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