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 #include <linux/smp_lock.h>
  24
  25 #define DM_MSG_PREFIX "core"
  26
  27 static const char *_name = DM_NAME;
  28
  29 static unsigned int major = 0;
  30 static unsigned int _major = 0;
  31
  32 static DEFINE_SPINLOCK(_minor_lock);
  33 /*
  34  * One of these is allocated per bio.
  35  */
  36 struct dm_io {
  37         struct mapped_device *md;
  38         int error;
  39         struct bio *bio;
  40         atomic_t io_count;
  41         unsigned long start_time;
  42 };
  43
  44 /*
  45  * One of these is allocated per target within a bio.  Hopefully
  46  * this will be simplified out one day.
  47  */
  48 struct dm_target_io {
  49         struct dm_io *io;
  50         struct dm_target *ti;
  51         union map_info info;
  52 };
  53
  54 union map_info *dm_get_mapinfo(struct bio *bio)
  55 {
  56         if (bio && bio->bi_private)
  57                 return &((struct dm_target_io *)bio->bi_private)->info;
  58         return NULL;
  59 }
  60
  61 #define MINOR_ALLOCED ((void *)-1)
  62
  63 /*
  64  * Bits for the md->flags field.
  65  */
  66 #define DMF_BLOCK_IO 0
  67 #define DMF_SUSPENDED 1
  68 #define DMF_FROZEN 2
  69 #define DMF_FREEING 3
  70 #define DMF_DELETING 4
  71 #define DMF_NOFLUSH_SUSPENDING 5
  72
  73 struct mapped_device {
  74         struct rw_semaphore io_lock;
  75         struct semaphore suspend_lock;
  76         spinlock_t pushback_lock;
  77         rwlock_t map_lock;
  78         atomic_t holders;
  79         atomic_t open_count;
  80
  81         unsigned long flags;
  82
  83         struct request_queue *queue;
  84         struct gendisk *disk;
  85         char name[16];
  86
  87         void *interface_ptr;
  88
  89         /*
  90          * A list of ios that arrived while we were suspended.
  91          */
  92         atomic_t pending;
  93         wait_queue_head_t wait;
  94         struct bio_list deferred;
  95         struct bio_list pushback;
  96
  97         /*
  98          * The current mapping.
  99          */
 100         struct dm_table *map;
 101
 102         /*
 103          * io objects are allocated from here.
 104          */
 105         mempool_t *io_pool;
 106         mempool_t *tio_pool;
 107
 108         struct bio_set *bs;
 109
 110         /*
 111          * Event handling.
 112          */
 113         atomic_t event_nr;
 114         wait_queue_head_t eventq;
 115
 116         /*
 117          * freeze/thaw support require holding onto a super block
 118          */
 119         struct super_block *frozen_sb;
 120         struct block_device *suspended_bdev;
 121
 122         /* forced geometry settings */
 123         struct hd_geometry geometry;
 124 };
 125
 126 #define MIN_IOS 256
 127 static struct kmem_cache *_io_cache;
 128 static struct kmem_cache *_tio_cache;
 129
 130 static int __init local_init(void)
 131 {
 132         int r;
 133
 134         /* allocate a slab for the dm_ios */
 135         _io_cache = KMEM_CACHE(dm_io, 0);
 136         if (!_io_cache)
 137                 return -ENOMEM;
 138
 139         /* allocate a slab for the target ios */
 140         _tio_cache = KMEM_CACHE(dm_target_io, 0);
 141         if (!_tio_cache) {
 142                 kmem_cache_destroy(_io_cache);
 143                 return -ENOMEM;
 144         }
 145
 146         _major = major;
 147         r = register_blkdev(_major, _name);
 148         if (r < 0) {
 149                 kmem_cache_destroy(_tio_cache);
 150                 kmem_cache_destroy(_io_cache);
 151                 return r;
 152         }
 153
 154         if (!_major)
 155                 _major = r;
 156
 157         return 0;
 158 }
 159
 160 static void local_exit(void)
 161 {
 162         kmem_cache_destroy(_tio_cache);
 163         kmem_cache_destroy(_io_cache);
 164         unregister_blkdev(_major, _name);
 165
 166         _major = 0;
 167
 168         DMINFO("cleaned up");
 169 }
 170
 171 int (*_inits[])(void) __initdata = {
 172         local_init,
 173         dm_target_init,
 174         dm_linear_init,
 175         dm_stripe_init,
 176         dm_interface_init,
 177 };
 178
 179 void (*_exits[])(void) = {
 180         local_exit,
 181         dm_target_exit,
 182         dm_linear_exit,
 183         dm_stripe_exit,
 184         dm_interface_exit,
 185 };
 186
 187 static int __init dm_init(void)
 188 {
 189         const int count = ARRAY_SIZE(_inits);
 190
 191         int r, i;
 192
 193         for (i = 0; i < count; i++) {
 194                 r = _inits[i]();
 195                 if (r)
 196                         goto bad;
 197         }
 198
 199         return 0;
 200
 201       bad:
 202         while (i--)
 203                 _exits[i]();
 204
 205         return r;
 206 }
 207
 208 static void __exit dm_exit(void)
 209 {
 210         int i = ARRAY_SIZE(_exits);
 211
 212         while (i--)
 213                 _exits[i]();
 214 }
 215
 216 /*
 217  * Block device functions
 218  */
 219 static int dm_blk_open(struct inode *inode, struct file *file)
 220 {
 221         struct mapped_device *md;
 222
 223         spin_lock(&_minor_lock);
 224
 225         md = inode->i_bdev->bd_disk->private_data;
 226         if (!md)
 227                 goto out;
 228
 229         if (test_bit(DMF_FREEING, &md->flags) ||
 230             test_bit(DMF_DELETING, &md->flags)) {
 231                 md = NULL;
 232                 goto out;
 233         }
 234
 235         dm_get(md);
 236         atomic_inc(&md->open_count);
 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         atomic_dec(&md->open_count);
 250         dm_put(md);
 251         return 0;
 252 }
 253
 254 int dm_open_count(struct mapped_device *md)
 255 {
 256         return atomic_read(&md->open_count);
 257 }
 258
 259 /*
 260  * Guarantees nothing is using the device before it's deleted.
 261  */
 262 int dm_lock_for_deletion(struct mapped_device *md)
 263 {
 264         int r = 0;
 265
 266         spin_lock(&_minor_lock);
 267
 268         if (dm_open_count(md))
 269                 r = -EBUSY;
 270         else
 271                 set_bit(DMF_DELETING, &md->flags);
 272
 273         spin_unlock(&_minor_lock);
 274
 275         return r;
 276 }
 277
 278 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 279 {
 280         struct mapped_device *md = bdev->bd_disk->private_data;
 281
 282         return dm_get_geometry(md, geo);
 283 }
 284
 285 static int dm_blk_ioctl(struct inode *inode, struct file *file,
 286                         unsigned int cmd, unsigned long arg)
 287 {
 288         struct mapped_device *md;
 289         struct dm_table *map;
 290         struct dm_target *tgt;
 291         int r = -ENOTTY;
 292
 293         /* We don't really need this lock, but we do need 'inode'. */
 294         unlock_kernel();
 295
 296         md = inode->i_bdev->bd_disk->private_data;
 297
 298         map = dm_get_table(md);
 299
 300         if (!map || !dm_table_get_size(map))
 301                 goto out;
 302
 303         /* We only support devices that have a single target */
 304         if (dm_table_get_num_targets(map) != 1)
 305                 goto out;
 306
 307         tgt = dm_table_get_target(map, 0);
 308
 309         if (dm_suspended(md)) {
 310                 r = -EAGAIN;
 311                 goto out;
 312         }
 313
 314         if (tgt->type->ioctl)
 315                 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
 316
 317 out:
 318         dm_table_put(map);
 319
 320         lock_kernel();
 321         return r;
 322 }
 323
 324 static struct dm_io *alloc_io(struct mapped_device *md)
 325 {
 326         return mempool_alloc(md->io_pool, GFP_NOIO);
 327 }
 328
 329 static void free_io(struct mapped_device *md, struct dm_io *io)
 330 {
 331         mempool_free(io, md->io_pool);
 332 }
 333
 334 static struct dm_target_io *alloc_tio(struct mapped_device *md)
 335 {
 336         return mempool_alloc(md->tio_pool, GFP_NOIO);
 337 }
 338
 339 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
 340 {
 341         mempool_free(tio, md->tio_pool);
 342 }
 343
 344 static void start_io_acct(struct dm_io *io)
 345 {
 346         struct mapped_device *md = io->md;
 347
 348         io->start_time = jiffies;
 349
 350         preempt_disable();
 351         disk_round_stats(dm_disk(md));
 352         preempt_enable();
 353         dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
 354 }
 355
 356 static int end_io_acct(struct dm_io *io)
 357 {
 358         struct mapped_device *md = io->md;
 359         struct bio *bio = io->bio;
 360         unsigned long duration = jiffies - io->start_time;
 361         int pending;
 362         int rw = bio_data_dir(bio);
 363
 364         preempt_disable();
 365         disk_round_stats(dm_disk(md));
 366         preempt_enable();
 367         dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
 368
 369         disk_stat_add(dm_disk(md), ticks[rw], duration);
 370
 371         return !pending;
 372 }
 373
 374 /*
 375  * Add the bio to the list of deferred io.
 376  */
 377 static int queue_io(struct mapped_device *md, struct bio *bio)
 378 {
 379         down_write(&md->io_lock);
 380
 381         if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
 382                 up_write(&md->io_lock);
 383                 return 1;
 384         }
 385
 386         bio_list_add(&md->deferred, bio);
 387
 388         up_write(&md->io_lock);
 389         return 0;               /* deferred successfully */
 390 }
 391
 392 /*
 393  * Everyone (including functions in this file), should use this
 394  * function to access the md->map field, and make sure they call
 395  * dm_table_put() when finished.
 396  */
 397 struct dm_table *dm_get_table(struct mapped_device *md)
 398 {
 399         struct dm_table *t;
 400
 401         read_lock(&md->map_lock);
 402         t = md->map;
 403         if (t)
 404                 dm_table_get(t);
 405         read_unlock(&md->map_lock);
 406
 407         return t;
 408 }
 409
 410 /*
 411  * Get the geometry associated with a dm device
 412  */
 413 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 414 {
 415         *geo = md->geometry;
 416
 417         return 0;
 418 }
 419
 420 /*
 421  * Set the geometry of a device.
 422  */
 423 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 424 {
 425         sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 426
 427         if (geo->start > sz) {
 428                 DMWARN("Start sector is beyond the geometry limits.");
 429                 return -EINVAL;
 430         }
 431
 432         md->geometry = *geo;
 433
 434         return 0;
 435 }
 436
 437 /*-----------------------------------------------------------------
 438  * CRUD START:
 439  *   A more elegant soln is in the works that uses the queue
 440  *   merge fn, unfortunately there are a couple of changes to
 441  *   the block layer that I want to make for this.  So in the
 442  *   interests of getting something for people to use I give
 443  *   you this clearly demarcated crap.
 444  *---------------------------------------------------------------*/
 445
 446 static int __noflush_suspending(struct mapped_device *md)
 447 {
 448         return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 449 }
 450
 451 /*
 452  * Decrements the number of outstanding ios that a bio has been
 453  * cloned into, completing the original io if necc.
 454  */
 455 static void dec_pending(struct dm_io *io, int error)
 456 {
 457         unsigned long flags;
 458
 459         /* Push-back supersedes any I/O errors */
 460         if (error && !(io->error > 0 && __noflush_suspending(io->md)))
 461                 io->error = error;
 462
 463         if (atomic_dec_and_test(&io->io_count)) {
 464                 if (io->error == DM_ENDIO_REQUEUE) {
 465                         /*
 466                          * Target requested pushing back the I/O.
 467                          * This must be handled before the sleeper on
 468                          * suspend queue merges the pushback list.
 469                          */
 470                         spin_lock_irqsave(&io->md->pushback_lock, flags);
 471                         if (__noflush_suspending(io->md))
 472                                 bio_list_add(&io->md->pushback, io->bio);
 473                         else
 474                                 /* noflush suspend was interrupted. */
 475                                 io->error = -EIO;
 476                         spin_unlock_irqrestore(&io->md->pushback_lock, flags);
 477                 }
 478
 479                 if (end_io_acct(io))
 480                         /* nudge anyone waiting on suspend queue */
 481                         wake_up(&io->md->wait);
 482
 483                 if (io->error != DM_ENDIO_REQUEUE) {
 484                         blk_add_trace_bio(io->md->queue, io->bio,
 485                                           BLK_TA_COMPLETE);
 486
 487                         bio_endio(io->bio, io->bio->bi_size, io->error);
 488                 }
 489
 490                 free_io(io->md, io);
 491         }
 492 }
 493
 494 static int clone_endio(struct bio *bio, unsigned int done, int error)
 495 {
 496         int r = 0;
 497         struct dm_target_io *tio = bio->bi_private;
 498         struct mapped_device *md = tio->io->md;
 499         dm_endio_fn endio = tio->ti->type->end_io;
 500
 501         if (bio->bi_size)
 502                 return 1;
 503
 504         if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 505                 error = -EIO;
 506
 507         if (endio) {
 508                 r = endio(tio->ti, bio, error, &tio->info);
 509                 if (r < 0 || r == DM_ENDIO_REQUEUE)
 510                         /*
 511                          * error and requeue request are handled
 512                          * in dec_pending().
 513                          */
 514                         error = r;
 515                 else if (r == DM_ENDIO_INCOMPLETE)
 516                         /* The target will handle the io */
 517                         return 1;
 518                 else if (r) {
 519                         DMWARN("unimplemented target endio return value: %d", r);
 520                         BUG();
 521                 }
 522         }
 523
 524         dec_pending(tio->io, error);
 525
 526         /*
 527          * Store md for cleanup instead of tio which is about to get freed.
 528          */
 529         bio->bi_private = md->bs;
 530
 531         bio_put(bio);
 532         free_tio(md, tio);
 533         return r;
 534 }
 535
 536 static sector_t max_io_len(struct mapped_device *md,
 537                            sector_t sector, struct dm_target *ti)
 538 {
 539         sector_t offset = sector - ti->begin;
 540         sector_t len = ti->len - offset;
 541
 542         /*
 543          * Does the target need to split even further ?
 544          */
 545         if (ti->split_io) {
 546                 sector_t boundary;
 547                 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
 548                            - offset;
 549                 if (len > boundary)
 550                         len = boundary;
 551         }
 552
 553         return len;
 554 }
 555
 556 static void __map_bio(struct dm_target *ti, struct bio *clone,
 557                       struct dm_target_io *tio)
 558 {
 559         int r;
 560         sector_t sector;
 561         struct mapped_device *md;
 562
 563         /*
 564          * Sanity checks.
 565          */
 566         BUG_ON(!clone->bi_size);
 567
 568         clone->bi_end_io = clone_endio;
 569         clone->bi_private = tio;
 570
 571         /*
 572          * Map the clone.  If r == 0 we don't need to do
 573          * anything, the target has assumed ownership of
 574          * this io.
 575          */
 576         atomic_inc(&tio->io->io_count);
 577         sector = clone->bi_sector;
 578         r = ti->type->map(ti, clone, &tio->info);
 579         if (r == DM_MAPIO_REMAPPED) {
 580                 /* the bio has been remapped so dispatch it */
 581
 582                 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
 583                                     tio->io->bio->bi_bdev->bd_dev,
 584                                     clone->bi_sector, sector);
 585
 586                 generic_make_request(clone);
 587         } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
 588                 /* error the io and bail out, or requeue it if needed */
 589                 md = tio->io->md;
 590                 dec_pending(tio->io, r);
 591                 /*
 592                  * Store bio_set for cleanup.
 593                  */
 594                 clone->bi_private = md->bs;
 595                 bio_put(clone);
 596                 free_tio(md, tio);
 597         } else if (r) {
 598                 DMWARN("unimplemented target map return value: %d", r);
 599                 BUG();
 600         }
 601 }
 602
 603 struct clone_info {
 604         struct mapped_device *md;
 605         struct dm_table *map;
 606         struct bio *bio;
 607         struct dm_io *io;
 608         sector_t sector;
 609         sector_t sector_count;
 610         unsigned short idx;
 611 };
 612
 613 static void dm_bio_destructor(struct bio *bio)
 614 {
 615         struct bio_set *bs = bio->bi_private;
 616
 617         bio_free(bio, bs);
 618 }
 619
 620 /*
 621  * Creates a little bio that is just does part of a bvec.
 622  */
 623 static struct bio *split_bvec(struct bio *bio, sector_t sector,
 624                               unsigned short idx, unsigned int offset,
 625                               unsigned int len, struct bio_set *bs)
 626 {
 627         struct bio *clone;
 628         struct bio_vec *bv = bio->bi_io_vec + idx;
 629
 630         clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
 631         clone->bi_destructor = dm_bio_destructor;
 632         *clone->bi_io_vec = *bv;
 633
 634         clone->bi_sector = sector;
 635         clone->bi_bdev = bio->bi_bdev;
 636         clone->bi_rw = bio->bi_rw;
 637         clone->bi_vcnt = 1;
 638         clone->bi_size = to_bytes(len);
 639         clone->bi_io_vec->bv_offset = offset;
 640         clone->bi_io_vec->bv_len = clone->bi_size;
 641
 642         return clone;
 643 }
 644
 645 /*
 646  * Creates a bio that consists of range of complete bvecs.
 647  */
 648 static struct bio *clone_bio(struct bio *bio, sector_t sector,
 649                              unsigned short idx, unsigned short bv_count,
 650                              unsigned int len, struct bio_set *bs)
 651 {
 652         struct bio *clone;
 653
 654         clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
 655         __bio_clone(clone, bio);
 656         clone->bi_destructor = dm_bio_destructor;
 657         clone->bi_sector = sector;
 658         clone->bi_idx = idx;
 659         clone->bi_vcnt = idx + bv_count;
 660         clone->bi_size = to_bytes(len);
 661         clone->bi_flags &= ~(1 << BIO_SEG_VALID);
 662
 663         return clone;
 664 }
 665
 666 static int __clone_and_map(struct clone_info *ci)
 667 {
 668         struct bio *clone, *bio = ci->bio;
 669         struct dm_target *ti;
 670         sector_t len = 0, max;
 671         struct dm_target_io *tio;
 672
 673         ti = dm_table_find_target(ci->map, ci->sector);
 674         if (!dm_target_is_valid(ti))
 675                 return -EIO;
 676
 677         max = max_io_len(ci->md, ci->sector, ti);
 678
 679         /*
 680          * Allocate a target io object.
 681          */
 682         tio = alloc_tio(ci->md);
 683         tio->io = ci->io;
 684         tio->ti = ti;
 685         memset(&tio->info, 0, sizeof(tio->info));
 686
 687         if (ci->sector_count <= max) {
 688                 /*
 689                  * Optimise for the simple case where we can do all of
 690                  * the remaining io with a single clone.
 691                  */
 692                 clone = clone_bio(bio, ci->sector, ci->idx,
 693                                   bio->bi_vcnt - ci->idx, ci->sector_count,
 694                                   ci->md->bs);
 695                 __map_bio(ti, clone, tio);
 696                 ci->sector_count = 0;
 697
 698         } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
 699                 /*
 700                  * There are some bvecs that don't span targets.
 701                  * Do as many of these as possible.
 702                  */
 703                 int i;
 704                 sector_t remaining = max;
 705                 sector_t bv_len;
 706
 707                 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
 708                         bv_len = to_sector(bio->bi_io_vec[i].bv_len);
 709
 710                         if (bv_len > remaining)
 711                                 break;
 712
 713                         remaining -= bv_len;
 714                         len += bv_len;
 715                 }
 716
 717                 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
 718                                   ci->md->bs);
 719                 __map_bio(ti, clone, tio);
 720
 721                 ci->sector += len;
 722                 ci->sector_count -= len;
 723                 ci->idx = i;
 724
 725         } else {
 726                 /*
 727                  * Handle a bvec that must be split between two or more targets.
 728                  */
 729                 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
 730                 sector_t remaining = to_sector(bv->bv_len);
 731                 unsigned int offset = 0;
 732
 733                 do {
 734                         if (offset) {
 735                                 ti = dm_table_find_target(ci->map, ci->sector);
 736                                 if (!dm_target_is_valid(ti))
 737                                         return -EIO;
 738
 739                                 max = max_io_len(ci->md, ci->sector, ti);
 740
 741                                 tio = alloc_tio(ci->md);
 742                                 tio->io = ci->io;
 743                                 tio->ti = ti;
 744                                 memset(&tio->info, 0, sizeof(tio->info));
 745                         }
 746
 747                         len = min(remaining, max);
 748
 749                         clone = split_bvec(bio, ci->sector, ci->idx,
 750                                            bv->bv_offset + offset, len,
 751                                            ci->md->bs);
 752
 753                         __map_bio(ti, clone, tio);
 754
 755                         ci->sector += len;
 756                         ci->sector_count -= len;
 757                         offset += to_bytes(len);
 758                 } while (remaining -= len);
 759
 760                 ci->idx++;
 761         }
 762
 763         return 0;
 764 }
 765
 766 /*
 767  * Split the bio into several clones.
 768  */
 769 static void __split_bio(struct mapped_device *md, struct bio *bio)
 770 {
 771         struct clone_info ci;
 772         int error = 0;
 773
 774         ci.map = dm_get_table(md);
 775         if (!ci.map) {
 776                 bio_io_error(bio, bio->bi_size);
 777                 return;
 778         }
 779
 780         ci.md = md;
 781         ci.bio = bio;
 782         ci.io = alloc_io(md);
 783         ci.io->error = 0;
 784         atomic_set(&ci.io->io_count, 1);
 785         ci.io->bio = bio;
 786         ci.io->md = md;
 787         ci.sector = bio->bi_sector;
 788         ci.sector_count = bio_sectors(bio);
 789         ci.idx = bio->bi_idx;
 790
 791         start_io_acct(ci.io);
 792         while (ci.sector_count && !error)
 793                 error = __clone_and_map(&ci);
 794
 795         /* drop the extra reference count */
 796         dec_pending(ci.io, error);
 797         dm_table_put(ci.map);
 798 }
 799 /*-----------------------------------------------------------------
 800  * CRUD END
 801  *---------------------------------------------------------------*/
 802
 803 /*
 804  * The request function that just remaps the bio built up by
 805  * dm_merge_bvec.
 806  */
 807 static int dm_request(struct request_queue *q, struct bio *bio)
 808 {
 809         int r;
 810         int rw = bio_data_dir(bio);
 811         struct mapped_device *md = q->queuedata;
 812
 813         /*
 814          * There is no use in forwarding any barrier request since we can't
 815          * guarantee it is (or can be) handled by the targets correctly.
 816          */
 817         if (unlikely(bio_barrier(bio))) {
 818                 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
 819                 return 0;
 820         }
 821
 822         down_read(&md->io_lock);
 823
 824         disk_stat_inc(dm_disk(md), ios[rw]);
 825         disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
 826
 827         /*
 828          * If we're suspended we have to queue
 829          * this io for later.
 830          */
 831         while (test_bit(DMF_BLOCK_IO, &md->flags)) {
 832                 up_read(&md->io_lock);
 833
 834                 if (bio_rw(bio) == READA) {
 835                         bio_io_error(bio, bio->bi_size);
 836                         return 0;
 837                 }
 838
 839                 r = queue_io(md, bio);
 840                 if (r < 0) {
 841                         bio_io_error(bio, bio->bi_size);
 842                         return 0;
 843
 844                 } else if (r == 0)
 845                         return 0;       /* deferred successfully */
 846
 847                 /*
 848                  * We're in a while loop, because someone could suspend
 849                  * before we get to the following read lock.
 850                  */
 851                 down_read(&md->io_lock);
 852         }
 853
 854         __split_bio(md, bio);
 855         up_read(&md->io_lock);
 856         return 0;
 857 }
 858
 859 static int dm_flush_all(struct request_queue *q, struct gendisk *disk,
 860                         sector_t *error_sector)
 861 {
 862         struct mapped_device *md = q->queuedata;
 863         struct dm_table *map = dm_get_table(md);
 864         int ret = -ENXIO;
 865
 866         if (map) {
 867                 ret = dm_table_flush_all(map);
 868                 dm_table_put(map);
 869         }
 870
 871         return ret;
 872 }
 873
 874 static void dm_unplug_all(struct request_queue *q)
 875 {
 876         struct mapped_device *md = q->queuedata;
 877         struct dm_table *map = dm_get_table(md);
 878
 879         if (map) {
 880                 dm_table_unplug_all(map);
 881                 dm_table_put(map);
 882         }
 883 }
 884
 885 static int dm_any_congested(void *congested_data, int bdi_bits)
 886 {
 887         int r;
 888         struct mapped_device *md = (struct mapped_device *) congested_data;
 889         struct dm_table *map = dm_get_table(md);
 890
 891         if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
 892                 r = bdi_bits;
 893         else
 894                 r = dm_table_any_congested(map, bdi_bits);
 895
 896         dm_table_put(map);
 897         return r;
 898 }
 899
 900 /*-----------------------------------------------------------------
 901  * An IDR is used to keep track of allocated minor numbers.
 902  *---------------------------------------------------------------*/
 903 static DEFINE_IDR(_minor_idr);
 904
 905 static void free_minor(int minor)
 906 {
 907         spin_lock(&_minor_lock);
 908         idr_remove(&_minor_idr, minor);
 909         spin_unlock(&_minor_lock);
 910 }
 911
 912 /*
 913  * See if the device with a specific minor # is free.
 914  */
 915 static int specific_minor(struct mapped_device *md, int minor)
 916 {
 917         int r, m;
 918
 919         if (minor >= (1 << MINORBITS))
 920                 return -EINVAL;
 921
 922         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
 923         if (!r)
 924                 return -ENOMEM;
 925
 926         spin_lock(&_minor_lock);
 927
 928         if (idr_find(&_minor_idr, minor)) {
 929                 r = -EBUSY;
 930                 goto out;
 931         }
 932
 933         r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
 934         if (r)
 935                 goto out;
 936
 937         if (m != minor) {
 938                 idr_remove(&_minor_idr, m);
 939                 r = -EBUSY;
 940                 goto out;
 941         }
 942
 943 out:
 944         spin_unlock(&_minor_lock);
 945         return r;
 946 }
 947
 948 static int next_free_minor(struct mapped_device *md, int *minor)
 949 {
 950         int r, m;
 951
 952         r = idr_pre_get(&_minor_idr, GFP_KERNEL);
 953         if (!r)
 954                 return -ENOMEM;
 955
 956         spin_lock(&_minor_lock);
 957
 958         r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
 959         if (r) {
 960                 goto out;
 961         }
 962
 963         if (m >= (1 << MINORBITS)) {
 964                 idr_remove(&_minor_idr, m);
 965                 r = -ENOSPC;
 966                 goto out;
 967         }
 968
 969         *minor = m;
 970
 971 out:
 972         spin_unlock(&_minor_lock);
 973         return r;
 974 }
 975
 976 static struct block_device_operations dm_blk_dops;
 977
 978 /*
 979  * Allocate and initialise a blank device with a given minor.
 980  */
 981 static struct mapped_device *alloc_dev(int minor)
 982 {
 983         int r;
 984         struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
 985         void *old_md;
 986
 987         if (!md) {
 988                 DMWARN("unable to allocate device, out of memory.");
 989                 return NULL;
 990         }
 991
 992         if (!try_module_get(THIS_MODULE))
 993                 goto bad0;
 994
 995         /* get a minor number for the dev */
 996         if (minor == DM_ANY_MINOR)
 997                 r = next_free_minor(md, &minor);
 998         else
 999                 r = specific_minor(md, minor);
1000         if (r < 0)
1001                 goto bad1;
1002
1003         memset(md, 0, sizeof(*md));
1004         init_rwsem(&md->io_lock);
1005         init_MUTEX(&md->suspend_lock);
1006         spin_lock_init(&md->pushback_lock);
1007         rwlock_init(&md->map_lock);
1008         atomic_set(&md->holders, 1);
1009         atomic_set(&md->open_count, 0);
1010         atomic_set(&md->event_nr, 0);
1011
1012         md->queue = blk_alloc_queue(GFP_KERNEL);
1013         if (!md->queue)
1014                 goto bad1_free_minor;
1015
1016         md->queue->queuedata = md;
1017         md->queue->backing_dev_info.congested_fn = dm_any_congested;
1018         md->queue->backing_dev_info.congested_data = md;
1019         blk_queue_make_request(md->queue, dm_request);
1020         blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1021         md->queue->unplug_fn = dm_unplug_all;
1022         md->queue->issue_flush_fn = dm_flush_all;
1023
1024         md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1025         if (!md->io_pool)
1026                 goto bad2;
1027
1028         md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1029         if (!md->tio_pool)
1030                 goto bad3;
1031
1032         md->bs = bioset_create(16, 16);
1033         if (!md->bs)
1034                 goto bad_no_bioset;
1035
1036         md->disk = alloc_disk(1);
1037         if (!md->disk)
1038                 goto bad4;
1039
1040         atomic_set(&md->pending, 0);
1041         init_waitqueue_head(&md->wait);
1042         init_waitqueue_head(&md->eventq);
1043
1044         md->disk->major = _major;
1045         md->disk->first_minor = minor;
1046         md->disk->fops = &dm_blk_dops;
1047         md->disk->queue = md->queue;
1048         md->disk->private_data = md;
1049         sprintf(md->disk->disk_name, "dm-%d", minor);
1050         add_disk(md->disk);
1051         format_dev_t(md->name, MKDEV(_major, minor));
1052
1053         /* Populate the mapping, nobody knows we exist yet */
1054         spin_lock(&_minor_lock);
1055         old_md = idr_replace(&_minor_idr, md, minor);
1056         spin_unlock(&_minor_lock);
1057
1058         BUG_ON(old_md != MINOR_ALLOCED);
1059
1060         return md;
1061
1062  bad4:
1063         bioset_free(md->bs);
1064  bad_no_bioset:
1065         mempool_destroy(md->tio_pool);
1066  bad3:
1067         mempool_destroy(md->io_pool);
1068  bad2:
1069         blk_cleanup_queue(md->queue);
1070  bad1_free_minor:
1071         free_minor(minor);
1072  bad1:
1073         module_put(THIS_MODULE);
1074  bad0:
1075         kfree(md);
1076         return NULL;
1077 }
1078
1079 static void unlock_fs(struct mapped_device *md);
1080
1081 static void free_dev(struct mapped_device *md)
1082 {
1083         int minor = md->disk->first_minor;
1084
1085         if (md->suspended_bdev) {
1086                 unlock_fs(md);
1087                 bdput(md->suspended_bdev);
1088         }
1089         mempool_destroy(md->tio_pool);
1090         mempool_destroy(md->io_pool);
1091         bioset_free(md->bs);
1092         del_gendisk(md->disk);
1093         free_minor(minor);
1094
1095         spin_lock(&_minor_lock);
1096         md->disk->private_data = NULL;
1097         spin_unlock(&_minor_lock);
1098
1099         put_disk(md->disk);
1100         blk_cleanup_queue(md->queue);
1101         module_put(THIS_MODULE);
1102         kfree(md);
1103 }
1104
1105 /*
1106  * Bind a table to the device.
1107  */
1108 static void event_callback(void *context)
1109 {
1110         struct mapped_device *md = (struct mapped_device *) context;
1111
1112         atomic_inc(&md->event_nr);
1113         wake_up(&md->eventq);
1114 }
1115
1116 static void __set_size(struct mapped_device *md, sector_t size)
1117 {
1118         set_capacity(md->disk, size);
1119
1120         mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1121         i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1122         mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1123 }
1124
1125 static int __bind(struct mapped_device *md, struct dm_table *t)
1126 {
1127         struct request_queue *q = md->queue;
1128         sector_t size;
1129
1130         size = dm_table_get_size(t);
1131
1132         /*
1133          * Wipe any geometry if the size of the table changed.
1134          */
1135         if (size != get_capacity(md->disk))
1136                 memset(&md->geometry, 0, sizeof(md->geometry));
1137
1138         if (md->suspended_bdev)
1139                 __set_size(md, size);
1140         if (size == 0)
1141                 return 0;
1142
1143         dm_table_get(t);
1144         dm_table_event_callback(t, event_callback, md);
1145
1146         write_lock(&md->map_lock);
1147         md->map = t;
1148         dm_table_set_restrictions(t, q);
1149         write_unlock(&md->map_lock);
1150
1151         return 0;
1152 }
1153
1154 static void __unbind(struct mapped_device *md)
1155 {
1156         struct dm_table *map = md->map;
1157
1158         if (!map)
1159                 return;
1160
1161         dm_table_event_callback(map, NULL, NULL);
1162         write_lock(&md->map_lock);
1163         md->map = NULL;
1164         write_unlock(&md->map_lock);
1165         dm_table_put(map);
1166 }
1167
1168 /*
1169  * Constructor for a new device.
1170  */
1171 int dm_create(int minor, struct mapped_device **result)
1172 {
1173         struct mapped_device *md;
1174
1175         md = alloc_dev(minor);
1176         if (!md)
1177                 return -ENXIO;
1178
1179         *result = md;
1180         return 0;
1181 }
1182
1183 static struct mapped_device *dm_find_md(dev_t dev)
1184 {
1185         struct mapped_device *md;
1186         unsigned minor = MINOR(dev);
1187
1188         if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1189                 return NULL;
1190
1191         spin_lock(&_minor_lock);
1192
1193         md = idr_find(&_minor_idr, minor);
1194         if (md && (md == MINOR_ALLOCED ||
1195                    (dm_disk(md)->first_minor != minor) ||
1196                    test_bit(DMF_FREEING, &md->flags))) {
1197                 md = NULL;
1198                 goto out;
1199         }
1200
1201 out:
1202         spin_unlock(&_minor_lock);
1203
1204         return md;
1205 }
1206
1207 struct mapped_device *dm_get_md(dev_t dev)
1208 {
1209         struct mapped_device *md = dm_find_md(dev);
1210
1211         if (md)
1212                 dm_get(md);
1213
1214         return md;
1215 }
1216
1217 void *dm_get_mdptr(struct mapped_device *md)
1218 {
1219         return md->interface_ptr;
1220 }
1221
1222 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1223 {
1224         md->interface_ptr = ptr;
1225 }
1226
1227 void dm_get(struct mapped_device *md)
1228 {
1229         atomic_inc(&md->holders);
1230 }
1231
1232 const char *dm_device_name(struct mapped_device *md)
1233 {
1234         return md->name;
1235 }
1236 EXPORT_SYMBOL_GPL(dm_device_name);
1237
1238 void dm_put(struct mapped_device *md)
1239 {
1240         struct dm_table *map;
1241
1242         BUG_ON(test_bit(DMF_FREEING, &md->flags));
1243
1244         if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1245                 map = dm_get_table(md);
1246                 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1247                 set_bit(DMF_FREEING, &md->flags);
1248                 spin_unlock(&_minor_lock);
1249                 if (!dm_suspended(md)) {
1250                         dm_table_presuspend_targets(map);
1251                         dm_table_postsuspend_targets(map);
1252                 }
1253                 __unbind(md);
1254                 dm_table_put(map);
1255                 free_dev(md);
1256         }
1257 }
1258 EXPORT_SYMBOL_GPL(dm_put);
1259
1260 /*
1261  * Process the deferred bios
1262  */
1263 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1264 {
1265         struct bio *n;
1266
1267         while (c) {
1268                 n = c->bi_next;
1269                 c->bi_next = NULL;
1270                 __split_bio(md, c);
1271                 c = n;
1272         }
1273 }
1274
1275 /*
1276  * Swap in a new table (destroying old one).
1277  */
1278 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1279 {
1280         int r = -EINVAL;
1281
1282         down(&md->suspend_lock);
1283
1284         /* device must be suspended */
1285         if (!dm_suspended(md))
1286                 goto out;
1287
1288         /* without bdev, the device size cannot be changed */
1289         if (!md->suspended_bdev)
1290                 if (get_capacity(md->disk) != dm_table_get_size(table))
1291                         goto out;
1292
1293         __unbind(md);
1294         r = __bind(md, table);
1295
1296 out:
1297         up(&md->suspend_lock);
1298         return r;
1299 }
1300
1301 /*
1302  * Functions to lock and unlock any filesystem running on the
1303  * device.
1304  */
1305 static int lock_fs(struct mapped_device *md)
1306 {
1307         int r;
1308
1309         WARN_ON(md->frozen_sb);
1310
1311         md->frozen_sb = freeze_bdev(md->suspended_bdev);
1312         if (IS_ERR(md->frozen_sb)) {
1313                 r = PTR_ERR(md->frozen_sb);
1314                 md->frozen_sb = NULL;
1315                 return r;
1316         }
1317
1318         set_bit(DMF_FROZEN, &md->flags);
1319
1320         /* don't bdput right now, we don't want the bdev
1321          * to go away while it is locked.
1322          */
1323         return 0;
1324 }
1325
1326 static void unlock_fs(struct mapped_device *md)
1327 {
1328         if (!test_bit(DMF_FROZEN, &md->flags))
1329                 return;
1330
1331         thaw_bdev(md->suspended_bdev, md->frozen_sb);
1332         md->frozen_sb = NULL;
1333         clear_bit(DMF_FROZEN, &md->flags);
1334 }
1335
1336 /*
1337  * We need to be able to change a mapping table under a mounted
1338  * filesystem.  For example we might want to move some data in
1339  * the background.  Before the table can be swapped with
1340  * dm_bind_table, dm_suspend must be called to flush any in
1341  * flight bios and ensure that any further io gets deferred.
1342  */
1343 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1344 {
1345         struct dm_table *map = NULL;
1346         unsigned long flags;
1347         DECLARE_WAITQUEUE(wait, current);
1348         struct bio *def;
1349         int r = -EINVAL;
1350         int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1351         int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1352
1353         down(&md->suspend_lock);
1354
1355         if (dm_suspended(md))
1356                 goto out_unlock;
1357
1358         map = dm_get_table(md);
1359
1360         /*
1361          * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1362          * This flag is cleared before dm_suspend returns.
1363          */
1364         if (noflush)
1365                 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1366
1367         /* This does not get reverted if there's an error later. */
1368         dm_table_presuspend_targets(map);
1369
1370         /* bdget() can stall if the pending I/Os are not flushed */
1371         if (!noflush) {
1372                 md->suspended_bdev = bdget_disk(md->disk, 0);
1373                 if (!md->suspended_bdev) {
1374                         DMWARN("bdget failed in dm_suspend");
1375                         r = -ENOMEM;
1376                         goto flush_and_out;
1377                 }
1378         }
1379
1380         /*
1381          * Flush I/O to the device.
1382          * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
1383          */
1384         if (do_lockfs && !noflush) {
1385                 r = lock_fs(md);
1386                 if (r)
1387                         goto out;
1388         }
1389
1390         /*
1391          * First we set the BLOCK_IO flag so no more ios will be mapped.
1392          */
1393         down_write(&md->io_lock);
1394         set_bit(DMF_BLOCK_IO, &md->flags);
1395
1396         add_wait_queue(&md->wait, &wait);
1397         up_write(&md->io_lock);
1398
1399         /* unplug */
1400         if (map)
1401                 dm_table_unplug_all(map);
1402
1403         /*
1404          * Then we wait for the already mapped ios to
1405          * complete.
1406          */
1407         while (1) {
1408                 set_current_state(TASK_INTERRUPTIBLE);
1409
1410                 if (!atomic_read(&md->pending) || signal_pending(current))
1411                         break;
1412
1413                 io_schedule();
1414         }
1415         set_current_state(TASK_RUNNING);
1416
1417         down_write(&md->io_lock);
1418         remove_wait_queue(&md->wait, &wait);
1419
1420         if (noflush) {
1421                 spin_lock_irqsave(&md->pushback_lock, flags);
1422                 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1423                 bio_list_merge_head(&md->deferred, &md->pushback);
1424                 bio_list_init(&md->pushback);
1425                 spin_unlock_irqrestore(&md->pushback_lock, flags);
1426         }
1427
1428         /* were we interrupted ? */
1429         r = -EINTR;
1430         if (atomic_read(&md->pending)) {
1431                 clear_bit(DMF_BLOCK_IO, &md->flags);
1432                 def = bio_list_get(&md->deferred);
1433                 __flush_deferred_io(md, def);
1434                 up_write(&md->io_lock);
1435                 unlock_fs(md);
1436                 goto out; /* pushback list is already flushed, so skip flush */
1437         }
1438         up_write(&md->io_lock);
1439
1440         dm_table_postsuspend_targets(map);
1441
1442         set_bit(DMF_SUSPENDED, &md->flags);
1443
1444         r = 0;
1445
1446 flush_and_out:
1447         if (r && noflush) {
1448                 /*
1449                  * Because there may be already I/Os in the pushback list,
1450                  * flush them before return.
1451                  */
1452                 down_write(&md->io_lock);
1453
1454                 spin_lock_irqsave(&md->pushback_lock, flags);
1455                 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1456                 bio_list_merge_head(&md->deferred, &md->pushback);
1457                 bio_list_init(&md->pushback);
1458                 spin_unlock_irqrestore(&md->pushback_lock, flags);
1459
1460                 def = bio_list_get(&md->deferred);
1461                 __flush_deferred_io(md, def);
1462                 up_write(&md->io_lock);
1463         }
1464
1465 out:
1466         if (r && md->suspended_bdev) {
1467                 bdput(md->suspended_bdev);
1468                 md->suspended_bdev = NULL;
1469         }
1470
1471         dm_table_put(map);
1472
1473 out_unlock:
1474         up(&md->suspend_lock);
1475         return r;
1476 }
1477
1478 int dm_resume(struct mapped_device *md)
1479 {
1480         int r = -EINVAL;
1481         struct bio *def;
1482         struct dm_table *map = NULL;
1483
1484         down(&md->suspend_lock);
1485         if (!dm_suspended(md))
1486                 goto out;
1487
1488         map = dm_get_table(md);
1489         if (!map || !dm_table_get_size(map))
1490                 goto out;
1491
1492         r = dm_table_resume_targets(map);
1493         if (r)
1494                 goto out;
1495
1496         down_write(&md->io_lock);
1497         clear_bit(DMF_BLOCK_IO, &md->flags);
1498
1499         def = bio_list_get(&md->deferred);
1500         __flush_deferred_io(md, def);
1501         up_write(&md->io_lock);
1502
1503         unlock_fs(md);
1504
1505         if (md->suspended_bdev) {
1506                 bdput(md->suspended_bdev);
1507                 md->suspended_bdev = NULL;
1508         }
1509
1510         clear_bit(DMF_SUSPENDED, &md->flags);
1511
1512         dm_table_unplug_all(map);
1513
1514         kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1515
1516         r = 0;
1517
1518 out:
1519         dm_table_put(map);
1520         up(&md->suspend_lock);
1521
1522         return r;
1523 }
1524
1525 /*-----------------------------------------------------------------
1526  * Event notification.
1527  *---------------------------------------------------------------*/
1528 uint32_t dm_get_event_nr(struct mapped_device *md)
1529 {
1530         return atomic_read(&md->event_nr);
1531 }
1532
1533 int dm_wait_event(struct mapped_device *md, int event_nr)
1534 {
1535         return wait_event_interruptible(md->eventq,
1536                         (event_nr != atomic_read(&md->event_nr)));
1537 }
1538
1539 /*
1540  * The gendisk is only valid as long as you have a reference
1541  * count on 'md'.
1542  */
1543 struct gendisk *dm_disk(struct mapped_device *md)
1544 {
1545         return md->disk;
1546 }
1547
1548 int dm_suspended(struct mapped_device *md)
1549 {
1550         return test_bit(DMF_SUSPENDED, &md->flags);
1551 }
1552
1553 int dm_noflush_suspending(struct dm_target *ti)
1554 {
1555         struct mapped_device *md = dm_table_get_md(ti->table);
1556         int r = __noflush_suspending(md);
1557
1558         dm_put(md);
1559
1560         return r;
1561 }
1562 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1563
1564 static struct block_device_operations dm_blk_dops = {
1565         .open = dm_blk_open,
1566         .release = dm_blk_close,
1567         .ioctl = dm_blk_ioctl,
1568         .getgeo = dm_blk_getgeo,
1569         .owner = THIS_MODULE
1570 };
1571
1572 EXPORT_SYMBOL(dm_get_mapinfo);
1573
1574 /*
1575  * module hooks
1576  */
1577 module_init(dm_init);
1578 module_exit(dm_exit);
1579
1580 module_param(major, uint, 0);
1581 MODULE_PARM_DESC(major, "The major number of the device mapper");
1582 MODULE_DESCRIPTION(DM_NAME " driver");
1583 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1584 MODULE_LICENSE("GPL");