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