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