1 /* 2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. 3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. 4 * 5 * This file is released under the GPL. 6 */ 7 8 #include "dm-core.h" 9 #include "dm-rq.h" 10 #include "dm-uevent.h" 11 12 #include <linux/init.h> 13 #include <linux/module.h> 14 #include <linux/mutex.h> 15 #include <linux/sched/mm.h> 16 #include <linux/sched/signal.h> 17 #include <linux/blkpg.h> 18 #include <linux/bio.h> 19 #include <linux/mempool.h> 20 #include <linux/dax.h> 21 #include <linux/slab.h> 22 #include <linux/idr.h> 23 #include <linux/uio.h> 24 #include <linux/hdreg.h> 25 #include <linux/delay.h> 26 #include <linux/wait.h> 27 #include <linux/pr.h> 28 #include <linux/refcount.h> 29 #include <linux/part_stat.h> 30 #include <linux/blk-crypto.h> 31 32 #define DM_MSG_PREFIX "core" 33 34 /* 35 * Cookies are numeric values sent with CHANGE and REMOVE 36 * uevents while resuming, removing or renaming the device. 37 */ 38 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" 39 #define DM_COOKIE_LENGTH 24 40 41 static const char *_name = DM_NAME; 42 43 static unsigned int major = 0; 44 static unsigned int _major = 0; 45 46 static DEFINE_IDR(_minor_idr); 47 48 static DEFINE_SPINLOCK(_minor_lock); 49 50 static void do_deferred_remove(struct work_struct *w); 51 52 static DECLARE_WORK(deferred_remove_work, do_deferred_remove); 53 54 static struct workqueue_struct *deferred_remove_workqueue; 55 56 atomic_t dm_global_event_nr = ATOMIC_INIT(0); 57 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq); 58 59 void dm_issue_global_event(void) 60 { 61 atomic_inc(&dm_global_event_nr); 62 wake_up(&dm_global_eventq); 63 } 64 65 /* 66 * One of these is allocated (on-stack) per original bio. 67 */ 68 struct clone_info { 69 struct dm_table *map; 70 struct bio *bio; 71 struct dm_io *io; 72 sector_t sector; 73 unsigned sector_count; 74 }; 75 76 /* 77 * One of these is allocated per clone bio. 78 */ 79 #define DM_TIO_MAGIC 7282014 80 struct dm_target_io { 81 unsigned magic; 82 struct dm_io *io; 83 struct dm_target *ti; 84 unsigned target_bio_nr; 85 unsigned *len_ptr; 86 bool inside_dm_io; 87 struct bio clone; 88 }; 89 90 /* 91 * One of these is allocated per original bio. 92 * It contains the first clone used for that original. 93 */ 94 #define DM_IO_MAGIC 5191977 95 struct dm_io { 96 unsigned magic; 97 struct mapped_device *md; 98 blk_status_t status; 99 atomic_t io_count; 100 struct bio *orig_bio; 101 unsigned long start_time; 102 spinlock_t endio_lock; 103 struct dm_stats_aux stats_aux; 104 /* last member of dm_target_io is 'struct bio' */ 105 struct dm_target_io tio; 106 }; 107 108 void *dm_per_bio_data(struct bio *bio, size_t data_size) 109 { 110 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); 111 if (!tio->inside_dm_io) 112 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size; 113 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size; 114 } 115 EXPORT_SYMBOL_GPL(dm_per_bio_data); 116 117 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size) 118 { 119 struct dm_io *io = (struct dm_io *)((char *)data + data_size); 120 if (io->magic == DM_IO_MAGIC) 121 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone)); 122 BUG_ON(io->magic != DM_TIO_MAGIC); 123 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone)); 124 } 125 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data); 126 127 unsigned dm_bio_get_target_bio_nr(const struct bio *bio) 128 { 129 return container_of(bio, struct dm_target_io, clone)->target_bio_nr; 130 } 131 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr); 132 133 #define MINOR_ALLOCED ((void *)-1) 134 135 /* 136 * Bits for the md->flags field. 137 */ 138 #define DMF_BLOCK_IO_FOR_SUSPEND 0 139 #define DMF_SUSPENDED 1 140 #define DMF_FROZEN 2 141 #define DMF_FREEING 3 142 #define DMF_DELETING 4 143 #define DMF_NOFLUSH_SUSPENDING 5 144 #define DMF_DEFERRED_REMOVE 6 145 #define DMF_SUSPENDED_INTERNALLY 7 146 #define DMF_POST_SUSPENDING 8 147 148 #define DM_NUMA_NODE NUMA_NO_NODE 149 static int dm_numa_node = DM_NUMA_NODE; 150 151 /* 152 * For mempools pre-allocation at the table loading time. 153 */ 154 struct dm_md_mempools { 155 struct bio_set bs; 156 struct bio_set io_bs; 157 }; 158 159 struct table_device { 160 struct list_head list; 161 refcount_t count; 162 struct dm_dev dm_dev; 163 }; 164 165 /* 166 * Bio-based DM's mempools' reserved IOs set by the user. 167 */ 168 #define RESERVED_BIO_BASED_IOS 16 169 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; 170 171 static int __dm_get_module_param_int(int *module_param, int min, int max) 172 { 173 int param = READ_ONCE(*module_param); 174 int modified_param = 0; 175 bool modified = true; 176 177 if (param < min) 178 modified_param = min; 179 else if (param > max) 180 modified_param = max; 181 else 182 modified = false; 183 184 if (modified) { 185 (void)cmpxchg(module_param, param, modified_param); 186 param = modified_param; 187 } 188 189 return param; 190 } 191 192 unsigned __dm_get_module_param(unsigned *module_param, 193 unsigned def, unsigned max) 194 { 195 unsigned param = READ_ONCE(*module_param); 196 unsigned modified_param = 0; 197 198 if (!param) 199 modified_param = def; 200 else if (param > max) 201 modified_param = max; 202 203 if (modified_param) { 204 (void)cmpxchg(module_param, param, modified_param); 205 param = modified_param; 206 } 207 208 return param; 209 } 210 211 unsigned dm_get_reserved_bio_based_ios(void) 212 { 213 return __dm_get_module_param(&reserved_bio_based_ios, 214 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); 215 } 216 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); 217 218 static unsigned dm_get_numa_node(void) 219 { 220 return __dm_get_module_param_int(&dm_numa_node, 221 DM_NUMA_NODE, num_online_nodes() - 1); 222 } 223 224 static int __init local_init(void) 225 { 226 int r; 227 228 r = dm_uevent_init(); 229 if (r) 230 return r; 231 232 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); 233 if (!deferred_remove_workqueue) { 234 r = -ENOMEM; 235 goto out_uevent_exit; 236 } 237 238 _major = major; 239 r = register_blkdev(_major, _name); 240 if (r < 0) 241 goto out_free_workqueue; 242 243 if (!_major) 244 _major = r; 245 246 return 0; 247 248 out_free_workqueue: 249 destroy_workqueue(deferred_remove_workqueue); 250 out_uevent_exit: 251 dm_uevent_exit(); 252 253 return r; 254 } 255 256 static void local_exit(void) 257 { 258 flush_scheduled_work(); 259 destroy_workqueue(deferred_remove_workqueue); 260 261 unregister_blkdev(_major, _name); 262 dm_uevent_exit(); 263 264 _major = 0; 265 266 DMINFO("cleaned up"); 267 } 268 269 static int (*_inits[])(void) __initdata = { 270 local_init, 271 dm_target_init, 272 dm_linear_init, 273 dm_stripe_init, 274 dm_io_init, 275 dm_kcopyd_init, 276 dm_interface_init, 277 dm_statistics_init, 278 }; 279 280 static void (*_exits[])(void) = { 281 local_exit, 282 dm_target_exit, 283 dm_linear_exit, 284 dm_stripe_exit, 285 dm_io_exit, 286 dm_kcopyd_exit, 287 dm_interface_exit, 288 dm_statistics_exit, 289 }; 290 291 static int __init dm_init(void) 292 { 293 const int count = ARRAY_SIZE(_inits); 294 295 int r, i; 296 297 for (i = 0; i < count; i++) { 298 r = _inits[i](); 299 if (r) 300 goto bad; 301 } 302 303 return 0; 304 305 bad: 306 while (i--) 307 _exits[i](); 308 309 return r; 310 } 311 312 static void __exit dm_exit(void) 313 { 314 int i = ARRAY_SIZE(_exits); 315 316 while (i--) 317 _exits[i](); 318 319 /* 320 * Should be empty by this point. 321 */ 322 idr_destroy(&_minor_idr); 323 } 324 325 /* 326 * Block device functions 327 */ 328 int dm_deleting_md(struct mapped_device *md) 329 { 330 return test_bit(DMF_DELETING, &md->flags); 331 } 332 333 static int dm_blk_open(struct block_device *bdev, fmode_t mode) 334 { 335 struct mapped_device *md; 336 337 spin_lock(&_minor_lock); 338 339 md = bdev->bd_disk->private_data; 340 if (!md) 341 goto out; 342 343 if (test_bit(DMF_FREEING, &md->flags) || 344 dm_deleting_md(md)) { 345 md = NULL; 346 goto out; 347 } 348 349 dm_get(md); 350 atomic_inc(&md->open_count); 351 out: 352 spin_unlock(&_minor_lock); 353 354 return md ? 0 : -ENXIO; 355 } 356 357 static void dm_blk_close(struct gendisk *disk, fmode_t mode) 358 { 359 struct mapped_device *md; 360 361 spin_lock(&_minor_lock); 362 363 md = disk->private_data; 364 if (WARN_ON(!md)) 365 goto out; 366 367 if (atomic_dec_and_test(&md->open_count) && 368 (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) 369 queue_work(deferred_remove_workqueue, &deferred_remove_work); 370 371 dm_put(md); 372 out: 373 spin_unlock(&_minor_lock); 374 } 375 376 int dm_open_count(struct mapped_device *md) 377 { 378 return atomic_read(&md->open_count); 379 } 380 381 /* 382 * Guarantees nothing is using the device before it's deleted. 383 */ 384 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) 385 { 386 int r = 0; 387 388 spin_lock(&_minor_lock); 389 390 if (dm_open_count(md)) { 391 r = -EBUSY; 392 if (mark_deferred) 393 set_bit(DMF_DEFERRED_REMOVE, &md->flags); 394 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) 395 r = -EEXIST; 396 else 397 set_bit(DMF_DELETING, &md->flags); 398 399 spin_unlock(&_minor_lock); 400 401 return r; 402 } 403 404 int dm_cancel_deferred_remove(struct mapped_device *md) 405 { 406 int r = 0; 407 408 spin_lock(&_minor_lock); 409 410 if (test_bit(DMF_DELETING, &md->flags)) 411 r = -EBUSY; 412 else 413 clear_bit(DMF_DEFERRED_REMOVE, &md->flags); 414 415 spin_unlock(&_minor_lock); 416 417 return r; 418 } 419 420 static void do_deferred_remove(struct work_struct *w) 421 { 422 dm_deferred_remove(); 423 } 424 425 sector_t dm_get_size(struct mapped_device *md) 426 { 427 return get_capacity(md->disk); 428 } 429 430 struct request_queue *dm_get_md_queue(struct mapped_device *md) 431 { 432 return md->queue; 433 } 434 435 struct dm_stats *dm_get_stats(struct mapped_device *md) 436 { 437 return &md->stats; 438 } 439 440 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) 441 { 442 struct mapped_device *md = bdev->bd_disk->private_data; 443 444 return dm_get_geometry(md, geo); 445 } 446 447 #ifdef CONFIG_BLK_DEV_ZONED 448 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data) 449 { 450 struct dm_report_zones_args *args = data; 451 sector_t sector_diff = args->tgt->begin - args->start; 452 453 /* 454 * Ignore zones beyond the target range. 455 */ 456 if (zone->start >= args->start + args->tgt->len) 457 return 0; 458 459 /* 460 * Remap the start sector and write pointer position of the zone 461 * to match its position in the target range. 462 */ 463 zone->start += sector_diff; 464 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { 465 if (zone->cond == BLK_ZONE_COND_FULL) 466 zone->wp = zone->start + zone->len; 467 else if (zone->cond == BLK_ZONE_COND_EMPTY) 468 zone->wp = zone->start; 469 else 470 zone->wp += sector_diff; 471 } 472 473 args->next_sector = zone->start + zone->len; 474 return args->orig_cb(zone, args->zone_idx++, args->orig_data); 475 } 476 EXPORT_SYMBOL_GPL(dm_report_zones_cb); 477 478 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector, 479 unsigned int nr_zones, report_zones_cb cb, void *data) 480 { 481 struct mapped_device *md = disk->private_data; 482 struct dm_table *map; 483 int srcu_idx, ret; 484 struct dm_report_zones_args args = { 485 .next_sector = sector, 486 .orig_data = data, 487 .orig_cb = cb, 488 }; 489 490 if (dm_suspended_md(md)) 491 return -EAGAIN; 492 493 map = dm_get_live_table(md, &srcu_idx); 494 if (!map) 495 return -EIO; 496 497 do { 498 struct dm_target *tgt; 499 500 tgt = dm_table_find_target(map, args.next_sector); 501 if (WARN_ON_ONCE(!tgt->type->report_zones)) { 502 ret = -EIO; 503 goto out; 504 } 505 506 args.tgt = tgt; 507 ret = tgt->type->report_zones(tgt, &args, 508 nr_zones - args.zone_idx); 509 if (ret < 0) 510 goto out; 511 } while (args.zone_idx < nr_zones && 512 args.next_sector < get_capacity(disk)); 513 514 ret = args.zone_idx; 515 out: 516 dm_put_live_table(md, srcu_idx); 517 return ret; 518 } 519 #else 520 #define dm_blk_report_zones NULL 521 #endif /* CONFIG_BLK_DEV_ZONED */ 522 523 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, 524 struct block_device **bdev) 525 __acquires(md->io_barrier) 526 { 527 struct dm_target *tgt; 528 struct dm_table *map; 529 int r; 530 531 retry: 532 r = -ENOTTY; 533 map = dm_get_live_table(md, srcu_idx); 534 if (!map || !dm_table_get_size(map)) 535 return r; 536 537 /* We only support devices that have a single target */ 538 if (dm_table_get_num_targets(map) != 1) 539 return r; 540 541 tgt = dm_table_get_target(map, 0); 542 if (!tgt->type->prepare_ioctl) 543 return r; 544 545 if (dm_suspended_md(md)) 546 return -EAGAIN; 547 548 r = tgt->type->prepare_ioctl(tgt, bdev); 549 if (r == -ENOTCONN && !fatal_signal_pending(current)) { 550 dm_put_live_table(md, *srcu_idx); 551 msleep(10); 552 goto retry; 553 } 554 555 return r; 556 } 557 558 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) 559 __releases(md->io_barrier) 560 { 561 dm_put_live_table(md, srcu_idx); 562 } 563 564 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, 565 unsigned int cmd, unsigned long arg) 566 { 567 struct mapped_device *md = bdev->bd_disk->private_data; 568 int r, srcu_idx; 569 570 r = dm_prepare_ioctl(md, &srcu_idx, &bdev); 571 if (r < 0) 572 goto out; 573 574 if (r > 0) { 575 /* 576 * Target determined this ioctl is being issued against a 577 * subset of the parent bdev; require extra privileges. 578 */ 579 if (!capable(CAP_SYS_RAWIO)) { 580 DMWARN_LIMIT( 581 "%s: sending ioctl %x to DM device without required privilege.", 582 current->comm, cmd); 583 r = -ENOIOCTLCMD; 584 goto out; 585 } 586 } 587 588 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); 589 out: 590 dm_unprepare_ioctl(md, srcu_idx); 591 return r; 592 } 593 594 static void start_io_acct(struct dm_io *io); 595 596 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) 597 { 598 struct dm_io *io; 599 struct dm_target_io *tio; 600 struct bio *clone; 601 602 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs); 603 if (!clone) 604 return NULL; 605 606 tio = container_of(clone, struct dm_target_io, clone); 607 tio->inside_dm_io = true; 608 tio->io = NULL; 609 610 io = container_of(tio, struct dm_io, tio); 611 io->magic = DM_IO_MAGIC; 612 io->status = 0; 613 atomic_set(&io->io_count, 1); 614 io->orig_bio = bio; 615 io->md = md; 616 spin_lock_init(&io->endio_lock); 617 618 start_io_acct(io); 619 620 return io; 621 } 622 623 static void free_io(struct mapped_device *md, struct dm_io *io) 624 { 625 bio_put(&io->tio.clone); 626 } 627 628 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, 629 unsigned target_bio_nr, gfp_t gfp_mask) 630 { 631 struct dm_target_io *tio; 632 633 if (!ci->io->tio.io) { 634 /* the dm_target_io embedded in ci->io is available */ 635 tio = &ci->io->tio; 636 } else { 637 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs); 638 if (!clone) 639 return NULL; 640 641 tio = container_of(clone, struct dm_target_io, clone); 642 tio->inside_dm_io = false; 643 } 644 645 tio->magic = DM_TIO_MAGIC; 646 tio->io = ci->io; 647 tio->ti = ti; 648 tio->target_bio_nr = target_bio_nr; 649 650 return tio; 651 } 652 653 static void free_tio(struct dm_target_io *tio) 654 { 655 if (tio->inside_dm_io) 656 return; 657 bio_put(&tio->clone); 658 } 659 660 u64 dm_start_time_ns_from_clone(struct bio *bio) 661 { 662 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); 663 struct dm_io *io = tio->io; 664 665 return jiffies_to_nsecs(io->start_time); 666 } 667 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone); 668 669 static void start_io_acct(struct dm_io *io) 670 { 671 struct mapped_device *md = io->md; 672 struct bio *bio = io->orig_bio; 673 674 io->start_time = bio_start_io_acct(bio); 675 if (unlikely(dm_stats_used(&md->stats))) 676 dm_stats_account_io(&md->stats, bio_data_dir(bio), 677 bio->bi_iter.bi_sector, bio_sectors(bio), 678 false, 0, &io->stats_aux); 679 } 680 681 static void end_io_acct(struct dm_io *io) 682 { 683 struct mapped_device *md = io->md; 684 struct bio *bio = io->orig_bio; 685 unsigned long duration = jiffies - io->start_time; 686 687 bio_end_io_acct(bio, io->start_time); 688 689 if (unlikely(dm_stats_used(&md->stats))) 690 dm_stats_account_io(&md->stats, bio_data_dir(bio), 691 bio->bi_iter.bi_sector, bio_sectors(bio), 692 true, duration, &io->stats_aux); 693 694 /* nudge anyone waiting on suspend queue */ 695 if (unlikely(wq_has_sleeper(&md->wait))) 696 wake_up(&md->wait); 697 } 698 699 /* 700 * Add the bio to the list of deferred io. 701 */ 702 static void queue_io(struct mapped_device *md, struct bio *bio) 703 { 704 unsigned long flags; 705 706 spin_lock_irqsave(&md->deferred_lock, flags); 707 bio_list_add(&md->deferred, bio); 708 spin_unlock_irqrestore(&md->deferred_lock, flags); 709 queue_work(md->wq, &md->work); 710 } 711 712 /* 713 * Everyone (including functions in this file), should use this 714 * function to access the md->map field, and make sure they call 715 * dm_put_live_table() when finished. 716 */ 717 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) 718 { 719 *srcu_idx = srcu_read_lock(&md->io_barrier); 720 721 return srcu_dereference(md->map, &md->io_barrier); 722 } 723 724 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) 725 { 726 srcu_read_unlock(&md->io_barrier, srcu_idx); 727 } 728 729 void dm_sync_table(struct mapped_device *md) 730 { 731 synchronize_srcu(&md->io_barrier); 732 synchronize_rcu_expedited(); 733 } 734 735 /* 736 * A fast alternative to dm_get_live_table/dm_put_live_table. 737 * The caller must not block between these two functions. 738 */ 739 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) 740 { 741 rcu_read_lock(); 742 return rcu_dereference(md->map); 743 } 744 745 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) 746 { 747 rcu_read_unlock(); 748 } 749 750 static char *_dm_claim_ptr = "I belong to device-mapper"; 751 752 /* 753 * Open a table device so we can use it as a map destination. 754 */ 755 static int open_table_device(struct table_device *td, dev_t dev, 756 struct mapped_device *md) 757 { 758 struct block_device *bdev; 759 760 int r; 761 762 BUG_ON(td->dm_dev.bdev); 763 764 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr); 765 if (IS_ERR(bdev)) 766 return PTR_ERR(bdev); 767 768 r = bd_link_disk_holder(bdev, dm_disk(md)); 769 if (r) { 770 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); 771 return r; 772 } 773 774 td->dm_dev.bdev = bdev; 775 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); 776 return 0; 777 } 778 779 /* 780 * Close a table device that we've been using. 781 */ 782 static void close_table_device(struct table_device *td, struct mapped_device *md) 783 { 784 if (!td->dm_dev.bdev) 785 return; 786 787 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); 788 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); 789 put_dax(td->dm_dev.dax_dev); 790 td->dm_dev.bdev = NULL; 791 td->dm_dev.dax_dev = NULL; 792 } 793 794 static struct table_device *find_table_device(struct list_head *l, dev_t dev, 795 fmode_t mode) 796 { 797 struct table_device *td; 798 799 list_for_each_entry(td, l, list) 800 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) 801 return td; 802 803 return NULL; 804 } 805 806 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, 807 struct dm_dev **result) 808 { 809 int r; 810 struct table_device *td; 811 812 mutex_lock(&md->table_devices_lock); 813 td = find_table_device(&md->table_devices, dev, mode); 814 if (!td) { 815 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); 816 if (!td) { 817 mutex_unlock(&md->table_devices_lock); 818 return -ENOMEM; 819 } 820 821 td->dm_dev.mode = mode; 822 td->dm_dev.bdev = NULL; 823 824 if ((r = open_table_device(td, dev, md))) { 825 mutex_unlock(&md->table_devices_lock); 826 kfree(td); 827 return r; 828 } 829 830 format_dev_t(td->dm_dev.name, dev); 831 832 refcount_set(&td->count, 1); 833 list_add(&td->list, &md->table_devices); 834 } else { 835 refcount_inc(&td->count); 836 } 837 mutex_unlock(&md->table_devices_lock); 838 839 *result = &td->dm_dev; 840 return 0; 841 } 842 EXPORT_SYMBOL_GPL(dm_get_table_device); 843 844 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) 845 { 846 struct table_device *td = container_of(d, struct table_device, dm_dev); 847 848 mutex_lock(&md->table_devices_lock); 849 if (refcount_dec_and_test(&td->count)) { 850 close_table_device(td, md); 851 list_del(&td->list); 852 kfree(td); 853 } 854 mutex_unlock(&md->table_devices_lock); 855 } 856 EXPORT_SYMBOL(dm_put_table_device); 857 858 static void free_table_devices(struct list_head *devices) 859 { 860 struct list_head *tmp, *next; 861 862 list_for_each_safe(tmp, next, devices) { 863 struct table_device *td = list_entry(tmp, struct table_device, list); 864 865 DMWARN("dm_destroy: %s still exists with %d references", 866 td->dm_dev.name, refcount_read(&td->count)); 867 kfree(td); 868 } 869 } 870 871 /* 872 * Get the geometry associated with a dm device 873 */ 874 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) 875 { 876 *geo = md->geometry; 877 878 return 0; 879 } 880 881 /* 882 * Set the geometry of a device. 883 */ 884 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) 885 { 886 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; 887 888 if (geo->start > sz) { 889 DMWARN("Start sector is beyond the geometry limits."); 890 return -EINVAL; 891 } 892 893 md->geometry = *geo; 894 895 return 0; 896 } 897 898 static int __noflush_suspending(struct mapped_device *md) 899 { 900 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 901 } 902 903 /* 904 * Decrements the number of outstanding ios that a bio has been 905 * cloned into, completing the original io if necc. 906 */ 907 static void dec_pending(struct dm_io *io, blk_status_t error) 908 { 909 unsigned long flags; 910 blk_status_t io_error; 911 struct bio *bio; 912 struct mapped_device *md = io->md; 913 914 /* Push-back supersedes any I/O errors */ 915 if (unlikely(error)) { 916 spin_lock_irqsave(&io->endio_lock, flags); 917 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) 918 io->status = error; 919 spin_unlock_irqrestore(&io->endio_lock, flags); 920 } 921 922 if (atomic_dec_and_test(&io->io_count)) { 923 if (io->status == BLK_STS_DM_REQUEUE) { 924 /* 925 * Target requested pushing back the I/O. 926 */ 927 spin_lock_irqsave(&md->deferred_lock, flags); 928 if (__noflush_suspending(md)) 929 /* NOTE early return due to BLK_STS_DM_REQUEUE below */ 930 bio_list_add_head(&md->deferred, io->orig_bio); 931 else 932 /* noflush suspend was interrupted. */ 933 io->status = BLK_STS_IOERR; 934 spin_unlock_irqrestore(&md->deferred_lock, flags); 935 } 936 937 io_error = io->status; 938 bio = io->orig_bio; 939 end_io_acct(io); 940 free_io(md, io); 941 942 if (io_error == BLK_STS_DM_REQUEUE) 943 return; 944 945 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { 946 /* 947 * Preflush done for flush with data, reissue 948 * without REQ_PREFLUSH. 949 */ 950 bio->bi_opf &= ~REQ_PREFLUSH; 951 queue_io(md, bio); 952 } else { 953 /* done with normal IO or empty flush */ 954 if (io_error) 955 bio->bi_status = io_error; 956 bio_endio(bio); 957 } 958 } 959 } 960 961 void disable_discard(struct mapped_device *md) 962 { 963 struct queue_limits *limits = dm_get_queue_limits(md); 964 965 /* device doesn't really support DISCARD, disable it */ 966 limits->max_discard_sectors = 0; 967 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue); 968 } 969 970 void disable_write_same(struct mapped_device *md) 971 { 972 struct queue_limits *limits = dm_get_queue_limits(md); 973 974 /* device doesn't really support WRITE SAME, disable it */ 975 limits->max_write_same_sectors = 0; 976 } 977 978 void disable_write_zeroes(struct mapped_device *md) 979 { 980 struct queue_limits *limits = dm_get_queue_limits(md); 981 982 /* device doesn't really support WRITE ZEROES, disable it */ 983 limits->max_write_zeroes_sectors = 0; 984 } 985 986 static void clone_endio(struct bio *bio) 987 { 988 blk_status_t error = bio->bi_status; 989 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); 990 struct dm_io *io = tio->io; 991 struct mapped_device *md = tio->io->md; 992 dm_endio_fn endio = tio->ti->type->end_io; 993 struct bio *orig_bio = io->orig_bio; 994 995 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { 996 if (bio_op(bio) == REQ_OP_DISCARD && 997 !bio->bi_disk->queue->limits.max_discard_sectors) 998 disable_discard(md); 999 else if (bio_op(bio) == REQ_OP_WRITE_SAME && 1000 !bio->bi_disk->queue->limits.max_write_same_sectors) 1001 disable_write_same(md); 1002 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES && 1003 !bio->bi_disk->queue->limits.max_write_zeroes_sectors) 1004 disable_write_zeroes(md); 1005 } 1006 1007 /* 1008 * For zone-append bios get offset in zone of the written 1009 * sector and add that to the original bio sector pos. 1010 */ 1011 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) { 1012 sector_t written_sector = bio->bi_iter.bi_sector; 1013 struct request_queue *q = orig_bio->bi_disk->queue; 1014 u64 mask = (u64)blk_queue_zone_sectors(q) - 1; 1015 1016 orig_bio->bi_iter.bi_sector += written_sector & mask; 1017 } 1018 1019 if (endio) { 1020 int r = endio(tio->ti, bio, &error); 1021 switch (r) { 1022 case DM_ENDIO_REQUEUE: 1023 error = BLK_STS_DM_REQUEUE; 1024 fallthrough; 1025 case DM_ENDIO_DONE: 1026 break; 1027 case DM_ENDIO_INCOMPLETE: 1028 /* The target will handle the io */ 1029 return; 1030 default: 1031 DMWARN("unimplemented target endio return value: %d", r); 1032 BUG(); 1033 } 1034 } 1035 1036 free_tio(tio); 1037 dec_pending(io, error); 1038 } 1039 1040 /* 1041 * Return maximum size of I/O possible at the supplied sector up to the current 1042 * target boundary. 1043 */ 1044 static inline sector_t max_io_len_target_boundary(struct dm_target *ti, 1045 sector_t target_offset) 1046 { 1047 return ti->len - target_offset; 1048 } 1049 1050 static sector_t max_io_len(struct dm_target *ti, sector_t sector) 1051 { 1052 sector_t target_offset = dm_target_offset(ti, sector); 1053 sector_t len = max_io_len_target_boundary(ti, target_offset); 1054 sector_t max_len; 1055 1056 /* 1057 * Does the target need to split even further? 1058 * - q->limits.chunk_sectors reflects ti->max_io_len so 1059 * blk_max_size_offset() provides required splitting. 1060 * - blk_max_size_offset() also respects q->limits.max_sectors 1061 */ 1062 max_len = blk_max_size_offset(dm_table_get_md(ti->table)->queue, 1063 target_offset); 1064 if (len > max_len) 1065 len = max_len; 1066 1067 return len; 1068 } 1069 1070 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) 1071 { 1072 if (len > UINT_MAX) { 1073 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", 1074 (unsigned long long)len, UINT_MAX); 1075 ti->error = "Maximum size of target IO is too large"; 1076 return -EINVAL; 1077 } 1078 1079 ti->max_io_len = (uint32_t) len; 1080 1081 return 0; 1082 } 1083 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); 1084 1085 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, 1086 sector_t sector, int *srcu_idx) 1087 __acquires(md->io_barrier) 1088 { 1089 struct dm_table *map; 1090 struct dm_target *ti; 1091 1092 map = dm_get_live_table(md, srcu_idx); 1093 if (!map) 1094 return NULL; 1095 1096 ti = dm_table_find_target(map, sector); 1097 if (!ti) 1098 return NULL; 1099 1100 return ti; 1101 } 1102 1103 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, 1104 long nr_pages, void **kaddr, pfn_t *pfn) 1105 { 1106 struct mapped_device *md = dax_get_private(dax_dev); 1107 sector_t sector = pgoff * PAGE_SECTORS; 1108 struct dm_target *ti; 1109 long len, ret = -EIO; 1110 int srcu_idx; 1111 1112 ti = dm_dax_get_live_target(md, sector, &srcu_idx); 1113 1114 if (!ti) 1115 goto out; 1116 if (!ti->type->direct_access) 1117 goto out; 1118 len = max_io_len(ti, sector) / PAGE_SECTORS; 1119 if (len < 1) 1120 goto out; 1121 nr_pages = min(len, nr_pages); 1122 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); 1123 1124 out: 1125 dm_put_live_table(md, srcu_idx); 1126 1127 return ret; 1128 } 1129 1130 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev, 1131 int blocksize, sector_t start, sector_t len) 1132 { 1133 struct mapped_device *md = dax_get_private(dax_dev); 1134 struct dm_table *map; 1135 bool ret = false; 1136 int srcu_idx; 1137 1138 map = dm_get_live_table(md, &srcu_idx); 1139 if (!map) 1140 goto out; 1141 1142 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize); 1143 1144 out: 1145 dm_put_live_table(md, srcu_idx); 1146 1147 return ret; 1148 } 1149 1150 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, 1151 void *addr, size_t bytes, struct iov_iter *i) 1152 { 1153 struct mapped_device *md = dax_get_private(dax_dev); 1154 sector_t sector = pgoff * PAGE_SECTORS; 1155 struct dm_target *ti; 1156 long ret = 0; 1157 int srcu_idx; 1158 1159 ti = dm_dax_get_live_target(md, sector, &srcu_idx); 1160 1161 if (!ti) 1162 goto out; 1163 if (!ti->type->dax_copy_from_iter) { 1164 ret = copy_from_iter(addr, bytes, i); 1165 goto out; 1166 } 1167 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); 1168 out: 1169 dm_put_live_table(md, srcu_idx); 1170 1171 return ret; 1172 } 1173 1174 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, 1175 void *addr, size_t bytes, struct iov_iter *i) 1176 { 1177 struct mapped_device *md = dax_get_private(dax_dev); 1178 sector_t sector = pgoff * PAGE_SECTORS; 1179 struct dm_target *ti; 1180 long ret = 0; 1181 int srcu_idx; 1182 1183 ti = dm_dax_get_live_target(md, sector, &srcu_idx); 1184 1185 if (!ti) 1186 goto out; 1187 if (!ti->type->dax_copy_to_iter) { 1188 ret = copy_to_iter(addr, bytes, i); 1189 goto out; 1190 } 1191 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i); 1192 out: 1193 dm_put_live_table(md, srcu_idx); 1194 1195 return ret; 1196 } 1197 1198 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, 1199 size_t nr_pages) 1200 { 1201 struct mapped_device *md = dax_get_private(dax_dev); 1202 sector_t sector = pgoff * PAGE_SECTORS; 1203 struct dm_target *ti; 1204 int ret = -EIO; 1205 int srcu_idx; 1206 1207 ti = dm_dax_get_live_target(md, sector, &srcu_idx); 1208 1209 if (!ti) 1210 goto out; 1211 if (WARN_ON(!ti->type->dax_zero_page_range)) { 1212 /* 1213 * ->zero_page_range() is mandatory dax operation. If we are 1214 * here, something is wrong. 1215 */ 1216 dm_put_live_table(md, srcu_idx); 1217 goto out; 1218 } 1219 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages); 1220 1221 out: 1222 dm_put_live_table(md, srcu_idx); 1223 1224 return ret; 1225 } 1226 1227 /* 1228 * A target may call dm_accept_partial_bio only from the map routine. It is 1229 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET, 1230 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH. 1231 * 1232 * dm_accept_partial_bio informs the dm that the target only wants to process 1233 * additional n_sectors sectors of the bio and the rest of the data should be 1234 * sent in a next bio. 1235 * 1236 * A diagram that explains the arithmetics: 1237 * +--------------------+---------------+-------+ 1238 * | 1 | 2 | 3 | 1239 * +--------------------+---------------+-------+ 1240 * 1241 * <-------------- *tio->len_ptr ---------------> 1242 * <------- bi_size -------> 1243 * <-- n_sectors --> 1244 * 1245 * Region 1 was already iterated over with bio_advance or similar function. 1246 * (it may be empty if the target doesn't use bio_advance) 1247 * Region 2 is the remaining bio size that the target wants to process. 1248 * (it may be empty if region 1 is non-empty, although there is no reason 1249 * to make it empty) 1250 * The target requires that region 3 is to be sent in the next bio. 1251 * 1252 * If the target wants to receive multiple copies of the bio (via num_*bios, etc), 1253 * the partially processed part (the sum of regions 1+2) must be the same for all 1254 * copies of the bio. 1255 */ 1256 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) 1257 { 1258 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); 1259 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; 1260 BUG_ON(bio->bi_opf & REQ_PREFLUSH); 1261 BUG_ON(bi_size > *tio->len_ptr); 1262 BUG_ON(n_sectors > bi_size); 1263 *tio->len_ptr -= bi_size - n_sectors; 1264 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; 1265 } 1266 EXPORT_SYMBOL_GPL(dm_accept_partial_bio); 1267 1268 static blk_qc_t __map_bio(struct dm_target_io *tio) 1269 { 1270 int r; 1271 sector_t sector; 1272 struct bio *clone = &tio->clone; 1273 struct dm_io *io = tio->io; 1274 struct dm_target *ti = tio->ti; 1275 blk_qc_t ret = BLK_QC_T_NONE; 1276 1277 clone->bi_end_io = clone_endio; 1278 1279 /* 1280 * Map the clone. If r == 0 we don't need to do 1281 * anything, the target has assumed ownership of 1282 * this io. 1283 */ 1284 atomic_inc(&io->io_count); 1285 sector = clone->bi_iter.bi_sector; 1286 1287 r = ti->type->map(ti, clone); 1288 switch (r) { 1289 case DM_MAPIO_SUBMITTED: 1290 break; 1291 case DM_MAPIO_REMAPPED: 1292 /* the bio has been remapped so dispatch it */ 1293 trace_block_bio_remap(clone->bi_disk->queue, clone, 1294 bio_dev(io->orig_bio), sector); 1295 ret = submit_bio_noacct(clone); 1296 break; 1297 case DM_MAPIO_KILL: 1298 free_tio(tio); 1299 dec_pending(io, BLK_STS_IOERR); 1300 break; 1301 case DM_MAPIO_REQUEUE: 1302 free_tio(tio); 1303 dec_pending(io, BLK_STS_DM_REQUEUE); 1304 break; 1305 default: 1306 DMWARN("unimplemented target map return value: %d", r); 1307 BUG(); 1308 } 1309 1310 return ret; 1311 } 1312 1313 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) 1314 { 1315 bio->bi_iter.bi_sector = sector; 1316 bio->bi_iter.bi_size = to_bytes(len); 1317 } 1318 1319 /* 1320 * Creates a bio that consists of range of complete bvecs. 1321 */ 1322 static int clone_bio(struct dm_target_io *tio, struct bio *bio, 1323 sector_t sector, unsigned len) 1324 { 1325 struct bio *clone = &tio->clone; 1326 1327 __bio_clone_fast(clone, bio); 1328 1329 bio_crypt_clone(clone, bio, GFP_NOIO); 1330 1331 if (bio_integrity(bio)) { 1332 int r; 1333 1334 if (unlikely(!dm_target_has_integrity(tio->ti->type) && 1335 !dm_target_passes_integrity(tio->ti->type))) { 1336 DMWARN("%s: the target %s doesn't support integrity data.", 1337 dm_device_name(tio->io->md), 1338 tio->ti->type->name); 1339 return -EIO; 1340 } 1341 1342 r = bio_integrity_clone(clone, bio, GFP_NOIO); 1343 if (r < 0) 1344 return r; 1345 } 1346 1347 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); 1348 clone->bi_iter.bi_size = to_bytes(len); 1349 1350 if (bio_integrity(bio)) 1351 bio_integrity_trim(clone); 1352 1353 return 0; 1354 } 1355 1356 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, 1357 struct dm_target *ti, unsigned num_bios) 1358 { 1359 struct dm_target_io *tio; 1360 int try; 1361 1362 if (!num_bios) 1363 return; 1364 1365 if (num_bios == 1) { 1366 tio = alloc_tio(ci, ti, 0, GFP_NOIO); 1367 bio_list_add(blist, &tio->clone); 1368 return; 1369 } 1370 1371 for (try = 0; try < 2; try++) { 1372 int bio_nr; 1373 struct bio *bio; 1374 1375 if (try) 1376 mutex_lock(&ci->io->md->table_devices_lock); 1377 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { 1378 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); 1379 if (!tio) 1380 break; 1381 1382 bio_list_add(blist, &tio->clone); 1383 } 1384 if (try) 1385 mutex_unlock(&ci->io->md->table_devices_lock); 1386 if (bio_nr == num_bios) 1387 return; 1388 1389 while ((bio = bio_list_pop(blist))) { 1390 tio = container_of(bio, struct dm_target_io, clone); 1391 free_tio(tio); 1392 } 1393 } 1394 } 1395 1396 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, 1397 struct dm_target_io *tio, unsigned *len) 1398 { 1399 struct bio *clone = &tio->clone; 1400 1401 tio->len_ptr = len; 1402 1403 __bio_clone_fast(clone, ci->bio); 1404 if (len) 1405 bio_setup_sector(clone, ci->sector, *len); 1406 1407 return __map_bio(tio); 1408 } 1409 1410 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, 1411 unsigned num_bios, unsigned *len) 1412 { 1413 struct bio_list blist = BIO_EMPTY_LIST; 1414 struct bio *bio; 1415 struct dm_target_io *tio; 1416 1417 alloc_multiple_bios(&blist, ci, ti, num_bios); 1418 1419 while ((bio = bio_list_pop(&blist))) { 1420 tio = container_of(bio, struct dm_target_io, clone); 1421 (void) __clone_and_map_simple_bio(ci, tio, len); 1422 } 1423 } 1424 1425 static int __send_empty_flush(struct clone_info *ci) 1426 { 1427 unsigned target_nr = 0; 1428 struct dm_target *ti; 1429 struct bio flush_bio; 1430 1431 /* 1432 * Use an on-stack bio for this, it's safe since we don't 1433 * need to reference it after submit. It's just used as 1434 * the basis for the clone(s). 1435 */ 1436 bio_init(&flush_bio, NULL, 0); 1437 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; 1438 ci->bio = &flush_bio; 1439 ci->sector_count = 0; 1440 1441 /* 1442 * Empty flush uses a statically initialized bio, as the base for 1443 * cloning. However, blkg association requires that a bdev is 1444 * associated with a gendisk, which doesn't happen until the bdev is 1445 * opened. So, blkg association is done at issue time of the flush 1446 * rather than when the device is created in alloc_dev(). 1447 */ 1448 bio_set_dev(ci->bio, ci->io->md->bdev); 1449 1450 BUG_ON(bio_has_data(ci->bio)); 1451 while ((ti = dm_table_get_target(ci->map, target_nr++))) 1452 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); 1453 1454 bio_uninit(ci->bio); 1455 return 0; 1456 } 1457 1458 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, 1459 sector_t sector, unsigned *len) 1460 { 1461 struct bio *bio = ci->bio; 1462 struct dm_target_io *tio; 1463 int r; 1464 1465 tio = alloc_tio(ci, ti, 0, GFP_NOIO); 1466 tio->len_ptr = len; 1467 r = clone_bio(tio, bio, sector, *len); 1468 if (r < 0) { 1469 free_tio(tio); 1470 return r; 1471 } 1472 (void) __map_bio(tio); 1473 1474 return 0; 1475 } 1476 1477 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); 1478 1479 static unsigned get_num_discard_bios(struct dm_target *ti) 1480 { 1481 return ti->num_discard_bios; 1482 } 1483 1484 static unsigned get_num_secure_erase_bios(struct dm_target *ti) 1485 { 1486 return ti->num_secure_erase_bios; 1487 } 1488 1489 static unsigned get_num_write_same_bios(struct dm_target *ti) 1490 { 1491 return ti->num_write_same_bios; 1492 } 1493 1494 static unsigned get_num_write_zeroes_bios(struct dm_target *ti) 1495 { 1496 return ti->num_write_zeroes_bios; 1497 } 1498 1499 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, 1500 unsigned num_bios) 1501 { 1502 unsigned len; 1503 1504 /* 1505 * Even though the device advertised support for this type of 1506 * request, that does not mean every target supports it, and 1507 * reconfiguration might also have changed that since the 1508 * check was performed. 1509 */ 1510 if (!num_bios) 1511 return -EOPNOTSUPP; 1512 1513 len = min_t(sector_t, ci->sector_count, 1514 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector))); 1515 1516 __send_duplicate_bios(ci, ti, num_bios, &len); 1517 1518 ci->sector += len; 1519 ci->sector_count -= len; 1520 1521 return 0; 1522 } 1523 1524 static int __send_discard(struct clone_info *ci, struct dm_target *ti) 1525 { 1526 return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti)); 1527 } 1528 1529 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti) 1530 { 1531 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti)); 1532 } 1533 1534 static int __send_write_same(struct clone_info *ci, struct dm_target *ti) 1535 { 1536 return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti)); 1537 } 1538 1539 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) 1540 { 1541 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti)); 1542 } 1543 1544 static bool is_abnormal_io(struct bio *bio) 1545 { 1546 bool r = false; 1547 1548 switch (bio_op(bio)) { 1549 case REQ_OP_DISCARD: 1550 case REQ_OP_SECURE_ERASE: 1551 case REQ_OP_WRITE_SAME: 1552 case REQ_OP_WRITE_ZEROES: 1553 r = true; 1554 break; 1555 } 1556 1557 return r; 1558 } 1559 1560 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti, 1561 int *result) 1562 { 1563 struct bio *bio = ci->bio; 1564 1565 if (bio_op(bio) == REQ_OP_DISCARD) 1566 *result = __send_discard(ci, ti); 1567 else if (bio_op(bio) == REQ_OP_SECURE_ERASE) 1568 *result = __send_secure_erase(ci, ti); 1569 else if (bio_op(bio) == REQ_OP_WRITE_SAME) 1570 *result = __send_write_same(ci, ti); 1571 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES) 1572 *result = __send_write_zeroes(ci, ti); 1573 else 1574 return false; 1575 1576 return true; 1577 } 1578 1579 /* 1580 * Select the correct strategy for processing a non-flush bio. 1581 */ 1582 static int __split_and_process_non_flush(struct clone_info *ci) 1583 { 1584 struct dm_target *ti; 1585 unsigned len; 1586 int r; 1587 1588 ti = dm_table_find_target(ci->map, ci->sector); 1589 if (!ti) 1590 return -EIO; 1591 1592 if (__process_abnormal_io(ci, ti, &r)) 1593 return r; 1594 1595 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count); 1596 1597 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); 1598 if (r < 0) 1599 return r; 1600 1601 ci->sector += len; 1602 ci->sector_count -= len; 1603 1604 return 0; 1605 } 1606 1607 static void init_clone_info(struct clone_info *ci, struct mapped_device *md, 1608 struct dm_table *map, struct bio *bio) 1609 { 1610 ci->map = map; 1611 ci->io = alloc_io(md, bio); 1612 ci->sector = bio->bi_iter.bi_sector; 1613 } 1614 1615 #define __dm_part_stat_sub(part, field, subnd) \ 1616 (part_stat_get(part, field) -= (subnd)) 1617 1618 /* 1619 * Entry point to split a bio into clones and submit them to the targets. 1620 */ 1621 static blk_qc_t __split_and_process_bio(struct mapped_device *md, 1622 struct dm_table *map, struct bio *bio) 1623 { 1624 struct clone_info ci; 1625 blk_qc_t ret = BLK_QC_T_NONE; 1626 int error = 0; 1627 1628 init_clone_info(&ci, md, map, bio); 1629 1630 if (bio->bi_opf & REQ_PREFLUSH) { 1631 error = __send_empty_flush(&ci); 1632 /* dec_pending submits any data associated with flush */ 1633 } else if (op_is_zone_mgmt(bio_op(bio))) { 1634 ci.bio = bio; 1635 ci.sector_count = 0; 1636 error = __split_and_process_non_flush(&ci); 1637 } else { 1638 ci.bio = bio; 1639 ci.sector_count = bio_sectors(bio); 1640 while (ci.sector_count && !error) { 1641 error = __split_and_process_non_flush(&ci); 1642 if (current->bio_list && ci.sector_count && !error) { 1643 /* 1644 * Remainder must be passed to submit_bio_noacct() 1645 * so that it gets handled *after* bios already submitted 1646 * have been completely processed. 1647 * We take a clone of the original to store in 1648 * ci.io->orig_bio to be used by end_io_acct() and 1649 * for dec_pending to use for completion handling. 1650 */ 1651 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count, 1652 GFP_NOIO, &md->queue->bio_split); 1653 ci.io->orig_bio = b; 1654 1655 /* 1656 * Adjust IO stats for each split, otherwise upon queue 1657 * reentry there will be redundant IO accounting. 1658 * NOTE: this is a stop-gap fix, a proper fix involves 1659 * significant refactoring of DM core's bio splitting 1660 * (by eliminating DM's splitting and just using bio_split) 1661 */ 1662 part_stat_lock(); 1663 __dm_part_stat_sub(&dm_disk(md)->part0, 1664 sectors[op_stat_group(bio_op(bio))], ci.sector_count); 1665 part_stat_unlock(); 1666 1667 bio_chain(b, bio); 1668 trace_block_split(md->queue, b, bio->bi_iter.bi_sector); 1669 ret = submit_bio_noacct(bio); 1670 break; 1671 } 1672 } 1673 } 1674 1675 /* drop the extra reference count */ 1676 dec_pending(ci.io, errno_to_blk_status(error)); 1677 return ret; 1678 } 1679 1680 /* 1681 * Optimized variant of __split_and_process_bio that leverages the 1682 * fact that targets that use it do _not_ have a need to split bios. 1683 */ 1684 static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map, 1685 struct bio *bio) 1686 { 1687 struct clone_info ci; 1688 blk_qc_t ret = BLK_QC_T_NONE; 1689 int error = 0; 1690 1691 init_clone_info(&ci, md, map, bio); 1692 1693 if (bio->bi_opf & REQ_PREFLUSH) { 1694 error = __send_empty_flush(&ci); 1695 /* dec_pending submits any data associated with flush */ 1696 } else { 1697 struct dm_target_io *tio; 1698 struct dm_target *ti = md->immutable_target; 1699 1700 if (WARN_ON_ONCE(!ti)) { 1701 error = -EIO; 1702 goto out; 1703 } 1704 1705 ci.bio = bio; 1706 ci.sector_count = bio_sectors(bio); 1707 if (__process_abnormal_io(&ci, ti, &error)) 1708 goto out; 1709 1710 tio = alloc_tio(&ci, ti, 0, GFP_NOIO); 1711 ret = __clone_and_map_simple_bio(&ci, tio, NULL); 1712 } 1713 out: 1714 /* drop the extra reference count */ 1715 dec_pending(ci.io, errno_to_blk_status(error)); 1716 return ret; 1717 } 1718 1719 static blk_qc_t dm_process_bio(struct mapped_device *md, 1720 struct dm_table *map, struct bio *bio) 1721 { 1722 blk_qc_t ret = BLK_QC_T_NONE; 1723 1724 if (unlikely(!map)) { 1725 bio_io_error(bio); 1726 return ret; 1727 } 1728 1729 /* 1730 * If in ->submit_bio we need to use blk_queue_split(), otherwise 1731 * queue_limits for abnormal requests (e.g. discard, writesame, etc) 1732 * won't be imposed. 1733 * If called from dm_wq_work() for deferred bio processing, bio 1734 * was already handled by following code with previous ->submit_bio. 1735 */ 1736 if (current->bio_list) { 1737 if (is_abnormal_io(bio)) 1738 blk_queue_split(&bio); 1739 /* regular IO is split by __split_and_process_bio */ 1740 } 1741 1742 if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED) 1743 return __process_bio(md, map, bio); 1744 return __split_and_process_bio(md, map, bio); 1745 } 1746 1747 static blk_qc_t dm_submit_bio(struct bio *bio) 1748 { 1749 struct mapped_device *md = bio->bi_disk->private_data; 1750 blk_qc_t ret = BLK_QC_T_NONE; 1751 int srcu_idx; 1752 struct dm_table *map; 1753 1754 if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) { 1755 /* 1756 * We are called with a live reference on q_usage_counter, but 1757 * that one will be released as soon as we return. Grab an 1758 * extra one as blk_mq_submit_bio expects to be able to consume 1759 * a reference (which lives until the request is freed in case a 1760 * request is allocated). 1761 */ 1762 percpu_ref_get(&bio->bi_disk->queue->q_usage_counter); 1763 return blk_mq_submit_bio(bio); 1764 } 1765 1766 map = dm_get_live_table(md, &srcu_idx); 1767 1768 /* if we're suspended, we have to queue this io for later */ 1769 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { 1770 dm_put_live_table(md, srcu_idx); 1771 1772 if (bio->bi_opf & REQ_NOWAIT) 1773 bio_wouldblock_error(bio); 1774 else if (!(bio->bi_opf & REQ_RAHEAD)) 1775 queue_io(md, bio); 1776 else 1777 bio_io_error(bio); 1778 return ret; 1779 } 1780 1781 ret = dm_process_bio(md, map, bio); 1782 1783 dm_put_live_table(md, srcu_idx); 1784 return ret; 1785 } 1786 1787 /*----------------------------------------------------------------- 1788 * An IDR is used to keep track of allocated minor numbers. 1789 *---------------------------------------------------------------*/ 1790 static void free_minor(int minor) 1791 { 1792 spin_lock(&_minor_lock); 1793 idr_remove(&_minor_idr, minor); 1794 spin_unlock(&_minor_lock); 1795 } 1796 1797 /* 1798 * See if the device with a specific minor # is free. 1799 */ 1800 static int specific_minor(int minor) 1801 { 1802 int r; 1803 1804 if (minor >= (1 << MINORBITS)) 1805 return -EINVAL; 1806 1807 idr_preload(GFP_KERNEL); 1808 spin_lock(&_minor_lock); 1809 1810 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); 1811 1812 spin_unlock(&_minor_lock); 1813 idr_preload_end(); 1814 if (r < 0) 1815 return r == -ENOSPC ? -EBUSY : r; 1816 return 0; 1817 } 1818 1819 static int next_free_minor(int *minor) 1820 { 1821 int r; 1822 1823 idr_preload(GFP_KERNEL); 1824 spin_lock(&_minor_lock); 1825 1826 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); 1827 1828 spin_unlock(&_minor_lock); 1829 idr_preload_end(); 1830 if (r < 0) 1831 return r; 1832 *minor = r; 1833 return 0; 1834 } 1835 1836 static const struct block_device_operations dm_blk_dops; 1837 static const struct dax_operations dm_dax_ops; 1838 1839 static void dm_wq_work(struct work_struct *work); 1840 1841 static void cleanup_mapped_device(struct mapped_device *md) 1842 { 1843 if (md->wq) 1844 destroy_workqueue(md->wq); 1845 bioset_exit(&md->bs); 1846 bioset_exit(&md->io_bs); 1847 1848 if (md->dax_dev) { 1849 kill_dax(md->dax_dev); 1850 put_dax(md->dax_dev); 1851 md->dax_dev = NULL; 1852 } 1853 1854 if (md->disk) { 1855 spin_lock(&_minor_lock); 1856 md->disk->private_data = NULL; 1857 spin_unlock(&_minor_lock); 1858 del_gendisk(md->disk); 1859 put_disk(md->disk); 1860 } 1861 1862 if (md->queue) 1863 blk_cleanup_queue(md->queue); 1864 1865 cleanup_srcu_struct(&md->io_barrier); 1866 1867 if (md->bdev) { 1868 bdput(md->bdev); 1869 md->bdev = NULL; 1870 } 1871 1872 mutex_destroy(&md->suspend_lock); 1873 mutex_destroy(&md->type_lock); 1874 mutex_destroy(&md->table_devices_lock); 1875 1876 dm_mq_cleanup_mapped_device(md); 1877 } 1878 1879 /* 1880 * Allocate and initialise a blank device with a given minor. 1881 */ 1882 static struct mapped_device *alloc_dev(int minor) 1883 { 1884 int r, numa_node_id = dm_get_numa_node(); 1885 struct mapped_device *md; 1886 void *old_md; 1887 1888 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); 1889 if (!md) { 1890 DMWARN("unable to allocate device, out of memory."); 1891 return NULL; 1892 } 1893 1894 if (!try_module_get(THIS_MODULE)) 1895 goto bad_module_get; 1896 1897 /* get a minor number for the dev */ 1898 if (minor == DM_ANY_MINOR) 1899 r = next_free_minor(&minor); 1900 else 1901 r = specific_minor(minor); 1902 if (r < 0) 1903 goto bad_minor; 1904 1905 r = init_srcu_struct(&md->io_barrier); 1906 if (r < 0) 1907 goto bad_io_barrier; 1908 1909 md->numa_node_id = numa_node_id; 1910 md->init_tio_pdu = false; 1911 md->type = DM_TYPE_NONE; 1912 mutex_init(&md->suspend_lock); 1913 mutex_init(&md->type_lock); 1914 mutex_init(&md->table_devices_lock); 1915 spin_lock_init(&md->deferred_lock); 1916 atomic_set(&md->holders, 1); 1917 atomic_set(&md->open_count, 0); 1918 atomic_set(&md->event_nr, 0); 1919 atomic_set(&md->uevent_seq, 0); 1920 INIT_LIST_HEAD(&md->uevent_list); 1921 INIT_LIST_HEAD(&md->table_devices); 1922 spin_lock_init(&md->uevent_lock); 1923 1924 /* 1925 * default to bio-based until DM table is loaded and md->type 1926 * established. If request-based table is loaded: blk-mq will 1927 * override accordingly. 1928 */ 1929 md->queue = blk_alloc_queue(numa_node_id); 1930 if (!md->queue) 1931 goto bad; 1932 1933 md->disk = alloc_disk_node(1, md->numa_node_id); 1934 if (!md->disk) 1935 goto bad; 1936 1937 init_waitqueue_head(&md->wait); 1938 INIT_WORK(&md->work, dm_wq_work); 1939 init_waitqueue_head(&md->eventq); 1940 init_completion(&md->kobj_holder.completion); 1941 1942 md->disk->major = _major; 1943 md->disk->first_minor = minor; 1944 md->disk->fops = &dm_blk_dops; 1945 md->disk->queue = md->queue; 1946 md->disk->private_data = md; 1947 sprintf(md->disk->disk_name, "dm-%d", minor); 1948 1949 if (IS_ENABLED(CONFIG_DAX_DRIVER)) { 1950 md->dax_dev = alloc_dax(md, md->disk->disk_name, 1951 &dm_dax_ops, 0); 1952 if (IS_ERR(md->dax_dev)) 1953 goto bad; 1954 } 1955 1956 add_disk_no_queue_reg(md->disk); 1957 format_dev_t(md->name, MKDEV(_major, minor)); 1958 1959 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); 1960 if (!md->wq) 1961 goto bad; 1962 1963 md->bdev = bdget_disk(md->disk, 0); 1964 if (!md->bdev) 1965 goto bad; 1966 1967 dm_stats_init(&md->stats); 1968 1969 /* Populate the mapping, nobody knows we exist yet */ 1970 spin_lock(&_minor_lock); 1971 old_md = idr_replace(&_minor_idr, md, minor); 1972 spin_unlock(&_minor_lock); 1973 1974 BUG_ON(old_md != MINOR_ALLOCED); 1975 1976 return md; 1977 1978 bad: 1979 cleanup_mapped_device(md); 1980 bad_io_barrier: 1981 free_minor(minor); 1982 bad_minor: 1983 module_put(THIS_MODULE); 1984 bad_module_get: 1985 kvfree(md); 1986 return NULL; 1987 } 1988 1989 static void unlock_fs(struct mapped_device *md); 1990 1991 static void free_dev(struct mapped_device *md) 1992 { 1993 int minor = MINOR(disk_devt(md->disk)); 1994 1995 unlock_fs(md); 1996 1997 cleanup_mapped_device(md); 1998 1999 free_table_devices(&md->table_devices); 2000 dm_stats_cleanup(&md->stats); 2001 free_minor(minor); 2002 2003 module_put(THIS_MODULE); 2004 kvfree(md); 2005 } 2006 2007 static int __bind_mempools(struct mapped_device *md, struct dm_table *t) 2008 { 2009 struct dm_md_mempools *p = dm_table_get_md_mempools(t); 2010 int ret = 0; 2011 2012 if (dm_table_bio_based(t)) { 2013 /* 2014 * The md may already have mempools that need changing. 2015 * If so, reload bioset because front_pad may have changed 2016 * because a different table was loaded. 2017 */ 2018 bioset_exit(&md->bs); 2019 bioset_exit(&md->io_bs); 2020 2021 } else if (bioset_initialized(&md->bs)) { 2022 /* 2023 * There's no need to reload with request-based dm 2024 * because the size of front_pad doesn't change. 2025 * Note for future: If you are to reload bioset, 2026 * prep-ed requests in the queue may refer 2027 * to bio from the old bioset, so you must walk 2028 * through the queue to unprep. 2029 */ 2030 goto out; 2031 } 2032 2033 BUG_ON(!p || 2034 bioset_initialized(&md->bs) || 2035 bioset_initialized(&md->io_bs)); 2036 2037 ret = bioset_init_from_src(&md->bs, &p->bs); 2038 if (ret) 2039 goto out; 2040 ret = bioset_init_from_src(&md->io_bs, &p->io_bs); 2041 if (ret) 2042 bioset_exit(&md->bs); 2043 out: 2044 /* mempool bind completed, no longer need any mempools in the table */ 2045 dm_table_free_md_mempools(t); 2046 return ret; 2047 } 2048 2049 /* 2050 * Bind a table to the device. 2051 */ 2052 static void event_callback(void *context) 2053 { 2054 unsigned long flags; 2055 LIST_HEAD(uevents); 2056 struct mapped_device *md = (struct mapped_device *) context; 2057 2058 spin_lock_irqsave(&md->uevent_lock, flags); 2059 list_splice_init(&md->uevent_list, &uevents); 2060 spin_unlock_irqrestore(&md->uevent_lock, flags); 2061 2062 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); 2063 2064 atomic_inc(&md->event_nr); 2065 wake_up(&md->eventq); 2066 dm_issue_global_event(); 2067 } 2068 2069 /* 2070 * Returns old map, which caller must destroy. 2071 */ 2072 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, 2073 struct queue_limits *limits) 2074 { 2075 struct dm_table *old_map; 2076 struct request_queue *q = md->queue; 2077 bool request_based = dm_table_request_based(t); 2078 sector_t size; 2079 int ret; 2080 2081 lockdep_assert_held(&md->suspend_lock); 2082 2083 size = dm_table_get_size(t); 2084 2085 /* 2086 * Wipe any geometry if the size of the table changed. 2087 */ 2088 if (size != dm_get_size(md)) 2089 memset(&md->geometry, 0, sizeof(md->geometry)); 2090 2091 set_capacity(md->disk, size); 2092 bd_set_nr_sectors(md->bdev, size); 2093 2094 dm_table_event_callback(t, event_callback, md); 2095 2096 /* 2097 * The queue hasn't been stopped yet, if the old table type wasn't 2098 * for request-based during suspension. So stop it to prevent 2099 * I/O mapping before resume. 2100 * This must be done before setting the queue restrictions, 2101 * because request-based dm may be run just after the setting. 2102 */ 2103 if (request_based) 2104 dm_stop_queue(q); 2105 2106 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { 2107 /* 2108 * Leverage the fact that request-based DM targets and 2109 * NVMe bio based targets are immutable singletons 2110 * - used to optimize both __process_bio and dm_mq_queue_rq 2111 */ 2112 md->immutable_target = dm_table_get_immutable_target(t); 2113 } 2114 2115 ret = __bind_mempools(md, t); 2116 if (ret) { 2117 old_map = ERR_PTR(ret); 2118 goto out; 2119 } 2120 2121 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 2122 rcu_assign_pointer(md->map, (void *)t); 2123 md->immutable_target_type = dm_table_get_immutable_target_type(t); 2124 2125 dm_table_set_restrictions(t, q, limits); 2126 if (old_map) 2127 dm_sync_table(md); 2128 2129 out: 2130 return old_map; 2131 } 2132 2133 /* 2134 * Returns unbound table for the caller to free. 2135 */ 2136 static struct dm_table *__unbind(struct mapped_device *md) 2137 { 2138 struct dm_table *map = rcu_dereference_protected(md->map, 1); 2139 2140 if (!map) 2141 return NULL; 2142 2143 dm_table_event_callback(map, NULL, NULL); 2144 RCU_INIT_POINTER(md->map, NULL); 2145 dm_sync_table(md); 2146 2147 return map; 2148 } 2149 2150 /* 2151 * Constructor for a new device. 2152 */ 2153 int dm_create(int minor, struct mapped_device **result) 2154 { 2155 int r; 2156 struct mapped_device *md; 2157 2158 md = alloc_dev(minor); 2159 if (!md) 2160 return -ENXIO; 2161 2162 r = dm_sysfs_init(md); 2163 if (r) { 2164 free_dev(md); 2165 return r; 2166 } 2167 2168 *result = md; 2169 return 0; 2170 } 2171 2172 /* 2173 * Functions to manage md->type. 2174 * All are required to hold md->type_lock. 2175 */ 2176 void dm_lock_md_type(struct mapped_device *md) 2177 { 2178 mutex_lock(&md->type_lock); 2179 } 2180 2181 void dm_unlock_md_type(struct mapped_device *md) 2182 { 2183 mutex_unlock(&md->type_lock); 2184 } 2185 2186 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) 2187 { 2188 BUG_ON(!mutex_is_locked(&md->type_lock)); 2189 md->type = type; 2190 } 2191 2192 enum dm_queue_mode dm_get_md_type(struct mapped_device *md) 2193 { 2194 return md->type; 2195 } 2196 2197 struct target_type *dm_get_immutable_target_type(struct mapped_device *md) 2198 { 2199 return md->immutable_target_type; 2200 } 2201 2202 /* 2203 * The queue_limits are only valid as long as you have a reference 2204 * count on 'md'. 2205 */ 2206 struct queue_limits *dm_get_queue_limits(struct mapped_device *md) 2207 { 2208 BUG_ON(!atomic_read(&md->holders)); 2209 return &md->queue->limits; 2210 } 2211 EXPORT_SYMBOL_GPL(dm_get_queue_limits); 2212 2213 /* 2214 * Setup the DM device's queue based on md's type 2215 */ 2216 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) 2217 { 2218 int r; 2219 struct queue_limits limits; 2220 enum dm_queue_mode type = dm_get_md_type(md); 2221 2222 switch (type) { 2223 case DM_TYPE_REQUEST_BASED: 2224 r = dm_mq_init_request_queue(md, t); 2225 if (r) { 2226 DMERR("Cannot initialize queue for request-based dm-mq mapped device"); 2227 return r; 2228 } 2229 break; 2230 case DM_TYPE_BIO_BASED: 2231 case DM_TYPE_DAX_BIO_BASED: 2232 case DM_TYPE_NVME_BIO_BASED: 2233 break; 2234 case DM_TYPE_NONE: 2235 WARN_ON_ONCE(true); 2236 break; 2237 } 2238 2239 r = dm_calculate_queue_limits(t, &limits); 2240 if (r) { 2241 DMERR("Cannot calculate initial queue limits"); 2242 return r; 2243 } 2244 dm_table_set_restrictions(t, md->queue, &limits); 2245 blk_register_queue(md->disk); 2246 2247 return 0; 2248 } 2249 2250 struct mapped_device *dm_get_md(dev_t dev) 2251 { 2252 struct mapped_device *md; 2253 unsigned minor = MINOR(dev); 2254 2255 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) 2256 return NULL; 2257 2258 spin_lock(&_minor_lock); 2259 2260 md = idr_find(&_minor_idr, minor); 2261 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || 2262 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { 2263 md = NULL; 2264 goto out; 2265 } 2266 dm_get(md); 2267 out: 2268 spin_unlock(&_minor_lock); 2269 2270 return md; 2271 } 2272 EXPORT_SYMBOL_GPL(dm_get_md); 2273 2274 void *dm_get_mdptr(struct mapped_device *md) 2275 { 2276 return md->interface_ptr; 2277 } 2278 2279 void dm_set_mdptr(struct mapped_device *md, void *ptr) 2280 { 2281 md->interface_ptr = ptr; 2282 } 2283 2284 void dm_get(struct mapped_device *md) 2285 { 2286 atomic_inc(&md->holders); 2287 BUG_ON(test_bit(DMF_FREEING, &md->flags)); 2288 } 2289 2290 int dm_hold(struct mapped_device *md) 2291 { 2292 spin_lock(&_minor_lock); 2293 if (test_bit(DMF_FREEING, &md->flags)) { 2294 spin_unlock(&_minor_lock); 2295 return -EBUSY; 2296 } 2297 dm_get(md); 2298 spin_unlock(&_minor_lock); 2299 return 0; 2300 } 2301 EXPORT_SYMBOL_GPL(dm_hold); 2302 2303 const char *dm_device_name(struct mapped_device *md) 2304 { 2305 return md->name; 2306 } 2307 EXPORT_SYMBOL_GPL(dm_device_name); 2308 2309 static void __dm_destroy(struct mapped_device *md, bool wait) 2310 { 2311 struct dm_table *map; 2312 int srcu_idx; 2313 2314 might_sleep(); 2315 2316 spin_lock(&_minor_lock); 2317 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); 2318 set_bit(DMF_FREEING, &md->flags); 2319 spin_unlock(&_minor_lock); 2320 2321 blk_set_queue_dying(md->queue); 2322 2323 /* 2324 * Take suspend_lock so that presuspend and postsuspend methods 2325 * do not race with internal suspend. 2326 */ 2327 mutex_lock(&md->suspend_lock); 2328 map = dm_get_live_table(md, &srcu_idx); 2329 if (!dm_suspended_md(md)) { 2330 dm_table_presuspend_targets(map); 2331 set_bit(DMF_SUSPENDED, &md->flags); 2332 set_bit(DMF_POST_SUSPENDING, &md->flags); 2333 dm_table_postsuspend_targets(map); 2334 } 2335 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ 2336 dm_put_live_table(md, srcu_idx); 2337 mutex_unlock(&md->suspend_lock); 2338 2339 /* 2340 * Rare, but there may be I/O requests still going to complete, 2341 * for example. Wait for all references to disappear. 2342 * No one should increment the reference count of the mapped_device, 2343 * after the mapped_device state becomes DMF_FREEING. 2344 */ 2345 if (wait) 2346 while (atomic_read(&md->holders)) 2347 msleep(1); 2348 else if (atomic_read(&md->holders)) 2349 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", 2350 dm_device_name(md), atomic_read(&md->holders)); 2351 2352 dm_sysfs_exit(md); 2353 dm_table_destroy(__unbind(md)); 2354 free_dev(md); 2355 } 2356 2357 void dm_destroy(struct mapped_device *md) 2358 { 2359 __dm_destroy(md, true); 2360 } 2361 2362 void dm_destroy_immediate(struct mapped_device *md) 2363 { 2364 __dm_destroy(md, false); 2365 } 2366 2367 void dm_put(struct mapped_device *md) 2368 { 2369 atomic_dec(&md->holders); 2370 } 2371 EXPORT_SYMBOL_GPL(dm_put); 2372 2373 static bool md_in_flight_bios(struct mapped_device *md) 2374 { 2375 int cpu; 2376 struct hd_struct *part = &dm_disk(md)->part0; 2377 long sum = 0; 2378 2379 for_each_possible_cpu(cpu) { 2380 sum += part_stat_local_read_cpu(part, in_flight[0], cpu); 2381 sum += part_stat_local_read_cpu(part, in_flight[1], cpu); 2382 } 2383 2384 return sum != 0; 2385 } 2386 2387 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state) 2388 { 2389 int r = 0; 2390 DEFINE_WAIT(wait); 2391 2392 while (true) { 2393 prepare_to_wait(&md->wait, &wait, task_state); 2394 2395 if (!md_in_flight_bios(md)) 2396 break; 2397 2398 if (signal_pending_state(task_state, current)) { 2399 r = -EINTR; 2400 break; 2401 } 2402 2403 io_schedule(); 2404 } 2405 finish_wait(&md->wait, &wait); 2406 2407 return r; 2408 } 2409 2410 static int dm_wait_for_completion(struct mapped_device *md, long task_state) 2411 { 2412 int r = 0; 2413 2414 if (!queue_is_mq(md->queue)) 2415 return dm_wait_for_bios_completion(md, task_state); 2416 2417 while (true) { 2418 if (!blk_mq_queue_inflight(md->queue)) 2419 break; 2420 2421 if (signal_pending_state(task_state, current)) { 2422 r = -EINTR; 2423 break; 2424 } 2425 2426 msleep(5); 2427 } 2428 2429 return r; 2430 } 2431 2432 /* 2433 * Process the deferred bios 2434 */ 2435 static void dm_wq_work(struct work_struct *work) 2436 { 2437 struct mapped_device *md = container_of(work, struct mapped_device, 2438 work); 2439 struct bio *c; 2440 int srcu_idx; 2441 struct dm_table *map; 2442 2443 map = dm_get_live_table(md, &srcu_idx); 2444 2445 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 2446 spin_lock_irq(&md->deferred_lock); 2447 c = bio_list_pop(&md->deferred); 2448 spin_unlock_irq(&md->deferred_lock); 2449 2450 if (!c) 2451 break; 2452 2453 if (dm_request_based(md)) 2454 (void) submit_bio_noacct(c); 2455 else 2456 (void) dm_process_bio(md, map, c); 2457 } 2458 2459 dm_put_live_table(md, srcu_idx); 2460 } 2461 2462 static void dm_queue_flush(struct mapped_device *md) 2463 { 2464 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2465 smp_mb__after_atomic(); 2466 queue_work(md->wq, &md->work); 2467 } 2468 2469 /* 2470 * Swap in a new table, returning the old one for the caller to destroy. 2471 */ 2472 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) 2473 { 2474 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); 2475 struct queue_limits limits; 2476 int r; 2477 2478 mutex_lock(&md->suspend_lock); 2479 2480 /* device must be suspended */ 2481 if (!dm_suspended_md(md)) 2482 goto out; 2483 2484 /* 2485 * If the new table has no data devices, retain the existing limits. 2486 * This helps multipath with queue_if_no_path if all paths disappear, 2487 * then new I/O is queued based on these limits, and then some paths 2488 * reappear. 2489 */ 2490 if (dm_table_has_no_data_devices(table)) { 2491 live_map = dm_get_live_table_fast(md); 2492 if (live_map) 2493 limits = md->queue->limits; 2494 dm_put_live_table_fast(md); 2495 } 2496 2497 if (!live_map) { 2498 r = dm_calculate_queue_limits(table, &limits); 2499 if (r) { 2500 map = ERR_PTR(r); 2501 goto out; 2502 } 2503 } 2504 2505 map = __bind(md, table, &limits); 2506 dm_issue_global_event(); 2507 2508 out: 2509 mutex_unlock(&md->suspend_lock); 2510 return map; 2511 } 2512 2513 /* 2514 * Functions to lock and unlock any filesystem running on the 2515 * device. 2516 */ 2517 static int lock_fs(struct mapped_device *md) 2518 { 2519 int r; 2520 2521 WARN_ON(md->frozen_sb); 2522 2523 md->frozen_sb = freeze_bdev(md->bdev); 2524 if (IS_ERR(md->frozen_sb)) { 2525 r = PTR_ERR(md->frozen_sb); 2526 md->frozen_sb = NULL; 2527 return r; 2528 } 2529 2530 set_bit(DMF_FROZEN, &md->flags); 2531 2532 return 0; 2533 } 2534 2535 static void unlock_fs(struct mapped_device *md) 2536 { 2537 if (!test_bit(DMF_FROZEN, &md->flags)) 2538 return; 2539 2540 thaw_bdev(md->bdev, md->frozen_sb); 2541 md->frozen_sb = NULL; 2542 clear_bit(DMF_FROZEN, &md->flags); 2543 } 2544 2545 /* 2546 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG 2547 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE 2548 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY 2549 * 2550 * If __dm_suspend returns 0, the device is completely quiescent 2551 * now. There is no request-processing activity. All new requests 2552 * are being added to md->deferred list. 2553 */ 2554 static int __dm_suspend(struct mapped_device *md, struct dm_table *map, 2555 unsigned suspend_flags, long task_state, 2556 int dmf_suspended_flag) 2557 { 2558 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; 2559 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; 2560 int r; 2561 2562 lockdep_assert_held(&md->suspend_lock); 2563 2564 /* 2565 * DMF_NOFLUSH_SUSPENDING must be set before presuspend. 2566 * This flag is cleared before dm_suspend returns. 2567 */ 2568 if (noflush) 2569 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 2570 else 2571 DMDEBUG("%s: suspending with flush", dm_device_name(md)); 2572 2573 /* 2574 * This gets reverted if there's an error later and the targets 2575 * provide the .presuspend_undo hook. 2576 */ 2577 dm_table_presuspend_targets(map); 2578 2579 /* 2580 * Flush I/O to the device. 2581 * Any I/O submitted after lock_fs() may not be flushed. 2582 * noflush takes precedence over do_lockfs. 2583 * (lock_fs() flushes I/Os and waits for them to complete.) 2584 */ 2585 if (!noflush && do_lockfs) { 2586 r = lock_fs(md); 2587 if (r) { 2588 dm_table_presuspend_undo_targets(map); 2589 return r; 2590 } 2591 } 2592 2593 /* 2594 * Here we must make sure that no processes are submitting requests 2595 * to target drivers i.e. no one may be executing 2596 * __split_and_process_bio. This is called from dm_request and 2597 * dm_wq_work. 2598 * 2599 * To get all processes out of __split_and_process_bio in dm_request, 2600 * we take the write lock. To prevent any process from reentering 2601 * __split_and_process_bio from dm_request and quiesce the thread 2602 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call 2603 * flush_workqueue(md->wq). 2604 */ 2605 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2606 if (map) 2607 synchronize_srcu(&md->io_barrier); 2608 2609 /* 2610 * Stop md->queue before flushing md->wq in case request-based 2611 * dm defers requests to md->wq from md->queue. 2612 */ 2613 if (dm_request_based(md)) 2614 dm_stop_queue(md->queue); 2615 2616 flush_workqueue(md->wq); 2617 2618 /* 2619 * At this point no more requests are entering target request routines. 2620 * We call dm_wait_for_completion to wait for all existing requests 2621 * to finish. 2622 */ 2623 r = dm_wait_for_completion(md, task_state); 2624 if (!r) 2625 set_bit(dmf_suspended_flag, &md->flags); 2626 2627 if (noflush) 2628 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 2629 if (map) 2630 synchronize_srcu(&md->io_barrier); 2631 2632 /* were we interrupted ? */ 2633 if (r < 0) { 2634 dm_queue_flush(md); 2635 2636 if (dm_request_based(md)) 2637 dm_start_queue(md->queue); 2638 2639 unlock_fs(md); 2640 dm_table_presuspend_undo_targets(map); 2641 /* pushback list is already flushed, so skip flush */ 2642 } 2643 2644 return r; 2645 } 2646 2647 /* 2648 * We need to be able to change a mapping table under a mounted 2649 * filesystem. For example we might want to move some data in 2650 * the background. Before the table can be swapped with 2651 * dm_bind_table, dm_suspend must be called to flush any in 2652 * flight bios and ensure that any further io gets deferred. 2653 */ 2654 /* 2655 * Suspend mechanism in request-based dm. 2656 * 2657 * 1. Flush all I/Os by lock_fs() if needed. 2658 * 2. Stop dispatching any I/O by stopping the request_queue. 2659 * 3. Wait for all in-flight I/Os to be completed or requeued. 2660 * 2661 * To abort suspend, start the request_queue. 2662 */ 2663 int dm_suspend(struct mapped_device *md, unsigned suspend_flags) 2664 { 2665 struct dm_table *map = NULL; 2666 int r = 0; 2667 2668 retry: 2669 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); 2670 2671 if (dm_suspended_md(md)) { 2672 r = -EINVAL; 2673 goto out_unlock; 2674 } 2675 2676 if (dm_suspended_internally_md(md)) { 2677 /* already internally suspended, wait for internal resume */ 2678 mutex_unlock(&md->suspend_lock); 2679 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); 2680 if (r) 2681 return r; 2682 goto retry; 2683 } 2684 2685 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 2686 2687 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); 2688 if (r) 2689 goto out_unlock; 2690 2691 set_bit(DMF_POST_SUSPENDING, &md->flags); 2692 dm_table_postsuspend_targets(map); 2693 clear_bit(DMF_POST_SUSPENDING, &md->flags); 2694 2695 out_unlock: 2696 mutex_unlock(&md->suspend_lock); 2697 return r; 2698 } 2699 2700 static int __dm_resume(struct mapped_device *md, struct dm_table *map) 2701 { 2702 if (map) { 2703 int r = dm_table_resume_targets(map); 2704 if (r) 2705 return r; 2706 } 2707 2708 dm_queue_flush(md); 2709 2710 /* 2711 * Flushing deferred I/Os must be done after targets are resumed 2712 * so that mapping of targets can work correctly. 2713 * Request-based dm is queueing the deferred I/Os in its request_queue. 2714 */ 2715 if (dm_request_based(md)) 2716 dm_start_queue(md->queue); 2717 2718 unlock_fs(md); 2719 2720 return 0; 2721 } 2722 2723 int dm_resume(struct mapped_device *md) 2724 { 2725 int r; 2726 struct dm_table *map = NULL; 2727 2728 retry: 2729 r = -EINVAL; 2730 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); 2731 2732 if (!dm_suspended_md(md)) 2733 goto out; 2734 2735 if (dm_suspended_internally_md(md)) { 2736 /* already internally suspended, wait for internal resume */ 2737 mutex_unlock(&md->suspend_lock); 2738 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); 2739 if (r) 2740 return r; 2741 goto retry; 2742 } 2743 2744 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 2745 if (!map || !dm_table_get_size(map)) 2746 goto out; 2747 2748 r = __dm_resume(md, map); 2749 if (r) 2750 goto out; 2751 2752 clear_bit(DMF_SUSPENDED, &md->flags); 2753 out: 2754 mutex_unlock(&md->suspend_lock); 2755 2756 return r; 2757 } 2758 2759 /* 2760 * Internal suspend/resume works like userspace-driven suspend. It waits 2761 * until all bios finish and prevents issuing new bios to the target drivers. 2762 * It may be used only from the kernel. 2763 */ 2764 2765 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) 2766 { 2767 struct dm_table *map = NULL; 2768 2769 lockdep_assert_held(&md->suspend_lock); 2770 2771 if (md->internal_suspend_count++) 2772 return; /* nested internal suspend */ 2773 2774 if (dm_suspended_md(md)) { 2775 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); 2776 return; /* nest suspend */ 2777 } 2778 2779 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 2780 2781 /* 2782 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is 2783 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend 2784 * would require changing .presuspend to return an error -- avoid this 2785 * until there is a need for more elaborate variants of internal suspend. 2786 */ 2787 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, 2788 DMF_SUSPENDED_INTERNALLY); 2789 2790 set_bit(DMF_POST_SUSPENDING, &md->flags); 2791 dm_table_postsuspend_targets(map); 2792 clear_bit(DMF_POST_SUSPENDING, &md->flags); 2793 } 2794 2795 static void __dm_internal_resume(struct mapped_device *md) 2796 { 2797 BUG_ON(!md->internal_suspend_count); 2798 2799 if (--md->internal_suspend_count) 2800 return; /* resume from nested internal suspend */ 2801 2802 if (dm_suspended_md(md)) 2803 goto done; /* resume from nested suspend */ 2804 2805 /* 2806 * NOTE: existing callers don't need to call dm_table_resume_targets 2807 * (which may fail -- so best to avoid it for now by passing NULL map) 2808 */ 2809 (void) __dm_resume(md, NULL); 2810 2811 done: 2812 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); 2813 smp_mb__after_atomic(); 2814 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); 2815 } 2816 2817 void dm_internal_suspend_noflush(struct mapped_device *md) 2818 { 2819 mutex_lock(&md->suspend_lock); 2820 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); 2821 mutex_unlock(&md->suspend_lock); 2822 } 2823 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); 2824 2825 void dm_internal_resume(struct mapped_device *md) 2826 { 2827 mutex_lock(&md->suspend_lock); 2828 __dm_internal_resume(md); 2829 mutex_unlock(&md->suspend_lock); 2830 } 2831 EXPORT_SYMBOL_GPL(dm_internal_resume); 2832 2833 /* 2834 * Fast variants of internal suspend/resume hold md->suspend_lock, 2835 * which prevents interaction with userspace-driven suspend. 2836 */ 2837 2838 void dm_internal_suspend_fast(struct mapped_device *md) 2839 { 2840 mutex_lock(&md->suspend_lock); 2841 if (dm_suspended_md(md) || dm_suspended_internally_md(md)) 2842 return; 2843 2844 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2845 synchronize_srcu(&md->io_barrier); 2846 flush_workqueue(md->wq); 2847 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); 2848 } 2849 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); 2850 2851 void dm_internal_resume_fast(struct mapped_device *md) 2852 { 2853 if (dm_suspended_md(md) || dm_suspended_internally_md(md)) 2854 goto done; 2855 2856 dm_queue_flush(md); 2857 2858 done: 2859 mutex_unlock(&md->suspend_lock); 2860 } 2861 EXPORT_SYMBOL_GPL(dm_internal_resume_fast); 2862 2863 /*----------------------------------------------------------------- 2864 * Event notification. 2865 *---------------------------------------------------------------*/ 2866 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, 2867 unsigned cookie) 2868 { 2869 int r; 2870 unsigned noio_flag; 2871 char udev_cookie[DM_COOKIE_LENGTH]; 2872 char *envp[] = { udev_cookie, NULL }; 2873 2874 noio_flag = memalloc_noio_save(); 2875 2876 if (!cookie) 2877 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action); 2878 else { 2879 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", 2880 DM_COOKIE_ENV_VAR_NAME, cookie); 2881 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, 2882 action, envp); 2883 } 2884 2885 memalloc_noio_restore(noio_flag); 2886 2887 return r; 2888 } 2889 2890 uint32_t dm_next_uevent_seq(struct mapped_device *md) 2891 { 2892 return atomic_add_return(1, &md->uevent_seq); 2893 } 2894 2895 uint32_t dm_get_event_nr(struct mapped_device *md) 2896 { 2897 return atomic_read(&md->event_nr); 2898 } 2899 2900 int dm_wait_event(struct mapped_device *md, int event_nr) 2901 { 2902 return wait_event_interruptible(md->eventq, 2903 (event_nr != atomic_read(&md->event_nr))); 2904 } 2905 2906 void dm_uevent_add(struct mapped_device *md, struct list_head *elist) 2907 { 2908 unsigned long flags; 2909 2910 spin_lock_irqsave(&md->uevent_lock, flags); 2911 list_add(elist, &md->uevent_list); 2912 spin_unlock_irqrestore(&md->uevent_lock, flags); 2913 } 2914 2915 /* 2916 * The gendisk is only valid as long as you have a reference 2917 * count on 'md'. 2918 */ 2919 struct gendisk *dm_disk(struct mapped_device *md) 2920 { 2921 return md->disk; 2922 } 2923 EXPORT_SYMBOL_GPL(dm_disk); 2924 2925 struct kobject *dm_kobject(struct mapped_device *md) 2926 { 2927 return &md->kobj_holder.kobj; 2928 } 2929 2930 struct mapped_device *dm_get_from_kobject(struct kobject *kobj) 2931 { 2932 struct mapped_device *md; 2933 2934 md = container_of(kobj, struct mapped_device, kobj_holder.kobj); 2935 2936 spin_lock(&_minor_lock); 2937 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { 2938 md = NULL; 2939 goto out; 2940 } 2941 dm_get(md); 2942 out: 2943 spin_unlock(&_minor_lock); 2944 2945 return md; 2946 } 2947 2948 int dm_suspended_md(struct mapped_device *md) 2949 { 2950 return test_bit(DMF_SUSPENDED, &md->flags); 2951 } 2952 2953 static int dm_post_suspending_md(struct mapped_device *md) 2954 { 2955 return test_bit(DMF_POST_SUSPENDING, &md->flags); 2956 } 2957 2958 int dm_suspended_internally_md(struct mapped_device *md) 2959 { 2960 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); 2961 } 2962 2963 int dm_test_deferred_remove_flag(struct mapped_device *md) 2964 { 2965 return test_bit(DMF_DEFERRED_REMOVE, &md->flags); 2966 } 2967 2968 int dm_suspended(struct dm_target *ti) 2969 { 2970 return dm_suspended_md(dm_table_get_md(ti->table)); 2971 } 2972 EXPORT_SYMBOL_GPL(dm_suspended); 2973 2974 int dm_post_suspending(struct dm_target *ti) 2975 { 2976 return dm_post_suspending_md(dm_table_get_md(ti->table)); 2977 } 2978 EXPORT_SYMBOL_GPL(dm_post_suspending); 2979 2980 int dm_noflush_suspending(struct dm_target *ti) 2981 { 2982 return __noflush_suspending(dm_table_get_md(ti->table)); 2983 } 2984 EXPORT_SYMBOL_GPL(dm_noflush_suspending); 2985 2986 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, 2987 unsigned integrity, unsigned per_io_data_size, 2988 unsigned min_pool_size) 2989 { 2990 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); 2991 unsigned int pool_size = 0; 2992 unsigned int front_pad, io_front_pad; 2993 int ret; 2994 2995 if (!pools) 2996 return NULL; 2997 2998 switch (type) { 2999 case DM_TYPE_BIO_BASED: 3000 case DM_TYPE_DAX_BIO_BASED: 3001 case DM_TYPE_NVME_BIO_BASED: 3002 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); 3003 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); 3004 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); 3005 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0); 3006 if (ret) 3007 goto out; 3008 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size)) 3009 goto out; 3010 break; 3011 case DM_TYPE_REQUEST_BASED: 3012 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); 3013 front_pad = offsetof(struct dm_rq_clone_bio_info, clone); 3014 /* per_io_data_size is used for blk-mq pdu at queue allocation */ 3015 break; 3016 default: 3017 BUG(); 3018 } 3019 3020 ret = bioset_init(&pools->bs, pool_size, front_pad, 0); 3021 if (ret) 3022 goto out; 3023 3024 if (integrity && bioset_integrity_create(&pools->bs, pool_size)) 3025 goto out; 3026 3027 return pools; 3028 3029 out: 3030 dm_free_md_mempools(pools); 3031 3032 return NULL; 3033 } 3034 3035 void dm_free_md_mempools(struct dm_md_mempools *pools) 3036 { 3037 if (!pools) 3038 return; 3039 3040 bioset_exit(&pools->bs); 3041 bioset_exit(&pools->io_bs); 3042 3043 kfree(pools); 3044 } 3045 3046 struct dm_pr { 3047 u64 old_key; 3048 u64 new_key; 3049 u32 flags; 3050 bool fail_early; 3051 }; 3052 3053 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, 3054 void *data) 3055 { 3056 struct mapped_device *md = bdev->bd_disk->private_data; 3057 struct dm_table *table; 3058 struct dm_target *ti; 3059 int ret = -ENOTTY, srcu_idx; 3060 3061 table = dm_get_live_table(md, &srcu_idx); 3062 if (!table || !dm_table_get_size(table)) 3063 goto out; 3064 3065 /* We only support devices that have a single target */ 3066 if (dm_table_get_num_targets(table) != 1) 3067 goto out; 3068 ti = dm_table_get_target(table, 0); 3069 3070 ret = -EINVAL; 3071 if (!ti->type->iterate_devices) 3072 goto out; 3073 3074 ret = ti->type->iterate_devices(ti, fn, data); 3075 out: 3076 dm_put_live_table(md, srcu_idx); 3077 return ret; 3078 } 3079 3080 /* 3081 * For register / unregister we need to manually call out to every path. 3082 */ 3083 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, 3084 sector_t start, sector_t len, void *data) 3085 { 3086 struct dm_pr *pr = data; 3087 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; 3088 3089 if (!ops || !ops->pr_register) 3090 return -EOPNOTSUPP; 3091 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); 3092 } 3093 3094 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, 3095 u32 flags) 3096 { 3097 struct dm_pr pr = { 3098 .old_key = old_key, 3099 .new_key = new_key, 3100 .flags = flags, 3101 .fail_early = true, 3102 }; 3103 int ret; 3104 3105 ret = dm_call_pr(bdev, __dm_pr_register, &pr); 3106 if (ret && new_key) { 3107 /* unregister all paths if we failed to register any path */ 3108 pr.old_key = new_key; 3109 pr.new_key = 0; 3110 pr.flags = 0; 3111 pr.fail_early = false; 3112 dm_call_pr(bdev, __dm_pr_register, &pr); 3113 } 3114 3115 return ret; 3116 } 3117 3118 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, 3119 u32 flags) 3120 { 3121 struct mapped_device *md = bdev->bd_disk->private_data; 3122 const struct pr_ops *ops; 3123 int r, srcu_idx; 3124 3125 r = dm_prepare_ioctl(md, &srcu_idx, &bdev); 3126 if (r < 0) 3127 goto out; 3128 3129 ops = bdev->bd_disk->fops->pr_ops; 3130 if (ops && ops->pr_reserve) 3131 r = ops->pr_reserve(bdev, key, type, flags); 3132 else 3133 r = -EOPNOTSUPP; 3134 out: 3135 dm_unprepare_ioctl(md, srcu_idx); 3136 return r; 3137 } 3138 3139 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) 3140 { 3141 struct mapped_device *md = bdev->bd_disk->private_data; 3142 const struct pr_ops *ops; 3143 int r, srcu_idx; 3144 3145 r = dm_prepare_ioctl(md, &srcu_idx, &bdev); 3146 if (r < 0) 3147 goto out; 3148 3149 ops = bdev->bd_disk->fops->pr_ops; 3150 if (ops && ops->pr_release) 3151 r = ops->pr_release(bdev, key, type); 3152 else 3153 r = -EOPNOTSUPP; 3154 out: 3155 dm_unprepare_ioctl(md, srcu_idx); 3156 return r; 3157 } 3158 3159 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, 3160 enum pr_type type, bool abort) 3161 { 3162 struct mapped_device *md = bdev->bd_disk->private_data; 3163 const struct pr_ops *ops; 3164 int r, srcu_idx; 3165 3166 r = dm_prepare_ioctl(md, &srcu_idx, &bdev); 3167 if (r < 0) 3168 goto out; 3169 3170 ops = bdev->bd_disk->fops->pr_ops; 3171 if (ops && ops->pr_preempt) 3172 r = ops->pr_preempt(bdev, old_key, new_key, type, abort); 3173 else 3174 r = -EOPNOTSUPP; 3175 out: 3176 dm_unprepare_ioctl(md, srcu_idx); 3177 return r; 3178 } 3179 3180 static int dm_pr_clear(struct block_device *bdev, u64 key) 3181 { 3182 struct mapped_device *md = bdev->bd_disk->private_data; 3183 const struct pr_ops *ops; 3184 int r, srcu_idx; 3185 3186 r = dm_prepare_ioctl(md, &srcu_idx, &bdev); 3187 if (r < 0) 3188 goto out; 3189 3190 ops = bdev->bd_disk->fops->pr_ops; 3191 if (ops && ops->pr_clear) 3192 r = ops->pr_clear(bdev, key); 3193 else 3194 r = -EOPNOTSUPP; 3195 out: 3196 dm_unprepare_ioctl(md, srcu_idx); 3197 return r; 3198 } 3199 3200 static const struct pr_ops dm_pr_ops = { 3201 .pr_register = dm_pr_register, 3202 .pr_reserve = dm_pr_reserve, 3203 .pr_release = dm_pr_release, 3204 .pr_preempt = dm_pr_preempt, 3205 .pr_clear = dm_pr_clear, 3206 }; 3207 3208 static const struct block_device_operations dm_blk_dops = { 3209 .submit_bio = dm_submit_bio, 3210 .open = dm_blk_open, 3211 .release = dm_blk_close, 3212 .ioctl = dm_blk_ioctl, 3213 .getgeo = dm_blk_getgeo, 3214 .report_zones = dm_blk_report_zones, 3215 .pr_ops = &dm_pr_ops, 3216 .owner = THIS_MODULE 3217 }; 3218 3219 static const struct dax_operations dm_dax_ops = { 3220 .direct_access = dm_dax_direct_access, 3221 .dax_supported = dm_dax_supported, 3222 .copy_from_iter = dm_dax_copy_from_iter, 3223 .copy_to_iter = dm_dax_copy_to_iter, 3224 .zero_page_range = dm_dax_zero_page_range, 3225 }; 3226 3227 /* 3228 * module hooks 3229 */ 3230 module_init(dm_init); 3231 module_exit(dm_exit); 3232 3233 module_param(major, uint, 0); 3234 MODULE_PARM_DESC(major, "The major number of the device mapper"); 3235 3236 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); 3237 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); 3238 3239 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); 3240 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); 3241 3242 MODULE_DESCRIPTION(DM_NAME " driver"); 3243 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); 3244 MODULE_LICENSE("GPL"); 3245