drivers/md/dm.c

   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/blktrace_api.h>
  23
  24 #include <trace/events/block.h>
  25
  26 #define DM_MSG_PREFIX "core"
  27
  28 static const char *_name = DM_NAME;
  29
  30 static unsigned int major = 0;
  31 static unsigned int _major = 0;
  32
  33 static DEFINE_SPINLOCK(_minor_lock);
  34 /*
  35  * For bio-based dm.
  36  * One of these is allocated per bio.
  37  */
  38 struct dm_io {
  39         struct mapped_device *md;
  40         int error;
  41         atomic_t io_count;
  42         struct bio *bio;
  43         unsigned long start_time;
  44 };
  45
  46 /*
  47  * For bio-based dm.
  48  * One of these is allocated per target within a bio.  Hopefully
  49  * this will be simplified out one day.
  50  */
  51 struct dm_target_io {
  52         struct dm_io *io;
  53         struct dm_target *ti;
  54         union map_info info;
  55 };
  56
  57 /*
  58  * For request-based dm.
  59  * One of these is allocated per request.
  60  */
  61 struct dm_rq_target_io {
  62         struct mapped_device *md;
  63         struct dm_target *ti;
  64         struct request *orig, clone;
  65         int error;
  66         union map_info info;
  67 };
  68
  69 /*
  70  * For request-based dm.
  71  * One of these is allocated per bio.
  72  */
  73 struct dm_rq_clone_bio_info {
  74         struct bio *orig;
  75         struct request *rq;
  76 };
  77
  78 union map_info *dm_get_mapinfo(struct bio *bio)
  79 {
  80         if (bio && bio->bi_private)
  81                 return &((struct dm_target_io *)bio->bi_private)->info;
  82         return NULL;
  83 }
  84
  85 #define MINOR_ALLOCED ((void *)-1)
  86
  87 /*
  88  * Bits for the md->flags field.
  89  */
  90 #define DMF_BLOCK_IO_FOR_SUSPEND 0
  91 #define DMF_SUSPENDED 1
  92 #define DMF_FROZEN 2
  93 #define DMF_FREEING 3
  94 #define DMF_DELETING 4
  95 #define DMF_NOFLUSH_SUSPENDING 5
  96 #define DMF_QUEUE_IO_TO_THREAD 6
  97
  98 /*
  99  * Work processed by per-device workqueue.
 100  */
 101 struct mapped_device {
 102         struct rw_semaphore io_lock;
 103         struct mutex suspend_lock;
 104         rwlock_t map_lock;
 105         atomic_t holders;
 106         atomic_t open_count;
 107
 108         unsigned long flags;
 109
 110         struct request_queue *queue;
 111         struct gendisk *disk;
 112         char name[16];
 113
 114         void *interface_ptr;
 115
 116         /*
 117          * A list of ios that arrived while we were suspended.
 118          */
 119         atomic_t pending;
 120         wait_queue_head_t wait;
 121         struct work_struct work;
 122         struct bio_list deferred;
 123         spinlock_t deferred_lock;
 124
 125         /*
 126          * An error from the barrier request currently being processed.
 127          */
 128         int barrier_error;
 129
 130         /*
 131          * Processing queue (flush/barriers)
 132          */
 133         struct workqueue_struct *wq;
 134
 135         /*
 136          * The current mapping.
 137          */
 138         struct dm_table *map;
 139
 140         /*
 141          * io objects are allocated from here.
 142          */
 143         mempool_t *io_pool;
 144         mempool_t *tio_pool;
 145
 146         struct bio_set *bs;
 147
 148         /*
 149          * Event handling.
 150          */
 151         atomic_t event_nr;
 152         wait_queue_head_t eventq;
 153         atomic_t uevent_seq;
 154         struct list_head uevent_list;
 155         spinlock_t uevent_lock; /* Protect access to uevent_list */
 156
 157         /*
 158          * freeze/thaw support require holding onto a super block
 159          */
 160         struct super_block *frozen_sb;
 161         struct block_device *suspended_bdev;
 162
 163         /* forced geometry settings */
 164         struct hd_geometry geometry;
 165
 166         /* sysfs handle */
 167         struct kobject kobj;
 168 };
 169
 170 #define MIN_IOS 256
 171 static struct kmem_cache *_io_cache;
 172 static struct kmem_cache *_tio_cache;
 173 static struct kmem_cache *_rq_tio_cache;
 174 static struct kmem_cache *_rq_bio_info_cache;
 175
 176 static int __init local_init(void)
 177 {
 178         int r = -ENOMEM;
 179
 180         /* allocate a slab for the dm_ios */
 181         _io_cache = KMEM_CACHE(dm_io, 0);
 182         if (!_io_cache)
 183                 return r;
 184
 185         /* allocate a slab for the target ios */
 186         _tio_cache = KMEM_CACHE(dm_target_io, 0);
 187         if (!_tio_cache)
 188                 goto out_free_io_cache;
 189
 190         _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
 191         if (!_rq_tio_cache)
 192                 goto out_free_tio_cache;
 193
 194         _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
 195         if (!_rq_bio_info_cache)
 196                 goto out_free_rq_tio_cache;
 197
 198         r = dm_uevent_init();
 199         if (r)
 200                 goto out_free_rq_bio_info_cache;
 201
 202         _major = major;
 203         r = register_blkdev(_major, _name);
 204         if (r < 0)
 205                 goto out_uevent_exit;
 206
 207         if (!_major)
 208                 _major = r;
 209
 210         return 0;
 211
 212 out_uevent_exit:
 213         dm_uevent_exit();
 214 out_free_rq_bio_info_cache:
 215         kmem_cache_destroy(_rq_bio_info_cache);
 216 out_free_rq_tio_cache:
 217         kmem_cache_destroy(_rq_tio_cache);
 218 out_free_tio_cache:
 219         kmem_cache_destroy(_tio_cache);
 220 out_free_io_cache:
 221         kmem_cache_destroy(_io_cache);
 222
 223         return r;
 224 }
 225
 226 static void local_exit(void)
 227 {
 228         kmem_cache_destroy(_rq_bio_info_cache);
 229         kmem_cache_destroy(_rq_tio_cache);
 230         kmem_cache_destroy(_tio_cache);
 231         kmem_cache_destroy(_io_cache);
 232         unregister_blkdev(_major, _name);
 233         dm_uevent_exit();
 234
 235         _major = 0;
 236
 237         DMINFO("cleaned up");
 238 }
 239
 240 static int (*_inits[])(void) __initdata = {
 241         local_init,
 242         dm_target_init,
 243         dm_linear_init,
 244         dm_stripe_init,
 245         dm_kcopyd_init,
 246         dm_interface_init,
 247 };
 248
 249 static void (*_exits[])(void) = {
 250         local_exit,
 251         dm_target_exit,
 252         dm_linear_exit,
 253         dm_stripe_exit,
 254         dm_kcopyd_exit,
 255         dm_interface_exit,
 256 };
 257
 258 static int __init dm_init(void)
 259 {
 260         const int count = ARRAY_SIZE(_inits);
 261
 262         int r, i;
 263
 264         for (i = 0; i < count; i++) {
 265                 r = _inits[i]();
 266                 if (r)
 267                         goto bad;
 268         }
 269
 270         return 0;
 271
 272       bad:
 273         while (i--)
 274                 _exits[i]();
 275
 276         return r;
 277 }
 278
 279 static void __exit dm_exit(void)
 280 {
 281         int i = ARRAY_SIZE(_exits);
 282
 283         while (i--)
 284                 _exits[i]();
 285 }
 286
 287 /*
 288  * Block device functions
 289  */
 290 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 291 {
 292         struct mapped_device *md;
 293
 294         spin_lock(&_minor_lock);
 295
 296         md = bdev->bd_disk->private_data;
 297         if (!md)
 298                 goto out;
 299
 300         if (test_bit(DMF_FREEING, &md->flags) ||
 301             test_bit(DMF_DELETING, &md->flags)) {
 302                 md = NULL;
 303                 goto out;
 304         }
 305
 306         dm_get(md);
 307         atomic_inc(&md->open_count);
 308
 309 out:
 310         spin_unlock(&_minor_lock);
 311
 312         return md ? 0 : -ENXIO;
 313 }
 314
 315 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
 316 {
 317         struct mapped_device *md = disk->private_data;
 318         atomic_dec(&md->open_count);
 319         dm_put(md);
 320         return 0;
 321 }
 322
 323 int dm_open_count(struct mapped_device *md)
 324 {
 325         return atomic_read(&md->open_count);
 326 }
 327
 328 /*
 329  * Guarantees nothing is using the device before it's deleted.
 330  */
 331 int dm_lock_for_deletion(struct mapped_device *md)
 332 {
 333         int r = 0;
 334
 335         spin_lock(&_minor_lock);
 336
 337         if (dm_open_count(md))
 338                 r = -EBUSY;
 339         else
 340                 set_bit(DMF_DELETING, &md->flags);
 341
 342         spin_unlock(&_minor_lock);
 343
 344         return r;
 345 }
 346
 347 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 348 {
 349         struct mapped_device *md = bdev->bd_disk->private_data;
 350
 351         return dm_get_geometry(md, geo);
 352 }
 353
 354 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 355                         unsigned int cmd, unsigned long arg)
 356 {
 357         struct mapped_device *md = bdev->bd_disk->private_data;
 358         struct dm_table *map = dm_get_table(md);
 359         struct dm_target *tgt;
 360         int r = -ENOTTY;
 361
 362         if (!map || !dm_table_get_size(map))
 363                 goto out;
 364
 365         /* We only support devices that have a single target */
 366         if (dm_table_get_num_targets(map) != 1)
 367                 goto out;
 368
 369         tgt = dm_table_get_target(map, 0);
 370
 371         if (dm_suspended(md)) {
 372                 r = -EAGAIN;
 373                 goto out;
 374         }
 375
 376         if (tgt->type->ioctl)
 377                 r = tgt->type->ioctl(tgt, cmd, arg);
 378
 379 out:
 380         dm_table_put(map);
 381
 382         return r;
 383 }
 384
 385 static struct dm_io *alloc_io(struct mapped_device *md)
 386 {
 387         return mempool_alloc(md->io_pool, GFP_NOIO);
 388 }
 389
 390 static void free_io(struct mapped_device *md, struct dm_io *io)
 391 {
 392         mempool_free(io, md->io_pool);
 393 }
 394
 395 static struct dm_target_io *alloc_tio(struct mapped_device *md)
 396 {
 397         return mempool_alloc(md->tio_pool, GFP_NOIO);
 398 }
 399
 400 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
 401 {
 402         mempool_free(tio, md->tio_pool);
 403 }
 404
 405 static void start_io_acct(struct dm_io *io)
 406 {
 407         struct mapped_device *md = io->md;
 408         int cpu;
 409
 410         io->start_time = jiffies;
 411
 412         cpu = part_stat_lock();
 413         part_round_stats(cpu, &dm_disk(md)->part0);
 414         part_stat_unlock();
 415         dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
 416 }
 417
 418 static void end_io_acct(struct dm_io *io)
 419 {
 420         struct mapped_device *md = io->md;
 421         struct bio *bio = io->bio;
 422         unsigned long duration = jiffies - io->start_time;
 423         int pending, cpu;
 424         int rw = bio_data_dir(bio);
 425
 426         cpu = part_stat_lock();
 427         part_round_stats(cpu, &dm_disk(md)->part0);
 428         part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
 429         part_stat_unlock();
 430
 431         /*
 432          * After this is decremented the bio must not be touched if it is
 433          * a barrier.
 434          */
 435         dm_disk(md)->part0.in_flight = pending =
 436                 atomic_dec_return(&md->pending);
 437
 438         /* nudge anyone waiting on suspend queue */
 439         if (!pending)
 440                 wake_up(&md->wait);
 441 }
 442
 443 /*
 444  * Add the bio to the list of deferred io.
 445  */
 446 static void queue_io(struct mapped_device *md, struct bio *bio)
 447 {
 448         down_write(&md->io_lock);
 449
 450         spin_lock_irq(&md->deferred_lock);
 451         bio_list_add(&md->deferred, bio);
 452         spin_unlock_irq(&md->deferred_lock);
 453
 454         if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
 455                 queue_work(md->wq, &md->work);
 456
 457         up_write(&md->io_lock);
 458 }
 459
 460 /*
 461  * Everyone (including functions in this file), should use this
 462  * function to access the md->map field, and make sure they call
 463  * dm_table_put() when finished.
 464  */
 465 struct dm_table *dm_get_table(struct mapped_device *md)
 466 {
 467         struct dm_table *t;
 468
 469         read_lock(&md->map_lock);
 470         t = md->map;
 471         if (t)
 472                 dm_table_get(t);
 473         read_unlock(&md->map_lock);
 474
 475         return t;
 476 }
 477
 478 /*
 479  * Get the geometry associated with a dm device
 480  */
 481 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 482 {
 483         *geo = md->geometry;
 484
 485         return 0;
 486 }
 487
 488 /*
 489  * Set the geometry of a device.
 490  */
 491 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 492 {
 493         sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 494
 495         if (geo->start > sz) {
 496                 DMWARN("Start sector is beyond the geometry limits.");
 497                 return -EINVAL;
 498         }
 499
 500         md->geometry = *geo;
 501
 502         return 0;
 503 }
 504
 505 /*-----------------------------------------------------------------
 506  * CRUD START:
 507  *   A more elegant soln is in the works that uses the queue
 508  *   merge fn, unfortunately there are a couple of changes to
 509  *   the block layer that I want to make for this.  So in the
 510  *   interests of getting something for people to use I give
 511  *   you this clearly demarcated crap.
 512  *---------------------------------------------------------------*/
 513
 514 static int __noflush_suspending(struct mapped_device *md)
 515 {
 516         return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 517 }
 518
 519 /*
 520  * Decrements the number of outstanding ios that a bio has been
 521  * cloned into, completing the original io if necc.
 522  */
 523 static void dec_pending(struct dm_io *io, int error)
 524 {
 525         unsigned long flags;
 526         int io_error;
 527         struct bio *bio;
 528         struct mapped_device *md = io->md;
 529
 530         /* Push-back supersedes any I/O errors */
 531         if (error && !(io->error > 0 && __noflush_suspending(md)))
 532                 io->error = error;
 533
 534         if (atomic_dec_and_test(&io->io_count)) {
 535                 if (io->error == DM_ENDIO_REQUEUE) {
 536                         /*
 537                          * Target requested pushing back the I/O.
 538                          */
 539                         spin_lock_irqsave(&md->deferred_lock, flags);
 540                         if (__noflush_suspending(md))
 541                                 bio_list_add_head(&md->deferred, io->bio);
 542                         else
 543                                 /* noflush suspend was interrupted. */
 544                                 io->error = -EIO;
 545                         spin_unlock_irqrestore(&md->deferred_lock, flags);
 546                 }
 547
 548                 io_error = io->error;
 549                 bio = io->bio;
 550
 551                 if (bio_barrier(bio)) {
 552                         /*
 553                          * There can be just one barrier request so we use
 554                          * a per-device variable for error reporting.
 555                          * Note that you can't touch the bio after end_io_acct
 556                          */
 557                         md->barrier_error = io_error;
 558                         end_io_acct(io);
 559                 } else {
 560                         end_io_acct(io);
 561
 562                         if (io_error != DM_ENDIO_REQUEUE) {
 563                                 trace_block_bio_complete(md->queue, bio);
 564
 565                                 bio_endio(bio, io_error);
 566                         }
 567                 }
 568
 569                 free_io(md, io);
 570         }
 571 }
 572
 573 static void clone_endio(struct bio *bio, int error)
 574 {
 575         int r = 0;
 576         struct dm_target_io *tio = bio->bi_private;
 577         struct dm_io *io = tio->io;
 578         struct mapped_device *md = tio->io->md;
 579         dm_endio_fn endio = tio->ti->type->end_io;
 580
 581         if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 582                 error = -EIO;
 583
 584         if (endio) {
 585                 r = endio(tio->ti, bio, error, &tio->info);
 586                 if (r < 0 || r == DM_ENDIO_REQUEUE)
 587                         /*
 588                          * error and requeue request are handled
 589                          * in dec_pending().
 590                          */
 591                         error = r;
 592                 else if (r == DM_ENDIO_INCOMPLETE)
 593                         /* The target will handle the io */
 594                         return;
 595                 else if (r) {
 596                         DMWARN("unimplemented target endio return value: %d", r);
 597                         BUG();
 598                 }
 599         }
 600
 601         /*
 602          * Store md for cleanup instead of tio which is about to get freed.
 603          */
 604         bio->bi_private = md->bs;
 605
 606         free_tio(md, tio);
 607         bio_put(bio);
 608         dec_pending(io, error);
 609 }
 610
 611 static sector_t max_io_len(struct mapped_device *md,
 612                            sector_t sector, struct dm_target *ti)
 613 {
 614         sector_t offset = sector - ti->begin;
 615         sector_t len = ti->len - offset;
 616
 617         /*
 618          * Does the target need to split even further ?
 619          */
 620         if (ti->split_io) {
 621                 sector_t boundary;
 622                 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
 623                            - offset;
 624                 if (len > boundary)
 625                         len = boundary;
 626         }
 627
 628         return len;
 629 }
 630
 631 static void __map_bio(struct dm_target *ti, struct bio *clone,
 632                       struct dm_target_io *tio)
 633 {
 634         int r;
 635         sector_t sector;
 636         struct mapped_device *md;
 637
 638         /*
 639          * Sanity checks.
 640          */
 641         BUG_ON(!clone->bi_size);
 642
 643         clone->bi_end_io = clone_endio;
 644         clone->bi_private = tio;
 645
 646         /*
 647          * Map the clone.  If r == 0 we don't need to do
 648          * anything, the target has assumed ownership of
 649          * this io.
 650          */
 651         atomic_inc(&tio->io->io_count);
 652         sector = clone->bi_sector;
 653         r = ti->type->map(ti, clone, &tio->info);
 654         if (r == DM_MAPIO_REMAPPED) {
 655                 /* the bio has been remapped so dispatch it */
 656
 657                 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
 658                                     tio->io->bio->bi_bdev->bd_dev, sector);
 659
 660                 generic_make_request(clone);
 661         } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
 662                 /* error the io and bail out, or requeue it if needed */
 663                 md = tio->io->md;
 664                 dec_pending(tio->io, r);
 665                 /*
 666                  * Store bio_set for cleanup.
 667                  */
 668                 clone->bi_private = md->bs;
 669                 bio_put(clone);
 670                 free_tio(md, tio);
 671         } else if (r) {
 672                 DMWARN("unimplemented target map return value: %d", r);
 673                 BUG();
 674         }
 675 }
 676
 677 struct clone_info {
 678         struct mapped_device *md;
 679         struct dm_table *map;
 680         struct bio *bio;
 681         struct dm_io *io;
 682         sector_t sector;
 683         sector_t sector_count;
 684         unsigned short idx;
 685 };
 686
 687 static void dm_bio_destructor(struct bio *bio)
 688 {
 689         struct bio_set *bs = bio->bi_private;
 690
 691         bio_free(bio, bs);
 692 }
 693
 694 /*
 695  * Creates a little bio that is just does part of a bvec.
 696  */
 697 static struct bio *split_bvec(struct bio *bio, sector_t sector,
 698                               unsigned short idx, unsigned int offset,
 699                               unsigned int len, struct bio_set *bs)
 700 {
 701         struct bio *clone;
 702         struct bio_vec *bv = bio->bi_io_vec + idx;
 703
 704         clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
 705         clone->bi_destructor = dm_bio_destructor;
 706         *clone->bi_io_vec = *bv;
 707
 708         clone->bi_sector = sector;
 709         clone->bi_bdev = bio->bi_bdev;
 710         clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
 711         clone->bi_vcnt = 1;
 712         clone->bi_size = to_bytes(len);
 713         clone->bi_io_vec->bv_offset = offset;
 714         clone->bi_io_vec->bv_len = clone->bi_size;
 715         clone->bi_flags |= 1 << BIO_CLONED;
 716
 717         if (bio_integrity(bio)) {
 718                 bio_integrity_clone(clone, bio, GFP_NOIO);
 719                 bio_integrity_trim(clone,
 720                                    bio_sector_offset(bio, idx, offset), len);
 721         }
 722
 723         return clone;
 724 }
 725
 726 /*
 727  * Creates a bio that consists of range of complete bvecs.
 728  */
 729 static struct bio *clone_bio(struct bio *bio, sector_t sector,
 730                              unsigned short idx, unsigned short bv_count,
 731                              unsigned int len, struct bio_set *bs)
 732 {
 733         struct bio *clone;
 734
 735         clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
 736         __bio_clone(clone, bio);
 737         clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
 738         clone->bi_destructor = dm_bio_destructor;
 739         clone->bi_sector = sector;
 740         clone->bi_idx = idx;
 741         clone->bi_vcnt = idx + bv_count;
 742         clone->bi_size = to_bytes(len);
 743         clone->bi_flags &= ~(1 << BIO_SEG_VALID);
 744
 745         if (bio_integrity(bio)) {
 746                 bio_integrity_clone(clone, bio, GFP_NOIO);
 747
 748                 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
 749                         bio_integrity_trim(clone,
 750                                            bio_sector_offset(bio, idx, 0), len);
 751         }
 752
 753         return clone;
 754 }
 755
 756 static int __clone_and_map(struct clone_info *ci)
 757 {
 758         struct bio *clone, *bio = ci->bio;
 759         struct dm_target *ti;
 760         sector_t len = 0, max;
 761         struct dm_target_io *tio;
 762
 763         ti = dm_table_find_target(ci->map, ci->sector);
 764         if (!dm_target_is_valid(ti))
 765                 return -EIO;
 766
 767         max = max_io_len(ci->md, ci->sector, ti);
 768
 769         /*
 770          * Allocate a target io object.
 771          */
 772         tio = alloc_tio(ci->md);
 773         tio->io = ci->io;
 774         tio->ti = ti;
 775         memset(&tio->info, 0, sizeof(tio->info));
 776
 777         if (ci->sector_count <= max) {
 778                 /*
 779                  * Optimise for the simple case where we can do all of
 780                  * the remaining io with a single clone.
 781                  */
 782                 clone = clone_bio(bio, ci->sector, ci->idx,
 783                                   bio->bi_vcnt - ci->idx, ci->sector_count,
 784                                   ci->md->bs);
 785                 __map_bio(ti, clone, tio);
 786                 ci->sector_count = 0;
 787
 788         } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
 789                 /*
 790                  * There are some bvecs that don't span targets.
 791                  * Do as many of these as possible.
 792                  */
 793                 int i;
 794                 sector_t remaining = max;
 795                 sector_t bv_len;
 796
 797                 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
 798                         bv_len = to_sector(bio->bi_io_vec[i].bv_len);
 799
 800                         if (bv_len > remaining)
 801                                 break;
 802
 803                         remaining -= bv_len;
 804                         len += bv_len;
 805                 }
 806
 807                 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
 808                                   ci->md->bs);
 809                 __map_bio(ti, clone, tio);
 810
 811                 ci->sector += len;
 812                 ci->sector_count -= len;
 813                 ci->idx = i;
 814
 815         } else {
 816                 /*
 817                  * Handle a bvec that must be split between two or more targets.
 818                  */
 819                 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
 820                 sector_t remaining = to_sector(bv->bv_len);
 821                 unsigned int offset = 0;
 822
 823                 do {
 824                         if (offset) {
 825                                 ti = dm_table_find_target(ci->map, ci->sector);
 826                                 if (!dm_target_is_valid(ti))
 827                                         return -EIO;
 828
 829                                 max = max_io_len(ci->md, ci->sector, ti);
 830
 831                                 tio = alloc_tio(ci->md);
 832                                 tio->io = ci->io;
 833                                 tio->ti = ti;
 834                                 memset(&tio->info, 0, sizeof(tio->info));
 835                         }
 836
 837                         len = min(remaining, max);
 838
 839                         clone = split_bvec(bio, ci->sector, ci->idx,
 840                                            bv->bv_offset + offset, len,
 841                                            ci->md->bs);
 842
 843                         __map_bio(ti, clone, tio);
 844
 845                         ci->sector += len;
 846                         ci->sector_count -= len;
 847                         offset += to_bytes(len);
 848                 } while (remaining -= len);
 849
 850                 ci->idx++;
 851         }
 852
 853         return 0;
 854 }
 855
 856 /*
 857  * Split the bio into several clones and submit it to targets.
 858  */
 859 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
 860 {
 861         struct clone_info ci;
 862         int error = 0;
 863
 864         ci.map = dm_get_table(md);
 865         if (unlikely(!ci.map)) {
 866                 if (!bio_barrier(bio))
 867                         bio_io_error(bio);
 868                 else
 869                         md->barrier_error = -EIO;
 870                 return;
 871         }
 872
 873         ci.md = md;
 874         ci.bio = bio;
 875         ci.io = alloc_io(md);
 876         ci.io->error = 0;
 877         atomic_set(&ci.io->io_count, 1);
 878         ci.io->bio = bio;
 879         ci.io->md = md;
 880         ci.sector = bio->bi_sector;
 881         ci.sector_count = bio_sectors(bio);
 882         ci.idx = bio->bi_idx;
 883
 884         start_io_acct(ci.io);
 885         while (ci.sector_count && !error)
 886                 error = __clone_and_map(&ci);
 887
 888         /* drop the extra reference count */
 889         dec_pending(ci.io, error);
 890         dm_table_put(ci.map);
 891 }
 892 /*-----------------------------------------------------------------
 893  * CRUD END
 894  *---------------------------------------------------------------*/
 895
 896 static int dm_merge_bvec(struct request_queue *q,
 897                          struct bvec_merge_data *bvm,
 898                          struct bio_vec *biovec)
 899 {
 900         struct mapped_device *md = q->queuedata;
 901         struct dm_table *map = dm_get_table(md);
 902         struct dm_target *ti;
 903         sector_t max_sectors;
 904         int max_size = 0;
 905
 906         if (unlikely(!map))
 907                 goto out;
 908
 909         ti = dm_table_find_target(map, bvm->bi_sector);
 910         if (!dm_target_is_valid(ti))
 911                 goto out_table;
 912
 913         /*
 914          * Find maximum amount of I/O that won't need splitting
 915          */
 916         max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
 917                           (sector_t) BIO_MAX_SECTORS);
 918         max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
 919         if (max_size < 0)
 920                 max_size = 0;
 921
 922         /*
 923          * merge_bvec_fn() returns number of bytes
 924          * it can accept at this offset
 925          * max is precomputed maximal io size
 926          */
 927         if (max_size && ti->type->merge)
 928                 max_size = ti->type->merge(ti, bvm, biovec, max_size);
 929
 930 out_table:
 931         dm_table_put(map);
 932
 933 out:
 934         /*
 935          * Always allow an entire first page
 936          */
 937         if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
 938                 max_size = biovec->bv_len;
 939
 940         return max_size;
 941 }
 942
 943 /*
 944  * The request function that just remaps the bio built up by
 945  * dm_merge_bvec.
 946  */
 947 static int dm_request(struct request_queue *q, struct bio *bio)
 948 {
 949         int rw = bio_data_dir(bio);
 950         struct mapped_device *md = q->queuedata;
 951         int cpu;
 952
 953         down_read(&md->io_lock);
 954
 955         cpu = part_stat_lock();
 956         part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
 957         part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
 958         part_stat_unlock();
 959
 960         /*
 961          * If we're suspended or the thread is processing barriers
 962          * we have to queue this io for later.
 963          */
 964         if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
 965             unlikely(bio_barrier(bio))) {
 966                 up_read(&md->io_lock);
 967
 968                 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
 969                     bio_rw(bio) == READA) {
 970                         bio_io_error(bio);
 971                         return 0;
 972                 }
 973
 974                 queue_io(md, bio);
 975
 976                 return 0;
 977         }
 978
 979         __split_and_process_bio(md, bio);
 980         up_read(&md->io_lock);
 981         return 0;
 982 }
 983
 984 static void dm_unplug_all(struct request_queue *q)
 985 {
 986         struct mapped_device *md = q->queuedata;
 987         struct dm_table *map = dm_get_table(md);
 988
 989         if (map) {
 990                 dm_table_unplug_all(map);
 991                 dm_table_put(map);
 992         }
 993 }
 994
 995 static int dm_any_congested(void *congested_data, int bdi_bits)
 996 {
 997         int r = bdi_bits;
 998         struct mapped_device *md = congested_data;
 999         struct dm_table *map;
1000
1001         if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1002                 map = dm_get_table(md);
1003                 if (map) {
1004                         r = dm_table_any_congested(map, bdi_bits);
1005                         dm_table_put(map);
1006                 }
1007         }
1008
1009         return r;
1010 }
1011
1012 /*-----------------------------------------------------------------
1013  * An IDR is used to keep track of allocated minor numbers.
1014  *---------------------------------------------------------------*/
1015 static DEFINE_IDR(_minor_idr);
1016
1017 static void free_minor(int minor)
1018 {
1019         spin_lock(&_minor_lock);
1020         idr_remove(&_minor_idr, minor);
1021         spin_unlock(&_minor_lock);
1022 }
1023
1024 /*
1025  * See if the device with a specific minor # is free.
1026  */
1027 static int specific_minor(int minor)
1028 {
1029         int r, m;
1030
1031         if (minor >= (1 << MINORBITS))
1032                 return -EINVAL;
1033
1034         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1035         if (!r)
1036                 return -ENOMEM;
1037
1038         spin_lock(&_minor_lock);
1039
1040         if (idr_find(&_minor_idr, minor)) {
1041                 r = -EBUSY;
1042                 goto out;
1043         }
1044
1045         r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1046         if (r)
1047                 goto out;
1048
1049         if (m != minor) {
1050                 idr_remove(&_minor_idr, m);
1051                 r = -EBUSY;
1052                 goto out;
1053         }
1054
1055 out:
1056         spin_unlock(&_minor_lock);
1057         return r;
1058 }
1059
1060 static int next_free_minor(int *minor)
1061 {
1062         int r, m;
1063
1064         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1065         if (!r)
1066                 return -ENOMEM;
1067
1068         spin_lock(&_minor_lock);
1069
1070         r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1071         if (r)
1072                 goto out;
1073
1074         if (m >= (1 << MINORBITS)) {
1075                 idr_remove(&_minor_idr, m);
1076                 r = -ENOSPC;
1077                 goto out;
1078         }
1079
1080         *minor = m;
1081
1082 out:
1083         spin_unlock(&_minor_lock);
1084         return r;
1085 }
1086
1087 static struct block_device_operations dm_blk_dops;
1088
1089 static void dm_wq_work(struct work_struct *work);
1090
1091 /*
1092  * Allocate and initialise a blank device with a given minor.
1093  */
1094 static struct mapped_device *alloc_dev(int minor)
1095 {
1096         int r;
1097         struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1098         void *old_md;
1099
1100         if (!md) {
1101                 DMWARN("unable to allocate device, out of memory.");
1102                 return NULL;
1103         }
1104
1105         if (!try_module_get(THIS_MODULE))
1106                 goto bad_module_get;
1107
1108         /* get a minor number for the dev */
1109         if (minor == DM_ANY_MINOR)
1110                 r = next_free_minor(&minor);
1111         else
1112                 r = specific_minor(minor);
1113         if (r < 0)
1114                 goto bad_minor;
1115
1116         init_rwsem(&md->io_lock);
1117         mutex_init(&md->suspend_lock);
1118         spin_lock_init(&md->deferred_lock);
1119         rwlock_init(&md->map_lock);
1120         atomic_set(&md->holders, 1);
1121         atomic_set(&md->open_count, 0);
1122         atomic_set(&md->event_nr, 0);
1123         atomic_set(&md->uevent_seq, 0);
1124         INIT_LIST_HEAD(&md->uevent_list);
1125         spin_lock_init(&md->uevent_lock);
1126
1127         md->queue = blk_alloc_queue(GFP_KERNEL);
1128         if (!md->queue)
1129                 goto bad_queue;
1130
1131         md->queue->queuedata = md;
1132         md->queue->backing_dev_info.congested_fn = dm_any_congested;
1133         md->queue->backing_dev_info.congested_data = md;
1134         blk_queue_make_request(md->queue, dm_request);
1135         blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
1136         blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1137         md->queue->unplug_fn = dm_unplug_all;
1138         blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1139
1140         md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1141         if (!md->io_pool)
1142                 goto bad_io_pool;
1143
1144         md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1145         if (!md->tio_pool)
1146                 goto bad_tio_pool;
1147
1148         md->bs = bioset_create(16, 0);
1149         if (!md->bs)
1150                 goto bad_no_bioset;
1151
1152         md->disk = alloc_disk(1);
1153         if (!md->disk)
1154                 goto bad_disk;
1155
1156         atomic_set(&md->pending, 0);
1157         init_waitqueue_head(&md->wait);
1158         INIT_WORK(&md->work, dm_wq_work);
1159         init_waitqueue_head(&md->eventq);
1160
1161         md->disk->major = _major;
1162         md->disk->first_minor = minor;
1163         md->disk->fops = &dm_blk_dops;
1164         md->disk->queue = md->queue;
1165         md->disk->private_data = md;
1166         sprintf(md->disk->disk_name, "dm-%d", minor);
1167         add_disk(md->disk);
1168         format_dev_t(md->name, MKDEV(_major, minor));
1169
1170         md->wq = create_singlethread_workqueue("kdmflush");
1171         if (!md->wq)
1172                 goto bad_thread;
1173
1174         /* Populate the mapping, nobody knows we exist yet */
1175         spin_lock(&_minor_lock);
1176         old_md = idr_replace(&_minor_idr, md, minor);
1177         spin_unlock(&_minor_lock);
1178
1179         BUG_ON(old_md != MINOR_ALLOCED);
1180
1181         return md;
1182
1183 bad_thread:
1184         put_disk(md->disk);
1185 bad_disk:
1186         bioset_free(md->bs);
1187 bad_no_bioset:
1188         mempool_destroy(md->tio_pool);
1189 bad_tio_pool:
1190         mempool_destroy(md->io_pool);
1191 bad_io_pool:
1192         blk_cleanup_queue(md->queue);
1193 bad_queue:
1194         free_minor(minor);
1195 bad_minor:
1196         module_put(THIS_MODULE);
1197 bad_module_get:
1198         kfree(md);
1199         return NULL;
1200 }
1201
1202 static void unlock_fs(struct mapped_device *md);
1203
1204 static void free_dev(struct mapped_device *md)
1205 {
1206         int minor = MINOR(disk_devt(md->disk));
1207
1208         if (md->suspended_bdev) {
1209                 unlock_fs(md);
1210                 bdput(md->suspended_bdev);
1211         }
1212         destroy_workqueue(md->wq);
1213         mempool_destroy(md->tio_pool);
1214         mempool_destroy(md->io_pool);
1215         bioset_free(md->bs);
1216         blk_integrity_unregister(md->disk);
1217         del_gendisk(md->disk);
1218         free_minor(minor);
1219
1220         spin_lock(&_minor_lock);
1221         md->disk->private_data = NULL;
1222         spin_unlock(&_minor_lock);
1223
1224         put_disk(md->disk);
1225         blk_cleanup_queue(md->queue);
1226         module_put(THIS_MODULE);
1227         kfree(md);
1228 }
1229
1230 /*
1231  * Bind a table to the device.
1232  */
1233 static void event_callback(void *context)
1234 {
1235         unsigned long flags;
1236         LIST_HEAD(uevents);
1237         struct mapped_device *md = (struct mapped_device *) context;
1238
1239         spin_lock_irqsave(&md->uevent_lock, flags);
1240         list_splice_init(&md->uevent_list, &uevents);
1241         spin_unlock_irqrestore(&md->uevent_lock, flags);
1242
1243         dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1244
1245         atomic_inc(&md->event_nr);
1246         wake_up(&md->eventq);
1247 }
1248
1249 static void __set_size(struct mapped_device *md, sector_t size)
1250 {
1251         set_capacity(md->disk, size);
1252
1253         mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1254         i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1255         mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1256 }
1257
1258 static int __bind(struct mapped_device *md, struct dm_table *t)
1259 {
1260         struct request_queue *q = md->queue;
1261         sector_t size;
1262
1263         size = dm_table_get_size(t);
1264
1265         /*
1266          * Wipe any geometry if the size of the table changed.
1267          */
1268         if (size != get_capacity(md->disk))
1269                 memset(&md->geometry, 0, sizeof(md->geometry));
1270
1271         if (md->suspended_bdev)
1272                 __set_size(md, size);
1273
1274         if (!size) {
1275                 dm_table_destroy(t);
1276                 return 0;
1277         }
1278
1279         dm_table_event_callback(t, event_callback, md);
1280
1281         write_lock(&md->map_lock);
1282         md->map = t;
1283         dm_table_set_restrictions(t, q);
1284         write_unlock(&md->map_lock);
1285
1286         return 0;
1287 }
1288
1289 static void __unbind(struct mapped_device *md)
1290 {
1291         struct dm_table *map = md->map;
1292
1293         if (!map)
1294                 return;
1295
1296         dm_table_event_callback(map, NULL, NULL);
1297         write_lock(&md->map_lock);
1298         md->map = NULL;
1299         write_unlock(&md->map_lock);
1300         dm_table_destroy(map);
1301 }
1302
1303 /*
1304  * Constructor for a new device.
1305  */
1306 int dm_create(int minor, struct mapped_device **result)
1307 {
1308         struct mapped_device *md;
1309
1310         md = alloc_dev(minor);
1311         if (!md)
1312                 return -ENXIO;
1313
1314         dm_sysfs_init(md);
1315
1316         *result = md;
1317         return 0;
1318 }
1319
1320 static struct mapped_device *dm_find_md(dev_t dev)
1321 {
1322         struct mapped_device *md;
1323         unsigned minor = MINOR(dev);
1324
1325         if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1326                 return NULL;
1327
1328         spin_lock(&_minor_lock);
1329
1330         md = idr_find(&_minor_idr, minor);
1331         if (md && (md == MINOR_ALLOCED ||
1332                    (MINOR(disk_devt(dm_disk(md))) != minor) ||
1333                    test_bit(DMF_FREEING, &md->flags))) {
1334                 md = NULL;
1335                 goto out;
1336         }
1337
1338 out:
1339         spin_unlock(&_minor_lock);
1340
1341         return md;
1342 }
1343
1344 struct mapped_device *dm_get_md(dev_t dev)
1345 {
1346         struct mapped_device *md = dm_find_md(dev);
1347
1348         if (md)
1349                 dm_get(md);
1350
1351         return md;
1352 }
1353
1354 void *dm_get_mdptr(struct mapped_device *md)
1355 {
1356         return md->interface_ptr;
1357 }
1358
1359 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1360 {
1361         md->interface_ptr = ptr;
1362 }
1363
1364 void dm_get(struct mapped_device *md)
1365 {
1366         atomic_inc(&md->holders);
1367 }
1368
1369 const char *dm_device_name(struct mapped_device *md)
1370 {
1371         return md->name;
1372 }
1373 EXPORT_SYMBOL_GPL(dm_device_name);
1374
1375 void dm_put(struct mapped_device *md)
1376 {
1377         struct dm_table *map;
1378
1379         BUG_ON(test_bit(DMF_FREEING, &md->flags));
1380
1381         if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1382                 map = dm_get_table(md);
1383                 idr_replace(&_minor_idr, MINOR_ALLOCED,
1384                             MINOR(disk_devt(dm_disk(md))));
1385                 set_bit(DMF_FREEING, &md->flags);
1386                 spin_unlock(&_minor_lock);
1387                 if (!dm_suspended(md)) {
1388                         dm_table_presuspend_targets(map);
1389                         dm_table_postsuspend_targets(map);
1390                 }
1391                 dm_sysfs_exit(md);
1392                 dm_table_put(map);
1393                 __unbind(md);
1394                 free_dev(md);
1395         }
1396 }
1397 EXPORT_SYMBOL_GPL(dm_put);
1398
1399 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1400 {
1401         int r = 0;
1402         DECLARE_WAITQUEUE(wait, current);
1403
1404         dm_unplug_all(md->queue);
1405
1406         add_wait_queue(&md->wait, &wait);
1407
1408         while (1) {
1409                 set_current_state(interruptible);
1410
1411                 smp_mb();
1412                 if (!atomic_read(&md->pending))
1413                         break;
1414
1415                 if (interruptible == TASK_INTERRUPTIBLE &&
1416                     signal_pending(current)) {
1417                         r = -EINTR;
1418                         break;
1419                 }
1420
1421                 io_schedule();
1422         }
1423         set_current_state(TASK_RUNNING);
1424
1425         remove_wait_queue(&md->wait, &wait);
1426
1427         return r;
1428 }
1429
1430 static int dm_flush(struct mapped_device *md)
1431 {
1432         dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
1433         return 0;
1434 }
1435
1436 static void process_barrier(struct mapped_device *md, struct bio *bio)
1437 {
1438         int error = dm_flush(md);
1439
1440         if (unlikely(error)) {
1441                 bio_endio(bio, error);
1442                 return;
1443         }
1444         if (bio_empty_barrier(bio)) {
1445                 bio_endio(bio, 0);
1446                 return;
1447         }
1448
1449         __split_and_process_bio(md, bio);
1450
1451         error = dm_flush(md);
1452
1453         if (!error && md->barrier_error)
1454                 error = md->barrier_error;
1455
1456         if (md->barrier_error != DM_ENDIO_REQUEUE)
1457                 bio_endio(bio, error);
1458 }
1459
1460 /*
1461  * Process the deferred bios
1462  */
1463 static void dm_wq_work(struct work_struct *work)
1464 {
1465         struct mapped_device *md = container_of(work, struct mapped_device,
1466                                                 work);
1467         struct bio *c;
1468
1469         down_write(&md->io_lock);
1470
1471         while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1472                 spin_lock_irq(&md->deferred_lock);
1473                 c = bio_list_pop(&md->deferred);
1474                 spin_unlock_irq(&md->deferred_lock);
1475
1476                 if (!c) {
1477                         clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1478                         break;
1479                 }
1480
1481                 up_write(&md->io_lock);
1482
1483                 if (bio_barrier(c))
1484                         process_barrier(md, c);
1485                 else
1486                         __split_and_process_bio(md, c);
1487
1488                 down_write(&md->io_lock);
1489         }
1490
1491         up_write(&md->io_lock);
1492 }
1493
1494 static void dm_queue_flush(struct mapped_device *md)
1495 {
1496         clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1497         smp_mb__after_clear_bit();
1498         queue_work(md->wq, &md->work);
1499 }
1500
1501 /*
1502  * Swap in a new table (destroying old one).
1503  */
1504 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1505 {
1506         int r = -EINVAL;
1507
1508         mutex_lock(&md->suspend_lock);
1509
1510         /* device must be suspended */
1511         if (!dm_suspended(md))
1512                 goto out;
1513
1514         /* without bdev, the device size cannot be changed */
1515         if (!md->suspended_bdev)
1516                 if (get_capacity(md->disk) != dm_table_get_size(table))
1517                         goto out;
1518
1519         __unbind(md);
1520         r = __bind(md, table);
1521
1522 out:
1523         mutex_unlock(&md->suspend_lock);
1524         return r;
1525 }
1526
1527 /*
1528  * Functions to lock and unlock any filesystem running on the
1529  * device.
1530  */
1531 static int lock_fs(struct mapped_device *md)
1532 {
1533         int r;
1534
1535         WARN_ON(md->frozen_sb);
1536
1537         md->frozen_sb = freeze_bdev(md->suspended_bdev);
1538         if (IS_ERR(md->frozen_sb)) {
1539                 r = PTR_ERR(md->frozen_sb);
1540                 md->frozen_sb = NULL;
1541                 return r;
1542         }
1543
1544         set_bit(DMF_FROZEN, &md->flags);
1545
1546         /* don't bdput right now, we don't want the bdev
1547          * to go away while it is locked.
1548          */
1549         return 0;
1550 }
1551
1552 static void unlock_fs(struct mapped_device *md)
1553 {
1554         if (!test_bit(DMF_FROZEN, &md->flags))
1555                 return;
1556
1557         thaw_bdev(md->suspended_bdev, md->frozen_sb);
1558         md->frozen_sb = NULL;
1559         clear_bit(DMF_FROZEN, &md->flags);
1560 }
1561
1562 /*
1563  * We need to be able to change a mapping table under a mounted
1564  * filesystem.  For example we might want to move some data in
1565  * the background.  Before the table can be swapped with
1566  * dm_bind_table, dm_suspend must be called to flush any in
1567  * flight bios and ensure that any further io gets deferred.
1568  */
1569 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1570 {
1571         struct dm_table *map = NULL;
1572         int r = 0;
1573         int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1574         int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1575
1576         mutex_lock(&md->suspend_lock);
1577
1578         if (dm_suspended(md)) {
1579                 r = -EINVAL;
1580                 goto out_unlock;
1581         }
1582
1583         map = dm_get_table(md);
1584
1585         /*
1586          * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1587          * This flag is cleared before dm_suspend returns.
1588          */
1589         if (noflush)
1590                 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1591
1592         /* This does not get reverted if there's an error later. */
1593         dm_table_presuspend_targets(map);
1594
1595         /* bdget() can stall if the pending I/Os are not flushed */
1596         if (!noflush) {
1597                 md->suspended_bdev = bdget_disk(md->disk, 0);
1598                 if (!md->suspended_bdev) {
1599                         DMWARN("bdget failed in dm_suspend");
1600                         r = -ENOMEM;
1601                         goto out;
1602                 }
1603
1604                 /*
1605                  * Flush I/O to the device. noflush supersedes do_lockfs,
1606                  * because lock_fs() needs to flush I/Os.
1607                  */
1608                 if (do_lockfs) {
1609                         r = lock_fs(md);
1610                         if (r)
1611                                 goto out;
1612                 }
1613         }
1614
1615         /*
1616          * Here we must make sure that no processes are submitting requests
1617          * to target drivers i.e. no one may be executing
1618          * __split_and_process_bio. This is called from dm_request and
1619          * dm_wq_work.
1620          *
1621          * To get all processes out of __split_and_process_bio in dm_request,
1622          * we take the write lock. To prevent any process from reentering
1623          * __split_and_process_bio from dm_request, we set
1624          * DMF_QUEUE_IO_TO_THREAD.
1625          *
1626          * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
1627          * and call flush_workqueue(md->wq). flush_workqueue will wait until
1628          * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
1629          * further calls to __split_and_process_bio from dm_wq_work.
1630          */
1631         down_write(&md->io_lock);
1632         set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1633         set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1634         up_write(&md->io_lock);
1635
1636         flush_workqueue(md->wq);
1637
1638         /*
1639          * At this point no more requests are entering target request routines.
1640          * We call dm_wait_for_completion to wait for all existing requests
1641          * to finish.
1642          */
1643         r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
1644
1645         down_write(&md->io_lock);
1646         if (noflush)
1647                 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1648         up_write(&md->io_lock);
1649
1650         /* were we interrupted ? */
1651         if (r < 0) {
1652                 dm_queue_flush(md);
1653
1654                 unlock_fs(md);
1655                 goto out; /* pushback list is already flushed, so skip flush */
1656         }
1657
1658         /*
1659          * If dm_wait_for_completion returned 0, the device is completely
1660          * quiescent now. There is no request-processing activity. All new
1661          * requests are being added to md->deferred list.
1662          */
1663
1664         dm_table_postsuspend_targets(map);
1665
1666         set_bit(DMF_SUSPENDED, &md->flags);
1667
1668 out:
1669         if (r && md->suspended_bdev) {
1670                 bdput(md->suspended_bdev);
1671                 md->suspended_bdev = NULL;
1672         }
1673
1674         dm_table_put(map);
1675
1676 out_unlock:
1677         mutex_unlock(&md->suspend_lock);
1678         return r;
1679 }
1680
1681 int dm_resume(struct mapped_device *md)
1682 {
1683         int r = -EINVAL;
1684         struct dm_table *map = NULL;
1685
1686         mutex_lock(&md->suspend_lock);
1687         if (!dm_suspended(md))
1688                 goto out;
1689
1690         map = dm_get_table(md);
1691         if (!map || !dm_table_get_size(map))
1692                 goto out;
1693
1694         r = dm_table_resume_targets(map);
1695         if (r)
1696                 goto out;
1697
1698         dm_queue_flush(md);
1699
1700         unlock_fs(md);
1701
1702         if (md->suspended_bdev) {
1703                 bdput(md->suspended_bdev);
1704                 md->suspended_bdev = NULL;
1705         }
1706
1707         clear_bit(DMF_SUSPENDED, &md->flags);
1708
1709         dm_table_unplug_all(map);
1710
1711         dm_kobject_uevent(md);
1712
1713         r = 0;
1714
1715 out:
1716         dm_table_put(map);
1717         mutex_unlock(&md->suspend_lock);
1718
1719         return r;
1720 }
1721
1722 /*-----------------------------------------------------------------
1723  * Event notification.
1724  *---------------------------------------------------------------*/
1725 void dm_kobject_uevent(struct mapped_device *md)
1726 {
1727         kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1728 }
1729
1730 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1731 {
1732         return atomic_add_return(1, &md->uevent_seq);
1733 }
1734
1735 uint32_t dm_get_event_nr(struct mapped_device *md)
1736 {
1737         return atomic_read(&md->event_nr);
1738 }
1739
1740 int dm_wait_event(struct mapped_device *md, int event_nr)
1741 {
1742         return wait_event_interruptible(md->eventq,
1743                         (event_nr != atomic_read(&md->event_nr)));
1744 }
1745
1746 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1747 {
1748         unsigned long flags;
1749
1750         spin_lock_irqsave(&md->uevent_lock, flags);
1751         list_add(elist, &md->uevent_list);
1752         spin_unlock_irqrestore(&md->uevent_lock, flags);
1753 }
1754
1755 /*
1756  * The gendisk is only valid as long as you have a reference
1757  * count on 'md'.
1758  */
1759 struct gendisk *dm_disk(struct mapped_device *md)
1760 {
1761         return md->disk;
1762 }
1763
1764 struct kobject *dm_kobject(struct mapped_device *md)
1765 {
1766         return &md->kobj;
1767 }
1768
1769 /*
1770  * struct mapped_device should not be exported outside of dm.c
1771  * so use this check to verify that kobj is part of md structure
1772  */
1773 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1774 {
1775         struct mapped_device *md;
1776
1777         md = container_of(kobj, struct mapped_device, kobj);
1778         if (&md->kobj != kobj)
1779                 return NULL;
1780
1781         dm_get(md);
1782         return md;
1783 }
1784
1785 int dm_suspended(struct mapped_device *md)
1786 {
1787         return test_bit(DMF_SUSPENDED, &md->flags);
1788 }
1789
1790 int dm_noflush_suspending(struct dm_target *ti)
1791 {
1792         struct mapped_device *md = dm_table_get_md(ti->table);
1793         int r = __noflush_suspending(md);
1794
1795         dm_put(md);
1796
1797         return r;
1798 }
1799 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1800
1801 static struct block_device_operations dm_blk_dops = {
1802         .open = dm_blk_open,
1803         .release = dm_blk_close,
1804         .ioctl = dm_blk_ioctl,
1805         .getgeo = dm_blk_getgeo,
1806         .owner = THIS_MODULE
1807 };
1808
1809 EXPORT_SYMBOL(dm_get_mapinfo);
1810
1811 /*
1812  * module hooks
1813  */
1814 module_init(dm_init);
1815 module_exit(dm_exit);
1816
1817 module_param(major, uint, 0);
1818 MODULE_PARM_DESC(major, "The major number of the device mapper");
1819 MODULE_DESCRIPTION(DM_NAME " driver");
1820 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1821 MODULE_LICENSE("GPL");