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 int __clone_and_map(struct clone_info *ci) 754 { 755 struct bio *clone, *bio = ci->bio; 756 struct dm_target *ti; 757 sector_t len = 0, max; 758 struct dm_target_io *tio; 759 760 ti = dm_table_find_target(ci->map, ci->sector); 761 if (!dm_target_is_valid(ti)) 762 return -EIO; 763 764 max = max_io_len(ci->md, ci->sector, ti); 765 766 /* 767 * Allocate a target io object. 768 */ 769 tio = alloc_tio(ci->md); 770 tio->io = ci->io; 771 tio->ti = ti; 772 memset(&tio->info, 0, sizeof(tio->info)); 773 774 if (ci->sector_count <= max) { 775 /* 776 * Optimise for the simple case where we can do all of 777 * the remaining io with a single clone. 778 */ 779 clone = clone_bio(bio, ci->sector, ci->idx, 780 bio->bi_vcnt - ci->idx, ci->sector_count, 781 ci->md->bs); 782 __map_bio(ti, clone, tio); 783 ci->sector_count = 0; 784 785 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { 786 /* 787 * There are some bvecs that don't span targets. 788 * Do as many of these as possible. 789 */ 790 int i; 791 sector_t remaining = max; 792 sector_t bv_len; 793 794 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { 795 bv_len = to_sector(bio->bi_io_vec[i].bv_len); 796 797 if (bv_len > remaining) 798 break; 799 800 remaining -= bv_len; 801 len += bv_len; 802 } 803 804 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, 805 ci->md->bs); 806 __map_bio(ti, clone, tio); 807 808 ci->sector += len; 809 ci->sector_count -= len; 810 ci->idx = i; 811 812 } else { 813 /* 814 * Handle a bvec that must be split between two or more targets. 815 */ 816 struct bio_vec *bv = bio->bi_io_vec + ci->idx; 817 sector_t remaining = to_sector(bv->bv_len); 818 unsigned int offset = 0; 819 820 do { 821 if (offset) { 822 ti = dm_table_find_target(ci->map, ci->sector); 823 if (!dm_target_is_valid(ti)) 824 return -EIO; 825 826 max = max_io_len(ci->md, ci->sector, ti); 827 828 tio = alloc_tio(ci->md); 829 tio->io = ci->io; 830 tio->ti = ti; 831 memset(&tio->info, 0, sizeof(tio->info)); 832 } 833 834 len = min(remaining, max); 835 836 clone = split_bvec(bio, ci->sector, ci->idx, 837 bv->bv_offset + offset, len, 838 ci->md->bs); 839 840 __map_bio(ti, clone, tio); 841 842 ci->sector += len; 843 ci->sector_count -= len; 844 offset += to_bytes(len); 845 } while (remaining -= len); 846 847 ci->idx++; 848 } 849 850 return 0; 851 } 852 853 /* 854 * Split the bio into several clones and submit it to targets. 855 */ 856 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) 857 { 858 struct clone_info ci; 859 int error = 0; 860 861 ci.map = dm_get_table(md); 862 if (unlikely(!ci.map)) { 863 if (!bio_barrier(bio)) 864 bio_io_error(bio); 865 else 866 if (!md->barrier_error) 867 md->barrier_error = -EIO; 868 return; 869 } 870 871 ci.md = md; 872 ci.bio = bio; 873 ci.io = alloc_io(md); 874 ci.io->error = 0; 875 atomic_set(&ci.io->io_count, 1); 876 ci.io->bio = bio; 877 ci.io->md = md; 878 ci.sector = bio->bi_sector; 879 ci.sector_count = bio_sectors(bio); 880 ci.idx = bio->bi_idx; 881 882 start_io_acct(ci.io); 883 while (ci.sector_count && !error) 884 error = __clone_and_map(&ci); 885 886 /* drop the extra reference count */ 887 dec_pending(ci.io, error); 888 dm_table_put(ci.map); 889 } 890 /*----------------------------------------------------------------- 891 * CRUD END 892 *---------------------------------------------------------------*/ 893 894 static int dm_merge_bvec(struct request_queue *q, 895 struct bvec_merge_data *bvm, 896 struct bio_vec *biovec) 897 { 898 struct mapped_device *md = q->queuedata; 899 struct dm_table *map = dm_get_table(md); 900 struct dm_target *ti; 901 sector_t max_sectors; 902 int max_size = 0; 903 904 if (unlikely(!map)) 905 goto out; 906 907 ti = dm_table_find_target(map, bvm->bi_sector); 908 if (!dm_target_is_valid(ti)) 909 goto out_table; 910 911 /* 912 * Find maximum amount of I/O that won't need splitting 913 */ 914 max_sectors = min(max_io_len(md, bvm->bi_sector, ti), 915 (sector_t) BIO_MAX_SECTORS); 916 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; 917 if (max_size < 0) 918 max_size = 0; 919 920 /* 921 * merge_bvec_fn() returns number of bytes 922 * it can accept at this offset 923 * max is precomputed maximal io size 924 */ 925 if (max_size && ti->type->merge) 926 max_size = ti->type->merge(ti, bvm, biovec, max_size); 927 /* 928 * If the target doesn't support merge method and some of the devices 929 * provided their merge_bvec method (we know this by looking at 930 * queue_max_hw_sectors), then we can't allow bios with multiple vector 931 * entries. So always set max_size to 0, and the code below allows 932 * just one page. 933 */ 934 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) 935 936 max_size = 0; 937 938 out_table: 939 dm_table_put(map); 940 941 out: 942 /* 943 * Always allow an entire first page 944 */ 945 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) 946 max_size = biovec->bv_len; 947 948 return max_size; 949 } 950 951 /* 952 * The request function that just remaps the bio built up by 953 * dm_merge_bvec. 954 */ 955 static int dm_request(struct request_queue *q, struct bio *bio) 956 { 957 int rw = bio_data_dir(bio); 958 struct mapped_device *md = q->queuedata; 959 int cpu; 960 961 down_read(&md->io_lock); 962 963 cpu = part_stat_lock(); 964 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); 965 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); 966 part_stat_unlock(); 967 968 /* 969 * If we're suspended or the thread is processing barriers 970 * we have to queue this io for later. 971 */ 972 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) || 973 unlikely(bio_barrier(bio))) { 974 up_read(&md->io_lock); 975 976 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) && 977 bio_rw(bio) == READA) { 978 bio_io_error(bio); 979 return 0; 980 } 981 982 queue_io(md, bio); 983 984 return 0; 985 } 986 987 __split_and_process_bio(md, bio); 988 up_read(&md->io_lock); 989 return 0; 990 } 991 992 static void dm_unplug_all(struct request_queue *q) 993 { 994 struct mapped_device *md = q->queuedata; 995 struct dm_table *map = dm_get_table(md); 996 997 if (map) { 998 dm_table_unplug_all(map); 999 dm_table_put(map); 1000 } 1001 } 1002 1003 static int dm_any_congested(void *congested_data, int bdi_bits) 1004 { 1005 int r = bdi_bits; 1006 struct mapped_device *md = congested_data; 1007 struct dm_table *map; 1008 1009 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 1010 map = dm_get_table(md); 1011 if (map) { 1012 r = dm_table_any_congested(map, bdi_bits); 1013 dm_table_put(map); 1014 } 1015 } 1016 1017 return r; 1018 } 1019 1020 /*----------------------------------------------------------------- 1021 * An IDR is used to keep track of allocated minor numbers. 1022 *---------------------------------------------------------------*/ 1023 static DEFINE_IDR(_minor_idr); 1024 1025 static void free_minor(int minor) 1026 { 1027 spin_lock(&_minor_lock); 1028 idr_remove(&_minor_idr, minor); 1029 spin_unlock(&_minor_lock); 1030 } 1031 1032 /* 1033 * See if the device with a specific minor # is free. 1034 */ 1035 static int specific_minor(int minor) 1036 { 1037 int r, m; 1038 1039 if (minor >= (1 << MINORBITS)) 1040 return -EINVAL; 1041 1042 r = idr_pre_get(&_minor_idr, GFP_KERNEL); 1043 if (!r) 1044 return -ENOMEM; 1045 1046 spin_lock(&_minor_lock); 1047 1048 if (idr_find(&_minor_idr, minor)) { 1049 r = -EBUSY; 1050 goto out; 1051 } 1052 1053 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); 1054 if (r) 1055 goto out; 1056 1057 if (m != minor) { 1058 idr_remove(&_minor_idr, m); 1059 r = -EBUSY; 1060 goto out; 1061 } 1062 1063 out: 1064 spin_unlock(&_minor_lock); 1065 return r; 1066 } 1067 1068 static int next_free_minor(int *minor) 1069 { 1070 int r, m; 1071 1072 r = idr_pre_get(&_minor_idr, GFP_KERNEL); 1073 if (!r) 1074 return -ENOMEM; 1075 1076 spin_lock(&_minor_lock); 1077 1078 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); 1079 if (r) 1080 goto out; 1081 1082 if (m >= (1 << MINORBITS)) { 1083 idr_remove(&_minor_idr, m); 1084 r = -ENOSPC; 1085 goto out; 1086 } 1087 1088 *minor = m; 1089 1090 out: 1091 spin_unlock(&_minor_lock); 1092 return r; 1093 } 1094 1095 static struct block_device_operations dm_blk_dops; 1096 1097 static void dm_wq_work(struct work_struct *work); 1098 1099 /* 1100 * Allocate and initialise a blank device with a given minor. 1101 */ 1102 static struct mapped_device *alloc_dev(int minor) 1103 { 1104 int r; 1105 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); 1106 void *old_md; 1107 1108 if (!md) { 1109 DMWARN("unable to allocate device, out of memory."); 1110 return NULL; 1111 } 1112 1113 if (!try_module_get(THIS_MODULE)) 1114 goto bad_module_get; 1115 1116 /* get a minor number for the dev */ 1117 if (minor == DM_ANY_MINOR) 1118 r = next_free_minor(&minor); 1119 else 1120 r = specific_minor(minor); 1121 if (r < 0) 1122 goto bad_minor; 1123 1124 init_rwsem(&md->io_lock); 1125 mutex_init(&md->suspend_lock); 1126 spin_lock_init(&md->deferred_lock); 1127 rwlock_init(&md->map_lock); 1128 atomic_set(&md->holders, 1); 1129 atomic_set(&md->open_count, 0); 1130 atomic_set(&md->event_nr, 0); 1131 atomic_set(&md->uevent_seq, 0); 1132 INIT_LIST_HEAD(&md->uevent_list); 1133 spin_lock_init(&md->uevent_lock); 1134 1135 md->queue = blk_alloc_queue(GFP_KERNEL); 1136 if (!md->queue) 1137 goto bad_queue; 1138 1139 md->queue->queuedata = md; 1140 md->queue->backing_dev_info.congested_fn = dm_any_congested; 1141 md->queue->backing_dev_info.congested_data = md; 1142 blk_queue_make_request(md->queue, dm_request); 1143 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL); 1144 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); 1145 md->queue->unplug_fn = dm_unplug_all; 1146 blk_queue_merge_bvec(md->queue, dm_merge_bvec); 1147 1148 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); 1149 if (!md->io_pool) 1150 goto bad_io_pool; 1151 1152 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); 1153 if (!md->tio_pool) 1154 goto bad_tio_pool; 1155 1156 md->bs = bioset_create(16, 0); 1157 if (!md->bs) 1158 goto bad_no_bioset; 1159 1160 md->disk = alloc_disk(1); 1161 if (!md->disk) 1162 goto bad_disk; 1163 1164 atomic_set(&md->pending, 0); 1165 init_waitqueue_head(&md->wait); 1166 INIT_WORK(&md->work, dm_wq_work); 1167 init_waitqueue_head(&md->eventq); 1168 1169 md->disk->major = _major; 1170 md->disk->first_minor = minor; 1171 md->disk->fops = &dm_blk_dops; 1172 md->disk->queue = md->queue; 1173 md->disk->private_data = md; 1174 sprintf(md->disk->disk_name, "dm-%d", minor); 1175 add_disk(md->disk); 1176 format_dev_t(md->name, MKDEV(_major, minor)); 1177 1178 md->wq = create_singlethread_workqueue("kdmflush"); 1179 if (!md->wq) 1180 goto bad_thread; 1181 1182 md->bdev = bdget_disk(md->disk, 0); 1183 if (!md->bdev) 1184 goto bad_bdev; 1185 1186 /* Populate the mapping, nobody knows we exist yet */ 1187 spin_lock(&_minor_lock); 1188 old_md = idr_replace(&_minor_idr, md, minor); 1189 spin_unlock(&_minor_lock); 1190 1191 BUG_ON(old_md != MINOR_ALLOCED); 1192 1193 return md; 1194 1195 bad_bdev: 1196 destroy_workqueue(md->wq); 1197 bad_thread: 1198 put_disk(md->disk); 1199 bad_disk: 1200 bioset_free(md->bs); 1201 bad_no_bioset: 1202 mempool_destroy(md->tio_pool); 1203 bad_tio_pool: 1204 mempool_destroy(md->io_pool); 1205 bad_io_pool: 1206 blk_cleanup_queue(md->queue); 1207 bad_queue: 1208 free_minor(minor); 1209 bad_minor: 1210 module_put(THIS_MODULE); 1211 bad_module_get: 1212 kfree(md); 1213 return NULL; 1214 } 1215 1216 static void unlock_fs(struct mapped_device *md); 1217 1218 static void free_dev(struct mapped_device *md) 1219 { 1220 int minor = MINOR(disk_devt(md->disk)); 1221 1222 unlock_fs(md); 1223 bdput(md->bdev); 1224 destroy_workqueue(md->wq); 1225 mempool_destroy(md->tio_pool); 1226 mempool_destroy(md->io_pool); 1227 bioset_free(md->bs); 1228 blk_integrity_unregister(md->disk); 1229 del_gendisk(md->disk); 1230 free_minor(minor); 1231 1232 spin_lock(&_minor_lock); 1233 md->disk->private_data = NULL; 1234 spin_unlock(&_minor_lock); 1235 1236 put_disk(md->disk); 1237 blk_cleanup_queue(md->queue); 1238 module_put(THIS_MODULE); 1239 kfree(md); 1240 } 1241 1242 /* 1243 * Bind a table to the device. 1244 */ 1245 static void event_callback(void *context) 1246 { 1247 unsigned long flags; 1248 LIST_HEAD(uevents); 1249 struct mapped_device *md = (struct mapped_device *) context; 1250 1251 spin_lock_irqsave(&md->uevent_lock, flags); 1252 list_splice_init(&md->uevent_list, &uevents); 1253 spin_unlock_irqrestore(&md->uevent_lock, flags); 1254 1255 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); 1256 1257 atomic_inc(&md->event_nr); 1258 wake_up(&md->eventq); 1259 } 1260 1261 static void __set_size(struct mapped_device *md, sector_t size) 1262 { 1263 set_capacity(md->disk, size); 1264 1265 mutex_lock(&md->bdev->bd_inode->i_mutex); 1266 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); 1267 mutex_unlock(&md->bdev->bd_inode->i_mutex); 1268 } 1269 1270 static int __bind(struct mapped_device *md, struct dm_table *t) 1271 { 1272 struct request_queue *q = md->queue; 1273 sector_t size; 1274 1275 size = dm_table_get_size(t); 1276 1277 /* 1278 * Wipe any geometry if the size of the table changed. 1279 */ 1280 if (size != get_capacity(md->disk)) 1281 memset(&md->geometry, 0, sizeof(md->geometry)); 1282 1283 __set_size(md, size); 1284 1285 if (!size) { 1286 dm_table_destroy(t); 1287 return 0; 1288 } 1289 1290 dm_table_event_callback(t, event_callback, md); 1291 1292 write_lock(&md->map_lock); 1293 md->map = t; 1294 dm_table_set_restrictions(t, q); 1295 write_unlock(&md->map_lock); 1296 1297 return 0; 1298 } 1299 1300 static void __unbind(struct mapped_device *md) 1301 { 1302 struct dm_table *map = md->map; 1303 1304 if (!map) 1305 return; 1306 1307 dm_table_event_callback(map, NULL, NULL); 1308 write_lock(&md->map_lock); 1309 md->map = NULL; 1310 write_unlock(&md->map_lock); 1311 dm_table_destroy(map); 1312 } 1313 1314 /* 1315 * Constructor for a new device. 1316 */ 1317 int dm_create(int minor, struct mapped_device **result) 1318 { 1319 struct mapped_device *md; 1320 1321 md = alloc_dev(minor); 1322 if (!md) 1323 return -ENXIO; 1324 1325 dm_sysfs_init(md); 1326 1327 *result = md; 1328 return 0; 1329 } 1330 1331 static struct mapped_device *dm_find_md(dev_t dev) 1332 { 1333 struct mapped_device *md; 1334 unsigned minor = MINOR(dev); 1335 1336 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) 1337 return NULL; 1338 1339 spin_lock(&_minor_lock); 1340 1341 md = idr_find(&_minor_idr, minor); 1342 if (md && (md == MINOR_ALLOCED || 1343 (MINOR(disk_devt(dm_disk(md))) != minor) || 1344 test_bit(DMF_FREEING, &md->flags))) { 1345 md = NULL; 1346 goto out; 1347 } 1348 1349 out: 1350 spin_unlock(&_minor_lock); 1351 1352 return md; 1353 } 1354 1355 struct mapped_device *dm_get_md(dev_t dev) 1356 { 1357 struct mapped_device *md = dm_find_md(dev); 1358 1359 if (md) 1360 dm_get(md); 1361 1362 return md; 1363 } 1364 1365 void *dm_get_mdptr(struct mapped_device *md) 1366 { 1367 return md->interface_ptr; 1368 } 1369 1370 void dm_set_mdptr(struct mapped_device *md, void *ptr) 1371 { 1372 md->interface_ptr = ptr; 1373 } 1374 1375 void dm_get(struct mapped_device *md) 1376 { 1377 atomic_inc(&md->holders); 1378 } 1379 1380 const char *dm_device_name(struct mapped_device *md) 1381 { 1382 return md->name; 1383 } 1384 EXPORT_SYMBOL_GPL(dm_device_name); 1385 1386 void dm_put(struct mapped_device *md) 1387 { 1388 struct dm_table *map; 1389 1390 BUG_ON(test_bit(DMF_FREEING, &md->flags)); 1391 1392 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { 1393 map = dm_get_table(md); 1394 idr_replace(&_minor_idr, MINOR_ALLOCED, 1395 MINOR(disk_devt(dm_disk(md)))); 1396 set_bit(DMF_FREEING, &md->flags); 1397 spin_unlock(&_minor_lock); 1398 if (!dm_suspended(md)) { 1399 dm_table_presuspend_targets(map); 1400 dm_table_postsuspend_targets(map); 1401 } 1402 dm_sysfs_exit(md); 1403 dm_table_put(map); 1404 __unbind(md); 1405 free_dev(md); 1406 } 1407 } 1408 EXPORT_SYMBOL_GPL(dm_put); 1409 1410 static int dm_wait_for_completion(struct mapped_device *md, int interruptible) 1411 { 1412 int r = 0; 1413 DECLARE_WAITQUEUE(wait, current); 1414 1415 dm_unplug_all(md->queue); 1416 1417 add_wait_queue(&md->wait, &wait); 1418 1419 while (1) { 1420 set_current_state(interruptible); 1421 1422 smp_mb(); 1423 if (!atomic_read(&md->pending)) 1424 break; 1425 1426 if (interruptible == TASK_INTERRUPTIBLE && 1427 signal_pending(current)) { 1428 r = -EINTR; 1429 break; 1430 } 1431 1432 io_schedule(); 1433 } 1434 set_current_state(TASK_RUNNING); 1435 1436 remove_wait_queue(&md->wait, &wait); 1437 1438 return r; 1439 } 1440 1441 static void dm_flush(struct mapped_device *md) 1442 { 1443 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); 1444 } 1445 1446 static void process_barrier(struct mapped_device *md, struct bio *bio) 1447 { 1448 md->barrier_error = 0; 1449 1450 dm_flush(md); 1451 1452 if (!bio_empty_barrier(bio)) { 1453 __split_and_process_bio(md, bio); 1454 dm_flush(md); 1455 } 1456 1457 if (md->barrier_error != DM_ENDIO_REQUEUE) 1458 bio_endio(bio, md->barrier_error); 1459 else { 1460 spin_lock_irq(&md->deferred_lock); 1461 bio_list_add_head(&md->deferred, bio); 1462 spin_unlock_irq(&md->deferred_lock); 1463 } 1464 } 1465 1466 /* 1467 * Process the deferred bios 1468 */ 1469 static void dm_wq_work(struct work_struct *work) 1470 { 1471 struct mapped_device *md = container_of(work, struct mapped_device, 1472 work); 1473 struct bio *c; 1474 1475 down_write(&md->io_lock); 1476 1477 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 1478 spin_lock_irq(&md->deferred_lock); 1479 c = bio_list_pop(&md->deferred); 1480 spin_unlock_irq(&md->deferred_lock); 1481 1482 if (!c) { 1483 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); 1484 break; 1485 } 1486 1487 up_write(&md->io_lock); 1488 1489 if (bio_barrier(c)) 1490 process_barrier(md, c); 1491 else 1492 __split_and_process_bio(md, c); 1493 1494 down_write(&md->io_lock); 1495 } 1496 1497 up_write(&md->io_lock); 1498 } 1499 1500 static void dm_queue_flush(struct mapped_device *md) 1501 { 1502 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 1503 smp_mb__after_clear_bit(); 1504 queue_work(md->wq, &md->work); 1505 } 1506 1507 /* 1508 * Swap in a new table (destroying old one). 1509 */ 1510 int dm_swap_table(struct mapped_device *md, struct dm_table *table) 1511 { 1512 int r = -EINVAL; 1513 1514 mutex_lock(&md->suspend_lock); 1515 1516 /* device must be suspended */ 1517 if (!dm_suspended(md)) 1518 goto out; 1519 1520 __unbind(md); 1521 r = __bind(md, table); 1522 1523 out: 1524 mutex_unlock(&md->suspend_lock); 1525 return r; 1526 } 1527 1528 /* 1529 * Functions to lock and unlock any filesystem running on the 1530 * device. 1531 */ 1532 static int lock_fs(struct mapped_device *md) 1533 { 1534 int r; 1535 1536 WARN_ON(md->frozen_sb); 1537 1538 md->frozen_sb = freeze_bdev(md->bdev); 1539 if (IS_ERR(md->frozen_sb)) { 1540 r = PTR_ERR(md->frozen_sb); 1541 md->frozen_sb = NULL; 1542 return r; 1543 } 1544 1545 set_bit(DMF_FROZEN, &md->flags); 1546 1547 return 0; 1548 } 1549 1550 static void unlock_fs(struct mapped_device *md) 1551 { 1552 if (!test_bit(DMF_FROZEN, &md->flags)) 1553 return; 1554 1555 thaw_bdev(md->bdev, md->frozen_sb); 1556 md->frozen_sb = NULL; 1557 clear_bit(DMF_FROZEN, &md->flags); 1558 } 1559 1560 /* 1561 * We need to be able to change a mapping table under a mounted 1562 * filesystem. For example we might want to move some data in 1563 * the background. Before the table can be swapped with 1564 * dm_bind_table, dm_suspend must be called to flush any in 1565 * flight bios and ensure that any further io gets deferred. 1566 */ 1567 int dm_suspend(struct mapped_device *md, unsigned suspend_flags) 1568 { 1569 struct dm_table *map = NULL; 1570 int r = 0; 1571 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; 1572 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; 1573 1574 mutex_lock(&md->suspend_lock); 1575 1576 if (dm_suspended(md)) { 1577 r = -EINVAL; 1578 goto out_unlock; 1579 } 1580 1581 map = dm_get_table(md); 1582 1583 /* 1584 * DMF_NOFLUSH_SUSPENDING must be set before presuspend. 1585 * This flag is cleared before dm_suspend returns. 1586 */ 1587 if (noflush) 1588 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 1589 1590 /* This does not get reverted if there's an error later. */ 1591 dm_table_presuspend_targets(map); 1592 1593 /* 1594 * Flush I/O to the device. noflush supersedes do_lockfs, 1595 * because lock_fs() needs to flush I/Os. 1596 */ 1597 if (!noflush && do_lockfs) { 1598 r = lock_fs(md); 1599 if (r) 1600 goto out; 1601 } 1602 1603 /* 1604 * Here we must make sure that no processes are submitting requests 1605 * to target drivers i.e. no one may be executing 1606 * __split_and_process_bio. This is called from dm_request and 1607 * dm_wq_work. 1608 * 1609 * To get all processes out of __split_and_process_bio in dm_request, 1610 * we take the write lock. To prevent any process from reentering 1611 * __split_and_process_bio from dm_request, we set 1612 * DMF_QUEUE_IO_TO_THREAD. 1613 * 1614 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND 1615 * and call flush_workqueue(md->wq). flush_workqueue will wait until 1616 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any 1617 * further calls to __split_and_process_bio from dm_wq_work. 1618 */ 1619 down_write(&md->io_lock); 1620 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 1621 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); 1622 up_write(&md->io_lock); 1623 1624 flush_workqueue(md->wq); 1625 1626 /* 1627 * At this point no more requests are entering target request routines. 1628 * We call dm_wait_for_completion to wait for all existing requests 1629 * to finish. 1630 */ 1631 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); 1632 1633 down_write(&md->io_lock); 1634 if (noflush) 1635 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 1636 up_write(&md->io_lock); 1637 1638 /* were we interrupted ? */ 1639 if (r < 0) { 1640 dm_queue_flush(md); 1641 1642 unlock_fs(md); 1643 goto out; /* pushback list is already flushed, so skip flush */ 1644 } 1645 1646 /* 1647 * If dm_wait_for_completion returned 0, the device is completely 1648 * quiescent now. There is no request-processing activity. All new 1649 * requests are being added to md->deferred list. 1650 */ 1651 1652 dm_table_postsuspend_targets(map); 1653 1654 set_bit(DMF_SUSPENDED, &md->flags); 1655 1656 out: 1657 dm_table_put(map); 1658 1659 out_unlock: 1660 mutex_unlock(&md->suspend_lock); 1661 return r; 1662 } 1663 1664 int dm_resume(struct mapped_device *md) 1665 { 1666 int r = -EINVAL; 1667 struct dm_table *map = NULL; 1668 1669 mutex_lock(&md->suspend_lock); 1670 if (!dm_suspended(md)) 1671 goto out; 1672 1673 map = dm_get_table(md); 1674 if (!map || !dm_table_get_size(map)) 1675 goto out; 1676 1677 r = dm_table_resume_targets(map); 1678 if (r) 1679 goto out; 1680 1681 dm_queue_flush(md); 1682 1683 unlock_fs(md); 1684 1685 clear_bit(DMF_SUSPENDED, &md->flags); 1686 1687 dm_table_unplug_all(map); 1688 1689 dm_kobject_uevent(md); 1690 1691 r = 0; 1692 1693 out: 1694 dm_table_put(map); 1695 mutex_unlock(&md->suspend_lock); 1696 1697 return r; 1698 } 1699 1700 /*----------------------------------------------------------------- 1701 * Event notification. 1702 *---------------------------------------------------------------*/ 1703 void dm_kobject_uevent(struct mapped_device *md) 1704 { 1705 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE); 1706 } 1707 1708 uint32_t dm_next_uevent_seq(struct mapped_device *md) 1709 { 1710 return atomic_add_return(1, &md->uevent_seq); 1711 } 1712 1713 uint32_t dm_get_event_nr(struct mapped_device *md) 1714 { 1715 return atomic_read(&md->event_nr); 1716 } 1717 1718 int dm_wait_event(struct mapped_device *md, int event_nr) 1719 { 1720 return wait_event_interruptible(md->eventq, 1721 (event_nr != atomic_read(&md->event_nr))); 1722 } 1723 1724 void dm_uevent_add(struct mapped_device *md, struct list_head *elist) 1725 { 1726 unsigned long flags; 1727 1728 spin_lock_irqsave(&md->uevent_lock, flags); 1729 list_add(elist, &md->uevent_list); 1730 spin_unlock_irqrestore(&md->uevent_lock, flags); 1731 } 1732 1733 /* 1734 * The gendisk is only valid as long as you have a reference 1735 * count on 'md'. 1736 */ 1737 struct gendisk *dm_disk(struct mapped_device *md) 1738 { 1739 return md->disk; 1740 } 1741 1742 struct kobject *dm_kobject(struct mapped_device *md) 1743 { 1744 return &md->kobj; 1745 } 1746 1747 /* 1748 * struct mapped_device should not be exported outside of dm.c 1749 * so use this check to verify that kobj is part of md structure 1750 */ 1751 struct mapped_device *dm_get_from_kobject(struct kobject *kobj) 1752 { 1753 struct mapped_device *md; 1754 1755 md = container_of(kobj, struct mapped_device, kobj); 1756 if (&md->kobj != kobj) 1757 return NULL; 1758 1759 if (test_bit(DMF_FREEING, &md->flags) || 1760 test_bit(DMF_DELETING, &md->flags)) 1761 return NULL; 1762 1763 dm_get(md); 1764 return md; 1765 } 1766 1767 int dm_suspended(struct mapped_device *md) 1768 { 1769 return test_bit(DMF_SUSPENDED, &md->flags); 1770 } 1771 1772 int dm_noflush_suspending(struct dm_target *ti) 1773 { 1774 struct mapped_device *md = dm_table_get_md(ti->table); 1775 int r = __noflush_suspending(md); 1776 1777 dm_put(md); 1778 1779 return r; 1780 } 1781 EXPORT_SYMBOL_GPL(dm_noflush_suspending); 1782 1783 static struct block_device_operations dm_blk_dops = { 1784 .open = dm_blk_open, 1785 .release = dm_blk_close, 1786 .ioctl = dm_blk_ioctl, 1787 .getgeo = dm_blk_getgeo, 1788 .owner = THIS_MODULE 1789 }; 1790 1791 EXPORT_SYMBOL(dm_get_mapinfo); 1792 1793 /* 1794 * module hooks 1795 */ 1796 module_init(dm_init); 1797 module_exit(dm_exit); 1798 1799 module_param(major, uint, 0); 1800 MODULE_PARM_DESC(major, "The major number of the device mapper"); 1801 MODULE_DESCRIPTION(DM_NAME " driver"); 1802 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); 1803 MODULE_LICENSE("GPL"); 1804