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