linux/block/blk-core.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 1991, 1992 Linus Torvalds
   3 * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
   4 * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
   5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
   6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
   7 *      -  July2000
   8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
   9 */
  10
  11/*
  12 * This handles all read/write requests to block devices
  13 */
  14#include <linux/kernel.h>
  15#include <linux/module.h>
  16#include <linux/backing-dev.h>
  17#include <linux/bio.h>
  18#include <linux/blkdev.h>
  19#include <linux/highmem.h>
  20#include <linux/mm.h>
  21#include <linux/kernel_stat.h>
  22#include <linux/string.h>
  23#include <linux/init.h>
  24#include <linux/completion.h>
  25#include <linux/slab.h>
  26#include <linux/swap.h>
  27#include <linux/writeback.h>
  28#include <linux/task_io_accounting_ops.h>
  29#include <linux/fault-inject.h>
  30#include <linux/list_sort.h>
  31
  32#define CREATE_TRACE_POINTS
  33#include <trace/events/block.h>
  34
  35#include "blk.h"
  36
  37EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
  38EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
  39EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
  40
  41static int __make_request(struct request_queue *q, struct bio *bio);
  42
  43/*
  44 * For the allocated request tables
  45 */
  46static struct kmem_cache *request_cachep;
  47
  48/*
  49 * For queue allocation
  50 */
  51struct kmem_cache *blk_requestq_cachep;
  52
  53/*
  54 * Controlling structure to kblockd
  55 */
  56static struct workqueue_struct *kblockd_workqueue;
  57
  58static void drive_stat_acct(struct request *rq, int new_io)
  59{
  60        struct hd_struct *part;
  61        int rw = rq_data_dir(rq);
  62        int cpu;
  63
  64        if (!blk_do_io_stat(rq))
  65                return;
  66
  67        cpu = part_stat_lock();
  68
  69        if (!new_io) {
  70                part = rq->part;
  71                part_stat_inc(cpu, part, merges[rw]);
  72        } else {
  73                part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
  74                if (!hd_struct_try_get(part)) {
  75                        /*
  76                         * The partition is already being removed,
  77                         * the request will be accounted on the disk only
  78                         *
  79                         * We take a reference on disk->part0 although that
  80                         * partition will never be deleted, so we can treat
  81                         * it as any other partition.
  82                         */
  83                        part = &rq->rq_disk->part0;
  84                        hd_struct_get(part);
  85                }
  86                part_round_stats(cpu, part);
  87                part_inc_in_flight(part, rw);
  88                rq->part = part;
  89        }
  90
  91        part_stat_unlock();
  92}
  93
  94void blk_queue_congestion_threshold(struct request_queue *q)
  95{
  96        int nr;
  97
  98        nr = q->nr_requests - (q->nr_requests / 8) + 1;
  99        if (nr > q->nr_requests)
 100                nr = q->nr_requests;
 101        q->nr_congestion_on = nr;
 102
 103        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
 104        if (nr < 1)
 105                nr = 1;
 106        q->nr_congestion_off = nr;
 107}
 108
 109/**
 110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
 111 * @bdev:       device
 112 *
 113 * Locates the passed device's request queue and returns the address of its
 114 * backing_dev_info
 115 *
 116 * Will return NULL if the request queue cannot be located.
 117 */
 118struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
 119{
 120        struct backing_dev_info *ret = NULL;
 121        struct request_queue *q = bdev_get_queue(bdev);
 122
 123        if (q)
 124                ret = &q->backing_dev_info;
 125        return ret;
 126}
 127EXPORT_SYMBOL(blk_get_backing_dev_info);
 128
 129void blk_rq_init(struct request_queue *q, struct request *rq)
 130{
 131        memset(rq, 0, sizeof(*rq));
 132
 133        INIT_LIST_HEAD(&rq->queuelist);
 134        INIT_LIST_HEAD(&rq->timeout_list);
 135        rq->cpu = -1;
 136        rq->q = q;
 137        rq->__sector = (sector_t) -1;
 138        INIT_HLIST_NODE(&rq->hash);
 139        RB_CLEAR_NODE(&rq->rb_node);
 140        rq->cmd = rq->__cmd;
 141        rq->cmd_len = BLK_MAX_CDB;
 142        rq->tag = -1;
 143        rq->ref_count = 1;
 144        rq->start_time = jiffies;
 145        set_start_time_ns(rq);
 146        rq->part = NULL;
 147}
 148EXPORT_SYMBOL(blk_rq_init);
 149
 150static void req_bio_endio(struct request *rq, struct bio *bio,
 151                          unsigned int nbytes, int error)
 152{
 153        if (error)
 154                clear_bit(BIO_UPTODATE, &bio->bi_flags);
 155        else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 156                error = -EIO;
 157
 158        if (unlikely(nbytes > bio->bi_size)) {
 159                printk(KERN_ERR "%s: want %u bytes done, %u left\n",
 160                       __func__, nbytes, bio->bi_size);
 161                nbytes = bio->bi_size;
 162        }
 163
 164        if (unlikely(rq->cmd_flags & REQ_QUIET))
 165                set_bit(BIO_QUIET, &bio->bi_flags);
 166
 167        bio->bi_size -= nbytes;
 168        bio->bi_sector += (nbytes >> 9);
 169
 170        if (bio_integrity(bio))
 171                bio_integrity_advance(bio, nbytes);
 172
 173        /* don't actually finish bio if it's part of flush sequence */
 174        if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
 175                bio_endio(bio, error);
 176}
 177
 178void blk_dump_rq_flags(struct request *rq, char *msg)
 179{
 180        int bit;
 181
 182        printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
 183                rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
 184                rq->cmd_flags);
 185
 186        printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
 187               (unsigned long long)blk_rq_pos(rq),
 188               blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
 189        printk(KERN_INFO "  bio %p, biotail %p, buffer %p, len %u\n",
 190               rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
 191
 192        if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 193                printk(KERN_INFO "  cdb: ");
 194                for (bit = 0; bit < BLK_MAX_CDB; bit++)
 195                        printk("%02x ", rq->cmd[bit]);
 196                printk("\n");
 197        }
 198}
 199EXPORT_SYMBOL(blk_dump_rq_flags);
 200
 201static void blk_delay_work(struct work_struct *work)
 202{
 203        struct request_queue *q;
 204
 205        q = container_of(work, struct request_queue, delay_work.work);
 206        spin_lock_irq(q->queue_lock);
 207        __blk_run_queue(q);
 208        spin_unlock_irq(q->queue_lock);
 209}
 210
 211/**
 212 * blk_delay_queue - restart queueing after defined interval
 213 * @q:          The &struct request_queue in question
 214 * @msecs:      Delay in msecs
 215 *
 216 * Description:
 217 *   Sometimes queueing needs to be postponed for a little while, to allow
 218 *   resources to come back. This function will make sure that queueing is
 219 *   restarted around the specified time.
 220 */
 221void blk_delay_queue(struct request_queue *q, unsigned long msecs)
 222{
 223        queue_delayed_work(kblockd_workqueue, &q->delay_work,
 224                                msecs_to_jiffies(msecs));
 225}
 226EXPORT_SYMBOL(blk_delay_queue);
 227
 228/**
 229 * blk_start_queue - restart a previously stopped queue
 230 * @q:    The &struct request_queue in question
 231 *
 232 * Description:
 233 *   blk_start_queue() will clear the stop flag on the queue, and call
 234 *   the request_fn for the queue if it was in a stopped state when
 235 *   entered. Also see blk_stop_queue(). Queue lock must be held.
 236 **/
 237void blk_start_queue(struct request_queue *q)
 238{
 239        WARN_ON(!irqs_disabled());
 240
 241        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 242        __blk_run_queue(q);
 243}
 244EXPORT_SYMBOL(blk_start_queue);
 245
 246/**
 247 * blk_stop_queue - stop a queue
 248 * @q:    The &struct request_queue in question
 249 *
 250 * Description:
 251 *   The Linux block layer assumes that a block driver will consume all
 252 *   entries on the request queue when the request_fn strategy is called.
 253 *   Often this will not happen, because of hardware limitations (queue
 254 *   depth settings). If a device driver gets a 'queue full' response,
 255 *   or if it simply chooses not to queue more I/O at one point, it can
 256 *   call this function to prevent the request_fn from being called until
 257 *   the driver has signalled it's ready to go again. This happens by calling
 258 *   blk_start_queue() to restart queue operations. Queue lock must be held.
 259 **/
 260void blk_stop_queue(struct request_queue *q)
 261{
 262        __cancel_delayed_work(&q->delay_work);
 263        queue_flag_set(QUEUE_FLAG_STOPPED, q);
 264}
 265EXPORT_SYMBOL(blk_stop_queue);
 266
 267/**
 268 * blk_sync_queue - cancel any pending callbacks on a queue
 269 * @q: the queue
 270 *
 271 * Description:
 272 *     The block layer may perform asynchronous callback activity
 273 *     on a queue, such as calling the unplug function after a timeout.
 274 *     A block device may call blk_sync_queue to ensure that any
 275 *     such activity is cancelled, thus allowing it to release resources
 276 *     that the callbacks might use. The caller must already have made sure
 277 *     that its ->make_request_fn will not re-add plugging prior to calling
 278 *     this function.
 279 *
 280 *     This function does not cancel any asynchronous activity arising
 281 *     out of elevator or throttling code. That would require elevaotor_exit()
 282 *     and blk_throtl_exit() to be called with queue lock initialized.
 283 *
 284 */
 285void blk_sync_queue(struct request_queue *q)
 286{
 287        del_timer_sync(&q->timeout);
 288        cancel_delayed_work_sync(&q->delay_work);
 289}
 290EXPORT_SYMBOL(blk_sync_queue);
 291
 292/**
 293 * __blk_run_queue - run a single device queue
 294 * @q:  The queue to run
 295 *
 296 * Description:
 297 *    See @blk_run_queue. This variant must be called with the queue lock
 298 *    held and interrupts disabled.
 299 */
 300void __blk_run_queue(struct request_queue *q)
 301{
 302        if (unlikely(blk_queue_stopped(q)))
 303                return;
 304
 305        q->request_fn(q);
 306}
 307EXPORT_SYMBOL(__blk_run_queue);
 308
 309/**
 310 * blk_run_queue_async - run a single device queue in workqueue context
 311 * @q:  The queue to run
 312 *
 313 * Description:
 314 *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
 315 *    of us.
 316 */
 317void blk_run_queue_async(struct request_queue *q)
 318{
 319        if (likely(!blk_queue_stopped(q))) {
 320                __cancel_delayed_work(&q->delay_work);
 321                queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
 322        }
 323}
 324EXPORT_SYMBOL(blk_run_queue_async);
 325
 326/**
 327 * blk_run_queue - run a single device queue
 328 * @q: The queue to run
 329 *
 330 * Description:
 331 *    Invoke request handling on this queue, if it has pending work to do.
 332 *    May be used to restart queueing when a request has completed.
 333 */
 334void blk_run_queue(struct request_queue *q)
 335{
 336        unsigned long flags;
 337
 338        spin_lock_irqsave(q->queue_lock, flags);
 339        __blk_run_queue(q);
 340        spin_unlock_irqrestore(q->queue_lock, flags);
 341}
 342EXPORT_SYMBOL(blk_run_queue);
 343
 344void blk_put_queue(struct request_queue *q)
 345{
 346        kobject_put(&q->kobj);
 347}
 348EXPORT_SYMBOL(blk_put_queue);
 349
 350/*
 351 * Note: If a driver supplied the queue lock, it should not zap that lock
 352 * unexpectedly as some queue cleanup components like elevator_exit() and
 353 * blk_throtl_exit() need queue lock.
 354 */
 355void blk_cleanup_queue(struct request_queue *q)
 356{
 357        /*
 358         * We know we have process context here, so we can be a little
 359         * cautious and ensure that pending block actions on this device
 360         * are done before moving on. Going into this function, we should
 361         * not have processes doing IO to this device.
 362         */
 363        blk_sync_queue(q);
 364
 365        del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
 366        mutex_lock(&q->sysfs_lock);
 367        queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
 368        mutex_unlock(&q->sysfs_lock);
 369
 370        if (q->elevator)
 371                elevator_exit(q->elevator);
 372
 373        blk_throtl_exit(q);
 374
 375        blk_put_queue(q);
 376}
 377EXPORT_SYMBOL(blk_cleanup_queue);
 378
 379static int blk_init_free_list(struct request_queue *q)
 380{
 381        struct request_list *rl = &q->rq;
 382
 383        if (unlikely(rl->rq_pool))
 384                return 0;
 385
 386        rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
 387        rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
 388        rl->elvpriv = 0;
 389        init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
 390        init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
 391
 392        rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
 393                                mempool_free_slab, request_cachep, q->node);
 394
 395        if (!rl->rq_pool)
 396                return -ENOMEM;
 397
 398        return 0;
 399}
 400
 401struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 402{
 403        return blk_alloc_queue_node(gfp_mask, -1);
 404}
 405EXPORT_SYMBOL(blk_alloc_queue);
 406
 407struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 408{
 409        struct request_queue *q;
 410        int err;
 411
 412        q = kmem_cache_alloc_node(blk_requestq_cachep,
 413                                gfp_mask | __GFP_ZERO, node_id);
 414        if (!q)
 415                return NULL;
 416
 417        q->backing_dev_info.ra_pages =
 418                        (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 419        q->backing_dev_info.state = 0;
 420        q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
 421        q->backing_dev_info.name = "block";
 422
 423        err = bdi_init(&q->backing_dev_info);
 424        if (err) {
 425                kmem_cache_free(blk_requestq_cachep, q);
 426                return NULL;
 427        }
 428
 429        if (blk_throtl_init(q)) {
 430                kmem_cache_free(blk_requestq_cachep, q);
 431                return NULL;
 432        }
 433
 434        setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
 435                    laptop_mode_timer_fn, (unsigned long) q);
 436        setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
 437        INIT_LIST_HEAD(&q->timeout_list);
 438        INIT_LIST_HEAD(&q->flush_queue[0]);
 439        INIT_LIST_HEAD(&q->flush_queue[1]);
 440        INIT_LIST_HEAD(&q->flush_data_in_flight);
 441        INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
 442
 443        kobject_init(&q->kobj, &blk_queue_ktype);
 444
 445        mutex_init(&q->sysfs_lock);
 446        spin_lock_init(&q->__queue_lock);
 447
 448        /*
 449         * By default initialize queue_lock to internal lock and driver can
 450         * override it later if need be.
 451         */
 452        q->queue_lock = &q->__queue_lock;
 453
 454        return q;
 455}
 456EXPORT_SYMBOL(blk_alloc_queue_node);
 457
 458/**
 459 * blk_init_queue  - prepare a request queue for use with a block device
 460 * @rfn:  The function to be called to process requests that have been
 461 *        placed on the queue.
 462 * @lock: Request queue spin lock
 463 *
 464 * Description:
 465 *    If a block device wishes to use the standard request handling procedures,
 466 *    which sorts requests and coalesces adjacent requests, then it must
 467 *    call blk_init_queue().  The function @rfn will be called when there
 468 *    are requests on the queue that need to be processed.  If the device
 469 *    supports plugging, then @rfn may not be called immediately when requests
 470 *    are available on the queue, but may be called at some time later instead.
 471 *    Plugged queues are generally unplugged when a buffer belonging to one
 472 *    of the requests on the queue is needed, or due to memory pressure.
 473 *
 474 *    @rfn is not required, or even expected, to remove all requests off the
 475 *    queue, but only as many as it can handle at a time.  If it does leave
 476 *    requests on the queue, it is responsible for arranging that the requests
 477 *    get dealt with eventually.
 478 *
 479 *    The queue spin lock must be held while manipulating the requests on the
 480 *    request queue; this lock will be taken also from interrupt context, so irq
 481 *    disabling is needed for it.
 482 *
 483 *    Function returns a pointer to the initialized request queue, or %NULL if
 484 *    it didn't succeed.
 485 *
 486 * Note:
 487 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
 488 *    when the block device is deactivated (such as at module unload).
 489 **/
 490
 491struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
 492{
 493        return blk_init_queue_node(rfn, lock, -1);
 494}
 495EXPORT_SYMBOL(blk_init_queue);
 496
 497struct request_queue *
 498blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
 499{
 500        struct request_queue *uninit_q, *q;
 501
 502        uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
 503        if (!uninit_q)
 504                return NULL;
 505
 506        q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
 507        if (!q)
 508                blk_cleanup_queue(uninit_q);
 509
 510        return q;
 511}
 512EXPORT_SYMBOL(blk_init_queue_node);
 513
 514struct request_queue *
 515blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
 516                         spinlock_t *lock)
 517{
 518        return blk_init_allocated_queue_node(q, rfn, lock, -1);
 519}
 520EXPORT_SYMBOL(blk_init_allocated_queue);
 521
 522struct request_queue *
 523blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
 524                              spinlock_t *lock, int node_id)
 525{
 526        if (!q)
 527                return NULL;
 528
 529        q->node = node_id;
 530        if (blk_init_free_list(q))
 531                return NULL;
 532
 533        q->request_fn           = rfn;
 534        q->prep_rq_fn           = NULL;
 535        q->unprep_rq_fn         = NULL;
 536        q->queue_flags          = QUEUE_FLAG_DEFAULT;
 537
 538        /* Override internal queue lock with supplied lock pointer */
 539        if (lock)
 540                q->queue_lock           = lock;
 541
 542        /*
 543         * This also sets hw/phys segments, boundary and size
 544         */
 545        blk_queue_make_request(q, __make_request);
 546
 547        q->sg_reserved_size = INT_MAX;
 548
 549        /*
 550         * all done
 551         */
 552        if (!elevator_init(q, NULL)) {
 553                blk_queue_congestion_threshold(q);
 554                return q;
 555        }
 556
 557        return NULL;
 558}
 559EXPORT_SYMBOL(blk_init_allocated_queue_node);
 560
 561int blk_get_queue(struct request_queue *q)
 562{
 563        if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
 564                kobject_get(&q->kobj);
 565                return 0;
 566        }
 567
 568        return 1;
 569}
 570EXPORT_SYMBOL(blk_get_queue);
 571
 572static inline void blk_free_request(struct request_queue *q, struct request *rq)
 573{
 574        BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
 575
 576        if (rq->cmd_flags & REQ_ELVPRIV)
 577                elv_put_request(q, rq);
 578        mempool_free(rq, q->rq.rq_pool);
 579}
 580
 581static struct request *
 582blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
 583{
 584        struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
 585
 586        if (!rq)
 587                return NULL;
 588
 589        blk_rq_init(q, rq);
 590
 591        rq->cmd_flags = flags | REQ_ALLOCED;
 592
 593        if (priv) {
 594                if (unlikely(elv_set_request(q, rq, gfp_mask))) {
 595                        mempool_free(rq, q->rq.rq_pool);
 596                        return NULL;
 597                }
 598                rq->cmd_flags |= REQ_ELVPRIV;
 599        }
 600
 601        return rq;
 602}
 603
 604/*
 605 * ioc_batching returns true if the ioc is a valid batching request and
 606 * should be given priority access to a request.
 607 */
 608static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
 609{
 610        if (!ioc)
 611                return 0;
 612
 613        /*
 614         * Make sure the process is able to allocate at least 1 request
 615         * even if the batch times out, otherwise we could theoretically
 616         * lose wakeups.
 617         */
 618        return ioc->nr_batch_requests == q->nr_batching ||
 619                (ioc->nr_batch_requests > 0
 620                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
 621}
 622
 623/*
 624 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
 625 * will cause the process to be a "batcher" on all queues in the system. This
 626 * is the behaviour we want though - once it gets a wakeup it should be given
 627 * a nice run.
 628 */
 629static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
 630{
 631        if (!ioc || ioc_batching(q, ioc))
 632                return;
 633
 634        ioc->nr_batch_requests = q->nr_batching;
 635        ioc->last_waited = jiffies;
 636}
 637
 638static void __freed_request(struct request_queue *q, int sync)
 639{
 640        struct request_list *rl = &q->rq;
 641
 642        if (rl->count[sync] < queue_congestion_off_threshold(q))
 643                blk_clear_queue_congested(q, sync);
 644
 645        if (rl->count[sync] + 1 <= q->nr_requests) {
 646                if (waitqueue_active(&rl->wait[sync]))
 647                        wake_up(&rl->wait[sync]);
 648
 649                blk_clear_queue_full(q, sync);
 650        }
 651}
 652
 653/*
 654 * A request has just been released.  Account for it, update the full and
 655 * congestion status, wake up any waiters.   Called under q->queue_lock.
 656 */
 657static void freed_request(struct request_queue *q, int sync, int priv)
 658{
 659        struct request_list *rl = &q->rq;
 660
 661        rl->count[sync]--;
 662        if (priv)
 663                rl->elvpriv--;
 664
 665        __freed_request(q, sync);
 666
 667        if (unlikely(rl->starved[sync ^ 1]))
 668                __freed_request(q, sync ^ 1);
 669}
 670
 671/*
 672 * Determine if elevator data should be initialized when allocating the
 673 * request associated with @bio.
 674 */
 675static bool blk_rq_should_init_elevator(struct bio *bio)
 676{
 677        if (!bio)
 678                return true;
 679
 680        /*
 681         * Flush requests do not use the elevator so skip initialization.
 682         * This allows a request to share the flush and elevator data.
 683         */
 684        if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
 685                return false;
 686
 687        return true;
 688}
 689
 690/*
 691 * Get a free request, queue_lock must be held.
 692 * Returns NULL on failure, with queue_lock held.
 693 * Returns !NULL on success, with queue_lock *not held*.
 694 */
 695static struct request *get_request(struct request_queue *q, int rw_flags,
 696                                   struct bio *bio, gfp_t gfp_mask)
 697{
 698        struct request *rq = NULL;
 699        struct request_list *rl = &q->rq;
 700        struct io_context *ioc = NULL;
 701        const bool is_sync = rw_is_sync(rw_flags) != 0;
 702        int may_queue, priv = 0;
 703
 704        may_queue = elv_may_queue(q, rw_flags);
 705        if (may_queue == ELV_MQUEUE_NO)
 706                goto rq_starved;
 707
 708        if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
 709                if (rl->count[is_sync]+1 >= q->nr_requests) {
 710                        ioc = current_io_context(GFP_ATOMIC, q->node);
 711                        /*
 712                         * The queue will fill after this allocation, so set
 713                         * it as full, and mark this process as "batching".
 714                         * This process will be allowed to complete a batch of
 715                         * requests, others will be blocked.
 716                         */
 717                        if (!blk_queue_full(q, is_sync)) {
 718                                ioc_set_batching(q, ioc);
 719                                blk_set_queue_full(q, is_sync);
 720                        } else {
 721                                if (may_queue != ELV_MQUEUE_MUST
 722                                                && !ioc_batching(q, ioc)) {
 723                                        /*
 724                                         * The queue is full and the allocating
 725                                         * process is not a "batcher", and not
 726                                         * exempted by the IO scheduler
 727                                         */
 728                                        goto out;
 729                                }
 730                        }
 731                }
 732                blk_set_queue_congested(q, is_sync);
 733        }
 734
 735        /*
 736         * Only allow batching queuers to allocate up to 50% over the defined
 737         * limit of requests, otherwise we could have thousands of requests
 738         * allocated with any setting of ->nr_requests
 739         */
 740        if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
 741                goto out;
 742
 743        rl->count[is_sync]++;
 744        rl->starved[is_sync] = 0;
 745
 746        if (blk_rq_should_init_elevator(bio)) {
 747                priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
 748                if (priv)
 749                        rl->elvpriv++;
 750        }
 751
 752        if (blk_queue_io_stat(q))
 753                rw_flags |= REQ_IO_STAT;
 754        spin_unlock_irq(q->queue_lock);
 755
 756        rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
 757        if (unlikely(!rq)) {
 758                /*
 759                 * Allocation failed presumably due to memory. Undo anything
 760                 * we might have messed up.
 761                 *
 762                 * Allocating task should really be put onto the front of the
 763                 * wait queue, but this is pretty rare.
 764                 */
 765                spin_lock_irq(q->queue_lock);
 766                freed_request(q, is_sync, priv);
 767
 768                /*
 769                 * in the very unlikely event that allocation failed and no
 770                 * requests for this direction was pending, mark us starved
 771                 * so that freeing of a request in the other direction will
 772                 * notice us. another possible fix would be to split the
 773                 * rq mempool into READ and WRITE
 774                 */
 775rq_starved:
 776                if (unlikely(rl->count[is_sync] == 0))
 777                        rl->starved[is_sync] = 1;
 778
 779                goto out;
 780        }
 781
 782        /*
 783         * ioc may be NULL here, and ioc_batching will be false. That's
 784         * OK, if the queue is under the request limit then requests need
 785         * not count toward the nr_batch_requests limit. There will always
 786         * be some limit enforced by BLK_BATCH_TIME.
 787         */
 788        if (ioc_batching(q, ioc))
 789                ioc->nr_batch_requests--;
 790
 791        trace_block_getrq(q, bio, rw_flags & 1);
 792out:
 793        return rq;
 794}
 795
 796/*
 797 * No available requests for this queue, wait for some requests to become
 798 * available.
 799 *
 800 * Called with q->queue_lock held, and returns with it unlocked.
 801 */
 802static struct request *get_request_wait(struct request_queue *q, int rw_flags,
 803                                        struct bio *bio)
 804{
 805        const bool is_sync = rw_is_sync(rw_flags) != 0;
 806        struct request *rq;
 807
 808        rq = get_request(q, rw_flags, bio, GFP_NOIO);
 809        while (!rq) {
 810                DEFINE_WAIT(wait);
 811                struct io_context *ioc;
 812                struct request_list *rl = &q->rq;
 813
 814                prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
 815                                TASK_UNINTERRUPTIBLE);
 816
 817                trace_block_sleeprq(q, bio, rw_flags & 1);
 818
 819                spin_unlock_irq(q->queue_lock);
 820                io_schedule();
 821
 822                /*
 823                 * After sleeping, we become a "batching" process and
 824                 * will be able to allocate at least one request, and
 825                 * up to a big batch of them for a small period time.
 826                 * See ioc_batching, ioc_set_batching
 827                 */
 828                ioc = current_io_context(GFP_NOIO, q->node);
 829                ioc_set_batching(q, ioc);
 830
 831                spin_lock_irq(q->queue_lock);
 832                finish_wait(&rl->wait[is_sync], &wait);
 833
 834                rq = get_request(q, rw_flags, bio, GFP_NOIO);
 835        };
 836
 837        return rq;
 838}
 839
 840struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
 841{
 842        struct request *rq;
 843
 844        BUG_ON(rw != READ && rw != WRITE);
 845
 846        spin_lock_irq(q->queue_lock);
 847        if (gfp_mask & __GFP_WAIT) {
 848                rq = get_request_wait(q, rw, NULL);
 849        } else {
 850                rq = get_request(q, rw, NULL, gfp_mask);
 851                if (!rq)
 852                        spin_unlock_irq(q->queue_lock);
 853        }
 854        /* q->queue_lock is unlocked at this point */
 855
 856        return rq;
 857}
 858EXPORT_SYMBOL(blk_get_request);
 859
 860/**
 861 * blk_make_request - given a bio, allocate a corresponding struct request.
 862 * @q: target request queue
 863 * @bio:  The bio describing the memory mappings that will be submitted for IO.
 864 *        It may be a chained-bio properly constructed by block/bio layer.
 865 * @gfp_mask: gfp flags to be used for memory allocation
 866 *
 867 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
 868 * type commands. Where the struct request needs to be farther initialized by
 869 * the caller. It is passed a &struct bio, which describes the memory info of
 870 * the I/O transfer.
 871 *
 872 * The caller of blk_make_request must make sure that bi_io_vec
 873 * are set to describe the memory buffers. That bio_data_dir() will return
 874 * the needed direction of the request. (And all bio's in the passed bio-chain
 875 * are properly set accordingly)
 876 *
 877 * If called under none-sleepable conditions, mapped bio buffers must not
 878 * need bouncing, by calling the appropriate masked or flagged allocator,
 879 * suitable for the target device. Otherwise the call to blk_queue_bounce will
 880 * BUG.
 881 *
 882 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
 883 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
 884 * anything but the first bio in the chain. Otherwise you risk waiting for IO
 885 * completion of a bio that hasn't been submitted yet, thus resulting in a
 886 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
 887 * of bio_alloc(), as that avoids the mempool deadlock.
 888 * If possible a big IO should be split into smaller parts when allocation
 889 * fails. Partial allocation should not be an error, or you risk a live-lock.
 890 */
 891struct request *blk_make_request(struct request_queue *q, struct bio *bio,
 892                                 gfp_t gfp_mask)
 893{
 894        struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
 895
 896        if (unlikely(!rq))
 897                return ERR_PTR(-ENOMEM);
 898
 899        for_each_bio(bio) {
 900                struct bio *bounce_bio = bio;
 901                int ret;
 902
 903                blk_queue_bounce(q, &bounce_bio);
 904                ret = blk_rq_append_bio(q, rq, bounce_bio);
 905                if (unlikely(ret)) {
 906                        blk_put_request(rq);
 907                        return ERR_PTR(ret);
 908                }
 909        }
 910
 911        return rq;
 912}
 913EXPORT_SYMBOL(blk_make_request);
 914
 915/**
 916 * blk_requeue_request - put a request back on queue
 917 * @q:          request queue where request should be inserted
 918 * @rq:         request to be inserted
 919 *
 920 * Description:
 921 *    Drivers often keep queueing requests until the hardware cannot accept
 922 *    more, when that condition happens we need to put the request back
 923 *    on the queue. Must be called with queue lock held.
 924 */
 925void blk_requeue_request(struct request_queue *q, struct request *rq)
 926{
 927        blk_delete_timer(rq);
 928        blk_clear_rq_complete(rq);
 929        trace_block_rq_requeue(q, rq);
 930
 931        if (blk_rq_tagged(rq))
 932                blk_queue_end_tag(q, rq);
 933
 934        BUG_ON(blk_queued_rq(rq));
 935
 936        elv_requeue_request(q, rq);
 937}
 938EXPORT_SYMBOL(blk_requeue_request);
 939
 940static void add_acct_request(struct request_queue *q, struct request *rq,
 941                             int where)
 942{
 943        drive_stat_acct(rq, 1);
 944        __elv_add_request(q, rq, where);
 945}
 946
 947/**
 948 * blk_insert_request - insert a special request into a request queue
 949 * @q:          request queue where request should be inserted
 950 * @rq:         request to be inserted
 951 * @at_head:    insert request at head or tail of queue
 952 * @data:       private data
 953 *
 954 * Description:
 955 *    Many block devices need to execute commands asynchronously, so they don't
 956 *    block the whole kernel from preemption during request execution.  This is
 957 *    accomplished normally by inserting aritficial requests tagged as
 958 *    REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
 959 *    be scheduled for actual execution by the request queue.
 960 *
 961 *    We have the option of inserting the head or the tail of the queue.
 962 *    Typically we use the tail for new ioctls and so forth.  We use the head
 963 *    of the queue for things like a QUEUE_FULL message from a device, or a
 964 *    host that is unable to accept a particular command.
 965 */
 966void blk_insert_request(struct request_queue *q, struct request *rq,
 967                        int at_head, void *data)
 968{
 969        int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
 970        unsigned long flags;
 971
 972        /*
 973         * tell I/O scheduler that this isn't a regular read/write (ie it
 974         * must not attempt merges on this) and that it acts as a soft
 975         * barrier
 976         */
 977        rq->cmd_type = REQ_TYPE_SPECIAL;
 978
 979        rq->special = data;
 980
 981        spin_lock_irqsave(q->queue_lock, flags);
 982
 983        /*
 984         * If command is tagged, release the tag
 985         */
 986        if (blk_rq_tagged(rq))
 987                blk_queue_end_tag(q, rq);
 988
 989        add_acct_request(q, rq, where);
 990        __blk_run_queue(q);
 991        spin_unlock_irqrestore(q->queue_lock, flags);
 992}
 993EXPORT_SYMBOL(blk_insert_request);
 994
 995static void part_round_stats_single(int cpu, struct hd_struct *part,
 996                                    unsigned long now)
 997{
 998        if (now == part->stamp)
 999                return;
1000
1001        if (part_in_flight(part)) {
1002                __part_stat_add(cpu, part, time_in_queue,
1003                                part_in_flight(part) * (now - part->stamp));
1004                __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1005        }
1006        part->stamp = now;
1007}
1008
1009/**
1010 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1011 * @cpu: cpu number for stats access
1012 * @part: target partition
1013 *
1014 * The average IO queue length and utilisation statistics are maintained
1015 * by observing the current state of the queue length and the amount of
1016 * time it has been in this state for.
1017 *
1018 * Normally, that accounting is done on IO completion, but that can result
1019 * in more than a second's worth of IO being accounted for within any one
1020 * second, leading to >100% utilisation.  To deal with that, we call this
1021 * function to do a round-off before returning the results when reading
1022 * /proc/diskstats.  This accounts immediately for all queue usage up to
1023 * the current jiffies and restarts the counters again.
1024 */
1025void part_round_stats(int cpu, struct hd_struct *part)
1026{
1027        unsigned long now = jiffies;
1028
1029        if (part->partno)
1030                part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1031        part_round_stats_single(cpu, part, now);
1032}
1033EXPORT_SYMBOL_GPL(part_round_stats);
1034
1035/*
1036 * queue lock must be held
1037 */
1038void __blk_put_request(struct request_queue *q, struct request *req)
1039{
1040        if (unlikely(!q))
1041                return;
1042        if (unlikely(--req->ref_count))
1043                return;
1044
1045        elv_completed_request(q, req);
1046
1047        /* this is a bio leak */
1048        WARN_ON(req->bio != NULL);
1049
1050        /*
1051         * Request may not have originated from ll_rw_blk. if not,
1052         * it didn't come out of our reserved rq pools
1053         */
1054        if (req->cmd_flags & REQ_ALLOCED) {
1055                int is_sync = rq_is_sync(req) != 0;
1056                int priv = req->cmd_flags & REQ_ELVPRIV;
1057
1058                BUG_ON(!list_empty(&req->queuelist));
1059                BUG_ON(!hlist_unhashed(&req->hash));
1060
1061                blk_free_request(q, req);
1062                freed_request(q, is_sync, priv);
1063        }
1064}
1065EXPORT_SYMBOL_GPL(__blk_put_request);
1066
1067void blk_put_request(struct request *req)
1068{
1069        unsigned long flags;
1070        struct request_queue *q = req->q;
1071
1072        spin_lock_irqsave(q->queue_lock, flags);
1073        __blk_put_request(q, req);
1074        spin_unlock_irqrestore(q->queue_lock, flags);
1075}
1076EXPORT_SYMBOL(blk_put_request);
1077
1078/**
1079 * blk_add_request_payload - add a payload to a request
1080 * @rq: request to update
1081 * @page: page backing the payload
1082 * @len: length of the payload.
1083 *
1084 * This allows to later add a payload to an already submitted request by
1085 * a block driver.  The driver needs to take care of freeing the payload
1086 * itself.
1087 *
1088 * Note that this is a quite horrible hack and nothing but handling of
1089 * discard requests should ever use it.
1090 */
1091void blk_add_request_payload(struct request *rq, struct page *page,
1092                unsigned int len)
1093{
1094        struct bio *bio = rq->bio;
1095
1096        bio->bi_io_vec->bv_page = page;
1097        bio->bi_io_vec->bv_offset = 0;
1098        bio->bi_io_vec->bv_len = len;
1099
1100        bio->bi_size = len;
1101        bio->bi_vcnt = 1;
1102        bio->bi_phys_segments = 1;
1103
1104        rq->__data_len = rq->resid_len = len;
1105        rq->nr_phys_segments = 1;
1106        rq->buffer = bio_data(bio);
1107}
1108EXPORT_SYMBOL_GPL(blk_add_request_payload);
1109
1110static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1111                                   struct bio *bio)
1112{
1113        const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1114
1115        /*
1116         * Debug stuff, kill later
1117         */
1118        if (!rq_mergeable(req)) {
1119                blk_dump_rq_flags(req, "back");
1120                return false;
1121        }
1122
1123        if (!ll_back_merge_fn(q, req, bio))
1124                return false;
1125
1126        trace_block_bio_backmerge(q, bio);
1127
1128        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1129                blk_rq_set_mixed_merge(req);
1130
1131        req->biotail->bi_next = bio;
1132        req->biotail = bio;
1133        req->__data_len += bio->bi_size;
1134        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1135
1136        drive_stat_acct(req, 0);
1137        return true;
1138}
1139
1140static bool bio_attempt_front_merge(struct request_queue *q,
1141                                    struct request *req, struct bio *bio)
1142{
1143        const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1144        sector_t sector;
1145
1146        /*
1147         * Debug stuff, kill later
1148         */
1149        if (!rq_mergeable(req)) {
1150                blk_dump_rq_flags(req, "front");
1151                return false;
1152        }
1153
1154        if (!ll_front_merge_fn(q, req, bio))
1155                return false;
1156
1157        trace_block_bio_frontmerge(q, bio);
1158
1159        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1160                blk_rq_set_mixed_merge(req);
1161
1162        sector = bio->bi_sector;
1163
1164        bio->bi_next = req->bio;
1165        req->bio = bio;
1166
1167        /*
1168         * may not be valid. if the low level driver said
1169         * it didn't need a bounce buffer then it better
1170         * not touch req->buffer either...
1171         */
1172        req->buffer = bio_data(bio);
1173        req->__sector = bio->bi_sector;
1174        req->__data_len += bio->bi_size;
1175        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1176
1177        drive_stat_acct(req, 0);
1178        return true;
1179}
1180
1181/*
1182 * Attempts to merge with the plugged list in the current process. Returns
1183 * true if merge was successful, otherwise false.
1184 */
1185static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1186                               struct bio *bio)
1187{
1188        struct blk_plug *plug;
1189        struct request *rq;
1190        bool ret = false;
1191
1192        plug = tsk->plug;
1193        if (!plug)
1194                goto out;
1195
1196        list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1197                int el_ret;
1198
1199                if (rq->q != q)
1200                        continue;
1201
1202                el_ret = elv_try_merge(rq, bio);
1203                if (el_ret == ELEVATOR_BACK_MERGE) {
1204                        ret = bio_attempt_back_merge(q, rq, bio);
1205                        if (ret)
1206                                break;
1207                } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1208                        ret = bio_attempt_front_merge(q, rq, bio);
1209                        if (ret)
1210                                break;
1211                }
1212        }
1213out:
1214        return ret;
1215}
1216
1217void init_request_from_bio(struct request *req, struct bio *bio)
1218{
1219        req->cpu = bio->bi_comp_cpu;
1220        req->cmd_type = REQ_TYPE_FS;
1221
1222        req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1223        if (bio->bi_rw & REQ_RAHEAD)
1224                req->cmd_flags |= REQ_FAILFAST_MASK;
1225
1226        req->errors = 0;
1227        req->__sector = bio->bi_sector;
1228        req->ioprio = bio_prio(bio);
1229        blk_rq_bio_prep(req->q, req, bio);
1230}
1231
1232static int __make_request(struct request_queue *q, struct bio *bio)
1233{
1234        const bool sync = !!(bio->bi_rw & REQ_SYNC);
1235        struct blk_plug *plug;
1236        int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1237        struct request *req;
1238
1239        /*
1240         * low level driver can indicate that it wants pages above a
1241         * certain limit bounced to low memory (ie for highmem, or even
1242         * ISA dma in theory)
1243         */
1244        blk_queue_bounce(q, &bio);
1245
1246        if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1247                spin_lock_irq(q->queue_lock);
1248                where = ELEVATOR_INSERT_FLUSH;
1249                goto get_rq;
1250        }
1251
1252        /*
1253         * Check if we can merge with the plugged list before grabbing
1254         * any locks.
1255         */
1256        if (attempt_plug_merge(current, q, bio))
1257                goto out;
1258
1259        spin_lock_irq(q->queue_lock);
1260
1261        el_ret = elv_merge(q, &req, bio);
1262        if (el_ret == ELEVATOR_BACK_MERGE) {
1263                BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1264                if (bio_attempt_back_merge(q, req, bio)) {
1265                        if (!attempt_back_merge(q, req))
1266                                elv_merged_request(q, req, el_ret);
1267                        goto out_unlock;
1268                }
1269        } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1270                BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1271                if (bio_attempt_front_merge(q, req, bio)) {
1272                        if (!attempt_front_merge(q, req))
1273                                elv_merged_request(q, req, el_ret);
1274                        goto out_unlock;
1275                }
1276        }
1277
1278get_rq:
1279        /*
1280         * This sync check and mask will be re-done in init_request_from_bio(),
1281         * but we need to set it earlier to expose the sync flag to the
1282         * rq allocator and io schedulers.
1283         */
1284        rw_flags = bio_data_dir(bio);
1285        if (sync)
1286                rw_flags |= REQ_SYNC;
1287
1288        /*
1289         * Grab a free request. This is might sleep but can not fail.
1290         * Returns with the queue unlocked.
1291         */
1292        req = get_request_wait(q, rw_flags, bio);
1293
1294        /*
1295         * After dropping the lock and possibly sleeping here, our request
1296         * may now be mergeable after it had proven unmergeable (above).
1297         * We don't worry about that case for efficiency. It won't happen
1298         * often, and the elevators are able to handle it.
1299         */
1300        init_request_from_bio(req, bio);
1301
1302        if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1303            bio_flagged(bio, BIO_CPU_AFFINE)) {
1304                req->cpu = blk_cpu_to_group(get_cpu());
1305                put_cpu();
1306        }
1307
1308        plug = current->plug;
1309        if (plug) {
1310                /*
1311                 * If this is the first request added after a plug, fire
1312                 * of a plug trace. If others have been added before, check
1313                 * if we have multiple devices in this plug. If so, make a
1314                 * note to sort the list before dispatch.
1315                 */
1316                if (list_empty(&plug->list))
1317                        trace_block_plug(q);
1318                else if (!plug->should_sort) {
1319                        struct request *__rq;
1320
1321                        __rq = list_entry_rq(plug->list.prev);
1322                        if (__rq->q != q)
1323                                plug->should_sort = 1;
1324                }
1325                /*
1326                 * Debug flag, kill later
1327                 */
1328                req->cmd_flags |= REQ_ON_PLUG;
1329                list_add_tail(&req->queuelist, &plug->list);
1330                drive_stat_acct(req, 1);
1331        } else {
1332                spin_lock_irq(q->queue_lock);
1333                add_acct_request(q, req, where);
1334                __blk_run_queue(q);
1335out_unlock:
1336                spin_unlock_irq(q->queue_lock);
1337        }
1338out:
1339        return 0;
1340}
1341
1342/*
1343 * If bio->bi_dev is a partition, remap the location
1344 */
1345static inline void blk_partition_remap(struct bio *bio)
1346{
1347        struct block_device *bdev = bio->bi_bdev;
1348
1349        if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1350                struct hd_struct *p = bdev->bd_part;
1351
1352                bio->bi_sector += p->start_sect;
1353                bio->bi_bdev = bdev->bd_contains;
1354
1355                trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1356                                      bdev->bd_dev,
1357                                      bio->bi_sector - p->start_sect);
1358        }
1359}
1360
1361static void handle_bad_sector(struct bio *bio)
1362{
1363        char b[BDEVNAME_SIZE];
1364
1365        printk(KERN_INFO "attempt to access beyond end of device\n");
1366        printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1367                        bdevname(bio->bi_bdev, b),
1368                        bio->bi_rw,
1369                        (unsigned long long)bio->bi_sector + bio_sectors(bio),
1370                        (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1371
1372        set_bit(BIO_EOF, &bio->bi_flags);
1373}
1374
1375#ifdef CONFIG_FAIL_MAKE_REQUEST
1376
1377static DECLARE_FAULT_ATTR(fail_make_request);
1378
1379static int __init setup_fail_make_request(char *str)
1380{
1381        return setup_fault_attr(&fail_make_request, str);
1382}
1383__setup("fail_make_request=", setup_fail_make_request);
1384
1385static int should_fail_request(struct bio *bio)
1386{
1387        struct hd_struct *part = bio->bi_bdev->bd_part;
1388
1389        if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1390                return should_fail(&fail_make_request, bio->bi_size);
1391
1392        return 0;
1393}
1394
1395static int __init fail_make_request_debugfs(void)
1396{
1397        return init_fault_attr_dentries(&fail_make_request,
1398                                        "fail_make_request");
1399}
1400
1401late_initcall(fail_make_request_debugfs);
1402
1403#else /* CONFIG_FAIL_MAKE_REQUEST */
1404
1405static inline int should_fail_request(struct bio *bio)
1406{
1407        return 0;
1408}
1409
1410#endif /* CONFIG_FAIL_MAKE_REQUEST */
1411
1412/*
1413 * Check whether this bio extends beyond the end of the device.
1414 */
1415static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1416{
1417        sector_t maxsector;
1418
1419        if (!nr_sectors)
1420                return 0;
1421
1422        /* Test device or partition size, when known. */
1423        maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1424        if (maxsector) {
1425                sector_t sector = bio->bi_sector;
1426
1427                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1428                        /*
1429                         * This may well happen - the kernel calls bread()
1430                         * without checking the size of the device, e.g., when
1431                         * mounting a device.
1432                         */
1433                        handle_bad_sector(bio);
1434                        return 1;
1435                }
1436        }
1437
1438        return 0;
1439}
1440
1441/**
1442 * generic_make_request - hand a buffer to its device driver for I/O
1443 * @bio:  The bio describing the location in memory and on the device.
1444 *
1445 * generic_make_request() is used to make I/O requests of block
1446 * devices. It is passed a &struct bio, which describes the I/O that needs
1447 * to be done.
1448 *
1449 * generic_make_request() does not return any status.  The
1450 * success/failure status of the request, along with notification of
1451 * completion, is delivered asynchronously through the bio->bi_end_io
1452 * function described (one day) else where.
1453 *
1454 * The caller of generic_make_request must make sure that bi_io_vec
1455 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1456 * set to describe the device address, and the
1457 * bi_end_io and optionally bi_private are set to describe how
1458 * completion notification should be signaled.
1459 *
1460 * generic_make_request and the drivers it calls may use bi_next if this
1461 * bio happens to be merged with someone else, and may change bi_dev and
1462 * bi_sector for remaps as it sees fit.  So the values of these fields
1463 * should NOT be depended on after the call to generic_make_request.
1464 */
1465static inline void __generic_make_request(struct bio *bio)
1466{
1467        struct request_queue *q;
1468        sector_t old_sector;
1469        int ret, nr_sectors = bio_sectors(bio);
1470        dev_t old_dev;
1471        int err = -EIO;
1472
1473        might_sleep();
1474
1475        if (bio_check_eod(bio, nr_sectors))
1476                goto end_io;
1477
1478        /*
1479         * Resolve the mapping until finished. (drivers are
1480         * still free to implement/resolve their own stacking
1481         * by explicitly returning 0)
1482         *
1483         * NOTE: we don't repeat the blk_size check for each new device.
1484         * Stacking drivers are expected to know what they are doing.
1485         */
1486        old_sector = -1;
1487        old_dev = 0;
1488        do {
1489                char b[BDEVNAME_SIZE];
1490
1491                q = bdev_get_queue(bio->bi_bdev);
1492                if (unlikely(!q)) {
1493                        printk(KERN_ERR
1494                               "generic_make_request: Trying to access "
1495                                "nonexistent block-device %s (%Lu)\n",
1496                                bdevname(bio->bi_bdev, b),
1497                                (long long) bio->bi_sector);
1498                        goto end_io;
1499                }
1500
1501                if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1502                             nr_sectors > queue_max_hw_sectors(q))) {
1503                        printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1504                               bdevname(bio->bi_bdev, b),
1505                               bio_sectors(bio),
1506                               queue_max_hw_sectors(q));
1507                        goto end_io;
1508                }
1509
1510                if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1511                        goto end_io;
1512
1513                if (should_fail_request(bio))
1514                        goto end_io;
1515
1516                /*
1517                 * If this device has partitions, remap block n
1518                 * of partition p to block n+start(p) of the disk.
1519                 */
1520                blk_partition_remap(bio);
1521
1522                if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1523                        goto end_io;
1524
1525                if (old_sector != -1)
1526                        trace_block_bio_remap(q, bio, old_dev, old_sector);
1527
1528                old_sector = bio->bi_sector;
1529                old_dev = bio->bi_bdev->bd_dev;
1530
1531                if (bio_check_eod(bio, nr_sectors))
1532                        goto end_io;
1533
1534                /*
1535                 * Filter flush bio's early so that make_request based
1536                 * drivers without flush support don't have to worry
1537                 * about them.
1538                 */
1539                if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1540                        bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1541                        if (!nr_sectors) {
1542                                err = 0;
1543                                goto end_io;
1544                        }
1545                }
1546
1547                if ((bio->bi_rw & REQ_DISCARD) &&
1548                    (!blk_queue_discard(q) ||
1549                     ((bio->bi_rw & REQ_SECURE) &&
1550                      !blk_queue_secdiscard(q)))) {
1551                        err = -EOPNOTSUPP;
1552                        goto end_io;
1553                }
1554
1555                blk_throtl_bio(q, &bio);
1556
1557                /*
1558                 * If bio = NULL, bio has been throttled and will be submitted
1559                 * later.
1560                 */
1561                if (!bio)
1562                        break;
1563
1564                trace_block_bio_queue(q, bio);
1565
1566                ret = q->make_request_fn(q, bio);
1567        } while (ret);
1568
1569        return;
1570
1571end_io:
1572        bio_endio(bio, err);
1573}
1574
1575/*
1576 * We only want one ->make_request_fn to be active at a time,
1577 * else stack usage with stacked devices could be a problem.
1578 * So use current->bio_list to keep a list of requests
1579 * submited by a make_request_fn function.
1580 * current->bio_list is also used as a flag to say if
1581 * generic_make_request is currently active in this task or not.
1582 * If it is NULL, then no make_request is active.  If it is non-NULL,
1583 * then a make_request is active, and new requests should be added
1584 * at the tail
1585 */
1586void generic_make_request(struct bio *bio)
1587{
1588        struct bio_list bio_list_on_stack;
1589
1590        if (current->bio_list) {
1591                /* make_request is active */
1592                bio_list_add(current->bio_list, bio);
1593                return;
1594        }
1595        /* following loop may be a bit non-obvious, and so deserves some
1596         * explanation.
1597         * Before entering the loop, bio->bi_next is NULL (as all callers
1598         * ensure that) so we have a list with a single bio.
1599         * We pretend that we have just taken it off a longer list, so
1600         * we assign bio_list to a pointer to the bio_list_on_stack,
1601         * thus initialising the bio_list of new bios to be
1602         * added.  __generic_make_request may indeed add some more bios
1603         * through a recursive call to generic_make_request.  If it
1604         * did, we find a non-NULL value in bio_list and re-enter the loop
1605         * from the top.  In this case we really did just take the bio
1606         * of the top of the list (no pretending) and so remove it from
1607         * bio_list, and call into __generic_make_request again.
1608         *
1609         * The loop was structured like this to make only one call to
1610         * __generic_make_request (which is important as it is large and
1611         * inlined) and to keep the structure simple.
1612         */
1613        BUG_ON(bio->bi_next);
1614        bio_list_init(&bio_list_on_stack);
1615        current->bio_list = &bio_list_on_stack;
1616        do {
1617                __generic_make_request(bio);
1618                bio = bio_list_pop(current->bio_list);
1619        } while (bio);
1620        current->bio_list = NULL; /* deactivate */
1621}
1622EXPORT_SYMBOL(generic_make_request);
1623
1624/**
1625 * submit_bio - submit a bio to the block device layer for I/O
1626 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1627 * @bio: The &struct bio which describes the I/O
1628 *
1629 * submit_bio() is very similar in purpose to generic_make_request(), and
1630 * uses that function to do most of the work. Both are fairly rough
1631 * interfaces; @bio must be presetup and ready for I/O.
1632 *
1633 */
1634void submit_bio(int rw, struct bio *bio)
1635{
1636        int count = bio_sectors(bio);
1637
1638        bio->bi_rw |= rw;
1639
1640        /*
1641         * If it's a regular read/write or a barrier with data attached,
1642         * go through the normal accounting stuff before submission.
1643         */
1644        if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1645                if (rw & WRITE) {
1646                        count_vm_events(PGPGOUT, count);
1647                } else {
1648                        task_io_account_read(bio->bi_size);
1649                        count_vm_events(PGPGIN, count);
1650                }
1651
1652                if (unlikely(block_dump)) {
1653                        char b[BDEVNAME_SIZE];
1654                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1655                        current->comm, task_pid_nr(current),
1656                                (rw & WRITE) ? "WRITE" : "READ",
1657                                (unsigned long long)bio->bi_sector,
1658                                bdevname(bio->bi_bdev, b),
1659                                count);
1660                }
1661        }
1662
1663        generic_make_request(bio);
1664}
1665EXPORT_SYMBOL(submit_bio);
1666
1667/**
1668 * blk_rq_check_limits - Helper function to check a request for the queue limit
1669 * @q:  the queue
1670 * @rq: the request being checked
1671 *
1672 * Description:
1673 *    @rq may have been made based on weaker limitations of upper-level queues
1674 *    in request stacking drivers, and it may violate the limitation of @q.
1675 *    Since the block layer and the underlying device driver trust @rq
1676 *    after it is inserted to @q, it should be checked against @q before
1677 *    the insertion using this generic function.
1678 *
1679 *    This function should also be useful for request stacking drivers
1680 *    in some cases below, so export this function.
1681 *    Request stacking drivers like request-based dm may change the queue
1682 *    limits while requests are in the queue (e.g. dm's table swapping).
1683 *    Such request stacking drivers should check those requests agaist
1684 *    the new queue limits again when they dispatch those requests,
1685 *    although such checkings are also done against the old queue limits
1686 *    when submitting requests.
1687 */
1688int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1689{
1690        if (rq->cmd_flags & REQ_DISCARD)
1691                return 0;
1692
1693        if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1694            blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1695                printk(KERN_ERR "%s: over max size limit.\n", __func__);
1696                return -EIO;
1697        }
1698
1699        /*
1700         * queue's settings related to segment counting like q->bounce_pfn
1701         * may differ from that of other stacking queues.
1702         * Recalculate it to check the request correctly on this queue's
1703         * limitation.
1704         */
1705        blk_recalc_rq_segments(rq);
1706        if (rq->nr_phys_segments > queue_max_segments(q)) {
1707                printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1708                return -EIO;
1709        }
1710
1711        return 0;
1712}
1713EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1714
1715/**
1716 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1717 * @q:  the queue to submit the request
1718 * @rq: the request being queued
1719 */
1720int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1721{
1722        unsigned long flags;
1723
1724        if (blk_rq_check_limits(q, rq))
1725                return -EIO;
1726
1727#ifdef CONFIG_FAIL_MAKE_REQUEST
1728        if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1729            should_fail(&fail_make_request, blk_rq_bytes(rq)))
1730                return -EIO;
1731#endif
1732
1733        spin_lock_irqsave(q->queue_lock, flags);
1734
1735        /*
1736         * Submitting request must be dequeued before calling this function
1737         * because it will be linked to another request_queue
1738         */
1739        BUG_ON(blk_queued_rq(rq));
1740
1741        add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1742        spin_unlock_irqrestore(q->queue_lock, flags);
1743
1744        return 0;
1745}
1746EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1747
1748/**
1749 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1750 * @rq: request to examine
1751 *
1752 * Description:
1753 *     A request could be merge of IOs which require different failure
1754 *     handling.  This function determines the number of bytes which
1755 *     can be failed from the beginning of the request without
1756 *     crossing into area which need to be retried further.
1757 *
1758 * Return:
1759 *     The number of bytes to fail.
1760 *
1761 * Context:
1762 *     queue_lock must be held.
1763 */
1764unsigned int blk_rq_err_bytes(const struct request *rq)
1765{
1766        unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1767        unsigned int bytes = 0;
1768        struct bio *bio;
1769
1770        if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1771                return blk_rq_bytes(rq);
1772
1773        /*
1774         * Currently the only 'mixing' which can happen is between
1775         * different fastfail types.  We can safely fail portions
1776         * which have all the failfast bits that the first one has -
1777         * the ones which are at least as eager to fail as the first
1778         * one.
1779         */
1780        for (bio = rq->bio; bio; bio = bio->bi_next) {
1781                if ((bio->bi_rw & ff) != ff)
1782                        break;
1783                bytes += bio->bi_size;
1784        }
1785
1786        /* this could lead to infinite loop */
1787        BUG_ON(blk_rq_bytes(rq) && !bytes);
1788        return bytes;
1789}
1790EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1791
1792static void blk_account_io_completion(struct request *req, unsigned int bytes)
1793{
1794        if (blk_do_io_stat(req)) {
1795                const int rw = rq_data_dir(req);
1796                struct hd_struct *part;
1797                int cpu;
1798
1799                cpu = part_stat_lock();
1800                part = req->part;
1801                part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1802                part_stat_unlock();
1803        }
1804}
1805
1806static void blk_account_io_done(struct request *req)
1807{
1808        /*
1809         * Account IO completion.  flush_rq isn't accounted as a
1810         * normal IO on queueing nor completion.  Accounting the
1811         * containing request is enough.
1812         */
1813        if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1814                unsigned long duration = jiffies - req->start_time;
1815                const int rw = rq_data_dir(req);
1816                struct hd_struct *part;
1817                int cpu;
1818
1819                cpu = part_stat_lock();
1820                part = req->part;
1821
1822                part_stat_inc(cpu, part, ios[rw]);
1823                part_stat_add(cpu, part, ticks[rw], duration);
1824                part_round_stats(cpu, part);
1825                part_dec_in_flight(part, rw);
1826
1827                hd_struct_put(part);
1828                part_stat_unlock();
1829        }
1830}
1831
1832/**
1833 * blk_peek_request - peek at the top of a request queue
1834 * @q: request queue to peek at
1835 *
1836 * Description:
1837 *     Return the request at the top of @q.  The returned request
1838 *     should be started using blk_start_request() before LLD starts
1839 *     processing it.
1840 *
1841 * Return:
1842 *     Pointer to the request at the top of @q if available.  Null
1843 *     otherwise.
1844 *
1845 * Context:
1846 *     queue_lock must be held.
1847 */
1848struct request *blk_peek_request(struct request_queue *q)
1849{
1850        struct request *rq;
1851        int ret;
1852
1853        while ((rq = __elv_next_request(q)) != NULL) {
1854                if (!(rq->cmd_flags & REQ_STARTED)) {
1855                        /*
1856                         * This is the first time the device driver
1857                         * sees this request (possibly after
1858                         * requeueing).  Notify IO scheduler.
1859                         */
1860                        if (rq->cmd_flags & REQ_SORTED)
1861                                elv_activate_rq(q, rq);
1862
1863                        /*
1864                         * just mark as started even if we don't start
1865                         * it, a request that has been delayed should
1866                         * not be passed by new incoming requests
1867                         */
1868                        rq->cmd_flags |= REQ_STARTED;
1869                        trace_block_rq_issue(q, rq);
1870                }
1871
1872                if (!q->boundary_rq || q->boundary_rq == rq) {
1873                        q->end_sector = rq_end_sector(rq);
1874                        q->boundary_rq = NULL;
1875                }
1876
1877                if (rq->cmd_flags & REQ_DONTPREP)
1878                        break;
1879
1880                if (q->dma_drain_size && blk_rq_bytes(rq)) {
1881                        /*
1882                         * make sure space for the drain appears we
1883                         * know we can do this because max_hw_segments
1884                         * has been adjusted to be one fewer than the
1885                         * device can handle
1886                         */
1887                        rq->nr_phys_segments++;
1888                }
1889
1890                if (!q->prep_rq_fn)
1891                        break;
1892
1893                ret = q->prep_rq_fn(q, rq);
1894                if (ret == BLKPREP_OK) {
1895                        break;
1896                } else if (ret == BLKPREP_DEFER) {
1897                        /*
1898                         * the request may have been (partially) prepped.
1899                         * we need to keep this request in the front to
1900                         * avoid resource deadlock.  REQ_STARTED will
1901                         * prevent other fs requests from passing this one.
1902                         */
1903                        if (q->dma_drain_size && blk_rq_bytes(rq) &&
1904                            !(rq->cmd_flags & REQ_DONTPREP)) {
1905                                /*
1906                                 * remove the space for the drain we added
1907                                 * so that we don't add it again
1908                                 */
1909                                --rq->nr_phys_segments;
1910                        }
1911
1912                        rq = NULL;
1913                        break;
1914                } else if (ret == BLKPREP_KILL) {
1915                        rq->cmd_flags |= REQ_QUIET;
1916                        /*
1917                         * Mark this request as started so we don't trigger
1918                         * any debug logic in the end I/O path.
1919                         */
1920                        blk_start_request(rq);
1921                        __blk_end_request_all(rq, -EIO);
1922                } else {
1923                        printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1924                        break;
1925                }
1926        }
1927
1928        return rq;
1929}
1930EXPORT_SYMBOL(blk_peek_request);
1931
1932void blk_dequeue_request(struct request *rq)
1933{
1934        struct request_queue *q = rq->q;
1935
1936        BUG_ON(list_empty(&rq->queuelist));
1937        BUG_ON(ELV_ON_HASH(rq));
1938
1939        list_del_init(&rq->queuelist);
1940
1941        /*
1942         * the time frame between a request being removed from the lists
1943         * and to it is freed is accounted as io that is in progress at
1944         * the driver side.
1945         */
1946        if (blk_account_rq(rq)) {
1947                q->in_flight[rq_is_sync(rq)]++;
1948                set_io_start_time_ns(rq);
1949        }
1950}
1951
1952/**
1953 * blk_start_request - start request processing on the driver
1954 * @req: request to dequeue
1955 *
1956 * Description:
1957 *     Dequeue @req and start timeout timer on it.  This hands off the
1958 *     request to the driver.
1959 *
1960 *     Block internal functions which don't want to start timer should
1961 *     call blk_dequeue_request().
1962 *
1963 * Context:
1964 *     queue_lock must be held.
1965 */
1966void blk_start_request(struct request *req)
1967{
1968        blk_dequeue_request(req);
1969
1970        /*
1971         * We are now handing the request to the hardware, initialize
1972         * resid_len to full count and add the timeout handler.
1973         */
1974        req->resid_len = blk_rq_bytes(req);
1975        if (unlikely(blk_bidi_rq(req)))
1976                req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1977
1978        blk_add_timer(req);
1979}
1980EXPORT_SYMBOL(blk_start_request);
1981
1982/**
1983 * blk_fetch_request - fetch a request from a request queue
1984 * @q: request queue to fetch a request from
1985 *
1986 * Description:
1987 *     Return the request at the top of @q.  The request is started on
1988 *     return and LLD can start processing it immediately.
1989 *
1990 * Return:
1991 *     Pointer to the request at the top of @q if available.  Null
1992 *     otherwise.
1993 *
1994 * Context:
1995 *     queue_lock must be held.
1996 */
1997struct request *blk_fetch_request(struct request_queue *q)
1998{
1999        struct request *rq;
2000
2001        rq = blk_peek_request(q);
2002        if (rq)
2003                blk_start_request(rq);
2004        return rq;
2005}
2006EXPORT_SYMBOL(blk_fetch_request);
2007
2008/**
2009 * blk_update_request - Special helper function for request stacking drivers
2010 * @req:      the request being processed
2011 * @error:    %0 for success, < %0 for error
2012 * @nr_bytes: number of bytes to complete @req
2013 *
2014 * Description:
2015 *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2016 *     the request structure even if @req doesn't have leftover.
2017 *     If @req has leftover, sets it up for the next range of segments.
2018 *
2019 *     This special helper function is only for request stacking drivers
2020 *     (e.g. request-based dm) so that they can handle partial completion.
2021 *     Actual device drivers should use blk_end_request instead.
2022 *
2023 *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2024 *     %false return from this function.
2025 *
2026 * Return:
2027 *     %false - this request doesn't have any more data
2028 *     %true  - this request has more data
2029 **/
2030bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2031{
2032        int total_bytes, bio_nbytes, next_idx = 0;
2033        struct bio *bio;
2034
2035        if (!req->bio)
2036                return false;
2037
2038        trace_block_rq_complete(req->q, req);
2039
2040        /*
2041         * For fs requests, rq is just carrier of independent bio's
2042         * and each partial completion should be handled separately.
2043         * Reset per-request error on each partial completion.
2044         *
2045         * TODO: tj: This is too subtle.  It would be better to let
2046         * low level drivers do what they see fit.
2047         */
2048        if (req->cmd_type == REQ_TYPE_FS)
2049                req->errors = 0;
2050
2051        if (error && req->cmd_type == REQ_TYPE_FS &&
2052            !(req->cmd_flags & REQ_QUIET)) {
2053                char *error_type;
2054
2055                switch (error) {
2056                case -ENOLINK:
2057                        error_type = "recoverable transport";
2058                        break;
2059                case -EREMOTEIO:
2060                        error_type = "critical target";
2061                        break;
2062                case -EBADE:
2063                        error_type = "critical nexus";
2064                        break;
2065                case -EIO:
2066                default:
2067                        error_type = "I/O";
2068                        break;
2069                }
2070                printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2071                       error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2072                       (unsigned long long)blk_rq_pos(req));
2073        }
2074
2075        blk_account_io_completion(req, nr_bytes);
2076
2077        total_bytes = bio_nbytes = 0;
2078        while ((bio = req->bio) != NULL) {
2079                int nbytes;
2080
2081                if (nr_bytes >= bio->bi_size) {
2082                        req->bio = bio->bi_next;
2083                        nbytes = bio->bi_size;
2084                        req_bio_endio(req, bio, nbytes, error);
2085                        next_idx = 0;
2086                        bio_nbytes = 0;
2087                } else {
2088                        int idx = bio->bi_idx + next_idx;
2089
2090                        if (unlikely(idx >= bio->bi_vcnt)) {
2091                                blk_dump_rq_flags(req, "__end_that");
2092                                printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2093                                       __func__, idx, bio->bi_vcnt);
2094                                break;
2095                        }
2096
2097                        nbytes = bio_iovec_idx(bio, idx)->bv_len;
2098                        BIO_BUG_ON(nbytes > bio->bi_size);
2099
2100                        /*
2101                         * not a complete bvec done
2102                         */
2103                        if (unlikely(nbytes > nr_bytes)) {
2104                                bio_nbytes += nr_bytes;
2105                                total_bytes += nr_bytes;
2106                                break;
2107                        }
2108
2109                        /*
2110                         * advance to the next vector
2111                         */
2112                        next_idx++;
2113                        bio_nbytes += nbytes;
2114                }
2115
2116                total_bytes += nbytes;
2117                nr_bytes -= nbytes;
2118
2119                bio = req->bio;
2120                if (bio) {
2121                        /*
2122                         * end more in this run, or just return 'not-done'
2123                         */
2124                        if (unlikely(nr_bytes <= 0))
2125                                break;
2126                }
2127        }
2128
2129        /*
2130         * completely done
2131         */
2132        if (!req->bio) {
2133                /*
2134                 * Reset counters so that the request stacking driver
2135                 * can find how many bytes remain in the request
2136                 * later.
2137                 */
2138                req->__data_len = 0;
2139                return false;
2140        }
2141
2142        /*
2143         * if the request wasn't completed, update state
2144         */
2145        if (bio_nbytes) {
2146                req_bio_endio(req, bio, bio_nbytes, error);
2147                bio->bi_idx += next_idx;
2148                bio_iovec(bio)->bv_offset += nr_bytes;
2149                bio_iovec(bio)->bv_len -= nr_bytes;
2150        }
2151
2152        req->__data_len -= total_bytes;
2153        req->buffer = bio_data(req->bio);
2154
2155        /* update sector only for requests with clear definition of sector */
2156        if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2157                req->__sector += total_bytes >> 9;
2158
2159        /* mixed attributes always follow the first bio */
2160        if (req->cmd_flags & REQ_MIXED_MERGE) {
2161                req->cmd_flags &= ~REQ_FAILFAST_MASK;
2162                req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2163        }
2164
2165        /*
2166         * If total number of sectors is less than the first segment
2167         * size, something has gone terribly wrong.
2168         */
2169        if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2170                blk_dump_rq_flags(req, "request botched");
2171                req->__data_len = blk_rq_cur_bytes(req);
2172        }
2173
2174        /* recalculate the number of segments */
2175        blk_recalc_rq_segments(req);
2176
2177        return true;
2178}
2179EXPORT_SYMBOL_GPL(blk_update_request);
2180
2181static bool blk_update_bidi_request(struct request *rq, int error,
2182                                    unsigned int nr_bytes,
2183                                    unsigned int bidi_bytes)
2184{
2185        if (blk_update_request(rq, error, nr_bytes))
2186                return true;
2187
2188        /* Bidi request must be completed as a whole */
2189        if (unlikely(blk_bidi_rq(rq)) &&
2190            blk_update_request(rq->next_rq, error, bidi_bytes))
2191                return true;
2192
2193        if (blk_queue_add_random(rq->q))
2194                add_disk_randomness(rq->rq_disk);
2195
2196        return false;
2197}
2198
2199/**
2200 * blk_unprep_request - unprepare a request
2201 * @req:        the request
2202 *
2203 * This function makes a request ready for complete resubmission (or
2204 * completion).  It happens only after all error handling is complete,
2205 * so represents the appropriate moment to deallocate any resources
2206 * that were allocated to the request in the prep_rq_fn.  The queue
2207 * lock is held when calling this.
2208 */
2209void blk_unprep_request(struct request *req)
2210{
2211        struct request_queue *q = req->q;
2212
2213        req->cmd_flags &= ~REQ_DONTPREP;
2214        if (q->unprep_rq_fn)
2215                q->unprep_rq_fn(q, req);
2216}
2217EXPORT_SYMBOL_GPL(blk_unprep_request);
2218
2219/*
2220 * queue lock must be held
2221 */
2222static void blk_finish_request(struct request *req, int error)
2223{
2224        if (blk_rq_tagged(req))
2225                blk_queue_end_tag(req->q, req);
2226
2227        BUG_ON(blk_queued_rq(req));
2228
2229        if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2230                laptop_io_completion(&req->q->backing_dev_info);
2231
2232        blk_delete_timer(req);
2233
2234        if (req->cmd_flags & REQ_DONTPREP)
2235                blk_unprep_request(req);
2236
2237
2238        blk_account_io_done(req);
2239
2240        if (req->end_io)
2241                req->end_io(req, error);
2242        else {
2243                if (blk_bidi_rq(req))
2244                        __blk_put_request(req->next_rq->q, req->next_rq);
2245
2246                __blk_put_request(req->q, req);
2247        }
2248}
2249
2250/**
2251 * blk_end_bidi_request - Complete a bidi request
2252 * @rq:         the request to complete
2253 * @error:      %0 for success, < %0 for error
2254 * @nr_bytes:   number of bytes to complete @rq
2255 * @bidi_bytes: number of bytes to complete @rq->next_rq
2256 *
2257 * Description:
2258 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2259 *     Drivers that supports bidi can safely call this member for any
2260 *     type of request, bidi or uni.  In the later case @bidi_bytes is
2261 *     just ignored.
2262 *
2263 * Return:
2264 *     %false - we are done with this request
2265 *     %true  - still buffers pending for this request
2266 **/
2267static bool blk_end_bidi_request(struct request *rq, int error,
2268                                 unsigned int nr_bytes, unsigned int bidi_bytes)
2269{
2270        struct request_queue *q = rq->q;
2271        unsigned long flags;
2272
2273        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2274                return true;
2275
2276        spin_lock_irqsave(q->queue_lock, flags);
2277        blk_finish_request(rq, error);
2278        spin_unlock_irqrestore(q->queue_lock, flags);
2279
2280        return false;
2281}
2282
2283/**
2284 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2285 * @rq:         the request to complete
2286 * @error:      %0 for success, < %0 for error
2287 * @nr_bytes:   number of bytes to complete @rq
2288 * @bidi_bytes: number of bytes to complete @rq->next_rq
2289 *
2290 * Description:
2291 *     Identical to blk_end_bidi_request() except that queue lock is
2292 *     assumed to be locked on entry and remains so on return.
2293 *
2294 * Return:
2295 *     %false - we are done with this request
2296 *     %true  - still buffers pending for this request
2297 **/
2298static bool __blk_end_bidi_request(struct request *rq, int error,
2299                                   unsigned int nr_bytes, unsigned int bidi_bytes)
2300{
2301        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2302                return true;
2303
2304        blk_finish_request(rq, error);
2305
2306        return false;
2307}
2308
2309/**
2310 * blk_end_request - Helper function for drivers to complete the request.
2311 * @rq:       the request being processed
2312 * @error:    %0 for success, < %0 for error
2313 * @nr_bytes: number of bytes to complete
2314 *
2315 * Description:
2316 *     Ends I/O on a number of bytes attached to @rq.
2317 *     If @rq has leftover, sets it up for the next range of segments.
2318 *
2319 * Return:
2320 *     %false - we are done with this request
2321 *     %true  - still buffers pending for this request
2322 **/
2323bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2324{
2325        return blk_end_bidi_request(rq, error, nr_bytes, 0);
2326}
2327EXPORT_SYMBOL(blk_end_request);
2328
2329/**
2330 * blk_end_request_all - Helper function for drives to finish the request.
2331 * @rq: the request to finish
2332 * @error: %0 for success, < %0 for error
2333 *
2334 * Description:
2335 *     Completely finish @rq.
2336 */
2337void blk_end_request_all(struct request *rq, int error)
2338{
2339        bool pending;
2340        unsigned int bidi_bytes = 0;
2341
2342        if (unlikely(blk_bidi_rq(rq)))
2343                bidi_bytes = blk_rq_bytes(rq->next_rq);
2344
2345        pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2346        BUG_ON(pending);
2347}
2348EXPORT_SYMBOL(blk_end_request_all);
2349
2350/**
2351 * blk_end_request_cur - Helper function to finish the current request chunk.
2352 * @rq: the request to finish the current chunk for
2353 * @error: %0 for success, < %0 for error
2354 *
2355 * Description:
2356 *     Complete the current consecutively mapped chunk from @rq.
2357 *
2358 * Return:
2359 *     %false - we are done with this request
2360 *     %true  - still buffers pending for this request
2361 */
2362bool blk_end_request_cur(struct request *rq, int error)
2363{
2364        return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2365}
2366EXPORT_SYMBOL(blk_end_request_cur);
2367
2368/**
2369 * blk_end_request_err - Finish a request till the next failure boundary.
2370 * @rq: the request to finish till the next failure boundary for
2371 * @error: must be negative errno
2372 *
2373 * Description:
2374 *     Complete @rq till the next failure boundary.
2375 *
2376 * Return:
2377 *     %false - we are done with this request
2378 *     %true  - still buffers pending for this request
2379 */
2380bool blk_end_request_err(struct request *rq, int error)
2381{
2382        WARN_ON(error >= 0);
2383        return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2384}
2385EXPORT_SYMBOL_GPL(blk_end_request_err);
2386
2387/**
2388 * __blk_end_request - Helper function for drivers to complete the request.
2389 * @rq:       the request being processed
2390 * @error:    %0 for success, < %0 for error
2391 * @nr_bytes: number of bytes to complete
2392 *
2393 * Description:
2394 *     Must be called with queue lock held unlike blk_end_request().
2395 *
2396 * Return:
2397 *     %false - we are done with this request
2398 *     %true  - still buffers pending for this request
2399 **/
2400bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2401{
2402        return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2403}
2404EXPORT_SYMBOL(__blk_end_request);
2405
2406/**
2407 * __blk_end_request_all - Helper function for drives to finish the request.
2408 * @rq: the request to finish
2409 * @error: %0 for success, < %0 for error
2410 *
2411 * Description:
2412 *     Completely finish @rq.  Must be called with queue lock held.
2413 */
2414void __blk_end_request_all(struct request *rq, int error)
2415{
2416        bool pending;
2417        unsigned int bidi_bytes = 0;
2418
2419        if (unlikely(blk_bidi_rq(rq)))
2420                bidi_bytes = blk_rq_bytes(rq->next_rq);
2421
2422        pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2423        BUG_ON(pending);
2424}
2425EXPORT_SYMBOL(__blk_end_request_all);
2426
2427/**
2428 * __blk_end_request_cur - Helper function to finish the current request chunk.
2429 * @rq: the request to finish the current chunk for
2430 * @error: %0 for success, < %0 for error
2431 *
2432 * Description:
2433 *     Complete the current consecutively mapped chunk from @rq.  Must
2434 *     be called with queue lock held.
2435 *
2436 * Return:
2437 *     %false - we are done with this request
2438 *     %true  - still buffers pending for this request
2439 */
2440bool __blk_end_request_cur(struct request *rq, int error)
2441{
2442        return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2443}
2444EXPORT_SYMBOL(__blk_end_request_cur);
2445
2446/**
2447 * __blk_end_request_err - Finish a request till the next failure boundary.
2448 * @rq: the request to finish till the next failure boundary for
2449 * @error: must be negative errno
2450 *
2451 * Description:
2452 *     Complete @rq till the next failure boundary.  Must be called
2453 *     with queue lock held.
2454 *
2455 * Return:
2456 *     %false - we are done with this request
2457 *     %true  - still buffers pending for this request
2458 */
2459bool __blk_end_request_err(struct request *rq, int error)
2460{
2461        WARN_ON(error >= 0);
2462        return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2463}
2464EXPORT_SYMBOL_GPL(__blk_end_request_err);
2465
2466void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2467                     struct bio *bio)
2468{
2469        /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2470        rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2471
2472        if (bio_has_data(bio)) {
2473                rq->nr_phys_segments = bio_phys_segments(q, bio);
2474                rq->buffer = bio_data(bio);
2475        }
2476        rq->__data_len = bio->bi_size;
2477        rq->bio = rq->biotail = bio;
2478
2479        if (bio->bi_bdev)
2480                rq->rq_disk = bio->bi_bdev->bd_disk;
2481}
2482
2483#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2484/**
2485 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2486 * @rq: the request to be flushed
2487 *
2488 * Description:
2489 *     Flush all pages in @rq.
2490 */
2491void rq_flush_dcache_pages(struct request *rq)
2492{
2493        struct req_iterator iter;
2494        struct bio_vec *bvec;
2495
2496        rq_for_each_segment(bvec, rq, iter)
2497                flush_dcache_page(bvec->bv_page);
2498}
2499EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2500#endif
2501
2502/**
2503 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2504 * @q : the queue of the device being checked
2505 *
2506 * Description:
2507 *    Check if underlying low-level drivers of a device are busy.
2508 *    If the drivers want to export their busy state, they must set own
2509 *    exporting function using blk_queue_lld_busy() first.
2510 *
2511 *    Basically, this function is used only by request stacking drivers
2512 *    to stop dispatching requests to underlying devices when underlying
2513 *    devices are busy.  This behavior helps more I/O merging on the queue
2514 *    of the request stacking driver and prevents I/O throughput regression
2515 *    on burst I/O load.
2516 *
2517 * Return:
2518 *    0 - Not busy (The request stacking driver should dispatch request)
2519 *    1 - Busy (The request stacking driver should stop dispatching request)
2520 */
2521int blk_lld_busy(struct request_queue *q)
2522{
2523        if (q->lld_busy_fn)
2524                return q->lld_busy_fn(q);
2525
2526        return 0;
2527}
2528EXPORT_SYMBOL_GPL(blk_lld_busy);
2529
2530/**
2531 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2532 * @rq: the clone request to be cleaned up
2533 *
2534 * Description:
2535 *     Free all bios in @rq for a cloned request.
2536 */
2537void blk_rq_unprep_clone(struct request *rq)
2538{
2539        struct bio *bio;
2540
2541        while ((bio = rq->bio) != NULL) {
2542                rq->bio = bio->bi_next;
2543
2544                bio_put(bio);
2545        }
2546}
2547EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2548
2549/*
2550 * Copy attributes of the original request to the clone request.
2551 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2552 */
2553static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2554{
2555        dst->cpu = src->cpu;
2556        dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2557        dst->cmd_type = src->cmd_type;
2558        dst->__sector = blk_rq_pos(src);
2559        dst->__data_len = blk_rq_bytes(src);
2560        dst->nr_phys_segments = src->nr_phys_segments;
2561        dst->ioprio = src->ioprio;
2562        dst->extra_len = src->extra_len;
2563}
2564
2565/**
2566 * blk_rq_prep_clone - Helper function to setup clone request
2567 * @rq: the request to be setup
2568 * @rq_src: original request to be cloned
2569 * @bs: bio_set that bios for clone are allocated from
2570 * @gfp_mask: memory allocation mask for bio
2571 * @bio_ctr: setup function to be called for each clone bio.
2572 *           Returns %0 for success, non %0 for failure.
2573 * @data: private data to be passed to @bio_ctr
2574 *
2575 * Description:
2576 *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2577 *     The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2578 *     are not copied, and copying such parts is the caller's responsibility.
2579 *     Also, pages which the original bios are pointing to are not copied
2580 *     and the cloned bios just point same pages.
2581 *     So cloned bios must be completed before original bios, which means
2582 *     the caller must complete @rq before @rq_src.
2583 */
2584int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2585                      struct bio_set *bs, gfp_t gfp_mask,
2586                      int (*bio_ctr)(struct bio *, struct bio *, void *),
2587                      void *data)
2588{
2589        struct bio *bio, *bio_src;
2590
2591        if (!bs)
2592                bs = fs_bio_set;
2593
2594        blk_rq_init(NULL, rq);
2595
2596        __rq_for_each_bio(bio_src, rq_src) {
2597                bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2598                if (!bio)
2599                        goto free_and_out;
2600
2601                __bio_clone(bio, bio_src);
2602
2603                if (bio_integrity(bio_src) &&
2604                    bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2605                        goto free_and_out;
2606
2607                if (bio_ctr && bio_ctr(bio, bio_src, data))
2608                        goto free_and_out;
2609
2610                if (rq->bio) {
2611                        rq->biotail->bi_next = bio;
2612                        rq->biotail = bio;
2613                } else
2614                        rq->bio = rq->biotail = bio;
2615        }
2616
2617        __blk_rq_prep_clone(rq, rq_src);
2618
2619        return 0;
2620
2621free_and_out:
2622        if (bio)
2623                bio_free(bio, bs);
2624        blk_rq_unprep_clone(rq);
2625
2626        return -ENOMEM;
2627}
2628EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2629
2630int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2631{
2632        return queue_work(kblockd_workqueue, work);
2633}
2634EXPORT_SYMBOL(kblockd_schedule_work);
2635
2636int kblockd_schedule_delayed_work(struct request_queue *q,
2637                        struct delayed_work *dwork, unsigned long delay)
2638{
2639        return queue_delayed_work(kblockd_workqueue, dwork, delay);
2640}
2641EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2642
2643#define PLUG_MAGIC      0x91827364
2644
2645void blk_start_plug(struct blk_plug *plug)
2646{
2647        struct task_struct *tsk = current;
2648
2649        plug->magic = PLUG_MAGIC;
2650        INIT_LIST_HEAD(&plug->list);
2651        INIT_LIST_HEAD(&plug->cb_list);
2652        plug->should_sort = 0;
2653
2654        /*
2655         * If this is a nested plug, don't actually assign it. It will be
2656         * flushed on its own.
2657         */
2658        if (!tsk->plug) {
2659                /*
2660                 * Store ordering should not be needed here, since a potential
2661                 * preempt will imply a full memory barrier
2662                 */
2663                tsk->plug = plug;
2664        }
2665}
2666EXPORT_SYMBOL(blk_start_plug);
2667
2668static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2669{
2670        struct request *rqa = container_of(a, struct request, queuelist);
2671        struct request *rqb = container_of(b, struct request, queuelist);
2672
2673        return !(rqa->q <= rqb->q);
2674}
2675
2676/*
2677 * If 'from_schedule' is true, then postpone the dispatch of requests
2678 * until a safe kblockd context. We due this to avoid accidental big
2679 * additional stack usage in driver dispatch, in places where the originally
2680 * plugger did not intend it.
2681 */
2682static void queue_unplugged(struct request_queue *q, unsigned int depth,
2683                            bool from_schedule)
2684        __releases(q->queue_lock)
2685{
2686        trace_block_unplug(q, depth, !from_schedule);
2687
2688        /*
2689         * If we are punting this to kblockd, then we can safely drop
2690         * the queue_lock before waking kblockd (which needs to take
2691         * this lock).
2692         */
2693        if (from_schedule) {
2694                spin_unlock(q->queue_lock);
2695                blk_run_queue_async(q);
2696        } else {
2697                __blk_run_queue(q);
2698                spin_unlock(q->queue_lock);
2699        }
2700
2701}
2702
2703static void flush_plug_callbacks(struct blk_plug *plug)
2704{
2705        LIST_HEAD(callbacks);
2706
2707        if (list_empty(&plug->cb_list))
2708                return;
2709
2710        list_splice_init(&plug->cb_list, &callbacks);
2711
2712        while (!list_empty(&callbacks)) {
2713                struct blk_plug_cb *cb = list_first_entry(&callbacks,
2714                                                          struct blk_plug_cb,
2715                                                          list);
2716                list_del(&cb->list);
2717                cb->callback(cb);
2718        }
2719}
2720
2721void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2722{
2723        struct request_queue *q;
2724        unsigned long flags;
2725        struct request *rq;
2726        LIST_HEAD(list);
2727        unsigned int depth;
2728
2729        BUG_ON(plug->magic != PLUG_MAGIC);
2730
2731        flush_plug_callbacks(plug);
2732        if (list_empty(&plug->list))
2733                return;
2734
2735        list_splice_init(&plug->list, &list);
2736
2737        if (plug->should_sort) {
2738                list_sort(NULL, &list, plug_rq_cmp);
2739                plug->should_sort = 0;
2740        }
2741
2742        q = NULL;
2743        depth = 0;
2744
2745        /*
2746         * Save and disable interrupts here, to avoid doing it for every
2747         * queue lock we have to take.
2748         */
2749        local_irq_save(flags);
2750        while (!list_empty(&list)) {
2751                rq = list_entry_rq(list.next);
2752                list_del_init(&rq->queuelist);
2753                BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2754                BUG_ON(!rq->q);
2755                if (rq->q != q) {
2756                        /*
2757                         * This drops the queue lock
2758                         */
2759                        if (q)
2760                                queue_unplugged(q, depth, from_schedule);
2761                        q = rq->q;
2762                        depth = 0;
2763                        spin_lock(q->queue_lock);
2764                }
2765                rq->cmd_flags &= ~REQ_ON_PLUG;
2766
2767                /*
2768                 * rq is already accounted, so use raw insert
2769                 */
2770                if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2771                        __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2772                else
2773                        __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2774
2775                depth++;
2776        }
2777
2778        /*
2779         * This drops the queue lock
2780         */
2781        if (q)
2782                queue_unplugged(q, depth, from_schedule);
2783
2784        local_irq_restore(flags);
2785}
2786
2787void blk_finish_plug(struct blk_plug *plug)
2788{
2789        blk_flush_plug_list(plug, false);
2790
2791        if (plug == current->plug)
2792                current->plug = NULL;
2793}
2794EXPORT_SYMBOL(blk_finish_plug);
2795
2796int __init blk_dev_init(void)
2797{
2798        BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2799                        sizeof(((struct request *)0)->cmd_flags));
2800
2801        /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2802        kblockd_workqueue = alloc_workqueue("kblockd",
2803                                            WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2804        if (!kblockd_workqueue)
2805                panic("Failed to create kblockd\n");
2806
2807        request_cachep = kmem_cache_create("blkdev_requests",
2808                        sizeof(struct request), 0, SLAB_PANIC, NULL);
2809
2810        blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2811                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2812
2813        return 0;
2814}
2815