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/blktrace_api.h>
  30#include <linux/fault-inject.h>
  31#include <trace/block.h>
  32
  33#include "blk.h"
  34
  35DEFINE_TRACE(block_plug);
  36DEFINE_TRACE(block_unplug_io);
  37DEFINE_TRACE(block_unplug_timer);
  38DEFINE_TRACE(block_getrq);
  39DEFINE_TRACE(block_sleeprq);
  40DEFINE_TRACE(block_rq_requeue);
  41DEFINE_TRACE(block_bio_backmerge);
  42DEFINE_TRACE(block_bio_frontmerge);
  43DEFINE_TRACE(block_bio_queue);
  44DEFINE_TRACE(block_rq_complete);
  45DEFINE_TRACE(block_remap);      /* Also used in drivers/md/dm.c */
  46EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
  47
  48static int __make_request(struct request_queue *q, struct bio *bio);
  49
  50/*
  51 * For the allocated request tables
  52 */
  53static struct kmem_cache *request_cachep;
  54
  55/*
  56 * For queue allocation
  57 */
  58struct kmem_cache *blk_requestq_cachep;
  59
  60/*
  61 * Controlling structure to kblockd
  62 */
  63static struct workqueue_struct *kblockd_workqueue;
  64
  65static void drive_stat_acct(struct request *rq, int new_io)
  66{
  67        struct hd_struct *part;
  68        int rw = rq_data_dir(rq);
  69        int cpu;
  70
  71        if (!blk_fs_request(rq) || !blk_do_io_stat(rq))
  72                return;
  73
  74        cpu = part_stat_lock();
  75        part = disk_map_sector_rcu(rq->rq_disk, rq->sector);
  76
  77        if (!new_io)
  78                part_stat_inc(cpu, part, merges[rw]);
  79        else {
  80                part_round_stats(cpu, part);
  81                part_inc_in_flight(part);
  82        }
  83
  84        part_stat_unlock();
  85}
  86
  87void blk_queue_congestion_threshold(struct request_queue *q)
  88{
  89        int nr;
  90
  91        nr = q->nr_requests - (q->nr_requests / 8) + 1;
  92        if (nr > q->nr_requests)
  93                nr = q->nr_requests;
  94        q->nr_congestion_on = nr;
  95
  96        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
  97        if (nr < 1)
  98                nr = 1;
  99        q->nr_congestion_off = nr;
 100}
 101
 102/**
 103 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
 104 * @bdev:       device
 105 *
 106 * Locates the passed device's request queue and returns the address of its
 107 * backing_dev_info
 108 *
 109 * Will return NULL if the request queue cannot be located.
 110 */
 111struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
 112{
 113        struct backing_dev_info *ret = NULL;
 114        struct request_queue *q = bdev_get_queue(bdev);
 115
 116        if (q)
 117                ret = &q->backing_dev_info;
 118        return ret;
 119}
 120EXPORT_SYMBOL(blk_get_backing_dev_info);
 121
 122void blk_rq_init(struct request_queue *q, struct request *rq)
 123{
 124        memset(rq, 0, sizeof(*rq));
 125
 126        INIT_LIST_HEAD(&rq->queuelist);
 127        INIT_LIST_HEAD(&rq->timeout_list);
 128        rq->cpu = -1;
 129        rq->q = q;
 130        rq->sector = rq->hard_sector = (sector_t) -1;
 131        INIT_HLIST_NODE(&rq->hash);
 132        RB_CLEAR_NODE(&rq->rb_node);
 133        rq->cmd = rq->__cmd;
 134        rq->cmd_len = BLK_MAX_CDB;
 135        rq->tag = -1;
 136        rq->ref_count = 1;
 137}
 138EXPORT_SYMBOL(blk_rq_init);
 139
 140static void req_bio_endio(struct request *rq, struct bio *bio,
 141                          unsigned int nbytes, int error)
 142{
 143        struct request_queue *q = rq->q;
 144
 145        if (&q->bar_rq != rq) {
 146                if (error)
 147                        clear_bit(BIO_UPTODATE, &bio->bi_flags);
 148                else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 149                        error = -EIO;
 150
 151                if (unlikely(nbytes > bio->bi_size)) {
 152                        printk(KERN_ERR "%s: want %u bytes done, %u left\n",
 153                               __func__, nbytes, bio->bi_size);
 154                        nbytes = bio->bi_size;
 155                }
 156
 157                if (unlikely(rq->cmd_flags & REQ_QUIET))
 158                        set_bit(BIO_QUIET, &bio->bi_flags);
 159
 160                bio->bi_size -= nbytes;
 161                bio->bi_sector += (nbytes >> 9);
 162
 163                if (bio_integrity(bio))
 164                        bio_integrity_advance(bio, nbytes);
 165
 166                if (bio->bi_size == 0)
 167                        bio_endio(bio, error);
 168        } else {
 169
 170                /*
 171                 * Okay, this is the barrier request in progress, just
 172                 * record the error;
 173                 */
 174                if (error && !q->orderr)
 175                        q->orderr = error;
 176        }
 177}
 178
 179void blk_dump_rq_flags(struct request *rq, char *msg)
 180{
 181        int bit;
 182
 183        printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
 184                rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
 185                rq->cmd_flags);
 186
 187        printk(KERN_INFO "  sector %llu, nr/cnr %lu/%u\n",
 188                                                (unsigned long long)rq->sector,
 189                                                rq->nr_sectors,
 190                                                rq->current_nr_sectors);
 191        printk(KERN_INFO "  bio %p, biotail %p, buffer %p, data %p, len %u\n",
 192                                                rq->bio, rq->biotail,
 193                                                rq->buffer, rq->data,
 194                                                rq->data_len);
 195
 196        if (blk_pc_request(rq)) {
 197                printk(KERN_INFO "  cdb: ");
 198                for (bit = 0; bit < BLK_MAX_CDB; bit++)
 199                        printk("%02x ", rq->cmd[bit]);
 200                printk("\n");
 201        }
 202}
 203EXPORT_SYMBOL(blk_dump_rq_flags);
 204
 205/*
 206 * "plug" the device if there are no outstanding requests: this will
 207 * force the transfer to start only after we have put all the requests
 208 * on the list.
 209 *
 210 * This is called with interrupts off and no requests on the queue and
 211 * with the queue lock held.
 212 */
 213void blk_plug_device(struct request_queue *q)
 214{
 215        WARN_ON(!irqs_disabled());
 216
 217        /*
 218         * don't plug a stopped queue, it must be paired with blk_start_queue()
 219         * which will restart the queueing
 220         */
 221        if (blk_queue_stopped(q))
 222                return;
 223
 224        if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
 225                mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
 226                trace_block_plug(q);
 227        }
 228}
 229EXPORT_SYMBOL(blk_plug_device);
 230
 231/**
 232 * blk_plug_device_unlocked - plug a device without queue lock held
 233 * @q:    The &struct request_queue to plug
 234 *
 235 * Description:
 236 *   Like @blk_plug_device(), but grabs the queue lock and disables
 237 *   interrupts.
 238 **/
 239void blk_plug_device_unlocked(struct request_queue *q)
 240{
 241        unsigned long flags;
 242
 243        spin_lock_irqsave(q->queue_lock, flags);
 244        blk_plug_device(q);
 245        spin_unlock_irqrestore(q->queue_lock, flags);
 246}
 247EXPORT_SYMBOL(blk_plug_device_unlocked);
 248
 249/*
 250 * remove the queue from the plugged list, if present. called with
 251 * queue lock held and interrupts disabled.
 252 */
 253int blk_remove_plug(struct request_queue *q)
 254{
 255        WARN_ON(!irqs_disabled());
 256
 257        if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
 258                return 0;
 259
 260        del_timer(&q->unplug_timer);
 261        return 1;
 262}
 263EXPORT_SYMBOL(blk_remove_plug);
 264
 265/*
 266 * remove the plug and let it rip..
 267 */
 268void __generic_unplug_device(struct request_queue *q)
 269{
 270        if (unlikely(blk_queue_stopped(q)))
 271                return;
 272        if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
 273                return;
 274
 275        q->request_fn(q);
 276}
 277
 278/**
 279 * generic_unplug_device - fire a request queue
 280 * @q:    The &struct request_queue in question
 281 *
 282 * Description:
 283 *   Linux uses plugging to build bigger requests queues before letting
 284 *   the device have at them. If a queue is plugged, the I/O scheduler
 285 *   is still adding and merging requests on the queue. Once the queue
 286 *   gets unplugged, the request_fn defined for the queue is invoked and
 287 *   transfers started.
 288 **/
 289void generic_unplug_device(struct request_queue *q)
 290{
 291        if (blk_queue_plugged(q)) {
 292                spin_lock_irq(q->queue_lock);
 293                __generic_unplug_device(q);
 294                spin_unlock_irq(q->queue_lock);
 295        }
 296}
 297EXPORT_SYMBOL(generic_unplug_device);
 298
 299static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
 300                                   struct page *page)
 301{
 302        struct request_queue *q = bdi->unplug_io_data;
 303
 304        blk_unplug(q);
 305}
 306
 307void blk_unplug_work(struct work_struct *work)
 308{
 309        struct request_queue *q =
 310                container_of(work, struct request_queue, unplug_work);
 311
 312        trace_block_unplug_io(q);
 313        q->unplug_fn(q);
 314}
 315
 316void blk_unplug_timeout(unsigned long data)
 317{
 318        struct request_queue *q = (struct request_queue *)data;
 319
 320        trace_block_unplug_timer(q);
 321        kblockd_schedule_work(q, &q->unplug_work);
 322}
 323
 324void blk_unplug(struct request_queue *q)
 325{
 326        /*
 327         * devices don't necessarily have an ->unplug_fn defined
 328         */
 329        if (q->unplug_fn) {
 330                trace_block_unplug_io(q);
 331                q->unplug_fn(q);
 332        }
 333}
 334EXPORT_SYMBOL(blk_unplug);
 335
 336static void blk_invoke_request_fn(struct request_queue *q)
 337{
 338        if (unlikely(blk_queue_stopped(q)))
 339                return;
 340
 341        /*
 342         * one level of recursion is ok and is much faster than kicking
 343         * the unplug handling
 344         */
 345        if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
 346                q->request_fn(q);
 347                queue_flag_clear(QUEUE_FLAG_REENTER, q);
 348        } else {
 349                queue_flag_set(QUEUE_FLAG_PLUGGED, q);
 350                kblockd_schedule_work(q, &q->unplug_work);
 351        }
 352}
 353
 354/**
 355 * blk_start_queue - restart a previously stopped queue
 356 * @q:    The &struct request_queue in question
 357 *
 358 * Description:
 359 *   blk_start_queue() will clear the stop flag on the queue, and call
 360 *   the request_fn for the queue if it was in a stopped state when
 361 *   entered. Also see blk_stop_queue(). Queue lock must be held.
 362 **/
 363void blk_start_queue(struct request_queue *q)
 364{
 365        WARN_ON(!irqs_disabled());
 366
 367        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 368        blk_invoke_request_fn(q);
 369}
 370EXPORT_SYMBOL(blk_start_queue);
 371
 372/**
 373 * blk_stop_queue - stop a queue
 374 * @q:    The &struct request_queue in question
 375 *
 376 * Description:
 377 *   The Linux block layer assumes that a block driver will consume all
 378 *   entries on the request queue when the request_fn strategy is called.
 379 *   Often this will not happen, because of hardware limitations (queue
 380 *   depth settings). If a device driver gets a 'queue full' response,
 381 *   or if it simply chooses not to queue more I/O at one point, it can
 382 *   call this function to prevent the request_fn from being called until
 383 *   the driver has signalled it's ready to go again. This happens by calling
 384 *   blk_start_queue() to restart queue operations. Queue lock must be held.
 385 **/
 386void blk_stop_queue(struct request_queue *q)
 387{
 388        blk_remove_plug(q);
 389        queue_flag_set(QUEUE_FLAG_STOPPED, q);
 390}
 391EXPORT_SYMBOL(blk_stop_queue);
 392
 393/**
 394 * blk_sync_queue - cancel any pending callbacks on a queue
 395 * @q: the queue
 396 *
 397 * Description:
 398 *     The block layer may perform asynchronous callback activity
 399 *     on a queue, such as calling the unplug function after a timeout.
 400 *     A block device may call blk_sync_queue to ensure that any
 401 *     such activity is cancelled, thus allowing it to release resources
 402 *     that the callbacks might use. The caller must already have made sure
 403 *     that its ->make_request_fn will not re-add plugging prior to calling
 404 *     this function.
 405 *
 406 */
 407void blk_sync_queue(struct request_queue *q)
 408{
 409        del_timer_sync(&q->unplug_timer);
 410        del_timer_sync(&q->timeout);
 411        cancel_work_sync(&q->unplug_work);
 412}
 413EXPORT_SYMBOL(blk_sync_queue);
 414
 415/**
 416 * __blk_run_queue - run a single device queue
 417 * @q:  The queue to run
 418 *
 419 * Description:
 420 *    See @blk_run_queue. This variant must be called with the queue lock
 421 *    held and interrupts disabled.
 422 *
 423 */
 424void __blk_run_queue(struct request_queue *q)
 425{
 426        blk_remove_plug(q);
 427
 428        /*
 429         * Only recurse once to avoid overrunning the stack, let the unplug
 430         * handling reinvoke the handler shortly if we already got there.
 431         */
 432        if (!elv_queue_empty(q))
 433                blk_invoke_request_fn(q);
 434}
 435EXPORT_SYMBOL(__blk_run_queue);
 436
 437/**
 438 * blk_run_queue - run a single device queue
 439 * @q: The queue to run
 440 *
 441 * Description:
 442 *    Invoke request handling on this queue, if it has pending work to do.
 443 *    May be used to restart queueing when a request has completed. Also
 444 *    See @blk_start_queueing.
 445 *
 446 */
 447void blk_run_queue(struct request_queue *q)
 448{
 449        unsigned long flags;
 450
 451        spin_lock_irqsave(q->queue_lock, flags);
 452        __blk_run_queue(q);
 453        spin_unlock_irqrestore(q->queue_lock, flags);
 454}
 455EXPORT_SYMBOL(blk_run_queue);
 456
 457void blk_put_queue(struct request_queue *q)
 458{
 459        kobject_put(&q->kobj);
 460}
 461
 462void blk_cleanup_queue(struct request_queue *q)
 463{
 464        /*
 465         * We know we have process context here, so we can be a little
 466         * cautious and ensure that pending block actions on this device
 467         * are done before moving on. Going into this function, we should
 468         * not have processes doing IO to this device.
 469         */
 470        blk_sync_queue(q);
 471
 472        mutex_lock(&q->sysfs_lock);
 473        queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
 474        mutex_unlock(&q->sysfs_lock);
 475
 476        if (q->elevator)
 477                elevator_exit(q->elevator);
 478
 479        blk_put_queue(q);
 480}
 481EXPORT_SYMBOL(blk_cleanup_queue);
 482
 483static int blk_init_free_list(struct request_queue *q)
 484{
 485        struct request_list *rl = &q->rq;
 486
 487        rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
 488        rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
 489        rl->elvpriv = 0;
 490        init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
 491        init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
 492
 493        rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
 494                                mempool_free_slab, request_cachep, q->node);
 495
 496        if (!rl->rq_pool)
 497                return -ENOMEM;
 498
 499        return 0;
 500}
 501
 502struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 503{
 504        return blk_alloc_queue_node(gfp_mask, -1);
 505}
 506EXPORT_SYMBOL(blk_alloc_queue);
 507
 508struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 509{
 510        struct request_queue *q;
 511        int err;
 512
 513        q = kmem_cache_alloc_node(blk_requestq_cachep,
 514                                gfp_mask | __GFP_ZERO, node_id);
 515        if (!q)
 516                return NULL;
 517
 518        q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
 519        q->backing_dev_info.unplug_io_data = q;
 520        err = bdi_init(&q->backing_dev_info);
 521        if (err) {
 522                kmem_cache_free(blk_requestq_cachep, q);
 523                return NULL;
 524        }
 525
 526        init_timer(&q->unplug_timer);
 527        setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
 528        INIT_LIST_HEAD(&q->timeout_list);
 529        INIT_WORK(&q->unplug_work, blk_unplug_work);
 530
 531        kobject_init(&q->kobj, &blk_queue_ktype);
 532
 533        mutex_init(&q->sysfs_lock);
 534        spin_lock_init(&q->__queue_lock);
 535
 536        return q;
 537}
 538EXPORT_SYMBOL(blk_alloc_queue_node);
 539
 540/**
 541 * blk_init_queue  - prepare a request queue for use with a block device
 542 * @rfn:  The function to be called to process requests that have been
 543 *        placed on the queue.
 544 * @lock: Request queue spin lock
 545 *
 546 * Description:
 547 *    If a block device wishes to use the standard request handling procedures,
 548 *    which sorts requests and coalesces adjacent requests, then it must
 549 *    call blk_init_queue().  The function @rfn will be called when there
 550 *    are requests on the queue that need to be processed.  If the device
 551 *    supports plugging, then @rfn may not be called immediately when requests
 552 *    are available on the queue, but may be called at some time later instead.
 553 *    Plugged queues are generally unplugged when a buffer belonging to one
 554 *    of the requests on the queue is needed, or due to memory pressure.
 555 *
 556 *    @rfn is not required, or even expected, to remove all requests off the
 557 *    queue, but only as many as it can handle at a time.  If it does leave
 558 *    requests on the queue, it is responsible for arranging that the requests
 559 *    get dealt with eventually.
 560 *
 561 *    The queue spin lock must be held while manipulating the requests on the
 562 *    request queue; this lock will be taken also from interrupt context, so irq
 563 *    disabling is needed for it.
 564 *
 565 *    Function returns a pointer to the initialized request queue, or %NULL if
 566 *    it didn't succeed.
 567 *
 568 * Note:
 569 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
 570 *    when the block device is deactivated (such as at module unload).
 571 **/
 572
 573struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
 574{
 575        return blk_init_queue_node(rfn, lock, -1);
 576}
 577EXPORT_SYMBOL(blk_init_queue);
 578
 579struct request_queue *
 580blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
 581{
 582        struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
 583
 584        if (!q)
 585                return NULL;
 586
 587        q->node = node_id;
 588        if (blk_init_free_list(q)) {
 589                kmem_cache_free(blk_requestq_cachep, q);
 590                return NULL;
 591        }
 592
 593        /*
 594         * if caller didn't supply a lock, they get per-queue locking with
 595         * our embedded lock
 596         */
 597        if (!lock)
 598                lock = &q->__queue_lock;
 599
 600        q->request_fn           = rfn;
 601        q->prep_rq_fn           = NULL;
 602        q->unplug_fn            = generic_unplug_device;
 603        q->queue_flags          = QUEUE_FLAG_DEFAULT;
 604        q->queue_lock           = lock;
 605
 606        /*
 607         * This also sets hw/phys segments, boundary and size
 608         */
 609        blk_queue_make_request(q, __make_request);
 610
 611        q->sg_reserved_size = INT_MAX;
 612
 613        blk_set_cmd_filter_defaults(&q->cmd_filter);
 614
 615        /*
 616         * all done
 617         */
 618        if (!elevator_init(q, NULL)) {
 619                blk_queue_congestion_threshold(q);
 620                return q;
 621        }
 622
 623        blk_put_queue(q);
 624        return NULL;
 625}
 626EXPORT_SYMBOL(blk_init_queue_node);
 627
 628int blk_get_queue(struct request_queue *q)
 629{
 630        if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
 631                kobject_get(&q->kobj);
 632                return 0;
 633        }
 634
 635        return 1;
 636}
 637
 638static inline void blk_free_request(struct request_queue *q, struct request *rq)
 639{
 640        if (rq->cmd_flags & REQ_ELVPRIV)
 641                elv_put_request(q, rq);
 642        mempool_free(rq, q->rq.rq_pool);
 643}
 644
 645static struct request *
 646blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
 647{
 648        struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
 649
 650        if (!rq)
 651                return NULL;
 652
 653        blk_rq_init(q, rq);
 654
 655        rq->cmd_flags = flags | REQ_ALLOCED;
 656
 657        if (priv) {
 658                if (unlikely(elv_set_request(q, rq, gfp_mask))) {
 659                        mempool_free(rq, q->rq.rq_pool);
 660                        return NULL;
 661                }
 662                rq->cmd_flags |= REQ_ELVPRIV;
 663        }
 664
 665        return rq;
 666}
 667
 668/*
 669 * ioc_batching returns true if the ioc is a valid batching request and
 670 * should be given priority access to a request.
 671 */
 672static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
 673{
 674        if (!ioc)
 675                return 0;
 676
 677        /*
 678         * Make sure the process is able to allocate at least 1 request
 679         * even if the batch times out, otherwise we could theoretically
 680         * lose wakeups.
 681         */
 682        return ioc->nr_batch_requests == q->nr_batching ||
 683                (ioc->nr_batch_requests > 0
 684                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
 685}
 686
 687/*
 688 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
 689 * will cause the process to be a "batcher" on all queues in the system. This
 690 * is the behaviour we want though - once it gets a wakeup it should be given
 691 * a nice run.
 692 */
 693static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
 694{
 695        if (!ioc || ioc_batching(q, ioc))
 696                return;
 697
 698        ioc->nr_batch_requests = q->nr_batching;
 699        ioc->last_waited = jiffies;
 700}
 701
 702static void __freed_request(struct request_queue *q, int sync)
 703{
 704        struct request_list *rl = &q->rq;
 705
 706        if (rl->count[sync] < queue_congestion_off_threshold(q))
 707                blk_clear_queue_congested(q, sync);
 708
 709        if (rl->count[sync] + 1 <= q->nr_requests) {
 710                if (waitqueue_active(&rl->wait[sync]))
 711                        wake_up(&rl->wait[sync]);
 712
 713                blk_clear_queue_full(q, sync);
 714        }
 715}
 716
 717/*
 718 * A request has just been released.  Account for it, update the full and
 719 * congestion status, wake up any waiters.   Called under q->queue_lock.
 720 */
 721static void freed_request(struct request_queue *q, int sync, int priv)
 722{
 723        struct request_list *rl = &q->rq;
 724
 725        rl->count[sync]--;
 726        if (priv)
 727                rl->elvpriv--;
 728
 729        __freed_request(q, sync);
 730
 731        if (unlikely(rl->starved[sync ^ 1]))
 732                __freed_request(q, sync ^ 1);
 733}
 734
 735/*
 736 * Get a free request, queue_lock must be held.
 737 * Returns NULL on failure, with queue_lock held.
 738 * Returns !NULL on success, with queue_lock *not held*.
 739 */
 740static struct request *get_request(struct request_queue *q, int rw_flags,
 741                                   struct bio *bio, gfp_t gfp_mask)
 742{
 743        struct request *rq = NULL;
 744        struct request_list *rl = &q->rq;
 745        struct io_context *ioc = NULL;
 746        const bool is_sync = rw_is_sync(rw_flags) != 0;
 747        int may_queue, priv;
 748
 749        may_queue = elv_may_queue(q, rw_flags);
 750        if (may_queue == ELV_MQUEUE_NO)
 751                goto rq_starved;
 752
 753        if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
 754                if (rl->count[is_sync]+1 >= q->nr_requests) {
 755                        ioc = current_io_context(GFP_ATOMIC, q->node);
 756                        /*
 757                         * The queue will fill after this allocation, so set
 758                         * it as full, and mark this process as "batching".
 759                         * This process will be allowed to complete a batch of
 760                         * requests, others will be blocked.
 761                         */
 762                        if (!blk_queue_full(q, is_sync)) {
 763                                ioc_set_batching(q, ioc);
 764                                blk_set_queue_full(q, is_sync);
 765                        } else {
 766                                if (may_queue != ELV_MQUEUE_MUST
 767                                                && !ioc_batching(q, ioc)) {
 768                                        /*
 769                                         * The queue is full and the allocating
 770                                         * process is not a "batcher", and not
 771                                         * exempted by the IO scheduler
 772                                         */
 773                                        goto out;
 774                                }
 775                        }
 776                }
 777                blk_set_queue_congested(q, is_sync);
 778        }
 779
 780        /*
 781         * Only allow batching queuers to allocate up to 50% over the defined
 782         * limit of requests, otherwise we could have thousands of requests
 783         * allocated with any setting of ->nr_requests
 784         */
 785        if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
 786                goto out;
 787
 788        rl->count[is_sync]++;
 789        rl->starved[is_sync] = 0;
 790
 791        priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
 792        if (priv)
 793                rl->elvpriv++;
 794
 795        if (blk_queue_io_stat(q))
 796                rw_flags |= REQ_IO_STAT;
 797        spin_unlock_irq(q->queue_lock);
 798
 799        rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
 800        if (unlikely(!rq)) {
 801                /*
 802                 * Allocation failed presumably due to memory. Undo anything
 803                 * we might have messed up.
 804                 *
 805                 * Allocating task should really be put onto the front of the
 806                 * wait queue, but this is pretty rare.
 807                 */
 808                spin_lock_irq(q->queue_lock);
 809                freed_request(q, is_sync, priv);
 810
 811                /*
 812                 * in the very unlikely event that allocation failed and no
 813                 * requests for this direction was pending, mark us starved
 814                 * so that freeing of a request in the other direction will
 815                 * notice us. another possible fix would be to split the
 816                 * rq mempool into READ and WRITE
 817                 */
 818rq_starved:
 819                if (unlikely(rl->count[is_sync] == 0))
 820                        rl->starved[is_sync] = 1;
 821
 822                goto out;
 823        }
 824
 825        /*
 826         * ioc may be NULL here, and ioc_batching will be false. That's
 827         * OK, if the queue is under the request limit then requests need
 828         * not count toward the nr_batch_requests limit. There will always
 829         * be some limit enforced by BLK_BATCH_TIME.
 830         */
 831        if (ioc_batching(q, ioc))
 832                ioc->nr_batch_requests--;
 833
 834        trace_block_getrq(q, bio, rw_flags & 1);
 835out:
 836        return rq;
 837}
 838
 839/*
 840 * No available requests for this queue, unplug the device and wait for some
 841 * requests to become available.
 842 *
 843 * Called with q->queue_lock held, and returns with it unlocked.
 844 */
 845static struct request *get_request_wait(struct request_queue *q, int rw_flags,
 846                                        struct bio *bio)
 847{
 848        const bool is_sync = rw_is_sync(rw_flags) != 0;
 849        struct request *rq;
 850
 851        rq = get_request(q, rw_flags, bio, GFP_NOIO);
 852        while (!rq) {
 853                DEFINE_WAIT(wait);
 854                struct io_context *ioc;
 855                struct request_list *rl = &q->rq;
 856
 857                prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
 858                                TASK_UNINTERRUPTIBLE);
 859
 860                trace_block_sleeprq(q, bio, rw_flags & 1);
 861
 862                __generic_unplug_device(q);
 863                spin_unlock_irq(q->queue_lock);
 864                io_schedule();
 865
 866                /*
 867                 * After sleeping, we become a "batching" process and
 868                 * will be able to allocate at least one request, and
 869                 * up to a big batch of them for a small period time.
 870                 * See ioc_batching, ioc_set_batching
 871                 */
 872                ioc = current_io_context(GFP_NOIO, q->node);
 873                ioc_set_batching(q, ioc);
 874
 875                spin_lock_irq(q->queue_lock);
 876                finish_wait(&rl->wait[is_sync], &wait);
 877
 878                rq = get_request(q, rw_flags, bio, GFP_NOIO);
 879        };
 880
 881        return rq;
 882}
 883
 884struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
 885{
 886        struct request *rq;
 887
 888        BUG_ON(rw != READ && rw != WRITE);
 889
 890        spin_lock_irq(q->queue_lock);
 891        if (gfp_mask & __GFP_WAIT) {
 892                rq = get_request_wait(q, rw, NULL);
 893        } else {
 894                rq = get_request(q, rw, NULL, gfp_mask);
 895                if (!rq)
 896                        spin_unlock_irq(q->queue_lock);
 897        }
 898        /* q->queue_lock is unlocked at this point */
 899
 900        return rq;
 901}
 902EXPORT_SYMBOL(blk_get_request);
 903
 904/**
 905 * blk_start_queueing - initiate dispatch of requests to device
 906 * @q:          request queue to kick into gear
 907 *
 908 * This is basically a helper to remove the need to know whether a queue
 909 * is plugged or not if someone just wants to initiate dispatch of requests
 910 * for this queue. Should be used to start queueing on a device outside
 911 * of ->request_fn() context. Also see @blk_run_queue.
 912 *
 913 * The queue lock must be held with interrupts disabled.
 914 */
 915void blk_start_queueing(struct request_queue *q)
 916{
 917        if (!blk_queue_plugged(q)) {
 918                if (unlikely(blk_queue_stopped(q)))
 919                        return;
 920                q->request_fn(q);
 921        } else
 922                __generic_unplug_device(q);
 923}
 924EXPORT_SYMBOL(blk_start_queueing);
 925
 926/**
 927 * blk_requeue_request - put a request back on queue
 928 * @q:          request queue where request should be inserted
 929 * @rq:         request to be inserted
 930 *
 931 * Description:
 932 *    Drivers often keep queueing requests until the hardware cannot accept
 933 *    more, when that condition happens we need to put the request back
 934 *    on the queue. Must be called with queue lock held.
 935 */
 936void blk_requeue_request(struct request_queue *q, struct request *rq)
 937{
 938        blk_delete_timer(rq);
 939        blk_clear_rq_complete(rq);
 940        trace_block_rq_requeue(q, rq);
 941
 942        if (blk_rq_tagged(rq))
 943                blk_queue_end_tag(q, rq);
 944
 945        elv_requeue_request(q, rq);
 946}
 947EXPORT_SYMBOL(blk_requeue_request);
 948
 949/**
 950 * blk_insert_request - insert a special request into a request queue
 951 * @q:          request queue where request should be inserted
 952 * @rq:         request to be inserted
 953 * @at_head:    insert request at head or tail of queue
 954 * @data:       private data
 955 *
 956 * Description:
 957 *    Many block devices need to execute commands asynchronously, so they don't
 958 *    block the whole kernel from preemption during request execution.  This is
 959 *    accomplished normally by inserting aritficial requests tagged as
 960 *    REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
 961 *    be scheduled for actual execution by the request queue.
 962 *
 963 *    We have the option of inserting the head or the tail of the queue.
 964 *    Typically we use the tail for new ioctls and so forth.  We use the head
 965 *    of the queue for things like a QUEUE_FULL message from a device, or a
 966 *    host that is unable to accept a particular command.
 967 */
 968void blk_insert_request(struct request_queue *q, struct request *rq,
 969                        int at_head, void *data)
 970{
 971        int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
 972        unsigned long flags;
 973
 974        /*
 975         * tell I/O scheduler that this isn't a regular read/write (ie it
 976         * must not attempt merges on this) and that it acts as a soft
 977         * barrier
 978         */
 979        rq->cmd_type = REQ_TYPE_SPECIAL;
 980        rq->cmd_flags |= REQ_SOFTBARRIER;
 981
 982        rq->special = data;
 983
 984        spin_lock_irqsave(q->queue_lock, flags);
 985
 986        /*
 987         * If command is tagged, release the tag
 988         */
 989        if (blk_rq_tagged(rq))
 990                blk_queue_end_tag(q, rq);
 991
 992        drive_stat_acct(rq, 1);
 993        __elv_add_request(q, rq, where, 0);
 994        blk_start_queueing(q);
 995        spin_unlock_irqrestore(q->queue_lock, flags);
 996}
 997EXPORT_SYMBOL(blk_insert_request);
 998
 999/*
1000 * add-request adds a request to the linked list.
1001 * queue lock is held and interrupts disabled, as we muck with the
1002 * request queue list.
1003 */
1004static inline void add_request(struct request_queue *q, struct request *req)
1005{
1006        drive_stat_acct(req, 1);
1007
1008        /*
1009         * elevator indicated where it wants this request to be
1010         * inserted at elevator_merge time
1011         */
1012        __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1013}
1014
1015static void part_round_stats_single(int cpu, struct hd_struct *part,
1016                                    unsigned long now)
1017{
1018        if (now == part->stamp)
1019                return;
1020
1021        if (part->in_flight) {
1022                __part_stat_add(cpu, part, time_in_queue,
1023                                part->in_flight * (now - part->stamp));
1024                __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1025        }
1026        part->stamp = now;
1027}
1028
1029/**
1030 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1031 * @cpu: cpu number for stats access
1032 * @part: target partition
1033 *
1034 * The average IO queue length and utilisation statistics are maintained
1035 * by observing the current state of the queue length and the amount of
1036 * time it has been in this state for.
1037 *
1038 * Normally, that accounting is done on IO completion, but that can result
1039 * in more than a second's worth of IO being accounted for within any one
1040 * second, leading to >100% utilisation.  To deal with that, we call this
1041 * function to do a round-off before returning the results when reading
1042 * /proc/diskstats.  This accounts immediately for all queue usage up to
1043 * the current jiffies and restarts the counters again.
1044 */
1045void part_round_stats(int cpu, struct hd_struct *part)
1046{
1047        unsigned long now = jiffies;
1048
1049        if (part->partno)
1050                part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1051        part_round_stats_single(cpu, part, now);
1052}
1053EXPORT_SYMBOL_GPL(part_round_stats);
1054
1055/*
1056 * queue lock must be held
1057 */
1058void __blk_put_request(struct request_queue *q, struct request *req)
1059{
1060        if (unlikely(!q))
1061                return;
1062        if (unlikely(--req->ref_count))
1063                return;
1064
1065        elv_completed_request(q, req);
1066
1067        /* this is a bio leak */
1068        WARN_ON(req->bio != NULL);
1069
1070        /*
1071         * Request may not have originated from ll_rw_blk. if not,
1072         * it didn't come out of our reserved rq pools
1073         */
1074        if (req->cmd_flags & REQ_ALLOCED) {
1075                int is_sync = rq_is_sync(req) != 0;
1076                int priv = req->cmd_flags & REQ_ELVPRIV;
1077
1078                BUG_ON(!list_empty(&req->queuelist));
1079                BUG_ON(!hlist_unhashed(&req->hash));
1080
1081                blk_free_request(q, req);
1082                freed_request(q, is_sync, priv);
1083        }
1084}
1085EXPORT_SYMBOL_GPL(__blk_put_request);
1086
1087void blk_put_request(struct request *req)
1088{
1089        unsigned long flags;
1090        struct request_queue *q = req->q;
1091
1092        spin_lock_irqsave(q->queue_lock, flags);
1093        __blk_put_request(q, req);
1094        spin_unlock_irqrestore(q->queue_lock, flags);
1095}
1096EXPORT_SYMBOL(blk_put_request);
1097
1098void init_request_from_bio(struct request *req, struct bio *bio)
1099{
1100        req->cpu = bio->bi_comp_cpu;
1101        req->cmd_type = REQ_TYPE_FS;
1102
1103        /*
1104         * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1105         */
1106        if (bio_rw_ahead(bio))
1107                req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1108                                   REQ_FAILFAST_DRIVER);
1109        if (bio_failfast_dev(bio))
1110                req->cmd_flags |= REQ_FAILFAST_DEV;
1111        if (bio_failfast_transport(bio))
1112                req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1113        if (bio_failfast_driver(bio))
1114                req->cmd_flags |= REQ_FAILFAST_DRIVER;
1115
1116        /*
1117         * REQ_BARRIER implies no merging, but lets make it explicit
1118         */
1119        if (unlikely(bio_discard(bio))) {
1120                req->cmd_flags |= REQ_DISCARD;
1121                if (bio_barrier(bio))
1122                        req->cmd_flags |= REQ_SOFTBARRIER;
1123                req->q->prepare_discard_fn(req->q, req);
1124        } else if (unlikely(bio_barrier(bio)))
1125                req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
1126
1127        if (bio_sync(bio))
1128                req->cmd_flags |= REQ_RW_SYNC;
1129        if (bio_rw_meta(bio))
1130                req->cmd_flags |= REQ_RW_META;
1131        if (bio_noidle(bio))
1132                req->cmd_flags |= REQ_NOIDLE;
1133
1134        req->errors = 0;
1135        req->hard_sector = req->sector = bio->bi_sector;
1136        req->ioprio = bio_prio(bio);
1137        req->start_time = jiffies;
1138        blk_rq_bio_prep(req->q, req, bio);
1139}
1140
1141/*
1142 * Only disabling plugging for non-rotational devices if it does tagging
1143 * as well, otherwise we do need the proper merging
1144 */
1145static inline bool queue_should_plug(struct request_queue *q)
1146{
1147        return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1148}
1149
1150static int __make_request(struct request_queue *q, struct bio *bio)
1151{
1152        struct request *req;
1153        int el_ret, nr_sectors;
1154        const unsigned short prio = bio_prio(bio);
1155        const int sync = bio_sync(bio);
1156        const int unplug = bio_unplug(bio);
1157        int rw_flags;
1158
1159        nr_sectors = bio_sectors(bio);
1160
1161        /*
1162         * low level driver can indicate that it wants pages above a
1163         * certain limit bounced to low memory (ie for highmem, or even
1164         * ISA dma in theory)
1165         */
1166        blk_queue_bounce(q, &bio);
1167
1168        spin_lock_irq(q->queue_lock);
1169
1170        if (unlikely(bio_barrier(bio)) || elv_queue_empty(q))
1171                goto get_rq;
1172
1173        el_ret = elv_merge(q, &req, bio);
1174        switch (el_ret) {
1175        case ELEVATOR_BACK_MERGE:
1176                BUG_ON(!rq_mergeable(req));
1177
1178                if (!ll_back_merge_fn(q, req, bio))
1179                        break;
1180
1181                trace_block_bio_backmerge(q, bio);
1182
1183                req->biotail->bi_next = bio;
1184                req->biotail = bio;
1185                req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1186                req->ioprio = ioprio_best(req->ioprio, prio);
1187                if (!blk_rq_cpu_valid(req))
1188                        req->cpu = bio->bi_comp_cpu;
1189                drive_stat_acct(req, 0);
1190                if (!attempt_back_merge(q, req))
1191                        elv_merged_request(q, req, el_ret);
1192                goto out;
1193
1194        case ELEVATOR_FRONT_MERGE:
1195                BUG_ON(!rq_mergeable(req));
1196
1197                if (!ll_front_merge_fn(q, req, bio))
1198                        break;
1199
1200                trace_block_bio_frontmerge(q, bio);
1201
1202                bio->bi_next = req->bio;
1203                req->bio = bio;
1204
1205                /*
1206                 * may not be valid. if the low level driver said
1207                 * it didn't need a bounce buffer then it better
1208                 * not touch req->buffer either...
1209                 */
1210                req->buffer = bio_data(bio);
1211                req->current_nr_sectors = bio_cur_sectors(bio);
1212                req->hard_cur_sectors = req->current_nr_sectors;
1213                req->sector = req->hard_sector = bio->bi_sector;
1214                req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1215                req->ioprio = ioprio_best(req->ioprio, prio);
1216                if (!blk_rq_cpu_valid(req))
1217                        req->cpu = bio->bi_comp_cpu;
1218                drive_stat_acct(req, 0);
1219                if (!attempt_front_merge(q, req))
1220                        elv_merged_request(q, req, el_ret);
1221                goto out;
1222
1223        /* ELV_NO_MERGE: elevator says don't/can't merge. */
1224        default:
1225                ;
1226        }
1227
1228get_rq:
1229        /*
1230         * This sync check and mask will be re-done in init_request_from_bio(),
1231         * but we need to set it earlier to expose the sync flag to the
1232         * rq allocator and io schedulers.
1233         */
1234        rw_flags = bio_data_dir(bio);
1235        if (sync)
1236                rw_flags |= REQ_RW_SYNC;
1237
1238        /*
1239         * Grab a free request. This is might sleep but can not fail.
1240         * Returns with the queue unlocked.
1241         */
1242        req = get_request_wait(q, rw_flags, bio);
1243
1244        /*
1245         * After dropping the lock and possibly sleeping here, our request
1246         * may now be mergeable after it had proven unmergeable (above).
1247         * We don't worry about that case for efficiency. It won't happen
1248         * often, and the elevators are able to handle it.
1249         */
1250        init_request_from_bio(req, bio);
1251
1252        spin_lock_irq(q->queue_lock);
1253        if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1254            bio_flagged(bio, BIO_CPU_AFFINE))
1255                req->cpu = blk_cpu_to_group(smp_processor_id());
1256        if (queue_should_plug(q) && elv_queue_empty(q))
1257                blk_plug_device(q);
1258        add_request(q, req);
1259out:
1260        if (unplug || !queue_should_plug(q))
1261                __generic_unplug_device(q);
1262        spin_unlock_irq(q->queue_lock);
1263        return 0;
1264}
1265
1266/*
1267 * If bio->bi_dev is a partition, remap the location
1268 */
1269static inline void blk_partition_remap(struct bio *bio)
1270{
1271        struct block_device *bdev = bio->bi_bdev;
1272
1273        if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1274                struct hd_struct *p = bdev->bd_part;
1275
1276                bio->bi_sector += p->start_sect;
1277                bio->bi_bdev = bdev->bd_contains;
1278
1279                trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
1280                                    bdev->bd_dev, bio->bi_sector,
1281                                    bio->bi_sector - p->start_sect);
1282        }
1283}
1284
1285static void handle_bad_sector(struct bio *bio)
1286{
1287        char b[BDEVNAME_SIZE];
1288
1289        printk(KERN_INFO "attempt to access beyond end of device\n");
1290        printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1291                        bdevname(bio->bi_bdev, b),
1292                        bio->bi_rw,
1293                        (unsigned long long)bio->bi_sector + bio_sectors(bio),
1294                        (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1295
1296        set_bit(BIO_EOF, &bio->bi_flags);
1297}
1298
1299#ifdef CONFIG_FAIL_MAKE_REQUEST
1300
1301static DECLARE_FAULT_ATTR(fail_make_request);
1302
1303static int __init setup_fail_make_request(char *str)
1304{
1305        return setup_fault_attr(&fail_make_request, str);
1306}
1307__setup("fail_make_request=", setup_fail_make_request);
1308
1309static int should_fail_request(struct bio *bio)
1310{
1311        struct hd_struct *part = bio->bi_bdev->bd_part;
1312
1313        if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1314                return should_fail(&fail_make_request, bio->bi_size);
1315
1316        return 0;
1317}
1318
1319static int __init fail_make_request_debugfs(void)
1320{
1321        return init_fault_attr_dentries(&fail_make_request,
1322                                        "fail_make_request");
1323}
1324
1325late_initcall(fail_make_request_debugfs);
1326
1327#else /* CONFIG_FAIL_MAKE_REQUEST */
1328
1329static inline int should_fail_request(struct bio *bio)
1330{
1331        return 0;
1332}
1333
1334#endif /* CONFIG_FAIL_MAKE_REQUEST */
1335
1336/*
1337 * Check whether this bio extends beyond the end of the device.
1338 */
1339static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1340{
1341        sector_t maxsector;
1342
1343        if (!nr_sectors)
1344                return 0;
1345
1346        /* Test device or partition size, when known. */
1347        maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1348        if (maxsector) {
1349                sector_t sector = bio->bi_sector;
1350
1351                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1352                        /*
1353                         * This may well happen - the kernel calls bread()
1354                         * without checking the size of the device, e.g., when
1355                         * mounting a device.
1356                         */
1357                        handle_bad_sector(bio);
1358                        return 1;
1359                }
1360        }
1361
1362        return 0;
1363}
1364
1365/**
1366 * generic_make_request - hand a buffer to its device driver for I/O
1367 * @bio:  The bio describing the location in memory and on the device.
1368 *
1369 * generic_make_request() is used to make I/O requests of block
1370 * devices. It is passed a &struct bio, which describes the I/O that needs
1371 * to be done.
1372 *
1373 * generic_make_request() does not return any status.  The
1374 * success/failure status of the request, along with notification of
1375 * completion, is delivered asynchronously through the bio->bi_end_io
1376 * function described (one day) else where.
1377 *
1378 * The caller of generic_make_request must make sure that bi_io_vec
1379 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1380 * set to describe the device address, and the
1381 * bi_end_io and optionally bi_private are set to describe how
1382 * completion notification should be signaled.
1383 *
1384 * generic_make_request and the drivers it calls may use bi_next if this
1385 * bio happens to be merged with someone else, and may change bi_dev and
1386 * bi_sector for remaps as it sees fit.  So the values of these fields
1387 * should NOT be depended on after the call to generic_make_request.
1388 */
1389static inline void __generic_make_request(struct bio *bio)
1390{
1391        struct request_queue *q;
1392        sector_t old_sector;
1393        int ret, nr_sectors = bio_sectors(bio);
1394        dev_t old_dev;
1395        int err = -EIO;
1396
1397        might_sleep();
1398
1399        if (bio_check_eod(bio, nr_sectors))
1400                goto end_io;
1401
1402        /*
1403         * Resolve the mapping until finished. (drivers are
1404         * still free to implement/resolve their own stacking
1405         * by explicitly returning 0)
1406         *
1407         * NOTE: we don't repeat the blk_size check for each new device.
1408         * Stacking drivers are expected to know what they are doing.
1409         */
1410        old_sector = -1;
1411        old_dev = 0;
1412        do {
1413                char b[BDEVNAME_SIZE];
1414
1415                q = bdev_get_queue(bio->bi_bdev);
1416                if (unlikely(!q)) {
1417                        printk(KERN_ERR
1418                               "generic_make_request: Trying to access "
1419                                "nonexistent block-device %s (%Lu)\n",
1420                                bdevname(bio->bi_bdev, b),
1421                                (long long) bio->bi_sector);
1422                        goto end_io;
1423                }
1424
1425                if (unlikely(nr_sectors > q->max_hw_sectors)) {
1426                        printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1427                                bdevname(bio->bi_bdev, b),
1428                                bio_sectors(bio),
1429                                q->max_hw_sectors);
1430                        goto end_io;
1431                }
1432
1433                if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1434                        goto end_io;
1435
1436                if (should_fail_request(bio))
1437                        goto end_io;
1438
1439                /*
1440                 * If this device has partitions, remap block n
1441                 * of partition p to block n+start(p) of the disk.
1442                 */
1443                blk_partition_remap(bio);
1444
1445                if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1446                        goto end_io;
1447
1448                if (old_sector != -1)
1449                        trace_block_remap(q, bio, old_dev, bio->bi_sector,
1450                                            old_sector);
1451
1452                trace_block_bio_queue(q, bio);
1453
1454                old_sector = bio->bi_sector;
1455                old_dev = bio->bi_bdev->bd_dev;
1456
1457                if (bio_check_eod(bio, nr_sectors))
1458                        goto end_io;
1459
1460                if (bio_discard(bio) && !q->prepare_discard_fn) {
1461                        err = -EOPNOTSUPP;
1462                        goto end_io;
1463                }
1464                if (bio_barrier(bio) && bio_has_data(bio) &&
1465                    (q->next_ordered == QUEUE_ORDERED_NONE)) {
1466                        err = -EOPNOTSUPP;
1467                        goto end_io;
1468                }
1469
1470                ret = q->make_request_fn(q, bio);
1471        } while (ret);
1472
1473        return;
1474
1475end_io:
1476        bio_endio(bio, err);
1477}
1478
1479/*
1480 * We only want one ->make_request_fn to be active at a time,
1481 * else stack usage with stacked devices could be a problem.
1482 * So use current->bio_{list,tail} to keep a list of requests
1483 * submited by a make_request_fn function.
1484 * current->bio_tail is also used as a flag to say if
1485 * generic_make_request is currently active in this task or not.
1486 * If it is NULL, then no make_request is active.  If it is non-NULL,
1487 * then a make_request is active, and new requests should be added
1488 * at the tail
1489 */
1490void generic_make_request(struct bio *bio)
1491{
1492        if (current->bio_tail) {
1493                /* make_request is active */
1494                *(current->bio_tail) = bio;
1495                bio->bi_next = NULL;
1496                current->bio_tail = &bio->bi_next;
1497                return;
1498        }
1499        /* following loop may be a bit non-obvious, and so deserves some
1500         * explanation.
1501         * Before entering the loop, bio->bi_next is NULL (as all callers
1502         * ensure that) so we have a list with a single bio.
1503         * We pretend that we have just taken it off a longer list, so
1504         * we assign bio_list to the next (which is NULL) and bio_tail
1505         * to &bio_list, thus initialising the bio_list of new bios to be
1506         * added.  __generic_make_request may indeed add some more bios
1507         * through a recursive call to generic_make_request.  If it
1508         * did, we find a non-NULL value in bio_list and re-enter the loop
1509         * from the top.  In this case we really did just take the bio
1510         * of the top of the list (no pretending) and so fixup bio_list and
1511         * bio_tail or bi_next, and call into __generic_make_request again.
1512         *
1513         * The loop was structured like this to make only one call to
1514         * __generic_make_request (which is important as it is large and
1515         * inlined) and to keep the structure simple.
1516         */
1517        BUG_ON(bio->bi_next);
1518        do {
1519                current->bio_list = bio->bi_next;
1520                if (bio->bi_next == NULL)
1521                        current->bio_tail = &current->bio_list;
1522                else
1523                        bio->bi_next = NULL;
1524                __generic_make_request(bio);
1525                bio = current->bio_list;
1526        } while (bio);
1527        current->bio_tail = NULL; /* deactivate */
1528}
1529EXPORT_SYMBOL(generic_make_request);
1530
1531/**
1532 * submit_bio - submit a bio to the block device layer for I/O
1533 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1534 * @bio: The &struct bio which describes the I/O
1535 *
1536 * submit_bio() is very similar in purpose to generic_make_request(), and
1537 * uses that function to do most of the work. Both are fairly rough
1538 * interfaces; @bio must be presetup and ready for I/O.
1539 *
1540 */
1541void submit_bio(int rw, struct bio *bio)
1542{
1543        int count = bio_sectors(bio);
1544
1545        bio->bi_rw |= rw;
1546
1547        /*
1548         * If it's a regular read/write or a barrier with data attached,
1549         * go through the normal accounting stuff before submission.
1550         */
1551        if (bio_has_data(bio)) {
1552                if (rw & WRITE) {
1553                        count_vm_events(PGPGOUT, count);
1554                } else {
1555                        task_io_account_read(bio->bi_size);
1556                        count_vm_events(PGPGIN, count);
1557                }
1558
1559                if (unlikely(block_dump)) {
1560                        char b[BDEVNAME_SIZE];
1561                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1562                        current->comm, task_pid_nr(current),
1563                                (rw & WRITE) ? "WRITE" : "READ",
1564                                (unsigned long long)bio->bi_sector,
1565                                bdevname(bio->bi_bdev, b));
1566                }
1567        }
1568
1569        generic_make_request(bio);
1570}
1571EXPORT_SYMBOL(submit_bio);
1572
1573/**
1574 * blk_rq_check_limits - Helper function to check a request for the queue limit
1575 * @q:  the queue
1576 * @rq: the request being checked
1577 *
1578 * Description:
1579 *    @rq may have been made based on weaker limitations of upper-level queues
1580 *    in request stacking drivers, and it may violate the limitation of @q.
1581 *    Since the block layer and the underlying device driver trust @rq
1582 *    after it is inserted to @q, it should be checked against @q before
1583 *    the insertion using this generic function.
1584 *
1585 *    This function should also be useful for request stacking drivers
1586 *    in some cases below, so export this fuction.
1587 *    Request stacking drivers like request-based dm may change the queue
1588 *    limits while requests are in the queue (e.g. dm's table swapping).
1589 *    Such request stacking drivers should check those requests agaist
1590 *    the new queue limits again when they dispatch those requests,
1591 *    although such checkings are also done against the old queue limits
1592 *    when submitting requests.
1593 */
1594int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1595{
1596        if (rq->nr_sectors > q->max_sectors ||
1597            rq->data_len > q->max_hw_sectors << 9) {
1598                printk(KERN_ERR "%s: over max size limit.\n", __func__);
1599                return -EIO;
1600        }
1601
1602        /*
1603         * queue's settings related to segment counting like q->bounce_pfn
1604         * may differ from that of other stacking queues.
1605         * Recalculate it to check the request correctly on this queue's
1606         * limitation.
1607         */
1608        blk_recalc_rq_segments(rq);
1609        if (rq->nr_phys_segments > q->max_phys_segments ||
1610            rq->nr_phys_segments > q->max_hw_segments) {
1611                printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1612                return -EIO;
1613        }
1614
1615        return 0;
1616}
1617EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1618
1619/**
1620 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1621 * @q:  the queue to submit the request
1622 * @rq: the request being queued
1623 */
1624int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1625{
1626        unsigned long flags;
1627
1628        if (blk_rq_check_limits(q, rq))
1629                return -EIO;
1630
1631#ifdef CONFIG_FAIL_MAKE_REQUEST
1632        if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1633            should_fail(&fail_make_request, blk_rq_bytes(rq)))
1634                return -EIO;
1635#endif
1636
1637        spin_lock_irqsave(q->queue_lock, flags);
1638
1639        /*
1640         * Submitting request must be dequeued before calling this function
1641         * because it will be linked to another request_queue
1642         */
1643        BUG_ON(blk_queued_rq(rq));
1644
1645        drive_stat_acct(rq, 1);
1646        __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1647
1648        spin_unlock_irqrestore(q->queue_lock, flags);
1649
1650        return 0;
1651}
1652EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1653
1654/**
1655 * blkdev_dequeue_request - dequeue request and start timeout timer
1656 * @req: request to dequeue
1657 *
1658 * Dequeue @req and start timeout timer on it.  This hands off the
1659 * request to the driver.
1660 *
1661 * Block internal functions which don't want to start timer should
1662 * call elv_dequeue_request().
1663 */
1664void blkdev_dequeue_request(struct request *req)
1665{
1666        elv_dequeue_request(req->q, req);
1667
1668        /*
1669         * We are now handing the request to the hardware, add the
1670         * timeout handler.
1671         */
1672        blk_add_timer(req);
1673}
1674EXPORT_SYMBOL(blkdev_dequeue_request);
1675
1676static void blk_account_io_completion(struct request *req, unsigned int bytes)
1677{
1678        if (!blk_do_io_stat(req))
1679                return;
1680
1681        if (blk_fs_request(req)) {
1682                const int rw = rq_data_dir(req);
1683                struct hd_struct *part;
1684                int cpu;
1685
1686                cpu = part_stat_lock();
1687                part = disk_map_sector_rcu(req->rq_disk, req->sector);
1688                part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1689                part_stat_unlock();
1690        }
1691}
1692
1693static void blk_account_io_done(struct request *req)
1694{
1695        if (!blk_do_io_stat(req))
1696                return;
1697
1698        /*
1699         * Account IO completion.  bar_rq isn't accounted as a normal
1700         * IO on queueing nor completion.  Accounting the containing
1701         * request is enough.
1702         */
1703        if (blk_fs_request(req) && req != &req->q->bar_rq) {
1704                unsigned long duration = jiffies - req->start_time;
1705                const int rw = rq_data_dir(req);
1706                struct hd_struct *part;
1707                int cpu;
1708
1709                cpu = part_stat_lock();
1710                part = disk_map_sector_rcu(req->rq_disk, req->sector);
1711
1712                part_stat_inc(cpu, part, ios[rw]);
1713                part_stat_add(cpu, part, ticks[rw], duration);
1714                part_round_stats(cpu, part);
1715                part_dec_in_flight(part);
1716
1717                part_stat_unlock();
1718        }
1719}
1720
1721/**
1722 * __end_that_request_first - end I/O on a request
1723 * @req:      the request being processed
1724 * @error:    %0 for success, < %0 for error
1725 * @nr_bytes: number of bytes to complete
1726 *
1727 * Description:
1728 *     Ends I/O on a number of bytes attached to @req, and sets it up
1729 *     for the next range of segments (if any) in the cluster.
1730 *
1731 * Return:
1732 *     %0 - we are done with this request, call end_that_request_last()
1733 *     %1 - still buffers pending for this request
1734 **/
1735static int __end_that_request_first(struct request *req, int error,
1736                                    int nr_bytes)
1737{
1738        int total_bytes, bio_nbytes, next_idx = 0;
1739        struct bio *bio;
1740
1741        trace_block_rq_complete(req->q, req);
1742
1743        /*
1744         * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1745         * sense key with us all the way through
1746         */
1747        if (!blk_pc_request(req))
1748                req->errors = 0;
1749
1750        if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1751                printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1752                                req->rq_disk ? req->rq_disk->disk_name : "?",
1753                                (unsigned long long)req->sector);
1754        }
1755
1756        blk_account_io_completion(req, nr_bytes);
1757
1758        total_bytes = bio_nbytes = 0;
1759        while ((bio = req->bio) != NULL) {
1760                int nbytes;
1761
1762                if (nr_bytes >= bio->bi_size) {
1763                        req->bio = bio->bi_next;
1764                        nbytes = bio->bi_size;
1765                        req_bio_endio(req, bio, nbytes, error);
1766                        next_idx = 0;
1767                        bio_nbytes = 0;
1768                } else {
1769                        int idx = bio->bi_idx + next_idx;
1770
1771                        if (unlikely(idx >= bio->bi_vcnt)) {
1772                                blk_dump_rq_flags(req, "__end_that");
1773                                printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1774                                       __func__, idx, bio->bi_vcnt);
1775                                break;
1776                        }
1777
1778                        nbytes = bio_iovec_idx(bio, idx)->bv_len;
1779                        BIO_BUG_ON(nbytes > bio->bi_size);
1780
1781                        /*
1782                         * not a complete bvec done
1783                         */
1784                        if (unlikely(nbytes > nr_bytes)) {
1785                                bio_nbytes += nr_bytes;
1786                                total_bytes += nr_bytes;
1787                                break;
1788                        }
1789
1790                        /*
1791                         * advance to the next vector
1792                         */
1793                        next_idx++;
1794                        bio_nbytes += nbytes;
1795                }
1796
1797                total_bytes += nbytes;
1798                nr_bytes -= nbytes;
1799
1800                bio = req->bio;
1801                if (bio) {
1802                        /*
1803                         * end more in this run, or just return 'not-done'
1804                         */
1805                        if (unlikely(nr_bytes <= 0))
1806                                break;
1807                }
1808        }
1809
1810        /*
1811         * completely done
1812         */
1813        if (!req->bio)
1814                return 0;
1815
1816        /*
1817         * if the request wasn't completed, update state
1818         */
1819        if (bio_nbytes) {
1820                req_bio_endio(req, bio, bio_nbytes, error);
1821                bio->bi_idx += next_idx;
1822                bio_iovec(bio)->bv_offset += nr_bytes;
1823                bio_iovec(bio)->bv_len -= nr_bytes;
1824        }
1825
1826        blk_recalc_rq_sectors(req, total_bytes >> 9);
1827        blk_recalc_rq_segments(req);
1828        return 1;
1829}
1830
1831/*
1832 * queue lock must be held
1833 */
1834static void end_that_request_last(struct request *req, int error)
1835{
1836        if (blk_rq_tagged(req))
1837                blk_queue_end_tag(req->q, req);
1838
1839        if (blk_queued_rq(req))
1840                elv_dequeue_request(req->q, req);
1841
1842        if (unlikely(laptop_mode) && blk_fs_request(req))
1843                laptop_io_completion();
1844
1845        blk_delete_timer(req);
1846
1847        blk_account_io_done(req);
1848
1849        if (req->end_io)
1850                req->end_io(req, error);
1851        else {
1852                if (blk_bidi_rq(req))
1853                        __blk_put_request(req->next_rq->q, req->next_rq);
1854
1855                __blk_put_request(req->q, req);
1856        }
1857}
1858
1859/**
1860 * blk_rq_bytes - Returns bytes left to complete in the entire request
1861 * @rq: the request being processed
1862 **/
1863unsigned int blk_rq_bytes(struct request *rq)
1864{
1865        if (blk_fs_request(rq))
1866                return rq->hard_nr_sectors << 9;
1867
1868        return rq->data_len;
1869}
1870EXPORT_SYMBOL_GPL(blk_rq_bytes);
1871
1872/**
1873 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1874 * @rq: the request being processed
1875 **/
1876unsigned int blk_rq_cur_bytes(struct request *rq)
1877{
1878        if (blk_fs_request(rq))
1879                return rq->current_nr_sectors << 9;
1880
1881        if (rq->bio)
1882                return rq->bio->bi_size;
1883
1884        return rq->data_len;
1885}
1886EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
1887
1888/**
1889 * end_request - end I/O on the current segment of the request
1890 * @req:        the request being processed
1891 * @uptodate:   error value or %0/%1 uptodate flag
1892 *
1893 * Description:
1894 *     Ends I/O on the current segment of a request. If that is the only
1895 *     remaining segment, the request is also completed and freed.
1896 *
1897 *     This is a remnant of how older block drivers handled I/O completions.
1898 *     Modern drivers typically end I/O on the full request in one go, unless
1899 *     they have a residual value to account for. For that case this function
1900 *     isn't really useful, unless the residual just happens to be the
1901 *     full current segment. In other words, don't use this function in new
1902 *     code. Use blk_end_request() or __blk_end_request() to end a request.
1903 **/
1904void end_request(struct request *req, int uptodate)
1905{
1906        int error = 0;
1907
1908        if (uptodate <= 0)
1909                error = uptodate ? uptodate : -EIO;
1910
1911        __blk_end_request(req, error, req->hard_cur_sectors << 9);
1912}
1913EXPORT_SYMBOL(end_request);
1914
1915static int end_that_request_data(struct request *rq, int error,
1916                                 unsigned int nr_bytes, unsigned int bidi_bytes)
1917{
1918        if (rq->bio) {
1919                if (__end_that_request_first(rq, error, nr_bytes))
1920                        return 1;
1921
1922                /* Bidi request must be completed as a whole */
1923                if (blk_bidi_rq(rq) &&
1924                    __end_that_request_first(rq->next_rq, error, bidi_bytes))
1925                        return 1;
1926        }
1927
1928        return 0;
1929}
1930
1931/**
1932 * blk_end_io - Generic end_io function to complete a request.
1933 * @rq:           the request being processed
1934 * @error:        %0 for success, < %0 for error
1935 * @nr_bytes:     number of bytes to complete @rq
1936 * @bidi_bytes:   number of bytes to complete @rq->next_rq
1937 * @drv_callback: function called between completion of bios in the request
1938 *                and completion of the request.
1939 *                If the callback returns non %0, this helper returns without
1940 *                completion of the request.
1941 *
1942 * Description:
1943 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1944 *     If @rq has leftover, sets it up for the next range of segments.
1945 *
1946 * Return:
1947 *     %0 - we are done with this request
1948 *     %1 - this request is not freed yet, it still has pending buffers.
1949 **/
1950static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1951                      unsigned int bidi_bytes,
1952                      int (drv_callback)(struct request *))
1953{
1954        struct request_queue *q = rq->q;
1955        unsigned long flags = 0UL;
1956
1957        if (end_that_request_data(rq, error, nr_bytes, bidi_bytes))
1958                return 1;
1959
1960        /* Special feature for tricky drivers */
1961        if (drv_callback && drv_callback(rq))
1962                return 1;
1963
1964        add_disk_randomness(rq->rq_disk);
1965
1966        spin_lock_irqsave(q->queue_lock, flags);
1967        end_that_request_last(rq, error);
1968        spin_unlock_irqrestore(q->queue_lock, flags);
1969
1970        return 0;
1971}
1972
1973/**
1974 * blk_end_request - Helper function for drivers to complete the request.
1975 * @rq:       the request being processed
1976 * @error:    %0 for success, < %0 for error
1977 * @nr_bytes: number of bytes to complete
1978 *
1979 * Description:
1980 *     Ends I/O on a number of bytes attached to @rq.
1981 *     If @rq has leftover, sets it up for the next range of segments.
1982 *
1983 * Return:
1984 *     %0 - we are done with this request
1985 *     %1 - still buffers pending for this request
1986 **/
1987int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1988{
1989        return blk_end_io(rq, error, nr_bytes, 0, NULL);
1990}
1991EXPORT_SYMBOL_GPL(blk_end_request);
1992
1993/**
1994 * __blk_end_request - Helper function for drivers to complete the request.
1995 * @rq:       the request being processed
1996 * @error:    %0 for success, < %0 for error
1997 * @nr_bytes: number of bytes to complete
1998 *
1999 * Description:
2000 *     Must be called with queue lock held unlike blk_end_request().
2001 *
2002 * Return:
2003 *     %0 - we are done with this request
2004 *     %1 - still buffers pending for this request
2005 **/
2006int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2007{
2008        if (rq->bio && __end_that_request_first(rq, error, nr_bytes))
2009                return 1;
2010
2011        add_disk_randomness(rq->rq_disk);
2012
2013        end_that_request_last(rq, error);
2014
2015        return 0;
2016}
2017EXPORT_SYMBOL_GPL(__blk_end_request);
2018
2019/**
2020 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
2021 * @rq:         the bidi request being processed
2022 * @error:      %0 for success, < %0 for error
2023 * @nr_bytes:   number of bytes to complete @rq
2024 * @bidi_bytes: number of bytes to complete @rq->next_rq
2025 *
2026 * Description:
2027 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2028 *
2029 * Return:
2030 *     %0 - we are done with this request
2031 *     %1 - still buffers pending for this request
2032 **/
2033int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
2034                         unsigned int bidi_bytes)
2035{
2036        return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
2037}
2038EXPORT_SYMBOL_GPL(blk_end_bidi_request);
2039
2040/**
2041 * blk_update_request - Special helper function for request stacking drivers
2042 * @rq:           the request being processed
2043 * @error:        %0 for success, < %0 for error
2044 * @nr_bytes:     number of bytes to complete @rq
2045 *
2046 * Description:
2047 *     Ends I/O on a number of bytes attached to @rq, but doesn't complete
2048 *     the request structure even if @rq doesn't have leftover.
2049 *     If @rq has leftover, sets it up for the next range of segments.
2050 *
2051 *     This special helper function is only for request stacking drivers
2052 *     (e.g. request-based dm) so that they can handle partial completion.
2053 *     Actual device drivers should use blk_end_request instead.
2054 */
2055void blk_update_request(struct request *rq, int error, unsigned int nr_bytes)
2056{
2057        if (!end_that_request_data(rq, error, nr_bytes, 0)) {
2058                /*
2059                 * These members are not updated in end_that_request_data()
2060                 * when all bios are completed.
2061                 * Update them so that the request stacking driver can find
2062                 * how many bytes remain in the request later.
2063                 */
2064                rq->nr_sectors = rq->hard_nr_sectors = 0;
2065                rq->current_nr_sectors = rq->hard_cur_sectors = 0;
2066        }
2067}
2068EXPORT_SYMBOL_GPL(blk_update_request);
2069
2070/**
2071 * blk_end_request_callback - Special helper function for tricky drivers
2072 * @rq:           the request being processed
2073 * @error:        %0 for success, < %0 for error
2074 * @nr_bytes:     number of bytes to complete
2075 * @drv_callback: function called between completion of bios in the request
2076 *                and completion of the request.
2077 *                If the callback returns non %0, this helper returns without
2078 *                completion of the request.
2079 *
2080 * Description:
2081 *     Ends I/O on a number of bytes attached to @rq.
2082 *     If @rq has leftover, sets it up for the next range of segments.
2083 *
2084 *     This special helper function is used only for existing tricky drivers.
2085 *     (e.g. cdrom_newpc_intr() of ide-cd)
2086 *     This interface will be removed when such drivers are rewritten.
2087 *     Don't use this interface in other places anymore.
2088 *
2089 * Return:
2090 *     %0 - we are done with this request
2091 *     %1 - this request is not freed yet.
2092 *          this request still has pending buffers or
2093 *          the driver doesn't want to finish this request yet.
2094 **/
2095int blk_end_request_callback(struct request *rq, int error,
2096                             unsigned int nr_bytes,
2097                             int (drv_callback)(struct request *))
2098{
2099        return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
2100}
2101EXPORT_SYMBOL_GPL(blk_end_request_callback);
2102
2103void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2104                     struct bio *bio)
2105{
2106        /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2107           we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2108        rq->cmd_flags |= (bio->bi_rw & 3);
2109
2110        if (bio_has_data(bio)) {
2111                rq->nr_phys_segments = bio_phys_segments(q, bio);
2112                rq->buffer = bio_data(bio);
2113        }
2114        rq->current_nr_sectors = bio_cur_sectors(bio);
2115        rq->hard_cur_sectors = rq->current_nr_sectors;
2116        rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2117        rq->data_len = bio->bi_size;
2118
2119        rq->bio = rq->biotail = bio;
2120
2121        if (bio->bi_bdev)
2122                rq->rq_disk = bio->bi_bdev->bd_disk;
2123}
2124
2125/**
2126 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2127 * @q : the queue of the device being checked
2128 *
2129 * Description:
2130 *    Check if underlying low-level drivers of a device are busy.
2131 *    If the drivers want to export their busy state, they must set own
2132 *    exporting function using blk_queue_lld_busy() first.
2133 *
2134 *    Basically, this function is used only by request stacking drivers
2135 *    to stop dispatching requests to underlying devices when underlying
2136 *    devices are busy.  This behavior helps more I/O merging on the queue
2137 *    of the request stacking driver and prevents I/O throughput regression
2138 *    on burst I/O load.
2139 *
2140 * Return:
2141 *    0 - Not busy (The request stacking driver should dispatch request)
2142 *    1 - Busy (The request stacking driver should stop dispatching request)
2143 */
2144int blk_lld_busy(struct request_queue *q)
2145{
2146        if (q->lld_busy_fn)
2147                return q->lld_busy_fn(q);
2148
2149        return 0;
2150}
2151EXPORT_SYMBOL_GPL(blk_lld_busy);
2152
2153int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2154{
2155        return queue_work(kblockd_workqueue, work);
2156}
2157EXPORT_SYMBOL(kblockd_schedule_work);
2158
2159int __init blk_dev_init(void)
2160{
2161        kblockd_workqueue = create_workqueue("kblockd");
2162        if (!kblockd_workqueue)
2163                panic("Failed to create kblockd\n");
2164
2165        request_cachep = kmem_cache_create("blkdev_requests",
2166                        sizeof(struct request), 0, SLAB_PANIC, NULL);
2167
2168        blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2169                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2170
2171        return 0;
2172}
2173
2174