* 475 * Bucket priorities/gens: 476 * 477 * For each bucket, we store on disk its 478 * 8 bit gen 479 * 16 bit priority 480 * 481 * See alloc.c for an explanation of the gen. The priority is used to implement 482 * lru (and in the future other) cache replacement policies; for most purposes 483 * it's just an opaque integer. 484 * 485 * The gens and the priorities don't have a whole lot to do with each other, and 486 * it's actually the gens that must be written out at specific times - it's no 487 * big deal if the priorities don't get written, if we lose them we just reuse 488 * buckets in suboptimal order. 489 * 490 * On disk they're stored in a packed array, and in as many buckets are required 491 * to fit them all. The buckets we use to store them form a list; the journal 492 * header points to the first bucket, the first bucket points to the second 493 * bucket, et cetera. 494 * 495 * This code is used by the allocation code; periodically (whenever it runs out 496 * of buckets to allocate from) the allocation code will invalidate some 497 * buckets, but it can't use those buckets until their new gens are safely on 498 * disk. 499 */ 500 501 static void prio_endio(struct bio *bio) 502 { 503 struct cache *ca = bio->bi_private; 504 505 cache_set_err_on(bio->bi_status, ca->set, "accessing priorities"); 506 bch_bbio_free(bio, ca->set); 507 closure_put(&ca->prio); 508 } 509 510 static void prio_io(struct cache *ca, uint64_t bucket, int op, 511 unsigned long op_flags) 512 { 513 struct closure *cl = &ca->prio; 514 struct bio *bio = bch_bbio_alloc(ca->set); 515 516 closure_init_stack(cl); 517 518 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size; 519 bio_set_dev(bio, ca->bdev); 520 bio->bi_iter.bi_size = bucket_bytes(ca); 521 522 bio->bi_end_io = prio_endio; 523 bio->bi_private = ca; 524 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags); 525 bch_bio_map(bio, ca->disk_buckets); 526 527 closure_bio_submit(ca->set, bio, &ca->prio); 528 closure_sync(cl); 529 } 530 531 int bch_prio_write(struct cache *ca, bool wait) 532 { 533 int i; 534 struct bucket *b; 535 struct closure cl; 536 537 pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu\n", 538 fifo_used(&ca->free[RESERVE_PRIO]), 539 fifo_used(&ca->free[RESERVE_NONE]), 540 fifo_used(&ca->free_inc)); 541 542 /* 543 * Pre-check if there are enough free buckets. In the non-blocking 544 * scenario it's better to fail early rather than starting to allocate 545 * buckets and do a cleanup later in case of failure. 546 */ 547 if (!wait) { 548 size_t avail = fifo_used(&ca->free[RESERVE_PRIO]) + 549 fifo_used(&ca->free[RESERVE_NONE]); 550 if (prio_buckets(ca) > avail) 551 return -ENOMEM; 552 } 553 554 closure_init_stack(&cl); 555 556 lockdep_assert_held(&ca->set->bucket_lock); 557 558 ca->disk_buckets->seq++; 559 560 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca), 561 &ca->meta_sectors_written); 562 563 for (i = prio_buckets(ca) - 1; i >= 0; --i) { 564 long bucket; 565 struct prio_set *p = ca->disk_buckets; 566 struct bucket_disk *d = p->data; 567 struct bucket_disk *end = d + prios_per_bucket(ca); 568 569 for (b = ca->buckets + i * prios_per_bucket(ca); 570 b < ca->buckets + ca->sb.nbuckets && d < end; 571 b++, d++) { 572 d->prio = cpu_to_le16(b->prio); 573 d->gen = b->gen; 574 } 575 576 p->next_bucket = ca->prio_buckets[i + 1]; 577 p->magic = pset_magic(&ca->sb); 578 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8); 579 580 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait); 581 BUG_ON(bucket == -1); 582 583 mutex_unlock(&ca->set->bucket_lock); 584 prio_io(ca, bucket, REQ_OP_WRITE, 0); 585 mutex_lock(&ca->set->bucket_lock); 586 587 ca->prio_buckets[i] = bucket; 588 atomic_dec_bug(&ca->buckets[bucket].pin); 589 } 590 591 mutex_unlock(&ca->set->bucket_lock); 592 593 bch_journal_meta(ca->set, &cl); 594 closure_sync(&cl); 595 596 mutex_lock(&ca->set->bucket_lock); 597 598 /* 599 * Don't want the old priorities to get garbage collected until after we 600 * finish writing the new ones, and they're journalled 601 */ 602 for (i = 0; i < prio_buckets(ca); i++) { 603 if (ca->prio_last_buckets[i]) 604 __bch_bucket_free(ca, 605 &ca->buckets[ca->prio_last_buckets[i]]); 606 607 ca->prio_last_buckets[i] = ca->prio_buckets[i]; 608 } 609 return 0; 610 } 611 612 static int prio_read(struct cache *ca, uint64_t bucket) 613 { 614 struct prio_set *p = ca->disk_buckets; 615 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d; 616 struct bucket *b; 617 unsigned int bucket_nr = 0; 618 int ret = -EIO; 619 620 for (b = ca->buckets; 621 b < ca->buckets + ca->sb.nbuckets; 622 b++, d++) { 623 if (d == end) { 624 ca->prio_buckets[bucket_nr] = bucket; 625 ca->prio_last_buckets[bucket_nr] = bucket; 626 bucket_nr++; 627 628 prio_io(ca, bucket, REQ_OP_READ, 0); 629 630 if (p->csum != 631 bch_crc64(&p->magic, bucket_bytes(ca) - 8)) { 632 pr_warn("bad csum reading priorities\n"); 633 goto out; 634 } 635 636 if (p->magic != pset_magic(&ca->sb)) { 637 pr_warn("bad magic reading priorities\n"); 638 goto out; 639 } 640 641 bucket = p->next_bucket; 642 d = p->data; 643 } 644 645 b->prio = le16_to_cpu(d->prio); 646 b->gen = b->last_gc = d->gen; 647 } 648 649 ret = 0; 650 out: 651 return ret; 652 } 653 654 /* Bcache device */ 655 656 static int open_dev(struct block_device *b, fmode_t mode) 657 { 658 struct bcache_device *d = b->bd_disk->private_data; 659 660 if (test_bit(BCACHE_DEV_CLOSING, &d->flags)) 661 return -ENXIO; 662 663 closure_get(&d->cl); 664 return 0; 665 } 666 667 static void release_dev(struct gendisk *b, fmode_t mode) 668 { 669 struct bcache_device *d = b->private_data; 670 671 closure_put(&d->cl); 672 } 673 674 static int ioctl_dev(struct block_device *b, fmode_t mode, 675 unsigned int cmd, unsigned long arg) 676 { 677 struct bcache_device *d = b->bd_disk->private_data; 678 679 return d->ioctl(d, mode, cmd, arg); 680 } 681 682 static const struct block_device_operations bcache_ops = { 683 .open = open_dev, 684 .release = release_dev, 685 .ioctl = ioctl_dev, 686 .owner = THIS_MODULE, 687 }; 688 689 void bcache_device_stop(struct bcache_device *d) 690 { 691 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags)) 692 /* 693 * closure_fn set to 694 * - cached device: cached_dev_flush() 695 * - flash dev: flash_dev_flush() 696 */ 697 closure_queue(&d->cl); 698 } 699 700 static void bcache_device_unlink(struct bcache_device *d) 701 { 702 lockdep_assert_held(&bch_register_lock); 703 704 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) { 705 unsigned int i; 706 struct cache *ca; 707 708 sysfs_remove_link(&d->c->kobj, d->name); 709 sysfs_remove_link(&d->kobj, "cache"); 710 711 for_each_cache(ca, d->c, i) 712 bd_unlink_disk_holder(ca->bdev, d->disk); 713 } 714 } 715 716 static void bcache_device_link(struct bcache_device *d, struct cache_set *c, 717 const char *name) 718 { 719 unsigned int i; 720 struct cache *ca; 721 int ret; 722 723 for_each_cache(ca, d->c, i) 724 bd_link_disk_holder(ca->bdev, d->disk); 725 726 snprintf(d->name, BCACHEDEVNAME_SIZE, 727 "%s%u", name, d->id); 728 729 ret = sysfs_create_link(&d->kobj, &c->kobj, "cache"); 730 if (ret < 0) 731 pr_err("Couldn't create device -> cache set symlink\n"); 732 733 ret = sysfs_create_link(&c->kobj, &d->kobj, d->name); 734 if (ret < 0) 735 pr_err("Couldn't create cache set -> device symlink\n"); 736 737 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags); 738 } 739 740 static void bcache_device_detach(struct bcache_device *d) 741 { 742 lockdep_assert_held(&bch_register_lock); 743 744 atomic_dec(&d->c->attached_dev_nr); 745 746 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) { 747 struct uuid_entry *u = d->c->uuids + d->id; 748 749 SET_UUID_FLASH_ONLY(u, 0); 750 memcpy(u->uuid, invalid_uuid, 16); 751 u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds()); 752 bch_uuid_write(d->c); 753 } 754 755 bcache_device_unlink(d); 756 757 d->c->devices[d->id] = NULL; 758 closure_put(&d->c->caching); 759 d->c = NULL; 760 } 761 762 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c, 763 unsigned int id) 764 { 765 d->id = id; 766 d->c = c; 767 c->devices[id] = d; 768 769 if (id >= c->devices_max_used) 770 c->devices_max_used = id + 1; 771 772 closure_get(&c->caching); 773 } 774 775 static inline int first_minor_to_idx(int first_minor) 776 { 777 return (first_minor/BCACHE_MINORS); 778 } 779 780 static inline int idx_to_first_minor(int idx) 781 { 782 return (idx * BCACHE_MINORS); 783 } 784 785 static void bcache_device_free(struct bcache_device *d) 786 { 787 struct gendisk *disk = d->disk; 788 789 lockdep_assert_held(&bch_register_lock); 790 791 if (disk) 792 pr_info("%s stopped\n", disk->disk_name); 793 else 794 pr_err("bcache device (NULL gendisk) stopped\n"); 795 796 if (d->c) 797 bcache_device_detach(d); 798 799 if (disk) { 800 if (disk->flags & GENHD_FL_UP) 801 del_gendisk(disk); 802 803 if (disk->queue) 804 blk_cleanup_queue(disk->queue); 805 806 ida_simple_remove(&bcache_device_idx, 807 first_minor_to_idx(disk->first_minor)); 808 put_disk(disk); 809 } 810 811 bioset_exit(&d->bio_split); 812 kvfree(d->full_dirty_stripes); 813 kvfree(d->stripe_sectors_dirty); 814 815 closure_debug_destroy(&d->cl); 816 } 817 818 static int bcache_device_init(struct bcache_device *d, unsigned int block_size, 819 sector_t sectors, make_request_fn make_request_fn) 820 { 821 struct request_queue *q; 822 const size_t max_stripes = min_t(size_t, INT_MAX, 823 SIZE_MAX / sizeof(atomic_t)); 824 size_t n; 825 int idx; 826 827 if (!d->stripe_size) 828 d->stripe_size = 1 << 31; 829 830 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size); 831 832 if (!d->nr_stripes || d->nr_stripes > max_stripes) { 833 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)\n", 834 (unsigned int)d->nr_stripes); 835 return -ENOMEM; 836 } 837 838 n = d->nr_stripes * sizeof(atomic_t); 839 d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL); 840 if (!d->stripe_sectors_dirty) 841 return -ENOMEM; 842 843 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long); 844 d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL); 845 if (!d->full_dirty_stripes) 846 return -ENOMEM; 847 848 idx = ida_simple_get(&bcache_device_idx, 0, 849 BCACHE_DEVICE_IDX_MAX, GFP_KERNEL); 850 if (idx < 0) 851 return idx; 852 853 if (bioset_init(&d->bio_split, 4, offsetof(struct bbio, bio), 854 BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER)) 855 goto err; 856 857 d->disk = alloc_disk(BCACHE_MINORS); 858 if (!d->disk) 859 goto err; 860 861 set_capacity(d->disk, sectors); 862 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx); 863 864 d->disk->major = bcache_major; 865 d->disk->first_minor = idx_to_first_minor(idx); 866 d->disk->fops = &bcache_ops; 867 d->disk->private_data = d; 868 869 q = blk_alloc_queue(make_request_fn, NUMA_NO_NODE); 870 if (!q) 871 return -ENOMEM; 872 873 d->disk->queue = q; 874 q->queuedata = d; 875 q->backing_dev_info->congested_data = d; 876 q->limits.max_hw_sectors = UINT_MAX; 877 q->limits.max_sectors = UINT_MAX; 878 q->limits.max_segment_size = UINT_MAX; 879 q->limits.max_segments = BIO_MAX_PAGES; 880 blk_queue_max_discard_sectors(q, UINT_MAX); 881 q->limits.discard_granularity = 512; 882 q->limits.io_min = block_size; 883 q->limits.logical_block_size = block_size; 884 q->limits.physical_block_size = block_size; 885 blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue); 886 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, d->disk->queue); 887 blk_queue_flag_set(QUEUE_FLAG_DISCARD, d->disk->queue); 888 889 blk_queue_write_cache(q, true, true); 890 891 return 0; 892 893 err: 894 ida_simple_remove(&bcache_device_idx, idx); 895 return -ENOMEM; 896 897 } 898 899 /* Cached device */ 900 901 static void calc_cached_dev_sectors(struct cache_set *c) 902 { 903 uint64_t sectors = 0; 904 struct cached_dev *dc; 905 906 list_for_each_entry(dc, &c->cached_devs, list) 907 sectors += bdev_sectors(dc->bdev); 908 909 c->cached_dev_sectors = sectors; 910 } 911 912 #define BACKING_DEV_OFFLINE_TIMEOUT 5 913 static int cached_dev_status_update(void *arg) 914 { 915 struct cached_dev *dc = arg; 916 struct request_queue *q; 917 918 /* 919 * If this delayed worker is stopping outside, directly quit here. 920 * dc->io_disable might be set via sysfs interface, so check it 921 * here too. 922 */ 923 while (!kthread_should_stop() && !dc->io_disable) { 924 q = bdev_get_queue(dc->bdev); 925 if (blk_queue_dying(q)) 926 dc->offline_seconds++; 927 else 928 dc->offline_seconds = 0; 929 930 if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) { 931 pr_err("%s: device offline for %d seconds\n", 932 dc->backing_dev_name, 933 BACKING_DEV_OFFLINE_TIMEOUT); 934 pr_err("%s: disable I/O request due to backing device offline\n", 935 dc->disk.name); 936 dc->io_disable = true; 937 /* let others know earlier that io_disable is true */ 938 smp_mb(); 939 bcache_device_stop(&dc->disk); 940 break; 941 } 942 schedule_timeout_interruptible(HZ); 943 } 944 945 wait_for_kthread_stop(); 946 return 0; 947 } 948 949 950 int bch_cached_dev_run(struct cached_dev *dc) 951 { 952 struct bcache_device *d = &dc->disk; 953 char *buf = kmemdup_nul(dc->sb.label, SB_LABEL_SIZE, GFP_KERNEL); 954 char *env[] = { 955 "DRIVER=bcache", 956 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid), 957 kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf ? : ""), 958 NULL, 959 }; 960 961 if (dc->io_disable) { 962 pr_err("I/O disabled on cached dev %s\n", 963 dc->backing_dev_name); 964 kfree(env[1]); 965 kfree(env[2]); 966 kfree(buf); 967 return -EIO; 968 } 969 970 if (atomic_xchg(&dc->running, 1)) { 971 kfree(env[1]); 972 kfree(env[2]); 973 kfree(buf); 974 pr_info("cached dev %s is running already\n", 975 dc->backing_dev_name); 976 return -EBUSY; 977 } 978 979 if (!d->c && 980 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) { 981 struct closure cl; 982 983 closure_init_stack(&cl); 984 985 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE); 986 bch_write_bdev_super(dc, &cl); 987 closure_sync(&cl); 988 } 989 990 add_disk(d->disk); 991 bd_link_disk_holder(dc->bdev, dc->disk.disk); 992 /* 993 * won't show up in the uevent file, use udevadm monitor -e instead 994 * only class / kset properties are persistent 995 */ 996 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env); 997 kfree(env[1]); 998 kfree(env[2]); 999 kfree(buf); 1000 1001 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") || 1002 sysfs_create_link(&disk_to_dev(d->disk)->kobj, 1003 &d->kobj, "bcache")) { 1004 pr_err("Couldn't create bcache dev <-> disk sysfs symlinks\n"); 1005 return -ENOMEM; 1006 } 1007 1008 dc->status_update_thread = kthread_run(cached_dev_status_update, 1009 dc, "bcache_status_update"); 1010 if (IS_ERR(dc->status_update_thread)) { 1011 pr_warn("failed to create bcache_status_update kthread, continue to run without monitoring backing device status\n"); 1012 } 1013 1014 return 0; 1015 } 1016 1017 /* 1018 * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed 1019 * work dc->writeback_rate_update is running. Wait until the routine 1020 * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to 1021 * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out 1022 * seconds, give up waiting here and continue to cancel it too. 1023 */ 1024 static void cancel_writeback_rate_update_dwork(struct cached_dev *dc) 1025 { 1026 int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ; 1027 1028 do { 1029 if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING, 1030 &dc->disk.flags)) 1031 break; 1032 time_out--; 1033 schedule_timeout_interruptible(1); 1034 } while (time_out > 0); 1035 1036 if (time_out == 0) 1037 pr_warn("give up waiting for dc->writeback_write_update to quit\n"); 1038 1039 cancel_delayed_work_sync(&dc->writeback_rate_update); 1040 } 1041 1042 static void cached_dev_detach_finish(struct work_struct *w) 1043 { 1044 struct cached_dev *dc = container_of(w, struct cached_dev, detach); 1045 struct closure cl; 1046 1047 closure_init_stack(&cl); 1048 1049 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)); 1050 BUG_ON(refcount_read(&dc->count)); 1051 1052 1053 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) 1054 cancel_writeback_rate_update_dwork(dc); 1055 1056 if (!IS_ERR_OR_NULL(dc->writeback_thread)) { 1057 kthread_stop(dc->writeback_thread); 1058 dc->writeback_thread = NULL; 1059 } 1060 1061 memset(&dc->sb.set_uuid, 0, 16); 1062 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE); 1063 1064 bch_write_bdev_super(dc, &cl); 1065 closure_sync(&cl); 1066 1067 mutex_lock(&bch_register_lock); 1068 1069 calc_cached_dev_sectors(dc->disk.c); 1070 bcache_device_detach(&dc->disk); 1071 list_move(&dc->list, &uncached_devices); 1072 1073 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags); 1074 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags); 1075 1076 mutex_unlock(&bch_register_lock); 1077 1078 pr_info("Caching disabled for %s\n", dc->backing_dev_name); 1079 1080 /* Drop ref we took in cached_dev_detach() */ 1081 closure_put(&dc->disk.cl); 1082 } 1083 1084 void bch_cached_dev_detach(struct cached_dev *dc) 1085 { 1086 lockdep_assert_held(&bch_register_lock); 1087 1088 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) 1089 return; 1090 1091 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) 1092 return; 1093 1094 /* 1095 * Block the device from being closed and freed until we're finished 1096 * detaching 1097 */ 1098 closure_get(&dc->disk.cl); 1099 1100 bch_writeback_queue(dc); 1101 1102 cached_dev_put(dc); 1103 } 1104 1105 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c, 1106 uint8_t *set_uuid) 1107 { 1108 uint32_t rtime = cpu_to_le32((u32)ktime_get_real_seconds()); 1109 struct uuid_entry *u; 1110 struct cached_dev *exist_dc, *t; 1111 int ret = 0; 1112 1113 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) || 1114 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))) 1115 return -ENOENT; 1116 1117 if (dc->disk.c) { 1118 pr_err("Can't attach %s: already attached\n", 1119 dc->backing_dev_name); 1120 return -EINVAL; 1121 } 1122 1123 if (test_bit(CACHE_SET_STOPPING, &c->flags)) { 1124 pr_err("Can't attach %s: shutting down\n", 1125 dc->backing_dev_name); 1126 return -EINVAL; 1127 } 1128 1129 if (dc->sb.block_size < c->sb.block_size) { 1130 /* Will die */ 1131 pr_err("Couldn't attach %s: block size less than set's block size\n", 1132 dc->backing_dev_name); 1133 return -EINVAL; 1134 } 1135 1136 /* Check whether already attached */ 1137 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) { 1138 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) { 1139 pr_err("Tried to attach %s but duplicate UUID already attached\n", 1140 dc->backing_dev_name); 1141 1142 return -EINVAL; 1143 } 1144 } 1145 1146 u = uuid_find(c, dc->sb.uuid); 1147 1148 if (u && 1149 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE || 1150 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) { 1151 memcpy(u->uuid, invalid_uuid, 16); 1152 u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds()); 1153 u = NULL; 1154 } 1155 1156 if (!u) { 1157 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1158 pr_err("Couldn't find uuid for %s in set\n", 1159 dc->backing_dev_name); 1160 return -ENOENT; 1161 } 1162 1163 u = uuid_find_empty(c); 1164 if (!u) { 1165 pr_err("Not caching %s, no room for UUID\n", 1166 dc->backing_dev_name); 1167 return -EINVAL; 1168 } 1169 } 1170 1171 /* 1172 * Deadlocks since we're called via sysfs... 1173 * sysfs_remove_file(&dc->kobj, &sysfs_attach); 1174 */ 1175 1176 if (bch_is_zero(u->uuid, 16)) { 1177 struct closure cl; 1178 1179 closure_init_stack(&cl); 1180 1181 memcpy(u->uuid, dc->sb.uuid, 16); 1182 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE); 1183 u->first_reg = u->last_reg = rtime; 1184 bch_uuid_write(c); 1185 1186 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16); 1187 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN); 1188 1189 bch_write_bdev_super(dc, &cl); 1190 closure_sync(&cl); 1191 } else { 1192 u->last_reg = rtime; 1193 bch_uuid_write(c); 1194 } 1195 1196 bcache_device_attach(&dc->disk, c, u - c->uuids); 1197 list_move(&dc->list, &c->cached_devs); 1198 calc_cached_dev_sectors(c); 1199 1200 /* 1201 * dc->c must be set before dc->count != 0 - paired with the mb in 1202 * cached_dev_get() 1203 */ 1204 smp_wmb(); 1205 refcount_set(&dc->count, 1); 1206 1207 /* Block writeback thread, but spawn it */ 1208 down_write(&dc->writeback_lock); 1209 if (bch_cached_dev_writeback_start(dc)) { 1210 up_write(&dc->writeback_lock); 1211 pr_err("Couldn't start writeback facilities for %s\n", 1212 dc->disk.disk->disk_name); 1213 return -ENOMEM; 1214 } 1215 1216 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { 1217 atomic_set(&dc->has_dirty, 1); 1218 bch_writeback_queue(dc); 1219 } 1220 1221 bch_sectors_dirty_init(&dc->disk); 1222 1223 ret = bch_cached_dev_run(dc); 1224 if (ret && (ret != -EBUSY)) { 1225 up_write(&dc->writeback_lock); 1226 /* 1227 * bch_register_lock is held, bcache_device_stop() is not 1228 * able to be directly called. The kthread and kworker 1229 * created previously in bch_cached_dev_writeback_start() 1230 * have to be stopped manually here. 1231 */ 1232 kthread_stop(dc->writeback_thread); 1233 cancel_writeback_rate_update_dwork(dc); 1234 pr_err("Couldn't run cached device %s\n", 1235 dc->backing_dev_name); 1236 return ret; 1237 } 1238 1239 bcache_device_link(&dc->disk, c, "bdev"); 1240 atomic_inc(&c->attached_dev_nr); 1241 1242 /* Allow the writeback thread to proceed */ 1243 up_write(&dc->writeback_lock); 1244 1245 pr_info("Caching %s as %s on set %pU\n", 1246 dc->backing_dev_name, 1247 dc->disk.disk->disk_name, 1248 dc->disk.c->sb.set_uuid); 1249 return 0; 1250 } 1251 1252 /* when dc->disk.kobj released */ 1253 void bch_cached_dev_release(struct kobject *kobj) 1254 { 1255 struct cached_dev *dc = container_of(kobj, struct cached_dev, 1256 disk.kobj); 1257 kfree(dc); 1258 module_put(THIS_MODULE); 1259 } 1260 1261 static void cached_dev_free(struct closure *cl) 1262 { 1263 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1264 1265 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)) 1266 cancel_writeback_rate_update_dwork(dc); 1267 1268 if (!IS_ERR_OR_NULL(dc->writeback_thread)) 1269 kthread_stop(dc->writeback_thread); 1270 if (!IS_ERR_OR_NULL(dc->status_update_thread)) 1271 kthread_stop(dc->status_update_thread); 1272 1273 mutex_lock(&bch_register_lock); 1274 1275 if (atomic_read(&dc->running)) 1276 bd_unlink_disk_holder(dc->bdev, dc->disk.disk); 1277 bcache_device_free(&dc->disk); 1278 list_del(&dc->list); 1279 1280 mutex_unlock(&bch_register_lock); 1281 1282 if (dc->sb_disk) 1283 put_page(virt_to_page(dc->sb_disk)); 1284 1285 if (!IS_ERR_OR_NULL(dc->bdev)) 1286 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1287 1288 wake_up(&unregister_wait); 1289 1290 kobject_put(&dc->disk.kobj); 1291 } 1292 1293 static void cached_dev_flush(struct closure *cl) 1294 { 1295 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl); 1296 struct bcache_device *d = &dc->disk; 1297 1298 mutex_lock(&bch_register_lock); 1299 bcache_device_unlink(d); 1300 mutex_unlock(&bch_register_lock); 1301 1302 bch_cache_accounting_destroy(&dc->accounting); 1303 kobject_del(&d->kobj); 1304 1305 continue_at(cl, cached_dev_free, system_wq); 1306 } 1307 1308 static int cached_dev_init(struct cached_dev *dc, unsigned int block_size) 1309 { 1310 int ret; 1311 struct io *io; 1312 struct request_queue *q = bdev_get_queue(dc->bdev); 1313 1314 __module_get(THIS_MODULE); 1315 INIT_LIST_HEAD(&dc->list); 1316 closure_init(&dc->disk.cl, NULL); 1317 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq); 1318 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype); 1319 INIT_WORK(&dc->detach, cached_dev_detach_finish); 1320 sema_init(&dc->sb_write_mutex, 1); 1321 INIT_LIST_HEAD(&dc->io_lru); 1322 spin_lock_init(&dc->io_lock); 1323 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl); 1324 1325 dc->sequential_cutoff = 4 << 20; 1326 1327 for (io = dc->io; io < dc->io + RECENT_IO; io++) { 1328 list_add(&io->lru, &dc->io_lru); 1329 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO); 1330 } 1331 1332 dc->disk.stripe_size = q->limits.io_opt >> 9; 1333 1334 if (dc->disk.stripe_size) 1335 dc->partial_stripes_expensive = 1336 q->limits.raid_partial_stripes_expensive; 1337 1338 ret = bcache_device_init(&dc->disk, block_size, 1339 dc->bdev->bd_part->nr_sects - dc->sb.data_offset, 1340 cached_dev_make_request); 1341 if (ret) 1342 return ret; 1343 1344 dc->disk.disk->queue->backing_dev_info->ra_pages = 1345 max(dc->disk.disk->queue->backing_dev_info->ra_pages, 1346 q->backing_dev_info->ra_pages); 1347 1348 atomic_set(&dc->io_errors, 0); 1349 dc->io_disable = false; 1350 dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT; 1351 /* default to auto */ 1352 dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO; 1353 1354 bch_cached_dev_request_init(dc); 1355 bch_cached_dev_writeback_init(dc); 1356 return 0; 1357 } 1358 1359 /* Cached device - bcache superblock */ 1360 1361 static int register_bdev(struct cache_sb *sb, struct cache_sb_disk *sb_disk, 1362 struct block_device *bdev, 1363 struct cached_dev *dc) 1364 { 1365 const char *err = "cannot allocate memory"; 1366 struct cache_set *c; 1367 int ret = -ENOMEM; 1368 1369 bdevname(bdev, dc->backing_dev_name); 1370 memcpy(&dc->sb, sb, sizeof(struct cache_sb)); 1371 dc->bdev = bdev; 1372 dc->bdev->bd_holder = dc; 1373 dc->sb_disk = sb_disk; 1374 1375 if (cached_dev_init(dc, sb->block_size << 9)) 1376 goto err; 1377 1378 err = "error creating kobject"; 1379 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj, 1380 "bcache")) 1381 goto err; 1382 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj)) 1383 goto err; 1384 1385 pr_info("registered backing device %s\n", dc->backing_dev_name); 1386 1387 list_add(&dc->list, &uncached_devices); 1388 /* attach to a matched cache set if it exists */ 1389 list_for_each_entry(c, &bch_cache_sets, list) 1390 bch_cached_dev_attach(dc, c, NULL); 1391 1392 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE || 1393 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) { 1394 err = "failed to run cached device"; 1395 ret = bch_cached_dev_run(dc); 1396 if (ret) 1397 goto err; 1398 } 1399 1400 return 0; 1401 err: 1402 pr_notice("error %s: %s\n", dc->backing_dev_name, err); 1403 bcache_device_stop(&dc->disk); 1404 return ret; 1405 } 1406 1407 /* Flash only volumes */ 1408 1409 /* When d->kobj released */ 1410 void bch_flash_dev_release(struct kobject *kobj) 1411 { 1412 struct bcache_device *d = container_of(kobj, struct bcache_device, 1413 kobj); 1414 kfree(d); 1415 } 1416 1417 static void flash_dev_free(struct closure *cl) 1418 { 1419 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1420 1421 mutex_lock(&bch_register_lock); 1422 atomic_long_sub(bcache_dev_sectors_dirty(d), 1423 &d->c->flash_dev_dirty_sectors); 1424 bcache_device_free(d); 1425 mutex_unlock(&bch_register_lock); 1426 kobject_put(&d->kobj); 1427 } 1428 1429 static void flash_dev_flush(struct closure *cl) 1430 { 1431 struct bcache_device *d = container_of(cl, struct bcache_device, cl); 1432 1433 mutex_lock(&bch_register_lock); 1434 bcache_device_unlink(d); 1435 mutex_unlock(&bch_register_lock); 1436 kobject_del(&d->kobj); 1437 continue_at(cl, flash_dev_free, system_wq); 1438 } 1439 1440 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) 1441 { 1442 struct bcache_device *d = kzalloc(sizeof(struct bcache_device), 1443 GFP_KERNEL); 1444 if (!d) 1445 return -ENOMEM; 1446 1447 closure_init(&d->cl, NULL); 1448 set_closure_fn(&d->cl, flash_dev_flush, system_wq); 1449 1450 kobject_init(&d->kobj, &bch_flash_dev_ktype); 1451 1452 if (bcache_device_init(d, block_bytes(c), u->sectors, 1453 flash_dev_make_request)) 1454 goto err; 1455 1456 bcache_device_attach(d, c, u - c->uuids); 1457 bch_sectors_dirty_init(d); 1458 bch_flash_dev_request_init(d); 1459 add_disk(d->disk); 1460 1461 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache")) 1462 goto err; 1463 1464 bcache_device_link(d, c, "volume"); 1465 1466 return 0; 1467 err: 1468 kobject_put(&d->kobj); 1469 return -ENOMEM; 1470 } 1471 1472 static int flash_devs_run(struct cache_set *c) 1473 { 1474 int ret = 0; 1475 struct uuid_entry *u; 1476 1477 for (u = c->uuids; 1478 u < c->uuids + c->nr_uuids && !ret; 1479 u++) 1480 if (UUID_FLASH_ONLY(u)) 1481 ret = flash_dev_run(c, u); 1482 1483 return ret; 1484 } 1485 1486 int bch_flash_dev_create(struct cache_set *c, uint64_t size) 1487 { 1488 struct uuid_entry *u; 1489 1490 if (test_bit(CACHE_SET_STOPPING, &c->flags)) 1491 return -EINTR; 1492 1493 if (!test_bit(CACHE_SET_RUNNING, &c->flags)) 1494 return -EPERM; 1495 1496 u = uuid_find_empty(c); 1497 if (!u) { 1498 pr_err("Can't create volume, no room for UUID\n"); 1499 return -EINVAL; 1500 } 1501 1502 get_random_bytes(u->uuid, 16); 1503 memset(u->label, 0, 32); 1504 u->first_reg = u->last_reg = cpu_to_le32((u32)ktime_get_real_seconds()); 1505 1506 SET_UUID_FLASH_ONLY(u, 1); 1507 u->sectors = size >> 9; 1508 1509 bch_uuid_write(c); 1510 1511 return flash_dev_run(c, u); 1512 } 1513 1514 bool bch_cached_dev_error(struct cached_dev *dc) 1515 { 1516 if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags)) 1517 return false; 1518 1519 dc->io_disable = true; 1520 /* make others know io_disable is true earlier */ 1521 smp_mb(); 1522 1523 pr_err("stop %s: too many IO errors on backing device %s\n", 1524 dc->disk.disk->disk_name, dc->backing_dev_name); 1525 1526 bcache_device_stop(&dc->disk); 1527 return true; 1528 } 1529 1530 /* Cache set */ 1531 1532 __printf(2, 3) 1533 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...) 1534 { 1535 struct va_format vaf; 1536 va_list args; 1537 1538 if (c->on_error != ON_ERROR_PANIC && 1539 test_bit(CACHE_SET_STOPPING, &c->flags)) 1540 return false; 1541 1542 if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags)) 1543 pr_info("CACHE_SET_IO_DISABLE already set\n"); 1544 1545 /* 1546 * XXX: we can be called from atomic context 1547 * acquire_console_sem(); 1548 */ 1549 1550 va_start(args, fmt); 1551 1552 vaf.fmt = fmt; 1553 vaf.va = &args; 1554 1555 pr_err("error on %pU: %pV, disabling caching\n", 1556 c->sb.set_uuid, &vaf); 1557 1558 va_end(args); 1559 1560 if (c->on_error == ON_ERROR_PANIC) 1561 panic("panic forced after error\n"); 1562 1563 bch_cache_set_unregister(c); 1564 return true; 1565 } 1566 1567 /* When c->kobj released */ 1568 void bch_cache_set_release(struct kobject *kobj) 1569 { 1570 struct cache_set *c = container_of(kobj, struct cache_set, kobj); 1571 1572 kfree(c); 1573 module_put(THIS_MODULE); 1574 } 1575 1576 static void cache_set_free(struct closure *cl) 1577 { 1578 struct cache_set *c = container_of(cl, struct cache_set, cl); 1579 struct cache *ca; 1580 unsigned int i; 1581 1582 debugfs_remove(c->debug); 1583 1584 bch_open_buckets_free(c); 1585 bch_btree_cache_free(c); 1586 bch_journal_free(c); 1587 1588 mutex_lock(&bch_register_lock); 1589 for_each_cache(ca, c, i) 1590 if (ca) { 1591 ca->set = NULL; 1592 c->cache[ca->sb.nr_this_dev] = NULL; 1593 kobject_put(&ca->kobj); 1594 } 1595 1596 bch_bset_sort_state_free(&c->sort); 1597 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c))); 1598 1599 if (c->moving_gc_wq) 1600 destroy_workqueue(c->moving_gc_wq); 1601 bioset_exit(&c->bio_split); 1602 mempool_exit(&c->fill_iter); 1603 mempool_exit(&c->bio_meta); 1604 mempool_exit(&c->search); 1605 kfree(c->devices); 1606 1607 list_del(&c->list); 1608 mutex_unlock(&bch_register_lock); 1609 1610 pr_info("Cache set %pU unregistered\n", c->sb.set_uuid); 1611 wake_up(&unregister_wait); 1612 1613 closure_debug_destroy(&c->cl); 1614 kobject_put(&c->kobj); 1615 } 1616 1617 static void cache_set_flush(struct closure *cl) 1618 { 1619 struct cache_set *c = container_of(cl, struct cache_set, caching); 1620 struct cache *ca; 1621 struct btree *b; 1622 unsigned int i; 1623 1624 bch_cache_accounting_destroy(&c->accounting); 1625 1626 kobject_put(&c->internal); 1627 kobject_del(&c->kobj); 1628 1629 if (!IS_ERR_OR_NULL(c->gc_thread)) 1630 kthread_stop(c->gc_thread); 1631 1632 if (!IS_ERR_OR_NULL(c->root)) 1633 list_add(&c->root->list, &c->btree_cache); 1634 1635 /* 1636 * Avoid flushing cached nodes if cache set is retiring 1637 * due to too many I/O errors detected. 1638 */ 1639 if (!test_bit(CACHE_SET_IO_DISABLE, &c->flags)) 1640 list_for_each_entry(b, &c->btree_cache, list) { 1641 mutex_lock(&b->write_lock); 1642 if (btree_node_dirty(b)) 1643 __bch_btree_node_write(b, NULL); 1644 mutex_unlock(&b->write_lock); 1645 } 1646 1647 for_each_cache(ca, c, i) 1648 if (ca->alloc_thread) 1649 kthread_stop(ca->alloc_thread); 1650 1651 if (c->journal.cur) { 1652 cancel_delayed_work_sync(&c->journal.work); 1653 /* flush last journal entry if needed */ 1654 c->journal.work.work.func(&c->journal.work.work); 1655 } 1656 1657 closure_return(cl); 1658 } 1659 1660 /* 1661 * This function is only called when CACHE_SET_IO_DISABLE is set, which means 1662 * cache set is unregistering due to too many I/O errors. In this condition, 1663 * the bcache device might be stopped, it depends on stop_when_cache_set_failed 1664 * value and whether the broken cache has dirty data: 1665 * 1666 * dc->stop_when_cache_set_failed dc->has_dirty stop bcache device 1667 * BCH_CACHED_STOP_AUTO 0 NO 1668 * BCH_CACHED_STOP_AUTO 1 YES 1669 * BCH_CACHED_DEV_STOP_ALWAYS 0 YES 1670 * BCH_CACHED_DEV_STOP_ALWAYS 1 YES 1671 * 1672 * The expected behavior is, if stop_when_cache_set_failed is configured to 1673 * "auto" via sysfs interface, the bcache device will not be stopped if the 1674 * backing device is clean on the broken cache device. 1675 */ 1676 static void conditional_stop_bcache_device(struct cache_set *c, 1677 struct bcache_device *d, 1678 struct cached_dev *dc) 1679 { 1680 if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) { 1681 pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.\n", 1682 d->disk->disk_name, c->sb.set_uuid); 1683 bcache_device_stop(d); 1684 } else if (atomic_read(&dc->has_dirty)) { 1685 /* 1686 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO 1687 * and dc->has_dirty == 1 1688 */ 1689 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.\n", 1690 d->disk->disk_name); 1691 /* 1692 * There might be a small time gap that cache set is 1693 * released but bcache device is not. Inside this time 1694 * gap, regular I/O requests will directly go into 1695 * backing device as no cache set attached to. This 1696 * behavior may also introduce potential inconsistence 1697 * data in writeback mode while cache is dirty. 1698 * Therefore before calling bcache_device_stop() due 1699 * to a broken cache device, dc->io_disable should be 1700 * explicitly set to true. 1701 */ 1702 dc->io_disable = true; 1703 /* make others know io_disable is true earlier */ 1704 smp_mb(); 1705 bcache_device_stop(d); 1706 } else { 1707 /* 1708 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO 1709 * and dc->has_dirty == 0 1710 */ 1711 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.\n", 1712 d->disk->disk_name); 1713 } 1714 } 1715 1716 static void __cache_set_unregister(struct closure *cl) 1717 { 1718 struct cache_set *c = container_of(cl, struct cache_set, caching); 1719 struct cached_dev *dc; 1720 struct bcache_device *d; 1721 size_t i; 1722 1723 mutex_lock(&bch_register_lock); 1724 1725 for (i = 0; i < c->devices_max_used; i++) { 1726 d = c->devices[i]; 1727 if (!d) 1728 continue; 1729 1730 if (!UUID_FLASH_ONLY(&c->uuids[i]) && 1731 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) { 1732 dc = container_of(d, struct cached_dev, disk); 1733 bch_cached_dev_detach(dc); 1734 if (test_bit(CACHE_SET_IO_DISABLE, &c->flags)) 1735 conditional_stop_bcache_device(c, d, dc); 1736 } else { 1737 bcache_device_stop(d); 1738 } 1739 } 1740 1741 mutex_unlock(&bch_register_lock); 1742 1743 continue_at(cl, cache_set_flush, system_wq); 1744 } 1745 1746 void bch_cache_set_stop(struct cache_set *c) 1747 { 1748 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags)) 1749 /* closure_fn set to __cache_set_unregister() */ 1750 closure_queue(&c->caching); 1751 } 1752 1753 void bch_cache_set_unregister(struct cache_set *c) 1754 { 1755 set_bit(CACHE_SET_UNREGISTERING, &c->flags); 1756 bch_cache_set_stop(c); 1757 } 1758 1759 #define alloc_bucket_pages(gfp, c) \ 1760 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c)))) 1761 1762 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb) 1763 { 1764 int iter_size; 1765 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL); 1766 1767 if (!c) 1768 return NULL; 1769 1770 __module_get(THIS_MODULE); 1771 closure_init(&c->cl, NULL); 1772 set_closure_fn(&c->cl, cache_set_free, system_wq); 1773 1774 closure_init(&c->caching, &c->cl); 1775 set_closure_fn(&c->caching, __cache_set_unregister, system_wq); 1776 1777 /* Maybe create continue_at_noreturn() and use it here? */ 1778 closure_set_stopped(&c->cl); 1779 closure_put(&c->cl); 1780 1781 kobject_init(&c->kobj, &bch_cache_set_ktype); 1782 kobject_init(&c->internal, &bch_cache_set_internal_ktype); 1783 1784 bch_cache_accounting_init(&c->accounting, &c->cl); 1785 1786 memcpy(c->sb.set_uuid, sb->set_uuid, 16); 1787 c->sb.block_size = sb->block_size; 1788 c->sb.bucket_size = sb->bucket_size; 1789 c->sb.nr_in_set = sb->nr_in_set; 1790 c->sb.last_mount = sb->last_mount; 1791 c->bucket_bits = ilog2(sb->bucket_size); 1792 c->block_bits = ilog2(sb->block_size); 1793 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry); 1794 c->devices_max_used = 0; 1795 atomic_set(&c->attached_dev_nr, 0); 1796 c->btree_pages = bucket_pages(c); 1797 if (c->btree_pages > BTREE_MAX_PAGES) 1798 c->btree_pages = max_t(int, c->btree_pages / 4, 1799 BTREE_MAX_PAGES); 1800 1801 sema_init(&c->sb_write_mutex, 1); 1802 mutex_init(&c->bucket_lock); 1803 init_waitqueue_head(&c->btree_cache_wait); 1804 spin_lock_init(&c->btree_cannibalize_lock); 1805 init_waitqueue_head(&c->bucket_wait); 1806 init_waitqueue_head(&c->gc_wait); 1807 sema_init(&c->uuid_write_mutex, 1); 1808 1809 spin_lock_init(&c->btree_gc_time.lock); 1810 spin_lock_init(&c->btree_split_time.lock); 1811 spin_lock_init(&c->btree_read_time.lock); 1812 1813 bch_moving_init_cache_set(c); 1814 1815 INIT_LIST_HEAD(&c->list); 1816 INIT_LIST_HEAD(&c->cached_devs); 1817 INIT_LIST_HEAD(&c->btree_cache); 1818 INIT_LIST_HEAD(&c->btree_cache_freeable); 1819 INIT_LIST_HEAD(&c->btree_cache_freed); 1820 INIT_LIST_HEAD(&c->data_buckets); 1821 1822 iter_size = (sb->bucket_size / sb->block_size + 1) * 1823 sizeof(struct btree_iter_set); 1824 1825 if (!(c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL)) || 1826 mempool_init_slab_pool(&c->search, 32, bch_search_cache) || 1827 mempool_init_kmalloc_pool(&c->bio_meta, 2, 1828 sizeof(struct bbio) + sizeof(struct bio_vec) * 1829 bucket_pages(c)) || 1830 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || 1831 bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio), 1832 BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER) || 1833 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) || 1834 !(c->moving_gc_wq = alloc_workqueue("bcache_gc", 1835 WQ_MEM_RECLAIM, 0)) || 1836 bch_journal_alloc(c) || 1837 bch_btree_cache_alloc(c) || 1838 bch_open_buckets_alloc(c) || 1839 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages))) 1840 goto err; 1841 1842 c->congested_read_threshold_us = 2000; 1843 c->congested_write_threshold_us = 20000; 1844 c->error_limit = DEFAULT_IO_ERROR_LIMIT; 1845 c->idle_max_writeback_rate_enabled = 1; 1846 WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags)); 1847 1848 return c; 1849 err: 1850 bch_cache_set_unregister(c); 1851 return NULL; 1852 } 1853 1854 static int run_cache_set(struct cache_set *c) 1855 { 1856 const char *err = "cannot allocate memory"; 1857 struct cached_dev *dc, *t; 1858 struct cache *ca; 1859 struct closure cl; 1860 unsigned int i; 1861 LIST_HEAD(journal); 1862 struct journal_replay *l; 1863 1864 closure_init_stack(&cl); 1865 1866 for_each_cache(ca, c, i) 1867 c->nbuckets += ca->sb.nbuckets; 1868 set_gc_sectors(c); 1869 1870 if (CACHE_SYNC(&c->sb)) { 1871 struct bkey *k; 1872 struct jset *j; 1873 1874 err = "cannot allocate memory for journal"; 1875 if (bch_journal_read(c, &journal)) 1876 goto err; 1877 1878 pr_debug("btree_journal_read() done\n"); 1879 1880 err = "no journal entries found"; 1881 if (list_empty(&journal)) 1882 goto err; 1883 1884 j = &list_entry(journal.prev, struct journal_replay, list)->j; 1885 1886 err = "IO error reading priorities"; 1887 for_each_cache(ca, c, i) { 1888 if (prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev])) 1889 goto err; 1890 } 1891 1892 /* 1893 * If prio_read() fails it'll call cache_set_error and we'll 1894 * tear everything down right away, but if we perhaps checked 1895 * sooner we could avoid journal replay. 1896 */ 1897 1898 k = &j->btree_root; 1899 1900 err = "bad btree root"; 1901 if (__bch_btree_ptr_invalid(c, k)) 1902 goto err; 1903 1904 err = "error reading btree root"; 1905 c->root = bch_btree_node_get(c, NULL, k, 1906 j->btree_level, 1907 true, NULL); 1908 if (IS_ERR_OR_NULL(c->root)) 1909 goto err; 1910 1911 list_del_init(&c->root->list); 1912 rw_unlock(true, c->root); 1913 1914 err = uuid_read(c, j, &cl); 1915 if (err) 1916 goto err; 1917 1918 err = "error in recovery"; 1919 if (bch_btree_check(c)) 1920 goto err; 1921 1922 bch_journal_mark(c, &journal); 1923 bch_initial_gc_finish(c); 1924 pr_debug("btree_check() done\n"); 1925 1926 /* 1927 * bcache_journal_next() can't happen sooner, or 1928 * btree_gc_finish() will give spurious errors about last_gc > 1929 * gc_gen - this is a hack but oh well. 1930 */ 1931 bch_journal_next(&c->journal); 1932 1933 err = "error starting allocator thread"; 1934 for_each_cache(ca, c, i) 1935 if (bch_cache_allocator_start(ca)) 1936 goto err; 1937 1938 /* 1939 * First place it's safe to allocate: btree_check() and 1940 * btree_gc_finish() have to run before we have buckets to 1941 * allocate, and bch_bucket_alloc_set() might cause a journal 1942 * entry to be written so bcache_journal_next() has to be called 1943 * first. 1944 * 1945 * If the uuids were in the old format we have to rewrite them 1946 * before the next journal entry is written: 1947 */ 1948 if (j->version < BCACHE_JSET_VERSION_UUID) 1949 __uuid_write(c); 1950 1951 err = "bcache: replay journal failed"; 1952 if (bch_journal_replay(c, &journal)) 1953 goto err; 1954 } else { 1955 pr_notice("invalidating existing data\n"); 1956 1957 for_each_cache(ca, c, i) { 1958 unsigned int j; 1959 1960 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7, 1961 2, SB_JOURNAL_BUCKETS); 1962 1963 for (j = 0; j < ca->sb.keys; j++) 1964 ca->sb.d[j] = ca->sb.first_bucket + j; 1965 } 1966 1967 bch_initial_gc_finish(c); 1968 1969 err = "error starting allocator thread"; 1970 for_each_cache(ca, c, i) 1971 if (bch_cache_allocator_start(ca)) 1972 goto err; 1973 1974 mutex_lock(&c->bucket_lock); 1975 for_each_cache(ca, c, i) 1976 bch_prio_write(ca, true); 1977 mutex_unlock(&c->bucket_lock); 1978 1979 err = "cannot allocate new UUID bucket"; 1980 if (__uuid_write(c)) 1981 goto err; 1982 1983 err = "cannot allocate new btree root"; 1984 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL); 1985 if (IS_ERR_OR_NULL(c->root)) 1986 goto err; 1987 1988 mutex_lock(&c->root->write_lock); 1989 bkey_copy_key(&c->root->key, &MAX_KEY); 1990 bch_btree_node_write(c->root, &cl); 1991 mutex_unlock(&c->root->write_lock); 1992 1993 bch_btree_set_root(c->root); 1994 rw_unlock(true, c->root); 1995 1996 /* 1997 * We don't want to write the first journal entry until 1998 * everything is set up - fortunately journal entries won't be 1999 * written until the SET_CACHE_SYNC() here: 2000 */ 2001 SET_CACHE_SYNC(&c->sb, true); 2002 2003 bch_journal_next(&c->journal); 2004 bch_journal_meta(c, &cl); 2005 } 2006 2007 err = "error starting gc thread"; 2008 if (bch_gc_thread_start(c)) 2009 goto err; 2010 2011 closure_sync(&cl); 2012 c->sb.last_mount = (u32)ktime_get_real_seconds(); 2013 bcache_write_super(c); 2014 2015 list_for_each_entry_safe(dc, t, &uncached_devices, list) 2016 bch_cached_dev_attach(dc, c, NULL); 2017 2018 flash_devs_run(c); 2019 2020 set_bit(CACHE_SET_RUNNING, &c->flags); 2021 return 0; 2022 err: 2023 while (!list_empty(&journal)) { 2024 l = list_first_entry(&journal, struct journal_replay, list); 2025 list_del(&l->list); 2026 kfree(l); 2027 } 2028 2029 closure_sync(&cl); 2030 2031 bch_cache_set_error(c, "%s", err); 2032 2033 return -EIO; 2034 } 2035 2036 static bool can_attach_cache(struct cache *ca, struct cache_set *c) 2037 { 2038 return ca->sb.block_size == c->sb.block_size && 2039 ca->sb.bucket_size == c->sb.bucket_size && 2040 ca->sb.nr_in_set == c->sb.nr_in_set; 2041 } 2042 2043 static const char *register_cache_set(struct cache *ca) 2044 { 2045 char buf[12]; 2046 const char *err = "cannot allocate memory"; 2047 struct cache_set *c; 2048 2049 list_for_each_entry(c, &bch_cache_sets, list) 2050 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) { 2051 if (c->cache[ca->sb.nr_this_dev]) 2052 return "duplicate cache set member"; 2053 2054 if (!can_attach_cache(ca, c)) 2055 return "cache sb does not match set"; 2056 2057 if (!CACHE_SYNC(&ca->sb)) 2058 SET_CACHE_SYNC(&c->sb, false); 2059 2060 goto found; 2061 } 2062 2063 c = bch_cache_set_alloc(&ca->sb); 2064 if (!c) 2065 return err; 2066 2067 err = "error creating kobject"; 2068 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) || 2069 kobject_add(&c->internal, &c->kobj, "internal")) 2070 goto err; 2071 2072 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj)) 2073 goto err; 2074 2075 bch_debug_init_cache_set(c); 2076 2077 list_add(&c->list, &bch_cache_sets); 2078 found: 2079 sprintf(buf, "cache%i", ca->sb.nr_this_dev); 2080 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") || 2081 sysfs_create_link(&c->kobj, &ca->kobj, buf)) 2082 goto err; 2083 2084 if (ca->sb.seq > c->sb.seq) { 2085 c->sb.version = ca->sb.version; 2086 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16); 2087 c->sb.flags = ca->sb.flags; 2088 c->sb.seq = ca->sb.seq; 2089 pr_debug("set version = %llu\n", c->sb.version); 2090 } 2091 2092 kobject_get(&ca->kobj); 2093 ca->set = c; 2094 ca->set->cache[ca->sb.nr_this_dev] = ca; 2095 c->cache_by_alloc[c->caches_loaded++] = ca; 2096 2097 if (c->caches_loaded == c->sb.nr_in_set) { 2098 err = "failed to run cache set"; 2099 if (run_cache_set(c) < 0) 2100 goto err; 2101 } 2102 2103 return NULL; 2104 err: 2105 bch_cache_set_unregister(c); 2106 return err; 2107 } 2108 2109 /* Cache device */ 2110 2111 /* When ca->kobj released */ 2112 void bch_cache_release(struct kobject *kobj) 2113 { 2114 struct cache *ca = container_of(kobj, struct cache, kobj); 2115 unsigned int i; 2116 2117 if (ca->set) { 2118 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca); 2119 ca->set->cache[ca->sb.nr_this_dev] = NULL; 2120 } 2121 2122 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca))); 2123 kfree(ca->prio_buckets); 2124 vfree(ca->buckets); 2125 2126 free_heap(&ca->heap); 2127 free_fifo(&ca->free_inc); 2128 2129 for (i = 0; i < RESERVE_NR; i++) 2130 free_fifo(&ca->free[i]); 2131 2132 if (ca->sb_disk) 2133 put_page(virt_to_page(ca->sb_disk)); 2134 2135 if (!IS_ERR_OR_NULL(ca->bdev)) 2136 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2137 2138 kfree(ca); 2139 module_put(THIS_MODULE); 2140 } 2141 2142 static int cache_alloc(struct cache *ca) 2143 { 2144 size_t free; 2145 size_t btree_buckets; 2146 struct bucket *b; 2147 int ret = -ENOMEM; 2148 const char *err = NULL; 2149 2150 __module_get(THIS_MODULE); 2151 kobject_init(&ca->kobj, &bch_cache_ktype); 2152 2153 bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8); 2154 2155 /* 2156 * when ca->sb.njournal_buckets is not zero, journal exists, 2157 * and in bch_journal_replay(), tree node may split, 2158 * so bucket of RESERVE_BTREE type is needed, 2159 * the worst situation is all journal buckets are valid journal, 2160 * and all the keys need to replay, 2161 * so the number of RESERVE_BTREE type buckets should be as much 2162 * as journal buckets 2163 */ 2164 btree_buckets = ca->sb.njournal_buckets ?: 8; 2165 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10; 2166 if (!free) { 2167 ret = -EPERM; 2168 err = "ca->sb.nbuckets is too small"; 2169 goto err_free; 2170 } 2171 2172 if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, 2173 GFP_KERNEL)) { 2174 err = "ca->free[RESERVE_BTREE] alloc failed"; 2175 goto err_btree_alloc; 2176 } 2177 2178 if (!init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), 2179 GFP_KERNEL)) { 2180 err = "ca->free[RESERVE_PRIO] alloc failed"; 2181 goto err_prio_alloc; 2182 } 2183 2184 if (!init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL)) { 2185 err = "ca->free[RESERVE_MOVINGGC] alloc failed"; 2186 goto err_movinggc_alloc; 2187 } 2188 2189 if (!init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL)) { 2190 err = "ca->free[RESERVE_NONE] alloc failed"; 2191 goto err_none_alloc; 2192 } 2193 2194 if (!init_fifo(&ca->free_inc, free << 2, GFP_KERNEL)) { 2195 err = "ca->free_inc alloc failed"; 2196 goto err_free_inc_alloc; 2197 } 2198 2199 if (!init_heap(&ca->heap, free << 3, GFP_KERNEL)) { 2200 err = "ca->heap alloc failed"; 2201 goto err_heap_alloc; 2202 } 2203 2204 ca->buckets = vzalloc(array_size(sizeof(struct bucket), 2205 ca->sb.nbuckets)); 2206 if (!ca->buckets) { 2207 err = "ca->buckets alloc failed"; 2208 goto err_buckets_alloc; 2209 } 2210 2211 ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t), 2212 prio_buckets(ca), 2), 2213 GFP_KERNEL); 2214 if (!ca->prio_buckets) { 2215 err = "ca->prio_buckets alloc failed"; 2216 goto err_prio_buckets_alloc; 2217 } 2218 2219 ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca); 2220 if (!ca->disk_buckets) { 2221 err = "ca->disk_buckets alloc failed"; 2222 goto err_disk_buckets_alloc; 2223 } 2224 2225 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca); 2226 2227 for_each_bucket(b, ca) 2228 atomic_set(&b->pin, 0); 2229 return 0; 2230 2231 err_disk_buckets_alloc: 2232 kfree(ca->prio_buckets); 2233 err_prio_buckets_alloc: 2234 vfree(ca->buckets); 2235 err_buckets_alloc: 2236 free_heap(&ca->heap); 2237 err_heap_alloc: 2238 free_fifo(&ca->free_inc); 2239 err_free_inc_alloc: 2240 free_fifo(&ca->free[RESERVE_NONE]); 2241 err_none_alloc: 2242 free_fifo(&ca->free[RESERVE_MOVINGGC]); 2243 err_movinggc_alloc: 2244 free_fifo(&ca->free[RESERVE_PRIO]); 2245 err_prio_alloc: 2246 free_fifo(&ca->free[RESERVE_BTREE]); 2247 err_btree_alloc: 2248 err_free: 2249 module_put(THIS_MODULE); 2250 if (err) 2251 pr_notice("error %s: %s\n", ca->cache_dev_name, err); 2252 return ret; 2253 } 2254 2255 static int register_cache(struct cache_sb *sb, struct cache_sb_disk *sb_disk, 2256 struct block_device *bdev, struct cache *ca) 2257 { 2258 const char *err = NULL; /* must be set for any error case */ 2259 int ret = 0; 2260 2261 bdevname(bdev, ca->cache_dev_name); 2262 memcpy(&ca->sb, sb, sizeof(struct cache_sb)); 2263 ca->bdev = bdev; 2264 ca->bdev->bd_holder = ca; 2265 ca->sb_disk = sb_disk; 2266 2267 if (blk_queue_discard(bdev_get_queue(bdev))) 2268 ca->discard = CACHE_DISCARD(&ca->sb); 2269 2270 ret = cache_alloc(ca); 2271 if (ret != 0) { 2272 /* 2273 * If we failed here, it means ca->kobj is not initialized yet, 2274 * kobject_put() won't be called and there is no chance to 2275 * call blkdev_put() to bdev in bch_cache_release(). So we 2276 * explicitly call blkdev_put() here. 2277 */ 2278 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 2279 if (ret == -ENOMEM) 2280 err = "cache_alloc(): -ENOMEM"; 2281 else if (ret == -EPERM) 2282 err = "cache_alloc(): cache device is too small"; 2283 else 2284 err = "cache_alloc(): unknown error"; 2285 goto err; 2286 } 2287 2288 if (kobject_add(&ca->kobj, 2289 &part_to_dev(bdev->bd_part)->kobj, 2290 "bcache")) { 2291 err = "error calling kobject_add"; 2292 ret = -ENOMEM; 2293 goto out; 2294 } 2295 2296 mutex_lock(&bch_register_lock); 2297 err = register_cache_set(ca); 2298 mutex_unlock(&bch_register_lock); 2299 2300 if (err) { 2301 ret = -ENODEV; 2302 goto out; 2303 } 2304 2305 pr_info("registered cache device %s\n", ca->cache_dev_name); 2306 2307 out: 2308 kobject_put(&ca->kobj); 2309 2310 err: 2311 if (err) 2312 pr_notice("error %s: %s\n", ca->cache_dev_name, err); 2313 2314 return ret; 2315 } 2316 2317 /* Global interfaces/init */ 2318 2319 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, 2320 const char *buffer, size_t size); 2321 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k, 2322 struct kobj_attribute *attr, 2323 const char *buffer, size_t size); 2324 2325 kobj_attribute_write(register, register_bcache); 2326 kobj_attribute_write(register_quiet, register_bcache); 2327 kobj_attribute_write(pendings_cleanup, bch_pending_bdevs_cleanup); 2328 2329 static bool bch_is_open_backing(struct block_device *bdev) 2330 { 2331 struct cache_set *c, *tc; 2332 struct cached_dev *dc, *t; 2333 2334 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2335 list_for_each_entry_safe(dc, t, &c->cached_devs, list) 2336 if (dc->bdev == bdev) 2337 return true; 2338 list_for_each_entry_safe(dc, t, &uncached_devices, list) 2339 if (dc->bdev == bdev) 2340 return true; 2341 return false; 2342 } 2343 2344 static bool bch_is_open_cache(struct block_device *bdev) 2345 { 2346 struct cache_set *c, *tc; 2347 struct cache *ca; 2348 unsigned int i; 2349 2350 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2351 for_each_cache(ca, c, i) 2352 if (ca->bdev == bdev) 2353 return true; 2354 return false; 2355 } 2356 2357 static bool bch_is_open(struct block_device *bdev) 2358 { 2359 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev); 2360 } 2361 2362 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, 2363 const char *buffer, size_t size) 2364 { 2365 const char *err; 2366 char *path = NULL; 2367 struct cache_sb *sb; 2368 struct cache_sb_disk *sb_disk; 2369 struct block_device *bdev; 2370 ssize_t ret; 2371 2372 ret = -EBUSY; 2373 err = "failed to reference bcache module"; 2374 if (!try_module_get(THIS_MODULE)) 2375 goto out; 2376 2377 /* For latest state of bcache_is_reboot */ 2378 smp_mb(); 2379 err = "bcache is in reboot"; 2380 if (bcache_is_reboot) 2381 goto out_module_put; 2382 2383 ret = -ENOMEM; 2384 err = "cannot allocate memory"; 2385 path = kstrndup(buffer, size, GFP_KERNEL); 2386 if (!path) 2387 goto out_module_put; 2388 2389 sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL); 2390 if (!sb) 2391 goto out_free_path; 2392 2393 ret = -EINVAL; 2394 err = "failed to open device"; 2395 bdev = blkdev_get_by_path(strim(path), 2396 FMODE_READ|FMODE_WRITE|FMODE_EXCL, 2397 sb); 2398 if (IS_ERR(bdev)) { 2399 if (bdev == ERR_PTR(-EBUSY)) { 2400 bdev = lookup_bdev(strim(path)); 2401 mutex_lock(&bch_register_lock); 2402 if (!IS_ERR(bdev) && bch_is_open(bdev)) 2403 err = "device already registered"; 2404 else 2405 err = "device busy"; 2406 mutex_unlock(&bch_register_lock); 2407 if (!IS_ERR(bdev)) 2408 bdput(bdev); 2409 if (attr == &ksysfs_register_quiet) 2410 goto done; 2411 } 2412 goto out_free_sb; 2413 } 2414 2415 err = "failed to set blocksize"; 2416 if (set_blocksize(bdev, 4096)) 2417 goto out_blkdev_put; 2418 2419 err = read_super(sb, bdev, &sb_disk); 2420 if (err) 2421 goto out_blkdev_put; 2422 2423 err = "failed to register device"; 2424 if (SB_IS_BDEV(sb)) { 2425 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL); 2426 2427 if (!dc) 2428 goto out_put_sb_page; 2429 2430 mutex_lock(&bch_register_lock); 2431 ret = register_bdev(sb, sb_disk, bdev, dc); 2432 mutex_unlock(&bch_register_lock); 2433 /* blkdev_put() will be called in cached_dev_free() */ 2434 if (ret < 0) 2435 goto out_free_sb; 2436 } else { 2437 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL); 2438 2439 if (!ca) 2440 goto out_put_sb_page; 2441 2442 /* blkdev_put() will be called in bch_cache_release() */ 2443 if (register_cache(sb, sb_disk, bdev, ca) != 0) 2444 goto out_free_sb; 2445 } 2446 2447 done: 2448 kfree(sb); 2449 kfree(path); 2450 module_put(THIS_MODULE); 2451 return size; 2452 2453 out_put_sb_page: 2454 put_page(virt_to_page(sb_disk)); 2455 out_blkdev_put: 2456 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); 2457 out_free_sb: 2458 kfree(sb); 2459 out_free_path: 2460 kfree(path); 2461 path = NULL; 2462 out_module_put: 2463 module_put(THIS_MODULE); 2464 out: 2465 pr_info("error %s: %s\n", path?path:"", err); 2466 return ret; 2467 } 2468 2469 2470 struct pdev { 2471 struct list_head list; 2472 struct cached_dev *dc; 2473 }; 2474 2475 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k, 2476 struct kobj_attribute *attr, 2477 const char *buffer, 2478 size_t size) 2479 { 2480 LIST_HEAD(pending_devs); 2481 ssize_t ret = size; 2482 struct cached_dev *dc, *tdc; 2483 struct pdev *pdev, *tpdev; 2484 struct cache_set *c, *tc; 2485 2486 mutex_lock(&bch_register_lock); 2487 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) { 2488 pdev = kmalloc(sizeof(struct pdev), GFP_KERNEL); 2489 if (!pdev) 2490 break; 2491 pdev->dc = dc; 2492 list_add(&pdev->list, &pending_devs); 2493 } 2494 2495 list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) { 2496 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) { 2497 char *pdev_set_uuid = pdev->dc->sb.set_uuid; 2498 char *set_uuid = c->sb.uuid; 2499 2500 if (!memcmp(pdev_set_uuid, set_uuid, 16)) { 2501 list_del(&pdev->list); 2502 kfree(pdev); 2503 break; 2504 } 2505 } 2506 } 2507 mutex_unlock(&bch_register_lock); 2508 2509 list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) { 2510 pr_info("delete pdev %p\n", pdev); 2511 list_del(&pdev->list); 2512 bcache_device_stop(&pdev->dc->disk); 2513 kfree(pdev); 2514 } 2515 2516 return ret; 2517 } 2518 2519 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x) 2520 { 2521 if (bcache_is_reboot) 2522 return NOTIFY_DONE; 2523 2524 if (code == SYS_DOWN || 2525 code == SYS_HALT || 2526 code == SYS_POWER_OFF) { 2527 DEFINE_WAIT(wait); 2528 unsigned long start = jiffies; 2529 bool stopped = false; 2530 2531 struct cache_set *c, *tc; 2532 struct cached_dev *dc, *tdc; 2533 2534 mutex_lock(&bch_register_lock); 2535 2536 if (bcache_is_reboot) 2537 goto out; 2538 2539 /* New registration is rejected since now */ 2540 bcache_is_reboot = true; 2541 /* 2542 * Make registering caller (if there is) on other CPU 2543 * core know bcache_is_reboot set to true earlier 2544 */ 2545 smp_mb(); 2546 2547 if (list_empty(&bch_cache_sets) && 2548 list_empty(&uncached_devices)) 2549 goto out; 2550 2551 mutex_unlock(&bch_register_lock); 2552 2553 pr_info("Stopping all devices:\n"); 2554 2555 /* 2556 * The reason bch_register_lock is not held to call 2557 * bch_cache_set_stop() and bcache_device_stop() is to 2558 * avoid potential deadlock during reboot, because cache 2559 * set or bcache device stopping process will acqurie 2560 * bch_register_lock too. 2561 * 2562 * We are safe here because bcache_is_reboot sets to 2563 * true already, register_bcache() will reject new 2564 * registration now. bcache_is_reboot also makes sure 2565 * bcache_reboot() won't be re-entered on by other thread, 2566 * so there is no race in following list iteration by 2567 * list_for_each_entry_safe(). 2568 */ 2569 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) 2570 bch_cache_set_stop(c); 2571 2572 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) 2573 bcache_device_stop(&dc->disk); 2574 2575 2576 /* 2577 * Give an early chance for other kthreads and 2578 * kworkers to stop themselves 2579 */ 2580 schedule(); 2581 2582 /* What's a condition variable? */ 2583 while (1) { 2584 long timeout = start + 10 * HZ - jiffies; 2585 2586 mutex_lock(&bch_register_lock); 2587 stopped = list_empty(&bch_cache_sets) && 2588 list_empty(&uncached_devices); 2589 2590 if (timeout < 0 || stopped) 2591 break; 2592 2593 prepare_to_wait(&unregister_wait, &wait, 2594 TASK_UNINTERRUPTIBLE); 2595 2596 mutex_unlock(&bch_register_lock); 2597 schedule_timeout(timeout); 2598 } 2599 2600 finish_wait(&unregister_wait, &wait); 2601 2602 if (stopped) 2603 pr_info("All devices stopped\n"); 2604 else 2605 pr_notice("Timeout waiting for devices to be closed\n"); 2606 out: 2607 mutex_unlock(&bch_register_lock); 2608 } 2609 2610 return NOTIFY_DONE; 2611 } 2612 2613 static struct notifier_block reboot = { 2614 .notifier_call = bcache_reboot, 2615 .priority = INT_MAX, /* before any real devices */ 2616 }; 2617 2618 static void bcache_exit(void) 2619 { 2620 bch_debug_exit(); 2621 bch_request_exit(); 2622 if (bcache_kobj) 2623 kobject_put(bcache_kobj); 2624 if (bcache_wq) 2625 destroy_workqueue(bcache_wq); 2626 if (bch_journal_wq) 2627 destroy_workqueue(bch_journal_wq); 2628 2629 if (bcache_major) 2630 unregister_blkdev(bcache_major, "bcache"); 2631 unregister_reboot_notifier(&reboot); 2632 mutex_destroy(&bch_register_lock); 2633 } 2634 2635 /* Check and fixup module parameters */ 2636 static void check_module_parameters(void) 2637 { 2638 if (bch_cutoff_writeback_sync == 0) 2639 bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC; 2640 else if (bch_cutoff_writeback_sync > CUTOFF_WRITEBACK_SYNC_MAX) { 2641 pr_warn("set bch_cutoff_writeback_sync (%u) to max value %u\n", 2642 bch_cutoff_writeback_sync, CUTOFF_WRITEBACK_SYNC_MAX); 2643 bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC_MAX; 2644 } 2645 2646 if (bch_cutoff_writeback == 0) 2647 bch_cutoff_writeback = CUTOFF_WRITEBACK; 2648 else if (bch_cutoff_writeback > CUTOFF_WRITEBACK_MAX) { 2649 pr_warn("set bch_cutoff_writeback (%u) to max value %u\n", 2650 bch_cutoff_writeback, CUTOFF_WRITEBACK_MAX); 2651 bch_cutoff_writeback = CUTOFF_WRITEBACK_MAX; 2652 } 2653 2654 if (bch_cutoff_writeback > bch_cutoff_writeback_sync) { 2655 pr_warn("set bch_cutoff_writeback (%u) to %u\n", 2656 bch_cutoff_writeback, bch_cutoff_writeback_sync); 2657 bch_cutoff_writeback = bch_cutoff_writeback_sync; 2658 } 2659 } 2660 2661 static int __init bcache_init(void) 2662 { 2663 static const struct attribute *files[] = { 2664 &ksysfs_register.attr, 2665 &ksysfs_register_quiet.attr, 2666 &ksysfs_pendings_cleanup.attr, 2667 NULL 2668 }; 2669 2670 check_module_parameters(); 2671 2672 mutex_init(&bch_register_lock); 2673 init_waitqueue_head(&unregister_wait); 2674 register_reboot_notifier(&reboot); 2675 2676 bcache_major = register_blkdev(0, "bcache"); 2677 if (bcache_major < 0) { 2678 unregister_reboot_notifier(&reboot); 2679 mutex_destroy(&bch_register_lock); 2680 return bcache_major; 2681 } 2682 2683 bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0); 2684 if (!bcache_wq) 2685 goto err; 2686 2687 bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0); 2688 if (!bch_journal_wq) 2689 goto err; 2690 2691 bcache_kobj = kobject_create_and_add("bcache", fs_kobj); 2692 if (!bcache_kobj) 2693 goto err; 2694 2695 if (bch_request_init() || 2696 sysfs_create_files(bcache_kobj, files)) 2697 goto err; 2698 2699 bch_debug_init(); 2700 closure_debug_init(); 2701 2702 bcache_is_reboot = false; 2703 2704 return 0; 2705 err: 2706 bcache_exit(); 2707 return -ENOMEM; 2708 } 2709 2710 /* 2711 * Module hooks 2712 */ 2713 module_exit(bcache_exit); 2714 module_init(bcache_init); 2715 2716 module_param(bch_cutoff_writeback, uint, 0); 2717 MODULE_PARM_DESC(bch_cutoff_writeback, "threshold to cutoff writeback"); 2718 2719 module_param(bch_cutoff_writeback_sync, uint, 0); 2720 MODULE_PARM_DESC(bch_cutoff_writeback_sync, "hard threshold to cutoff writeback"); 2721 2722 MODULE_DESCRIPTION("Bcache: a Linux block layer cache"); 2723 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); 2724 MODULE_LICENSE("GPL"); 2725