drivers/md/dm.c

   1 /*
   2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include "dm.h"
   9 #include "dm-bio-list.h"
  10
  11 #include <linux/init.h>
  12 #include <linux/module.h>
  13 #include <linux/moduleparam.h>
  14 #include <linux/blkpg.h>
  15 #include <linux/bio.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/mempool.h>
  18 #include <linux/slab.h>
  19 #include <linux/idr.h>
  20
  21 static const char *_name = DM_NAME;
  22
  23 static unsigned int major = 0;
  24 static unsigned int _major = 0;
  25
  26 /*
  27  * One of these is allocated per bio.
  28  */
  29 struct dm_io {
  30         struct mapped_device *md;
  31         int error;
  32         struct bio *bio;
  33         atomic_t io_count;
  34         unsigned long start_time;
  35 };
  36
  37 /*
  38  * One of these is allocated per target within a bio.  Hopefully
  39  * this will be simplified out one day.
  40  */
  41 struct target_io {
  42         struct dm_io *io;
  43         struct dm_target *ti;
  44         union map_info info;
  45 };
  46
  47 union map_info *dm_get_mapinfo(struct bio *bio)
  48 {
  49         if (bio && bio->bi_private)
  50                 return &((struct target_io *)bio->bi_private)->info;
  51         return NULL;
  52 }
  53
  54 /*
  55  * Bits for the md->flags field.
  56  */
  57 #define DMF_BLOCK_IO 0
  58 #define DMF_SUSPENDED 1
  59 #define DMF_FROZEN 2
  60
  61 struct mapped_device {
  62         struct rw_semaphore io_lock;
  63         struct semaphore suspend_lock;
  64         rwlock_t map_lock;
  65         atomic_t holders;
  66
  67         unsigned long flags;
  68
  69         request_queue_t *queue;
  70         struct gendisk *disk;
  71
  72         void *interface_ptr;
  73
  74         /*
  75          * A list of ios that arrived while we were suspended.
  76          */
  77         atomic_t pending;
  78         wait_queue_head_t wait;
  79         struct bio_list deferred;
  80
  81         /*
  82          * The current mapping.
  83          */
  84         struct dm_table *map;
  85
  86         /*
  87          * io objects are allocated from here.
  88          */
  89         mempool_t *io_pool;
  90         mempool_t *tio_pool;
  91
  92         /*
  93          * Event handling.
  94          */
  95         atomic_t event_nr;
  96         wait_queue_head_t eventq;
  97
  98         /*
  99          * freeze/thaw support require holding onto a super block
 100          */
 101         struct super_block *frozen_sb;
 102         struct block_device *suspended_bdev;
 103 };
 104
 105 #define MIN_IOS 256
 106 static kmem_cache_t *_io_cache;
 107 static kmem_cache_t *_tio_cache;
 108
 109 static struct bio_set *dm_set;
 110
 111 static int __init local_init(void)
 112 {
 113         int r;
 114
 115         dm_set = bioset_create(16, 16, 4);
 116         if (!dm_set)
 117                 return -ENOMEM;
 118
 119         /* allocate a slab for the dm_ios */
 120         _io_cache = kmem_cache_create("dm_io",
 121                                       sizeof(struct dm_io), 0, 0, NULL, NULL);
 122         if (!_io_cache)
 123                 return -ENOMEM;
 124
 125         /* allocate a slab for the target ios */
 126         _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
 127                                        0, 0, NULL, NULL);
 128         if (!_tio_cache) {
 129                 kmem_cache_destroy(_io_cache);
 130                 return -ENOMEM;
 131         }
 132
 133         _major = major;
 134         r = register_blkdev(_major, _name);
 135         if (r < 0) {
 136                 kmem_cache_destroy(_tio_cache);
 137                 kmem_cache_destroy(_io_cache);
 138                 return r;
 139         }
 140
 141         if (!_major)
 142                 _major = r;
 143
 144         return 0;
 145 }
 146
 147 static void local_exit(void)
 148 {
 149         kmem_cache_destroy(_tio_cache);
 150         kmem_cache_destroy(_io_cache);
 151
 152         bioset_free(dm_set);
 153
 154         if (unregister_blkdev(_major, _name) < 0)
 155                 DMERR("devfs_unregister_blkdev failed");
 156
 157         _major = 0;
 158
 159         DMINFO("cleaned up");
 160 }
 161
 162 int (*_inits[])(void) __initdata = {
 163         local_init,
 164         dm_target_init,
 165         dm_linear_init,
 166         dm_stripe_init,
 167         dm_interface_init,
 168 };
 169
 170 void (*_exits[])(void) = {
 171         local_exit,
 172         dm_target_exit,
 173         dm_linear_exit,
 174         dm_stripe_exit,
 175         dm_interface_exit,
 176 };
 177
 178 static int __init dm_init(void)
 179 {
 180         const int count = ARRAY_SIZE(_inits);
 181
 182         int r, i;
 183
 184         for (i = 0; i < count; i++) {
 185                 r = _inits[i]();
 186                 if (r)
 187                         goto bad;
 188         }
 189
 190         return 0;
 191
 192       bad:
 193         while (i--)
 194                 _exits[i]();
 195
 196         return r;
 197 }
 198
 199 static void __exit dm_exit(void)
 200 {
 201         int i = ARRAY_SIZE(_exits);
 202
 203         while (i--)
 204                 _exits[i]();
 205 }
 206
 207 /*
 208  * Block device functions
 209  */
 210 static int dm_blk_open(struct inode *inode, struct file *file)
 211 {
 212         struct mapped_device *md;
 213
 214         md = inode->i_bdev->bd_disk->private_data;
 215         dm_get(md);
 216         return 0;
 217 }
 218
 219 static int dm_blk_close(struct inode *inode, struct file *file)
 220 {
 221         struct mapped_device *md;
 222
 223         md = inode->i_bdev->bd_disk->private_data;
 224         dm_put(md);
 225         return 0;
 226 }
 227
 228 static inline struct dm_io *alloc_io(struct mapped_device *md)
 229 {
 230         return mempool_alloc(md->io_pool, GFP_NOIO);
 231 }
 232
 233 static inline void free_io(struct mapped_device *md, struct dm_io *io)
 234 {
 235         mempool_free(io, md->io_pool);
 236 }
 237
 238 static inline struct target_io *alloc_tio(struct mapped_device *md)
 239 {
 240         return mempool_alloc(md->tio_pool, GFP_NOIO);
 241 }
 242
 243 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
 244 {
 245         mempool_free(tio, md->tio_pool);
 246 }
 247
 248 static void start_io_acct(struct dm_io *io)
 249 {
 250         struct mapped_device *md = io->md;
 251
 252         io->start_time = jiffies;
 253
 254         preempt_disable();
 255         disk_round_stats(dm_disk(md));
 256         preempt_enable();
 257         dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
 258 }
 259
 260 static int end_io_acct(struct dm_io *io)
 261 {
 262         struct mapped_device *md = io->md;
 263         struct bio *bio = io->bio;
 264         unsigned long duration = jiffies - io->start_time;
 265         int pending;
 266         int rw = bio_data_dir(bio);
 267
 268         preempt_disable();
 269         disk_round_stats(dm_disk(md));
 270         preempt_enable();
 271         dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
 272
 273         disk_stat_add(dm_disk(md), ticks[rw], duration);
 274
 275         return !pending;
 276 }
 277
 278 /*
 279  * Add the bio to the list of deferred io.
 280  */
 281 static int queue_io(struct mapped_device *md, struct bio *bio)
 282 {
 283         down_write(&md->io_lock);
 284
 285         if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
 286                 up_write(&md->io_lock);
 287                 return 1;
 288         }
 289
 290         bio_list_add(&md->deferred, bio);
 291
 292         up_write(&md->io_lock);
 293         return 0;               /* deferred successfully */
 294 }
 295
 296 /*
 297  * Everyone (including functions in this file), should use this
 298  * function to access the md->map field, and make sure they call
 299  * dm_table_put() when finished.
 300  */
 301 struct dm_table *dm_get_table(struct mapped_device *md)
 302 {
 303         struct dm_table *t;
 304
 305         read_lock(&md->map_lock);
 306         t = md->map;
 307         if (t)
 308                 dm_table_get(t);
 309         read_unlock(&md->map_lock);
 310
 311         return t;
 312 }
 313
 314 /*-----------------------------------------------------------------
 315  * CRUD START:
 316  *   A more elegant soln is in the works that uses the queue
 317  *   merge fn, unfortunately there are a couple of changes to
 318  *   the block layer that I want to make for this.  So in the
 319  *   interests of getting something for people to use I give
 320  *   you this clearly demarcated crap.
 321  *---------------------------------------------------------------*/
 322
 323 /*
 324  * Decrements the number of outstanding ios that a bio has been
 325  * cloned into, completing the original io if necc.
 326  */
 327 static void dec_pending(struct dm_io *io, int error)
 328 {
 329         if (error)
 330                 io->error = error;
 331
 332         if (atomic_dec_and_test(&io->io_count)) {
 333                 if (end_io_acct(io))
 334                         /* nudge anyone waiting on suspend queue */
 335                         wake_up(&io->md->wait);
 336
 337                 bio_endio(io->bio, io->bio->bi_size, io->error);
 338                 free_io(io->md, io);
 339         }
 340 }
 341
 342 static int clone_endio(struct bio *bio, unsigned int done, int error)
 343 {
 344         int r = 0;
 345         struct target_io *tio = bio->bi_private;
 346         struct dm_io *io = tio->io;
 347         dm_endio_fn endio = tio->ti->type->end_io;
 348
 349         if (bio->bi_size)
 350                 return 1;
 351
 352         if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 353                 error = -EIO;
 354
 355         if (endio) {
 356                 r = endio(tio->ti, bio, error, &tio->info);
 357                 if (r < 0)
 358                         error = r;
 359
 360                 else if (r > 0)
 361                         /* the target wants another shot at the io */
 362                         return 1;
 363         }
 364
 365         free_tio(io->md, tio);
 366         dec_pending(io, error);
 367         bio_put(bio);
 368         return r;
 369 }
 370
 371 static sector_t max_io_len(struct mapped_device *md,
 372                            sector_t sector, struct dm_target *ti)
 373 {
 374         sector_t offset = sector - ti->begin;
 375         sector_t len = ti->len - offset;
 376
 377         /*
 378          * Does the target need to split even further ?
 379          */
 380         if (ti->split_io) {
 381                 sector_t boundary;
 382                 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
 383                            - offset;
 384                 if (len > boundary)
 385                         len = boundary;
 386         }
 387
 388         return len;
 389 }
 390
 391 static void __map_bio(struct dm_target *ti, struct bio *clone,
 392                       struct target_io *tio)
 393 {
 394         int r;
 395
 396         /*
 397          * Sanity checks.
 398          */
 399         BUG_ON(!clone->bi_size);
 400
 401         clone->bi_end_io = clone_endio;
 402         clone->bi_private = tio;
 403
 404         /*
 405          * Map the clone.  If r == 0 we don't need to do
 406          * anything, the target has assumed ownership of
 407          * this io.
 408          */
 409         atomic_inc(&tio->io->io_count);
 410         r = ti->type->map(ti, clone, &tio->info);
 411         if (r > 0)
 412                 /* the bio has been remapped so dispatch it */
 413                 generic_make_request(clone);
 414
 415         else if (r < 0) {
 416                 /* error the io and bail out */
 417                 struct dm_io *io = tio->io;
 418                 free_tio(tio->io->md, tio);
 419                 dec_pending(io, r);
 420                 bio_put(clone);
 421         }
 422 }
 423
 424 struct clone_info {
 425         struct mapped_device *md;
 426         struct dm_table *map;
 427         struct bio *bio;
 428         struct dm_io *io;
 429         sector_t sector;
 430         sector_t sector_count;
 431         unsigned short idx;
 432 };
 433
 434 static void dm_bio_destructor(struct bio *bio)
 435 {
 436         bio_free(bio, dm_set);
 437 }
 438
 439 /*
 440  * Creates a little bio that is just does part of a bvec.
 441  */
 442 static struct bio *split_bvec(struct bio *bio, sector_t sector,
 443                               unsigned short idx, unsigned int offset,
 444                               unsigned int len)
 445 {
 446         struct bio *clone;
 447         struct bio_vec *bv = bio->bi_io_vec + idx;
 448
 449         clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set);
 450         clone->bi_destructor = dm_bio_destructor;
 451         *clone->bi_io_vec = *bv;
 452
 453         clone->bi_sector = sector;
 454         clone->bi_bdev = bio->bi_bdev;
 455         clone->bi_rw = bio->bi_rw;
 456         clone->bi_vcnt = 1;
 457         clone->bi_size = to_bytes(len);
 458         clone->bi_io_vec->bv_offset = offset;
 459         clone->bi_io_vec->bv_len = clone->bi_size;
 460
 461         return clone;
 462 }
 463
 464 /*
 465  * Creates a bio that consists of range of complete bvecs.
 466  */
 467 static struct bio *clone_bio(struct bio *bio, sector_t sector,
 468                              unsigned short idx, unsigned short bv_count,
 469                              unsigned int len)
 470 {
 471         struct bio *clone;
 472
 473         clone = bio_clone(bio, GFP_NOIO);
 474         clone->bi_sector = sector;
 475         clone->bi_idx = idx;
 476         clone->bi_vcnt = idx + bv_count;
 477         clone->bi_size = to_bytes(len);
 478         clone->bi_flags &= ~(1 << BIO_SEG_VALID);
 479
 480         return clone;
 481 }
 482
 483 static void __clone_and_map(struct clone_info *ci)
 484 {
 485         struct bio *clone, *bio = ci->bio;
 486         struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
 487         sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
 488         struct target_io *tio;
 489
 490         /*
 491          * Allocate a target io object.
 492          */
 493         tio = alloc_tio(ci->md);
 494         tio->io = ci->io;
 495         tio->ti = ti;
 496         memset(&tio->info, 0, sizeof(tio->info));
 497
 498         if (ci->sector_count <= max) {
 499                 /*
 500                  * Optimise for the simple case where we can do all of
 501                  * the remaining io with a single clone.
 502                  */
 503                 clone = clone_bio(bio, ci->sector, ci->idx,
 504                                   bio->bi_vcnt - ci->idx, ci->sector_count);
 505                 __map_bio(ti, clone, tio);
 506                 ci->sector_count = 0;
 507
 508         } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
 509                 /*
 510                  * There are some bvecs that don't span targets.
 511                  * Do as many of these as possible.
 512                  */
 513                 int i;
 514                 sector_t remaining = max;
 515                 sector_t bv_len;
 516
 517                 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
 518                         bv_len = to_sector(bio->bi_io_vec[i].bv_len);
 519
 520                         if (bv_len > remaining)
 521                                 break;
 522
 523                         remaining -= bv_len;
 524                         len += bv_len;
 525                 }
 526
 527                 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len);
 528                 __map_bio(ti, clone, tio);
 529
 530                 ci->sector += len;
 531                 ci->sector_count -= len;
 532                 ci->idx = i;
 533
 534         } else {
 535                 /*
 536                  * Handle a bvec that must be split between two or more targets.
 537                  */
 538                 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
 539                 sector_t remaining = to_sector(bv->bv_len);
 540                 unsigned int offset = 0;
 541
 542                 do {
 543                         if (offset) {
 544                                 ti = dm_table_find_target(ci->map, ci->sector);
 545                                 max = max_io_len(ci->md, ci->sector, ti);
 546
 547                                 tio = alloc_tio(ci->md);
 548                                 tio->io = ci->io;
 549                                 tio->ti = ti;
 550                                 memset(&tio->info, 0, sizeof(tio->info));
 551                         }
 552
 553                         len = min(remaining, max);
 554
 555                         clone = split_bvec(bio, ci->sector, ci->idx,
 556                                            bv->bv_offset + offset, len);
 557
 558                         __map_bio(ti, clone, tio);
 559
 560                         ci->sector += len;
 561                         ci->sector_count -= len;
 562                         offset += to_bytes(len);
 563                 } while (remaining -= len);
 564
 565                 ci->idx++;
 566         }
 567 }
 568
 569 /*
 570  * Split the bio into several clones.
 571  */
 572 static void __split_bio(struct mapped_device *md, struct bio *bio)
 573 {
 574         struct clone_info ci;
 575
 576         ci.map = dm_get_table(md);
 577         if (!ci.map) {
 578                 bio_io_error(bio, bio->bi_size);
 579                 return;
 580         }
 581
 582         ci.md = md;
 583         ci.bio = bio;
 584         ci.io = alloc_io(md);
 585         ci.io->error = 0;
 586         atomic_set(&ci.io->io_count, 1);
 587         ci.io->bio = bio;
 588         ci.io->md = md;
 589         ci.sector = bio->bi_sector;
 590         ci.sector_count = bio_sectors(bio);
 591         ci.idx = bio->bi_idx;
 592
 593         start_io_acct(ci.io);
 594         while (ci.sector_count)
 595                 __clone_and_map(&ci);
 596
 597         /* drop the extra reference count */
 598         dec_pending(ci.io, 0);
 599         dm_table_put(ci.map);
 600 }
 601 /*-----------------------------------------------------------------
 602  * CRUD END
 603  *---------------------------------------------------------------*/
 604
 605 /*
 606  * The request function that just remaps the bio built up by
 607  * dm_merge_bvec.
 608  */
 609 static int dm_request(request_queue_t *q, struct bio *bio)
 610 {
 611         int r;
 612         int rw = bio_data_dir(bio);
 613         struct mapped_device *md = q->queuedata;
 614
 615         down_read(&md->io_lock);
 616
 617         disk_stat_inc(dm_disk(md), ios[rw]);
 618         disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
 619
 620         /*
 621          * If we're suspended we have to queue
 622          * this io for later.
 623          */
 624         while (test_bit(DMF_BLOCK_IO, &md->flags)) {
 625                 up_read(&md->io_lock);
 626
 627                 if (bio_rw(bio) == READA) {
 628                         bio_io_error(bio, bio->bi_size);
 629                         return 0;
 630                 }
 631
 632                 r = queue_io(md, bio);
 633                 if (r < 0) {
 634                         bio_io_error(bio, bio->bi_size);
 635                         return 0;
 636
 637                 } else if (r == 0)
 638                         return 0;       /* deferred successfully */
 639
 640                 /*
 641                  * We're in a while loop, because someone could suspend
 642                  * before we get to the following read lock.
 643                  */
 644                 down_read(&md->io_lock);
 645         }
 646
 647         __split_bio(md, bio);
 648         up_read(&md->io_lock);
 649         return 0;
 650 }
 651
 652 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
 653                         sector_t *error_sector)
 654 {
 655         struct mapped_device *md = q->queuedata;
 656         struct dm_table *map = dm_get_table(md);
 657         int ret = -ENXIO;
 658
 659         if (map) {
 660                 ret = dm_table_flush_all(map);
 661                 dm_table_put(map);
 662         }
 663
 664         return ret;
 665 }
 666
 667 static void dm_unplug_all(request_queue_t *q)
 668 {
 669         struct mapped_device *md = q->queuedata;
 670         struct dm_table *map = dm_get_table(md);
 671
 672         if (map) {
 673                 dm_table_unplug_all(map);
 674                 dm_table_put(map);
 675         }
 676 }
 677
 678 static int dm_any_congested(void *congested_data, int bdi_bits)
 679 {
 680         int r;
 681         struct mapped_device *md = (struct mapped_device *) congested_data;
 682         struct dm_table *map = dm_get_table(md);
 683
 684         if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
 685                 r = bdi_bits;
 686         else
 687                 r = dm_table_any_congested(map, bdi_bits);
 688
 689         dm_table_put(map);
 690         return r;
 691 }
 692
 693 /*-----------------------------------------------------------------
 694  * An IDR is used to keep track of allocated minor numbers.
 695  *---------------------------------------------------------------*/
 696 static DECLARE_MUTEX(_minor_lock);
 697 static DEFINE_IDR(_minor_idr);
 698
 699 static void free_minor(unsigned int minor)
 700 {
 701         down(&_minor_lock);
 702         idr_remove(&_minor_idr, minor);
 703         up(&_minor_lock);
 704 }
 705
 706 /*
 707  * See if the device with a specific minor # is free.
 708  */
 709 static int specific_minor(struct mapped_device *md, unsigned int minor)
 710 {
 711         int r, m;
 712
 713         if (minor >= (1 << MINORBITS))
 714                 return -EINVAL;
 715
 716         down(&_minor_lock);
 717
 718         if (idr_find(&_minor_idr, minor)) {
 719                 r = -EBUSY;
 720                 goto out;
 721         }
 722
 723         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
 724         if (!r) {
 725                 r = -ENOMEM;
 726                 goto out;
 727         }
 728
 729         r = idr_get_new_above(&_minor_idr, md, minor, &m);
 730         if (r) {
 731                 goto out;
 732         }
 733
 734         if (m != minor) {
 735                 idr_remove(&_minor_idr, m);
 736                 r = -EBUSY;
 737                 goto out;
 738         }
 739
 740 out:
 741         up(&_minor_lock);
 742         return r;
 743 }
 744
 745 static int next_free_minor(struct mapped_device *md, unsigned int *minor)
 746 {
 747         int r;
 748         unsigned int m;
 749
 750         down(&_minor_lock);
 751
 752         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
 753         if (!r) {
 754                 r = -ENOMEM;
 755                 goto out;
 756         }
 757
 758         r = idr_get_new(&_minor_idr, md, &m);
 759         if (r) {
 760                 goto out;
 761         }
 762
 763         if (m >= (1 << MINORBITS)) {
 764                 idr_remove(&_minor_idr, m);
 765                 r = -ENOSPC;
 766                 goto out;
 767         }
 768
 769         *minor = m;
 770
 771 out:
 772         up(&_minor_lock);
 773         return r;
 774 }
 775
 776 static struct block_device_operations dm_blk_dops;
 777
 778 /*
 779  * Allocate and initialise a blank device with a given minor.
 780  */
 781 static struct mapped_device *alloc_dev(unsigned int minor, int persistent)
 782 {
 783         int r;
 784         struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
 785
 786         if (!md) {
 787                 DMWARN("unable to allocate device, out of memory.");
 788                 return NULL;
 789         }
 790
 791         if (!try_module_get(THIS_MODULE))
 792                 goto bad0;
 793
 794         /* get a minor number for the dev */
 795         r = persistent ? specific_minor(md, minor) : next_free_minor(md, &minor);
 796         if (r < 0)
 797                 goto bad1;
 798
 799         memset(md, 0, sizeof(*md));
 800         init_rwsem(&md->io_lock);
 801         init_MUTEX(&md->suspend_lock);
 802         rwlock_init(&md->map_lock);
 803         atomic_set(&md->holders, 1);
 804         atomic_set(&md->event_nr, 0);
 805
 806         md->queue = blk_alloc_queue(GFP_KERNEL);
 807         if (!md->queue)
 808                 goto bad1;
 809
 810         md->queue->queuedata = md;
 811         md->queue->backing_dev_info.congested_fn = dm_any_congested;
 812         md->queue->backing_dev_info.congested_data = md;
 813         blk_queue_make_request(md->queue, dm_request);
 814         blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
 815         md->queue->unplug_fn = dm_unplug_all;
 816         md->queue->issue_flush_fn = dm_flush_all;
 817
 818         md->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab,
 819                                      mempool_free_slab, _io_cache);
 820         if (!md->io_pool)
 821                 goto bad2;
 822
 823         md->tio_pool = mempool_create(MIN_IOS, mempool_alloc_slab,
 824                                       mempool_free_slab, _tio_cache);
 825         if (!md->tio_pool)
 826                 goto bad3;
 827
 828         md->disk = alloc_disk(1);
 829         if (!md->disk)
 830                 goto bad4;
 831
 832         md->disk->major = _major;
 833         md->disk->first_minor = minor;
 834         md->disk->fops = &dm_blk_dops;
 835         md->disk->queue = md->queue;
 836         md->disk->private_data = md;
 837         sprintf(md->disk->disk_name, "dm-%d", minor);
 838         add_disk(md->disk);
 839
 840         atomic_set(&md->pending, 0);
 841         init_waitqueue_head(&md->wait);
 842         init_waitqueue_head(&md->eventq);
 843
 844         return md;
 845
 846  bad4:
 847         mempool_destroy(md->tio_pool);
 848  bad3:
 849         mempool_destroy(md->io_pool);
 850  bad2:
 851         blk_put_queue(md->queue);
 852         free_minor(minor);
 853  bad1:
 854         module_put(THIS_MODULE);
 855  bad0:
 856         kfree(md);
 857         return NULL;
 858 }
 859
 860 static void free_dev(struct mapped_device *md)
 861 {
 862         unsigned int minor = md->disk->first_minor;
 863
 864         if (md->suspended_bdev) {
 865                 thaw_bdev(md->suspended_bdev, NULL);
 866                 bdput(md->suspended_bdev);
 867         }
 868         mempool_destroy(md->tio_pool);
 869         mempool_destroy(md->io_pool);
 870         del_gendisk(md->disk);
 871         free_minor(minor);
 872         put_disk(md->disk);
 873         blk_put_queue(md->queue);
 874         module_put(THIS_MODULE);
 875         kfree(md);
 876 }
 877
 878 /*
 879  * Bind a table to the device.
 880  */
 881 static void event_callback(void *context)
 882 {
 883         struct mapped_device *md = (struct mapped_device *) context;
 884
 885         atomic_inc(&md->event_nr);
 886         wake_up(&md->eventq);
 887 }
 888
 889 static void __set_size(struct mapped_device *md, sector_t size)
 890 {
 891         set_capacity(md->disk, size);
 892
 893         mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
 894         i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
 895         mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
 896 }
 897
 898 static int __bind(struct mapped_device *md, struct dm_table *t)
 899 {
 900         request_queue_t *q = md->queue;
 901         sector_t size;
 902
 903         size = dm_table_get_size(t);
 904         __set_size(md, size);
 905         if (size == 0)
 906                 return 0;
 907
 908         dm_table_get(t);
 909         dm_table_event_callback(t, event_callback, md);
 910
 911         write_lock(&md->map_lock);
 912         md->map = t;
 913         dm_table_set_restrictions(t, q);
 914         write_unlock(&md->map_lock);
 915
 916         return 0;
 917 }
 918
 919 static void __unbind(struct mapped_device *md)
 920 {
 921         struct dm_table *map = md->map;
 922
 923         if (!map)
 924                 return;
 925
 926         dm_table_event_callback(map, NULL, NULL);
 927         write_lock(&md->map_lock);
 928         md->map = NULL;
 929         write_unlock(&md->map_lock);
 930         dm_table_put(map);
 931 }
 932
 933 /*
 934  * Constructor for a new device.
 935  */
 936 static int create_aux(unsigned int minor, int persistent,
 937                       struct mapped_device **result)
 938 {
 939         struct mapped_device *md;
 940
 941         md = alloc_dev(minor, persistent);
 942         if (!md)
 943                 return -ENXIO;
 944
 945         *result = md;
 946         return 0;
 947 }
 948
 949 int dm_create(struct mapped_device **result)
 950 {
 951         return create_aux(0, 0, result);
 952 }
 953
 954 int dm_create_with_minor(unsigned int minor, struct mapped_device **result)
 955 {
 956         return create_aux(minor, 1, result);
 957 }
 958
 959 static struct mapped_device *dm_find_md(dev_t dev)
 960 {
 961         struct mapped_device *md;
 962         unsigned minor = MINOR(dev);
 963
 964         if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
 965                 return NULL;
 966
 967         down(&_minor_lock);
 968
 969         md = idr_find(&_minor_idr, minor);
 970         if (!md || (dm_disk(md)->first_minor != minor))
 971                 md = NULL;
 972
 973         up(&_minor_lock);
 974
 975         return md;
 976 }
 977
 978 struct mapped_device *dm_get_md(dev_t dev)
 979 {
 980         struct mapped_device *md = dm_find_md(dev);
 981
 982         if (md)
 983                 dm_get(md);
 984
 985         return md;
 986 }
 987
 988 void *dm_get_mdptr(dev_t dev)
 989 {
 990         struct mapped_device *md;
 991         void *mdptr = NULL;
 992
 993         md = dm_find_md(dev);
 994         if (md)
 995                 mdptr = md->interface_ptr;
 996         return mdptr;
 997 }
 998
 999 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1000 {
1001         md->interface_ptr = ptr;
1002 }
1003
1004 void dm_get(struct mapped_device *md)
1005 {
1006         atomic_inc(&md->holders);
1007 }
1008
1009 void dm_put(struct mapped_device *md)
1010 {
1011         struct dm_table *map = dm_get_table(md);
1012
1013         if (atomic_dec_and_test(&md->holders)) {
1014                 if (!dm_suspended(md)) {
1015                         dm_table_presuspend_targets(map);
1016                         dm_table_postsuspend_targets(map);
1017                 }
1018                 __unbind(md);
1019                 free_dev(md);
1020         }
1021
1022         dm_table_put(map);
1023 }
1024
1025 /*
1026  * Process the deferred bios
1027  */
1028 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1029 {
1030         struct bio *n;
1031
1032         while (c) {
1033                 n = c->bi_next;
1034                 c->bi_next = NULL;
1035                 __split_bio(md, c);
1036                 c = n;
1037         }
1038 }
1039
1040 /*
1041  * Swap in a new table (destroying old one).
1042  */
1043 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1044 {
1045         int r = -EINVAL;
1046
1047         down(&md->suspend_lock);
1048
1049         /* device must be suspended */
1050         if (!dm_suspended(md))
1051                 goto out;
1052
1053         __unbind(md);
1054         r = __bind(md, table);
1055
1056 out:
1057         up(&md->suspend_lock);
1058         return r;
1059 }
1060
1061 /*
1062  * Functions to lock and unlock any filesystem running on the
1063  * device.
1064  */
1065 static int lock_fs(struct mapped_device *md)
1066 {
1067         int r;
1068
1069         WARN_ON(md->frozen_sb);
1070
1071         md->frozen_sb = freeze_bdev(md->suspended_bdev);
1072         if (IS_ERR(md->frozen_sb)) {
1073                 r = PTR_ERR(md->frozen_sb);
1074                 md->frozen_sb = NULL;
1075                 return r;
1076         }
1077
1078         set_bit(DMF_FROZEN, &md->flags);
1079
1080         /* don't bdput right now, we don't want the bdev
1081          * to go away while it is locked.
1082          */
1083         return 0;
1084 }
1085
1086 static void unlock_fs(struct mapped_device *md)
1087 {
1088         if (!test_bit(DMF_FROZEN, &md->flags))
1089                 return;
1090
1091         thaw_bdev(md->suspended_bdev, md->frozen_sb);
1092         md->frozen_sb = NULL;
1093         clear_bit(DMF_FROZEN, &md->flags);
1094 }
1095
1096 /*
1097  * We need to be able to change a mapping table under a mounted
1098  * filesystem.  For example we might want to move some data in
1099  * the background.  Before the table can be swapped with
1100  * dm_bind_table, dm_suspend must be called to flush any in
1101  * flight bios and ensure that any further io gets deferred.
1102  */
1103 int dm_suspend(struct mapped_device *md, int do_lockfs)
1104 {
1105         struct dm_table *map = NULL;
1106         DECLARE_WAITQUEUE(wait, current);
1107         struct bio *def;
1108         int r = -EINVAL;
1109
1110         down(&md->suspend_lock);
1111
1112         if (dm_suspended(md))
1113                 goto out;
1114
1115         map = dm_get_table(md);
1116
1117         /* This does not get reverted if there's an error later. */
1118         dm_table_presuspend_targets(map);
1119
1120         md->suspended_bdev = bdget_disk(md->disk, 0);
1121         if (!md->suspended_bdev) {
1122                 DMWARN("bdget failed in dm_suspend");
1123                 r = -ENOMEM;
1124                 goto out;
1125         }
1126
1127         /* Flush I/O to the device. */
1128         if (do_lockfs) {
1129                 r = lock_fs(md);
1130                 if (r)
1131                         goto out;
1132         }
1133
1134         /*
1135          * First we set the BLOCK_IO flag so no more ios will be mapped.
1136          */
1137         down_write(&md->io_lock);
1138         set_bit(DMF_BLOCK_IO, &md->flags);
1139
1140         add_wait_queue(&md->wait, &wait);
1141         up_write(&md->io_lock);
1142
1143         /* unplug */
1144         if (map)
1145                 dm_table_unplug_all(map);
1146
1147         /*
1148          * Then we wait for the already mapped ios to
1149          * complete.
1150          */
1151         while (1) {
1152                 set_current_state(TASK_INTERRUPTIBLE);
1153
1154                 if (!atomic_read(&md->pending) || signal_pending(current))
1155                         break;
1156
1157                 io_schedule();
1158         }
1159         set_current_state(TASK_RUNNING);
1160
1161         down_write(&md->io_lock);
1162         remove_wait_queue(&md->wait, &wait);
1163
1164         /* were we interrupted ? */
1165         r = -EINTR;
1166         if (atomic_read(&md->pending)) {
1167                 clear_bit(DMF_BLOCK_IO, &md->flags);
1168                 def = bio_list_get(&md->deferred);
1169                 __flush_deferred_io(md, def);
1170                 up_write(&md->io_lock);
1171                 unlock_fs(md);
1172                 goto out;
1173         }
1174         up_write(&md->io_lock);
1175
1176         dm_table_postsuspend_targets(map);
1177
1178         set_bit(DMF_SUSPENDED, &md->flags);
1179
1180         r = 0;
1181
1182 out:
1183         if (r && md->suspended_bdev) {
1184                 bdput(md->suspended_bdev);
1185                 md->suspended_bdev = NULL;
1186         }
1187
1188         dm_table_put(map);
1189         up(&md->suspend_lock);
1190         return r;
1191 }
1192
1193 int dm_resume(struct mapped_device *md)
1194 {
1195         int r = -EINVAL;
1196         struct bio *def;
1197         struct dm_table *map = NULL;
1198
1199         down(&md->suspend_lock);
1200         if (!dm_suspended(md))
1201                 goto out;
1202
1203         map = dm_get_table(md);
1204         if (!map || !dm_table_get_size(map))
1205                 goto out;
1206
1207         dm_table_resume_targets(map);
1208
1209         down_write(&md->io_lock);
1210         clear_bit(DMF_BLOCK_IO, &md->flags);
1211
1212         def = bio_list_get(&md->deferred);
1213         __flush_deferred_io(md, def);
1214         up_write(&md->io_lock);
1215
1216         unlock_fs(md);
1217
1218         bdput(md->suspended_bdev);
1219         md->suspended_bdev = NULL;
1220
1221         clear_bit(DMF_SUSPENDED, &md->flags);
1222
1223         dm_table_unplug_all(map);
1224
1225         r = 0;
1226
1227 out:
1228         dm_table_put(map);
1229         up(&md->suspend_lock);
1230
1231         return r;
1232 }
1233
1234 /*-----------------------------------------------------------------
1235  * Event notification.
1236  *---------------------------------------------------------------*/
1237 uint32_t dm_get_event_nr(struct mapped_device *md)
1238 {
1239         return atomic_read(&md->event_nr);
1240 }
1241
1242 int dm_wait_event(struct mapped_device *md, int event_nr)
1243 {
1244         return wait_event_interruptible(md->eventq,
1245                         (event_nr != atomic_read(&md->event_nr)));
1246 }
1247
1248 /*
1249  * The gendisk is only valid as long as you have a reference
1250  * count on 'md'.
1251  */
1252 struct gendisk *dm_disk(struct mapped_device *md)
1253 {
1254         return md->disk;
1255 }
1256
1257 int dm_suspended(struct mapped_device *md)
1258 {
1259         return test_bit(DMF_SUSPENDED, &md->flags);
1260 }
1261
1262 static struct block_device_operations dm_blk_dops = {
1263         .open = dm_blk_open,
1264         .release = dm_blk_close,
1265         .owner = THIS_MODULE
1266 };
1267
1268 EXPORT_SYMBOL(dm_get_mapinfo);
1269
1270 /*
1271  * module hooks
1272  */
1273 module_init(dm_init);
1274 module_exit(dm_exit);
1275
1276 module_param(major, uint, 0);
1277 MODULE_PARM_DESC(major, "The major number of the device mapper");
1278 MODULE_DESCRIPTION(DM_NAME " driver");
1279 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1280 MODULE_LICENSE("GPL");