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.h" 9 #include "dm-uevent.h" 10 11 #include <linux/init.h> 12 #include <linux/module.h> 13 #include <linux/mutex.h> 14 #include <linux/moduleparam.h> 15 #include <linux/blkpg.h> 16 #include <linux/bio.h> 17 #include <linux/buffer_head.h> 18 #include <linux/mempool.h> 19 #include <linux/slab.h> 20 #include <linux/idr.h> 21 #include <linux/hdreg.h> 22 23 #include <trace/events/block.h> 24 25 #define DM_MSG_PREFIX "core" 26 27 static const char *_name = DM_NAME; 28 29 static unsigned int major = 0; 30 static unsigned int _major = 0; 31 32 static DEFINE_SPINLOCK(_minor_lock); 33 /* 34 * For bio-based dm. 35 * One of these is allocated per bio. 36 */ 37 struct dm_io { 38 struct mapped_device *md; 39 int error; 40 atomic_t io_count; 41 struct bio *bio; 42 unsigned long start_time; 43 }; 44 45 /* 46 * For bio-based dm. 47 * One of these is allocated per target within a bio. Hopefully 48 * this will be simplified out one day. 49 */ 50 struct dm_target_io { 51 struct dm_io *io; 52 struct dm_target *ti; 53 union map_info info; 54 }; 55 56 /* 57 * For request-based dm. 58 * One of these is allocated per request. 59 */ 60 struct dm_rq_target_io { 61 struct mapped_device *md; 62 struct dm_target *ti; 63 struct request *orig, clone; 64 int error; 65 union map_info info; 66 }; 67 68 /* 69 * For request-based dm. 70 * One of these is allocated per bio. 71 */ 72 struct dm_rq_clone_bio_info { 73 struct bio *orig; 74 struct request *rq; 75 }; 76 77 union map_info *dm_get_mapinfo(struct bio *bio) 78 { 79 if (bio && bio->bi_private) 80 return &((struct dm_target_io *)bio->bi_private)->info; 81 return NULL; 82 } 83 84 #define MINOR_ALLOCED ((void *)-1) 85 86 /* 87 * Bits for the md->flags field. 88 */ 89 #define DMF_BLOCK_IO_FOR_SUSPEND 0 90 #define DMF_SUSPENDED 1 91 #define DMF_FROZEN 2 92 #define DMF_FREEING 3 93 #define DMF_DELETING 4 94 #define DMF_NOFLUSH_SUSPENDING 5 95 #define DMF_QUEUE_IO_TO_THREAD 6 96 97 /* 98 * Work processed by per-device workqueue. 99 */ 100 struct mapped_device { 101 struct rw_semaphore io_lock; 102 struct mutex suspend_lock; 103 rwlock_t map_lock; 104 atomic_t holders; 105 atomic_t open_count; 106 107 unsigned long flags; 108 109 struct request_queue *queue; 110 struct gendisk *disk; 111 char name[16]; 112 113 void *interface_ptr; 114 115 /* 116 * A list of ios that arrived while we were suspended. 117 */ 118 atomic_t pending; 119 wait_queue_head_t wait; 120 struct work_struct work; 121 struct bio_list deferred; 122 spinlock_t deferred_lock; 123 124 /* 125 * An error from the barrier request currently being processed. 126 */ 127 int barrier_error; 128 129 /* 130 * Processing queue (flush/barriers) 131 */ 132 struct workqueue_struct *wq; 133 134 /* 135 * The current mapping. 136 */ 137 struct dm_table *map; 138 139 /* 140 * io objects are allocated from here. 141 */ 142 mempool_t *io_pool; 143 mempool_t *tio_pool; 144 145 struct bio_set *bs; 146 147 /* 148 * Event handling. 149 */ 150 atomic_t event_nr; 151 wait_queue_head_t eventq; 152 atomic_t uevent_seq; 153 struct list_head uevent_list; 154 spinlock_t uevent_lock; /* Protect access to uevent_list */ 155 156 /* 157 * freeze/thaw support require holding onto a super block 158 */ 159 struct super_block *frozen_sb; 160 struct block_device *bdev; 161 162 /* forced geometry settings */ 163 struct hd_geometry geometry; 164 165 /* sysfs handle */ 166 struct kobject kobj; 167 }; 168 169 #define MIN_IOS 256 170 static struct kmem_cache *_io_cache; 171 static struct kmem_cache *_tio_cache; 172 static struct kmem_cache *_rq_tio_cache; 173 static struct kmem_cache *_rq_bio_info_cache; 174 175 static int __init local_init(void) 176 { 177 int r = -ENOMEM; 178 179 /* allocate a slab for the dm_ios */ 180 _io_cache = KMEM_CACHE(dm_io, 0); 181 if (!_io_cache) 182 return r; 183 184 /* allocate a slab for the target ios */ 185 _tio_cache = KMEM_CACHE(dm_target_io, 0); 186 if (!_tio_cache) 187 goto out_free_io_cache; 188 189 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); 190 if (!_rq_tio_cache) 191 goto out_free_tio_cache; 192 193 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); 194 if (!_rq_bio_info_cache) 195 goto out_free_rq_tio_cache; 196 197 r = dm_uevent_init(); 198 if (r) 199 goto out_free_rq_bio_info_cache; 200 201 _major = major; 202 r = register_blkdev(_major, _name); 203 if (r < 0) 204 goto out_uevent_exit; 205 206 if (!_major) 207 _major = r; 208 209 return 0; 210 211 out_uevent_exit: 212 dm_uevent_exit(); 213 out_free_rq_bio_info_cache: 214 kmem_cache_destroy(_rq_bio_info_cache); 215 out_free_rq_tio_cache: 216 kmem_cache_destroy(_rq_tio_cache); 217 out_free_tio_cache: 218 kmem_cache_destroy(_tio_cache); 219 out_free_io_cache: 220 kmem_cache_destroy(_io_cache); 221 222 return r; 223 } 224 225 static void local_exit(void) 226 { 227 kmem_cache_destroy(_rq_bio_info_cache); 228 kmem_cache_destroy(_rq_tio_cache); 229 kmem_cache_destroy(_tio_cache); 230 kmem_cache_destroy(_io_cache); 231 unregister_blkdev(_major, _name); 232 dm_uevent_exit(); 233 234 _major = 0; 235 236 DMINFO("cleaned up"); 237 } 238 239 static int (*_inits[])(void) __initdata = { 240 local_init, 241 dm_target_init, 242 dm_linear_init, 243 dm_stripe_init, 244 dm_kcopyd_init, 245 dm_interface_init, 246 }; 247 248 static void (*_exits[])(void) = { 249 local_exit, 250 dm_target_exit, 251 dm_linear_exit, 252 dm_stripe_exit, 253 dm_kcopyd_exit, 254 dm_interface_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 /* 287 * Block device functions 288 */ 289 static int dm_blk_open(struct block_device *bdev, fmode_t mode) 290 { 291 struct mapped_device *md; 292 293 spin_lock(&_minor_lock); 294 295 md = bdev->bd_disk->private_data; 296 if (!md) 297 goto out; 298 299 if (test_bit(DMF_FREEING, &md->flags) || 300 test_bit(DMF_DELETING, &md->flags)) { 301 md = NULL; 302 goto out; 303 } 304 305 dm_get(md); 306 atomic_inc(&md->open_count); 307 308 out: 309 spin_unlock(&_minor_lock); 310 311 return md ? 0 : -ENXIO; 312 } 313 314 static int dm_blk_close(struct gendisk *disk, fmode_t mode) 315 { 316 struct mapped_device *md = disk->private_data; 317 atomic_dec(&md->open_count); 318 dm_put(md); 319 return 0; 320 } 321 322 int dm_open_count(struct mapped_device *md) 323 { 324 return atomic_read(&md->open_count); 325 } 326 327 /* 328 * Guarantees nothing is using the device before it's deleted. 329 */ 330 int dm_lock_for_deletion(struct mapped_device *md) 331 { 332 int r = 0; 333 334 spin_lock(&_minor_lock); 335 336 if (dm_open_count(md)) 337 r = -EBUSY; 338 else 339 set_bit(DMF_DELETING, &md->flags); 340 341 spin_unlock(&_minor_lock); 342 343 return r; 344 } 345 346 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) 347 { 348 struct mapped_device *md = bdev->bd_disk->private_data; 349 350 return dm_get_geometry(md, geo); 351 } 352 353 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, 354 unsigned int cmd, unsigned long arg) 355 { 356 struct mapped_device *md = bdev->bd_disk->private_data; 357 struct dm_table *map = dm_get_table(md); 358 struct dm_target *tgt; 359 int r = -ENOTTY; 360 361 if (!map || !dm_table_get_size(map)) 362 goto out; 363 364 /* We only support devices that have a single target */ 365 if (dm_table_get_num_targets(map) != 1) 366 goto out; 367 368 tgt = dm_table_get_target(map, 0); 369 370 if (dm_suspended(md)) { 371 r = -EAGAIN; 372 goto out; 373 } 374 375 if (tgt->type->ioctl) 376 r = tgt->type->ioctl(tgt, cmd, arg); 377 378 out: 379 dm_table_put(map); 380 381 return r; 382 } 383 384 static struct dm_io *alloc_io(struct mapped_device *md) 385 { 386 return mempool_alloc(md->io_pool, GFP_NOIO); 387 } 388 389 static void free_io(struct mapped_device *md, struct dm_io *io) 390 { 391 mempool_free(io, md->io_pool); 392 } 393 394 static struct dm_target_io *alloc_tio(struct mapped_device *md) 395 { 396 return mempool_alloc(md->tio_pool, GFP_NOIO); 397 } 398 399 static void free_tio(struct mapped_device *md, struct dm_target_io *tio) 400 { 401 mempool_free(tio, md->tio_pool); 402 } 403 404 static void start_io_acct(struct dm_io *io) 405 { 406 struct mapped_device *md = io->md; 407 int cpu; 408 409 io->start_time = jiffies; 410 411 cpu = part_stat_lock(); 412 part_round_stats(cpu, &dm_disk(md)->part0); 413 part_stat_unlock(); 414 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending); 415 } 416 417 static void end_io_acct(struct dm_io *io) 418 { 419 struct mapped_device *md = io->md; 420 struct bio *bio = io->bio; 421 unsigned long duration = jiffies - io->start_time; 422 int pending, cpu; 423 int rw = bio_data_dir(bio); 424 425 cpu = part_stat_lock(); 426 part_round_stats(cpu, &dm_disk(md)->part0); 427 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); 428 part_stat_unlock(); 429 430 /* 431 * After this is decremented the bio must not be touched if it is 432 * a barrier. 433 */ 434 dm_disk(md)->part0.in_flight = pending = 435 atomic_dec_return(&md->pending); 436 437 /* nudge anyone waiting on suspend queue */ 438 if (!pending) 439 wake_up(&md->wait); 440 } 441 442 /* 443 * Add the bio to the list of deferred io. 444 */ 445 static void queue_io(struct mapped_device *md, struct bio *bio) 446 { 447 down_write(&md->io_lock); 448 449 spin_lock_irq(&md->deferred_lock); 450 bio_list_add(&md->deferred, bio); 451 spin_unlock_irq(&md->deferred_lock); 452 453 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) 454 queue_work(md->wq, &md->work); 455 456 up_write(&md->io_lock); 457 } 458 459 /* 460 * Everyone (including functions in this file), should use this 461 * function to access the md->map field, and make sure they call 462 * dm_table_put() when finished. 463 */ 464 struct dm_table *dm_get_table(struct mapped_device *md) 465 { 466 struct dm_table *t; 467 468 read_lock(&md->map_lock); 469 t = md->map; 470 if (t) 471 dm_table_get(t); 472 read_unlock(&md->map_lock); 473 474 return t; 475 } 476 477 /* 478 * Get the geometry associated with a dm device 479 */ 480 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) 481 { 482 *geo = md->geometry; 483 484 return 0; 485 } 486 487 /* 488 * Set the geometry of a device. 489 */ 490 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) 491 { 492 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; 493 494 if (geo->start > sz) { 495 DMWARN("Start sector is beyond the geometry limits."); 496 return -EINVAL; 497 } 498 499 md->geometry = *geo; 500 501 return 0; 502 } 503 504 /*----------------------------------------------------------------- 505 * CRUD START: 506 * A more elegant soln is in the works that uses the queue 507 * merge fn, unfortunately there are a couple of changes to 508 * the block layer that I want to make for this. So in the 509 * interests of getting something for people to use I give 510 * you this clearly demarcated crap. 511 *---------------------------------------------------------------*/ 512 513 static int __noflush_suspending(struct mapped_device *md) 514 { 515 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 516 } 517 518 /* 519 * Decrements the number of outstanding ios that a bio has been 520 * cloned into, completing the original io if necc. 521 */ 522 static void dec_pending(struct dm_io *io, int error) 523 { 524 unsigned long flags; 525 int io_error; 526 struct bio *bio; 527 struct mapped_device *md = io->md; 528 529 /* Push-back supersedes any I/O errors */ 530 if (error && !(io->error > 0 && __noflush_suspending(md))) 531 io->error = error; 532 533 if (atomic_dec_and_test(&io->io_count)) { 534 if (io->error == DM_ENDIO_REQUEUE) { 535 /* 536 * Target requested pushing back the I/O. 537 */ 538 spin_lock_irqsave(&md->deferred_lock, flags); 539 if (__noflush_suspending(md)) { 540 if (!bio_barrier(io->bio)) 541 bio_list_add_head(&md->deferred, 542 io->bio); 543 } else 544 /* noflush suspend was interrupted. */ 545 io->error = -EIO; 546 spin_unlock_irqrestore(&md->deferred_lock, flags); 547 } 548 549 io_error = io->error; 550 bio = io->bio; 551 552 if (bio_barrier(bio)) { 553 /* 554 * There can be just one barrier request so we use 555 * a per-device variable for error reporting. 556 * Note that you can't touch the bio after end_io_acct 557 */ 558 if (!md->barrier_error && io_error != -EOPNOTSUPP) 559 md->barrier_error = io_error; 560 end_io_acct(io); 561 } else { 562 end_io_acct(io); 563 564 if (io_error != DM_ENDIO_REQUEUE) { 565 trace_block_bio_complete(md->queue, bio); 566 567 bio_endio(bio, io_error); 568 } 569 } 570 571 free_io(md, io); 572 } 573 } 574 575 static void clone_endio(struct bio *bio, int error) 576 { 577 int r = 0; 578 struct dm_target_io *tio = bio->bi_private; 579 struct dm_io *io = tio->io; 580 struct mapped_device *md = tio->io->md; 581 dm_endio_fn endio = tio->ti->type->end_io; 582 583 if (!bio_flagged(bio, BIO_UPTODATE) && !error) 584 error = -EIO; 585 586 if (endio) { 587 r = endio(tio->ti, bio, error, &tio->info); 588 if (r < 0 || r == DM_ENDIO_REQUEUE) 589 /* 590 * error and requeue request are handled 591 * in dec_pending(). 592 */ 593 error = r; 594 else if (r == DM_ENDIO_INCOMPLETE) 595 /* The target will handle the io */ 596 return; 597 else if (r) { 598 DMWARN("unimplemented target endio return value: %d", r); 599 BUG(); 600 } 601 } 602 603 /* 604 * Store md for cleanup instead of tio which is about to get freed. 605 */ 606 bio->bi_private = md->bs; 607 608 free_tio(md, tio); 609 bio_put(bio); 610 dec_pending(io, error); 611 } 612 613 static sector_t max_io_len(struct mapped_device *md, 614 sector_t sector, struct dm_target *ti) 615 { 616 sector_t offset = sector - ti->begin; 617 sector_t len = ti->len - offset; 618 619 /* 620 * Does the target need to split even further ? 621 */ 622 if (ti->split_io) { 623 sector_t boundary; 624 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) 625 - offset; 626 if (len > boundary) 627 len = boundary; 628 } 629 630 return len; 631 } 632 633 static void __map_bio(struct dm_target *ti, struct bio *clone, 634 struct dm_target_io *tio) 635 { 636 int r; 637 sector_t sector; 638 struct mapped_device *md; 639 640 clone->bi_end_io = clone_endio; 641 clone->bi_private = tio; 642 643 /* 644 * Map the clone. If r == 0 we don't need to do 645 * anything, the target has assumed ownership of 646 * this io. 647 */ 648 atomic_inc(&tio->io->io_count); 649 sector = clone->bi_sector; 650 r = ti->type->map(ti, clone, &tio->info); 651 if (r == DM_MAPIO_REMAPPED) { 652 /* the bio has been remapped so dispatch it */ 653 654 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone, 655 tio->io->bio->bi_bdev->bd_dev, sector); 656 657 generic_make_request(clone); 658 } else if (r < 0 || r == DM_MAPIO_REQUEUE) { 659 /* error the io and bail out, or requeue it if needed */ 660 md = tio->io->md; 661 dec_pending(tio->io, r); 662 /* 663 * Store bio_set for cleanup. 664 */ 665 clone->bi_private = md->bs; 666 bio_put(clone); 667 free_tio(md, tio); 668 } else if (r) { 669 DMWARN("unimplemented target map return value: %d", r); 670 BUG(); 671 } 672 } 673 674 struct clone_info { 675 struct mapped_device *md; 676 struct dm_table *map; 677 struct bio *bio; 678 struct dm_io *io; 679 sector_t sector; 680 sector_t sector_count; 681 unsigned short idx; 682 }; 683 684 static void dm_bio_destructor(struct bio *bio) 685 { 686 struct bio_set *bs = bio->bi_private; 687 688 bio_free(bio, bs); 689 } 690 691 /* 692 * Creates a little bio that is just does part of a bvec. 693 */ 694 static struct bio *split_bvec(struct bio *bio, sector_t sector, 695 unsigned short idx, unsigned int offset, 696 unsigned int len, struct bio_set *bs) 697 { 698 struct bio *clone; 699 struct bio_vec *bv = bio->bi_io_vec + idx; 700 701 clone = bio_alloc_bioset(GFP_NOIO, 1, bs); 702 clone->bi_destructor = dm_bio_destructor; 703 *clone->bi_io_vec = *bv; 704 705 clone->bi_sector = sector; 706 clone->bi_bdev = bio->bi_bdev; 707 clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER); 708 clone->bi_vcnt = 1; 709 clone->bi_size = to_bytes(len); 710 clone->bi_io_vec->bv_offset = offset; 711 clone->bi_io_vec->bv_len = clone->bi_size; 712 clone->bi_flags |= 1 << BIO_CLONED; 713 714 if (bio_integrity(bio)) { 715 bio_integrity_clone(clone, bio, GFP_NOIO); 716 bio_integrity_trim(clone, 717 bio_sector_offset(bio, idx, offset), len); 718 } 719 720 return clone; 721 } 722 723 /* 724 * Creates a bio that consists of range of complete bvecs. 725 */ 726 static struct bio *clone_bio(struct bio *bio, sector_t sector, 727 unsigned short idx, unsigned short bv_count, 728 unsigned int len, struct bio_set *bs) 729 { 730 struct bio *clone; 731 732 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); 733 __bio_clone(clone, bio); 734 clone->bi_rw &= ~(1 << BIO_RW_BARRIER); 735 clone->bi_destructor = dm_bio_destructor; 736 clone->bi_sector = sector; 737 clone->bi_idx = idx; 738 clone->bi_vcnt = idx + bv_count; 739 clone->bi_size = to_bytes(len); 740 clone->bi_flags &= ~(1 << BIO_SEG_VALID); 741 742 if (bio_integrity(bio)) { 743 bio_integrity_clone(clone, bio, GFP_NOIO); 744 745 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) 746 bio_integrity_trim(clone, 747 bio_sector_offset(bio, idx, 0), len); 748 } 749 750 return clone; 751 } 752 753 static void __flush_target(struct clone_info *ci, struct dm_target *ti, 754 unsigned flush_nr) 755 { 756 struct dm_target_io *tio = alloc_tio(ci->md); 757 struct bio *clone; 758 759 tio->io = ci->io; 760 tio->ti = ti; 761 762 memset(&tio->info, 0, sizeof(tio->info)); 763 tio->info.flush_request = flush_nr; 764 765 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); 766 __bio_clone(clone, ci->bio); 767 clone->bi_destructor = dm_bio_destructor; 768 769 __map_bio(ti, clone, tio); 770 } 771 772 static int __clone_and_map_empty_barrier(struct clone_info *ci) 773 { 774 unsigned target_nr = 0, flush_nr; 775 struct dm_target *ti; 776 777 while ((ti = dm_table_get_target(ci->map, target_nr++))) 778 for (flush_nr = 0; flush_nr < ti->num_flush_requests; 779 flush_nr++) 780 __flush_target(ci, ti, flush_nr); 781 782 ci->sector_count = 0; 783 784 return 0; 785 } 786 787 static int __clone_and_map(struct clone_info *ci) 788 { 789 struct bio *clone, *bio = ci->bio; 790 struct dm_target *ti; 791 sector_t len = 0, max; 792 struct dm_target_io *tio; 793 794 if (unlikely(bio_empty_barrier(bio))) 795 return __clone_and_map_empty_barrier(ci); 796 797 ti = dm_table_find_target(ci->map, ci->sector); 798 if (!dm_target_is_valid(ti)) 799 return -EIO; 800 801 max = max_io_len(ci->md, ci->sector, ti); 802 803 /* 804 * Allocate a target io object. 805 */ 806 tio = alloc_tio(ci->md); 807 tio->io = ci->io; 808 tio->ti = ti; 809 memset(&tio->info, 0, sizeof(tio->info)); 810 811 if (ci->sector_count <= max) { 812 /* 813 * Optimise for the simple case where we can do all of 814 * the remaining io with a single clone. 815 */ 816 clone = clone_bio(bio, ci->sector, ci->idx, 817 bio->bi_vcnt - ci->idx, ci->sector_count, 818 ci->md->bs); 819 __map_bio(ti, clone, tio); 820 ci->sector_count = 0; 821 822 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { 823 /* 824 * There are some bvecs that don't span targets. 825 * Do as many of these as possible. 826 */ 827 int i; 828 sector_t remaining = max; 829 sector_t bv_len; 830 831 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { 832 bv_len = to_sector(bio->bi_io_vec[i].bv_len); 833 834 if (bv_len > remaining) 835 break; 836 837 remaining -= bv_len; 838 len += bv_len; 839 } 840 841 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, 842 ci->md->bs); 843 __map_bio(ti, clone, tio); 844 845 ci->sector += len; 846 ci->sector_count -= len; 847 ci->idx = i; 848 849 } else { 850 /* 851 * Handle a bvec that must be split between two or more targets. 852 */ 853 struct bio_vec *bv = bio->bi_io_vec + ci->idx; 854 sector_t remaining = to_sector(bv->bv_len); 855 unsigned int offset = 0; 856 857 do { 858 if (offset) { 859 ti = dm_table_find_target(ci->map, ci->sector); 860 if (!dm_target_is_valid(ti)) 861 return -EIO; 862 863 max = max_io_len(ci->md, ci->sector, ti); 864 865 tio = alloc_tio(ci->md); 866 tio->io = ci->io; 867 tio->ti = ti; 868 memset(&tio->info, 0, sizeof(tio->info)); 869 } 870 871 len = min(remaining, max); 872 873 clone = split_bvec(bio, ci->sector, ci->idx, 874 bv->bv_offset + offset, len, 875 ci->md->bs); 876 877 __map_bio(ti, clone, tio); 878 879 ci->sector += len; 880 ci->sector_count -= len; 881 offset += to_bytes(len); 882 } while (remaining -= len); 883 884 ci->idx++; 885 } 886 887 return 0; 888 } 889 890 /* 891 * Split the bio into several clones and submit it to targets. 892 */ 893 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) 894 { 895 struct clone_info ci; 896 int error = 0; 897 898 ci.map = dm_get_table(md); 899 if (unlikely(!ci.map)) { 900 if (!bio_barrier(bio)) 901 bio_io_error(bio); 902 else 903 if (!md->barrier_error) 904 md->barrier_error = -EIO; 905 return; 906 } 907 908 ci.md = md; 909 ci.bio = bio; 910 ci.io = alloc_io(md); 911 ci.io->error = 0; 912 atomic_set(&ci.io->io_count, 1); 913 ci.io->bio = bio; 914 ci.io->md = md; 915 ci.sector = bio->bi_sector; 916 ci.sector_count = bio_sectors(bio); 917 if (unlikely(bio_empty_barrier(bio))) 918 ci.sector_count = 1; 919 ci.idx = bio->bi_idx; 920 921 start_io_acct(ci.io); 922 while (ci.sector_count && !error) 923 error = __clone_and_map(&ci); 924 925 /* drop the extra reference count */ 926 dec_pending(ci.io, error); 927 dm_table_put(ci.map); 928 } 929 /*----------------------------------------------------------------- 930 * CRUD END 931 *---------------------------------------------------------------*/ 932 933 static int dm_merge_bvec(struct request_queue *q, 934 struct bvec_merge_data *bvm, 935 struct bio_vec *biovec) 936 { 937 struct mapped_device *md = q->queuedata; 938 struct dm_table *map = dm_get_table(md); 939 struct dm_target *ti; 940 sector_t max_sectors; 941 int max_size = 0; 942 943 if (unlikely(!map)) 944 goto out; 945 946 ti = dm_table_find_target(map, bvm->bi_sector); 947 if (!dm_target_is_valid(ti)) 948 goto out_table; 949 950 /* 951 * Find maximum amount of I/O that won't need splitting 952 */ 953 max_sectors = min(max_io_len(md, bvm->bi_sector, ti), 954 (sector_t) BIO_MAX_SECTORS); 955 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; 956 if (max_size < 0) 957 max_size = 0; 958 959 /* 960 * merge_bvec_fn() returns number of bytes 961 * it can accept at this offset 962 * max is precomputed maximal io size 963 */ 964 if (max_size && ti->type->merge) 965 max_size = ti->type->merge(ti, bvm, biovec, max_size); 966 /* 967 * If the target doesn't support merge method and some of the devices 968 * provided their merge_bvec method (we know this by looking at 969 * queue_max_hw_sectors), then we can't allow bios with multiple vector 970 * entries. So always set max_size to 0, and the code below allows 971 * just one page. 972 */ 973 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) 974 975 max_size = 0; 976 977 out_table: 978 dm_table_put(map); 979 980 out: 981 /* 982 * Always allow an entire first page 983 */ 984 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) 985 max_size = biovec->bv_len; 986 987 return max_size; 988 } 989 990 /* 991 * The request function that just remaps the bio built up by 992 * dm_merge_bvec. 993 */ 994 static int dm_request(struct request_queue *q, struct bio *bio) 995 { 996 int rw = bio_data_dir(bio); 997 struct mapped_device *md = q->queuedata; 998 int cpu; 999 1000 down_read(&md->io_lock); 1001 1002 cpu = part_stat_lock(); 1003 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); 1004 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); 1005 part_stat_unlock(); 1006 1007 /* 1008 * If we're suspended or the thread is processing barriers 1009 * we have to queue this io for later. 1010 */ 1011 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) || 1012 unlikely(bio_barrier(bio))) { 1013 up_read(&md->io_lock); 1014 1015 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) && 1016 bio_rw(bio) == READA) { 1017 bio_io_error(bio); 1018 return 0; 1019 } 1020 1021 queue_io(md, bio); 1022 1023 return 0; 1024 } 1025 1026 __split_and_process_bio(md, bio); 1027 up_read(&md->io_lock); 1028 return 0; 1029 } 1030 1031 static void dm_unplug_all(struct request_queue *q) 1032 { 1033 struct mapped_device *md = q->queuedata; 1034 struct dm_table *map = dm_get_table(md); 1035 1036 if (map) { 1037 dm_table_unplug_all(map); 1038 dm_table_put(map); 1039 } 1040 } 1041 1042 static int dm_any_congested(void *congested_data, int bdi_bits) 1043 { 1044 int r = bdi_bits; 1045 struct mapped_device *md = congested_data; 1046 struct dm_table *map; 1047 1048 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 1049 map = dm_get_table(md); 1050 if (map) { 1051 r = dm_table_any_congested(map, bdi_bits); 1052 dm_table_put(map); 1053 } 1054 } 1055 1056 return r; 1057 } 1058 1059 /*----------------------------------------------------------------- 1060 * An IDR is used to keep track of allocated minor numbers. 1061 *---------------------------------------------------------------*/ 1062 static DEFINE_IDR(_minor_idr); 1063 1064 static void free_minor(int minor) 1065 { 1066 spin_lock(&_minor_lock); 1067 idr_remove(&_minor_idr, minor); 1068 spin_unlock(&_minor_lock); 1069 } 1070 1071 /* 1072 * See if the device with a specific minor # is free. 1073 */ 1074 static int specific_minor(int minor) 1075 { 1076 int r, m; 1077 1078 if (minor >= (1 << MINORBITS)) 1079 return -EINVAL; 1080 1081 r = idr_pre_get(&_minor_idr, GFP_KERNEL); 1082 if (!r) 1083 return -ENOMEM; 1084 1085 spin_lock(&_minor_lock); 1086 1087 if (idr_find(&_minor_idr, minor)) { 1088 r = -EBUSY; 1089 goto out; 1090 } 1091 1092 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); 1093 if (r) 1094 goto out; 1095 1096 if (m != minor) { 1097 idr_remove(&_minor_idr, m); 1098 r = -EBUSY; 1099 goto out; 1100 } 1101 1102 out: 1103 spin_unlock(&_minor_lock); 1104 return r; 1105 } 1106 1107 static int next_free_minor(int *minor) 1108 { 1109 int r, m; 1110 1111 r = idr_pre_get(&_minor_idr, GFP_KERNEL); 1112 if (!r) 1113 return -ENOMEM; 1114 1115 spin_lock(&_minor_lock); 1116 1117 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); 1118 if (r) 1119 goto out; 1120 1121 if (m >= (1 << MINORBITS)) { 1122 idr_remove(&_minor_idr, m); 1123 r = -ENOSPC; 1124 goto out; 1125 } 1126 1127 *minor = m; 1128 1129 out: 1130 spin_unlock(&_minor_lock); 1131 return r; 1132 } 1133 1134 static struct block_device_operations dm_blk_dops; 1135 1136 static void dm_wq_work(struct work_struct *work); 1137 1138 /* 1139 * Allocate and initialise a blank device with a given minor. 1140 */ 1141 static struct mapped_device *alloc_dev(int minor) 1142 { 1143 int r; 1144 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); 1145 void *old_md; 1146 1147 if (!md) { 1148 DMWARN("unable to allocate device, out of memory."); 1149 return NULL; 1150 } 1151 1152 if (!try_module_get(THIS_MODULE)) 1153 goto bad_module_get; 1154 1155 /* get a minor number for the dev */ 1156 if (minor == DM_ANY_MINOR) 1157 r = next_free_minor(&minor); 1158 else 1159 r = specific_minor(minor); 1160 if (r < 0) 1161 goto bad_minor; 1162 1163 init_rwsem(&md->io_lock); 1164 mutex_init(&md->suspend_lock); 1165 spin_lock_init(&md->deferred_lock); 1166 rwlock_init(&md->map_lock); 1167 atomic_set(&md->holders, 1); 1168 atomic_set(&md->open_count, 0); 1169 atomic_set(&md->event_nr, 0); 1170 atomic_set(&md->uevent_seq, 0); 1171 INIT_LIST_HEAD(&md->uevent_list); 1172 spin_lock_init(&md->uevent_lock); 1173 1174 md->queue = blk_alloc_queue(GFP_KERNEL); 1175 if (!md->queue) 1176 goto bad_queue; 1177 1178 md->queue->queuedata = md; 1179 md->queue->backing_dev_info.congested_fn = dm_any_congested; 1180 md->queue->backing_dev_info.congested_data = md; 1181 blk_queue_make_request(md->queue, dm_request); 1182 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL); 1183 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); 1184 md->queue->unplug_fn = dm_unplug_all; 1185 blk_queue_merge_bvec(md->queue, dm_merge_bvec); 1186 1187 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); 1188 if (!md->io_pool) 1189 goto bad_io_pool; 1190 1191 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); 1192 if (!md->tio_pool) 1193 goto bad_tio_pool; 1194 1195 md->bs = bioset_create(16, 0); 1196 if (!md->bs) 1197 goto bad_no_bioset; 1198 1199 md->disk = alloc_disk(1); 1200 if (!md->disk) 1201 goto bad_disk; 1202 1203 atomic_set(&md->pending, 0); 1204 init_waitqueue_head(&md->wait); 1205 INIT_WORK(&md->work, dm_wq_work); 1206 init_waitqueue_head(&md->eventq); 1207 1208 md->disk->major = _major; 1209 md->disk->first_minor = minor; 1210 md->disk->fops = &dm_blk_dops; 1211 md->disk->queue = md->queue; 1212 md->disk->private_data = md; 1213 sprintf(md->disk->disk_name, "dm-%d", minor); 1214 add_disk(md->disk); 1215 format_dev_t(md->name, MKDEV(_major, minor)); 1216 1217 md->wq = create_singlethread_workqueue("kdmflush"); 1218 if (!md->wq) 1219 goto bad_thread; 1220 1221 md->bdev = bdget_disk(md->disk, 0); 1222 if (!md->bdev) 1223 goto bad_bdev; 1224 1225 /* Populate the mapping, nobody knows we exist yet */ 1226 spin_lock(&_minor_lock); 1227 old_md = idr_replace(&_minor_idr, md, minor); 1228 spin_unlock(&_minor_lock); 1229 1230 BUG_ON(old_md != MINOR_ALLOCED); 1231 1232 return md; 1233 1234 bad_bdev: 1235 destroy_workqueue(md->wq); 1236 bad_thread: 1237 put_disk(md->disk); 1238 bad_disk: 1239 bioset_free(md->bs); 1240 bad_no_bioset: 1241 mempool_destroy(md->tio_pool); 1242 bad_tio_pool: 1243 mempool_destroy(md->io_pool); 1244 bad_io_pool: 1245 blk_cleanup_queue(md->queue); 1246 bad_queue: 1247 free_minor(minor); 1248 bad_minor: 1249 module_put(THIS_MODULE); 1250 bad_module_get: 1251 kfree(md); 1252 return NULL; 1253 } 1254 1255 static void unlock_fs(struct mapped_device *md); 1256 1257 static void free_dev(struct mapped_device *md) 1258 { 1259 int minor = MINOR(disk_devt(md->disk)); 1260 1261 unlock_fs(md); 1262 bdput(md->bdev); 1263 destroy_workqueue(md->wq); 1264 mempool_destroy(md->tio_pool); 1265 mempool_destroy(md->io_pool); 1266 bioset_free(md->bs); 1267 blk_integrity_unregister(md->disk); 1268 del_gendisk(md->disk); 1269 free_minor(minor); 1270 1271 spin_lock(&_minor_lock); 1272 md->disk->private_data = NULL; 1273 spin_unlock(&_minor_lock); 1274 1275 put_disk(md->disk); 1276 blk_cleanup_queue(md->queue); 1277 module_put(THIS_MODULE); 1278 kfree(md); 1279 } 1280 1281 /* 1282 * Bind a table to the device. 1283 */ 1284 static void event_callback(void *context) 1285 { 1286 unsigned long flags; 1287 LIST_HEAD(uevents); 1288 struct mapped_device *md = (struct mapped_device *) context; 1289 1290 spin_lock_irqsave(&md->uevent_lock, flags); 1291 list_splice_init(&md->uevent_list, &uevents); 1292 spin_unlock_irqrestore(&md->uevent_lock, flags); 1293 1294 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); 1295 1296 atomic_inc(&md->event_nr); 1297 wake_up(&md->eventq); 1298 } 1299 1300 static void __set_size(struct mapped_device *md, sector_t size) 1301 { 1302 set_capacity(md->disk, size); 1303 1304 mutex_lock(&md->bdev->bd_inode->i_mutex); 1305 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); 1306 mutex_unlock(&md->bdev->bd_inode->i_mutex); 1307 } 1308 1309 static int __bind(struct mapped_device *md, struct dm_table *t) 1310 { 1311 struct request_queue *q = md->queue; 1312 sector_t size; 1313 1314 size = dm_table_get_size(t); 1315 1316 /* 1317 * Wipe any geometry if the size of the table changed. 1318 */ 1319 if (size != get_capacity(md->disk)) 1320 memset(&md->geometry, 0, sizeof(md->geometry)); 1321 1322 __set_size(md, size); 1323 1324 if (!size) { 1325 dm_table_destroy(t); 1326 return 0; 1327 } 1328 1329 dm_table_event_callback(t, event_callback, md); 1330 1331 write_lock(&md->map_lock); 1332 md->map = t; 1333 dm_table_set_restrictions(t, q); 1334 write_unlock(&md->map_lock); 1335 1336 return 0; 1337 } 1338 1339 static void __unbind(struct mapped_device *md) 1340 { 1341 struct dm_table *map = md->map; 1342 1343 if (!map) 1344 return; 1345 1346 dm_table_event_callback(map, NULL, NULL); 1347 write_lock(&md->map_lock); 1348 md->map = NULL; 1349 write_unlock(&md->map_lock); 1350 dm_table_destroy(map); 1351 } 1352 1353 /* 1354 * Constructor for a new device. 1355 */ 1356 int dm_create(int minor, struct mapped_device **result) 1357 { 1358 struct mapped_device *md; 1359 1360 md = alloc_dev(minor); 1361 if (!md) 1362 return -ENXIO; 1363 1364 dm_sysfs_init(md); 1365 1366 *result = md; 1367 return 0; 1368 } 1369 1370 static struct mapped_device *dm_find_md(dev_t dev) 1371 { 1372 struct mapped_device *md; 1373 unsigned minor = MINOR(dev); 1374 1375 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) 1376 return NULL; 1377 1378 spin_lock(&_minor_lock); 1379 1380 md = idr_find(&_minor_idr, minor); 1381 if (md && (md == MINOR_ALLOCED || 1382 (MINOR(disk_devt(dm_disk(md))) != minor) || 1383 test_bit(DMF_FREEING, &md->flags))) { 1384 md = NULL; 1385 goto out; 1386 } 1387 1388 out: 1389 spin_unlock(&_minor_lock); 1390 1391 return md; 1392 } 1393 1394 struct mapped_device *dm_get_md(dev_t dev) 1395 { 1396 struct mapped_device *md = dm_find_md(dev); 1397 1398 if (md) 1399 dm_get(md); 1400 1401 return md; 1402 } 1403 1404 void *dm_get_mdptr(struct mapped_device *md) 1405 { 1406 return md->interface_ptr; 1407 } 1408 1409 void dm_set_mdptr(struct mapped_device *md, void *ptr) 1410 { 1411 md->interface_ptr = ptr; 1412 } 1413 1414 void dm_get(struct mapped_device *md) 1415 { 1416 atomic_inc(&md->holders); 1417 } 1418 1419 const char *dm_device_name(struct mapped_device *md) 1420 { 1421 return md->name; 1422 } 1423 EXPORT_SYMBOL_GPL(dm_device_name); 1424 1425 void dm_put(struct mapped_device *md) 1426 { 1427 struct dm_table *map; 1428 1429 BUG_ON(test_bit(DMF_FREEING, &md->flags)); 1430 1431 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { 1432 map = dm_get_table(md); 1433 idr_replace(&_minor_idr, MINOR_ALLOCED, 1434 MINOR(disk_devt(dm_disk(md)))); 1435 set_bit(DMF_FREEING, &md->flags); 1436 spin_unlock(&_minor_lock); 1437 if (!dm_suspended(md)) { 1438 dm_table_presuspend_targets(map); 1439 dm_table_postsuspend_targets(map); 1440 } 1441 dm_sysfs_exit(md); 1442 dm_table_put(map); 1443 __unbind(md); 1444 free_dev(md); 1445 } 1446 } 1447 EXPORT_SYMBOL_GPL(dm_put); 1448 1449 static int dm_wait_for_completion(struct mapped_device *md, int interruptible) 1450 { 1451 int r = 0; 1452 DECLARE_WAITQUEUE(wait, current); 1453 1454 dm_unplug_all(md->queue); 1455 1456 add_wait_queue(&md->wait, &wait); 1457 1458 while (1) { 1459 set_current_state(interruptible); 1460 1461 smp_mb(); 1462 if (!atomic_read(&md->pending)) 1463 break; 1464 1465 if (interruptible == TASK_INTERRUPTIBLE && 1466 signal_pending(current)) { 1467 r = -EINTR; 1468 break; 1469 } 1470 1471 io_schedule(); 1472 } 1473 set_current_state(TASK_RUNNING); 1474 1475 remove_wait_queue(&md->wait, &wait); 1476 1477 return r; 1478 } 1479 1480 static void dm_flush(struct mapped_device *md) 1481 { 1482 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); 1483 } 1484 1485 static void process_barrier(struct mapped_device *md, struct bio *bio) 1486 { 1487 md->barrier_error = 0; 1488 1489 dm_flush(md); 1490 1491 if (!bio_empty_barrier(bio)) { 1492 __split_and_process_bio(md, bio); 1493 dm_flush(md); 1494 } 1495 1496 if (md->barrier_error != DM_ENDIO_REQUEUE) 1497 bio_endio(bio, md->barrier_error); 1498 else { 1499 spin_lock_irq(&md->deferred_lock); 1500 bio_list_add_head(&md->deferred, bio); 1501 spin_unlock_irq(&md->deferred_lock); 1502 } 1503 } 1504 1505 /* 1506 * Process the deferred bios 1507 */ 1508 static void dm_wq_work(struct work_struct *work) 1509 { 1510 struct mapped_device *md = container_of(work, struct mapped_device, 1511 work); 1512 struct bio *c; 1513 1514 down_write(&md->io_lock); 1515 1516 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 1517 spin_lock_irq(&md->deferred_lock); 1518 c = bio_list_pop(&md->deferred); 1519 spin_unlock_irq(&md->deferred_lock); 1520 1521 if (!c) { 1522 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); 1523 break; 1524 } 1525 1526 up_write(&md->io_lock); 1527 1528 if (bio_barrier(c)) 1529 process_barrier(md, c); 1530 else 1531 __split_and_process_bio(md, c); 1532 1533 down_write(&md->io_lock); 1534 } 1535 1536 up_write(&md->io_lock); 1537 } 1538 1539 static void dm_queue_flush(struct mapped_device *md) 1540 { 1541 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 1542 smp_mb__after_clear_bit(); 1543 queue_work(md->wq, &md->work); 1544 } 1545 1546 /* 1547 * Swap in a new table (destroying old one). 1548 */ 1549 int dm_swap_table(struct mapped_device *md, struct dm_table *table) 1550 { 1551 int r = -EINVAL; 1552 1553 mutex_lock(&md->suspend_lock); 1554 1555 /* device must be suspended */ 1556 if (!dm_suspended(md)) 1557 goto out; 1558 1559 __unbind(md); 1560 r = __bind(md, table); 1561 1562 out: 1563 mutex_unlock(&md->suspend_lock); 1564 return r; 1565 } 1566 1567 /* 1568 * Functions to lock and unlock any filesystem running on the 1569 * device. 1570 */ 1571 static int lock_fs(struct mapped_device *md) 1572 { 1573 int r; 1574 1575 WARN_ON(md->frozen_sb); 1576 1577 md->frozen_sb = freeze_bdev(md->bdev); 1578 if (IS_ERR(md->frozen_sb)) { 1579 r = PTR_ERR(md->frozen_sb); 1580 md->frozen_sb = NULL; 1581 return r; 1582 } 1583 1584 set_bit(DMF_FROZEN, &md->flags); 1585 1586 return 0; 1587 } 1588 1589 static void unlock_fs(struct mapped_device *md) 1590 { 1591 if (!test_bit(DMF_FROZEN, &md->flags)) 1592 return; 1593 1594 thaw_bdev(md->bdev, md->frozen_sb); 1595 md->frozen_sb = NULL; 1596 clear_bit(DMF_FROZEN, &md->flags); 1597 } 1598 1599 /* 1600 * We need to be able to change a mapping table under a mounted 1601 * filesystem. For example we might want to move some data in 1602 * the background. Before the table can be swapped with 1603 * dm_bind_table, dm_suspend must be called to flush any in 1604 * flight bios and ensure that any further io gets deferred. 1605 */ 1606 int dm_suspend(struct mapped_device *md, unsigned suspend_flags) 1607 { 1608 struct dm_table *map = NULL; 1609 int r = 0; 1610 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; 1611 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; 1612 1613 mutex_lock(&md->suspend_lock); 1614 1615 if (dm_suspended(md)) { 1616 r = -EINVAL; 1617 goto out_unlock; 1618 } 1619 1620 map = dm_get_table(md); 1621 1622 /* 1623 * DMF_NOFLUSH_SUSPENDING must be set before presuspend. 1624 * This flag is cleared before dm_suspend returns. 1625 */ 1626 if (noflush) 1627 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 1628 1629 /* This does not get reverted if there's an error later. */ 1630 dm_table_presuspend_targets(map); 1631 1632 /* 1633 * Flush I/O to the device. noflush supersedes do_lockfs, 1634 * because lock_fs() needs to flush I/Os. 1635 */ 1636 if (!noflush && do_lockfs) { 1637 r = lock_fs(md); 1638 if (r) 1639 goto out; 1640 } 1641 1642 /* 1643 * Here we must make sure that no processes are submitting requests 1644 * to target drivers i.e. no one may be executing 1645 * __split_and_process_bio. This is called from dm_request and 1646 * dm_wq_work. 1647 * 1648 * To get all processes out of __split_and_process_bio in dm_request, 1649 * we take the write lock. To prevent any process from reentering 1650 * __split_and_process_bio from dm_request, we set 1651 * DMF_QUEUE_IO_TO_THREAD. 1652 * 1653 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND 1654 * and call flush_workqueue(md->wq). flush_workqueue will wait until 1655 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any 1656 * further calls to __split_and_process_bio from dm_wq_work. 1657 */ 1658 down_write(&md->io_lock); 1659 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 1660 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); 1661 up_write(&md->io_lock); 1662 1663 flush_workqueue(md->wq); 1664 1665 /* 1666 * At this point no more requests are entering target request routines. 1667 * We call dm_wait_for_completion to wait for all existing requests 1668 * to finish. 1669 */ 1670 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); 1671 1672 down_write(&md->io_lock); 1673 if (noflush) 1674 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 1675 up_write(&md->io_lock); 1676 1677 /* were we interrupted ? */ 1678 if (r < 0) { 1679 dm_queue_flush(md); 1680 1681 unlock_fs(md); 1682 goto out; /* pushback list is already flushed, so skip flush */ 1683 } 1684 1685 /* 1686 * If dm_wait_for_completion returned 0, the device is completely 1687 * quiescent now. There is no request-processing activity. All new 1688 * requests are being added to md->deferred list. 1689 */ 1690 1691 dm_table_postsuspend_targets(map); 1692 1693 set_bit(DMF_SUSPENDED, &md->flags); 1694 1695 out: 1696 dm_table_put(map); 1697 1698 out_unlock: 1699 mutex_unlock(&md->suspend_lock); 1700 return r; 1701 } 1702 1703 int dm_resume(struct mapped_device *md) 1704 { 1705 int r = -EINVAL; 1706 struct dm_table *map = NULL; 1707 1708 mutex_lock(&md->suspend_lock); 1709 if (!dm_suspended(md)) 1710 goto out; 1711 1712 map = dm_get_table(md); 1713 if (!map || !dm_table_get_size(map)) 1714 goto out; 1715 1716 r = dm_table_resume_targets(map); 1717 if (r) 1718 goto out; 1719 1720 dm_queue_flush(md); 1721 1722 unlock_fs(md); 1723 1724 clear_bit(DMF_SUSPENDED, &md->flags); 1725 1726 dm_table_unplug_all(map); 1727 1728 dm_kobject_uevent(md); 1729 1730 r = 0; 1731 1732 out: 1733 dm_table_put(map); 1734 mutex_unlock(&md->suspend_lock); 1735 1736 return r; 1737 } 1738 1739 /*----------------------------------------------------------------- 1740 * Event notification. 1741 *---------------------------------------------------------------*/ 1742 void dm_kobject_uevent(struct mapped_device *md) 1743 { 1744 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE); 1745 } 1746 1747 uint32_t dm_next_uevent_seq(struct mapped_device *md) 1748 { 1749 return atomic_add_return(1, &md->uevent_seq); 1750 } 1751 1752 uint32_t dm_get_event_nr(struct mapped_device *md) 1753 { 1754 return atomic_read(&md->event_nr); 1755 } 1756 1757 int dm_wait_event(struct mapped_device *md, int event_nr) 1758 { 1759 return wait_event_interruptible(md->eventq, 1760 (event_nr != atomic_read(&md->event_nr))); 1761 } 1762 1763 void dm_uevent_add(struct mapped_device *md, struct list_head *elist) 1764 { 1765 unsigned long flags; 1766 1767 spin_lock_irqsave(&md->uevent_lock, flags); 1768 list_add(elist, &md->uevent_list); 1769 spin_unlock_irqrestore(&md->uevent_lock, flags); 1770 } 1771 1772 /* 1773 * The gendisk is only valid as long as you have a reference 1774 * count on 'md'. 1775 */ 1776 struct gendisk *dm_disk(struct mapped_device *md) 1777 { 1778 return md->disk; 1779 } 1780 1781 struct kobject *dm_kobject(struct mapped_device *md) 1782 { 1783 return &md->kobj; 1784 } 1785 1786 /* 1787 * struct mapped_device should not be exported outside of dm.c 1788 * so use this check to verify that kobj is part of md structure 1789 */ 1790 struct mapped_device *dm_get_from_kobject(struct kobject *kobj) 1791 { 1792 struct mapped_device *md; 1793 1794 md = container_of(kobj, struct mapped_device, kobj); 1795 if (&md->kobj != kobj) 1796 return NULL; 1797 1798 if (test_bit(DMF_FREEING, &md->flags) || 1799 test_bit(DMF_DELETING, &md->flags)) 1800 return NULL; 1801 1802 dm_get(md); 1803 return md; 1804 } 1805 1806 int dm_suspended(struct mapped_device *md) 1807 { 1808 return test_bit(DMF_SUSPENDED, &md->flags); 1809 } 1810 1811 int dm_noflush_suspending(struct dm_target *ti) 1812 { 1813 struct mapped_device *md = dm_table_get_md(ti->table); 1814 int r = __noflush_suspending(md); 1815 1816 dm_put(md); 1817 1818 return r; 1819 } 1820 EXPORT_SYMBOL_GPL(dm_noflush_suspending); 1821 1822 static struct block_device_operations dm_blk_dops = { 1823 .open = dm_blk_open, 1824 .release = dm_blk_close, 1825 .ioctl = dm_blk_ioctl, 1826 .getgeo = dm_blk_getgeo, 1827 .owner = THIS_MODULE 1828 }; 1829 1830 EXPORT_SYMBOL(dm_get_mapinfo); 1831 1832 /* 1833 * module hooks 1834 */ 1835 module_init(dm_init); 1836 module_exit(dm_exit); 1837 1838 module_param(major, uint, 0); 1839 MODULE_PARM_DESC(major, "The major number of the device mapper"); 1840 MODULE_DESCRIPTION(DM_NAME " driver"); 1841 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); 1842 MODULE_LICENSE("GPL"); 1843