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 #include <linux/delay.h> 23 24 #include <trace/events/block.h> 25 26 #define DM_MSG_PREFIX "core" 27 28 /* 29 * Cookies are numeric values sent with CHANGE and REMOVE 30 * uevents while resuming, removing or renaming the device. 31 */ 32 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" 33 #define DM_COOKIE_LENGTH 24 34 35 static const char *_name = DM_NAME; 36 37 static unsigned int major = 0; 38 static unsigned int _major = 0; 39 40 static DEFINE_SPINLOCK(_minor_lock); 41 /* 42 * For bio-based dm. 43 * One of these is allocated per bio. 44 */ 45 struct dm_io { 46 struct mapped_device *md; 47 int error; 48 atomic_t io_count; 49 struct bio *bio; 50 unsigned long start_time; 51 spinlock_t endio_lock; 52 }; 53 54 /* 55 * For bio-based dm. 56 * One of these is allocated per target within a bio. Hopefully 57 * this will be simplified out one day. 58 */ 59 struct dm_target_io { 60 struct dm_io *io; 61 struct dm_target *ti; 62 union map_info info; 63 }; 64 65 /* 66 * For request-based dm. 67 * One of these is allocated per request. 68 */ 69 struct dm_rq_target_io { 70 struct mapped_device *md; 71 struct dm_target *ti; 72 struct request *orig, clone; 73 int error; 74 union map_info info; 75 }; 76 77 /* 78 * For request-based dm. 79 * One of these is allocated per bio. 80 */ 81 struct dm_rq_clone_bio_info { 82 struct bio *orig; 83 struct dm_rq_target_io *tio; 84 }; 85 86 union map_info *dm_get_mapinfo(struct bio *bio) 87 { 88 if (bio && bio->bi_private) 89 return &((struct dm_target_io *)bio->bi_private)->info; 90 return NULL; 91 } 92 93 union map_info *dm_get_rq_mapinfo(struct request *rq) 94 { 95 if (rq && rq->end_io_data) 96 return &((struct dm_rq_target_io *)rq->end_io_data)->info; 97 return NULL; 98 } 99 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); 100 101 #define MINOR_ALLOCED ((void *)-1) 102 103 /* 104 * Bits for the md->flags field. 105 */ 106 #define DMF_BLOCK_IO_FOR_SUSPEND 0 107 #define DMF_SUSPENDED 1 108 #define DMF_FROZEN 2 109 #define DMF_FREEING 3 110 #define DMF_DELETING 4 111 #define DMF_NOFLUSH_SUSPENDING 5 112 113 /* 114 * Work processed by per-device workqueue. 115 */ 116 struct mapped_device { 117 struct rw_semaphore io_lock; 118 struct mutex suspend_lock; 119 rwlock_t map_lock; 120 atomic_t holders; 121 atomic_t open_count; 122 123 unsigned long flags; 124 125 struct request_queue *queue; 126 unsigned type; 127 /* Protect queue and type against concurrent access. */ 128 struct mutex type_lock; 129 130 struct gendisk *disk; 131 char name[16]; 132 133 void *interface_ptr; 134 135 /* 136 * A list of ios that arrived while we were suspended. 137 */ 138 atomic_t pending[2]; 139 wait_queue_head_t wait; 140 struct work_struct work; 141 struct bio_list deferred; 142 spinlock_t deferred_lock; 143 144 /* 145 * Processing queue (flush) 146 */ 147 struct workqueue_struct *wq; 148 149 /* 150 * The current mapping. 151 */ 152 struct dm_table *map; 153 154 /* 155 * io objects are allocated from here. 156 */ 157 mempool_t *io_pool; 158 mempool_t *tio_pool; 159 160 struct bio_set *bs; 161 162 /* 163 * Event handling. 164 */ 165 atomic_t event_nr; 166 wait_queue_head_t eventq; 167 atomic_t uevent_seq; 168 struct list_head uevent_list; 169 spinlock_t uevent_lock; /* Protect access to uevent_list */ 170 171 /* 172 * freeze/thaw support require holding onto a super block 173 */ 174 struct super_block *frozen_sb; 175 struct block_device *bdev; 176 177 /* forced geometry settings */ 178 struct hd_geometry geometry; 179 180 /* For saving the address of __make_request for request based dm */ 181 make_request_fn *saved_make_request_fn; 182 183 /* sysfs handle */ 184 struct kobject kobj; 185 186 /* zero-length flush that will be cloned and submitted to targets */ 187 struct bio flush_bio; 188 }; 189 190 /* 191 * For mempools pre-allocation at the table loading time. 192 */ 193 struct dm_md_mempools { 194 mempool_t *io_pool; 195 mempool_t *tio_pool; 196 struct bio_set *bs; 197 }; 198 199 #define MIN_IOS 256 200 static struct kmem_cache *_io_cache; 201 static struct kmem_cache *_tio_cache; 202 static struct kmem_cache *_rq_tio_cache; 203 static struct kmem_cache *_rq_bio_info_cache; 204 205 static int __init local_init(void) 206 { 207 int r = -ENOMEM; 208 209 /* allocate a slab for the dm_ios */ 210 _io_cache = KMEM_CACHE(dm_io, 0); 211 if (!_io_cache) 212 return r; 213 214 /* allocate a slab for the target ios */ 215 _tio_cache = KMEM_CACHE(dm_target_io, 0); 216 if (!_tio_cache) 217 goto out_free_io_cache; 218 219 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); 220 if (!_rq_tio_cache) 221 goto out_free_tio_cache; 222 223 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); 224 if (!_rq_bio_info_cache) 225 goto out_free_rq_tio_cache; 226 227 r = dm_uevent_init(); 228 if (r) 229 goto out_free_rq_bio_info_cache; 230 231 _major = major; 232 r = register_blkdev(_major, _name); 233 if (r < 0) 234 goto out_uevent_exit; 235 236 if (!_major) 237 _major = r; 238 239 return 0; 240 241 out_uevent_exit: 242 dm_uevent_exit(); 243 out_free_rq_bio_info_cache: 244 kmem_cache_destroy(_rq_bio_info_cache); 245 out_free_rq_tio_cache: 246 kmem_cache_destroy(_rq_tio_cache); 247 out_free_tio_cache: 248 kmem_cache_destroy(_tio_cache); 249 out_free_io_cache: 250 kmem_cache_destroy(_io_cache); 251 252 return r; 253 } 254 255 static void local_exit(void) 256 { 257 kmem_cache_destroy(_rq_bio_info_cache); 258 kmem_cache_destroy(_rq_tio_cache); 259 kmem_cache_destroy(_tio_cache); 260 kmem_cache_destroy(_io_cache); 261 unregister_blkdev(_major, _name); 262 dm_uevent_exit(); 263 264 _major = 0; 265 266 DMINFO("cleaned up"); 267 } 268 269 static int (*_inits[])(void) __initdata = { 270 local_init, 271 dm_target_init, 272 dm_linear_init, 273 dm_stripe_init, 274 dm_io_init, 275 dm_kcopyd_init, 276 dm_interface_init, 277 }; 278 279 static void (*_exits[])(void) = { 280 local_exit, 281 dm_target_exit, 282 dm_linear_exit, 283 dm_stripe_exit, 284 dm_io_exit, 285 dm_kcopyd_exit, 286 dm_interface_exit, 287 }; 288 289 static int __init dm_init(void) 290 { 291 const int count = ARRAY_SIZE(_inits); 292 293 int r, i; 294 295 for (i = 0; i < count; i++) { 296 r = _inits[i](); 297 if (r) 298 goto bad; 299 } 300 301 return 0; 302 303 bad: 304 while (i--) 305 _exits[i](); 306 307 return r; 308 } 309 310 static void __exit dm_exit(void) 311 { 312 int i = ARRAY_SIZE(_exits); 313 314 while (i--) 315 _exits[i](); 316 } 317 318 /* 319 * Block device functions 320 */ 321 int dm_deleting_md(struct mapped_device *md) 322 { 323 return test_bit(DMF_DELETING, &md->flags); 324 } 325 326 static int dm_blk_open(struct block_device *bdev, fmode_t mode) 327 { 328 struct mapped_device *md; 329 330 spin_lock(&_minor_lock); 331 332 md = bdev->bd_disk->private_data; 333 if (!md) 334 goto out; 335 336 if (test_bit(DMF_FREEING, &md->flags) || 337 dm_deleting_md(md)) { 338 md = NULL; 339 goto out; 340 } 341 342 dm_get(md); 343 atomic_inc(&md->open_count); 344 345 out: 346 spin_unlock(&_minor_lock); 347 348 return md ? 0 : -ENXIO; 349 } 350 351 static int dm_blk_close(struct gendisk *disk, fmode_t mode) 352 { 353 struct mapped_device *md = disk->private_data; 354 355 spin_lock(&_minor_lock); 356 357 atomic_dec(&md->open_count); 358 dm_put(md); 359 360 spin_unlock(&_minor_lock); 361 362 return 0; 363 } 364 365 int dm_open_count(struct mapped_device *md) 366 { 367 return atomic_read(&md->open_count); 368 } 369 370 /* 371 * Guarantees nothing is using the device before it's deleted. 372 */ 373 int dm_lock_for_deletion(struct mapped_device *md) 374 { 375 int r = 0; 376 377 spin_lock(&_minor_lock); 378 379 if (dm_open_count(md)) 380 r = -EBUSY; 381 else 382 set_bit(DMF_DELETING, &md->flags); 383 384 spin_unlock(&_minor_lock); 385 386 return r; 387 } 388 389 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) 390 { 391 struct mapped_device *md = bdev->bd_disk->private_data; 392 393 return dm_get_geometry(md, geo); 394 } 395 396 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, 397 unsigned int cmd, unsigned long arg) 398 { 399 struct mapped_device *md = bdev->bd_disk->private_data; 400 struct dm_table *map = dm_get_live_table(md); 401 struct dm_target *tgt; 402 int r = -ENOTTY; 403 404 if (!map || !dm_table_get_size(map)) 405 goto out; 406 407 /* We only support devices that have a single target */ 408 if (dm_table_get_num_targets(map) != 1) 409 goto out; 410 411 tgt = dm_table_get_target(map, 0); 412 413 if (dm_suspended_md(md)) { 414 r = -EAGAIN; 415 goto out; 416 } 417 418 if (tgt->type->ioctl) 419 r = tgt->type->ioctl(tgt, cmd, arg); 420 421 out: 422 dm_table_put(map); 423 424 return r; 425 } 426 427 static struct dm_io *alloc_io(struct mapped_device *md) 428 { 429 return mempool_alloc(md->io_pool, GFP_NOIO); 430 } 431 432 static void free_io(struct mapped_device *md, struct dm_io *io) 433 { 434 mempool_free(io, md->io_pool); 435 } 436 437 static void free_tio(struct mapped_device *md, struct dm_target_io *tio) 438 { 439 mempool_free(tio, md->tio_pool); 440 } 441 442 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, 443 gfp_t gfp_mask) 444 { 445 return mempool_alloc(md->tio_pool, gfp_mask); 446 } 447 448 static void free_rq_tio(struct dm_rq_target_io *tio) 449 { 450 mempool_free(tio, tio->md->tio_pool); 451 } 452 453 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md) 454 { 455 return mempool_alloc(md->io_pool, GFP_ATOMIC); 456 } 457 458 static void free_bio_info(struct dm_rq_clone_bio_info *info) 459 { 460 mempool_free(info, info->tio->md->io_pool); 461 } 462 463 static int md_in_flight(struct mapped_device *md) 464 { 465 return atomic_read(&md->pending[READ]) + 466 atomic_read(&md->pending[WRITE]); 467 } 468 469 static void start_io_acct(struct dm_io *io) 470 { 471 struct mapped_device *md = io->md; 472 int cpu; 473 int rw = bio_data_dir(io->bio); 474 475 io->start_time = jiffies; 476 477 cpu = part_stat_lock(); 478 part_round_stats(cpu, &dm_disk(md)->part0); 479 part_stat_unlock(); 480 atomic_set(&dm_disk(md)->part0.in_flight[rw], 481 atomic_inc_return(&md->pending[rw])); 482 } 483 484 static void end_io_acct(struct dm_io *io) 485 { 486 struct mapped_device *md = io->md; 487 struct bio *bio = io->bio; 488 unsigned long duration = jiffies - io->start_time; 489 int pending, cpu; 490 int rw = bio_data_dir(bio); 491 492 cpu = part_stat_lock(); 493 part_round_stats(cpu, &dm_disk(md)->part0); 494 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); 495 part_stat_unlock(); 496 497 /* 498 * After this is decremented the bio must not be touched if it is 499 * a flush. 500 */ 501 pending = atomic_dec_return(&md->pending[rw]); 502 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); 503 pending += atomic_read(&md->pending[rw^0x1]); 504 505 /* nudge anyone waiting on suspend queue */ 506 if (!pending) 507 wake_up(&md->wait); 508 } 509 510 /* 511 * Add the bio to the list of deferred io. 512 */ 513 static void queue_io(struct mapped_device *md, struct bio *bio) 514 { 515 unsigned long flags; 516 517 spin_lock_irqsave(&md->deferred_lock, flags); 518 bio_list_add(&md->deferred, bio); 519 spin_unlock_irqrestore(&md->deferred_lock, flags); 520 queue_work(md->wq, &md->work); 521 } 522 523 /* 524 * Everyone (including functions in this file), should use this 525 * function to access the md->map field, and make sure they call 526 * dm_table_put() when finished. 527 */ 528 struct dm_table *dm_get_live_table(struct mapped_device *md) 529 { 530 struct dm_table *t; 531 unsigned long flags; 532 533 read_lock_irqsave(&md->map_lock, flags); 534 t = md->map; 535 if (t) 536 dm_table_get(t); 537 read_unlock_irqrestore(&md->map_lock, flags); 538 539 return t; 540 } 541 542 /* 543 * Get the geometry associated with a dm device 544 */ 545 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) 546 { 547 *geo = md->geometry; 548 549 return 0; 550 } 551 552 /* 553 * Set the geometry of a device. 554 */ 555 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) 556 { 557 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; 558 559 if (geo->start > sz) { 560 DMWARN("Start sector is beyond the geometry limits."); 561 return -EINVAL; 562 } 563 564 md->geometry = *geo; 565 566 return 0; 567 } 568 569 /*----------------------------------------------------------------- 570 * CRUD START: 571 * A more elegant soln is in the works that uses the queue 572 * merge fn, unfortunately there are a couple of changes to 573 * the block layer that I want to make for this. So in the 574 * interests of getting something for people to use I give 575 * you this clearly demarcated crap. 576 *---------------------------------------------------------------*/ 577 578 static int __noflush_suspending(struct mapped_device *md) 579 { 580 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 581 } 582 583 /* 584 * Decrements the number of outstanding ios that a bio has been 585 * cloned into, completing the original io if necc. 586 */ 587 static void dec_pending(struct dm_io *io, int error) 588 { 589 unsigned long flags; 590 int io_error; 591 struct bio *bio; 592 struct mapped_device *md = io->md; 593 594 /* Push-back supersedes any I/O errors */ 595 if (unlikely(error)) { 596 spin_lock_irqsave(&io->endio_lock, flags); 597 if (!(io->error > 0 && __noflush_suspending(md))) 598 io->error = error; 599 spin_unlock_irqrestore(&io->endio_lock, flags); 600 } 601 602 if (atomic_dec_and_test(&io->io_count)) { 603 if (io->error == DM_ENDIO_REQUEUE) { 604 /* 605 * Target requested pushing back the I/O. 606 */ 607 spin_lock_irqsave(&md->deferred_lock, flags); 608 if (__noflush_suspending(md)) 609 bio_list_add_head(&md->deferred, io->bio); 610 else 611 /* noflush suspend was interrupted. */ 612 io->error = -EIO; 613 spin_unlock_irqrestore(&md->deferred_lock, flags); 614 } 615 616 io_error = io->error; 617 bio = io->bio; 618 end_io_acct(io); 619 free_io(md, io); 620 621 if (io_error == DM_ENDIO_REQUEUE) 622 return; 623 624 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { 625 /* 626 * Preflush done for flush with data, reissue 627 * without REQ_FLUSH. 628 */ 629 bio->bi_rw &= ~REQ_FLUSH; 630 queue_io(md, bio); 631 } else { 632 /* done with normal IO or empty flush */ 633 trace_block_bio_complete(md->queue, bio, io_error); 634 bio_endio(bio, io_error); 635 } 636 } 637 } 638 639 static void clone_endio(struct bio *bio, int error) 640 { 641 int r = 0; 642 struct dm_target_io *tio = bio->bi_private; 643 struct dm_io *io = tio->io; 644 struct mapped_device *md = tio->io->md; 645 dm_endio_fn endio = tio->ti->type->end_io; 646 647 if (!bio_flagged(bio, BIO_UPTODATE) && !error) 648 error = -EIO; 649 650 if (endio) { 651 r = endio(tio->ti, bio, error, &tio->info); 652 if (r < 0 || r == DM_ENDIO_REQUEUE) 653 /* 654 * error and requeue request are handled 655 * in dec_pending(). 656 */ 657 error = r; 658 else if (r == DM_ENDIO_INCOMPLETE) 659 /* The target will handle the io */ 660 return; 661 else if (r) { 662 DMWARN("unimplemented target endio return value: %d", r); 663 BUG(); 664 } 665 } 666 667 /* 668 * Store md for cleanup instead of tio which is about to get freed. 669 */ 670 bio->bi_private = md->bs; 671 672 free_tio(md, tio); 673 bio_put(bio); 674 dec_pending(io, error); 675 } 676 677 /* 678 * Partial completion handling for request-based dm 679 */ 680 static void end_clone_bio(struct bio *clone, int error) 681 { 682 struct dm_rq_clone_bio_info *info = clone->bi_private; 683 struct dm_rq_target_io *tio = info->tio; 684 struct bio *bio = info->orig; 685 unsigned int nr_bytes = info->orig->bi_size; 686 687 bio_put(clone); 688 689 if (tio->error) 690 /* 691 * An error has already been detected on the request. 692 * Once error occurred, just let clone->end_io() handle 693 * the remainder. 694 */ 695 return; 696 else if (error) { 697 /* 698 * Don't notice the error to the upper layer yet. 699 * The error handling decision is made by the target driver, 700 * when the request is completed. 701 */ 702 tio->error = error; 703 return; 704 } 705 706 /* 707 * I/O for the bio successfully completed. 708 * Notice the data completion to the upper layer. 709 */ 710 711 /* 712 * bios are processed from the head of the list. 713 * So the completing bio should always be rq->bio. 714 * If it's not, something wrong is happening. 715 */ 716 if (tio->orig->bio != bio) 717 DMERR("bio completion is going in the middle of the request"); 718 719 /* 720 * Update the original request. 721 * Do not use blk_end_request() here, because it may complete 722 * the original request before the clone, and break the ordering. 723 */ 724 blk_update_request(tio->orig, 0, nr_bytes); 725 } 726 727 /* 728 * Don't touch any member of the md after calling this function because 729 * the md may be freed in dm_put() at the end of this function. 730 * Or do dm_get() before calling this function and dm_put() later. 731 */ 732 static void rq_completed(struct mapped_device *md, int rw, int run_queue) 733 { 734 atomic_dec(&md->pending[rw]); 735 736 /* nudge anyone waiting on suspend queue */ 737 if (!md_in_flight(md)) 738 wake_up(&md->wait); 739 740 if (run_queue) 741 blk_run_queue(md->queue); 742 743 /* 744 * dm_put() must be at the end of this function. See the comment above 745 */ 746 dm_put(md); 747 } 748 749 static void free_rq_clone(struct request *clone) 750 { 751 struct dm_rq_target_io *tio = clone->end_io_data; 752 753 blk_rq_unprep_clone(clone); 754 free_rq_tio(tio); 755 } 756 757 /* 758 * Complete the clone and the original request. 759 * Must be called without queue lock. 760 */ 761 static void dm_end_request(struct request *clone, int error) 762 { 763 int rw = rq_data_dir(clone); 764 struct dm_rq_target_io *tio = clone->end_io_data; 765 struct mapped_device *md = tio->md; 766 struct request *rq = tio->orig; 767 768 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { 769 rq->errors = clone->errors; 770 rq->resid_len = clone->resid_len; 771 772 if (rq->sense) 773 /* 774 * We are using the sense buffer of the original 775 * request. 776 * So setting the length of the sense data is enough. 777 */ 778 rq->sense_len = clone->sense_len; 779 } 780 781 free_rq_clone(clone); 782 blk_end_request_all(rq, error); 783 rq_completed(md, rw, true); 784 } 785 786 static void dm_unprep_request(struct request *rq) 787 { 788 struct request *clone = rq->special; 789 790 rq->special = NULL; 791 rq->cmd_flags &= ~REQ_DONTPREP; 792 793 free_rq_clone(clone); 794 } 795 796 /* 797 * Requeue the original request of a clone. 798 */ 799 void dm_requeue_unmapped_request(struct request *clone) 800 { 801 int rw = rq_data_dir(clone); 802 struct dm_rq_target_io *tio = clone->end_io_data; 803 struct mapped_device *md = tio->md; 804 struct request *rq = tio->orig; 805 struct request_queue *q = rq->q; 806 unsigned long flags; 807 808 dm_unprep_request(rq); 809 810 spin_lock_irqsave(q->queue_lock, flags); 811 blk_requeue_request(q, rq); 812 spin_unlock_irqrestore(q->queue_lock, flags); 813 814 rq_completed(md, rw, 0); 815 } 816 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); 817 818 static void __stop_queue(struct request_queue *q) 819 { 820 blk_stop_queue(q); 821 } 822 823 static void stop_queue(struct request_queue *q) 824 { 825 unsigned long flags; 826 827 spin_lock_irqsave(q->queue_lock, flags); 828 __stop_queue(q); 829 spin_unlock_irqrestore(q->queue_lock, flags); 830 } 831 832 static void __start_queue(struct request_queue *q) 833 { 834 if (blk_queue_stopped(q)) 835 blk_start_queue(q); 836 } 837 838 static void start_queue(struct request_queue *q) 839 { 840 unsigned long flags; 841 842 spin_lock_irqsave(q->queue_lock, flags); 843 __start_queue(q); 844 spin_unlock_irqrestore(q->queue_lock, flags); 845 } 846 847 static void dm_done(struct request *clone, int error, bool mapped) 848 { 849 int r = error; 850 struct dm_rq_target_io *tio = clone->end_io_data; 851 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io; 852 853 if (mapped && rq_end_io) 854 r = rq_end_io(tio->ti, clone, error, &tio->info); 855 856 if (r <= 0) 857 /* The target wants to complete the I/O */ 858 dm_end_request(clone, r); 859 else if (r == DM_ENDIO_INCOMPLETE) 860 /* The target will handle the I/O */ 861 return; 862 else if (r == DM_ENDIO_REQUEUE) 863 /* The target wants to requeue the I/O */ 864 dm_requeue_unmapped_request(clone); 865 else { 866 DMWARN("unimplemented target endio return value: %d", r); 867 BUG(); 868 } 869 } 870 871 /* 872 * Request completion handler for request-based dm 873 */ 874 static void dm_softirq_done(struct request *rq) 875 { 876 bool mapped = true; 877 struct request *clone = rq->completion_data; 878 struct dm_rq_target_io *tio = clone->end_io_data; 879 880 if (rq->cmd_flags & REQ_FAILED) 881 mapped = false; 882 883 dm_done(clone, tio->error, mapped); 884 } 885 886 /* 887 * Complete the clone and the original request with the error status 888 * through softirq context. 889 */ 890 static void dm_complete_request(struct request *clone, int error) 891 { 892 struct dm_rq_target_io *tio = clone->end_io_data; 893 struct request *rq = tio->orig; 894 895 tio->error = error; 896 rq->completion_data = clone; 897 blk_complete_request(rq); 898 } 899 900 /* 901 * Complete the not-mapped clone and the original request with the error status 902 * through softirq context. 903 * Target's rq_end_io() function isn't called. 904 * This may be used when the target's map_rq() function fails. 905 */ 906 void dm_kill_unmapped_request(struct request *clone, int error) 907 { 908 struct dm_rq_target_io *tio = clone->end_io_data; 909 struct request *rq = tio->orig; 910 911 rq->cmd_flags |= REQ_FAILED; 912 dm_complete_request(clone, error); 913 } 914 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); 915 916 /* 917 * Called with the queue lock held 918 */ 919 static void end_clone_request(struct request *clone, int error) 920 { 921 /* 922 * For just cleaning up the information of the queue in which 923 * the clone was dispatched. 924 * The clone is *NOT* freed actually here because it is alloced from 925 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. 926 */ 927 __blk_put_request(clone->q, clone); 928 929 /* 930 * Actual request completion is done in a softirq context which doesn't 931 * hold the queue lock. Otherwise, deadlock could occur because: 932 * - another request may be submitted by the upper level driver 933 * of the stacking during the completion 934 * - the submission which requires queue lock may be done 935 * against this queue 936 */ 937 dm_complete_request(clone, error); 938 } 939 940 /* 941 * Return maximum size of I/O possible at the supplied sector up to the current 942 * target boundary. 943 */ 944 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) 945 { 946 sector_t target_offset = dm_target_offset(ti, sector); 947 948 return ti->len - target_offset; 949 } 950 951 static sector_t max_io_len(sector_t sector, struct dm_target *ti) 952 { 953 sector_t len = max_io_len_target_boundary(sector, ti); 954 955 /* 956 * Does the target need to split even further ? 957 */ 958 if (ti->split_io) { 959 sector_t boundary; 960 sector_t offset = dm_target_offset(ti, sector); 961 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) 962 - offset; 963 if (len > boundary) 964 len = boundary; 965 } 966 967 return len; 968 } 969 970 static void __map_bio(struct dm_target *ti, struct bio *clone, 971 struct dm_target_io *tio) 972 { 973 int r; 974 sector_t sector; 975 struct mapped_device *md; 976 977 clone->bi_end_io = clone_endio; 978 clone->bi_private = tio; 979 980 /* 981 * Map the clone. If r == 0 we don't need to do 982 * anything, the target has assumed ownership of 983 * this io. 984 */ 985 atomic_inc(&tio->io->io_count); 986 sector = clone->bi_sector; 987 r = ti->type->map(ti, clone, &tio->info); 988 if (r == DM_MAPIO_REMAPPED) { 989 /* the bio has been remapped so dispatch it */ 990 991 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, 992 tio->io->bio->bi_bdev->bd_dev, sector); 993 994 generic_make_request(clone); 995 } else if (r < 0 || r == DM_MAPIO_REQUEUE) { 996 /* error the io and bail out, or requeue it if needed */ 997 md = tio->io->md; 998 dec_pending(tio->io, r); 999 /* 1000 * Store bio_set for cleanup. 1001 */ 1002 clone->bi_private = md->bs; 1003 bio_put(clone); 1004 free_tio(md, tio); 1005 } else if (r) { 1006 DMWARN("unimplemented target map return value: %d", r); 1007 BUG(); 1008 } 1009 } 1010 1011 struct clone_info { 1012 struct mapped_device *md; 1013 struct dm_table *map; 1014 struct bio *bio; 1015 struct dm_io *io; 1016 sector_t sector; 1017 sector_t sector_count; 1018 unsigned short idx; 1019 }; 1020 1021 static void dm_bio_destructor(struct bio *bio) 1022 { 1023 struct bio_set *bs = bio->bi_private; 1024 1025 bio_free(bio, bs); 1026 } 1027 1028 /* 1029 * Creates a little bio that just does part of a bvec. 1030 */ 1031 static struct bio *split_bvec(struct bio *bio, sector_t sector, 1032 unsigned short idx, unsigned int offset, 1033 unsigned int len, struct bio_set *bs) 1034 { 1035 struct bio *clone; 1036 struct bio_vec *bv = bio->bi_io_vec + idx; 1037 1038 clone = bio_alloc_bioset(GFP_NOIO, 1, bs); 1039 clone->bi_destructor = dm_bio_destructor; 1040 *clone->bi_io_vec = *bv; 1041 1042 clone->bi_sector = sector; 1043 clone->bi_bdev = bio->bi_bdev; 1044 clone->bi_rw = bio->bi_rw; 1045 clone->bi_vcnt = 1; 1046 clone->bi_size = to_bytes(len); 1047 clone->bi_io_vec->bv_offset = offset; 1048 clone->bi_io_vec->bv_len = clone->bi_size; 1049 clone->bi_flags |= 1 << BIO_CLONED; 1050 1051 if (bio_integrity(bio)) { 1052 bio_integrity_clone(clone, bio, GFP_NOIO, bs); 1053 bio_integrity_trim(clone, 1054 bio_sector_offset(bio, idx, offset), len); 1055 } 1056 1057 return clone; 1058 } 1059 1060 /* 1061 * Creates a bio that consists of range of complete bvecs. 1062 */ 1063 static struct bio *clone_bio(struct bio *bio, sector_t sector, 1064 unsigned short idx, unsigned short bv_count, 1065 unsigned int len, struct bio_set *bs) 1066 { 1067 struct bio *clone; 1068 1069 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); 1070 __bio_clone(clone, bio); 1071 clone->bi_destructor = dm_bio_destructor; 1072 clone->bi_sector = sector; 1073 clone->bi_idx = idx; 1074 clone->bi_vcnt = idx + bv_count; 1075 clone->bi_size = to_bytes(len); 1076 clone->bi_flags &= ~(1 << BIO_SEG_VALID); 1077 1078 if (bio_integrity(bio)) { 1079 bio_integrity_clone(clone, bio, GFP_NOIO, bs); 1080 1081 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) 1082 bio_integrity_trim(clone, 1083 bio_sector_offset(bio, idx, 0), len); 1084 } 1085 1086 return clone; 1087 } 1088 1089 static struct dm_target_io *alloc_tio(struct clone_info *ci, 1090 struct dm_target *ti) 1091 { 1092 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); 1093 1094 tio->io = ci->io; 1095 tio->ti = ti; 1096 memset(&tio->info, 0, sizeof(tio->info)); 1097 1098 return tio; 1099 } 1100 1101 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti, 1102 unsigned request_nr, sector_t len) 1103 { 1104 struct dm_target_io *tio = alloc_tio(ci, ti); 1105 struct bio *clone; 1106 1107 tio->info.target_request_nr = request_nr; 1108 1109 /* 1110 * Discard requests require the bio's inline iovecs be initialized. 1111 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush 1112 * and discard, so no need for concern about wasted bvec allocations. 1113 */ 1114 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs); 1115 __bio_clone(clone, ci->bio); 1116 clone->bi_destructor = dm_bio_destructor; 1117 if (len) { 1118 clone->bi_sector = ci->sector; 1119 clone->bi_size = to_bytes(len); 1120 } 1121 1122 __map_bio(ti, clone, tio); 1123 } 1124 1125 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti, 1126 unsigned num_requests, sector_t len) 1127 { 1128 unsigned request_nr; 1129 1130 for (request_nr = 0; request_nr < num_requests; request_nr++) 1131 __issue_target_request(ci, ti, request_nr, len); 1132 } 1133 1134 static int __clone_and_map_empty_flush(struct clone_info *ci) 1135 { 1136 unsigned target_nr = 0; 1137 struct dm_target *ti; 1138 1139 BUG_ON(bio_has_data(ci->bio)); 1140 while ((ti = dm_table_get_target(ci->map, target_nr++))) 1141 __issue_target_requests(ci, ti, ti->num_flush_requests, 0); 1142 1143 return 0; 1144 } 1145 1146 /* 1147 * Perform all io with a single clone. 1148 */ 1149 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti) 1150 { 1151 struct bio *clone, *bio = ci->bio; 1152 struct dm_target_io *tio; 1153 1154 tio = alloc_tio(ci, ti); 1155 clone = clone_bio(bio, ci->sector, ci->idx, 1156 bio->bi_vcnt - ci->idx, ci->sector_count, 1157 ci->md->bs); 1158 __map_bio(ti, clone, tio); 1159 ci->sector_count = 0; 1160 } 1161 1162 static int __clone_and_map_discard(struct clone_info *ci) 1163 { 1164 struct dm_target *ti; 1165 sector_t len; 1166 1167 do { 1168 ti = dm_table_find_target(ci->map, ci->sector); 1169 if (!dm_target_is_valid(ti)) 1170 return -EIO; 1171 1172 /* 1173 * Even though the device advertised discard support, 1174 * reconfiguration might have changed that since the 1175 * check was performed. 1176 */ 1177 if (!ti->num_discard_requests) 1178 return -EOPNOTSUPP; 1179 1180 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); 1181 1182 __issue_target_requests(ci, ti, ti->num_discard_requests, len); 1183 1184 ci->sector += len; 1185 } while (ci->sector_count -= len); 1186 1187 return 0; 1188 } 1189 1190 static int __clone_and_map(struct clone_info *ci) 1191 { 1192 struct bio *clone, *bio = ci->bio; 1193 struct dm_target *ti; 1194 sector_t len = 0, max; 1195 struct dm_target_io *tio; 1196 1197 if (unlikely(bio->bi_rw & REQ_DISCARD)) 1198 return __clone_and_map_discard(ci); 1199 1200 ti = dm_table_find_target(ci->map, ci->sector); 1201 if (!dm_target_is_valid(ti)) 1202 return -EIO; 1203 1204 max = max_io_len(ci->sector, ti); 1205 1206 if (ci->sector_count <= max) { 1207 /* 1208 * Optimise for the simple case where we can do all of 1209 * the remaining io with a single clone. 1210 */ 1211 __clone_and_map_simple(ci, ti); 1212 1213 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { 1214 /* 1215 * There are some bvecs that don't span targets. 1216 * Do as many of these as possible. 1217 */ 1218 int i; 1219 sector_t remaining = max; 1220 sector_t bv_len; 1221 1222 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { 1223 bv_len = to_sector(bio->bi_io_vec[i].bv_len); 1224 1225 if (bv_len > remaining) 1226 break; 1227 1228 remaining -= bv_len; 1229 len += bv_len; 1230 } 1231 1232 tio = alloc_tio(ci, ti); 1233 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, 1234 ci->md->bs); 1235 __map_bio(ti, clone, tio); 1236 1237 ci->sector += len; 1238 ci->sector_count -= len; 1239 ci->idx = i; 1240 1241 } else { 1242 /* 1243 * Handle a bvec that must be split between two or more targets. 1244 */ 1245 struct bio_vec *bv = bio->bi_io_vec + ci->idx; 1246 sector_t remaining = to_sector(bv->bv_len); 1247 unsigned int offset = 0; 1248 1249 do { 1250 if (offset) { 1251 ti = dm_table_find_target(ci->map, ci->sector); 1252 if (!dm_target_is_valid(ti)) 1253 return -EIO; 1254 1255 max = max_io_len(ci->sector, ti); 1256 } 1257 1258 len = min(remaining, max); 1259 1260 tio = alloc_tio(ci, ti); 1261 clone = split_bvec(bio, ci->sector, ci->idx, 1262 bv->bv_offset + offset, len, 1263 ci->md->bs); 1264 1265 __map_bio(ti, clone, tio); 1266 1267 ci->sector += len; 1268 ci->sector_count -= len; 1269 offset += to_bytes(len); 1270 } while (remaining -= len); 1271 1272 ci->idx++; 1273 } 1274 1275 return 0; 1276 } 1277 1278 /* 1279 * Split the bio into several clones and submit it to targets. 1280 */ 1281 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) 1282 { 1283 struct clone_info ci; 1284 int error = 0; 1285 1286 ci.map = dm_get_live_table(md); 1287 if (unlikely(!ci.map)) { 1288 bio_io_error(bio); 1289 return; 1290 } 1291 1292 ci.md = md; 1293 ci.io = alloc_io(md); 1294 ci.io->error = 0; 1295 atomic_set(&ci.io->io_count, 1); 1296 ci.io->bio = bio; 1297 ci.io->md = md; 1298 spin_lock_init(&ci.io->endio_lock); 1299 ci.sector = bio->bi_sector; 1300 ci.idx = bio->bi_idx; 1301 1302 start_io_acct(ci.io); 1303 if (bio->bi_rw & REQ_FLUSH) { 1304 ci.bio = &ci.md->flush_bio; 1305 ci.sector_count = 0; 1306 error = __clone_and_map_empty_flush(&ci); 1307 /* dec_pending submits any data associated with flush */ 1308 } else { 1309 ci.bio = bio; 1310 ci.sector_count = bio_sectors(bio); 1311 while (ci.sector_count && !error) 1312 error = __clone_and_map(&ci); 1313 } 1314 1315 /* drop the extra reference count */ 1316 dec_pending(ci.io, error); 1317 dm_table_put(ci.map); 1318 } 1319 /*----------------------------------------------------------------- 1320 * CRUD END 1321 *---------------------------------------------------------------*/ 1322 1323 static int dm_merge_bvec(struct request_queue *q, 1324 struct bvec_merge_data *bvm, 1325 struct bio_vec *biovec) 1326 { 1327 struct mapped_device *md = q->queuedata; 1328 struct dm_table *map = dm_get_live_table(md); 1329 struct dm_target *ti; 1330 sector_t max_sectors; 1331 int max_size = 0; 1332 1333 if (unlikely(!map)) 1334 goto out; 1335 1336 ti = dm_table_find_target(map, bvm->bi_sector); 1337 if (!dm_target_is_valid(ti)) 1338 goto out_table; 1339 1340 /* 1341 * Find maximum amount of I/O that won't need splitting 1342 */ 1343 max_sectors = min(max_io_len(bvm->bi_sector, ti), 1344 (sector_t) BIO_MAX_SECTORS); 1345 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; 1346 if (max_size < 0) 1347 max_size = 0; 1348 1349 /* 1350 * merge_bvec_fn() returns number of bytes 1351 * it can accept at this offset 1352 * max is precomputed maximal io size 1353 */ 1354 if (max_size && ti->type->merge) 1355 max_size = ti->type->merge(ti, bvm, biovec, max_size); 1356 /* 1357 * If the target doesn't support merge method and some of the devices 1358 * provided their merge_bvec method (we know this by looking at 1359 * queue_max_hw_sectors), then we can't allow bios with multiple vector 1360 * entries. So always set max_size to 0, and the code below allows 1361 * just one page. 1362 */ 1363 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) 1364 1365 max_size = 0; 1366 1367 out_table: 1368 dm_table_put(map); 1369 1370 out: 1371 /* 1372 * Always allow an entire first page 1373 */ 1374 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) 1375 max_size = biovec->bv_len; 1376 1377 return max_size; 1378 } 1379 1380 /* 1381 * The request function that just remaps the bio built up by 1382 * dm_merge_bvec. 1383 */ 1384 static int _dm_request(struct request_queue *q, struct bio *bio) 1385 { 1386 int rw = bio_data_dir(bio); 1387 struct mapped_device *md = q->queuedata; 1388 int cpu; 1389 1390 down_read(&md->io_lock); 1391 1392 cpu = part_stat_lock(); 1393 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); 1394 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); 1395 part_stat_unlock(); 1396 1397 /* if we're suspended, we have to queue this io for later */ 1398 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { 1399 up_read(&md->io_lock); 1400 1401 if (bio_rw(bio) != READA) 1402 queue_io(md, bio); 1403 else 1404 bio_io_error(bio); 1405 return 0; 1406 } 1407 1408 __split_and_process_bio(md, bio); 1409 up_read(&md->io_lock); 1410 return 0; 1411 } 1412 1413 static int dm_make_request(struct request_queue *q, struct bio *bio) 1414 { 1415 struct mapped_device *md = q->queuedata; 1416 1417 return md->saved_make_request_fn(q, bio); /* call __make_request() */ 1418 } 1419 1420 static int dm_request_based(struct mapped_device *md) 1421 { 1422 return blk_queue_stackable(md->queue); 1423 } 1424 1425 static int dm_request(struct request_queue *q, struct bio *bio) 1426 { 1427 struct mapped_device *md = q->queuedata; 1428 1429 if (dm_request_based(md)) 1430 return dm_make_request(q, bio); 1431 1432 return _dm_request(q, bio); 1433 } 1434 1435 void dm_dispatch_request(struct request *rq) 1436 { 1437 int r; 1438 1439 if (blk_queue_io_stat(rq->q)) 1440 rq->cmd_flags |= REQ_IO_STAT; 1441 1442 rq->start_time = jiffies; 1443 r = blk_insert_cloned_request(rq->q, rq); 1444 if (r) 1445 dm_complete_request(rq, r); 1446 } 1447 EXPORT_SYMBOL_GPL(dm_dispatch_request); 1448 1449 static void dm_rq_bio_destructor(struct bio *bio) 1450 { 1451 struct dm_rq_clone_bio_info *info = bio->bi_private; 1452 struct mapped_device *md = info->tio->md; 1453 1454 free_bio_info(info); 1455 bio_free(bio, md->bs); 1456 } 1457 1458 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, 1459 void *data) 1460 { 1461 struct dm_rq_target_io *tio = data; 1462 struct mapped_device *md = tio->md; 1463 struct dm_rq_clone_bio_info *info = alloc_bio_info(md); 1464 1465 if (!info) 1466 return -ENOMEM; 1467 1468 info->orig = bio_orig; 1469 info->tio = tio; 1470 bio->bi_end_io = end_clone_bio; 1471 bio->bi_private = info; 1472 bio->bi_destructor = dm_rq_bio_destructor; 1473 1474 return 0; 1475 } 1476 1477 static int setup_clone(struct request *clone, struct request *rq, 1478 struct dm_rq_target_io *tio) 1479 { 1480 int r; 1481 1482 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, 1483 dm_rq_bio_constructor, tio); 1484 if (r) 1485 return r; 1486 1487 clone->cmd = rq->cmd; 1488 clone->cmd_len = rq->cmd_len; 1489 clone->sense = rq->sense; 1490 clone->buffer = rq->buffer; 1491 clone->end_io = end_clone_request; 1492 clone->end_io_data = tio; 1493 1494 return 0; 1495 } 1496 1497 static struct request *clone_rq(struct request *rq, struct mapped_device *md, 1498 gfp_t gfp_mask) 1499 { 1500 struct request *clone; 1501 struct dm_rq_target_io *tio; 1502 1503 tio = alloc_rq_tio(md, gfp_mask); 1504 if (!tio) 1505 return NULL; 1506 1507 tio->md = md; 1508 tio->ti = NULL; 1509 tio->orig = rq; 1510 tio->error = 0; 1511 memset(&tio->info, 0, sizeof(tio->info)); 1512 1513 clone = &tio->clone; 1514 if (setup_clone(clone, rq, tio)) { 1515 /* -ENOMEM */ 1516 free_rq_tio(tio); 1517 return NULL; 1518 } 1519 1520 return clone; 1521 } 1522 1523 /* 1524 * Called with the queue lock held. 1525 */ 1526 static int dm_prep_fn(struct request_queue *q, struct request *rq) 1527 { 1528 struct mapped_device *md = q->queuedata; 1529 struct request *clone; 1530 1531 if (unlikely(rq->special)) { 1532 DMWARN("Already has something in rq->special."); 1533 return BLKPREP_KILL; 1534 } 1535 1536 clone = clone_rq(rq, md, GFP_ATOMIC); 1537 if (!clone) 1538 return BLKPREP_DEFER; 1539 1540 rq->special = clone; 1541 rq->cmd_flags |= REQ_DONTPREP; 1542 1543 return BLKPREP_OK; 1544 } 1545 1546 /* 1547 * Returns: 1548 * 0 : the request has been processed (not requeued) 1549 * !0 : the request has been requeued 1550 */ 1551 static int map_request(struct dm_target *ti, struct request *clone, 1552 struct mapped_device *md) 1553 { 1554 int r, requeued = 0; 1555 struct dm_rq_target_io *tio = clone->end_io_data; 1556 1557 /* 1558 * Hold the md reference here for the in-flight I/O. 1559 * We can't rely on the reference count by device opener, 1560 * because the device may be closed during the request completion 1561 * when all bios are completed. 1562 * See the comment in rq_completed() too. 1563 */ 1564 dm_get(md); 1565 1566 tio->ti = ti; 1567 r = ti->type->map_rq(ti, clone, &tio->info); 1568 switch (r) { 1569 case DM_MAPIO_SUBMITTED: 1570 /* The target has taken the I/O to submit by itself later */ 1571 break; 1572 case DM_MAPIO_REMAPPED: 1573 /* The target has remapped the I/O so dispatch it */ 1574 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), 1575 blk_rq_pos(tio->orig)); 1576 dm_dispatch_request(clone); 1577 break; 1578 case DM_MAPIO_REQUEUE: 1579 /* The target wants to requeue the I/O */ 1580 dm_requeue_unmapped_request(clone); 1581 requeued = 1; 1582 break; 1583 default: 1584 if (r > 0) { 1585 DMWARN("unimplemented target map return value: %d", r); 1586 BUG(); 1587 } 1588 1589 /* The target wants to complete the I/O */ 1590 dm_kill_unmapped_request(clone, r); 1591 break; 1592 } 1593 1594 return requeued; 1595 } 1596 1597 /* 1598 * q->request_fn for request-based dm. 1599 * Called with the queue lock held. 1600 */ 1601 static void dm_request_fn(struct request_queue *q) 1602 { 1603 struct mapped_device *md = q->queuedata; 1604 struct dm_table *map = dm_get_live_table(md); 1605 struct dm_target *ti; 1606 struct request *rq, *clone; 1607 sector_t pos; 1608 1609 /* 1610 * For suspend, check blk_queue_stopped() and increment 1611 * ->pending within a single queue_lock not to increment the 1612 * number of in-flight I/Os after the queue is stopped in 1613 * dm_suspend(). 1614 */ 1615 while (!blk_queue_stopped(q)) { 1616 rq = blk_peek_request(q); 1617 if (!rq) 1618 goto delay_and_out; 1619 1620 /* always use block 0 to find the target for flushes for now */ 1621 pos = 0; 1622 if (!(rq->cmd_flags & REQ_FLUSH)) 1623 pos = blk_rq_pos(rq); 1624 1625 ti = dm_table_find_target(map, pos); 1626 BUG_ON(!dm_target_is_valid(ti)); 1627 1628 if (ti->type->busy && ti->type->busy(ti)) 1629 goto delay_and_out; 1630 1631 blk_start_request(rq); 1632 clone = rq->special; 1633 atomic_inc(&md->pending[rq_data_dir(clone)]); 1634 1635 spin_unlock(q->queue_lock); 1636 if (map_request(ti, clone, md)) 1637 goto requeued; 1638 1639 BUG_ON(!irqs_disabled()); 1640 spin_lock(q->queue_lock); 1641 } 1642 1643 goto out; 1644 1645 requeued: 1646 BUG_ON(!irqs_disabled()); 1647 spin_lock(q->queue_lock); 1648 1649 delay_and_out: 1650 blk_delay_queue(q, HZ / 10); 1651 out: 1652 dm_table_put(map); 1653 1654 return; 1655 } 1656 1657 int dm_underlying_device_busy(struct request_queue *q) 1658 { 1659 return blk_lld_busy(q); 1660 } 1661 EXPORT_SYMBOL_GPL(dm_underlying_device_busy); 1662 1663 static int dm_lld_busy(struct request_queue *q) 1664 { 1665 int r; 1666 struct mapped_device *md = q->queuedata; 1667 struct dm_table *map = dm_get_live_table(md); 1668 1669 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) 1670 r = 1; 1671 else 1672 r = dm_table_any_busy_target(map); 1673 1674 dm_table_put(map); 1675 1676 return r; 1677 } 1678 1679 static int dm_any_congested(void *congested_data, int bdi_bits) 1680 { 1681 int r = bdi_bits; 1682 struct mapped_device *md = congested_data; 1683 struct dm_table *map; 1684 1685 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 1686 map = dm_get_live_table(md); 1687 if (map) { 1688 /* 1689 * Request-based dm cares about only own queue for 1690 * the query about congestion status of request_queue 1691 */ 1692 if (dm_request_based(md)) 1693 r = md->queue->backing_dev_info.state & 1694 bdi_bits; 1695 else 1696 r = dm_table_any_congested(map, bdi_bits); 1697 1698 dm_table_put(map); 1699 } 1700 } 1701 1702 return r; 1703 } 1704 1705 /*----------------------------------------------------------------- 1706 * An IDR is used to keep track of allocated minor numbers. 1707 *---------------------------------------------------------------*/ 1708 static DEFINE_IDR(_minor_idr); 1709 1710 static void free_minor(int minor) 1711 { 1712 spin_lock(&_minor_lock); 1713 idr_remove(&_minor_idr, minor); 1714 spin_unlock(&_minor_lock); 1715 } 1716 1717 /* 1718 * See if the device with a specific minor # is free. 1719 */ 1720 static int specific_minor(int minor) 1721 { 1722 int r, m; 1723 1724 if (minor >= (1 << MINORBITS)) 1725 return -EINVAL; 1726 1727 r = idr_pre_get(&_minor_idr, GFP_KERNEL); 1728 if (!r) 1729 return -ENOMEM; 1730 1731 spin_lock(&_minor_lock); 1732 1733 if (idr_find(&_minor_idr, minor)) { 1734 r = -EBUSY; 1735 goto out; 1736 } 1737 1738 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); 1739 if (r) 1740 goto out; 1741 1742 if (m != minor) { 1743 idr_remove(&_minor_idr, m); 1744 r = -EBUSY; 1745 goto out; 1746 } 1747 1748 out: 1749 spin_unlock(&_minor_lock); 1750 return r; 1751 } 1752 1753 static int next_free_minor(int *minor) 1754 { 1755 int r, m; 1756 1757 r = idr_pre_get(&_minor_idr, GFP_KERNEL); 1758 if (!r) 1759 return -ENOMEM; 1760 1761 spin_lock(&_minor_lock); 1762 1763 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); 1764 if (r) 1765 goto out; 1766 1767 if (m >= (1 << MINORBITS)) { 1768 idr_remove(&_minor_idr, m); 1769 r = -ENOSPC; 1770 goto out; 1771 } 1772 1773 *minor = m; 1774 1775 out: 1776 spin_unlock(&_minor_lock); 1777 return r; 1778 } 1779 1780 static const struct block_device_operations dm_blk_dops; 1781 1782 static void dm_wq_work(struct work_struct *work); 1783 1784 static void dm_init_md_queue(struct mapped_device *md) 1785 { 1786 /* 1787 * Request-based dm devices cannot be stacked on top of bio-based dm 1788 * devices. The type of this dm device has not been decided yet. 1789 * The type is decided at the first table loading time. 1790 * To prevent problematic device stacking, clear the queue flag 1791 * for request stacking support until then. 1792 * 1793 * This queue is new, so no concurrency on the queue_flags. 1794 */ 1795 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); 1796 1797 md->queue->queuedata = md; 1798 md->queue->backing_dev_info.congested_fn = dm_any_congested; 1799 md->queue->backing_dev_info.congested_data = md; 1800 blk_queue_make_request(md->queue, dm_request); 1801 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); 1802 blk_queue_merge_bvec(md->queue, dm_merge_bvec); 1803 blk_queue_flush(md->queue, REQ_FLUSH | REQ_FUA); 1804 } 1805 1806 /* 1807 * Allocate and initialise a blank device with a given minor. 1808 */ 1809 static struct mapped_device *alloc_dev(int minor) 1810 { 1811 int r; 1812 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); 1813 void *old_md; 1814 1815 if (!md) { 1816 DMWARN("unable to allocate device, out of memory."); 1817 return NULL; 1818 } 1819 1820 if (!try_module_get(THIS_MODULE)) 1821 goto bad_module_get; 1822 1823 /* get a minor number for the dev */ 1824 if (minor == DM_ANY_MINOR) 1825 r = next_free_minor(&minor); 1826 else 1827 r = specific_minor(minor); 1828 if (r < 0) 1829 goto bad_minor; 1830 1831 md->type = DM_TYPE_NONE; 1832 init_rwsem(&md->io_lock); 1833 mutex_init(&md->suspend_lock); 1834 mutex_init(&md->type_lock); 1835 spin_lock_init(&md->deferred_lock); 1836 rwlock_init(&md->map_lock); 1837 atomic_set(&md->holders, 1); 1838 atomic_set(&md->open_count, 0); 1839 atomic_set(&md->event_nr, 0); 1840 atomic_set(&md->uevent_seq, 0); 1841 INIT_LIST_HEAD(&md->uevent_list); 1842 spin_lock_init(&md->uevent_lock); 1843 1844 md->queue = blk_alloc_queue(GFP_KERNEL); 1845 if (!md->queue) 1846 goto bad_queue; 1847 1848 dm_init_md_queue(md); 1849 1850 md->disk = alloc_disk(1); 1851 if (!md->disk) 1852 goto bad_disk; 1853 1854 atomic_set(&md->pending[0], 0); 1855 atomic_set(&md->pending[1], 0); 1856 init_waitqueue_head(&md->wait); 1857 INIT_WORK(&md->work, dm_wq_work); 1858 init_waitqueue_head(&md->eventq); 1859 1860 md->disk->major = _major; 1861 md->disk->first_minor = minor; 1862 md->disk->fops = &dm_blk_dops; 1863 md->disk->queue = md->queue; 1864 md->disk->private_data = md; 1865 sprintf(md->disk->disk_name, "dm-%d", minor); 1866 add_disk(md->disk); 1867 format_dev_t(md->name, MKDEV(_major, minor)); 1868 1869 md->wq = alloc_workqueue("kdmflush", 1870 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); 1871 if (!md->wq) 1872 goto bad_thread; 1873 1874 md->bdev = bdget_disk(md->disk, 0); 1875 if (!md->bdev) 1876 goto bad_bdev; 1877 1878 bio_init(&md->flush_bio); 1879 md->flush_bio.bi_bdev = md->bdev; 1880 md->flush_bio.bi_rw = WRITE_FLUSH; 1881 1882 /* Populate the mapping, nobody knows we exist yet */ 1883 spin_lock(&_minor_lock); 1884 old_md = idr_replace(&_minor_idr, md, minor); 1885 spin_unlock(&_minor_lock); 1886 1887 BUG_ON(old_md != MINOR_ALLOCED); 1888 1889 return md; 1890 1891 bad_bdev: 1892 destroy_workqueue(md->wq); 1893 bad_thread: 1894 del_gendisk(md->disk); 1895 put_disk(md->disk); 1896 bad_disk: 1897 blk_cleanup_queue(md->queue); 1898 bad_queue: 1899 free_minor(minor); 1900 bad_minor: 1901 module_put(THIS_MODULE); 1902 bad_module_get: 1903 kfree(md); 1904 return NULL; 1905 } 1906 1907 static void unlock_fs(struct mapped_device *md); 1908 1909 static void free_dev(struct mapped_device *md) 1910 { 1911 int minor = MINOR(disk_devt(md->disk)); 1912 1913 unlock_fs(md); 1914 bdput(md->bdev); 1915 destroy_workqueue(md->wq); 1916 if (md->tio_pool) 1917 mempool_destroy(md->tio_pool); 1918 if (md->io_pool) 1919 mempool_destroy(md->io_pool); 1920 if (md->bs) 1921 bioset_free(md->bs); 1922 blk_integrity_unregister(md->disk); 1923 del_gendisk(md->disk); 1924 free_minor(minor); 1925 1926 spin_lock(&_minor_lock); 1927 md->disk->private_data = NULL; 1928 spin_unlock(&_minor_lock); 1929 1930 put_disk(md->disk); 1931 blk_cleanup_queue(md->queue); 1932 module_put(THIS_MODULE); 1933 kfree(md); 1934 } 1935 1936 static void __bind_mempools(struct mapped_device *md, struct dm_table *t) 1937 { 1938 struct dm_md_mempools *p; 1939 1940 if (md->io_pool && md->tio_pool && md->bs) 1941 /* the md already has necessary mempools */ 1942 goto out; 1943 1944 p = dm_table_get_md_mempools(t); 1945 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); 1946 1947 md->io_pool = p->io_pool; 1948 p->io_pool = NULL; 1949 md->tio_pool = p->tio_pool; 1950 p->tio_pool = NULL; 1951 md->bs = p->bs; 1952 p->bs = NULL; 1953 1954 out: 1955 /* mempool bind completed, now no need any mempools in the table */ 1956 dm_table_free_md_mempools(t); 1957 } 1958 1959 /* 1960 * Bind a table to the device. 1961 */ 1962 static void event_callback(void *context) 1963 { 1964 unsigned long flags; 1965 LIST_HEAD(uevents); 1966 struct mapped_device *md = (struct mapped_device *) context; 1967 1968 spin_lock_irqsave(&md->uevent_lock, flags); 1969 list_splice_init(&md->uevent_list, &uevents); 1970 spin_unlock_irqrestore(&md->uevent_lock, flags); 1971 1972 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); 1973 1974 atomic_inc(&md->event_nr); 1975 wake_up(&md->eventq); 1976 } 1977 1978 /* 1979 * Protected by md->suspend_lock obtained by dm_swap_table(). 1980 */ 1981 static void __set_size(struct mapped_device *md, sector_t size) 1982 { 1983 set_capacity(md->disk, size); 1984 1985 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); 1986 } 1987 1988 /* 1989 * Returns old map, which caller must destroy. 1990 */ 1991 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, 1992 struct queue_limits *limits) 1993 { 1994 struct dm_table *old_map; 1995 struct request_queue *q = md->queue; 1996 sector_t size; 1997 unsigned long flags; 1998 1999 size = dm_table_get_size(t); 2000 2001 /* 2002 * Wipe any geometry if the size of the table changed. 2003 */ 2004 if (size != get_capacity(md->disk)) 2005 memset(&md->geometry, 0, sizeof(md->geometry)); 2006 2007 __set_size(md, size); 2008 2009 dm_table_event_callback(t, event_callback, md); 2010 2011 /* 2012 * The queue hasn't been stopped yet, if the old table type wasn't 2013 * for request-based during suspension. So stop it to prevent 2014 * I/O mapping before resume. 2015 * This must be done before setting the queue restrictions, 2016 * because request-based dm may be run just after the setting. 2017 */ 2018 if (dm_table_request_based(t) && !blk_queue_stopped(q)) 2019 stop_queue(q); 2020 2021 __bind_mempools(md, t); 2022 2023 write_lock_irqsave(&md->map_lock, flags); 2024 old_map = md->map; 2025 md->map = t; 2026 dm_table_set_restrictions(t, q, limits); 2027 write_unlock_irqrestore(&md->map_lock, flags); 2028 2029 return old_map; 2030 } 2031 2032 /* 2033 * Returns unbound table for the caller to free. 2034 */ 2035 static struct dm_table *__unbind(struct mapped_device *md) 2036 { 2037 struct dm_table *map = md->map; 2038 unsigned long flags; 2039 2040 if (!map) 2041 return NULL; 2042 2043 dm_table_event_callback(map, NULL, NULL); 2044 write_lock_irqsave(&md->map_lock, flags); 2045 md->map = NULL; 2046 write_unlock_irqrestore(&md->map_lock, flags); 2047 2048 return map; 2049 } 2050 2051 /* 2052 * Constructor for a new device. 2053 */ 2054 int dm_create(int minor, struct mapped_device **result) 2055 { 2056 struct mapped_device *md; 2057 2058 md = alloc_dev(minor); 2059 if (!md) 2060 return -ENXIO; 2061 2062 dm_sysfs_init(md); 2063 2064 *result = md; 2065 return 0; 2066 } 2067 2068 /* 2069 * Functions to manage md->type. 2070 * All are required to hold md->type_lock. 2071 */ 2072 void dm_lock_md_type(struct mapped_device *md) 2073 { 2074 mutex_lock(&md->type_lock); 2075 } 2076 2077 void dm_unlock_md_type(struct mapped_device *md) 2078 { 2079 mutex_unlock(&md->type_lock); 2080 } 2081 2082 void dm_set_md_type(struct mapped_device *md, unsigned type) 2083 { 2084 md->type = type; 2085 } 2086 2087 unsigned dm_get_md_type(struct mapped_device *md) 2088 { 2089 return md->type; 2090 } 2091 2092 /* 2093 * Fully initialize a request-based queue (->elevator, ->request_fn, etc). 2094 */ 2095 static int dm_init_request_based_queue(struct mapped_device *md) 2096 { 2097 struct request_queue *q = NULL; 2098 2099 if (md->queue->elevator) 2100 return 1; 2101 2102 /* Fully initialize the queue */ 2103 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); 2104 if (!q) 2105 return 0; 2106 2107 md->queue = q; 2108 md->saved_make_request_fn = md->queue->make_request_fn; 2109 dm_init_md_queue(md); 2110 blk_queue_softirq_done(md->queue, dm_softirq_done); 2111 blk_queue_prep_rq(md->queue, dm_prep_fn); 2112 blk_queue_lld_busy(md->queue, dm_lld_busy); 2113 2114 elv_register_queue(md->queue); 2115 2116 return 1; 2117 } 2118 2119 /* 2120 * Setup the DM device's queue based on md's type 2121 */ 2122 int dm_setup_md_queue(struct mapped_device *md) 2123 { 2124 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && 2125 !dm_init_request_based_queue(md)) { 2126 DMWARN("Cannot initialize queue for request-based mapped device"); 2127 return -EINVAL; 2128 } 2129 2130 return 0; 2131 } 2132 2133 static struct mapped_device *dm_find_md(dev_t dev) 2134 { 2135 struct mapped_device *md; 2136 unsigned minor = MINOR(dev); 2137 2138 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) 2139 return NULL; 2140 2141 spin_lock(&_minor_lock); 2142 2143 md = idr_find(&_minor_idr, minor); 2144 if (md && (md == MINOR_ALLOCED || 2145 (MINOR(disk_devt(dm_disk(md))) != minor) || 2146 dm_deleting_md(md) || 2147 test_bit(DMF_FREEING, &md->flags))) { 2148 md = NULL; 2149 goto out; 2150 } 2151 2152 out: 2153 spin_unlock(&_minor_lock); 2154 2155 return md; 2156 } 2157 2158 struct mapped_device *dm_get_md(dev_t dev) 2159 { 2160 struct mapped_device *md = dm_find_md(dev); 2161 2162 if (md) 2163 dm_get(md); 2164 2165 return md; 2166 } 2167 2168 void *dm_get_mdptr(struct mapped_device *md) 2169 { 2170 return md->interface_ptr; 2171 } 2172 2173 void dm_set_mdptr(struct mapped_device *md, void *ptr) 2174 { 2175 md->interface_ptr = ptr; 2176 } 2177 2178 void dm_get(struct mapped_device *md) 2179 { 2180 atomic_inc(&md->holders); 2181 BUG_ON(test_bit(DMF_FREEING, &md->flags)); 2182 } 2183 2184 const char *dm_device_name(struct mapped_device *md) 2185 { 2186 return md->name; 2187 } 2188 EXPORT_SYMBOL_GPL(dm_device_name); 2189 2190 static void __dm_destroy(struct mapped_device *md, bool wait) 2191 { 2192 struct dm_table *map; 2193 2194 might_sleep(); 2195 2196 spin_lock(&_minor_lock); 2197 map = dm_get_live_table(md); 2198 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); 2199 set_bit(DMF_FREEING, &md->flags); 2200 spin_unlock(&_minor_lock); 2201 2202 if (!dm_suspended_md(md)) { 2203 dm_table_presuspend_targets(map); 2204 dm_table_postsuspend_targets(map); 2205 } 2206 2207 /* 2208 * Rare, but there may be I/O requests still going to complete, 2209 * for example. Wait for all references to disappear. 2210 * No one should increment the reference count of the mapped_device, 2211 * after the mapped_device state becomes DMF_FREEING. 2212 */ 2213 if (wait) 2214 while (atomic_read(&md->holders)) 2215 msleep(1); 2216 else if (atomic_read(&md->holders)) 2217 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", 2218 dm_device_name(md), atomic_read(&md->holders)); 2219 2220 dm_sysfs_exit(md); 2221 dm_table_put(map); 2222 dm_table_destroy(__unbind(md)); 2223 free_dev(md); 2224 } 2225 2226 void dm_destroy(struct mapped_device *md) 2227 { 2228 __dm_destroy(md, true); 2229 } 2230 2231 void dm_destroy_immediate(struct mapped_device *md) 2232 { 2233 __dm_destroy(md, false); 2234 } 2235 2236 void dm_put(struct mapped_device *md) 2237 { 2238 atomic_dec(&md->holders); 2239 } 2240 EXPORT_SYMBOL_GPL(dm_put); 2241 2242 static int dm_wait_for_completion(struct mapped_device *md, int interruptible) 2243 { 2244 int r = 0; 2245 DECLARE_WAITQUEUE(wait, current); 2246 2247 add_wait_queue(&md->wait, &wait); 2248 2249 while (1) { 2250 set_current_state(interruptible); 2251 2252 smp_mb(); 2253 if (!md_in_flight(md)) 2254 break; 2255 2256 if (interruptible == TASK_INTERRUPTIBLE && 2257 signal_pending(current)) { 2258 r = -EINTR; 2259 break; 2260 } 2261 2262 io_schedule(); 2263 } 2264 set_current_state(TASK_RUNNING); 2265 2266 remove_wait_queue(&md->wait, &wait); 2267 2268 return r; 2269 } 2270 2271 /* 2272 * Process the deferred bios 2273 */ 2274 static void dm_wq_work(struct work_struct *work) 2275 { 2276 struct mapped_device *md = container_of(work, struct mapped_device, 2277 work); 2278 struct bio *c; 2279 2280 down_read(&md->io_lock); 2281 2282 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { 2283 spin_lock_irq(&md->deferred_lock); 2284 c = bio_list_pop(&md->deferred); 2285 spin_unlock_irq(&md->deferred_lock); 2286 2287 if (!c) 2288 break; 2289 2290 up_read(&md->io_lock); 2291 2292 if (dm_request_based(md)) 2293 generic_make_request(c); 2294 else 2295 __split_and_process_bio(md, c); 2296 2297 down_read(&md->io_lock); 2298 } 2299 2300 up_read(&md->io_lock); 2301 } 2302 2303 static void dm_queue_flush(struct mapped_device *md) 2304 { 2305 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2306 smp_mb__after_clear_bit(); 2307 queue_work(md->wq, &md->work); 2308 } 2309 2310 /* 2311 * Swap in a new table, returning the old one for the caller to destroy. 2312 */ 2313 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) 2314 { 2315 struct dm_table *map = ERR_PTR(-EINVAL); 2316 struct queue_limits limits; 2317 int r; 2318 2319 mutex_lock(&md->suspend_lock); 2320 2321 /* device must be suspended */ 2322 if (!dm_suspended_md(md)) 2323 goto out; 2324 2325 r = dm_calculate_queue_limits(table, &limits); 2326 if (r) { 2327 map = ERR_PTR(r); 2328 goto out; 2329 } 2330 2331 map = __bind(md, table, &limits); 2332 2333 out: 2334 mutex_unlock(&md->suspend_lock); 2335 return map; 2336 } 2337 2338 /* 2339 * Functions to lock and unlock any filesystem running on the 2340 * device. 2341 */ 2342 static int lock_fs(struct mapped_device *md) 2343 { 2344 int r; 2345 2346 WARN_ON(md->frozen_sb); 2347 2348 md->frozen_sb = freeze_bdev(md->bdev); 2349 if (IS_ERR(md->frozen_sb)) { 2350 r = PTR_ERR(md->frozen_sb); 2351 md->frozen_sb = NULL; 2352 return r; 2353 } 2354 2355 set_bit(DMF_FROZEN, &md->flags); 2356 2357 return 0; 2358 } 2359 2360 static void unlock_fs(struct mapped_device *md) 2361 { 2362 if (!test_bit(DMF_FROZEN, &md->flags)) 2363 return; 2364 2365 thaw_bdev(md->bdev, md->frozen_sb); 2366 md->frozen_sb = NULL; 2367 clear_bit(DMF_FROZEN, &md->flags); 2368 } 2369 2370 /* 2371 * We need to be able to change a mapping table under a mounted 2372 * filesystem. For example we might want to move some data in 2373 * the background. Before the table can be swapped with 2374 * dm_bind_table, dm_suspend must be called to flush any in 2375 * flight bios and ensure that any further io gets deferred. 2376 */ 2377 /* 2378 * Suspend mechanism in request-based dm. 2379 * 2380 * 1. Flush all I/Os by lock_fs() if needed. 2381 * 2. Stop dispatching any I/O by stopping the request_queue. 2382 * 3. Wait for all in-flight I/Os to be completed or requeued. 2383 * 2384 * To abort suspend, start the request_queue. 2385 */ 2386 int dm_suspend(struct mapped_device *md, unsigned suspend_flags) 2387 { 2388 struct dm_table *map = NULL; 2389 int r = 0; 2390 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; 2391 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; 2392 2393 mutex_lock(&md->suspend_lock); 2394 2395 if (dm_suspended_md(md)) { 2396 r = -EINVAL; 2397 goto out_unlock; 2398 } 2399 2400 map = dm_get_live_table(md); 2401 2402 /* 2403 * DMF_NOFLUSH_SUSPENDING must be set before presuspend. 2404 * This flag is cleared before dm_suspend returns. 2405 */ 2406 if (noflush) 2407 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 2408 2409 /* This does not get reverted if there's an error later. */ 2410 dm_table_presuspend_targets(map); 2411 2412 /* 2413 * Flush I/O to the device. 2414 * Any I/O submitted after lock_fs() may not be flushed. 2415 * noflush takes precedence over do_lockfs. 2416 * (lock_fs() flushes I/Os and waits for them to complete.) 2417 */ 2418 if (!noflush && do_lockfs) { 2419 r = lock_fs(md); 2420 if (r) 2421 goto out; 2422 } 2423 2424 /* 2425 * Here we must make sure that no processes are submitting requests 2426 * to target drivers i.e. no one may be executing 2427 * __split_and_process_bio. This is called from dm_request and 2428 * dm_wq_work. 2429 * 2430 * To get all processes out of __split_and_process_bio in dm_request, 2431 * we take the write lock. To prevent any process from reentering 2432 * __split_and_process_bio from dm_request and quiesce the thread 2433 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call 2434 * flush_workqueue(md->wq). 2435 */ 2436 down_write(&md->io_lock); 2437 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); 2438 up_write(&md->io_lock); 2439 2440 /* 2441 * Stop md->queue before flushing md->wq in case request-based 2442 * dm defers requests to md->wq from md->queue. 2443 */ 2444 if (dm_request_based(md)) 2445 stop_queue(md->queue); 2446 2447 flush_workqueue(md->wq); 2448 2449 /* 2450 * At this point no more requests are entering target request routines. 2451 * We call dm_wait_for_completion to wait for all existing requests 2452 * to finish. 2453 */ 2454 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); 2455 2456 down_write(&md->io_lock); 2457 if (noflush) 2458 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); 2459 up_write(&md->io_lock); 2460 2461 /* were we interrupted ? */ 2462 if (r < 0) { 2463 dm_queue_flush(md); 2464 2465 if (dm_request_based(md)) 2466 start_queue(md->queue); 2467 2468 unlock_fs(md); 2469 goto out; /* pushback list is already flushed, so skip flush */ 2470 } 2471 2472 /* 2473 * If dm_wait_for_completion returned 0, the device is completely 2474 * quiescent now. There is no request-processing activity. All new 2475 * requests are being added to md->deferred list. 2476 */ 2477 2478 set_bit(DMF_SUSPENDED, &md->flags); 2479 2480 dm_table_postsuspend_targets(map); 2481 2482 out: 2483 dm_table_put(map); 2484 2485 out_unlock: 2486 mutex_unlock(&md->suspend_lock); 2487 return r; 2488 } 2489 2490 int dm_resume(struct mapped_device *md) 2491 { 2492 int r = -EINVAL; 2493 struct dm_table *map = NULL; 2494 2495 mutex_lock(&md->suspend_lock); 2496 if (!dm_suspended_md(md)) 2497 goto out; 2498 2499 map = dm_get_live_table(md); 2500 if (!map || !dm_table_get_size(map)) 2501 goto out; 2502 2503 r = dm_table_resume_targets(map); 2504 if (r) 2505 goto out; 2506 2507 dm_queue_flush(md); 2508 2509 /* 2510 * Flushing deferred I/Os must be done after targets are resumed 2511 * so that mapping of targets can work correctly. 2512 * Request-based dm is queueing the deferred I/Os in its request_queue. 2513 */ 2514 if (dm_request_based(md)) 2515 start_queue(md->queue); 2516 2517 unlock_fs(md); 2518 2519 clear_bit(DMF_SUSPENDED, &md->flags); 2520 2521 r = 0; 2522 out: 2523 dm_table_put(map); 2524 mutex_unlock(&md->suspend_lock); 2525 2526 return r; 2527 } 2528 2529 /*----------------------------------------------------------------- 2530 * Event notification. 2531 *---------------------------------------------------------------*/ 2532 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, 2533 unsigned cookie) 2534 { 2535 char udev_cookie[DM_COOKIE_LENGTH]; 2536 char *envp[] = { udev_cookie, NULL }; 2537 2538 if (!cookie) 2539 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); 2540 else { 2541 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", 2542 DM_COOKIE_ENV_VAR_NAME, cookie); 2543 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, 2544 action, envp); 2545 } 2546 } 2547 2548 uint32_t dm_next_uevent_seq(struct mapped_device *md) 2549 { 2550 return atomic_add_return(1, &md->uevent_seq); 2551 } 2552 2553 uint32_t dm_get_event_nr(struct mapped_device *md) 2554 { 2555 return atomic_read(&md->event_nr); 2556 } 2557 2558 int dm_wait_event(struct mapped_device *md, int event_nr) 2559 { 2560 return wait_event_interruptible(md->eventq, 2561 (event_nr != atomic_read(&md->event_nr))); 2562 } 2563 2564 void dm_uevent_add(struct mapped_device *md, struct list_head *elist) 2565 { 2566 unsigned long flags; 2567 2568 spin_lock_irqsave(&md->uevent_lock, flags); 2569 list_add(elist, &md->uevent_list); 2570 spin_unlock_irqrestore(&md->uevent_lock, flags); 2571 } 2572 2573 /* 2574 * The gendisk is only valid as long as you have a reference 2575 * count on 'md'. 2576 */ 2577 struct gendisk *dm_disk(struct mapped_device *md) 2578 { 2579 return md->disk; 2580 } 2581 2582 struct kobject *dm_kobject(struct mapped_device *md) 2583 { 2584 return &md->kobj; 2585 } 2586 2587 /* 2588 * struct mapped_device should not be exported outside of dm.c 2589 * so use this check to verify that kobj is part of md structure 2590 */ 2591 struct mapped_device *dm_get_from_kobject(struct kobject *kobj) 2592 { 2593 struct mapped_device *md; 2594 2595 md = container_of(kobj, struct mapped_device, kobj); 2596 if (&md->kobj != kobj) 2597 return NULL; 2598 2599 if (test_bit(DMF_FREEING, &md->flags) || 2600 dm_deleting_md(md)) 2601 return NULL; 2602 2603 dm_get(md); 2604 return md; 2605 } 2606 2607 int dm_suspended_md(struct mapped_device *md) 2608 { 2609 return test_bit(DMF_SUSPENDED, &md->flags); 2610 } 2611 2612 int dm_suspended(struct dm_target *ti) 2613 { 2614 return dm_suspended_md(dm_table_get_md(ti->table)); 2615 } 2616 EXPORT_SYMBOL_GPL(dm_suspended); 2617 2618 int dm_noflush_suspending(struct dm_target *ti) 2619 { 2620 return __noflush_suspending(dm_table_get_md(ti->table)); 2621 } 2622 EXPORT_SYMBOL_GPL(dm_noflush_suspending); 2623 2624 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity) 2625 { 2626 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); 2627 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS; 2628 2629 if (!pools) 2630 return NULL; 2631 2632 pools->io_pool = (type == DM_TYPE_BIO_BASED) ? 2633 mempool_create_slab_pool(MIN_IOS, _io_cache) : 2634 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); 2635 if (!pools->io_pool) 2636 goto free_pools_and_out; 2637 2638 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? 2639 mempool_create_slab_pool(MIN_IOS, _tio_cache) : 2640 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); 2641 if (!pools->tio_pool) 2642 goto free_io_pool_and_out; 2643 2644 pools->bs = bioset_create(pool_size, 0); 2645 if (!pools->bs) 2646 goto free_tio_pool_and_out; 2647 2648 if (integrity && bioset_integrity_create(pools->bs, pool_size)) 2649 goto free_bioset_and_out; 2650 2651 return pools; 2652 2653 free_bioset_and_out: 2654 bioset_free(pools->bs); 2655 2656 free_tio_pool_and_out: 2657 mempool_destroy(pools->tio_pool); 2658 2659 free_io_pool_and_out: 2660 mempool_destroy(pools->io_pool); 2661 2662 free_pools_and_out: 2663 kfree(pools); 2664 2665 return NULL; 2666 } 2667 2668 void dm_free_md_mempools(struct dm_md_mempools *pools) 2669 { 2670 if (!pools) 2671 return; 2672 2673 if (pools->io_pool) 2674 mempool_destroy(pools->io_pool); 2675 2676 if (pools->tio_pool) 2677 mempool_destroy(pools->tio_pool); 2678 2679 if (pools->bs) 2680 bioset_free(pools->bs); 2681 2682 kfree(pools); 2683 } 2684 2685 static const struct block_device_operations dm_blk_dops = { 2686 .open = dm_blk_open, 2687 .release = dm_blk_close, 2688 .ioctl = dm_blk_ioctl, 2689 .getgeo = dm_blk_getgeo, 2690 .owner = THIS_MODULE 2691 }; 2692 2693 EXPORT_SYMBOL(dm_get_mapinfo); 2694 2695 /* 2696 * module hooks 2697 */ 2698 module_init(dm_init); 2699 module_exit(dm_exit); 2700 2701 module_param(major, uint, 0); 2702 MODULE_PARM_DESC(major, "The major number of the device mapper"); 2703 MODULE_DESCRIPTION(DM_NAME " driver"); 2704 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); 2705 MODULE_LICENSE("GPL"); 2706