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