linux/drivers/md/dm.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4 *
   5 * This file is released under the GPL.
   6 */
   7
   8#include "dm.h"
   9#include "dm-uevent.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/mempool.h>
  18#include <linux/slab.h>
  19#include <linux/idr.h>
  20#include <linux/hdreg.h>
  21#include <linux/delay.h>
  22
  23#include <trace/events/block.h>
  24
  25#define DM_MSG_PREFIX "core"
  26
  27#ifdef CONFIG_PRINTK
  28/*
  29 * ratelimit state to be used in DMXXX_LIMIT().
  30 */
  31DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
  32                       DEFAULT_RATELIMIT_INTERVAL,
  33                       DEFAULT_RATELIMIT_BURST);
  34EXPORT_SYMBOL(dm_ratelimit_state);
  35#endif
  36
  37/*
  38 * Cookies are numeric values sent with CHANGE and REMOVE
  39 * uevents while resuming, removing or renaming the device.
  40 */
  41#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
  42#define DM_COOKIE_LENGTH 24
  43
  44static const char *_name = DM_NAME;
  45
  46static unsigned int major = 0;
  47static unsigned int _major = 0;
  48
  49static DEFINE_IDR(_minor_idr);
  50
  51static DEFINE_SPINLOCK(_minor_lock);
  52/*
  53 * For bio-based dm.
  54 * One of these is allocated per bio.
  55 */
  56struct dm_io {
  57        struct mapped_device *md;
  58        int error;
  59        atomic_t io_count;
  60        struct bio *bio;
  61        unsigned long start_time;
  62        spinlock_t endio_lock;
  63        struct dm_stats_aux stats_aux;
  64};
  65
  66/*
  67 * For request-based dm.
  68 * One of these is allocated per request.
  69 */
  70struct dm_rq_target_io {
  71        struct mapped_device *md;
  72        struct dm_target *ti;
  73        struct request *orig, clone;
  74        int error;
  75        union map_info info;
  76};
  77
  78/*
  79 * For request-based dm - the bio clones we allocate are embedded in these
  80 * structs.
  81 *
  82 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
  83 * the bioset is created - this means the bio has to come at the end of the
  84 * struct.
  85 */
  86struct dm_rq_clone_bio_info {
  87        struct bio *orig;
  88        struct dm_rq_target_io *tio;
  89        struct bio clone;
  90};
  91
  92union map_info *dm_get_mapinfo(struct bio *bio)
  93{
  94        if (bio && bio->bi_private)
  95                return &((struct dm_target_io *)bio->bi_private)->info;
  96        return NULL;
  97}
  98
  99union map_info *dm_get_rq_mapinfo(struct request *rq)
 100{
 101        if (rq && rq->end_io_data)
 102                return &((struct dm_rq_target_io *)rq->end_io_data)->info;
 103        return NULL;
 104}
 105EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
 106
 107#define MINOR_ALLOCED ((void *)-1)
 108
 109/*
 110 * Bits for the md->flags field.
 111 */
 112#define DMF_BLOCK_IO_FOR_SUSPEND 0
 113#define DMF_SUSPENDED 1
 114#define DMF_FROZEN 2
 115#define DMF_FREEING 3
 116#define DMF_DELETING 4
 117#define DMF_NOFLUSH_SUSPENDING 5
 118#define DMF_MERGE_IS_OPTIONAL 6
 119
 120/*
 121 * A dummy definition to make RCU happy.
 122 * struct dm_table should never be dereferenced in this file.
 123 */
 124struct dm_table {
 125        int undefined__;
 126};
 127
 128/*
 129 * Work processed by per-device workqueue.
 130 */
 131struct mapped_device {
 132        struct srcu_struct io_barrier;
 133        struct mutex suspend_lock;
 134        atomic_t holders;
 135        atomic_t open_count;
 136
 137        /*
 138         * The current mapping.
 139         * Use dm_get_live_table{_fast} or take suspend_lock for
 140         * dereference.
 141         */
 142        struct dm_table *map;
 143
 144        unsigned long flags;
 145
 146        struct request_queue *queue;
 147        unsigned type;
 148        /* Protect queue and type against concurrent access. */
 149        struct mutex type_lock;
 150
 151        struct target_type *immutable_target_type;
 152
 153        struct gendisk *disk;
 154        char name[16];
 155
 156        void *interface_ptr;
 157
 158        /*
 159         * A list of ios that arrived while we were suspended.
 160         */
 161        atomic_t pending[2];
 162        wait_queue_head_t wait;
 163        struct work_struct work;
 164        struct bio_list deferred;
 165        spinlock_t deferred_lock;
 166
 167        /*
 168         * Processing queue (flush)
 169         */
 170        struct workqueue_struct *wq;
 171
 172        /*
 173         * io objects are allocated from here.
 174         */
 175        mempool_t *io_pool;
 176
 177        struct bio_set *bs;
 178
 179        /*
 180         * Event handling.
 181         */
 182        atomic_t event_nr;
 183        wait_queue_head_t eventq;
 184        atomic_t uevent_seq;
 185        struct list_head uevent_list;
 186        spinlock_t uevent_lock; /* Protect access to uevent_list */
 187
 188        /*
 189         * freeze/thaw support require holding onto a super block
 190         */
 191        struct super_block *frozen_sb;
 192        struct block_device *bdev;
 193
 194        /* forced geometry settings */
 195        struct hd_geometry geometry;
 196
 197        /* sysfs handle */
 198        struct kobject kobj;
 199
 200        /* zero-length flush that will be cloned and submitted to targets */
 201        struct bio flush_bio;
 202
 203        struct dm_stats stats;
 204};
 205
 206/*
 207 * For mempools pre-allocation at the table loading time.
 208 */
 209struct dm_md_mempools {
 210        mempool_t *io_pool;
 211        struct bio_set *bs;
 212};
 213
 214#define RESERVED_BIO_BASED_IOS          16
 215#define RESERVED_REQUEST_BASED_IOS      256
 216#define RESERVED_MAX_IOS                1024
 217static struct kmem_cache *_io_cache;
 218static struct kmem_cache *_rq_tio_cache;
 219
 220/*
 221 * Bio-based DM's mempools' reserved IOs set by the user.
 222 */
 223static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
 224
 225/*
 226 * Request-based DM's mempools' reserved IOs set by the user.
 227 */
 228static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
 229
 230static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
 231                                      unsigned def, unsigned max)
 232{
 233        unsigned ios = ACCESS_ONCE(*reserved_ios);
 234        unsigned modified_ios = 0;
 235
 236        if (!ios)
 237                modified_ios = def;
 238        else if (ios > max)
 239                modified_ios = max;
 240
 241        if (modified_ios) {
 242                (void)cmpxchg(reserved_ios, ios, modified_ios);
 243                ios = modified_ios;
 244        }
 245
 246        return ios;
 247}
 248
 249unsigned dm_get_reserved_bio_based_ios(void)
 250{
 251        return __dm_get_reserved_ios(&reserved_bio_based_ios,
 252                                     RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
 253}
 254EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
 255
 256unsigned dm_get_reserved_rq_based_ios(void)
 257{
 258        return __dm_get_reserved_ios(&reserved_rq_based_ios,
 259                                     RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
 260}
 261EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
 262
 263static int __init local_init(void)
 264{
 265        int r = -ENOMEM;
 266
 267        /* allocate a slab for the dm_ios */
 268        _io_cache = KMEM_CACHE(dm_io, 0);
 269        if (!_io_cache)
 270                return r;
 271
 272        _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
 273        if (!_rq_tio_cache)
 274                goto out_free_io_cache;
 275
 276        r = dm_uevent_init();
 277        if (r)
 278                goto out_free_rq_tio_cache;
 279
 280        _major = major;
 281        r = register_blkdev(_major, _name);
 282        if (r < 0)
 283                goto out_uevent_exit;
 284
 285        if (!_major)
 286                _major = r;
 287
 288        return 0;
 289
 290out_uevent_exit:
 291        dm_uevent_exit();
 292out_free_rq_tio_cache:
 293        kmem_cache_destroy(_rq_tio_cache);
 294out_free_io_cache:
 295        kmem_cache_destroy(_io_cache);
 296
 297        return r;
 298}
 299
 300static void local_exit(void)
 301{
 302        kmem_cache_destroy(_rq_tio_cache);
 303        kmem_cache_destroy(_io_cache);
 304        unregister_blkdev(_major, _name);
 305        dm_uevent_exit();
 306
 307        _major = 0;
 308
 309        DMINFO("cleaned up");
 310}
 311
 312static int (*_inits[])(void) __initdata = {
 313        local_init,
 314        dm_target_init,
 315        dm_linear_init,
 316        dm_stripe_init,
 317        dm_io_init,
 318        dm_kcopyd_init,
 319        dm_interface_init,
 320        dm_statistics_init,
 321};
 322
 323static void (*_exits[])(void) = {
 324        local_exit,
 325        dm_target_exit,
 326        dm_linear_exit,
 327        dm_stripe_exit,
 328        dm_io_exit,
 329        dm_kcopyd_exit,
 330        dm_interface_exit,
 331        dm_statistics_exit,
 332};
 333
 334static int __init dm_init(void)
 335{
 336        const int count = ARRAY_SIZE(_inits);
 337
 338        int r, i;
 339
 340        for (i = 0; i < count; i++) {
 341                r = _inits[i]();
 342                if (r)
 343                        goto bad;
 344        }
 345
 346        return 0;
 347
 348      bad:
 349        while (i--)
 350                _exits[i]();
 351
 352        return r;
 353}
 354
 355static void __exit dm_exit(void)
 356{
 357        int i = ARRAY_SIZE(_exits);
 358
 359        while (i--)
 360                _exits[i]();
 361
 362        /*
 363         * Should be empty by this point.
 364         */
 365        idr_destroy(&_minor_idr);
 366}
 367
 368/*
 369 * Block device functions
 370 */
 371int dm_deleting_md(struct mapped_device *md)
 372{
 373        return test_bit(DMF_DELETING, &md->flags);
 374}
 375
 376static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 377{
 378        struct mapped_device *md;
 379
 380        spin_lock(&_minor_lock);
 381
 382        md = bdev->bd_disk->private_data;
 383        if (!md)
 384                goto out;
 385
 386        if (test_bit(DMF_FREEING, &md->flags) ||
 387            dm_deleting_md(md)) {
 388                md = NULL;
 389                goto out;
 390        }
 391
 392        dm_get(md);
 393        atomic_inc(&md->open_count);
 394
 395out:
 396        spin_unlock(&_minor_lock);
 397
 398        return md ? 0 : -ENXIO;
 399}
 400
 401static void dm_blk_close(struct gendisk *disk, fmode_t mode)
 402{
 403        struct mapped_device *md = disk->private_data;
 404
 405        spin_lock(&_minor_lock);
 406
 407        atomic_dec(&md->open_count);
 408        dm_put(md);
 409
 410        spin_unlock(&_minor_lock);
 411}
 412
 413int dm_open_count(struct mapped_device *md)
 414{
 415        return atomic_read(&md->open_count);
 416}
 417
 418/*
 419 * Guarantees nothing is using the device before it's deleted.
 420 */
 421int dm_lock_for_deletion(struct mapped_device *md)
 422{
 423        int r = 0;
 424
 425        spin_lock(&_minor_lock);
 426
 427        if (dm_open_count(md))
 428                r = -EBUSY;
 429        else
 430                set_bit(DMF_DELETING, &md->flags);
 431
 432        spin_unlock(&_minor_lock);
 433
 434        return r;
 435}
 436
 437sector_t dm_get_size(struct mapped_device *md)
 438{
 439        return get_capacity(md->disk);
 440}
 441
 442struct dm_stats *dm_get_stats(struct mapped_device *md)
 443{
 444        return &md->stats;
 445}
 446
 447static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 448{
 449        struct mapped_device *md = bdev->bd_disk->private_data;
 450
 451        return dm_get_geometry(md, geo);
 452}
 453
 454static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 455                        unsigned int cmd, unsigned long arg)
 456{
 457        struct mapped_device *md = bdev->bd_disk->private_data;
 458        int srcu_idx;
 459        struct dm_table *map;
 460        struct dm_target *tgt;
 461        int r = -ENOTTY;
 462
 463retry:
 464        map = dm_get_live_table(md, &srcu_idx);
 465
 466        if (!map || !dm_table_get_size(map))
 467                goto out;
 468
 469        /* We only support devices that have a single target */
 470        if (dm_table_get_num_targets(map) != 1)
 471                goto out;
 472
 473        tgt = dm_table_get_target(map, 0);
 474
 475        if (dm_suspended_md(md)) {
 476                r = -EAGAIN;
 477                goto out;
 478        }
 479
 480        if (tgt->type->ioctl)
 481                r = tgt->type->ioctl(tgt, cmd, arg);
 482
 483out:
 484        dm_put_live_table(md, srcu_idx);
 485
 486        if (r == -ENOTCONN) {
 487                msleep(10);
 488                goto retry;
 489        }
 490
 491        return r;
 492}
 493
 494static struct dm_io *alloc_io(struct mapped_device *md)
 495{
 496        return mempool_alloc(md->io_pool, GFP_NOIO);
 497}
 498
 499static void free_io(struct mapped_device *md, struct dm_io *io)
 500{
 501        mempool_free(io, md->io_pool);
 502}
 503
 504static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
 505{
 506        bio_put(&tio->clone);
 507}
 508
 509static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
 510                                            gfp_t gfp_mask)
 511{
 512        return mempool_alloc(md->io_pool, gfp_mask);
 513}
 514
 515static void free_rq_tio(struct dm_rq_target_io *tio)
 516{
 517        mempool_free(tio, tio->md->io_pool);
 518}
 519
 520static int md_in_flight(struct mapped_device *md)
 521{
 522        return atomic_read(&md->pending[READ]) +
 523               atomic_read(&md->pending[WRITE]);
 524}
 525
 526static void start_io_acct(struct dm_io *io)
 527{
 528        struct mapped_device *md = io->md;
 529        struct bio *bio = io->bio;
 530        int cpu;
 531        int rw = bio_data_dir(bio);
 532
 533        io->start_time = jiffies;
 534
 535        cpu = part_stat_lock();
 536        part_round_stats(cpu, &dm_disk(md)->part0);
 537        part_stat_unlock();
 538        atomic_set(&dm_disk(md)->part0.in_flight[rw],
 539                atomic_inc_return(&md->pending[rw]));
 540
 541        if (unlikely(dm_stats_used(&md->stats)))
 542                dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector,
 543                                    bio_sectors(bio), false, 0, &io->stats_aux);
 544}
 545
 546static void end_io_acct(struct dm_io *io)
 547{
 548        struct mapped_device *md = io->md;
 549        struct bio *bio = io->bio;
 550        unsigned long duration = jiffies - io->start_time;
 551        int pending, cpu;
 552        int rw = bio_data_dir(bio);
 553
 554        cpu = part_stat_lock();
 555        part_round_stats(cpu, &dm_disk(md)->part0);
 556        part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
 557        part_stat_unlock();
 558
 559        if (unlikely(dm_stats_used(&md->stats)))
 560                dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector,
 561                                    bio_sectors(bio), true, duration, &io->stats_aux);
 562
 563        /*
 564         * After this is decremented the bio must not be touched if it is
 565         * a flush.
 566         */
 567        pending = atomic_dec_return(&md->pending[rw]);
 568        atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
 569        pending += atomic_read(&md->pending[rw^0x1]);
 570
 571        /* nudge anyone waiting on suspend queue */
 572        if (!pending)
 573                wake_up(&md->wait);
 574}
 575
 576/*
 577 * Add the bio to the list of deferred io.
 578 */
 579static void queue_io(struct mapped_device *md, struct bio *bio)
 580{
 581        unsigned long flags;
 582
 583        spin_lock_irqsave(&md->deferred_lock, flags);
 584        bio_list_add(&md->deferred, bio);
 585        spin_unlock_irqrestore(&md->deferred_lock, flags);
 586        queue_work(md->wq, &md->work);
 587}
 588
 589/*
 590 * Everyone (including functions in this file), should use this
 591 * function to access the md->map field, and make sure they call
 592 * dm_put_live_table() when finished.
 593 */
 594struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
 595{
 596        *srcu_idx = srcu_read_lock(&md->io_barrier);
 597
 598        return srcu_dereference(md->map, &md->io_barrier);
 599}
 600
 601void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
 602{
 603        srcu_read_unlock(&md->io_barrier, srcu_idx);
 604}
 605
 606void dm_sync_table(struct mapped_device *md)
 607{
 608        synchronize_srcu(&md->io_barrier);
 609        synchronize_rcu_expedited();
 610}
 611
 612/*
 613 * A fast alternative to dm_get_live_table/dm_put_live_table.
 614 * The caller must not block between these two functions.
 615 */
 616static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
 617{
 618        rcu_read_lock();
 619        return rcu_dereference(md->map);
 620}
 621
 622static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
 623{
 624        rcu_read_unlock();
 625}
 626
 627/*
 628 * Get the geometry associated with a dm device
 629 */
 630int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 631{
 632        *geo = md->geometry;
 633
 634        return 0;
 635}
 636
 637/*
 638 * Set the geometry of a device.
 639 */
 640int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 641{
 642        sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 643
 644        if (geo->start > sz) {
 645                DMWARN("Start sector is beyond the geometry limits.");
 646                return -EINVAL;
 647        }
 648
 649        md->geometry = *geo;
 650
 651        return 0;
 652}
 653
 654/*-----------------------------------------------------------------
 655 * CRUD START:
 656 *   A more elegant soln is in the works that uses the queue
 657 *   merge fn, unfortunately there are a couple of changes to
 658 *   the block layer that I want to make for this.  So in the
 659 *   interests of getting something for people to use I give
 660 *   you this clearly demarcated crap.
 661 *---------------------------------------------------------------*/
 662
 663static int __noflush_suspending(struct mapped_device *md)
 664{
 665        return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 666}
 667
 668/*
 669 * Decrements the number of outstanding ios that a bio has been
 670 * cloned into, completing the original io if necc.
 671 */
 672static void dec_pending(struct dm_io *io, int error)
 673{
 674        unsigned long flags;
 675        int io_error;
 676        struct bio *bio;
 677        struct mapped_device *md = io->md;
 678
 679        /* Push-back supersedes any I/O errors */
 680        if (unlikely(error)) {
 681                spin_lock_irqsave(&io->endio_lock, flags);
 682                if (!(io->error > 0 && __noflush_suspending(md)))
 683                        io->error = error;
 684                spin_unlock_irqrestore(&io->endio_lock, flags);
 685        }
 686
 687        if (atomic_dec_and_test(&io->io_count)) {
 688                if (io->error == DM_ENDIO_REQUEUE) {
 689                        /*
 690                         * Target requested pushing back the I/O.
 691                         */
 692                        spin_lock_irqsave(&md->deferred_lock, flags);
 693                        if (__noflush_suspending(md))
 694                                bio_list_add_head(&md->deferred, io->bio);
 695                        else
 696                                /* noflush suspend was interrupted. */
 697                                io->error = -EIO;
 698                        spin_unlock_irqrestore(&md->deferred_lock, flags);
 699                }
 700
 701                io_error = io->error;
 702                bio = io->bio;
 703                end_io_acct(io);
 704                free_io(md, io);
 705
 706                if (io_error == DM_ENDIO_REQUEUE)
 707                        return;
 708
 709                if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
 710                        /*
 711                         * Preflush done for flush with data, reissue
 712                         * without REQ_FLUSH.
 713                         */
 714                        bio->bi_rw &= ~REQ_FLUSH;
 715                        queue_io(md, bio);
 716                } else {
 717                        /* done with normal IO or empty flush */
 718                        trace_block_bio_complete(md->queue, bio, io_error);
 719                        bio_endio(bio, io_error);
 720                }
 721        }
 722}
 723
 724static void clone_endio(struct bio *bio, int error)
 725{
 726        int r = 0;
 727        struct dm_target_io *tio = bio->bi_private;
 728        struct dm_io *io = tio->io;
 729        struct mapped_device *md = tio->io->md;
 730        dm_endio_fn endio = tio->ti->type->end_io;
 731
 732        if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 733                error = -EIO;
 734
 735        if (endio) {
 736                r = endio(tio->ti, bio, error);
 737                if (r < 0 || r == DM_ENDIO_REQUEUE)
 738                        /*
 739                         * error and requeue request are handled
 740                         * in dec_pending().
 741                         */
 742                        error = r;
 743                else if (r == DM_ENDIO_INCOMPLETE)
 744                        /* The target will handle the io */
 745                        return;
 746                else if (r) {
 747                        DMWARN("unimplemented target endio return value: %d", r);
 748                        BUG();
 749                }
 750        }
 751
 752        free_tio(md, tio);
 753        dec_pending(io, error);
 754}
 755
 756/*
 757 * Partial completion handling for request-based dm
 758 */
 759static void end_clone_bio(struct bio *clone, int error)
 760{
 761        struct dm_rq_clone_bio_info *info = clone->bi_private;
 762        struct dm_rq_target_io *tio = info->tio;
 763        struct bio *bio = info->orig;
 764        unsigned int nr_bytes = info->orig->bi_size;
 765
 766        bio_put(clone);
 767
 768        if (tio->error)
 769                /*
 770                 * An error has already been detected on the request.
 771                 * Once error occurred, just let clone->end_io() handle
 772                 * the remainder.
 773                 */
 774                return;
 775        else if (error) {
 776                /*
 777                 * Don't notice the error to the upper layer yet.
 778                 * The error handling decision is made by the target driver,
 779                 * when the request is completed.
 780                 */
 781                tio->error = error;
 782                return;
 783        }
 784
 785        /*
 786         * I/O for the bio successfully completed.
 787         * Notice the data completion to the upper layer.
 788         */
 789
 790        /*
 791         * bios are processed from the head of the list.
 792         * So the completing bio should always be rq->bio.
 793         * If it's not, something wrong is happening.
 794         */
 795        if (tio->orig->bio != bio)
 796                DMERR("bio completion is going in the middle of the request");
 797
 798        /*
 799         * Update the original request.
 800         * Do not use blk_end_request() here, because it may complete
 801         * the original request before the clone, and break the ordering.
 802         */
 803        blk_update_request(tio->orig, 0, nr_bytes);
 804}
 805
 806/*
 807 * Don't touch any member of the md after calling this function because
 808 * the md may be freed in dm_put() at the end of this function.
 809 * Or do dm_get() before calling this function and dm_put() later.
 810 */
 811static void rq_completed(struct mapped_device *md, int rw, int run_queue)
 812{
 813        atomic_dec(&md->pending[rw]);
 814
 815        /* nudge anyone waiting on suspend queue */
 816        if (!md_in_flight(md))
 817                wake_up(&md->wait);
 818
 819        /*
 820         * Run this off this callpath, as drivers could invoke end_io while
 821         * inside their request_fn (and holding the queue lock). Calling
 822         * back into ->request_fn() could deadlock attempting to grab the
 823         * queue lock again.
 824         */
 825        if (run_queue)
 826                blk_run_queue_async(md->queue);
 827
 828        /*
 829         * dm_put() must be at the end of this function. See the comment above
 830         */
 831        dm_put(md);
 832}
 833
 834static void free_rq_clone(struct request *clone)
 835{
 836        struct dm_rq_target_io *tio = clone->end_io_data;
 837
 838        blk_rq_unprep_clone(clone);
 839        free_rq_tio(tio);
 840}
 841
 842/*
 843 * Complete the clone and the original request.
 844 * Must be called without queue lock.
 845 */
 846static void dm_end_request(struct request *clone, int error)
 847{
 848        int rw = rq_data_dir(clone);
 849        struct dm_rq_target_io *tio = clone->end_io_data;
 850        struct mapped_device *md = tio->md;
 851        struct request *rq = tio->orig;
 852
 853        if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 854                rq->errors = clone->errors;
 855                rq->resid_len = clone->resid_len;
 856
 857                if (rq->sense)
 858                        /*
 859                         * We are using the sense buffer of the original
 860                         * request.
 861                         * So setting the length of the sense data is enough.
 862                         */
 863                        rq->sense_len = clone->sense_len;
 864        }
 865
 866        free_rq_clone(clone);
 867        blk_end_request_all(rq, error);
 868        rq_completed(md, rw, true);
 869}
 870
 871static void dm_unprep_request(struct request *rq)
 872{
 873        struct request *clone = rq->special;
 874
 875        rq->special = NULL;
 876        rq->cmd_flags &= ~REQ_DONTPREP;
 877
 878        free_rq_clone(clone);
 879}
 880
 881/*
 882 * Requeue the original request of a clone.
 883 */
 884void dm_requeue_unmapped_request(struct request *clone)
 885{
 886        int rw = rq_data_dir(clone);
 887        struct dm_rq_target_io *tio = clone->end_io_data;
 888        struct mapped_device *md = tio->md;
 889        struct request *rq = tio->orig;
 890        struct request_queue *q = rq->q;
 891        unsigned long flags;
 892
 893        dm_unprep_request(rq);
 894
 895        spin_lock_irqsave(q->queue_lock, flags);
 896        blk_requeue_request(q, rq);
 897        spin_unlock_irqrestore(q->queue_lock, flags);
 898
 899        rq_completed(md, rw, 0);
 900}
 901EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
 902
 903static void __stop_queue(struct request_queue *q)
 904{
 905        blk_stop_queue(q);
 906}
 907
 908static void stop_queue(struct request_queue *q)
 909{
 910        unsigned long flags;
 911
 912        spin_lock_irqsave(q->queue_lock, flags);
 913        __stop_queue(q);
 914        spin_unlock_irqrestore(q->queue_lock, flags);
 915}
 916
 917static void __start_queue(struct request_queue *q)
 918{
 919        if (blk_queue_stopped(q))
 920                blk_start_queue(q);
 921}
 922
 923static void start_queue(struct request_queue *q)
 924{
 925        unsigned long flags;
 926
 927        spin_lock_irqsave(q->queue_lock, flags);
 928        __start_queue(q);
 929        spin_unlock_irqrestore(q->queue_lock, flags);
 930}
 931
 932static void dm_done(struct request *clone, int error, bool mapped)
 933{
 934        int r = error;
 935        struct dm_rq_target_io *tio = clone->end_io_data;
 936        dm_request_endio_fn rq_end_io = NULL;
 937
 938        if (tio->ti) {
 939                rq_end_io = tio->ti->type->rq_end_io;
 940
 941                if (mapped && rq_end_io)
 942                        r = rq_end_io(tio->ti, clone, error, &tio->info);
 943        }
 944
 945        if (r <= 0)
 946                /* The target wants to complete the I/O */
 947                dm_end_request(clone, r);
 948        else if (r == DM_ENDIO_INCOMPLETE)
 949                /* The target will handle the I/O */
 950                return;
 951        else if (r == DM_ENDIO_REQUEUE)
 952                /* The target wants to requeue the I/O */
 953                dm_requeue_unmapped_request(clone);
 954        else {
 955                DMWARN("unimplemented target endio return value: %d", r);
 956                BUG();
 957        }
 958}
 959
 960/*
 961 * Request completion handler for request-based dm
 962 */
 963static void dm_softirq_done(struct request *rq)
 964{
 965        bool mapped = true;
 966        struct request *clone = rq->completion_data;
 967        struct dm_rq_target_io *tio = clone->end_io_data;
 968
 969        if (rq->cmd_flags & REQ_FAILED)
 970                mapped = false;
 971
 972        dm_done(clone, tio->error, mapped);
 973}
 974
 975/*
 976 * Complete the clone and the original request with the error status
 977 * through softirq context.
 978 */
 979static void dm_complete_request(struct request *clone, int error)
 980{
 981        struct dm_rq_target_io *tio = clone->end_io_data;
 982        struct request *rq = tio->orig;
 983
 984        tio->error = error;
 985        rq->completion_data = clone;
 986        blk_complete_request(rq);
 987}
 988
 989/*
 990 * Complete the not-mapped clone and the original request with the error status
 991 * through softirq context.
 992 * Target's rq_end_io() function isn't called.
 993 * This may be used when the target's map_rq() function fails.
 994 */
 995void dm_kill_unmapped_request(struct request *clone, int error)
 996{
 997        struct dm_rq_target_io *tio = clone->end_io_data;
 998        struct request *rq = tio->orig;
 999
1000        rq->cmd_flags |= REQ_FAILED;
1001        dm_complete_request(clone, error);
1002}
1003EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1004
1005/*
1006 * Called with the queue lock held
1007 */
1008static void end_clone_request(struct request *clone, int error)
1009{
1010        /*
1011         * For just cleaning up the information of the queue in which
1012         * the clone was dispatched.
1013         * The clone is *NOT* freed actually here because it is alloced from
1014         * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1015         */
1016        __blk_put_request(clone->q, clone);
1017
1018        /*
1019         * Actual request completion is done in a softirq context which doesn't
1020         * hold the queue lock.  Otherwise, deadlock could occur because:
1021         *     - another request may be submitted by the upper level driver
1022         *       of the stacking during the completion
1023         *     - the submission which requires queue lock may be done
1024         *       against this queue
1025         */
1026        dm_complete_request(clone, error);
1027}
1028
1029/*
1030 * Return maximum size of I/O possible at the supplied sector up to the current
1031 * target boundary.
1032 */
1033static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1034{
1035        sector_t target_offset = dm_target_offset(ti, sector);
1036
1037        return ti->len - target_offset;
1038}
1039
1040static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1041{
1042        sector_t len = max_io_len_target_boundary(sector, ti);
1043        sector_t offset, max_len;
1044
1045        /*
1046         * Does the target need to split even further?
1047         */
1048        if (ti->max_io_len) {
1049                offset = dm_target_offset(ti, sector);
1050                if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1051                        max_len = sector_div(offset, ti->max_io_len);
1052                else
1053                        max_len = offset & (ti->max_io_len - 1);
1054                max_len = ti->max_io_len - max_len;
1055
1056                if (len > max_len)
1057                        len = max_len;
1058        }
1059
1060        return len;
1061}
1062
1063int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1064{
1065        if (len > UINT_MAX) {
1066                DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1067                      (unsigned long long)len, UINT_MAX);
1068                ti->error = "Maximum size of target IO is too large";
1069                return -EINVAL;
1070        }
1071
1072        ti->max_io_len = (uint32_t) len;
1073
1074        return 0;
1075}
1076EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1077
1078static void __map_bio(struct dm_target_io *tio)
1079{
1080        int r;
1081        sector_t sector;
1082        struct mapped_device *md;
1083        struct bio *clone = &tio->clone;
1084        struct dm_target *ti = tio->ti;
1085
1086        clone->bi_end_io = clone_endio;
1087        clone->bi_private = tio;
1088
1089        /*
1090         * Map the clone.  If r == 0 we don't need to do
1091         * anything, the target has assumed ownership of
1092         * this io.
1093         */
1094        atomic_inc(&tio->io->io_count);
1095        sector = clone->bi_sector;
1096        r = ti->type->map(ti, clone);
1097        if (r == DM_MAPIO_REMAPPED) {
1098                /* the bio has been remapped so dispatch it */
1099
1100                trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1101                                      tio->io->bio->bi_bdev->bd_dev, sector);
1102
1103                generic_make_request(clone);
1104        } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1105                /* error the io and bail out, or requeue it if needed */
1106                md = tio->io->md;
1107                dec_pending(tio->io, r);
1108                free_tio(md, tio);
1109        } else if (r) {
1110                DMWARN("unimplemented target map return value: %d", r);
1111                BUG();
1112        }
1113}
1114
1115struct clone_info {
1116        struct mapped_device *md;
1117        struct dm_table *map;
1118        struct bio *bio;
1119        struct dm_io *io;
1120        sector_t sector;
1121        sector_t sector_count;
1122        unsigned short idx;
1123};
1124
1125static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1126{
1127        bio->bi_sector = sector;
1128        bio->bi_size = to_bytes(len);
1129}
1130
1131static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count)
1132{
1133        bio->bi_idx = idx;
1134        bio->bi_vcnt = idx + bv_count;
1135        bio->bi_flags &= ~(1 << BIO_SEG_VALID);
1136}
1137
1138static void clone_bio_integrity(struct bio *bio, struct bio *clone,
1139                                unsigned short idx, unsigned len, unsigned offset,
1140                                unsigned trim)
1141{
1142        if (!bio_integrity(bio))
1143                return;
1144
1145        bio_integrity_clone(clone, bio, GFP_NOIO);
1146
1147        if (trim)
1148                bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len);
1149}
1150
1151/*
1152 * Creates a little bio that just does part of a bvec.
1153 */
1154static void clone_split_bio(struct dm_target_io *tio, struct bio *bio,
1155                            sector_t sector, unsigned short idx,
1156                            unsigned offset, unsigned len)
1157{
1158        struct bio *clone = &tio->clone;
1159        struct bio_vec *bv = bio->bi_io_vec + idx;
1160
1161        *clone->bi_io_vec = *bv;
1162
1163        bio_setup_sector(clone, sector, len);
1164
1165        clone->bi_bdev = bio->bi_bdev;
1166        clone->bi_rw = bio->bi_rw;
1167        clone->bi_vcnt = 1;
1168        clone->bi_io_vec->bv_offset = offset;
1169        clone->bi_io_vec->bv_len = clone->bi_size;
1170        clone->bi_flags |= 1 << BIO_CLONED;
1171
1172        clone_bio_integrity(bio, clone, idx, len, offset, 1);
1173}
1174
1175/*
1176 * Creates a bio that consists of range of complete bvecs.
1177 */
1178static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1179                      sector_t sector, unsigned short idx,
1180                      unsigned short bv_count, unsigned len)
1181{
1182        struct bio *clone = &tio->clone;
1183        unsigned trim = 0;
1184
1185        __bio_clone(clone, bio);
1186        bio_setup_sector(clone, sector, len);
1187        bio_setup_bv(clone, idx, bv_count);
1188
1189        if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1190                trim = 1;
1191        clone_bio_integrity(bio, clone, idx, len, 0, trim);
1192}
1193
1194static struct dm_target_io *alloc_tio(struct clone_info *ci,
1195                                      struct dm_target *ti, int nr_iovecs,
1196                                      unsigned target_bio_nr)
1197{
1198        struct dm_target_io *tio;
1199        struct bio *clone;
1200
1201        clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1202        tio = container_of(clone, struct dm_target_io, clone);
1203
1204        tio->io = ci->io;
1205        tio->ti = ti;
1206        memset(&tio->info, 0, sizeof(tio->info));
1207        tio->target_bio_nr = target_bio_nr;
1208
1209        return tio;
1210}
1211
1212static void __clone_and_map_simple_bio(struct clone_info *ci,
1213                                       struct dm_target *ti,
1214                                       unsigned target_bio_nr, sector_t len)
1215{
1216        struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1217        struct bio *clone = &tio->clone;
1218
1219        /*
1220         * Discard requests require the bio's inline iovecs be initialized.
1221         * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1222         * and discard, so no need for concern about wasted bvec allocations.
1223         */
1224         __bio_clone(clone, ci->bio);
1225        if (len)
1226                bio_setup_sector(clone, ci->sector, len);
1227
1228        __map_bio(tio);
1229}
1230
1231static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1232                                  unsigned num_bios, sector_t len)
1233{
1234        unsigned target_bio_nr;
1235
1236        for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1237                __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1238}
1239
1240static int __send_empty_flush(struct clone_info *ci)
1241{
1242        unsigned target_nr = 0;
1243        struct dm_target *ti;
1244
1245        BUG_ON(bio_has_data(ci->bio));
1246        while ((ti = dm_table_get_target(ci->map, target_nr++)))
1247                __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1248
1249        return 0;
1250}
1251
1252static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1253                                     sector_t sector, int nr_iovecs,
1254                                     unsigned short idx, unsigned short bv_count,
1255                                     unsigned offset, unsigned len,
1256                                     unsigned split_bvec)
1257{
1258        struct bio *bio = ci->bio;
1259        struct dm_target_io *tio;
1260        unsigned target_bio_nr;
1261        unsigned num_target_bios = 1;
1262
1263        /*
1264         * Does the target want to receive duplicate copies of the bio?
1265         */
1266        if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1267                num_target_bios = ti->num_write_bios(ti, bio);
1268
1269        for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1270                tio = alloc_tio(ci, ti, nr_iovecs, target_bio_nr);
1271                if (split_bvec)
1272                        clone_split_bio(tio, bio, sector, idx, offset, len);
1273                else
1274                        clone_bio(tio, bio, sector, idx, bv_count, len);
1275                __map_bio(tio);
1276        }
1277}
1278
1279typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1280
1281static unsigned get_num_discard_bios(struct dm_target *ti)
1282{
1283        return ti->num_discard_bios;
1284}
1285
1286static unsigned get_num_write_same_bios(struct dm_target *ti)
1287{
1288        return ti->num_write_same_bios;
1289}
1290
1291typedef bool (*is_split_required_fn)(struct dm_target *ti);
1292
1293static bool is_split_required_for_discard(struct dm_target *ti)
1294{
1295        return ti->split_discard_bios;
1296}
1297
1298static int __send_changing_extent_only(struct clone_info *ci,
1299                                       get_num_bios_fn get_num_bios,
1300                                       is_split_required_fn is_split_required)
1301{
1302        struct dm_target *ti;
1303        sector_t len;
1304        unsigned num_bios;
1305
1306        do {
1307                ti = dm_table_find_target(ci->map, ci->sector);
1308                if (!dm_target_is_valid(ti))
1309                        return -EIO;
1310
1311                /*
1312                 * Even though the device advertised support for this type of
1313                 * request, that does not mean every target supports it, and
1314                 * reconfiguration might also have changed that since the
1315                 * check was performed.
1316                 */
1317                num_bios = get_num_bios ? get_num_bios(ti) : 0;
1318                if (!num_bios)
1319                        return -EOPNOTSUPP;
1320
1321                if (is_split_required && !is_split_required(ti))
1322                        len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1323                else
1324                        len = min(ci->sector_count, max_io_len(ci->sector, ti));
1325
1326                __send_duplicate_bios(ci, ti, num_bios, len);
1327
1328                ci->sector += len;
1329        } while (ci->sector_count -= len);
1330
1331        return 0;
1332}
1333
1334static int __send_discard(struct clone_info *ci)
1335{
1336        return __send_changing_extent_only(ci, get_num_discard_bios,
1337                                           is_split_required_for_discard);
1338}
1339
1340static int __send_write_same(struct clone_info *ci)
1341{
1342        return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1343}
1344
1345/*
1346 * Find maximum number of sectors / bvecs we can process with a single bio.
1347 */
1348static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx)
1349{
1350        struct bio *bio = ci->bio;
1351        sector_t bv_len, total_len = 0;
1352
1353        for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) {
1354                bv_len = to_sector(bio->bi_io_vec[*idx].bv_len);
1355
1356                if (bv_len > max)
1357                        break;
1358
1359                max -= bv_len;
1360                total_len += bv_len;
1361        }
1362
1363        return total_len;
1364}
1365
1366static int __split_bvec_across_targets(struct clone_info *ci,
1367                                       struct dm_target *ti, sector_t max)
1368{
1369        struct bio *bio = ci->bio;
1370        struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1371        sector_t remaining = to_sector(bv->bv_len);
1372        unsigned offset = 0;
1373        sector_t len;
1374
1375        do {
1376                if (offset) {
1377                        ti = dm_table_find_target(ci->map, ci->sector);
1378                        if (!dm_target_is_valid(ti))
1379                                return -EIO;
1380
1381                        max = max_io_len(ci->sector, ti);
1382                }
1383
1384                len = min(remaining, max);
1385
1386                __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0,
1387                                         bv->bv_offset + offset, len, 1);
1388
1389                ci->sector += len;
1390                ci->sector_count -= len;
1391                offset += to_bytes(len);
1392        } while (remaining -= len);
1393
1394        ci->idx++;
1395
1396        return 0;
1397}
1398
1399/*
1400 * Select the correct strategy for processing a non-flush bio.
1401 */
1402static int __split_and_process_non_flush(struct clone_info *ci)
1403{
1404        struct bio *bio = ci->bio;
1405        struct dm_target *ti;
1406        sector_t len, max;
1407        int idx;
1408
1409        if (unlikely(bio->bi_rw & REQ_DISCARD))
1410                return __send_discard(ci);
1411        else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1412                return __send_write_same(ci);
1413
1414        ti = dm_table_find_target(ci->map, ci->sector);
1415        if (!dm_target_is_valid(ti))
1416                return -EIO;
1417
1418        max = max_io_len(ci->sector, ti);
1419
1420        /*
1421         * Optimise for the simple case where we can do all of
1422         * the remaining io with a single clone.
1423         */
1424        if (ci->sector_count <= max) {
1425                __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs,
1426                                         ci->idx, bio->bi_vcnt - ci->idx, 0,
1427                                         ci->sector_count, 0);
1428                ci->sector_count = 0;
1429                return 0;
1430        }
1431
1432        /*
1433         * There are some bvecs that don't span targets.
1434         * Do as many of these as possible.
1435         */
1436        if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1437                len = __len_within_target(ci, max, &idx);
1438
1439                __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs,
1440                                         ci->idx, idx - ci->idx, 0, len, 0);
1441
1442                ci->sector += len;
1443                ci->sector_count -= len;
1444                ci->idx = idx;
1445
1446                return 0;
1447        }
1448
1449        /*
1450         * Handle a bvec that must be split between two or more targets.
1451         */
1452        return __split_bvec_across_targets(ci, ti, max);
1453}
1454
1455/*
1456 * Entry point to split a bio into clones and submit them to the targets.
1457 */
1458static void __split_and_process_bio(struct mapped_device *md,
1459                                    struct dm_table *map, struct bio *bio)
1460{
1461        struct clone_info ci;
1462        int error = 0;
1463
1464        if (unlikely(!map)) {
1465                bio_io_error(bio);
1466                return;
1467        }
1468
1469        ci.map = map;
1470        ci.md = md;
1471        ci.io = alloc_io(md);
1472        ci.io->error = 0;
1473        atomic_set(&ci.io->io_count, 1);
1474        ci.io->bio = bio;
1475        ci.io->md = md;
1476        spin_lock_init(&ci.io->endio_lock);
1477        ci.sector = bio->bi_sector;
1478        ci.idx = bio->bi_idx;
1479
1480        start_io_acct(ci.io);
1481
1482        if (bio->bi_rw & REQ_FLUSH) {
1483                ci.bio = &ci.md->flush_bio;
1484                ci.sector_count = 0;
1485                error = __send_empty_flush(&ci);
1486                /* dec_pending submits any data associated with flush */
1487        } else {
1488                ci.bio = bio;
1489                ci.sector_count = bio_sectors(bio);
1490                while (ci.sector_count && !error)
1491                        error = __split_and_process_non_flush(&ci);
1492        }
1493
1494        /* drop the extra reference count */
1495        dec_pending(ci.io, error);
1496}
1497/*-----------------------------------------------------------------
1498 * CRUD END
1499 *---------------------------------------------------------------*/
1500
1501static int dm_merge_bvec(struct request_queue *q,
1502                         struct bvec_merge_data *bvm,
1503                         struct bio_vec *biovec)
1504{
1505        struct mapped_device *md = q->queuedata;
1506        struct dm_table *map = dm_get_live_table_fast(md);
1507        struct dm_target *ti;
1508        sector_t max_sectors;
1509        int max_size = 0;
1510
1511        if (unlikely(!map))
1512                goto out;
1513
1514        ti = dm_table_find_target(map, bvm->bi_sector);
1515        if (!dm_target_is_valid(ti))
1516                goto out;
1517
1518        /*
1519         * Find maximum amount of I/O that won't need splitting
1520         */
1521        max_sectors = min(max_io_len(bvm->bi_sector, ti),
1522                          (sector_t) BIO_MAX_SECTORS);
1523        max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1524        if (max_size < 0)
1525                max_size = 0;
1526
1527        /*
1528         * merge_bvec_fn() returns number of bytes
1529         * it can accept at this offset
1530         * max is precomputed maximal io size
1531         */
1532        if (max_size && ti->type->merge)
1533                max_size = ti->type->merge(ti, bvm, biovec, max_size);
1534        /*
1535         * If the target doesn't support merge method and some of the devices
1536         * provided their merge_bvec method (we know this by looking at
1537         * queue_max_hw_sectors), then we can't allow bios with multiple vector
1538         * entries.  So always set max_size to 0, and the code below allows
1539         * just one page.
1540         */
1541        else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1542
1543                max_size = 0;
1544
1545out:
1546        dm_put_live_table_fast(md);
1547        /*
1548         * Always allow an entire first page
1549         */
1550        if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1551                max_size = biovec->bv_len;
1552
1553        return max_size;
1554}
1555
1556/*
1557 * The request function that just remaps the bio built up by
1558 * dm_merge_bvec.
1559 */
1560static void _dm_request(struct request_queue *q, struct bio *bio)
1561{
1562        int rw = bio_data_dir(bio);
1563        struct mapped_device *md = q->queuedata;
1564        int cpu;
1565        int srcu_idx;
1566        struct dm_table *map;
1567
1568        map = dm_get_live_table(md, &srcu_idx);
1569
1570        cpu = part_stat_lock();
1571        part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1572        part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1573        part_stat_unlock();
1574
1575        /* if we're suspended, we have to queue this io for later */
1576        if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1577                dm_put_live_table(md, srcu_idx);
1578
1579                if (bio_rw(bio) != READA)
1580                        queue_io(md, bio);
1581                else
1582                        bio_io_error(bio);
1583                return;
1584        }
1585
1586        __split_and_process_bio(md, map, bio);
1587        dm_put_live_table(md, srcu_idx);
1588        return;
1589}
1590
1591int dm_request_based(struct mapped_device *md)
1592{
1593        return blk_queue_stackable(md->queue);
1594}
1595
1596static void dm_request(struct request_queue *q, struct bio *bio)
1597{
1598        struct mapped_device *md = q->queuedata;
1599
1600        if (dm_request_based(md))
1601                blk_queue_bio(q, bio);
1602        else
1603                _dm_request(q, bio);
1604}
1605
1606void dm_dispatch_request(struct request *rq)
1607{
1608        int r;
1609
1610        if (blk_queue_io_stat(rq->q))
1611                rq->cmd_flags |= REQ_IO_STAT;
1612
1613        rq->start_time = jiffies;
1614        r = blk_insert_cloned_request(rq->q, rq);
1615        if (r)
1616                dm_complete_request(rq, r);
1617}
1618EXPORT_SYMBOL_GPL(dm_dispatch_request);
1619
1620static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1621                                 void *data)
1622{
1623        struct dm_rq_target_io *tio = data;
1624        struct dm_rq_clone_bio_info *info =
1625                container_of(bio, struct dm_rq_clone_bio_info, clone);
1626
1627        info->orig = bio_orig;
1628        info->tio = tio;
1629        bio->bi_end_io = end_clone_bio;
1630        bio->bi_private = info;
1631
1632        return 0;
1633}
1634
1635static int setup_clone(struct request *clone, struct request *rq,
1636                       struct dm_rq_target_io *tio)
1637{
1638        int r;
1639
1640        r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1641                              dm_rq_bio_constructor, tio);
1642        if (r)
1643                return r;
1644
1645        clone->cmd = rq->cmd;
1646        clone->cmd_len = rq->cmd_len;
1647        clone->sense = rq->sense;
1648        clone->buffer = rq->buffer;
1649        clone->end_io = end_clone_request;
1650        clone->end_io_data = tio;
1651
1652        return 0;
1653}
1654
1655static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1656                                gfp_t gfp_mask)
1657{
1658        struct request *clone;
1659        struct dm_rq_target_io *tio;
1660
1661        tio = alloc_rq_tio(md, gfp_mask);
1662        if (!tio)
1663                return NULL;
1664
1665        tio->md = md;
1666        tio->ti = NULL;
1667        tio->orig = rq;
1668        tio->error = 0;
1669        memset(&tio->info, 0, sizeof(tio->info));
1670
1671        clone = &tio->clone;
1672        if (setup_clone(clone, rq, tio)) {
1673                /* -ENOMEM */
1674                free_rq_tio(tio);
1675                return NULL;
1676        }
1677
1678        return clone;
1679}
1680
1681/*
1682 * Called with the queue lock held.
1683 */
1684static int dm_prep_fn(struct request_queue *q, struct request *rq)
1685{
1686        struct mapped_device *md = q->queuedata;
1687        struct request *clone;
1688
1689        if (unlikely(rq->special)) {
1690                DMWARN("Already has something in rq->special.");
1691                return BLKPREP_KILL;
1692        }
1693
1694        clone = clone_rq(rq, md, GFP_ATOMIC);
1695        if (!clone)
1696                return BLKPREP_DEFER;
1697
1698        rq->special = clone;
1699        rq->cmd_flags |= REQ_DONTPREP;
1700
1701        return BLKPREP_OK;
1702}
1703
1704/*
1705 * Returns:
1706 * 0  : the request has been processed (not requeued)
1707 * !0 : the request has been requeued
1708 */
1709static int map_request(struct dm_target *ti, struct request *clone,
1710                       struct mapped_device *md)
1711{
1712        int r, requeued = 0;
1713        struct dm_rq_target_io *tio = clone->end_io_data;
1714
1715        tio->ti = ti;
1716        r = ti->type->map_rq(ti, clone, &tio->info);
1717        switch (r) {
1718        case DM_MAPIO_SUBMITTED:
1719                /* The target has taken the I/O to submit by itself later */
1720                break;
1721        case DM_MAPIO_REMAPPED:
1722                /* The target has remapped the I/O so dispatch it */
1723                trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1724                                     blk_rq_pos(tio->orig));
1725                dm_dispatch_request(clone);
1726                break;
1727        case DM_MAPIO_REQUEUE:
1728                /* The target wants to requeue the I/O */
1729                dm_requeue_unmapped_request(clone);
1730                requeued = 1;
1731                break;
1732        default:
1733                if (r > 0) {
1734                        DMWARN("unimplemented target map return value: %d", r);
1735                        BUG();
1736                }
1737
1738                /* The target wants to complete the I/O */
1739                dm_kill_unmapped_request(clone, r);
1740                break;
1741        }
1742
1743        return requeued;
1744}
1745
1746static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1747{
1748        struct request *clone;
1749
1750        blk_start_request(orig);
1751        clone = orig->special;
1752        atomic_inc(&md->pending[rq_data_dir(clone)]);
1753
1754        /*
1755         * Hold the md reference here for the in-flight I/O.
1756         * We can't rely on the reference count by device opener,
1757         * because the device may be closed during the request completion
1758         * when all bios are completed.
1759         * See the comment in rq_completed() too.
1760         */
1761        dm_get(md);
1762
1763        return clone;
1764}
1765
1766/*
1767 * q->request_fn for request-based dm.
1768 * Called with the queue lock held.
1769 */
1770static void dm_request_fn(struct request_queue *q)
1771{
1772        struct mapped_device *md = q->queuedata;
1773        int srcu_idx;
1774        struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1775        struct dm_target *ti;
1776        struct request *rq, *clone;
1777        sector_t pos;
1778
1779        /*
1780         * For suspend, check blk_queue_stopped() and increment
1781         * ->pending within a single queue_lock not to increment the
1782         * number of in-flight I/Os after the queue is stopped in
1783         * dm_suspend().
1784         */
1785        while (!blk_queue_stopped(q)) {
1786                rq = blk_peek_request(q);
1787                if (!rq)
1788                        goto delay_and_out;
1789
1790                /* always use block 0 to find the target for flushes for now */
1791                pos = 0;
1792                if (!(rq->cmd_flags & REQ_FLUSH))
1793                        pos = blk_rq_pos(rq);
1794
1795                ti = dm_table_find_target(map, pos);
1796                if (!dm_target_is_valid(ti)) {
1797                        /*
1798                         * Must perform setup, that dm_done() requires,
1799                         * before calling dm_kill_unmapped_request
1800                         */
1801                        DMERR_LIMIT("request attempted access beyond the end of device");
1802                        clone = dm_start_request(md, rq);
1803                        dm_kill_unmapped_request(clone, -EIO);
1804                        continue;
1805                }
1806
1807                if (ti->type->busy && ti->type->busy(ti))
1808                        goto delay_and_out;
1809
1810                clone = dm_start_request(md, rq);
1811
1812                spin_unlock(q->queue_lock);
1813                if (map_request(ti, clone, md))
1814                        goto requeued;
1815
1816                BUG_ON(!irqs_disabled());
1817                spin_lock(q->queue_lock);
1818        }
1819
1820        goto out;
1821
1822requeued:
1823        BUG_ON(!irqs_disabled());
1824        spin_lock(q->queue_lock);
1825
1826delay_and_out:
1827        blk_delay_queue(q, HZ / 10);
1828out:
1829        dm_put_live_table(md, srcu_idx);
1830}
1831
1832int dm_underlying_device_busy(struct request_queue *q)
1833{
1834        return blk_lld_busy(q);
1835}
1836EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1837
1838static int dm_lld_busy(struct request_queue *q)
1839{
1840        int r;
1841        struct mapped_device *md = q->queuedata;
1842        struct dm_table *map = dm_get_live_table_fast(md);
1843
1844        if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1845                r = 1;
1846        else
1847                r = dm_table_any_busy_target(map);
1848
1849        dm_put_live_table_fast(md);
1850
1851        return r;
1852}
1853
1854static int dm_any_congested(void *congested_data, int bdi_bits)
1855{
1856        int r = bdi_bits;
1857        struct mapped_device *md = congested_data;
1858        struct dm_table *map;
1859
1860        if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1861                map = dm_get_live_table_fast(md);
1862                if (map) {
1863                        /*
1864                         * Request-based dm cares about only own queue for
1865                         * the query about congestion status of request_queue
1866                         */
1867                        if (dm_request_based(md))
1868                                r = md->queue->backing_dev_info.state &
1869                                    bdi_bits;
1870                        else
1871                                r = dm_table_any_congested(map, bdi_bits);
1872                }
1873                dm_put_live_table_fast(md);
1874        }
1875
1876        return r;
1877}
1878
1879/*-----------------------------------------------------------------
1880 * An IDR is used to keep track of allocated minor numbers.
1881 *---------------------------------------------------------------*/
1882static void free_minor(int minor)
1883{
1884        spin_lock(&_minor_lock);
1885        idr_remove(&_minor_idr, minor);
1886        spin_unlock(&_minor_lock);
1887}
1888
1889/*
1890 * See if the device with a specific minor # is free.
1891 */
1892static int specific_minor(int minor)
1893{
1894        int r;
1895
1896        if (minor >= (1 << MINORBITS))
1897                return -EINVAL;
1898
1899        idr_preload(GFP_KERNEL);
1900        spin_lock(&_minor_lock);
1901
1902        r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1903
1904        spin_unlock(&_minor_lock);
1905        idr_preload_end();
1906        if (r < 0)
1907                return r == -ENOSPC ? -EBUSY : r;
1908        return 0;
1909}
1910
1911static int next_free_minor(int *minor)
1912{
1913        int r;
1914
1915        idr_preload(GFP_KERNEL);
1916        spin_lock(&_minor_lock);
1917
1918        r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1919
1920        spin_unlock(&_minor_lock);
1921        idr_preload_end();
1922        if (r < 0)
1923                return r;
1924        *minor = r;
1925        return 0;
1926}
1927
1928static const struct block_device_operations dm_blk_dops;
1929
1930static void dm_wq_work(struct work_struct *work);
1931
1932static void dm_init_md_queue(struct mapped_device *md)
1933{
1934        /*
1935         * Request-based dm devices cannot be stacked on top of bio-based dm
1936         * devices.  The type of this dm device has not been decided yet.
1937         * The type is decided at the first table loading time.
1938         * To prevent problematic device stacking, clear the queue flag
1939         * for request stacking support until then.
1940         *
1941         * This queue is new, so no concurrency on the queue_flags.
1942         */
1943        queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1944
1945        md->queue->queuedata = md;
1946        md->queue->backing_dev_info.congested_fn = dm_any_congested;
1947        md->queue->backing_dev_info.congested_data = md;
1948        blk_queue_make_request(md->queue, dm_request);
1949        blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1950        blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1951}
1952
1953/*
1954 * Allocate and initialise a blank device with a given minor.
1955 */
1956static struct mapped_device *alloc_dev(int minor)
1957{
1958        int r;
1959        struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1960        void *old_md;
1961
1962        if (!md) {
1963                DMWARN("unable to allocate device, out of memory.");
1964                return NULL;
1965        }
1966
1967        if (!try_module_get(THIS_MODULE))
1968                goto bad_module_get;
1969
1970        /* get a minor number for the dev */
1971        if (minor == DM_ANY_MINOR)
1972                r = next_free_minor(&minor);
1973        else
1974                r = specific_minor(minor);
1975        if (r < 0)
1976                goto bad_minor;
1977
1978        r = init_srcu_struct(&md->io_barrier);
1979        if (r < 0)
1980                goto bad_io_barrier;
1981
1982        md->type = DM_TYPE_NONE;
1983        mutex_init(&md->suspend_lock);
1984        mutex_init(&md->type_lock);
1985        spin_lock_init(&md->deferred_lock);
1986        atomic_set(&md->holders, 1);
1987        atomic_set(&md->open_count, 0);
1988        atomic_set(&md->event_nr, 0);
1989        atomic_set(&md->uevent_seq, 0);
1990        INIT_LIST_HEAD(&md->uevent_list);
1991        spin_lock_init(&md->uevent_lock);
1992
1993        md->queue = blk_alloc_queue(GFP_KERNEL);
1994        if (!md->queue)
1995                goto bad_queue;
1996
1997        dm_init_md_queue(md);
1998
1999        md->disk = alloc_disk(1);
2000        if (!md->disk)
2001                goto bad_disk;
2002
2003        atomic_set(&md->pending[0], 0);
2004        atomic_set(&md->pending[1], 0);
2005        init_waitqueue_head(&md->wait);
2006        INIT_WORK(&md->work, dm_wq_work);
2007        init_waitqueue_head(&md->eventq);
2008
2009        md->disk->major = _major;
2010        md->disk->first_minor = minor;
2011        md->disk->fops = &dm_blk_dops;
2012        md->disk->queue = md->queue;
2013        md->disk->private_data = md;
2014        sprintf(md->disk->disk_name, "dm-%d", minor);
2015        add_disk(md->disk);
2016        format_dev_t(md->name, MKDEV(_major, minor));
2017
2018        md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2019        if (!md->wq)
2020                goto bad_thread;
2021
2022        md->bdev = bdget_disk(md->disk, 0);
2023        if (!md->bdev)
2024                goto bad_bdev;
2025
2026        bio_init(&md->flush_bio);
2027        md->flush_bio.bi_bdev = md->bdev;
2028        md->flush_bio.bi_rw = WRITE_FLUSH;
2029
2030        dm_stats_init(&md->stats);
2031
2032        /* Populate the mapping, nobody knows we exist yet */
2033        spin_lock(&_minor_lock);
2034        old_md = idr_replace(&_minor_idr, md, minor);
2035        spin_unlock(&_minor_lock);
2036
2037        BUG_ON(old_md != MINOR_ALLOCED);
2038
2039        return md;
2040
2041bad_bdev:
2042        destroy_workqueue(md->wq);
2043bad_thread:
2044        del_gendisk(md->disk);
2045        put_disk(md->disk);
2046bad_disk:
2047        blk_cleanup_queue(md->queue);
2048bad_queue:
2049        cleanup_srcu_struct(&md->io_barrier);
2050bad_io_barrier:
2051        free_minor(minor);
2052bad_minor:
2053        module_put(THIS_MODULE);
2054bad_module_get:
2055        kfree(md);
2056        return NULL;
2057}
2058
2059static void unlock_fs(struct mapped_device *md);
2060
2061static void free_dev(struct mapped_device *md)
2062{
2063        int minor = MINOR(disk_devt(md->disk));
2064
2065        unlock_fs(md);
2066        bdput(md->bdev);
2067        destroy_workqueue(md->wq);
2068        if (md->io_pool)
2069                mempool_destroy(md->io_pool);
2070        if (md->bs)
2071                bioset_free(md->bs);
2072        blk_integrity_unregister(md->disk);
2073        del_gendisk(md->disk);
2074        cleanup_srcu_struct(&md->io_barrier);
2075        free_minor(minor);
2076
2077        spin_lock(&_minor_lock);
2078        md->disk->private_data = NULL;
2079        spin_unlock(&_minor_lock);
2080
2081        put_disk(md->disk);
2082        blk_cleanup_queue(md->queue);
2083        dm_stats_cleanup(&md->stats);
2084        module_put(THIS_MODULE);
2085        kfree(md);
2086}
2087
2088static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2089{
2090        struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2091
2092        if (md->io_pool && md->bs) {
2093                /* The md already has necessary mempools. */
2094                if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2095                        /*
2096                         * Reload bioset because front_pad may have changed
2097                         * because a different table was loaded.
2098                         */
2099                        bioset_free(md->bs);
2100                        md->bs = p->bs;
2101                        p->bs = NULL;
2102                } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2103                        /*
2104                         * There's no need to reload with request-based dm
2105                         * because the size of front_pad doesn't change.
2106                         * Note for future: If you are to reload bioset,
2107                         * prep-ed requests in the queue may refer
2108                         * to bio from the old bioset, so you must walk
2109                         * through the queue to unprep.
2110                         */
2111                }
2112                goto out;
2113        }
2114
2115        BUG_ON(!p || md->io_pool || md->bs);
2116
2117        md->io_pool = p->io_pool;
2118        p->io_pool = NULL;
2119        md->bs = p->bs;
2120        p->bs = NULL;
2121
2122out:
2123        /* mempool bind completed, now no need any mempools in the table */
2124        dm_table_free_md_mempools(t);
2125}
2126
2127/*
2128 * Bind a table to the device.
2129 */
2130static void event_callback(void *context)
2131{
2132        unsigned long flags;
2133        LIST_HEAD(uevents);
2134        struct mapped_device *md = (struct mapped_device *) context;
2135
2136        spin_lock_irqsave(&md->uevent_lock, flags);
2137        list_splice_init(&md->uevent_list, &uevents);
2138        spin_unlock_irqrestore(&md->uevent_lock, flags);
2139
2140        dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2141
2142        atomic_inc(&md->event_nr);
2143        wake_up(&md->eventq);
2144}
2145
2146/*
2147 * Protected by md->suspend_lock obtained by dm_swap_table().
2148 */
2149static void __set_size(struct mapped_device *md, sector_t size)
2150{
2151        set_capacity(md->disk, size);
2152
2153        i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2154}
2155
2156/*
2157 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2158 *
2159 * If this function returns 0, then the device is either a non-dm
2160 * device without a merge_bvec_fn, or it is a dm device that is
2161 * able to split any bios it receives that are too big.
2162 */
2163int dm_queue_merge_is_compulsory(struct request_queue *q)
2164{
2165        struct mapped_device *dev_md;
2166
2167        if (!q->merge_bvec_fn)
2168                return 0;
2169
2170        if (q->make_request_fn == dm_request) {
2171                dev_md = q->queuedata;
2172                if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2173                        return 0;
2174        }
2175
2176        return 1;
2177}
2178
2179static int dm_device_merge_is_compulsory(struct dm_target *ti,
2180                                         struct dm_dev *dev, sector_t start,
2181                                         sector_t len, void *data)
2182{
2183        struct block_device *bdev = dev->bdev;
2184        struct request_queue *q = bdev_get_queue(bdev);
2185
2186        return dm_queue_merge_is_compulsory(q);
2187}
2188
2189/*
2190 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2191 * on the properties of the underlying devices.
2192 */
2193static int dm_table_merge_is_optional(struct dm_table *table)
2194{
2195        unsigned i = 0;
2196        struct dm_target *ti;
2197
2198        while (i < dm_table_get_num_targets(table)) {
2199                ti = dm_table_get_target(table, i++);
2200
2201                if (ti->type->iterate_devices &&
2202                    ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2203                        return 0;
2204        }
2205
2206        return 1;
2207}
2208
2209/*
2210 * Returns old map, which caller must destroy.
2211 */
2212static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2213                               struct queue_limits *limits)
2214{
2215        struct dm_table *old_map;
2216        struct request_queue *q = md->queue;
2217        sector_t size;
2218        int merge_is_optional;
2219
2220        size = dm_table_get_size(t);
2221
2222        /*
2223         * Wipe any geometry if the size of the table changed.
2224         */
2225        if (size != dm_get_size(md))
2226                memset(&md->geometry, 0, sizeof(md->geometry));
2227
2228        __set_size(md, size);
2229
2230        dm_table_event_callback(t, event_callback, md);
2231
2232        /*
2233         * The queue hasn't been stopped yet, if the old table type wasn't
2234         * for request-based during suspension.  So stop it to prevent
2235         * I/O mapping before resume.
2236         * This must be done before setting the queue restrictions,
2237         * because request-based dm may be run just after the setting.
2238         */
2239        if (dm_table_request_based(t) && !blk_queue_stopped(q))
2240                stop_queue(q);
2241
2242        __bind_mempools(md, t);
2243
2244        merge_is_optional = dm_table_merge_is_optional(t);
2245
2246        old_map = md->map;
2247        rcu_assign_pointer(md->map, t);
2248        md->immutable_target_type = dm_table_get_immutable_target_type(t);
2249
2250        dm_table_set_restrictions(t, q, limits);
2251        if (merge_is_optional)
2252                set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2253        else
2254                clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2255        dm_sync_table(md);
2256
2257        return old_map;
2258}
2259
2260/*
2261 * Returns unbound table for the caller to free.
2262 */
2263static struct dm_table *__unbind(struct mapped_device *md)
2264{
2265        struct dm_table *map = md->map;
2266
2267        if (!map)
2268                return NULL;
2269
2270        dm_table_event_callback(map, NULL, NULL);
2271        rcu_assign_pointer(md->map, NULL);
2272        dm_sync_table(md);
2273
2274        return map;
2275}
2276
2277/*
2278 * Constructor for a new device.
2279 */
2280int dm_create(int minor, struct mapped_device **result)
2281{
2282        struct mapped_device *md;
2283
2284        md = alloc_dev(minor);
2285        if (!md)
2286                return -ENXIO;
2287
2288        dm_sysfs_init(md);
2289
2290        *result = md;
2291        return 0;
2292}
2293
2294/*
2295 * Functions to manage md->type.
2296 * All are required to hold md->type_lock.
2297 */
2298void dm_lock_md_type(struct mapped_device *md)
2299{
2300        mutex_lock(&md->type_lock);
2301}
2302
2303void dm_unlock_md_type(struct mapped_device *md)
2304{
2305        mutex_unlock(&md->type_lock);
2306}
2307
2308void dm_set_md_type(struct mapped_device *md, unsigned type)
2309{
2310        BUG_ON(!mutex_is_locked(&md->type_lock));
2311        md->type = type;
2312}
2313
2314unsigned dm_get_md_type(struct mapped_device *md)
2315{
2316        BUG_ON(!mutex_is_locked(&md->type_lock));
2317        return md->type;
2318}
2319
2320struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2321{
2322        return md->immutable_target_type;
2323}
2324
2325/*
2326 * The queue_limits are only valid as long as you have a reference
2327 * count on 'md'.
2328 */
2329struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2330{
2331        BUG_ON(!atomic_read(&md->holders));
2332        return &md->queue->limits;
2333}
2334EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2335
2336/*
2337 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2338 */
2339static int dm_init_request_based_queue(struct mapped_device *md)
2340{
2341        struct request_queue *q = NULL;
2342
2343        if (md->queue->elevator)
2344                return 1;
2345
2346        /* Fully initialize the queue */
2347        q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2348        if (!q)
2349                return 0;
2350
2351        md->queue = q;
2352        dm_init_md_queue(md);
2353        blk_queue_softirq_done(md->queue, dm_softirq_done);
2354        blk_queue_prep_rq(md->queue, dm_prep_fn);
2355        blk_queue_lld_busy(md->queue, dm_lld_busy);
2356
2357        elv_register_queue(md->queue);
2358
2359        return 1;
2360}
2361
2362/*
2363 * Setup the DM device's queue based on md's type
2364 */
2365int dm_setup_md_queue(struct mapped_device *md)
2366{
2367        if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2368            !dm_init_request_based_queue(md)) {
2369                DMWARN("Cannot initialize queue for request-based mapped device");
2370                return -EINVAL;
2371        }
2372
2373        return 0;
2374}
2375
2376static struct mapped_device *dm_find_md(dev_t dev)
2377{
2378        struct mapped_device *md;
2379        unsigned minor = MINOR(dev);
2380
2381        if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2382                return NULL;
2383
2384        spin_lock(&_minor_lock);
2385
2386        md = idr_find(&_minor_idr, minor);
2387        if (md && (md == MINOR_ALLOCED ||
2388                   (MINOR(disk_devt(dm_disk(md))) != minor) ||
2389                   dm_deleting_md(md) ||
2390                   test_bit(DMF_FREEING, &md->flags))) {
2391                md = NULL;
2392                goto out;
2393        }
2394
2395out:
2396        spin_unlock(&_minor_lock);
2397
2398        return md;
2399}
2400
2401struct mapped_device *dm_get_md(dev_t dev)
2402{
2403        struct mapped_device *md = dm_find_md(dev);
2404
2405        if (md)
2406                dm_get(md);
2407
2408        return md;
2409}
2410EXPORT_SYMBOL_GPL(dm_get_md);
2411
2412void *dm_get_mdptr(struct mapped_device *md)
2413{
2414        return md->interface_ptr;
2415}
2416
2417void dm_set_mdptr(struct mapped_device *md, void *ptr)
2418{
2419        md->interface_ptr = ptr;
2420}
2421
2422void dm_get(struct mapped_device *md)
2423{
2424        atomic_inc(&md->holders);
2425        BUG_ON(test_bit(DMF_FREEING, &md->flags));
2426}
2427
2428const char *dm_device_name(struct mapped_device *md)
2429{
2430        return md->name;
2431}
2432EXPORT_SYMBOL_GPL(dm_device_name);
2433
2434static void __dm_destroy(struct mapped_device *md, bool wait)
2435{
2436        struct dm_table *map;
2437        int srcu_idx;
2438
2439        might_sleep();
2440
2441        spin_lock(&_minor_lock);
2442        map = dm_get_live_table(md, &srcu_idx);
2443        idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2444        set_bit(DMF_FREEING, &md->flags);
2445        spin_unlock(&_minor_lock);
2446
2447        if (!dm_suspended_md(md)) {
2448                dm_table_presuspend_targets(map);
2449                dm_table_postsuspend_targets(map);
2450        }
2451
2452        /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2453        dm_put_live_table(md, srcu_idx);
2454
2455        /*
2456         * Rare, but there may be I/O requests still going to complete,
2457         * for example.  Wait for all references to disappear.
2458         * No one should increment the reference count of the mapped_device,
2459         * after the mapped_device state becomes DMF_FREEING.
2460         */
2461        if (wait)
2462                while (atomic_read(&md->holders))
2463                        msleep(1);
2464        else if (atomic_read(&md->holders))
2465                DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2466                       dm_device_name(md), atomic_read(&md->holders));
2467
2468        dm_sysfs_exit(md);
2469        dm_table_destroy(__unbind(md));
2470        free_dev(md);
2471}
2472
2473void dm_destroy(struct mapped_device *md)
2474{
2475        __dm_destroy(md, true);
2476}
2477
2478void dm_destroy_immediate(struct mapped_device *md)
2479{
2480        __dm_destroy(md, false);
2481}
2482
2483void dm_put(struct mapped_device *md)
2484{
2485        atomic_dec(&md->holders);
2486}
2487EXPORT_SYMBOL_GPL(dm_put);
2488
2489static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2490{
2491        int r = 0;
2492        DECLARE_WAITQUEUE(wait, current);
2493
2494        add_wait_queue(&md->wait, &wait);
2495
2496        while (1) {
2497                set_current_state(interruptible);
2498
2499                if (!md_in_flight(md))
2500                        break;
2501
2502                if (interruptible == TASK_INTERRUPTIBLE &&
2503                    signal_pending(current)) {
2504                        r = -EINTR;
2505                        break;
2506                }
2507
2508                io_schedule();
2509        }
2510        set_current_state(TASK_RUNNING);
2511
2512        remove_wait_queue(&md->wait, &wait);
2513
2514        return r;
2515}
2516
2517/*
2518 * Process the deferred bios
2519 */
2520static void dm_wq_work(struct work_struct *work)
2521{
2522        struct mapped_device *md = container_of(work, struct mapped_device,
2523                                                work);
2524        struct bio *c;
2525        int srcu_idx;
2526        struct dm_table *map;
2527
2528        map = dm_get_live_table(md, &srcu_idx);
2529
2530        while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2531                spin_lock_irq(&md->deferred_lock);
2532                c = bio_list_pop(&md->deferred);
2533                spin_unlock_irq(&md->deferred_lock);
2534
2535                if (!c)
2536                        break;
2537
2538                if (dm_request_based(md))
2539                        generic_make_request(c);
2540                else
2541                        __split_and_process_bio(md, map, c);
2542        }
2543
2544        dm_put_live_table(md, srcu_idx);
2545}
2546
2547static void dm_queue_flush(struct mapped_device *md)
2548{
2549        clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2550        smp_mb__after_clear_bit();
2551        queue_work(md->wq, &md->work);
2552}
2553
2554/*
2555 * Swap in a new table, returning the old one for the caller to destroy.
2556 */
2557struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2558{
2559        struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2560        struct queue_limits limits;
2561        int r;
2562
2563        mutex_lock(&md->suspend_lock);
2564
2565        /* device must be suspended */
2566        if (!dm_suspended_md(md))
2567                goto out;
2568
2569        /*
2570         * If the new table has no data devices, retain the existing limits.
2571         * This helps multipath with queue_if_no_path if all paths disappear,
2572         * then new I/O is queued based on these limits, and then some paths
2573         * reappear.
2574         */
2575        if (dm_table_has_no_data_devices(table)) {
2576                live_map = dm_get_live_table_fast(md);
2577                if (live_map)
2578                        limits = md->queue->limits;
2579                dm_put_live_table_fast(md);
2580        }
2581
2582        if (!live_map) {
2583                r = dm_calculate_queue_limits(table, &limits);
2584                if (r) {
2585                        map = ERR_PTR(r);
2586                        goto out;
2587                }
2588        }
2589
2590        map = __bind(md, table, &limits);
2591
2592out:
2593        mutex_unlock(&md->suspend_lock);
2594        return map;
2595}
2596
2597/*
2598 * Functions to lock and unlock any filesystem running on the
2599 * device.
2600 */
2601static int lock_fs(struct mapped_device *md)
2602{
2603        int r;
2604
2605        WARN_ON(md->frozen_sb);
2606
2607        md->frozen_sb = freeze_bdev(md->bdev);
2608        if (IS_ERR(md->frozen_sb)) {
2609                r = PTR_ERR(md->frozen_sb);
2610                md->frozen_sb = NULL;
2611                return r;
2612        }
2613
2614        set_bit(DMF_FROZEN, &md->flags);
2615
2616        return 0;
2617}
2618
2619static void unlock_fs(struct mapped_device *md)
2620{
2621        if (!test_bit(DMF_FROZEN, &md->flags))
2622                return;
2623
2624        thaw_bdev(md->bdev, md->frozen_sb);
2625        md->frozen_sb = NULL;
2626        clear_bit(DMF_FROZEN, &md->flags);
2627}
2628
2629/*
2630 * We need to be able to change a mapping table under a mounted
2631 * filesystem.  For example we might want to move some data in
2632 * the background.  Before the table can be swapped with
2633 * dm_bind_table, dm_suspend must be called to flush any in
2634 * flight bios and ensure that any further io gets deferred.
2635 */
2636/*
2637 * Suspend mechanism in request-based dm.
2638 *
2639 * 1. Flush all I/Os by lock_fs() if needed.
2640 * 2. Stop dispatching any I/O by stopping the request_queue.
2641 * 3. Wait for all in-flight I/Os to be completed or requeued.
2642 *
2643 * To abort suspend, start the request_queue.
2644 */
2645int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2646{
2647        struct dm_table *map = NULL;
2648        int r = 0;
2649        int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2650        int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2651
2652        mutex_lock(&md->suspend_lock);
2653
2654        if (dm_suspended_md(md)) {
2655                r = -EINVAL;
2656                goto out_unlock;
2657        }
2658
2659        map = md->map;
2660
2661        /*
2662         * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2663         * This flag is cleared before dm_suspend returns.
2664         */
2665        if (noflush)
2666                set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2667
2668        /* This does not get reverted if there's an error later. */
2669        dm_table_presuspend_targets(map);
2670
2671        /*
2672         * Flush I/O to the device.
2673         * Any I/O submitted after lock_fs() may not be flushed.
2674         * noflush takes precedence over do_lockfs.
2675         * (lock_fs() flushes I/Os and waits for them to complete.)
2676         */
2677        if (!noflush && do_lockfs) {
2678                r = lock_fs(md);
2679                if (r)
2680                        goto out_unlock;
2681        }
2682
2683        /*
2684         * Here we must make sure that no processes are submitting requests
2685         * to target drivers i.e. no one may be executing
2686         * __split_and_process_bio. This is called from dm_request and
2687         * dm_wq_work.
2688         *
2689         * To get all processes out of __split_and_process_bio in dm_request,
2690         * we take the write lock. To prevent any process from reentering
2691         * __split_and_process_bio from dm_request and quiesce the thread
2692         * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2693         * flush_workqueue(md->wq).
2694         */
2695        set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2696        synchronize_srcu(&md->io_barrier);
2697
2698        /*
2699         * Stop md->queue before flushing md->wq in case request-based
2700         * dm defers requests to md->wq from md->queue.
2701         */
2702        if (dm_request_based(md))
2703                stop_queue(md->queue);
2704
2705        flush_workqueue(md->wq);
2706
2707        /*
2708         * At this point no more requests are entering target request routines.
2709         * We call dm_wait_for_completion to wait for all existing requests
2710         * to finish.
2711         */
2712        r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2713
2714        if (noflush)
2715                clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2716        synchronize_srcu(&md->io_barrier);
2717
2718        /* were we interrupted ? */
2719        if (r < 0) {
2720                dm_queue_flush(md);
2721
2722                if (dm_request_based(md))
2723                        start_queue(md->queue);
2724
2725                unlock_fs(md);
2726                goto out_unlock; /* pushback list is already flushed, so skip flush */
2727        }
2728
2729        /*
2730         * If dm_wait_for_completion returned 0, the device is completely
2731         * quiescent now. There is no request-processing activity. All new
2732         * requests are being added to md->deferred list.
2733         */
2734
2735        set_bit(DMF_SUSPENDED, &md->flags);
2736
2737        dm_table_postsuspend_targets(map);
2738
2739out_unlock:
2740        mutex_unlock(&md->suspend_lock);
2741        return r;
2742}
2743
2744int dm_resume(struct mapped_device *md)
2745{
2746        int r = -EINVAL;
2747        struct dm_table *map = NULL;
2748
2749        mutex_lock(&md->suspend_lock);
2750        if (!dm_suspended_md(md))
2751                goto out;
2752
2753        map = md->map;
2754        if (!map || !dm_table_get_size(map))
2755                goto out;
2756
2757        r = dm_table_resume_targets(map);
2758        if (r)
2759                goto out;
2760
2761        dm_queue_flush(md);
2762
2763        /*
2764         * Flushing deferred I/Os must be done after targets are resumed
2765         * so that mapping of targets can work correctly.
2766         * Request-based dm is queueing the deferred I/Os in its request_queue.
2767         */
2768        if (dm_request_based(md))
2769                start_queue(md->queue);
2770
2771        unlock_fs(md);
2772
2773        clear_bit(DMF_SUSPENDED, &md->flags);
2774
2775        r = 0;
2776out:
2777        mutex_unlock(&md->suspend_lock);
2778
2779        return r;
2780}
2781
2782/*
2783 * Internal suspend/resume works like userspace-driven suspend. It waits
2784 * until all bios finish and prevents issuing new bios to the target drivers.
2785 * It may be used only from the kernel.
2786 *
2787 * Internal suspend holds md->suspend_lock, which prevents interaction with
2788 * userspace-driven suspend.
2789 */
2790
2791void dm_internal_suspend(struct mapped_device *md)
2792{
2793        mutex_lock(&md->suspend_lock);
2794        if (dm_suspended_md(md))
2795                return;
2796
2797        set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2798        synchronize_srcu(&md->io_barrier);
2799        flush_workqueue(md->wq);
2800        dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2801}
2802
2803void dm_internal_resume(struct mapped_device *md)
2804{
2805        if (dm_suspended_md(md))
2806                goto done;
2807
2808        dm_queue_flush(md);
2809
2810done:
2811        mutex_unlock(&md->suspend_lock);
2812}
2813
2814/*-----------------------------------------------------------------
2815 * Event notification.
2816 *---------------------------------------------------------------*/
2817int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2818                       unsigned cookie)
2819{
2820        char udev_cookie[DM_COOKIE_LENGTH];
2821        char *envp[] = { udev_cookie, NULL };
2822
2823        if (!cookie)
2824                return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2825        else {
2826                snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2827                         DM_COOKIE_ENV_VAR_NAME, cookie);
2828                return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2829                                          action, envp);
2830        }
2831}
2832
2833uint32_t dm_next_uevent_seq(struct mapped_device *md)
2834{
2835        return atomic_add_return(1, &md->uevent_seq);
2836}
2837
2838uint32_t dm_get_event_nr(struct mapped_device *md)
2839{
2840        return atomic_read(&md->event_nr);
2841}
2842
2843int dm_wait_event(struct mapped_device *md, int event_nr)
2844{
2845        return wait_event_interruptible(md->eventq,
2846                        (event_nr != atomic_read(&md->event_nr)));
2847}
2848
2849void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2850{
2851        unsigned long flags;
2852
2853        spin_lock_irqsave(&md->uevent_lock, flags);
2854        list_add(elist, &md->uevent_list);
2855        spin_unlock_irqrestore(&md->uevent_lock, flags);
2856}
2857
2858/*
2859 * The gendisk is only valid as long as you have a reference
2860 * count on 'md'.
2861 */
2862struct gendisk *dm_disk(struct mapped_device *md)
2863{
2864        return md->disk;
2865}
2866
2867struct kobject *dm_kobject(struct mapped_device *md)
2868{
2869        return &md->kobj;
2870}
2871
2872/*
2873 * struct mapped_device should not be exported outside of dm.c
2874 * so use this check to verify that kobj is part of md structure
2875 */
2876struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2877{
2878        struct mapped_device *md;
2879
2880        md = container_of(kobj, struct mapped_device, kobj);
2881        if (&md->kobj != kobj)
2882                return NULL;
2883
2884        if (test_bit(DMF_FREEING, &md->flags) ||
2885            dm_deleting_md(md))
2886                return NULL;
2887
2888        dm_get(md);
2889        return md;
2890}
2891
2892int dm_suspended_md(struct mapped_device *md)
2893{
2894        return test_bit(DMF_SUSPENDED, &md->flags);
2895}
2896
2897int dm_suspended(struct dm_target *ti)
2898{
2899        return dm_suspended_md(dm_table_get_md(ti->table));
2900}
2901EXPORT_SYMBOL_GPL(dm_suspended);
2902
2903int dm_noflush_suspending(struct dm_target *ti)
2904{
2905        return __noflush_suspending(dm_table_get_md(ti->table));
2906}
2907EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2908
2909struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2910{
2911        struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2912        struct kmem_cache *cachep;
2913        unsigned int pool_size;
2914        unsigned int front_pad;
2915
2916        if (!pools)
2917                return NULL;
2918
2919        if (type == DM_TYPE_BIO_BASED) {
2920                cachep = _io_cache;
2921                pool_size = dm_get_reserved_bio_based_ios();
2922                front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2923        } else if (type == DM_TYPE_REQUEST_BASED) {
2924                cachep = _rq_tio_cache;
2925                pool_size = dm_get_reserved_rq_based_ios();
2926                front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2927                /* per_bio_data_size is not used. See __bind_mempools(). */
2928                WARN_ON(per_bio_data_size != 0);
2929        } else
2930                goto out;
2931
2932        pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2933        if (!pools->io_pool)
2934                goto out;
2935
2936        pools->bs = bioset_create(pool_size, front_pad);
2937        if (!pools->bs)
2938                goto out;
2939
2940        if (integrity && bioset_integrity_create(pools->bs, pool_size))
2941                goto out;
2942
2943        return pools;
2944
2945out:
2946        dm_free_md_mempools(pools);
2947
2948        return NULL;
2949}
2950
2951void dm_free_md_mempools(struct dm_md_mempools *pools)
2952{
2953        if (!pools)
2954                return;
2955
2956        if (pools->io_pool)
2957                mempool_destroy(pools->io_pool);
2958
2959        if (pools->bs)
2960                bioset_free(pools->bs);
2961
2962        kfree(pools);
2963}
2964
2965static const struct block_device_operations dm_blk_dops = {
2966        .open = dm_blk_open,
2967        .release = dm_blk_close,
2968        .ioctl = dm_blk_ioctl,
2969        .getgeo = dm_blk_getgeo,
2970        .owner = THIS_MODULE
2971};
2972
2973EXPORT_SYMBOL(dm_get_mapinfo);
2974
2975/*
2976 * module hooks
2977 */
2978module_init(dm_init);
2979module_exit(dm_exit);
2980
2981module_param(major, uint, 0);
2982MODULE_PARM_DESC(major, "The major number of the device mapper");
2983
2984module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2985MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2986
2987module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2988MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2989
2990MODULE_DESCRIPTION(DM_NAME " driver");
2991MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2992MODULE_LICENSE("GPL");
2993
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.