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