drivers/md/dm.c

   1 /*
   2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004-2006 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/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 #define DM_MSG_PREFIX "core"
  25
  26 static const char *_name = DM_NAME;
  27
  28 static unsigned int major = 0;
  29 static unsigned int _major = 0;
  30
  31 static DEFINE_SPINLOCK(_minor_lock);
  32 /*
  33  * One of these is allocated per bio.
  34  */
  35 struct dm_io {
  36         struct mapped_device *md;
  37         int error;
  38         struct bio *bio;
  39         atomic_t io_count;
  40         unsigned long start_time;
  41 };
  42
  43 /*
  44  * One of these is allocated per target within a bio.  Hopefully
  45  * this will be simplified out one day.
  46  */
  47 struct target_io {
  48         struct dm_io *io;
  49         struct dm_target *ti;
  50         union map_info info;
  51 };
  52
  53 union map_info *dm_get_mapinfo(struct bio *bio)
  54 {
  55         if (bio && bio->bi_private)
  56                 return &((struct target_io *)bio->bi_private)->info;
  57         return NULL;
  58 }
  59
  60 #define MINOR_ALLOCED ((void *)-1)
  61
  62 /*
  63  * Bits for the md->flags field.
  64  */
  65 #define DMF_BLOCK_IO 0
  66 #define DMF_SUSPENDED 1
  67 #define DMF_FROZEN 2
  68 #define DMF_FREEING 3
  69 #define DMF_DELETING 4
  70
  71 struct mapped_device {
  72         struct rw_semaphore io_lock;
  73         struct semaphore suspend_lock;
  74         rwlock_t map_lock;
  75         atomic_t holders;
  76         atomic_t open_count;
  77
  78         unsigned long flags;
  79
  80         request_queue_t *queue;
  81         struct gendisk *disk;
  82         char name[16];
  83
  84         void *interface_ptr;
  85
  86         /*
  87          * A list of ios that arrived while we were suspended.
  88          */
  89         atomic_t pending;
  90         wait_queue_head_t wait;
  91         struct bio_list deferred;
  92
  93         /*
  94          * The current mapping.
  95          */
  96         struct dm_table *map;
  97
  98         /*
  99          * io objects are allocated from here.
 100          */
 101         mempool_t *io_pool;
 102         mempool_t *tio_pool;
 103
 104         /*
 105          * Event handling.
 106          */
 107         atomic_t event_nr;
 108         wait_queue_head_t eventq;
 109
 110         /*
 111          * freeze/thaw support require holding onto a super block
 112          */
 113         struct super_block *frozen_sb;
 114         struct block_device *suspended_bdev;
 115
 116         /* forced geometry settings */
 117         struct hd_geometry geometry;
 118 };
 119
 120 #define MIN_IOS 256
 121 static kmem_cache_t *_io_cache;
 122 static kmem_cache_t *_tio_cache;
 123
 124 static struct bio_set *dm_set;
 125
 126 static int __init local_init(void)
 127 {
 128         int r;
 129
 130         dm_set = bioset_create(16, 16, 4);
 131         if (!dm_set)
 132                 return -ENOMEM;
 133
 134         /* allocate a slab for the dm_ios */
 135         _io_cache = kmem_cache_create("dm_io",
 136                                       sizeof(struct dm_io), 0, 0, NULL, NULL);
 137         if (!_io_cache)
 138                 return -ENOMEM;
 139
 140         /* allocate a slab for the target ios */
 141         _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
 142                                        0, 0, NULL, NULL);
 143         if (!_tio_cache) {
 144                 kmem_cache_destroy(_io_cache);
 145                 return -ENOMEM;
 146         }
 147
 148         _major = major;
 149         r = register_blkdev(_major, _name);
 150         if (r < 0) {
 151                 kmem_cache_destroy(_tio_cache);
 152                 kmem_cache_destroy(_io_cache);
 153                 return r;
 154         }
 155
 156         if (!_major)
 157                 _major = r;
 158
 159         return 0;
 160 }
 161
 162 static void local_exit(void)
 163 {
 164         kmem_cache_destroy(_tio_cache);
 165         kmem_cache_destroy(_io_cache);
 166
 167         bioset_free(dm_set);
 168
 169         if (unregister_blkdev(_major, _name) < 0)
 170                 DMERR("unregister_blkdev failed");
 171
 172         _major = 0;
 173
 174         DMINFO("cleaned up");
 175 }
 176
 177 int (*_inits[])(void) __initdata = {
 178         local_init,
 179         dm_target_init,
 180         dm_linear_init,
 181         dm_stripe_init,
 182         dm_interface_init,
 183 };
 184
 185 void (*_exits[])(void) = {
 186         local_exit,
 187         dm_target_exit,
 188         dm_linear_exit,
 189         dm_stripe_exit,
 190         dm_interface_exit,
 191 };
 192
 193 static int __init dm_init(void)
 194 {
 195         const int count = ARRAY_SIZE(_inits);
 196
 197         int r, i;
 198
 199         for (i = 0; i < count; i++) {
 200                 r = _inits[i]();
 201                 if (r)
 202                         goto bad;
 203         }
 204
 205         return 0;
 206
 207       bad:
 208         while (i--)
 209                 _exits[i]();
 210
 211         return r;
 212 }
 213
 214 static void __exit dm_exit(void)
 215 {
 216         int i = ARRAY_SIZE(_exits);
 217
 218         while (i--)
 219                 _exits[i]();
 220 }
 221
 222 /*
 223  * Block device functions
 224  */
 225 static int dm_blk_open(struct inode *inode, struct file *file)
 226 {
 227         struct mapped_device *md;
 228
 229         spin_lock(&_minor_lock);
 230
 231         md = inode->i_bdev->bd_disk->private_data;
 232         if (!md)
 233                 goto out;
 234
 235         if (test_bit(DMF_FREEING, &md->flags) ||
 236             test_bit(DMF_DELETING, &md->flags)) {
 237                 md = NULL;
 238                 goto out;
 239         }
 240
 241         dm_get(md);
 242         atomic_inc(&md->open_count);
 243
 244 out:
 245         spin_unlock(&_minor_lock);
 246
 247         return md ? 0 : -ENXIO;
 248 }
 249
 250 static int dm_blk_close(struct inode *inode, struct file *file)
 251 {
 252         struct mapped_device *md;
 253
 254         md = inode->i_bdev->bd_disk->private_data;
 255         atomic_dec(&md->open_count);
 256         dm_put(md);
 257         return 0;
 258 }
 259
 260 int dm_open_count(struct mapped_device *md)
 261 {
 262         return atomic_read(&md->open_count);
 263 }
 264
 265 /*
 266  * Guarantees nothing is using the device before it's deleted.
 267  */
 268 int dm_lock_for_deletion(struct mapped_device *md)
 269 {
 270         int r = 0;
 271
 272         spin_lock(&_minor_lock);
 273
 274         if (dm_open_count(md))
 275                 r = -EBUSY;
 276         else
 277                 set_bit(DMF_DELETING, &md->flags);
 278
 279         spin_unlock(&_minor_lock);
 280
 281         return r;
 282 }
 283
 284 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 285 {
 286         struct mapped_device *md = bdev->bd_disk->private_data;
 287
 288         return dm_get_geometry(md, geo);
 289 }
 290
 291 static inline struct dm_io *alloc_io(struct mapped_device *md)
 292 {
 293         return mempool_alloc(md->io_pool, GFP_NOIO);
 294 }
 295
 296 static inline void free_io(struct mapped_device *md, struct dm_io *io)
 297 {
 298         mempool_free(io, md->io_pool);
 299 }
 300
 301 static inline struct target_io *alloc_tio(struct mapped_device *md)
 302 {
 303         return mempool_alloc(md->tio_pool, GFP_NOIO);
 304 }
 305
 306 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
 307 {
 308         mempool_free(tio, md->tio_pool);
 309 }
 310
 311 static void start_io_acct(struct dm_io *io)
 312 {
 313         struct mapped_device *md = io->md;
 314
 315         io->start_time = jiffies;
 316
 317         preempt_disable();
 318         disk_round_stats(dm_disk(md));
 319         preempt_enable();
 320         dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
 321 }
 322
 323 static int end_io_acct(struct dm_io *io)
 324 {
 325         struct mapped_device *md = io->md;
 326         struct bio *bio = io->bio;
 327         unsigned long duration = jiffies - io->start_time;
 328         int pending;
 329         int rw = bio_data_dir(bio);
 330
 331         preempt_disable();
 332         disk_round_stats(dm_disk(md));
 333         preempt_enable();
 334         dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
 335
 336         disk_stat_add(dm_disk(md), ticks[rw], duration);
 337
 338         return !pending;
 339 }
 340
 341 /*
 342  * Add the bio to the list of deferred io.
 343  */
 344 static int queue_io(struct mapped_device *md, struct bio *bio)
 345 {
 346         down_write(&md->io_lock);
 347
 348         if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
 349                 up_write(&md->io_lock);
 350                 return 1;
 351         }
 352
 353         bio_list_add(&md->deferred, bio);
 354
 355         up_write(&md->io_lock);
 356         return 0;               /* deferred successfully */
 357 }
 358
 359 /*
 360  * Everyone (including functions in this file), should use this
 361  * function to access the md->map field, and make sure they call
 362  * dm_table_put() when finished.
 363  */
 364 struct dm_table *dm_get_table(struct mapped_device *md)
 365 {
 366         struct dm_table *t;
 367
 368         read_lock(&md->map_lock);
 369         t = md->map;
 370         if (t)
 371                 dm_table_get(t);
 372         read_unlock(&md->map_lock);
 373
 374         return t;
 375 }
 376
 377 /*
 378  * Get the geometry associated with a dm device
 379  */
 380 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 381 {
 382         *geo = md->geometry;
 383
 384         return 0;
 385 }
 386
 387 /*
 388  * Set the geometry of a device.
 389  */
 390 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 391 {
 392         sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 393
 394         if (geo->start > sz) {
 395                 DMWARN("Start sector is beyond the geometry limits.");
 396                 return -EINVAL;
 397         }
 398
 399         md->geometry = *geo;
 400
 401         return 0;
 402 }
 403
 404 /*-----------------------------------------------------------------
 405  * CRUD START:
 406  *   A more elegant soln is in the works that uses the queue
 407  *   merge fn, unfortunately there are a couple of changes to
 408  *   the block layer that I want to make for this.  So in the
 409  *   interests of getting something for people to use I give
 410  *   you this clearly demarcated crap.
 411  *---------------------------------------------------------------*/
 412
 413 /*
 414  * Decrements the number of outstanding ios that a bio has been
 415  * cloned into, completing the original io if necc.
 416  */
 417 static void dec_pending(struct dm_io *io, int error)
 418 {
 419         if (error)
 420                 io->error = error;
 421
 422         if (atomic_dec_and_test(&io->io_count)) {
 423                 if (end_io_acct(io))
 424                         /* nudge anyone waiting on suspend queue */
 425                         wake_up(&io->md->wait);
 426
 427                 blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE);
 428
 429                 bio_endio(io->bio, io->bio->bi_size, io->error);
 430                 free_io(io->md, io);
 431         }
 432 }
 433
 434 static int clone_endio(struct bio *bio, unsigned int done, int error)
 435 {
 436         int r = 0;
 437         struct target_io *tio = bio->bi_private;
 438         struct dm_io *io = tio->io;
 439         dm_endio_fn endio = tio->ti->type->end_io;
 440
 441         if (bio->bi_size)
 442                 return 1;
 443
 444         if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 445                 error = -EIO;
 446
 447         if (endio) {
 448                 r = endio(tio->ti, bio, error, &tio->info);
 449                 if (r < 0)
 450                         error = r;
 451
 452                 else if (r > 0)
 453                         /* the target wants another shot at the io */
 454                         return 1;
 455         }
 456
 457         free_tio(io->md, tio);
 458         dec_pending(io, error);
 459         bio_put(bio);
 460         return r;
 461 }
 462
 463 static sector_t max_io_len(struct mapped_device *md,
 464                            sector_t sector, struct dm_target *ti)
 465 {
 466         sector_t offset = sector - ti->begin;
 467         sector_t len = ti->len - offset;
 468
 469         /*
 470          * Does the target need to split even further ?
 471          */
 472         if (ti->split_io) {
 473                 sector_t boundary;
 474                 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
 475                            - offset;
 476                 if (len > boundary)
 477                         len = boundary;
 478         }
 479
 480         return len;
 481 }
 482
 483 static void __map_bio(struct dm_target *ti, struct bio *clone,
 484                       struct target_io *tio)
 485 {
 486         int r;
 487         sector_t sector;
 488
 489         /*
 490          * Sanity checks.
 491          */
 492         BUG_ON(!clone->bi_size);
 493
 494         clone->bi_end_io = clone_endio;
 495         clone->bi_private = tio;
 496
 497         /*
 498          * Map the clone.  If r == 0 we don't need to do
 499          * anything, the target has assumed ownership of
 500          * this io.
 501          */
 502         atomic_inc(&tio->io->io_count);
 503         sector = clone->bi_sector;
 504         r = ti->type->map(ti, clone, &tio->info);
 505         if (r > 0) {
 506                 /* the bio has been remapped so dispatch it */
 507
 508                 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
 509                                     tio->io->bio->bi_bdev->bd_dev, sector,
 510                                     clone->bi_sector);
 511
 512                 generic_make_request(clone);
 513         }
 514
 515         else if (r < 0) {
 516                 /* error the io and bail out */
 517                 struct dm_io *io = tio->io;
 518                 free_tio(tio->io->md, tio);
 519                 dec_pending(io, r);
 520                 bio_put(clone);
 521         }
 522 }
 523
 524 struct clone_info {
 525         struct mapped_device *md;
 526         struct dm_table *map;
 527         struct bio *bio;
 528         struct dm_io *io;
 529         sector_t sector;
 530         sector_t sector_count;
 531         unsigned short idx;
 532 };
 533
 534 static void dm_bio_destructor(struct bio *bio)
 535 {
 536         bio_free(bio, dm_set);
 537 }
 538
 539 /*
 540  * Creates a little bio that is just does part of a bvec.
 541  */
 542 static struct bio *split_bvec(struct bio *bio, sector_t sector,
 543                               unsigned short idx, unsigned int offset,
 544                               unsigned int len)
 545 {
 546         struct bio *clone;
 547         struct bio_vec *bv = bio->bi_io_vec + idx;
 548
 549         clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set);
 550         clone->bi_destructor = dm_bio_destructor;
 551         *clone->bi_io_vec = *bv;
 552
 553         clone->bi_sector = sector;
 554         clone->bi_bdev = bio->bi_bdev;
 555         clone->bi_rw = bio->bi_rw;
 556         clone->bi_vcnt = 1;
 557         clone->bi_size = to_bytes(len);
 558         clone->bi_io_vec->bv_offset = offset;
 559         clone->bi_io_vec->bv_len = clone->bi_size;
 560
 561         return clone;
 562 }
 563
 564 /*
 565  * Creates a bio that consists of range of complete bvecs.
 566  */
 567 static struct bio *clone_bio(struct bio *bio, sector_t sector,
 568                              unsigned short idx, unsigned short bv_count,
 569                              unsigned int len)
 570 {
 571         struct bio *clone;
 572
 573         clone = bio_clone(bio, GFP_NOIO);
 574         clone->bi_sector = sector;
 575         clone->bi_idx = idx;
 576         clone->bi_vcnt = idx + bv_count;
 577         clone->bi_size = to_bytes(len);
 578         clone->bi_flags &= ~(1 << BIO_SEG_VALID);
 579
 580         return clone;
 581 }
 582
 583 static void __clone_and_map(struct clone_info *ci)
 584 {
 585         struct bio *clone, *bio = ci->bio;
 586         struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
 587         sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
 588         struct target_io *tio;
 589
 590         /*
 591          * Allocate a target io object.
 592          */
 593         tio = alloc_tio(ci->md);
 594         tio->io = ci->io;
 595         tio->ti = ti;
 596         memset(&tio->info, 0, sizeof(tio->info));
 597
 598         if (ci->sector_count <= max) {
 599                 /*
 600                  * Optimise for the simple case where we can do all of
 601                  * the remaining io with a single clone.
 602                  */
 603                 clone = clone_bio(bio, ci->sector, ci->idx,
 604                                   bio->bi_vcnt - ci->idx, ci->sector_count);
 605                 __map_bio(ti, clone, tio);
 606                 ci->sector_count = 0;
 607
 608         } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
 609                 /*
 610                  * There are some bvecs that don't span targets.
 611                  * Do as many of these as possible.
 612                  */
 613                 int i;
 614                 sector_t remaining = max;
 615                 sector_t bv_len;
 616
 617                 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
 618                         bv_len = to_sector(bio->bi_io_vec[i].bv_len);
 619
 620                         if (bv_len > remaining)
 621                                 break;
 622
 623                         remaining -= bv_len;
 624                         len += bv_len;
 625                 }
 626
 627                 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len);
 628                 __map_bio(ti, clone, tio);
 629
 630                 ci->sector += len;
 631                 ci->sector_count -= len;
 632                 ci->idx = i;
 633
 634         } else {
 635                 /*
 636                  * Handle a bvec that must be split between two or more targets.
 637                  */
 638                 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
 639                 sector_t remaining = to_sector(bv->bv_len);
 640                 unsigned int offset = 0;
 641
 642                 do {
 643                         if (offset) {
 644                                 ti = dm_table_find_target(ci->map, ci->sector);
 645                                 max = max_io_len(ci->md, ci->sector, ti);
 646
 647                                 tio = alloc_tio(ci->md);
 648                                 tio->io = ci->io;
 649                                 tio->ti = ti;
 650                                 memset(&tio->info, 0, sizeof(tio->info));
 651                         }
 652
 653                         len = min(remaining, max);
 654
 655                         clone = split_bvec(bio, ci->sector, ci->idx,
 656                                            bv->bv_offset + offset, len);
 657
 658                         __map_bio(ti, clone, tio);
 659
 660                         ci->sector += len;
 661                         ci->sector_count -= len;
 662                         offset += to_bytes(len);
 663                 } while (remaining -= len);
 664
 665                 ci->idx++;
 666         }
 667 }
 668
 669 /*
 670  * Split the bio into several clones.
 671  */
 672 static void __split_bio(struct mapped_device *md, struct bio *bio)
 673 {
 674         struct clone_info ci;
 675
 676         ci.map = dm_get_table(md);
 677         if (!ci.map) {
 678                 bio_io_error(bio, bio->bi_size);
 679                 return;
 680         }
 681
 682         ci.md = md;
 683         ci.bio = bio;
 684         ci.io = alloc_io(md);
 685         ci.io->error = 0;
 686         atomic_set(&ci.io->io_count, 1);
 687         ci.io->bio = bio;
 688         ci.io->md = md;
 689         ci.sector = bio->bi_sector;
 690         ci.sector_count = bio_sectors(bio);
 691         ci.idx = bio->bi_idx;
 692
 693         start_io_acct(ci.io);
 694         while (ci.sector_count)
 695                 __clone_and_map(&ci);
 696
 697         /* drop the extra reference count */
 698         dec_pending(ci.io, 0);
 699         dm_table_put(ci.map);
 700 }
 701 /*-----------------------------------------------------------------
 702  * CRUD END
 703  *---------------------------------------------------------------*/
 704
 705 /*
 706  * The request function that just remaps the bio built up by
 707  * dm_merge_bvec.
 708  */
 709 static int dm_request(request_queue_t *q, struct bio *bio)
 710 {
 711         int r;
 712         int rw = bio_data_dir(bio);
 713         struct mapped_device *md = q->queuedata;
 714
 715         down_read(&md->io_lock);
 716
 717         disk_stat_inc(dm_disk(md), ios[rw]);
 718         disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
 719
 720         /*
 721          * If we're suspended we have to queue
 722          * this io for later.
 723          */
 724         while (test_bit(DMF_BLOCK_IO, &md->flags)) {
 725                 up_read(&md->io_lock);
 726
 727                 if (bio_rw(bio) == READA) {
 728                         bio_io_error(bio, bio->bi_size);
 729                         return 0;
 730                 }
 731
 732                 r = queue_io(md, bio);
 733                 if (r < 0) {
 734                         bio_io_error(bio, bio->bi_size);
 735                         return 0;
 736
 737                 } else if (r == 0)
 738                         return 0;       /* deferred successfully */
 739
 740                 /*
 741                  * We're in a while loop, because someone could suspend
 742                  * before we get to the following read lock.
 743                  */
 744                 down_read(&md->io_lock);
 745         }
 746
 747         __split_bio(md, bio);
 748         up_read(&md->io_lock);
 749         return 0;
 750 }
 751
 752 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
 753                         sector_t *error_sector)
 754 {
 755         struct mapped_device *md = q->queuedata;
 756         struct dm_table *map = dm_get_table(md);
 757         int ret = -ENXIO;
 758
 759         if (map) {
 760                 ret = dm_table_flush_all(map);
 761                 dm_table_put(map);
 762         }
 763
 764         return ret;
 765 }
 766
 767 static void dm_unplug_all(request_queue_t *q)
 768 {
 769         struct mapped_device *md = q->queuedata;
 770         struct dm_table *map = dm_get_table(md);
 771
 772         if (map) {
 773                 dm_table_unplug_all(map);
 774                 dm_table_put(map);
 775         }
 776 }
 777
 778 static int dm_any_congested(void *congested_data, int bdi_bits)
 779 {
 780         int r;
 781         struct mapped_device *md = (struct mapped_device *) congested_data;
 782         struct dm_table *map = dm_get_table(md);
 783
 784         if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
 785                 r = bdi_bits;
 786         else
 787                 r = dm_table_any_congested(map, bdi_bits);
 788
 789         dm_table_put(map);
 790         return r;
 791 }
 792
 793 /*-----------------------------------------------------------------
 794  * An IDR is used to keep track of allocated minor numbers.
 795  *---------------------------------------------------------------*/
 796 static DEFINE_IDR(_minor_idr);
 797
 798 static void free_minor(int minor)
 799 {
 800         spin_lock(&_minor_lock);
 801         idr_remove(&_minor_idr, minor);
 802         spin_unlock(&_minor_lock);
 803 }
 804
 805 /*
 806  * See if the device with a specific minor # is free.
 807  */
 808 static int specific_minor(struct mapped_device *md, int minor)
 809 {
 810         int r, m;
 811
 812         if (minor >= (1 << MINORBITS))
 813                 return -EINVAL;
 814
 815         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
 816         if (!r)
 817                 return -ENOMEM;
 818
 819         spin_lock(&_minor_lock);
 820
 821         if (idr_find(&_minor_idr, minor)) {
 822                 r = -EBUSY;
 823                 goto out;
 824         }
 825
 826         r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
 827         if (r)
 828                 goto out;
 829
 830         if (m != minor) {
 831                 idr_remove(&_minor_idr, m);
 832                 r = -EBUSY;
 833                 goto out;
 834         }
 835
 836 out:
 837         spin_unlock(&_minor_lock);
 838         return r;
 839 }
 840
 841 static int next_free_minor(struct mapped_device *md, int *minor)
 842 {
 843         int r, m;
 844
 845         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
 846         if (!r)
 847                 return -ENOMEM;
 848
 849         spin_lock(&_minor_lock);
 850
 851         r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
 852         if (r) {
 853                 goto out;
 854         }
 855
 856         if (m >= (1 << MINORBITS)) {
 857                 idr_remove(&_minor_idr, m);
 858                 r = -ENOSPC;
 859                 goto out;
 860         }
 861
 862         *minor = m;
 863
 864 out:
 865         spin_unlock(&_minor_lock);
 866         return r;
 867 }
 868
 869 static struct block_device_operations dm_blk_dops;
 870
 871 /*
 872  * Allocate and initialise a blank device with a given minor.
 873  */
 874 static struct mapped_device *alloc_dev(int minor)
 875 {
 876         int r;
 877         struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
 878         void *old_md;
 879
 880         if (!md) {
 881                 DMWARN("unable to allocate device, out of memory.");
 882                 return NULL;
 883         }
 884
 885         if (!try_module_get(THIS_MODULE))
 886                 goto bad0;
 887
 888         /* get a minor number for the dev */
 889         if (minor == DM_ANY_MINOR)
 890                 r = next_free_minor(md, &minor);
 891         else
 892                 r = specific_minor(md, minor);
 893         if (r < 0)
 894                 goto bad1;
 895
 896         memset(md, 0, sizeof(*md));
 897         init_rwsem(&md->io_lock);
 898         init_MUTEX(&md->suspend_lock);
 899         rwlock_init(&md->map_lock);
 900         atomic_set(&md->holders, 1);
 901         atomic_set(&md->open_count, 0);
 902         atomic_set(&md->event_nr, 0);
 903
 904         md->queue = blk_alloc_queue(GFP_KERNEL);
 905         if (!md->queue)
 906                 goto bad1;
 907
 908         md->queue->queuedata = md;
 909         md->queue->backing_dev_info.congested_fn = dm_any_congested;
 910         md->queue->backing_dev_info.congested_data = md;
 911         blk_queue_make_request(md->queue, dm_request);
 912         blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
 913         md->queue->unplug_fn = dm_unplug_all;
 914         md->queue->issue_flush_fn = dm_flush_all;
 915
 916         md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
 917         if (!md->io_pool)
 918                 goto bad2;
 919
 920         md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
 921         if (!md->tio_pool)
 922                 goto bad3;
 923
 924         md->disk = alloc_disk(1);
 925         if (!md->disk)
 926                 goto bad4;
 927
 928         atomic_set(&md->pending, 0);
 929         init_waitqueue_head(&md->wait);
 930         init_waitqueue_head(&md->eventq);
 931
 932         md->disk->major = _major;
 933         md->disk->first_minor = minor;
 934         md->disk->fops = &dm_blk_dops;
 935         md->disk->queue = md->queue;
 936         md->disk->private_data = md;
 937         sprintf(md->disk->disk_name, "dm-%d", minor);
 938         add_disk(md->disk);
 939         format_dev_t(md->name, MKDEV(_major, minor));
 940
 941         /* Populate the mapping, nobody knows we exist yet */
 942         spin_lock(&_minor_lock);
 943         old_md = idr_replace(&_minor_idr, md, minor);
 944         spin_unlock(&_minor_lock);
 945
 946         BUG_ON(old_md != MINOR_ALLOCED);
 947
 948         return md;
 949
 950  bad4:
 951         mempool_destroy(md->tio_pool);
 952  bad3:
 953         mempool_destroy(md->io_pool);
 954  bad2:
 955         blk_cleanup_queue(md->queue);
 956         free_minor(minor);
 957  bad1:
 958         module_put(THIS_MODULE);
 959  bad0:
 960         kfree(md);
 961         return NULL;
 962 }
 963
 964 static void free_dev(struct mapped_device *md)
 965 {
 966         int minor = md->disk->first_minor;
 967
 968         if (md->suspended_bdev) {
 969                 thaw_bdev(md->suspended_bdev, NULL);
 970                 bdput(md->suspended_bdev);
 971         }
 972         mempool_destroy(md->tio_pool);
 973         mempool_destroy(md->io_pool);
 974         del_gendisk(md->disk);
 975         free_minor(minor);
 976
 977         spin_lock(&_minor_lock);
 978         md->disk->private_data = NULL;
 979         spin_unlock(&_minor_lock);
 980
 981         put_disk(md->disk);
 982         blk_cleanup_queue(md->queue);
 983         module_put(THIS_MODULE);
 984         kfree(md);
 985 }
 986
 987 /*
 988  * Bind a table to the device.
 989  */
 990 static void event_callback(void *context)
 991 {
 992         struct mapped_device *md = (struct mapped_device *) context;
 993
 994         atomic_inc(&md->event_nr);
 995         wake_up(&md->eventq);
 996 }
 997
 998 static void __set_size(struct mapped_device *md, sector_t size)
 999 {
1000         set_capacity(md->disk, size);
1001
1002         mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1003         i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1004         mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1005 }
1006
1007 static int __bind(struct mapped_device *md, struct dm_table *t)
1008 {
1009         request_queue_t *q = md->queue;
1010         sector_t size;
1011
1012         size = dm_table_get_size(t);
1013
1014         /*
1015          * Wipe any geometry if the size of the table changed.
1016          */
1017         if (size != get_capacity(md->disk))
1018                 memset(&md->geometry, 0, sizeof(md->geometry));
1019
1020         __set_size(md, size);
1021         if (size == 0)
1022                 return 0;
1023
1024         dm_table_get(t);
1025         dm_table_event_callback(t, event_callback, md);
1026
1027         write_lock(&md->map_lock);
1028         md->map = t;
1029         dm_table_set_restrictions(t, q);
1030         write_unlock(&md->map_lock);
1031
1032         return 0;
1033 }
1034
1035 static void __unbind(struct mapped_device *md)
1036 {
1037         struct dm_table *map = md->map;
1038
1039         if (!map)
1040                 return;
1041
1042         dm_table_event_callback(map, NULL, NULL);
1043         write_lock(&md->map_lock);
1044         md->map = NULL;
1045         write_unlock(&md->map_lock);
1046         dm_table_put(map);
1047 }
1048
1049 /*
1050  * Constructor for a new device.
1051  */
1052 int dm_create(int minor, struct mapped_device **result)
1053 {
1054         struct mapped_device *md;
1055
1056         md = alloc_dev(minor);
1057         if (!md)
1058                 return -ENXIO;
1059
1060         *result = md;
1061         return 0;
1062 }
1063
1064 static struct mapped_device *dm_find_md(dev_t dev)
1065 {
1066         struct mapped_device *md;
1067         unsigned minor = MINOR(dev);
1068
1069         if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1070                 return NULL;
1071
1072         spin_lock(&_minor_lock);
1073
1074         md = idr_find(&_minor_idr, minor);
1075         if (md && (md == MINOR_ALLOCED ||
1076                    (dm_disk(md)->first_minor != minor) ||
1077                    test_bit(DMF_FREEING, &md->flags))) {
1078                 md = NULL;
1079                 goto out;
1080         }
1081
1082 out:
1083         spin_unlock(&_minor_lock);
1084
1085         return md;
1086 }
1087
1088 struct mapped_device *dm_get_md(dev_t dev)
1089 {
1090         struct mapped_device *md = dm_find_md(dev);
1091
1092         if (md)
1093                 dm_get(md);
1094
1095         return md;
1096 }
1097
1098 void *dm_get_mdptr(struct mapped_device *md)
1099 {
1100         return md->interface_ptr;
1101 }
1102
1103 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1104 {
1105         md->interface_ptr = ptr;
1106 }
1107
1108 void dm_get(struct mapped_device *md)
1109 {
1110         atomic_inc(&md->holders);
1111 }
1112
1113 const char *dm_device_name(struct mapped_device *md)
1114 {
1115         return md->name;
1116 }
1117 EXPORT_SYMBOL_GPL(dm_device_name);
1118
1119 void dm_put(struct mapped_device *md)
1120 {
1121         struct dm_table *map;
1122
1123         BUG_ON(test_bit(DMF_FREEING, &md->flags));
1124
1125         if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1126                 map = dm_get_table(md);
1127                 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1128                 set_bit(DMF_FREEING, &md->flags);
1129                 spin_unlock(&_minor_lock);
1130                 if (!dm_suspended(md)) {
1131                         dm_table_presuspend_targets(map);
1132                         dm_table_postsuspend_targets(map);
1133                 }
1134                 __unbind(md);
1135                 dm_table_put(map);
1136                 free_dev(md);
1137         }
1138 }
1139
1140 /*
1141  * Process the deferred bios
1142  */
1143 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1144 {
1145         struct bio *n;
1146
1147         while (c) {
1148                 n = c->bi_next;
1149                 c->bi_next = NULL;
1150                 __split_bio(md, c);
1151                 c = n;
1152         }
1153 }
1154
1155 /*
1156  * Swap in a new table (destroying old one).
1157  */
1158 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1159 {
1160         int r = -EINVAL;
1161
1162         down(&md->suspend_lock);
1163
1164         /* device must be suspended */
1165         if (!dm_suspended(md))
1166                 goto out;
1167
1168         __unbind(md);
1169         r = __bind(md, table);
1170
1171 out:
1172         up(&md->suspend_lock);
1173         return r;
1174 }
1175
1176 /*
1177  * Functions to lock and unlock any filesystem running on the
1178  * device.
1179  */
1180 static int lock_fs(struct mapped_device *md)
1181 {
1182         int r;
1183
1184         WARN_ON(md->frozen_sb);
1185
1186         md->frozen_sb = freeze_bdev(md->suspended_bdev);
1187         if (IS_ERR(md->frozen_sb)) {
1188                 r = PTR_ERR(md->frozen_sb);
1189                 md->frozen_sb = NULL;
1190                 return r;
1191         }
1192
1193         set_bit(DMF_FROZEN, &md->flags);
1194
1195         /* don't bdput right now, we don't want the bdev
1196          * to go away while it is locked.
1197          */
1198         return 0;
1199 }
1200
1201 static void unlock_fs(struct mapped_device *md)
1202 {
1203         if (!test_bit(DMF_FROZEN, &md->flags))
1204                 return;
1205
1206         thaw_bdev(md->suspended_bdev, md->frozen_sb);
1207         md->frozen_sb = NULL;
1208         clear_bit(DMF_FROZEN, &md->flags);
1209 }
1210
1211 /*
1212  * We need to be able to change a mapping table under a mounted
1213  * filesystem.  For example we might want to move some data in
1214  * the background.  Before the table can be swapped with
1215  * dm_bind_table, dm_suspend must be called to flush any in
1216  * flight bios and ensure that any further io gets deferred.
1217  */
1218 int dm_suspend(struct mapped_device *md, int do_lockfs)
1219 {
1220         struct dm_table *map = NULL;
1221         DECLARE_WAITQUEUE(wait, current);
1222         struct bio *def;
1223         int r = -EINVAL;
1224
1225         down(&md->suspend_lock);
1226
1227         if (dm_suspended(md))
1228                 goto out;
1229
1230         map = dm_get_table(md);
1231
1232         /* This does not get reverted if there's an error later. */
1233         dm_table_presuspend_targets(map);
1234
1235         md->suspended_bdev = bdget_disk(md->disk, 0);
1236         if (!md->suspended_bdev) {
1237                 DMWARN("bdget failed in dm_suspend");
1238                 r = -ENOMEM;
1239                 goto out;
1240         }
1241
1242         /* Flush I/O to the device. */
1243         if (do_lockfs) {
1244                 r = lock_fs(md);
1245                 if (r)
1246                         goto out;
1247         }
1248
1249         /*
1250          * First we set the BLOCK_IO flag so no more ios will be mapped.
1251          */
1252         down_write(&md->io_lock);
1253         set_bit(DMF_BLOCK_IO, &md->flags);
1254
1255         add_wait_queue(&md->wait, &wait);
1256         up_write(&md->io_lock);
1257
1258         /* unplug */
1259         if (map)
1260                 dm_table_unplug_all(map);
1261
1262         /*
1263          * Then we wait for the already mapped ios to
1264          * complete.
1265          */
1266         while (1) {
1267                 set_current_state(TASK_INTERRUPTIBLE);
1268
1269                 if (!atomic_read(&md->pending) || signal_pending(current))
1270                         break;
1271
1272                 io_schedule();
1273         }
1274         set_current_state(TASK_RUNNING);
1275
1276         down_write(&md->io_lock);
1277         remove_wait_queue(&md->wait, &wait);
1278
1279         /* were we interrupted ? */
1280         r = -EINTR;
1281         if (atomic_read(&md->pending)) {
1282                 clear_bit(DMF_BLOCK_IO, &md->flags);
1283                 def = bio_list_get(&md->deferred);
1284                 __flush_deferred_io(md, def);
1285                 up_write(&md->io_lock);
1286                 unlock_fs(md);
1287                 goto out;
1288         }
1289         up_write(&md->io_lock);
1290
1291         dm_table_postsuspend_targets(map);
1292
1293         set_bit(DMF_SUSPENDED, &md->flags);
1294
1295         r = 0;
1296
1297 out:
1298         if (r && md->suspended_bdev) {
1299                 bdput(md->suspended_bdev);
1300                 md->suspended_bdev = NULL;
1301         }
1302
1303         dm_table_put(map);
1304         up(&md->suspend_lock);
1305         return r;
1306 }
1307
1308 int dm_resume(struct mapped_device *md)
1309 {
1310         int r = -EINVAL;
1311         struct bio *def;
1312         struct dm_table *map = NULL;
1313
1314         down(&md->suspend_lock);
1315         if (!dm_suspended(md))
1316                 goto out;
1317
1318         map = dm_get_table(md);
1319         if (!map || !dm_table_get_size(map))
1320                 goto out;
1321
1322         dm_table_resume_targets(map);
1323
1324         down_write(&md->io_lock);
1325         clear_bit(DMF_BLOCK_IO, &md->flags);
1326
1327         def = bio_list_get(&md->deferred);
1328         __flush_deferred_io(md, def);
1329         up_write(&md->io_lock);
1330
1331         unlock_fs(md);
1332
1333         bdput(md->suspended_bdev);
1334         md->suspended_bdev = NULL;
1335
1336         clear_bit(DMF_SUSPENDED, &md->flags);
1337
1338         dm_table_unplug_all(map);
1339
1340         r = 0;
1341
1342 out:
1343         dm_table_put(map);
1344         up(&md->suspend_lock);
1345
1346         return r;
1347 }
1348
1349 /*-----------------------------------------------------------------
1350  * Event notification.
1351  *---------------------------------------------------------------*/
1352 uint32_t dm_get_event_nr(struct mapped_device *md)
1353 {
1354         return atomic_read(&md->event_nr);
1355 }
1356
1357 int dm_wait_event(struct mapped_device *md, int event_nr)
1358 {
1359         return wait_event_interruptible(md->eventq,
1360                         (event_nr != atomic_read(&md->event_nr)));
1361 }
1362
1363 /*
1364  * The gendisk is only valid as long as you have a reference
1365  * count on 'md'.
1366  */
1367 struct gendisk *dm_disk(struct mapped_device *md)
1368 {
1369         return md->disk;
1370 }
1371
1372 int dm_suspended(struct mapped_device *md)
1373 {
1374         return test_bit(DMF_SUSPENDED, &md->flags);
1375 }
1376
1377 static struct block_device_operations dm_blk_dops = {
1378         .open = dm_blk_open,
1379         .release = dm_blk_close,
1380         .getgeo = dm_blk_getgeo,
1381         .owner = THIS_MODULE
1382 };
1383
1384 EXPORT_SYMBOL(dm_get_mapinfo);
1385
1386 /*
1387  * module hooks
1388  */
1389 module_init(dm_init);
1390 module_exit(dm_exit);
1391
1392 module_param(major, uint, 0);
1393 MODULE_PARM_DESC(major, "The major number of the device mapper");
1394 MODULE_DESCRIPTION(DM_NAME " driver");
1395 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1396 MODULE_LICENSE("GPL");