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