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/signal.h> 16 #include <linux/blkpg.h> 17 #include <linux/bio.h> 18 #include <linux/mempool.h> 19 #include <linux/slab.h> 20 #include <linux/idr.h> 21 #include <linux/hdreg.h> 22 #include <linux/delay.h> 23 #include <linux/wait.h> 24 #include <linux/pr.h> 25 26 #define DM_MSG_PREFIX "core" 27 28 #ifdef CONFIG_PRINTK 29 /* 30 * ratelimit state to be used in DMXXX_LIMIT(). 31 */ 32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state, 33 DEFAULT_RATELIMIT_INTERVAL, 34 DEFAULT_RATELIMIT_BURST); 35 EXPORT_SYMBOL(dm_ratelimit_state); 36 #endif 37 38 /* 39 * Cookies are numeric values sent with CHANGE and REMOVE 40 * uevents while resuming, removing or renaming the device. 41 */ 42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" 43 #define DM_COOKIE_LENGTH 24 44 45 static const char *_name = DM_NAME; 46 47 static unsigned int major = 0; 48 static unsigned int _major = 0; 49 50 static DEFINE_IDR(_minor_idr); 51 52 static DEFINE_SPINLOCK(_minor_lock); 53 54 static void do_deferred_remove(struct work_struct *w); 55 56 static DECLARE_WORK(deferred_remove_work, do_deferred_remove); 57 58 static struct workqueue_struct *deferred_remove_workqueue; 59 60 /* 61 * One of these is allocated per bio. 62 */ 63 struct dm_io { 64 struct mapped_device *md; 65 int error; 66 atomic_t io_count; 67 struct bio *bio; 68 unsigned long start_time; 69 spinlock_t endio_lock; 70 struct dm_stats_aux stats_aux; 71 }; 72 73 #define MINOR_ALLOCED ((void *)-1) 74 75 /* 76 * Bits for the md->flags field. 77 */ 78 #define DMF_BLOCK_IO_FOR_SUSPEND 0 79 #define DMF_SUSPENDED 1 80 #define DMF_FROZEN 2 81 #define DMF_FREEING 3 82 #define DMF_DELETING 4 83 #define DMF_NOFLUSH_SUSPENDING 5 84 #define DMF_DEFERRED_REMOVE 6 85 #define DMF_SUSPENDED_INTERNALLY 7 86 87 #define DM_NUMA_NODE NUMA_NO_NODE 88 static int dm_numa_node = DM_NUMA_NODE; 89 90 /* 91 * For mempools pre-allocation at the table loading time. 92 */ 93 struct dm_md_mempools { 94 mempool_t *io_pool; 95 struct bio_set *bs; 96 }; 97 98 struct table_device { 99 struct list_head list; 100 atomic_t count; 101 struct dm_dev dm_dev; 102 }; 103 104 static struct kmem_cache *_io_cache; 105 static struct kmem_cache *_rq_tio_cache; 106 static struct kmem_cache *_rq_cache; 107 108 /* 109 * Bio-based DM's mempools' reserved IOs set by the user. 110 */ 111 #define RESERVED_BIO_BASED_IOS 16 112 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; 113 114 static int __dm_get_module_param_int(int *module_param, int min, int max) 115 { 116 int param = ACCESS_ONCE(*module_param); 117 int modified_param = 0; 118 bool modified = true; 119 120 if (param < min) 121 modified_param = min; 122 else if (param > max) 123 modified_param = max; 124 else 125 modified = false; 126 127 if (modified) { 128 (void)cmpxchg(module_param, param, modified_param); 129 param = modified_param; 130 } 131 132 return param; 133 } 134 135 unsigned __dm_get_module_param(unsigned *module_param, 136 unsigned def, unsigned max) 137 { 138 unsigned param = ACCESS_ONCE(*module_param); 139 unsigned modified_param = 0; 140 141 if (!param) 142 modified_param = def; 143 else if (param > max) 144 modified_param = max; 145 146 if (modified_param) { 147 (void)cmpxchg(module_param, param, modified_param); 148 param = modified_param; 149 } 150 151 return param; 152 } 153 154 unsigned dm_get_reserved_bio_based_ios(void) 155 { 156 return __dm_get_module_param(&reserved_bio_based_ios, 157 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); 158 } 159 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); 160 161 static unsigned dm_get_numa_node(void) 162 { 163 return __dm_get_module_param_int(&dm_numa_node, 164 DM_NUMA_NODE, num_online_nodes() - 1); 165 } 166 167 static int __init local_init(void) 168 { 169 int r = -ENOMEM; 170 171 /* allocate a slab for the dm_ios */ 172 _io_cache = KMEM_CACHE(dm_io, 0); 173 if (!_io_cache) 174 return r; 175 176 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); 177 if (!_rq_tio_cache) 178 goto out_free_io_cache; 179 180 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), 181 __alignof__(struct request), 0, NULL); 182 if (!_rq_cache) 183 goto out_free_rq_tio_cache; 184 185 r = dm_uevent_init(); 186 if (r) 187 goto out_free_rq_cache; 188 189 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); 190 if (!deferred_remove_workqueue) { 191 r = -ENOMEM; 192 goto out_uevent_exit; 193 } 194 195 _major = major; 196 r = register_blkdev(_major, _name); 197 if (r < 0) 198 goto out_free_workqueue; 199 200 if (!_major) 201 _major = r; 202 203 return 0; 204 205 out_free_workqueue: 206 destroy_workqueue(deferred_remove_workqueue); 207 out_uevent_exit: 208 dm_uevent_exit(); 209 out_free_rq_cache: 210 kmem_cache_destroy(_rq_cache); 211 out_free_rq_tio_cache: 212 kmem_cache_destroy(_rq_tio_cache); 213 out_free_io_cache: 214 kmem_cache_destroy(_io_cache); 215 216 return r; 217 } 218 219 static void local_exit(void) 220 { 221 flush_scheduled_work(); 222 destroy_workqueue(deferred_remove_workqueue); 223 224 kmem_cache_destroy(_rq_cache); 225 kmem_cache_destroy(_rq_tio_cache); 226 kmem_cache_destroy(_io_cache); 227 unregister_blkdev(_major, _name); 228 dm_uevent_exit(); 229 230 _major = 0; 231 232 DMINFO("cleaned up"); 233 } 234 235 static int (*_inits[])(void) __initdata = { 236 local_init, 237 dm_target_init, 238 dm_linear_init, 239 dm_stripe_init, 240 dm_io_init, 241 dm_kcopyd_init, 242 dm_interface_init, 243 dm_statistics_init, 244 }; 245 246 static void (*_exits[])(void) = { 247 local_exit, 248 dm_target_exit, 249 dm_linear_exit, 250 dm_stripe_exit, 251 dm_io_exit, 252 dm_kcopyd_exit, 253 dm_interface_exit, 254 dm_statistics_exit, 255 }; 256 257 static int __init dm_init(void) 258 { 259 const int count = ARRAY_SIZE(_inits); 260 261 int r, i; 262 263 for (i = 0; i < count; i++) { 264 r = _inits[i](); 265 if (r) 266 goto bad; 267 } 268 269 return 0; 270 271 bad: 272 while (i--) 273 _exits[i](); 274 275 return r; 276 } 277 278 static void __exit dm_exit(void) 279 { 280 int i = ARRAY_SIZE(_exits); 281 282 while (i--) 283 _exits[i](); 284 285 /* 286 * Should be empty by this point. 287 */ 288 idr_destroy(&_minor_idr); 289 } 290 291 /* 292 * Block device functions 293 */ 294 int dm_deleting_md(struct mapped_device *md) 295 { 296 return test_bit(DMF_DELETING, &md->flags); 297 } 298 299 static int dm_blk_open(struct block_device *bdev, fmode_t mode) 300 { 301 struct mapped_device *md; 302 303 spin_lock(&_minor_lock); 304 305 md = bdev->bd_disk->private_data; 306 if (!md) 307 goto out; 308 309 if (test_bit(DMF_FREEING, &md->flags) || 310 dm_deleting_md(md)) { 311 md = NULL; 312 goto out; 313 } 314 315 dm_get(md); 316 atomic_inc(&md->open_count); 317 out: 318 spin_unlock(&_minor_lock); 319 320 return md ? 0 : -ENXIO; 321 } 322 323 static void dm_blk_close(struct gendisk *disk, fmode_t mode) 324 { 325 struct mapped_device *md; 326 327 spin_lock(&_minor_lock); 328 329 md = disk->private_data; 330 if (WARN_ON(!md)) 331 goto out; 332 333 if (atomic_dec_and_test(&md->open_count) && 334 (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) 335 queue_work(deferred_remove_workqueue, &deferred_remove_work); 336 337 dm_put(md); 338 out: 339 spin_unlock(&_minor_lock); 340 } 341 342 int dm_open_count(struct mapped_device *md) 343 { 344 return atomic_read(&md->open_count); 345 } 346 347 /* 348 * Guarantees nothing is using the device before it's deleted. 349 */ 350 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) 351 { 352 int r = 0; 353 354 spin_lock(&_minor_lock); 355 356 if (dm_open_count(md)) { 357 r = -EBUSY; 358 if (mark_deferred) 359 set_bit(DMF_DEFERRED_REMOVE, &md->flags); 360 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) 361 r = -EEXIST; 362 else 363 set_bit(DMF_DELETING, &md->flags); 364 365 spin_unlock(&_minor_lock); 366 367 return r; 368 } 369 370 int dm_cancel_deferred_remove(struct mapped_device *md) 371 { 372 int r = 0; 373 374 spin_lock(&_minor_lock); 375 376 if (test_bit(DMF_DELETING, &md->flags)) 377 r = -EBUSY; 378 else 379 clear_bit(DMF_DEFERRED_REMOVE, &md->flags); 380 381 spin_unlock(&_minor_lock); 382 383 return r; 384 } 385 386 static void do_deferred_remove(struct work_struct *w) 387 { 388 dm_deferred_remove(); 389 } 390 391 sector_t dm_get_size(struct mapped_device *md) 392 { 393 return get_capacity(md->disk); 394 } 395 396 struct request_queue *dm_get_md_queue(struct mapped_device *md) 397 { 398 return md->queue; 399 } 400 401 struct dm_stats *dm_get_stats(struct mapped_device *md) 402 { 403 return &md->stats; 404 } 405 406 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) 407 { 408 struct mapped_device *md = bdev->bd_disk->private_data; 409 410 return dm_get_geometry(md, geo); 411 } 412 413 static int dm_grab_bdev_for_ioctl(struct mapped_device *md, 414 struct block_device **bdev, 415 fmode_t *mode) 416 { 417 struct dm_target *tgt; 418 struct dm_table *map; 419 int srcu_idx, r; 420 421 retry: 422 r = -ENOTTY; 423 map = dm_get_live_table(md, &srcu_idx); 424 if (!map || !dm_table_get_size(map)) 425 goto out; 426 427 /* We only support devices that have a single target */ 428 if (dm_table_get_num_targets(map) != 1) 429 goto out; 430 431 tgt = dm_table_get_target(map, 0); 432 if (!tgt->type->prepare_ioctl) 433 goto out; 434 435 if (dm_suspended_md(md)) { 436 r = -EAGAIN; 437 goto out; 438 } 439 440 r = tgt->type->prepare_ioctl(tgt, bdev, mode); 441 if (r < 0) 442 goto out; 443 444 bdgrab(*bdev); 445 dm_put_live_table(md, srcu_idx); 446 return r; 447 448 out: 449 dm_put_live_table(md, srcu_idx); 450 if (r == -ENOTCONN && !fatal_signal_pending(current)) { 451 msleep(10); 452 goto retry; 453 } 454 return r; 455 } 456 457 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, 458 unsigned int cmd, unsigned long arg) 459 { 460 struct mapped_device *md = bdev->bd_disk->private_data; 461 int r; 462 463 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); 464 if (r < 0) 465 return r; 466 467 if (r > 0) { 468 /* 469 * Target determined this ioctl is being issued against a 470 * subset of the parent bdev; require extra privileges. 471 */ 472 if (!capable(CAP_SYS_RAWIO)) { 473 DMWARN_LIMIT( 474 "%s: sending ioctl %x to DM device without required privilege.", 475 current->comm, cmd); 476 r = -ENOIOCTLCMD; 477 goto out; 478 } 479 } 480 481 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); 482 out: 483 bdput(bdev); 484 return r; 485 } 486 487 static struct dm_io *alloc_io(struct mapped_device *md) 488 { 489 return mempool_alloc(md->io_pool, GFP_NOIO); 490 } 491 492 static void free_io(struct mapped_device *md, struct dm_io *io) 493 { 494 mempool_free(io, md->io_pool); 495 } 496 497 static void free_tio(struct dm_target_io *tio) 498 { 499 bio_put(&tio->clone); 500 } 501 502 int md_in_flight(struct mapped_device *md) 503 { 504 return atomic_read(&md->pending[READ]) + 505 atomic_read(&md->pending[WRITE]); 506 } 507 508 static void start_io_acct(struct dm_io *io) 509 { 510 struct mapped_device *md = io->md; 511 struct bio *bio = io->bio; 512 int cpu; 513 int rw = bio_data_dir(bio); 514 515 io->start_time = jiffies; 516 517 cpu = part_stat_lock(); 518 part_round_stats(cpu, &dm_disk(md)->part0); 519 part_stat_unlock(); 520 atomic_set(&dm_disk(md)->part0.in_flight[rw], 521 atomic_inc_return(&md->pending[rw])); 522 523 if (unlikely(dm_stats_used(&md->stats))) 524 dm_stats_account_io(&md->stats, bio_data_dir(bio), 525 bio->bi_iter.bi_sector, bio_sectors(bio), 526 false, 0, &io->stats_aux); 527 } 528 529 static void end_io_acct(struct dm_io *io) 530 { 531 struct mapped_device *md = io->md; 532 struct bio *bio = io->bio; 533 unsigned long duration = jiffies - io->start_time; 534 int pending; 535 int rw = bio_data_dir(bio); 536 537 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); 538 539 if (unlikely(dm_stats_used(&md->stats))) 540 dm_stats_account_io(&md->stats, bio_data_dir(bio), 541 bio->bi_iter.bi_sector, bio_sectors(bio), 542 true, duration, &io->stats_aux); 543 544 /* 545 * After this is decremented the bio must not be touched if it is 546 * a flush. 547 */ 548 pending = atomic_dec_return(&md->pending[rw]); 549 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); 550 pending += atomic_read(&md->pending[rw^0x1]); 551 552 /* nudge anyone waiting on suspend queue */ 553 if (!pending) 554 wake_up(&md->wait); 555 } 556 557 /* 558 * Add the bio to the list of deferred io. 559 */ 560 static void queue_io(struct mapped_device *md, struct bio *bio) 561 { 562 unsigned long flags; 563 564 spin_lock_irqsave(&md->deferred_lock, flags); 565 bio_list_add(&md->deferred, bio); 566 spin_unlock_irqrestore(&md->deferred_lock, flags); 567 queue_work(md->wq, &md->work); 568 } 569 570 /* 571 * Everyone (including functions in this file), should use this 572 * function to access the md->map field, and make sure they call 573 * dm_put_live_table() when finished. 574 */ 575 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) 576 { 577 *srcu_idx = srcu_read_lock(&md->io_barrier); 578 579 return srcu_dereference(md->map, &md->io_barrier); 580 } 581 582 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) 583 { 584 srcu_read_unlock(&md->io_barrier, srcu_idx); 585 } 586 587 void dm_sync_table(struct mapped_device *md) 588 { 589 synchronize_srcu(&md->io_barrier); 590 synchronize_rcu_expedited(); 591 } 592 593 /* 594 * A fast alternative to dm_get_live_table/dm_put_live_table. 595 * The caller must not block between these two functions. 596 */ 597 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) 598 { 599 rcu_read_lock(); 600 return rcu_dereference(md->map); 601 } 602 603 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) 604 { 605 rcu_read_unlock(); 606 } 607 608 /* 609 * Open a table device so we can use it as a map destination. 610 */ 611 static int open_table_device(struct table_device *td, dev_t dev, 612 struct mapped_device *md) 613 { 614 static char *_claim_ptr = "I belong to device-mapper"; 615 struct block_device *bdev; 616 617 int r; 618 619 BUG_ON(td->dm_dev.bdev); 620 621 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); 622 if (IS_ERR(bdev)) 623 return PTR_ERR(bdev); 624 625 r = bd_link_disk_holder(bdev, dm_disk(md)); 626 if (r) { 627 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); 628 return r; 629 } 630 631 td->dm_dev.bdev = bdev; 632 return 0; 633 } 634 635 /* 636 * Close a table device that we've been using. 637 */ 638 static void close_table_device(struct table_device *td, struct mapped_device *md) 639 { 640 if (!td->dm_dev.bdev) 641 return; 642 643 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); 644 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); 645 td->dm_dev.bdev = NULL; 646 } 647 648 static struct table_device *find_table_device(struct list_head *l, dev_t dev, 649 fmode_t mode) { 650 struct table_device *td; 651 652 list_for_each_entry(td, l, list) 653 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) 654 return td; 655 656 return NULL; 657 } 658 659 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, 660 struct dm_dev **result) { 661 int r; 662 struct table_device *td; 663 664 mutex_lock(&md->table_devices_lock); 665 td = find_table_device(&md->table_devices, dev, mode); 666 if (!td) { 667 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); 668 if (!td) { 669 mutex_unlock(&md->table_devices_lock); 670 return -ENOMEM; 671 } 672 673 td->dm_dev.mode = mode; 674 td->dm_dev.bdev = NULL; 675 676 if ((r = open_table_device(td, dev, md))) { 677 mutex_unlock(&md->table_devices_lock); 678 kfree(td); 679 return r; 680 } 681 682 format_dev_t(td->dm_dev.name, dev); 683 684 atomic_set(&td->count, 0); 685 list_add(&td->list, &md->table_devices); 686 } 687 atomic_inc(&td->count); 688 mutex_unlock(&md->table_devices_lock); 689 690 *result = &td->dm_dev; 691 return 0; 692 } 693 EXPORT_SYMBOL_GPL(dm_get_table_device); 694 695 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) 696 { 697 struct table_device *td = container_of(d, struct table_device, dm_dev); 698 699 mutex_lock(&md->table_devices_lock); 700 if (atomic_dec_and_test(&td->count)) { 701 close_table_device(td, md); 702 list_del(&td->list); 703 kfree(td); 704 } 705 mutex_unlock(&md->table_devices_lock); 706 } 707 EXPORT_SYMBOL(dm_put_table_device); 708 709 static void free_table_devices(struct list_head *devices) 710 { 711 struct list_head *tmp, *next; 712 713 list_for_each_safe(tmp, next, devices) { 714 struct table_device *td = list_entry(tmp, struct table_device, list); 715 716 DMWARN("dm_destroy: %s still exists with %d references", 717 td->dm_dev.name, atomic_read(&td->count)); 718 kfree(td); 719 } 720 } 721 722 /* 723 * Get the geometry associated with a dm device 724 */ 725 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) 726 { 727 *geo = md->geometry; 728 729 return 0; 730 } 731 732 /* 733 * Set the geometry of a device. 734 */ 735 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) 736 { 737 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; 738 739 if (geo->start > sz) { 740 DMWARN("Start sector is beyond the geometry limits."); 741 return -EINVAL; 742 } 743 744 md->geometry = *geo; 745 746 return 0; 747 } 748 749 /*----------------------------------------------------------------- 750 * CRUD START: 751 * A more elegant soln is in the works that uses the queue 752 * merge fn, unfortunately there are a couple of changes to 753 * the block layer that I want to make for this. So in the 754 * interests of getting something for people to use I give 755 * you this clearly demarcated crap. 756 *---------------------------------------------------------------*/ 757 758 static int __noflush_suspending(struct mapped_device *md) 759 { 760 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 761 } 762 763 /* 764 * Decrements the number of outstanding ios that a bio has been 765 * cloned into, completing the original io if necc. 766 */ 767 static void dec_pending(struct dm_io *io, int error) 768 { 769 unsigned long flags; 770 int io_error; 771 struct bio *bio; 772 struct mapped_device *md = io->md; 773 774 /* Push-back supersedes any I/O errors */ 775 if (unlikely(error)) { 776 spin_lock_irqsave(&io->endio_lock, flags); 777 if (!(io->error > 0 && __noflush_suspending(md))) 778 io->error = error; 779 spin_unlock_irqrestore(&io->endio_lock, flags); 780 } 781 782 if (atomic_dec_and_test(&io->io_count)) { 783 if (io->error == DM_ENDIO_REQUEUE) { 784 /* 785 * Target requested pushing back the I/O. 786 */ 787 spin_lock_irqsave(&md->deferred_lock, flags); 788 if (__noflush_suspending(md)) 789 bio_list_add_head(&md->deferred, io->bio); 790 else 791 /* noflush suspend was interrupted. */ 792 io->error = -EIO; 793 spin_unlock_irqrestore(&md->deferred_lock, flags); 794 } 795 796 io_error = io->error; 797 bio = io->bio; 798 end_io_acct(io); 799 free_io(md, io); 800 801 if (io_error == DM_ENDIO_REQUEUE) 802 return; 803 804 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { 805 /* 806 * Preflush done for flush with data, reissue 807 * without REQ_PREFLUSH. 808 */ 809 bio->bi_opf &= ~REQ_PREFLUSH; 810 queue_io(md, bio); 811 } else { 812 /* done with normal IO or empty flush */ 813 trace_block_bio_complete(md->queue, bio, io_error); 814 bio->bi_error = io_error; 815 bio_endio(bio); 816 } 817 } 818 } 819 820 void disable_write_same(struct mapped_device *md) 821 { 822 struct queue_limits *limits = dm_get_queue_limits(md); 823 824 /* device doesn't really support WRITE SAME, disable it */ 825 limits->max_write_same_sectors = 0; 826 } 827 828 static void clone_endio(struct bio *bio) 829 { 830 int error = bio->bi_error; 831 int r = error; 832 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); 833 struct dm_io *io = tio->io; 834 struct mapped_device *md = tio->io->md; 835 dm_endio_fn endio = tio->ti->type->end_io; 836 837 if (endio) { 838 r = endio(tio->ti, bio, error); 839 if (r < 0 || r == DM_ENDIO_REQUEUE) 840 /* 841 * error and requeue request are handled 842 * in dec_pending(). 843 */ 844 error = r; 845 else if (r == DM_ENDIO_INCOMPLETE) 846 /* The target will handle the io */ 847 return; 848 else if (r) { 849 DMWARN("unimplemented target endio return value: %d", r); 850 BUG(); 851 } 852 } 853 854 if (unlikely(r == -EREMOTEIO && (bio_op(bio) == REQ_OP_WRITE_SAME) && 855 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) 856 disable_write_same(md); 857 858 free_tio(tio); 859 dec_pending(io, error); 860 } 861 862 /* 863 * Return maximum size of I/O possible at the supplied sector up to the current 864 * target boundary. 865 */ 866 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) 867 { 868 sector_t target_offset = dm_target_offset(ti, sector); 869 870 return ti->len - target_offset; 871 } 872 873 static sector_t max_io_len(sector_t sector, struct dm_target *ti) 874 { 875 sector_t len = max_io_len_target_boundary(sector, ti); 876 sector_t offset, max_len; 877 878 /* 879 * Does the target need to split even further? 880 */ 881 if (ti->max_io_len) { 882 offset = dm_target_offset(ti, sector); 883 if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) 884 max_len = sector_div(offset, ti->max_io_len); 885 else 886 max_len = offset & (ti->max_io_len - 1); 887 max_len = ti->max_io_len - max_len; 888 889 if (len > max_len) 890 len = max_len; 891 } 892 893 return len; 894 } 895 896 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) 897 { 898 if (len > UINT_MAX) { 899 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", 900 (unsigned long long)len, UINT_MAX); 901 ti->error = "Maximum size of target IO is too large"; 902 return -EINVAL; 903 } 904 905 ti->max_io_len = (uint32_t) len; 906 907 return 0; 908 } 909 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); 910 911 static long dm_blk_direct_access(struct block_device *bdev, sector_t sector, 912 void **kaddr, pfn_t *pfn, long size) 913 { 914 struct mapped_device *md = bdev->bd_disk->private_data; 915 struct dm_table *map; 916 struct dm_target *ti; 917 int srcu_idx; 918 long len, ret = -EIO; 919 920 map = dm_get_live_table(md, &srcu_idx); 921 if (!map) 922 goto out; 923 924 ti = dm_table_find_target(map, sector); 925 if (!dm_target_is_valid(ti)) 926 goto out; 927 928 len = max_io_len(sector, ti) << SECTOR_SHIFT; 929 size = min(len, size); 930 931 if (ti->type->direct_access) 932 ret = ti->type->direct_access(ti, sector, kaddr, pfn, size); 933 out: 934 dm_put_live_table(md, srcu_idx); 935 return min(ret, size); 936 } 937 938 /* 939 * A target may call dm_accept_partial_bio only from the map routine. It is 940 * allowed for all bio types except REQ_PREFLUSH. 941 * 942 * dm_accept_partial_bio informs the dm that the target only wants to process 943 * additional n_sectors sectors of the bio and the rest of the data should be 944 * sent in a next bio. 945 * 946 * A diagram that explains the arithmetics: 947 * +--------------------+---------------+-------+ 948 * | 1 | 2 | 3 | 949 * +--------------------+---------------+-------+ 950 * 951 * <-------------- *tio->len_ptr ---------------> 952 * <------- bi_size -------> 953 * <-- n_sectors --> 954 * 955 * Region 1 was already iterated over with bio_advance or similar function. 956 * (it may be empty if the target doesn't use bio_advance) 957 * Region 2 is the remaining bio size that the target wants to process. 958 * (it may be empty if region 1 is non-empty, although there is no reason 959 * to make it empty) 960 * The target requires that region 3 is to be sent in the next bio. 961 * 962 * If the target wants to receive multiple copies of the bio (via num_*bios, etc), 963 * the partially processed part (the sum of regions 1+2) must be the same for all 964 * copies of the bio. 965 */ 966 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) 967 { 968 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); 969 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; 970 BUG_ON(bio->bi_opf & REQ_PREFLUSH); 971 BUG_ON(bi_size > *tio->len_ptr); 972 BUG_ON(n_sectors > bi_size); 973 *tio->len_ptr -= bi_size - n_sectors; 974 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; 975 } 976 EXPORT_SYMBOL_GPL(dm_accept_partial_bio); 977 978 /* 979 * Flush current->bio_list when the target map method blocks. 980 * This fixes deadlocks in snapshot and possibly in other targets. 981 */ 982 struct dm_offload { 983 struct blk_plug plug; 984 struct blk_plug_cb cb; 985 }; 986 987 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule) 988 { 989 struct dm_offload *o = container_of(cb, struct dm_offload, cb); 990 struct bio_list list; 991 struct bio *bio; 992 993 INIT_LIST_HEAD(&o->cb.list); 994 995 if (unlikely(!current->bio_list)) 996 return; 997 998 list = *current->bio_list; 999 bio_list_init(current->bio_list); 1000 1001 while ((bio = bio_list_pop(&list))) { 1002 struct bio_set *bs = bio->bi_pool; 1003 if (unlikely(!bs) || bs == fs_bio_set) { 1004 bio_list_add(current->bio_list, bio); 1005 continue; 1006 } 1007 1008 spin_lock(&bs->rescue_lock); 1009 bio_list_add(&bs->rescue_list, bio); 1010 queue_work(bs->rescue_workqueue, &bs->rescue_work); 1011 spin_unlock(&bs->rescue_lock); 1012 } 1013 } 1014 1015 static void dm_offload_start(struct dm_offload *o) 1016 { 1017 blk_start_plug(&o->plug); 1018 o->cb.callback = flush_current_bio_list; 1019 list_add(&o->cb.list, ¤t->plug->cb_list); 1020 } 1021 1022 static void dm_offload_end(struct dm_offload *o) 1023 { 1024 list_del(&o->cb.list); 1025 blk_finish_plug(&o->plug); 1026 } 1027 1028 static void __map_bio(struct dm_target_io *tio) 1029 { 1030 int r; 1031 sector_t sector; 1032 struct dm_offload o; 1033 struct bio *clone = &tio->clone; 1034 struct dm_target *ti = tio->ti; 1035 1036 clone->bi_end_io = clone_endio; 1037 1038 /* 1039 * Map the clone. If r == 0 we don't need to do 1040 * anything, the target has assumed ownership of 1041 * this io. 1042 */ 1043 atomic_inc(&tio->io->io_count); 1044 sector = clone->bi_iter.bi_sector; 1045 1046 dm_offload_start(&o); 1047 r = ti->type->map(ti, clone); 1048 dm_offload_end(&o); 1049 1050 if (r == DM_MAPIO_REMAPPED) { 1051 /* the bio has been remapped so dispatch it */ 1052 1053 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, 1054 tio->io->bio->bi_bdev->bd_dev, sector); 1055 1056 generic_make_request(clone); 1057 } else if (r < 0 || r == DM_MAPIO_REQUEUE) { 1058 /* error the io and bail out, or requeue it if needed */ 1059 dec_pending(tio->io, r); 1060 free_tio(tio); 1061 } else if (r != DM_MAPIO_SUBMITTED) { 1062 DMWARN("unimplemented target map return value: %d", r); 1063 BUG(); 1064 } 1065 } 1066 1067 struct clone_info { 1068 struct mapped_device *md; 1069 struct dm_table *map; 1070 struct bio *bio; 1071 struct dm_io *io; 1072 sector_t sector; 1073 unsigned sector_count; 1074 }; 1075 1076 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) 1077 { 1078 bio->bi_iter.bi_sector = sector; 1079 bio->bi_iter.bi_size = to_bytes(len); 1080 } 1081 1082 /* 1083 * Creates a bio that consists of range of complete bvecs. 1084 */ 1085 static int clone_bio(struct dm_target_io *tio, struct bio *bio, 1086 sector_t sector, unsigned len) 1087 { 1088 struct bio *clone = &tio->clone; 1089 1090 __bio_clone_fast(clone, bio); 1091 1092 if (bio_integrity(bio)) { 1093 int r = bio_integrity_clone(clone, bio, GFP_NOIO); 1094 if (r < 0) 1095 return r; 1096 } 1097 1098 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); 1099 clone->bi_iter.bi_size = to_bytes(len); 1100 1101 if (bio_integrity(bio)) 1102 bio_integrity_trim(clone, 0, len); 1103 1104 return 0; 1105 } 1106 1107 static struct dm_target_io *alloc_tio(struct clone_info *ci, 1108 struct dm_target *ti, 1109 unsigned target_bio_nr) 1110 { 1111 struct dm_target_io *tio; 1112 struct bio *clone; 1113 1114 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); 1115 tio = container_of(clone, struct dm_target_io, clone); 1116 1117 tio->io = ci->io; 1118 tio->ti = ti; 1119 tio->target_bio_nr = target_bio_nr; 1120 1121 return tio; 1122 } 1123 1124 static void __clone_and_map_simple_bio(struct clone_info *ci, 1125 struct dm_target *ti, 1126 unsigned target_bio_nr, unsigned *len) 1127 { 1128 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); 1129 struct bio *clone = &tio->clone; 1130 1131 tio->len_ptr = len; 1132 1133 __bio_clone_fast(clone, ci->bio); 1134 if (len) 1135 bio_setup_sector(clone, ci->sector, *len); 1136 1137 __map_bio(tio); 1138 } 1139 1140 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, 1141 unsigned num_bios, unsigned *len) 1142 { 1143 unsigned target_bio_nr; 1144 1145 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) 1146 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); 1147 } 1148 1149 static int __send_empty_flush(struct clone_info *ci) 1150 { 1151 unsigned target_nr = 0; 1152 struct dm_target *ti; 1153 1154 BUG_ON(bio_has_data(ci->bio)); 1155 while ((ti = dm_table_get_target(ci->map, target_nr++))) 1156 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); 1157 1158 return 0; 1159 } 1160 1161 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, 1162 sector_t sector, unsigned *len) 1163 { 1164 struct bio *bio = ci->bio; 1165 struct dm_target_io *tio; 1166 unsigned target_bio_nr; 1167 unsigned num_target_bios = 1; 1168 int r = 0; 1169 1170 /* 1171 * Does the target want to receive duplicate copies of the bio? 1172 */ 1173 if (bio_data_dir(bio) == WRITE && ti->num_write_bios) 1174 num_target_bios = ti->num_write_bios(ti, bio); 1175 1176 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { 1177 tio = alloc_tio(ci, ti, target_bio_nr); 1178 tio->len_ptr = len; 1179 r = clone_bio(tio, bio, sector, *len); 1180 if (r < 0) { 1181 free_tio(tio); 1182 break; 1183 } 1184 __map_bio(tio); 1185 } 1186 1187 return r; 1188 } 1189 1190 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); 1191 1192 static unsigned get_num_discard_bios(struct dm_target *ti) 1193 { 1194 return ti->num_discard_bios; 1195 } 1196 1197 static unsigned get_num_write_same_bios(struct dm_target *ti) 1198 { 1199 return ti->num_write_same_bios; 1200 } 1201 1202 typedef bool (*is_split_required_fn)(struct dm_target *ti); 1203 1204 static bool is_split_required_for_discard(struct dm_target *ti) 1205 { 1206 return ti->split_discard_bios; 1207 } 1208 1209 static int __send_changing_extent_only(struct clone_info *ci, 1210 get_num_bios_fn get_num_bios, 1211 is_split_required_fn is_split_required) 1212 { 1213 struct dm_target *ti; 1214 unsigned len; 1215 unsigned num_bios; 1216 1217 do { 1218 ti = dm_table_find_target(ci->map, ci->sector); 1219 if (!dm_target_is_valid(ti)) 1220 return -EIO; 1221 1222 /* 1223 * Even though the device advertised support for this type of 1224 * request, that does not mean every target supports it, and 1225 * reconfiguration might also have changed that since the 1226 * check was performed. 1227 */ 1228 num_bios = get_num_bios ? get_num_bios(ti) : 0; 1229 if (!num_bios) 1230 return -EOPNOTSUPP; 1231 1232 if (is_split_required && !is_split_required(ti)) 1233 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); 1234 else 1235 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); 1236 1237 __send_duplicate_bios(ci, ti, num_bios, &len); 1238 1239 ci->sector += len; 1240 } while (ci->sector_count -= len); 1241 1242 return 0; 1243 } 1244 1245 static int __send_discard(struct clone_info *ci) 1246 { 1247 return __send_changing_extent_only(ci, get_num_discard_bios, 1248 is_split_required_for_discard); 1249 } 1250 1251 static int __send_write_same(struct clone_info *ci) 1252 { 1253 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); 1254 } 1255 1256 /* 1257 * Select the correct strategy for processing a non-flush bio. 1258 */ 1259 static int __split_and_process_non_flush(struct clone_info *ci) 1260 { 1261 struct bio *bio = ci->bio; 1262 struct dm_target *ti; 1263 unsigned len; 1264 int r; 1265 1266 if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) 1267 return __send_discard(ci); 1268 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) 1269 return __send_write_same(ci); 1270 1271 ti = dm_table_find_target(ci->map, ci->sector); 1272 if (!dm_target_is_valid(ti)) 1273 return -EIO; 1274 1275 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); 1276 1277 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); 1278 if (r < 0) 1279 return r; 1280 1281 ci->sector += len; 1282 ci->sector_count -= len; 1283 1284 return 0; 1285 } 1286 1287 /* 1288 * Entry point to split a bio into clones and submit them to the targets. 1289 */ 1290 static void __split_and_process_bio(struct mapped_device *md, 1291 struct dm_table *map, struct bio *bio) 1292 { 1293 struct clone_info ci; 1294 int error = 0; 1295 1296 if (unlikely(!map)) { 1297 bio_io_error(bio); 1298 return; 1299 } 1300 1301 ci.map = map; 1302 ci.md = md; 1303 ci.io = alloc_io(md); 1304 ci.io->error = 0; 1305 atomic_set(&ci.io->io_count, 1); 1306 ci.io->bio = bio; 1307 ci.io->md = md; 1308 spin_lock_init(&ci.io->endio_lock); 1309 ci.sector = bio->bi_iter.bi_sector; 1310 1311 start_io_acct(ci.io); 1312 1313 if (bio->bi_opf & REQ_PREFLUSH) { 1314 ci.bio = &ci.md->flush_bio; 1315 ci.sector_count = 0; 1316 error = __send_empty_flush(&ci); 1317 /* dec_pending submits any data associated with flush */ 1318 } else { 1319 ci.bio = bio; 1320 ci.sector_count = bio_sectors(bio); 1321 while (ci.sector_count && !error) 1322 error = __split_and_process_non_flush(&ci); 1323 } 1324 1325 /* drop the extra reference count */ 1326 dec_pending(ci.io, error); 1327 } 1328 /*----------------------------------------------------------------- 1329 * CRUD END 1330 *---------------------------------------------------------------*/ 1331 1332 /* 1333 * The request function that just remaps the bio built up by 1334 * dm_merge_bvec. 1335 */ 1336 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) 1337 { 1338 int rw = bio_data_dir(bio); 1339 struct mapped_device *md = q->queuedata; 1340 int srcu_idx; 1341 struct dm_table *map; 1342 1343 map = dm_get_live_table(md, &srcu_idx); 1344 1345 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); 1346 1347 /* if we're suspended, we have to queue this io for later */ 1348 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { 1349 dm_put_live_table(md, srcu_idx); 1350 1351 if (!(bio->bi_opf & REQ_RAHEAD)) 1352 queue_io(md, bio); 1353 else 1354 bio_io_error(bio); 1355 return BLK_QC_T_NONE; 1356 } 1357 1358 __split_and_process_bio(md, map, bio); 1359 dm_put_live_table(md, srcu_idx); 1360 return BLK_QC_T_NONE; 1361 } 1362 1363 static int dm_any_congested(void *congested_data, int bdi_bits) 1364 { 1365 int r = bdi_bits; 1366 struct mapped_device *md = congested_data; 1367 struct dm_table *map; 1368 1369 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 1370 if (dm_request_based(md)) { 1371 /* 1372 * With request-based DM we only need to check the 1373 * top-level queue for congestion. 1374 */ 1375 r = md->queue->backing_dev_info->wb.state & bdi_bits; 1376 } else { 1377 map = dm_get_live_table_fast(md); 1378 if (map) 1379 r = dm_table_any_congested(map, bdi_bits); 1380 dm_put_live_table_fast(md); 1381 } 1382 } 1383 1384 return r; 1385 } 1386 1387 /*----------------------------------------------------------------- 1388 * An IDR is used to keep track of allocated minor numbers. 1389 *---------------------------------------------------------------*/ 1390 static void free_minor(int minor) 1391 { 1392 spin_lock(&_minor_lock); 1393 idr_remove(&_minor_idr, minor); 1394 spin_unlock(&_minor_lock); 1395 } 1396 1397 /* 1398 * See if the device with a specific minor # is free. 1399 */ 1400 static int specific_minor(int minor) 1401 { 1402 int r; 1403 1404 if (minor >= (1 << MINORBITS)) 1405 return -EINVAL; 1406 1407 idr_preload(GFP_KERNEL); 1408 spin_lock(&_minor_lock); 1409 1410 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); 1411 1412 spin_unlock(&_minor_lock); 1413 idr_preload_end(); 1414 if (r < 0) 1415 return r == -ENOSPC ? -EBUSY : r; 1416 return 0; 1417 } 1418 1419 static int next_free_minor(int *minor) 1420 { 1421 int r; 1422 1423 idr_preload(GFP_KERNEL); 1424 spin_lock(&_minor_lock); 1425 1426 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); 1427 1428 spin_unlock(&_minor_lock); 1429 idr_preload_end(); 1430 if (r < 0) 1431 return r; 1432 *minor = r; 1433 return 0; 1434 } 1435 1436 static const struct block_device_operations dm_blk_dops; 1437 1438 static void dm_wq_work(struct work_struct *work); 1439 1440 void dm_init_md_queue(struct mapped_device *md) 1441 { 1442 /* 1443 * Request-based dm devices cannot be stacked on top of bio-based dm 1444 * devices. The type of this dm device may not have been decided yet. 1445 * The type is decided at the first table loading time. 1446 * To prevent problematic device stacking, clear the queue flag 1447 * for request stacking support until then. 1448 * 1449 * This queue is new, so no concurrency on the queue_flags. 1450 */ 1451 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); 1452 1453 /* 1454 * Initialize data that will only be used by a non-blk-mq DM queue 1455 * - must do so here (in alloc_dev callchain) before queue is used 1456 */ 1457 md->queue->queuedata = md; 1458 md->queue->backing_dev_info->congested_data = md; 1459 } 1460 1461 void dm_init_normal_md_queue(struct mapped_device *md) 1462 { 1463 md->use_blk_mq = false; 1464 dm_init_md_queue(md); 1465 1466 /* 1467 * Initialize aspects of queue that aren't relevant for blk-mq 1468 */ 1469 md->queue->backing_dev_info->congested_fn = dm_any_congested; 1470 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); 1471 } 1472 1473 static void cleanup_mapped_device(struct mapped_device *md) 1474 { 1475 if (md->wq) 1476 destroy_workqueue(md->wq); 1477 if (md->kworker_task) 1478 kthread_stop(md->kworker_task); 1479 mempool_destroy(md->io_pool); 1480 if (md->bs) 1481 bioset_free(md->bs); 1482 1483 if (md->disk) { 1484 spin_lock(&_minor_lock); 1485 md->disk->private_data = NULL; 1486 spin_unlock(&_minor_lock); 1487 del_gendisk(md->disk); 1488 put_disk(md->disk); 1489 } 1490 1491 if (md->queue) 1492 blk_cleanup_queue(md->queue); 1493 1494 cleanup_srcu_struct(&md->io_barrier); 1495 1496 if (md->bdev) { 1497 bdput(md->bdev); 1498 md->bdev = NULL; 1499 } 1500 1501 dm_mq_cleanup_mapped_device(md); 1502 } 1503 1504 /* 1505 * Allocate and initialise a blank device with a given minor. 1506 */ 1507 static struct mapped_device *alloc_dev(int minor) 1508 { 1509 int r, numa_node_id = dm_get_numa_node(); 1510 struct mapped_device *md; 1511 void *old_md; 1512 1513 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); 1514 if (!md) { 1515 DMWARN("unable to allocate device, out of memory."); 1516 return NULL; 1517 } 1518 1519 if (!try_module_get(THIS_MODULE)) 1520 goto bad_module_get; 1521 1522 /* get a minor number for the dev */ 1523 if (minor == DM_ANY_MINOR) 1524 r = next_free_minor(&minor); 1525 else 1526 r = specific_minor(minor); 1527 if (r < 0) 1528 goto bad_minor; 1529 1530 r = init_srcu_struct(&md->io_barrier); 1531 if (r < 0) 1532 goto bad_io_barrier; 1533 1534 md->numa_node_id = numa_node_id; 1535 md->use_blk_mq = dm_use_blk_mq_default(); 1536 md->init_tio_pdu = false; 1537 md->type = DM_TYPE_NONE; 1538 mutex_init(&md->suspend_lock); 1539 mutex_init(&md->type_lock); 1540 mutex_init(&md->table_devices_lock); 1541 spin_lock_init(&md->deferred_lock); 1542 atomic_set(&md->holders, 1); 1543 atomic_set(&md->open_count, 0); 1544 atomic_set(&md->event_nr, 0); 1545 atomic_set(&md->uevent_seq, 0); 1546 INIT_LIST_HEAD(&md->uevent_list); 1547 INIT_LIST_HEAD(&md->table_devices); 1548 spin_lock_init(&md->uevent_lock); 1549 1550 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); 1551 if (!md->queue) 1552 goto bad; 1553 1554 dm_init_md_queue(md); 1555 1556 md->disk = alloc_disk_node(1, numa_node_id); 1557 if (!md->disk) 1558 goto bad; 1559 1560 atomic_set(&md->pending[0], 0); 1561 atomic_set(&md->pending[1], 0); 1562 init_waitqueue_head(&md->wait); 1563 INIT_WORK(&md->work, dm_wq_work); 1564 init_waitqueue_head(&md->eventq); 1565 init_completion(&md->kobj_holder.completion); 1566 md->kworker_task = NULL; 1567 1568 md->disk->major = _major; 1569 md->disk->first_minor = minor; 1570 md->disk->fops = &dm_blk_dops; 1571 md->disk->queue = md->queue; 1572 md->disk->private_data = md; 1573 sprintf(md->disk->disk_name, "dm-%d", minor); 1574 add_disk(md->disk); 1575 format_dev_t(md->name, MKDEV(_major, minor)); 1576 1577 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); 1578 if (!md->wq) 1579 goto bad; 1580 1581 md->bdev = bdget_disk(md->disk, 0); 1582 if (!md->bdev) 1583 goto bad; 1584 1585 bio_init(&md->flush_bio, NULL, 0); 1586 md->flush_bio.bi_bdev = md->bdev; 1587 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; 1588 1589 dm_stats_init(&md->stats); 1590 1591 /* Populate the mapping, nobody knows we exist yet */ 1592 spin_lock(&_minor_lock); 1593 old_md = idr_replace(&_minor_idr, md, minor); 1594 spin_unlock(&_minor_lock); 1595 1596 BUG_ON(old_md != MINOR_ALLOCED); 1597 1598 return md; 1599 1600 bad: 1601 cleanup_mapped_device(md); 1602 bad_io_barrier: 1603 free_minor(minor); 1604 bad_minor: 1605 module_put(THIS_MODULE); 1606 bad_module_get: 1607 kfree(md); 1608 return NULL; 1609 } 1610 1611 static void unlock_fs(struct mapped_device *md); 1612 1613 static void free_dev(struct mapped_device *md) 1614 { 1615 int minor = MINOR(disk_devt(md->disk)); 1616 1617 unlock_fs(md); 1618 1619 cleanup_mapped_device(md); 1620 1621 free_table_devices(&md->table_devices); 1622 dm_stats_cleanup(&md->stats); 1623 free_minor(minor); 1624 1625 module_put(THIS_MODULE); 1626 kfree(md); 1627 } 1628 1629 static void __bind_mempools(struct mapped_device *md, struct dm_table *t) 1630 { 1631 struct dm_md_mempools *p = dm_table_get_md_mempools(t); 1632 1633 if (md->bs) { 1634 /* The md already has necessary mempools. */ 1635 if (dm_table_bio_based(t)) { 1636 /* 1637 * Reload bioset because front_pad may have changed 1638 * because a different table was loaded. 1639 */ 1640 bioset_free(md->bs); 1641 md->bs = p->bs; 1642 p->bs = NULL; 1643 } 1644 /* 1645 * There's no need to reload with request-based dm 1646 * because the size of front_pad doesn't change. 1647 * Note for future: If you are to reload bioset, 1648 * prep-ed requests in the queue may refer 1649 * to bio from the old bioset, so you must walk 1650 * through the queue to unprep. 1651 */ 1652 goto out; 1653 } 1654 1655 BUG_ON(!p || md->io_pool || md->bs); 1656 1657 md->io_pool = p->io_pool; 1658 p->io_pool = NULL; 1659 md->bs = p->bs; 1660 p->bs = NULL; 1661 1662 out: 1663 /* mempool bind completed, no longer need any mempools in the table */ 1664 dm_table_free_md_mempools(t); 1665 } 1666 1667 /* 1668 * Bind a table to the device. 1669 */ 1670 static void event_callback(void *context) 1671 { 1672 unsigned long flags; 1673 LIST_HEAD(uevents); 1674 struct mapped_device *md = (struct mapped_device *) context; 1675 1676 spin_lock_irqsave(&md->uevent_lock, flags); 1677 list_splice_init(&md->uevent_list, &uevents); 1678 spin_unlock_irqrestore(&md->uevent_lock, flags); 1679 1680 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); 1681 1682 atomic_inc(&md->event_nr); 1683 wake_up(&md->eventq); 1684 } 1685 1686 /* 1687 * Protected by md->suspend_lock obtained by dm_swap_table(). 1688 */ 1689 static void __set_size(struct mapped_device *md, sector_t size) 1690 { 1691 set_capacity(md->disk, size); 1692 1693 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); 1694 } 1695 1696 /* 1697 * Returns old map, which caller must destroy. 1698 */ 1699 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, 1700 struct queue_limits *limits) 1701 { 1702 struct dm_table *old_map; 1703 struct request_queue *q = md->queue; 1704 sector_t size; 1705 1706 lockdep_assert_held(&md->suspend_lock); 1707 1708 size = dm_table_get_size(t); 1709 1710 /* 1711 * Wipe any geometry if the size of the table changed. 1712 */ 1713 if (size != dm_get_size(md)) 1714 memset(&md->geometry, 0, sizeof(md->geometry)); 1715 1716 __set_size(md, size); 1717 1718 dm_table_event_callback(t, event_callback, md); 1719 1720 /* 1721 * The queue hasn't been stopped yet, if the old table type wasn't 1722 * for request-based during suspension. So stop it to prevent 1723 * I/O mapping before resume. 1724 * This must be done before setting the queue restrictions, 1725 * because request-based dm may be run just after the setting. 1726 */ 1727 if (dm_table_request_based(t)) { 1728 dm_stop_queue(q); 1729 /* 1730 * Leverage the fact that request-based DM targets are 1731 * immutable singletons and establish md->immutable_target 1732 * - used to optimize both dm_request_fn and dm_mq_queue_rq 1733 */ 1734 md->immutable_target = dm_table_get_immutable_target(t); 1735 } 1736 1737 __bind_mempools(md, t); 1738 1739 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 1740 rcu_assign_pointer(md->map, (void *)t); 1741 md->immutable_target_type = dm_table_get_immutable_target_type(t); 1742 1743 dm_table_set_restrictions(t, q, limits); 1744 if (old_map) 1745 dm_sync_table(md); 1746 1747 return old_map; 1748 } 1749 1750 /* 1751 * Returns unbound table for the caller to free. 1752 */ 1753 static struct dm_table *__unbind(struct mapped_device *md) 1754 { 1755 struct dm_table *map = rcu_dereference_protected(md->map, 1); 1756 1757 if (!map) 1758 return NULL; 1759 1760 dm_table_event_callback(map, NULL, NULL); 1761 RCU_INIT_POINTER(md->map, NULL); 1762 dm_sync_table(md); 1763 1764 return map; 1765 } 1766 1767 /* 1768 * Constructor for a new device. 1769 */ 1770 int dm_create(int minor, struct mapped_device **result) 1771 { 1772 struct mapped_device *md; 1773 1774 md = alloc_dev(minor); 1775 if (!md) 1776 return -ENXIO; 1777 1778 dm_sysfs_init(md); 1779 1780 *result = md; 1781 return 0; 1782 } 1783 1784 /* 1785 * Functions to manage md->type. 1786 * All are required to hold md->type_lock. 1787 */ 1788 void dm_lock_md_type(struct mapped_device *md) 1789 { 1790 mutex_lock(&md->type_lock); 1791 } 1792 1793 void dm_unlock_md_type(struct mapped_device *md) 1794 { 1795 mutex_unlock(&md->type_lock); 1796 } 1797 1798 void dm_set_md_type(struct mapped_device *md, unsigned type) 1799 { 1800 BUG_ON(!mutex_is_locked(&md->type_lock)); 1801 md->type = type; 1802 } 1803 1804 unsigned dm_get_md_type(struct mapped_device *md) 1805 { 1806 return md->type; 1807 } 1808 1809 struct target_type *dm_get_immutable_target_type(struct mapped_device *md) 1810 { 1811 return md->immutable_target_type; 1812 } 1813 1814 /* 1815 * The queue_limits are only valid as long as you have a reference 1816 * count on 'md'. 1817 */ 1818 struct queue_limits *dm_get_queue_limits(struct mapped_device *md) 1819 { 1820 BUG_ON(!atomic_read(&md->holders)); 1821 return &md->queue->limits; 1822 } 1823 EXPORT_SYMBOL_GPL(dm_get_queue_limits); 1824 1825 /* 1826 * Setup the DM device's queue based on md's type 1827 */ 1828 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) 1829 { 1830 int r; 1831 unsigned type = dm_get_md_type(md); 1832 1833 switch (type) { 1834 case DM_TYPE_REQUEST_BASED: 1835 r = dm_old_init_request_queue(md, t); 1836 if (r) { 1837 DMERR("Cannot initialize queue for request-based mapped device"); 1838 return r; 1839 } 1840 break; 1841 case DM_TYPE_MQ_REQUEST_BASED: 1842 r = dm_mq_init_request_queue(md, t); 1843 if (r) { 1844 DMERR("Cannot initialize queue for request-based dm-mq mapped device"); 1845 return r; 1846 } 1847 break; 1848 case DM_TYPE_BIO_BASED: 1849 case DM_TYPE_DAX_BIO_BASED: 1850 dm_init_normal_md_queue(md); 1851 blk_queue_make_request(md->queue, dm_make_request); 1852 /* 1853 * DM handles splitting bios as needed. Free the bio_split bioset 1854 * since it won't be used (saves 1 process per bio-based DM device). 1855 */ 1856 bioset_free(md->queue->bio_split); 1857 md->queue->bio_split = NULL; 1858 1859 if (type == DM_TYPE_DAX_BIO_BASED) 1860 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue); 1861 break; 1862 } 1863 1864 return 0; 1865 } 1866 1867 struct mapped_device *dm_get_md(dev_t dev) 1868 { 1869 struct mapped_device *md; 1870 unsigned minor = MINOR(dev); 1871 1872 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) 1873 return NULL; 1874 1875 spin_lock(&_minor_lock); 1876 1877 md = idr_find(&_minor_idr, minor); 1878 if (md) { 1879 if ((md == MINOR_ALLOCED || 1880 (MINOR(disk_devt(dm_disk(md))) != minor) || 1881 dm_deleting_md(md) || 1882 test_bit(DMF_FREEING, &md->flags))) { 1883 md = NULL; 1884 goto out; 1885 } 1886 dm_get(md); 1887 } 1888 1889 out: 1890 spin_unlock(&_minor_lock); 1891 1892 return md; 1893 } 1894 EXPORT_SYMBOL_GPL(dm_get_md); 1895 1896 void *dm_get_mdptr(struct mapped_device *md) 1897 { 1898 return md->interface_ptr; 1899 } 1900 1901 void dm_set_mdptr(struct mapped_device *md, void *ptr) 1902 { 1903 md->interface_ptr = ptr; 1904 } 1905 1906 void dm_get(struct mapped_device *md) 1907 { 1908 atomic_inc(&md->holders); 1909 BUG_ON(test_bit(DMF_FREEING, &md->flags)); 1910 } 1911 1912 int dm_hold(struct mapped_device *md) 1913 { 1914 spin_lock(&_minor_lock); 1915 if (test_bit(DMF_FREEING, &md->flags)) { 1916 spin_unlock(&_minor_lock); 1917 return -EBUSY; 1918 } 1919 dm_get(md); 1920 spin_unlock(&_minor_lock); 1921 return 0; 1922 } 1923 EXPORT_SYMBOL_GPL(dm_hold); 1924 1925 const char *dm_device_name(struct mapped_device *md) 1926 { 1927 return md->name; 1928 } 1929 EXPORT_SYMBOL_GPL(dm_device_name); 1930 1931 static void __dm_destroy(struct mapped_device *md, bool wait) 1932 { 1933 struct request_queue *q = dm_get_md_queue(md); 1934 struct dm_table *map; 1935 int srcu_idx; 1936 1937 might_sleep(); 1938 1939 spin_lock(&_minor_lock); 1940 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); 1941 set_bit(DMF_FREEING, &md->flags); 1942 spin_unlock(&_minor_lock); 1943 1944 blk_set_queue_dying(q); 1945 1946 if (dm_request_based(md) && md->kworker_task) 1947 kthread_flush_worker(&md->kworker); 1948 1949 /* 1950 * Take suspend_lock so that presuspend and postsuspend methods 1951 * do not race with internal suspend. 1952 */ 1953 mutex_lock(&md->suspend_lock); 1954 map = dm_get_live_table(md, &srcu_idx); 1955 if (!dm_suspended_md(md)) { 1956 dm_table_presuspend_targets(map); 1957 dm_table_postsuspend_targets(map); 1958 } 1959 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ 1960 dm_put_live_table(md, srcu_idx); 1961 mutex_unlock(&md->suspend_lock); 1962 1963 /* 1964 * Rare, but there may be I/O requests still going to complete, 1965 * for example. Wait for all references to disappear. 1966 * No one should increment the reference count of the mapped_device, 1967 * after the mapped_device state becomes DMF_FREEING. 1968 */ 1969 if (wait) 1970 while (atomic_read(&md->holders)) 1971 msleep(1); 1972 else if (atomic_read(&md->holders)) 1973 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", 1974 dm_device_name(md), atomic_read(&md->holders)); 1975 1976 dm_sysfs_exit(md); 1977 dm_table_destroy(__unbind(md)); 1978 free_dev(md); 1979 } 1980 1981 void dm_destroy(struct mapped_device *md) 1982 { 1983 __dm_destroy(md, true); 1984 } 1985 1986 void dm_destroy_immediate(struct mapped_device *md) 1987 { 1988 __dm_destroy(md, false); 1989 } 1990 1991 void dm_put(struct mapped_device *md) 1992 { 1993 atomic_dec(&md->holders); 1994 } 1995 EXPORT_SYMBOL_GPL(dm_put); 1996 1997 static int dm_wait_for_completion(struct mapped_device *md, long task_state) 1998 { 1999 int r = 0; 2000 DEFINE_WAIT(wait); 2001 2002 while (1) { 2003 prepare_to_wait(&md->wait, &wait, task_state); 2004 2005 if (!md_in_flight(md)) 2006 break; 2007 2008 if (signal_pending_state(task_state, current)) { 2009 r = -EINTR; 2010 break; 2011 } 2012 2013 io_schedule(); 2014 } 2015 finish_wait(&md->wait, &wait); 2016 2017 return r; 2018 } 2019 2020 /* 2021 * Process the deferred bios 2022 */ 2023 static void dm_wq_work(struct work_struct *work) 2024 { 2025 struct mapped_device *md = container_of(work, struct mapped_device, 2026 work); 2027 struct bio *c; 2028 int srcu_idx; 2029 struct dm_table *map; 2030 2031 map = dm_get_live_table(md, &srcu_idx); 2032 2033 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 2034 spin_lock_irq(&md->deferred_lock); 2035 c = bio_list_pop(&md->deferred); 2036 spin_unlock_irq(&md->deferred_lock); 2037 2038 if (!c) 2039 break; 2040 2041 if (dm_request_based(md)) 2042 generic_make_request(c); 2043 else 2044 __split_and_process_bio(md, map, c); 2045 } 2046 2047 dm_put_live_table(md, srcu_idx); 2048 } 2049 2050 static void dm_queue_flush(struct mapped_device *md) 2051 { 2052 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2053 smp_mb__after_atomic(); 2054 queue_work(md->wq, &md->work); 2055 } 2056 2057 /* 2058 * Swap in a new table, returning the old one for the caller to destroy. 2059 */ 2060 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) 2061 { 2062 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); 2063 struct queue_limits limits; 2064 int r; 2065 2066 mutex_lock(&md->suspend_lock); 2067 2068 /* device must be suspended */ 2069 if (!dm_suspended_md(md)) 2070 goto out; 2071 2072 /* 2073 * If the new table has no data devices, retain the existing limits. 2074 * This helps multipath with queue_if_no_path if all paths disappear, 2075 * then new I/O is queued based on these limits, and then some paths 2076 * reappear. 2077 */ 2078 if (dm_table_has_no_data_devices(table)) { 2079 live_map = dm_get_live_table_fast(md); 2080 if (live_map) 2081 limits = md->queue->limits; 2082 dm_put_live_table_fast(md); 2083 } 2084 2085 if (!live_map) { 2086 r = dm_calculate_queue_limits(table, &limits); 2087 if (r) { 2088 map = ERR_PTR(r); 2089 goto out; 2090 } 2091 } 2092 2093 map = __bind(md, table, &limits); 2094 2095 out: 2096 mutex_unlock(&md->suspend_lock); 2097 return map; 2098 } 2099 2100 /* 2101 * Functions to lock and unlock any filesystem running on the 2102 * device. 2103 */ 2104 static int lock_fs(struct mapped_device *md) 2105 { 2106 int r; 2107 2108 WARN_ON(md->frozen_sb); 2109 2110 md->frozen_sb = freeze_bdev(md->bdev); 2111 if (IS_ERR(md->frozen_sb)) { 2112 r = PTR_ERR(md->frozen_sb); 2113 md->frozen_sb = NULL; 2114 return r; 2115 } 2116 2117 set_bit(DMF_FROZEN, &md->flags); 2118 2119 return 0; 2120 } 2121 2122 static void unlock_fs(struct mapped_device *md) 2123 { 2124 if (!test_bit(DMF_FROZEN, &md->flags)) 2125 return; 2126 2127 thaw_bdev(md->bdev, md->frozen_sb); 2128 md->frozen_sb = NULL; 2129 clear_bit(DMF_FROZEN, &md->flags); 2130 } 2131 2132 /* 2133 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG 2134 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE 2135 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY 2136 * 2137 * If __dm_suspend returns 0, the device is completely quiescent 2138 * now. There is no request-processing activity. All new requests 2139 * are being added to md->deferred list. 2140 * 2141 * Caller must hold md->suspend_lock 2142 */ 2143 static int __dm_suspend(struct mapped_device *md, struct dm_table *map, 2144 unsigned suspend_flags, long task_state, 2145 int dmf_suspended_flag) 2146 { 2147 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; 2148 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; 2149 int r; 2150 2151 lockdep_assert_held(&md->suspend_lock); 2152 2153 /* 2154 * DMF_NOFLUSH_SUSPENDING must be set before presuspend. 2155 * This flag is cleared before dm_suspend returns. 2156 */ 2157 if (noflush) 2158 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 2159 2160 /* 2161 * This gets reverted if there's an error later and the targets 2162 * provide the .presuspend_undo hook. 2163 */ 2164 dm_table_presuspend_targets(map); 2165 2166 /* 2167 * Flush I/O to the device. 2168 * Any I/O submitted after lock_fs() may not be flushed. 2169 * noflush takes precedence over do_lockfs. 2170 * (lock_fs() flushes I/Os and waits for them to complete.) 2171 */ 2172 if (!noflush && do_lockfs) { 2173 r = lock_fs(md); 2174 if (r) { 2175 dm_table_presuspend_undo_targets(map); 2176 return r; 2177 } 2178 } 2179 2180 /* 2181 * Here we must make sure that no processes are submitting requests 2182 * to target drivers i.e. no one may be executing 2183 * __split_and_process_bio. This is called from dm_request and 2184 * dm_wq_work. 2185 * 2186 * To get all processes out of __split_and_process_bio in dm_request, 2187 * we take the write lock. To prevent any process from reentering 2188 * __split_and_process_bio from dm_request and quiesce the thread 2189 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call 2190 * flush_workqueue(md->wq). 2191 */ 2192 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2193 if (map) 2194 synchronize_srcu(&md->io_barrier); 2195 2196 /* 2197 * Stop md->queue before flushing md->wq in case request-based 2198 * dm defers requests to md->wq from md->queue. 2199 */ 2200 if (dm_request_based(md)) { 2201 dm_stop_queue(md->queue); 2202 if (md->kworker_task) 2203 kthread_flush_worker(&md->kworker); 2204 } 2205 2206 flush_workqueue(md->wq); 2207 2208 /* 2209 * At this point no more requests are entering target request routines. 2210 * We call dm_wait_for_completion to wait for all existing requests 2211 * to finish. 2212 */ 2213 r = dm_wait_for_completion(md, task_state); 2214 if (!r) 2215 set_bit(dmf_suspended_flag, &md->flags); 2216 2217 if (noflush) 2218 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 2219 if (map) 2220 synchronize_srcu(&md->io_barrier); 2221 2222 /* were we interrupted ? */ 2223 if (r < 0) { 2224 dm_queue_flush(md); 2225 2226 if (dm_request_based(md)) 2227 dm_start_queue(md->queue); 2228 2229 unlock_fs(md); 2230 dm_table_presuspend_undo_targets(map); 2231 /* pushback list is already flushed, so skip flush */ 2232 } 2233 2234 return r; 2235 } 2236 2237 /* 2238 * We need to be able to change a mapping table under a mounted 2239 * filesystem. For example we might want to move some data in 2240 * the background. Before the table can be swapped with 2241 * dm_bind_table, dm_suspend must be called to flush any in 2242 * flight bios and ensure that any further io gets deferred. 2243 */ 2244 /* 2245 * Suspend mechanism in request-based dm. 2246 * 2247 * 1. Flush all I/Os by lock_fs() if needed. 2248 * 2. Stop dispatching any I/O by stopping the request_queue. 2249 * 3. Wait for all in-flight I/Os to be completed or requeued. 2250 * 2251 * To abort suspend, start the request_queue. 2252 */ 2253 int dm_suspend(struct mapped_device *md, unsigned suspend_flags) 2254 { 2255 struct dm_table *map = NULL; 2256 int r = 0; 2257 2258 retry: 2259 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); 2260 2261 if (dm_suspended_md(md)) { 2262 r = -EINVAL; 2263 goto out_unlock; 2264 } 2265 2266 if (dm_suspended_internally_md(md)) { 2267 /* already internally suspended, wait for internal resume */ 2268 mutex_unlock(&md->suspend_lock); 2269 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); 2270 if (r) 2271 return r; 2272 goto retry; 2273 } 2274 2275 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 2276 2277 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); 2278 if (r) 2279 goto out_unlock; 2280 2281 dm_table_postsuspend_targets(map); 2282 2283 out_unlock: 2284 mutex_unlock(&md->suspend_lock); 2285 return r; 2286 } 2287 2288 static int __dm_resume(struct mapped_device *md, struct dm_table *map) 2289 { 2290 if (map) { 2291 int r = dm_table_resume_targets(map); 2292 if (r) 2293 return r; 2294 } 2295 2296 dm_queue_flush(md); 2297 2298 /* 2299 * Flushing deferred I/Os must be done after targets are resumed 2300 * so that mapping of targets can work correctly. 2301 * Request-based dm is queueing the deferred I/Os in its request_queue. 2302 */ 2303 if (dm_request_based(md)) 2304 dm_start_queue(md->queue); 2305 2306 unlock_fs(md); 2307 2308 return 0; 2309 } 2310 2311 int dm_resume(struct mapped_device *md) 2312 { 2313 int r; 2314 struct dm_table *map = NULL; 2315 2316 retry: 2317 r = -EINVAL; 2318 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); 2319 2320 if (!dm_suspended_md(md)) 2321 goto out; 2322 2323 if (dm_suspended_internally_md(md)) { 2324 /* already internally suspended, wait for internal resume */ 2325 mutex_unlock(&md->suspend_lock); 2326 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); 2327 if (r) 2328 return r; 2329 goto retry; 2330 } 2331 2332 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 2333 if (!map || !dm_table_get_size(map)) 2334 goto out; 2335 2336 r = __dm_resume(md, map); 2337 if (r) 2338 goto out; 2339 2340 clear_bit(DMF_SUSPENDED, &md->flags); 2341 out: 2342 mutex_unlock(&md->suspend_lock); 2343 2344 return r; 2345 } 2346 2347 /* 2348 * Internal suspend/resume works like userspace-driven suspend. It waits 2349 * until all bios finish and prevents issuing new bios to the target drivers. 2350 * It may be used only from the kernel. 2351 */ 2352 2353 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) 2354 { 2355 struct dm_table *map = NULL; 2356 2357 if (md->internal_suspend_count++) 2358 return; /* nested internal suspend */ 2359 2360 if (dm_suspended_md(md)) { 2361 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); 2362 return; /* nest suspend */ 2363 } 2364 2365 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); 2366 2367 /* 2368 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is 2369 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend 2370 * would require changing .presuspend to return an error -- avoid this 2371 * until there is a need for more elaborate variants of internal suspend. 2372 */ 2373 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, 2374 DMF_SUSPENDED_INTERNALLY); 2375 2376 dm_table_postsuspend_targets(map); 2377 } 2378 2379 static void __dm_internal_resume(struct mapped_device *md) 2380 { 2381 BUG_ON(!md->internal_suspend_count); 2382 2383 if (--md->internal_suspend_count) 2384 return; /* resume from nested internal suspend */ 2385 2386 if (dm_suspended_md(md)) 2387 goto done; /* resume from nested suspend */ 2388 2389 /* 2390 * NOTE: existing callers don't need to call dm_table_resume_targets 2391 * (which may fail -- so best to avoid it for now by passing NULL map) 2392 */ 2393 (void) __dm_resume(md, NULL); 2394 2395 done: 2396 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); 2397 smp_mb__after_atomic(); 2398 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); 2399 } 2400 2401 void dm_internal_suspend_noflush(struct mapped_device *md) 2402 { 2403 mutex_lock(&md->suspend_lock); 2404 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); 2405 mutex_unlock(&md->suspend_lock); 2406 } 2407 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); 2408 2409 void dm_internal_resume(struct mapped_device *md) 2410 { 2411 mutex_lock(&md->suspend_lock); 2412 __dm_internal_resume(md); 2413 mutex_unlock(&md->suspend_lock); 2414 } 2415 EXPORT_SYMBOL_GPL(dm_internal_resume); 2416 2417 /* 2418 * Fast variants of internal suspend/resume hold md->suspend_lock, 2419 * which prevents interaction with userspace-driven suspend. 2420 */ 2421 2422 void dm_internal_suspend_fast(struct mapped_device *md) 2423 { 2424 mutex_lock(&md->suspend_lock); 2425 if (dm_suspended_md(md) || dm_suspended_internally_md(md)) 2426 return; 2427 2428 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2429 synchronize_srcu(&md->io_barrier); 2430 flush_workqueue(md->wq); 2431 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); 2432 } 2433 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); 2434 2435 void dm_internal_resume_fast(struct mapped_device *md) 2436 { 2437 if (dm_suspended_md(md) || dm_suspended_internally_md(md)) 2438 goto done; 2439 2440 dm_queue_flush(md); 2441 2442 done: 2443 mutex_unlock(&md->suspend_lock); 2444 } 2445 EXPORT_SYMBOL_GPL(dm_internal_resume_fast); 2446 2447 /*----------------------------------------------------------------- 2448 * Event notification. 2449 *---------------------------------------------------------------*/ 2450 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, 2451 unsigned cookie) 2452 { 2453 char udev_cookie[DM_COOKIE_LENGTH]; 2454 char *envp[] = { udev_cookie, NULL }; 2455 2456 if (!cookie) 2457 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); 2458 else { 2459 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", 2460 DM_COOKIE_ENV_VAR_NAME, cookie); 2461 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, 2462 action, envp); 2463 } 2464 } 2465 2466 uint32_t dm_next_uevent_seq(struct mapped_device *md) 2467 { 2468 return atomic_add_return(1, &md->uevent_seq); 2469 } 2470 2471 uint32_t dm_get_event_nr(struct mapped_device *md) 2472 { 2473 return atomic_read(&md->event_nr); 2474 } 2475 2476 int dm_wait_event(struct mapped_device *md, int event_nr) 2477 { 2478 return wait_event_interruptible(md->eventq, 2479 (event_nr != atomic_read(&md->event_nr))); 2480 } 2481 2482 void dm_uevent_add(struct mapped_device *md, struct list_head *elist) 2483 { 2484 unsigned long flags; 2485 2486 spin_lock_irqsave(&md->uevent_lock, flags); 2487 list_add(elist, &md->uevent_list); 2488 spin_unlock_irqrestore(&md->uevent_lock, flags); 2489 } 2490 2491 /* 2492 * The gendisk is only valid as long as you have a reference 2493 * count on 'md'. 2494 */ 2495 struct gendisk *dm_disk(struct mapped_device *md) 2496 { 2497 return md->disk; 2498 } 2499 EXPORT_SYMBOL_GPL(dm_disk); 2500 2501 struct kobject *dm_kobject(struct mapped_device *md) 2502 { 2503 return &md->kobj_holder.kobj; 2504 } 2505 2506 struct mapped_device *dm_get_from_kobject(struct kobject *kobj) 2507 { 2508 struct mapped_device *md; 2509 2510 md = container_of(kobj, struct mapped_device, kobj_holder.kobj); 2511 2512 if (test_bit(DMF_FREEING, &md->flags) || 2513 dm_deleting_md(md)) 2514 return NULL; 2515 2516 dm_get(md); 2517 return md; 2518 } 2519 2520 int dm_suspended_md(struct mapped_device *md) 2521 { 2522 return test_bit(DMF_SUSPENDED, &md->flags); 2523 } 2524 2525 int dm_suspended_internally_md(struct mapped_device *md) 2526 { 2527 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); 2528 } 2529 2530 int dm_test_deferred_remove_flag(struct mapped_device *md) 2531 { 2532 return test_bit(DMF_DEFERRED_REMOVE, &md->flags); 2533 } 2534 2535 int dm_suspended(struct dm_target *ti) 2536 { 2537 return dm_suspended_md(dm_table_get_md(ti->table)); 2538 } 2539 EXPORT_SYMBOL_GPL(dm_suspended); 2540 2541 int dm_noflush_suspending(struct dm_target *ti) 2542 { 2543 return __noflush_suspending(dm_table_get_md(ti->table)); 2544 } 2545 EXPORT_SYMBOL_GPL(dm_noflush_suspending); 2546 2547 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, 2548 unsigned integrity, unsigned per_io_data_size) 2549 { 2550 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); 2551 unsigned int pool_size = 0; 2552 unsigned int front_pad; 2553 2554 if (!pools) 2555 return NULL; 2556 2557 switch (type) { 2558 case DM_TYPE_BIO_BASED: 2559 case DM_TYPE_DAX_BIO_BASED: 2560 pool_size = dm_get_reserved_bio_based_ios(); 2561 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); 2562 2563 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); 2564 if (!pools->io_pool) 2565 goto out; 2566 break; 2567 case DM_TYPE_REQUEST_BASED: 2568 case DM_TYPE_MQ_REQUEST_BASED: 2569 pool_size = dm_get_reserved_rq_based_ios(); 2570 front_pad = offsetof(struct dm_rq_clone_bio_info, clone); 2571 /* per_io_data_size is used for blk-mq pdu at queue allocation */ 2572 break; 2573 default: 2574 BUG(); 2575 } 2576 2577 pools->bs = bioset_create_nobvec(pool_size, front_pad); 2578 if (!pools->bs) 2579 goto out; 2580 2581 if (integrity && bioset_integrity_create(pools->bs, pool_size)) 2582 goto out; 2583 2584 return pools; 2585 2586 out: 2587 dm_free_md_mempools(pools); 2588 2589 return NULL; 2590 } 2591 2592 void dm_free_md_mempools(struct dm_md_mempools *pools) 2593 { 2594 if (!pools) 2595 return; 2596 2597 mempool_destroy(pools->io_pool); 2598 2599 if (pools->bs) 2600 bioset_free(pools->bs); 2601 2602 kfree(pools); 2603 } 2604 2605 struct dm_pr { 2606 u64 old_key; 2607 u64 new_key; 2608 u32 flags; 2609 bool fail_early; 2610 }; 2611 2612 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, 2613 void *data) 2614 { 2615 struct mapped_device *md = bdev->bd_disk->private_data; 2616 struct dm_table *table; 2617 struct dm_target *ti; 2618 int ret = -ENOTTY, srcu_idx; 2619 2620 table = dm_get_live_table(md, &srcu_idx); 2621 if (!table || !dm_table_get_size(table)) 2622 goto out; 2623 2624 /* We only support devices that have a single target */ 2625 if (dm_table_get_num_targets(table) != 1) 2626 goto out; 2627 ti = dm_table_get_target(table, 0); 2628 2629 ret = -EINVAL; 2630 if (!ti->type->iterate_devices) 2631 goto out; 2632 2633 ret = ti->type->iterate_devices(ti, fn, data); 2634 out: 2635 dm_put_live_table(md, srcu_idx); 2636 return ret; 2637 } 2638 2639 /* 2640 * For register / unregister we need to manually call out to every path. 2641 */ 2642 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, 2643 sector_t start, sector_t len, void *data) 2644 { 2645 struct dm_pr *pr = data; 2646 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; 2647 2648 if (!ops || !ops->pr_register) 2649 return -EOPNOTSUPP; 2650 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); 2651 } 2652 2653 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, 2654 u32 flags) 2655 { 2656 struct dm_pr pr = { 2657 .old_key = old_key, 2658 .new_key = new_key, 2659 .flags = flags, 2660 .fail_early = true, 2661 }; 2662 int ret; 2663 2664 ret = dm_call_pr(bdev, __dm_pr_register, &pr); 2665 if (ret && new_key) { 2666 /* unregister all paths if we failed to register any path */ 2667 pr.old_key = new_key; 2668 pr.new_key = 0; 2669 pr.flags = 0; 2670 pr.fail_early = false; 2671 dm_call_pr(bdev, __dm_pr_register, &pr); 2672 } 2673 2674 return ret; 2675 } 2676 2677 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, 2678 u32 flags) 2679 { 2680 struct mapped_device *md = bdev->bd_disk->private_data; 2681 const struct pr_ops *ops; 2682 fmode_t mode; 2683 int r; 2684 2685 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); 2686 if (r < 0) 2687 return r; 2688 2689 ops = bdev->bd_disk->fops->pr_ops; 2690 if (ops && ops->pr_reserve) 2691 r = ops->pr_reserve(bdev, key, type, flags); 2692 else 2693 r = -EOPNOTSUPP; 2694 2695 bdput(bdev); 2696 return r; 2697 } 2698 2699 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) 2700 { 2701 struct mapped_device *md = bdev->bd_disk->private_data; 2702 const struct pr_ops *ops; 2703 fmode_t mode; 2704 int r; 2705 2706 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); 2707 if (r < 0) 2708 return r; 2709 2710 ops = bdev->bd_disk->fops->pr_ops; 2711 if (ops && ops->pr_release) 2712 r = ops->pr_release(bdev, key, type); 2713 else 2714 r = -EOPNOTSUPP; 2715 2716 bdput(bdev); 2717 return r; 2718 } 2719 2720 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, 2721 enum pr_type type, bool abort) 2722 { 2723 struct mapped_device *md = bdev->bd_disk->private_data; 2724 const struct pr_ops *ops; 2725 fmode_t mode; 2726 int r; 2727 2728 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); 2729 if (r < 0) 2730 return r; 2731 2732 ops = bdev->bd_disk->fops->pr_ops; 2733 if (ops && ops->pr_preempt) 2734 r = ops->pr_preempt(bdev, old_key, new_key, type, abort); 2735 else 2736 r = -EOPNOTSUPP; 2737 2738 bdput(bdev); 2739 return r; 2740 } 2741 2742 static int dm_pr_clear(struct block_device *bdev, u64 key) 2743 { 2744 struct mapped_device *md = bdev->bd_disk->private_data; 2745 const struct pr_ops *ops; 2746 fmode_t mode; 2747 int r; 2748 2749 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); 2750 if (r < 0) 2751 return r; 2752 2753 ops = bdev->bd_disk->fops->pr_ops; 2754 if (ops && ops->pr_clear) 2755 r = ops->pr_clear(bdev, key); 2756 else 2757 r = -EOPNOTSUPP; 2758 2759 bdput(bdev); 2760 return r; 2761 } 2762 2763 static const struct pr_ops dm_pr_ops = { 2764 .pr_register = dm_pr_register, 2765 .pr_reserve = dm_pr_reserve, 2766 .pr_release = dm_pr_release, 2767 .pr_preempt = dm_pr_preempt, 2768 .pr_clear = dm_pr_clear, 2769 }; 2770 2771 static const struct block_device_operations dm_blk_dops = { 2772 .open = dm_blk_open, 2773 .release = dm_blk_close, 2774 .ioctl = dm_blk_ioctl, 2775 .direct_access = dm_blk_direct_access, 2776 .getgeo = dm_blk_getgeo, 2777 .pr_ops = &dm_pr_ops, 2778 .owner = THIS_MODULE 2779 }; 2780 2781 /* 2782 * module hooks 2783 */ 2784 module_init(dm_init); 2785 module_exit(dm_exit); 2786 2787 module_param(major, uint, 0); 2788 MODULE_PARM_DESC(major, "The major number of the device mapper"); 2789 2790 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); 2791 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); 2792 2793 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); 2794 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); 2795 2796 MODULE_DESCRIPTION(DM_NAME " driver"); 2797 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); 2798 MODULE_LICENSE("GPL"); 2799