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
  32#include "blk.h"
  33
  34static int __make_request(struct request_queue *q, struct bio *bio);
  35
  36/*
  37 * For the allocated request tables
  38 */
  39static struct kmem_cache *request_cachep;
  40
  41/*
  42 * For queue allocation
  43 */
  44struct kmem_cache *blk_requestq_cachep;
  45
  46/*
  47 * Controlling structure to kblockd
  48 */
  49static struct workqueue_struct *kblockd_workqueue;
  50
  51static void drive_stat_acct(struct request *rq, int new_io)
  52{
  53        struct hd_struct *part;
  54        int rw = rq_data_dir(rq);
  55        int cpu;
  56
  57        if (!blk_fs_request(rq) || !rq->rq_disk)
  58                return;
  59
  60        cpu = part_stat_lock();
  61        part = disk_map_sector_rcu(rq->rq_disk, rq->sector);
  62
  63        if (!new_io)
  64                part_stat_inc(cpu, part, merges[rw]);
  65        else {
  66                part_round_stats(cpu, part);
  67                part_inc_in_flight(part);
  68        }
  69
  70        part_stat_unlock();
  71}
  72
  73void blk_queue_congestion_threshold(struct request_queue *q)
  74{
  75        int nr;
  76
  77        nr = q->nr_requests - (q->nr_requests / 8) + 1;
  78        if (nr > q->nr_requests)
  79                nr = q->nr_requests;
  80        q->nr_congestion_on = nr;
  81
  82        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
  83        if (nr < 1)
  84                nr = 1;
  85        q->nr_congestion_off = nr;
  86}
  87
  88/**
  89 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
  90 * @bdev:       device
  91 *
  92 * Locates the passed device's request queue and returns the address of its
  93 * backing_dev_info
  94 *
  95 * Will return NULL if the request queue cannot be located.
  96 */
  97struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
  98{
  99        struct backing_dev_info *ret = NULL;
 100        struct request_queue *q = bdev_get_queue(bdev);
 101
 102        if (q)
 103                ret = &q->backing_dev_info;
 104        return ret;
 105}
 106EXPORT_SYMBOL(blk_get_backing_dev_info);
 107
 108void blk_rq_init(struct request_queue *q, struct request *rq)
 109{
 110        memset(rq, 0, sizeof(*rq));
 111
 112        INIT_LIST_HEAD(&rq->queuelist);
 113        INIT_LIST_HEAD(&rq->timeout_list);
 114        rq->cpu = -1;
 115        rq->q = q;
 116        rq->sector = rq->hard_sector = (sector_t) -1;
 117        INIT_HLIST_NODE(&rq->hash);
 118        RB_CLEAR_NODE(&rq->rb_node);
 119        rq->cmd = rq->__cmd;
 120        rq->tag = -1;
 121        rq->ref_count = 1;
 122}
 123EXPORT_SYMBOL(blk_rq_init);
 124
 125static void req_bio_endio(struct request *rq, struct bio *bio,
 126                          unsigned int nbytes, int error)
 127{
 128        struct request_queue *q = rq->q;
 129
 130        if (&q->bar_rq != rq) {
 131                if (error)
 132                        clear_bit(BIO_UPTODATE, &bio->bi_flags);
 133                else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 134                        error = -EIO;
 135
 136                if (unlikely(nbytes > bio->bi_size)) {
 137                        printk(KERN_ERR "%s: want %u bytes done, %u left\n",
 138                               __func__, nbytes, bio->bi_size);
 139                        nbytes = bio->bi_size;
 140                }
 141
 142                bio->bi_size -= nbytes;
 143                bio->bi_sector += (nbytes >> 9);
 144
 145                if (bio_integrity(bio))
 146                        bio_integrity_advance(bio, nbytes);
 147
 148                if (bio->bi_size == 0)
 149                        bio_endio(bio, error);
 150        } else {
 151
 152                /*
 153                 * Okay, this is the barrier request in progress, just
 154                 * record the error;
 155                 */
 156                if (error && !q->orderr)
 157                        q->orderr = error;
 158        }
 159}
 160
 161void blk_dump_rq_flags(struct request *rq, char *msg)
 162{
 163        int bit;
 164
 165        printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
 166                rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
 167                rq->cmd_flags);
 168
 169        printk(KERN_INFO "  sector %llu, nr/cnr %lu/%u\n",
 170                                                (unsigned long long)rq->sector,
 171                                                rq->nr_sectors,
 172                                                rq->current_nr_sectors);
 173        printk(KERN_INFO "  bio %p, biotail %p, buffer %p, data %p, len %u\n",
 174                                                rq->bio, rq->biotail,
 175                                                rq->buffer, rq->data,
 176                                                rq->data_len);
 177
 178        if (blk_pc_request(rq)) {
 179                printk(KERN_INFO "  cdb: ");
 180                for (bit = 0; bit < BLK_MAX_CDB; bit++)
 181                        printk("%02x ", rq->cmd[bit]);
 182                printk("\n");
 183        }
 184}
 185EXPORT_SYMBOL(blk_dump_rq_flags);
 186
 187/*
 188 * "plug" the device if there are no outstanding requests: this will
 189 * force the transfer to start only after we have put all the requests
 190 * on the list.
 191 *
 192 * This is called with interrupts off and no requests on the queue and
 193 * with the queue lock held.
 194 */
 195void blk_plug_device(struct request_queue *q)
 196{
 197        WARN_ON(!irqs_disabled());
 198
 199        /*
 200         * don't plug a stopped queue, it must be paired with blk_start_queue()
 201         * which will restart the queueing
 202         */
 203        if (blk_queue_stopped(q))
 204                return;
 205
 206        if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
 207                mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
 208                blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
 209        }
 210}
 211EXPORT_SYMBOL(blk_plug_device);
 212
 213/**
 214 * blk_plug_device_unlocked - plug a device without queue lock held
 215 * @q:    The &struct request_queue to plug
 216 *
 217 * Description:
 218 *   Like @blk_plug_device(), but grabs the queue lock and disables
 219 *   interrupts.
 220 **/
 221void blk_plug_device_unlocked(struct request_queue *q)
 222{
 223        unsigned long flags;
 224
 225        spin_lock_irqsave(q->queue_lock, flags);
 226        blk_plug_device(q);
 227        spin_unlock_irqrestore(q->queue_lock, flags);
 228}
 229EXPORT_SYMBOL(blk_plug_device_unlocked);
 230
 231/*
 232 * remove the queue from the plugged list, if present. called with
 233 * queue lock held and interrupts disabled.
 234 */
 235int blk_remove_plug(struct request_queue *q)
 236{
 237        WARN_ON(!irqs_disabled());
 238
 239        if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
 240                return 0;
 241
 242        del_timer(&q->unplug_timer);
 243        return 1;
 244}
 245EXPORT_SYMBOL(blk_remove_plug);
 246
 247/*
 248 * remove the plug and let it rip..
 249 */
 250void __generic_unplug_device(struct request_queue *q)
 251{
 252        if (unlikely(blk_queue_stopped(q)))
 253                return;
 254
 255        if (!blk_remove_plug(q))
 256                return;
 257
 258        q->request_fn(q);
 259}
 260
 261/**
 262 * generic_unplug_device - fire a request queue
 263 * @q:    The &struct request_queue in question
 264 *
 265 * Description:
 266 *   Linux uses plugging to build bigger requests queues before letting
 267 *   the device have at them. If a queue is plugged, the I/O scheduler
 268 *   is still adding and merging requests on the queue. Once the queue
 269 *   gets unplugged, the request_fn defined for the queue is invoked and
 270 *   transfers started.
 271 **/
 272void generic_unplug_device(struct request_queue *q)
 273{
 274        if (blk_queue_plugged(q)) {
 275                spin_lock_irq(q->queue_lock);
 276                __generic_unplug_device(q);
 277                spin_unlock_irq(q->queue_lock);
 278        }
 279}
 280EXPORT_SYMBOL(generic_unplug_device);
 281
 282static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
 283                                   struct page *page)
 284{
 285        struct request_queue *q = bdi->unplug_io_data;
 286
 287        blk_unplug(q);
 288}
 289
 290void blk_unplug_work(struct work_struct *work)
 291{
 292        struct request_queue *q =
 293                container_of(work, struct request_queue, unplug_work);
 294
 295        blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
 296                                q->rq.count[READ] + q->rq.count[WRITE]);
 297
 298        q->unplug_fn(q);
 299}
 300
 301void blk_unplug_timeout(unsigned long data)
 302{
 303        struct request_queue *q = (struct request_queue *)data;
 304
 305        blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
 306                                q->rq.count[READ] + q->rq.count[WRITE]);
 307
 308        kblockd_schedule_work(q, &q->unplug_work);
 309}
 310
 311void blk_unplug(struct request_queue *q)
 312{
 313        /*
 314         * devices don't necessarily have an ->unplug_fn defined
 315         */
 316        if (q->unplug_fn) {
 317                blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
 318                                        q->rq.count[READ] + q->rq.count[WRITE]);
 319
 320                q->unplug_fn(q);
 321        }
 322}
 323EXPORT_SYMBOL(blk_unplug);
 324
 325static void blk_invoke_request_fn(struct request_queue *q)
 326{
 327        if (unlikely(blk_queue_stopped(q)))
 328                return;
 329
 330        /*
 331         * one level of recursion is ok and is much faster than kicking
 332         * the unplug handling
 333         */
 334        if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
 335                q->request_fn(q);
 336                queue_flag_clear(QUEUE_FLAG_REENTER, q);
 337        } else {
 338                queue_flag_set(QUEUE_FLAG_PLUGGED, q);
 339                kblockd_schedule_work(q, &q->unplug_work);
 340        }
 341}
 342
 343/**
 344 * blk_start_queue - restart a previously stopped queue
 345 * @q:    The &struct request_queue in question
 346 *
 347 * Description:
 348 *   blk_start_queue() will clear the stop flag on the queue, and call
 349 *   the request_fn for the queue if it was in a stopped state when
 350 *   entered. Also see blk_stop_queue(). Queue lock must be held.
 351 **/
 352void blk_start_queue(struct request_queue *q)
 353{
 354        WARN_ON(!irqs_disabled());
 355
 356        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 357        blk_invoke_request_fn(q);
 358}
 359EXPORT_SYMBOL(blk_start_queue);
 360
 361/**
 362 * blk_stop_queue - stop a queue
 363 * @q:    The &struct request_queue in question
 364 *
 365 * Description:
 366 *   The Linux block layer assumes that a block driver will consume all
 367 *   entries on the request queue when the request_fn strategy is called.
 368 *   Often this will not happen, because of hardware limitations (queue
 369 *   depth settings). If a device driver gets a 'queue full' response,
 370 *   or if it simply chooses not to queue more I/O at one point, it can
 371 *   call this function to prevent the request_fn from being called until
 372 *   the driver has signalled it's ready to go again. This happens by calling
 373 *   blk_start_queue() to restart queue operations. Queue lock must be held.
 374 **/
 375void blk_stop_queue(struct request_queue *q)
 376{
 377        blk_remove_plug(q);
 378        queue_flag_set(QUEUE_FLAG_STOPPED, q);
 379}
 380EXPORT_SYMBOL(blk_stop_queue);
 381
 382/**
 383 * blk_sync_queue - cancel any pending callbacks on a queue
 384 * @q: the queue
 385 *
 386 * Description:
 387 *     The block layer may perform asynchronous callback activity
 388 *     on a queue, such as calling the unplug function after a timeout.
 389 *     A block device may call blk_sync_queue to ensure that any
 390 *     such activity is cancelled, thus allowing it to release resources
 391 *     that the callbacks might use. The caller must already have made sure
 392 *     that its ->make_request_fn will not re-add plugging prior to calling
 393 *     this function.
 394 *
 395 */
 396void blk_sync_queue(struct request_queue *q)
 397{
 398        del_timer_sync(&q->unplug_timer);
 399        kblockd_flush_work(&q->unplug_work);
 400}
 401EXPORT_SYMBOL(blk_sync_queue);
 402
 403/**
 404 * __blk_run_queue - run a single device queue
 405 * @q:  The queue to run
 406 *
 407 * Description:
 408 *    See @blk_run_queue. This variant must be called with the queue lock
 409 *    held and interrupts disabled.
 410 *
 411 */
 412void __blk_run_queue(struct request_queue *q)
 413{
 414        blk_remove_plug(q);
 415
 416        /*
 417         * Only recurse once to avoid overrunning the stack, let the unplug
 418         * handling reinvoke the handler shortly if we already got there.
 419         */
 420        if (!elv_queue_empty(q))
 421                blk_invoke_request_fn(q);
 422}
 423EXPORT_SYMBOL(__blk_run_queue);
 424
 425/**
 426 * blk_run_queue - run a single device queue
 427 * @q: The queue to run
 428 *
 429 * Description:
 430 *    Invoke request handling on this queue, if it has pending work to do.
 431 *    May be used to restart queueing when a request has completed. Also
 432 *    See @blk_start_queueing.
 433 *
 434 */
 435void blk_run_queue(struct request_queue *q)
 436{
 437        unsigned long flags;
 438
 439        spin_lock_irqsave(q->queue_lock, flags);
 440        __blk_run_queue(q);
 441        spin_unlock_irqrestore(q->queue_lock, flags);
 442}
 443EXPORT_SYMBOL(blk_run_queue);
 444
 445void blk_put_queue(struct request_queue *q)
 446{
 447        kobject_put(&q->kobj);
 448}
 449
 450void blk_cleanup_queue(struct request_queue *q)
 451{
 452        /*
 453         * We know we have process context here, so we can be a little
 454         * cautious and ensure that pending block actions on this device
 455         * are done before moving on. Going into this function, we should
 456         * not have processes doing IO to this device.
 457         */
 458        blk_sync_queue(q);
 459
 460        mutex_lock(&q->sysfs_lock);
 461        queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
 462        mutex_unlock(&q->sysfs_lock);
 463
 464        if (q->elevator)
 465                elevator_exit(q->elevator);
 466
 467        blk_put_queue(q);
 468}
 469EXPORT_SYMBOL(blk_cleanup_queue);
 470
 471static int blk_init_free_list(struct request_queue *q)
 472{
 473        struct request_list *rl = &q->rq;
 474
 475        rl->count[READ] = rl->count[WRITE] = 0;
 476        rl->starved[READ] = rl->starved[WRITE] = 0;
 477        rl->elvpriv = 0;
 478        init_waitqueue_head(&rl->wait[READ]);
 479        init_waitqueue_head(&rl->wait[WRITE]);
 480
 481        rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
 482                                mempool_free_slab, request_cachep, q->node);
 483
 484        if (!rl->rq_pool)
 485                return -ENOMEM;
 486
 487        return 0;
 488}
 489
 490struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 491{
 492        return blk_alloc_queue_node(gfp_mask, -1);
 493}
 494EXPORT_SYMBOL(blk_alloc_queue);
 495
 496struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 497{
 498        struct request_queue *q;
 499        int err;
 500
 501        q = kmem_cache_alloc_node(blk_requestq_cachep,
 502                                gfp_mask | __GFP_ZERO, node_id);
 503        if (!q)
 504                return NULL;
 505
 506        q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
 507        q->backing_dev_info.unplug_io_data = q;
 508        err = bdi_init(&q->backing_dev_info);
 509        if (err) {
 510                kmem_cache_free(blk_requestq_cachep, q);
 511                return NULL;
 512        }
 513
 514        init_timer(&q->unplug_timer);
 515        setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
 516        INIT_LIST_HEAD(&q->timeout_list);
 517        INIT_WORK(&q->unplug_work, blk_unplug_work);
 518
 519        kobject_init(&q->kobj, &blk_queue_ktype);
 520
 521        mutex_init(&q->sysfs_lock);
 522        spin_lock_init(&q->__queue_lock);
 523
 524        return q;
 525}
 526EXPORT_SYMBOL(blk_alloc_queue_node);
 527
 528/**
 529 * blk_init_queue  - prepare a request queue for use with a block device
 530 * @rfn:  The function to be called to process requests that have been
 531 *        placed on the queue.
 532 * @lock: Request queue spin lock
 533 *
 534 * Description:
 535 *    If a block device wishes to use the standard request handling procedures,
 536 *    which sorts requests and coalesces adjacent requests, then it must
 537 *    call blk_init_queue().  The function @rfn will be called when there
 538 *    are requests on the queue that need to be processed.  If the device
 539 *    supports plugging, then @rfn may not be called immediately when requests
 540 *    are available on the queue, but may be called at some time later instead.
 541 *    Plugged queues are generally unplugged when a buffer belonging to one
 542 *    of the requests on the queue is needed, or due to memory pressure.
 543 *
 544 *    @rfn is not required, or even expected, to remove all requests off the
 545 *    queue, but only as many as it can handle at a time.  If it does leave
 546 *    requests on the queue, it is responsible for arranging that the requests
 547 *    get dealt with eventually.
 548 *
 549 *    The queue spin lock must be held while manipulating the requests on the
 550 *    request queue; this lock will be taken also from interrupt context, so irq
 551 *    disabling is needed for it.
 552 *
 553 *    Function returns a pointer to the initialized request queue, or %NULL if
 554 *    it didn't succeed.
 555 *
 556 * Note:
 557 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
 558 *    when the block device is deactivated (such as at module unload).
 559 **/
 560
 561struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
 562{
 563        return blk_init_queue_node(rfn, lock, -1);
 564}
 565EXPORT_SYMBOL(blk_init_queue);
 566
 567struct request_queue *
 568blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
 569{
 570        struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
 571
 572        if (!q)
 573                return NULL;
 574
 575        q->node = node_id;
 576        if (blk_init_free_list(q)) {
 577                kmem_cache_free(blk_requestq_cachep, q);
 578                return NULL;
 579        }
 580
 581        /*
 582         * if caller didn't supply a lock, they get per-queue locking with
 583         * our embedded lock
 584         */
 585        if (!lock)
 586                lock = &q->__queue_lock;
 587
 588        q->request_fn           = rfn;
 589        q->prep_rq_fn           = NULL;
 590        q->unplug_fn            = generic_unplug_device;
 591        q->queue_flags          = (1 << QUEUE_FLAG_CLUSTER |
 592                                   1 << QUEUE_FLAG_STACKABLE);
 593        q->queue_lock           = lock;
 594
 595        blk_queue_segment_boundary(q, BLK_SEG_BOUNDARY_MASK);
 596
 597        blk_queue_make_request(q, __make_request);
 598        blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
 599
 600        blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
 601        blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
 602
 603        q->sg_reserved_size = INT_MAX;
 604
 605        blk_set_cmd_filter_defaults(&q->cmd_filter);
 606
 607        /*
 608         * all done
 609         */
 610        if (!elevator_init(q, NULL)) {
 611                blk_queue_congestion_threshold(q);
 612                return q;
 613        }
 614
 615        blk_put_queue(q);
 616        return NULL;
 617}
 618EXPORT_SYMBOL(blk_init_queue_node);
 619
 620int blk_get_queue(struct request_queue *q)
 621{
 622        if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
 623                kobject_get(&q->kobj);
 624                return 0;
 625        }
 626
 627        return 1;
 628}
 629
 630static inline void blk_free_request(struct request_queue *q, struct request *rq)
 631{
 632        if (rq->cmd_flags & REQ_ELVPRIV)
 633                elv_put_request(q, rq);
 634        mempool_free(rq, q->rq.rq_pool);
 635}
 636
 637static struct request *
 638blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
 639{
 640        struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
 641
 642        if (!rq)
 643                return NULL;
 644
 645        blk_rq_init(q, rq);
 646
 647        rq->cmd_flags = rw | REQ_ALLOCED;
 648
 649        if (priv) {
 650                if (unlikely(elv_set_request(q, rq, gfp_mask))) {
 651                        mempool_free(rq, q->rq.rq_pool);
 652                        return NULL;
 653                }
 654                rq->cmd_flags |= REQ_ELVPRIV;
 655        }
 656
 657        return rq;
 658}
 659
 660/*
 661 * ioc_batching returns true if the ioc is a valid batching request and
 662 * should be given priority access to a request.
 663 */
 664static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
 665{
 666        if (!ioc)
 667                return 0;
 668
 669        /*
 670         * Make sure the process is able to allocate at least 1 request
 671         * even if the batch times out, otherwise we could theoretically
 672         * lose wakeups.
 673         */
 674        return ioc->nr_batch_requests == q->nr_batching ||
 675                (ioc->nr_batch_requests > 0
 676                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
 677}
 678
 679/*
 680 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
 681 * will cause the process to be a "batcher" on all queues in the system. This
 682 * is the behaviour we want though - once it gets a wakeup it should be given
 683 * a nice run.
 684 */
 685static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
 686{
 687        if (!ioc || ioc_batching(q, ioc))
 688                return;
 689
 690        ioc->nr_batch_requests = q->nr_batching;
 691        ioc->last_waited = jiffies;
 692}
 693
 694static void __freed_request(struct request_queue *q, int rw)
 695{
 696        struct request_list *rl = &q->rq;
 697
 698        if (rl->count[rw] < queue_congestion_off_threshold(q))
 699                blk_clear_queue_congested(q, rw);
 700
 701        if (rl->count[rw] + 1 <= q->nr_requests) {
 702                if (waitqueue_active(&rl->wait[rw]))
 703                        wake_up(&rl->wait[rw]);
 704
 705                blk_clear_queue_full(q, rw);
 706        }
 707}
 708
 709/*
 710 * A request has just been released.  Account for it, update the full and
 711 * congestion status, wake up any waiters.   Called under q->queue_lock.
 712 */
 713static void freed_request(struct request_queue *q, int rw, int priv)
 714{
 715        struct request_list *rl = &q->rq;
 716
 717        rl->count[rw]--;
 718        if (priv)
 719                rl->elvpriv--;
 720
 721        __freed_request(q, rw);
 722
 723        if (unlikely(rl->starved[rw ^ 1]))
 724                __freed_request(q, rw ^ 1);
 725}
 726
 727#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
 728/*
 729 * Get a free request, queue_lock must be held.
 730 * Returns NULL on failure, with queue_lock held.
 731 * Returns !NULL on success, with queue_lock *not held*.
 732 */
 733static struct request *get_request(struct request_queue *q, int rw_flags,
 734                                   struct bio *bio, gfp_t gfp_mask)
 735{
 736        struct request *rq = NULL;
 737        struct request_list *rl = &q->rq;
 738        struct io_context *ioc = NULL;
 739        const int rw = rw_flags & 0x01;
 740        int may_queue, priv;
 741
 742        may_queue = elv_may_queue(q, rw_flags);
 743        if (may_queue == ELV_MQUEUE_NO)
 744                goto rq_starved;
 745
 746        if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
 747                if (rl->count[rw]+1 >= q->nr_requests) {
 748                        ioc = current_io_context(GFP_ATOMIC, q->node);
 749                        /*
 750                         * The queue will fill after this allocation, so set
 751                         * it as full, and mark this process as "batching".
 752                         * This process will be allowed to complete a batch of
 753                         * requests, others will be blocked.
 754                         */
 755                        if (!blk_queue_full(q, rw)) {
 756                                ioc_set_batching(q, ioc);
 757                                blk_set_queue_full(q, rw);
 758                        } else {
 759                                if (may_queue != ELV_MQUEUE_MUST
 760                                                && !ioc_batching(q, ioc)) {
 761                                        /*
 762                                         * The queue is full and the allocating
 763                                         * process is not a "batcher", and not
 764                                         * exempted by the IO scheduler
 765                                         */
 766                                        goto out;
 767                                }
 768                        }
 769                }
 770                blk_set_queue_congested(q, rw);
 771        }
 772
 773        /*
 774         * Only allow batching queuers to allocate up to 50% over the defined
 775         * limit of requests, otherwise we could have thousands of requests
 776         * allocated with any setting of ->nr_requests
 777         */
 778        if (rl->count[rw] >= (3 * q->nr_requests / 2))
 779                goto out;
 780
 781        rl->count[rw]++;
 782        rl->starved[rw] = 0;
 783
 784        priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
 785        if (priv)
 786                rl->elvpriv++;
 787
 788        spin_unlock_irq(q->queue_lock);
 789
 790        rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
 791        if (unlikely(!rq)) {
 792                /*
 793                 * Allocation failed presumably due to memory. Undo anything
 794                 * we might have messed up.
 795                 *
 796                 * Allocating task should really be put onto the front of the
 797                 * wait queue, but this is pretty rare.
 798                 */
 799                spin_lock_irq(q->queue_lock);
 800                freed_request(q, rw, priv);
 801
 802                /*
 803                 * in the very unlikely event that allocation failed and no
 804                 * requests for this direction was pending, mark us starved
 805                 * so that freeing of a request in the other direction will
 806                 * notice us. another possible fix would be to split the
 807                 * rq mempool into READ and WRITE
 808                 */
 809rq_starved:
 810                if (unlikely(rl->count[rw] == 0))
 811                        rl->starved[rw] = 1;
 812
 813                goto out;
 814        }
 815
 816        /*
 817         * ioc may be NULL here, and ioc_batching will be false. That's
 818         * OK, if the queue is under the request limit then requests need
 819         * not count toward the nr_batch_requests limit. There will always
 820         * be some limit enforced by BLK_BATCH_TIME.
 821         */
 822        if (ioc_batching(q, ioc))
 823                ioc->nr_batch_requests--;
 824
 825        blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
 826out:
 827        return rq;
 828}
 829
 830/*
 831 * No available requests for this queue, unplug the device and wait for some
 832 * requests to become available.
 833 *
 834 * Called with q->queue_lock held, and returns with it unlocked.
 835 */
 836static struct request *get_request_wait(struct request_queue *q, int rw_flags,
 837                                        struct bio *bio)
 838{
 839        const int rw = rw_flags & 0x01;
 840        struct request *rq;
 841
 842        rq = get_request(q, rw_flags, bio, GFP_NOIO);
 843        while (!rq) {
 844                DEFINE_WAIT(wait);
 845                struct io_context *ioc;
 846                struct request_list *rl = &q->rq;
 847
 848                prepare_to_wait_exclusive(&rl->wait[rw], &wait,
 849                                TASK_UNINTERRUPTIBLE);
 850
 851                blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
 852
 853                __generic_unplug_device(q);
 854                spin_unlock_irq(q->queue_lock);
 855                io_schedule();
 856
 857                /*
 858                 * After sleeping, we become a "batching" process and
 859                 * will be able to allocate at least one request, and
 860                 * up to a big batch of them for a small period time.
 861                 * See ioc_batching, ioc_set_batching
 862                 */
 863                ioc = current_io_context(GFP_NOIO, q->node);
 864                ioc_set_batching(q, ioc);
 865
 866                spin_lock_irq(q->queue_lock);
 867                finish_wait(&rl->wait[rw], &wait);
 868
 869                rq = get_request(q, rw_flags, bio, GFP_NOIO);
 870        };
 871
 872        return rq;
 873}
 874
 875struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
 876{
 877        struct request *rq;
 878
 879        BUG_ON(rw != READ && rw != WRITE);
 880
 881        spin_lock_irq(q->queue_lock);
 882        if (gfp_mask & __GFP_WAIT) {
 883                rq = get_request_wait(q, rw, NULL);
 884        } else {
 885                rq = get_request(q, rw, NULL, gfp_mask);
 886                if (!rq)
 887                        spin_unlock_irq(q->queue_lock);
 888        }
 889        /* q->queue_lock is unlocked at this point */
 890
 891        return rq;
 892}
 893EXPORT_SYMBOL(blk_get_request);
 894
 895/**
 896 * blk_start_queueing - initiate dispatch of requests to device
 897 * @q:          request queue to kick into gear
 898 *
 899 * This is basically a helper to remove the need to know whether a queue
 900 * is plugged or not if someone just wants to initiate dispatch of requests
 901 * for this queue. Should be used to start queueing on a device outside
 902 * of ->request_fn() context. Also see @blk_run_queue.
 903 *
 904 * The queue lock must be held with interrupts disabled.
 905 */
 906void blk_start_queueing(struct request_queue *q)
 907{
 908        if (!blk_queue_plugged(q)) {
 909                if (unlikely(blk_queue_stopped(q)))
 910                        return;
 911                q->request_fn(q);
 912        } else
 913                __generic_unplug_device(q);
 914}
 915EXPORT_SYMBOL(blk_start_queueing);
 916
 917/**
 918 * blk_requeue_request - put a request back on queue
 919 * @q:          request queue where request should be inserted
 920 * @rq:         request to be inserted
 921 *
 922 * Description:
 923 *    Drivers often keep queueing requests until the hardware cannot accept
 924 *    more, when that condition happens we need to put the request back
 925 *    on the queue. Must be called with queue lock held.
 926 */
 927void blk_requeue_request(struct request_queue *q, struct request *rq)
 928{
 929        blk_delete_timer(rq);
 930        blk_clear_rq_complete(rq);
 931        blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 932
 933        if (blk_rq_tagged(rq))
 934                blk_queue_end_tag(q, rq);
 935
 936        elv_requeue_request(q, rq);
 937}
 938EXPORT_SYMBOL(blk_requeue_request);
 939
 940/**
 941 * blk_insert_request - insert a special request into a request queue
 942 * @q:          request queue where request should be inserted
 943 * @rq:         request to be inserted
 944 * @at_head:    insert request at head or tail of queue
 945 * @data:       private data
 946 *
 947 * Description:
 948 *    Many block devices need to execute commands asynchronously, so they don't
 949 *    block the whole kernel from preemption during request execution.  This is
 950 *    accomplished normally by inserting aritficial requests tagged as
 951 *    REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
 952 *    be scheduled for actual execution by the request queue.
 953 *
 954 *    We have the option of inserting the head or the tail of the queue.
 955 *    Typically we use the tail for new ioctls and so forth.  We use the head
 956 *    of the queue for things like a QUEUE_FULL message from a device, or a
 957 *    host that is unable to accept a particular command.
 958 */
 959void blk_insert_request(struct request_queue *q, struct request *rq,
 960                        int at_head, void *data)
 961{
 962        int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
 963        unsigned long flags;
 964
 965        /*
 966         * tell I/O scheduler that this isn't a regular read/write (ie it
 967         * must not attempt merges on this) and that it acts as a soft
 968         * barrier
 969         */
 970        rq->cmd_type = REQ_TYPE_SPECIAL;
 971        rq->cmd_flags |= REQ_SOFTBARRIER;
 972
 973        rq->special = data;
 974
 975        spin_lock_irqsave(q->queue_lock, flags);
 976
 977        /*
 978         * If command is tagged, release the tag
 979         */
 980        if (blk_rq_tagged(rq))
 981                blk_queue_end_tag(q, rq);
 982
 983        drive_stat_acct(rq, 1);
 984        __elv_add_request(q, rq, where, 0);
 985        blk_start_queueing(q);
 986        spin_unlock_irqrestore(q->queue_lock, flags);
 987}
 988EXPORT_SYMBOL(blk_insert_request);
 989
 990/*
 991 * add-request adds a request to the linked list.
 992 * queue lock is held and interrupts disabled, as we muck with the
 993 * request queue list.
 994 */
 995static inline void add_request(struct request_queue *q, struct request *req)
 996{
 997        drive_stat_acct(req, 1);
 998
 999        /*
1000         * elevator indicated where it wants this request to be
1001         * inserted at elevator_merge time
1002         */
1003        __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1004}
1005
1006static void part_round_stats_single(int cpu, struct hd_struct *part,
1007                                    unsigned long now)
1008{
1009        if (now == part->stamp)
1010                return;
1011
1012        if (part->in_flight) {
1013                __part_stat_add(cpu, part, time_in_queue,
1014                                part->in_flight * (now - part->stamp));
1015                __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1016        }
1017        part->stamp = now;
1018}
1019
1020/**
1021 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1022 * @cpu: cpu number for stats access
1023 * @part: target partition
1024 *
1025 * The average IO queue length and utilisation statistics are maintained
1026 * by observing the current state of the queue length and the amount of
1027 * time it has been in this state for.
1028 *
1029 * Normally, that accounting is done on IO completion, but that can result
1030 * in more than a second's worth of IO being accounted for within any one
1031 * second, leading to >100% utilisation.  To deal with that, we call this
1032 * function to do a round-off before returning the results when reading
1033 * /proc/diskstats.  This accounts immediately for all queue usage up to
1034 * the current jiffies and restarts the counters again.
1035 */
1036void part_round_stats(int cpu, struct hd_struct *part)
1037{
1038        unsigned long now = jiffies;
1039
1040        if (part->partno)
1041                part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1042        part_round_stats_single(cpu, part, now);
1043}
1044EXPORT_SYMBOL_GPL(part_round_stats);
1045
1046/*
1047 * queue lock must be held
1048 */
1049void __blk_put_request(struct request_queue *q, struct request *req)
1050{
1051        if (unlikely(!q))
1052                return;
1053        if (unlikely(--req->ref_count))
1054                return;
1055
1056        elv_completed_request(q, req);
1057
1058        /*
1059         * Request may not have originated from ll_rw_blk. if not,
1060         * it didn't come out of our reserved rq pools
1061         */
1062        if (req->cmd_flags & REQ_ALLOCED) {
1063                int rw = rq_data_dir(req);
1064                int priv = req->cmd_flags & REQ_ELVPRIV;
1065
1066                BUG_ON(!list_empty(&req->queuelist));
1067                BUG_ON(!hlist_unhashed(&req->hash));
1068
1069                blk_free_request(q, req);
1070                freed_request(q, rw, priv);
1071        }
1072}
1073EXPORT_SYMBOL_GPL(__blk_put_request);
1074
1075void blk_put_request(struct request *req)
1076{
1077        unsigned long flags;
1078        struct request_queue *q = req->q;
1079
1080        spin_lock_irqsave(q->queue_lock, flags);
1081        __blk_put_request(q, req);
1082        spin_unlock_irqrestore(q->queue_lock, flags);
1083}
1084EXPORT_SYMBOL(blk_put_request);
1085
1086void init_request_from_bio(struct request *req, struct bio *bio)
1087{
1088        req->cpu = bio->bi_comp_cpu;
1089        req->cmd_type = REQ_TYPE_FS;
1090
1091        /*
1092         * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1093         */
1094        if (bio_rw_ahead(bio))
1095                req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1096                                   REQ_FAILFAST_DRIVER);
1097        if (bio_failfast_dev(bio))
1098                req->cmd_flags |= REQ_FAILFAST_DEV;
1099        if (bio_failfast_transport(bio))
1100                req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1101        if (bio_failfast_driver(bio))
1102                req->cmd_flags |= REQ_FAILFAST_DRIVER;
1103
1104        /*
1105         * REQ_BARRIER implies no merging, but lets make it explicit
1106         */
1107        if (unlikely(bio_discard(bio))) {
1108                req->cmd_flags |= REQ_DISCARD;
1109                if (bio_barrier(bio))
1110                        req->cmd_flags |= REQ_SOFTBARRIER;
1111                req->q->prepare_discard_fn(req->q, req);
1112        } else if (unlikely(bio_barrier(bio)))
1113                req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
1114
1115        if (bio_sync(bio))
1116                req->cmd_flags |= REQ_RW_SYNC;
1117        if (bio_rw_meta(bio))
1118                req->cmd_flags |= REQ_RW_META;
1119
1120        req->errors = 0;
1121        req->hard_sector = req->sector = bio->bi_sector;
1122        req->ioprio = bio_prio(bio);
1123        req->start_time = jiffies;
1124        blk_rq_bio_prep(req->q, req, bio);
1125}
1126
1127static int __make_request(struct request_queue *q, struct bio *bio)
1128{
1129        struct request *req;
1130        int el_ret, nr_sectors, barrier, discard, err;
1131        const unsigned short prio = bio_prio(bio);
1132        const int sync = bio_sync(bio);
1133        int rw_flags;
1134
1135        nr_sectors = bio_sectors(bio);
1136
1137        /*
1138         * low level driver can indicate that it wants pages above a
1139         * certain limit bounced to low memory (ie for highmem, or even
1140         * ISA dma in theory)
1141         */
1142        blk_queue_bounce(q, &bio);
1143
1144        barrier = bio_barrier(bio);
1145        if (unlikely(barrier) && bio_has_data(bio) &&
1146            (q->next_ordered == QUEUE_ORDERED_NONE)) {
1147                err = -EOPNOTSUPP;
1148                goto end_io;
1149        }
1150
1151        discard = bio_discard(bio);
1152        if (unlikely(discard) && !q->prepare_discard_fn) {
1153                err = -EOPNOTSUPP;
1154                goto end_io;
1155        }
1156
1157        spin_lock_irq(q->queue_lock);
1158
1159        if (unlikely(barrier) || elv_queue_empty(q))
1160                goto get_rq;
1161
1162        el_ret = elv_merge(q, &req, bio);
1163        switch (el_ret) {
1164        case ELEVATOR_BACK_MERGE:
1165                BUG_ON(!rq_mergeable(req));
1166
1167                if (!ll_back_merge_fn(q, req, bio))
1168                        break;
1169
1170                blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
1171
1172                req->biotail->bi_next = bio;
1173                req->biotail = bio;
1174                req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1175                req->ioprio = ioprio_best(req->ioprio, prio);
1176                if (!blk_rq_cpu_valid(req))
1177                        req->cpu = bio->bi_comp_cpu;
1178                drive_stat_acct(req, 0);
1179                if (!attempt_back_merge(q, req))
1180                        elv_merged_request(q, req, el_ret);
1181                goto out;
1182
1183        case ELEVATOR_FRONT_MERGE:
1184                BUG_ON(!rq_mergeable(req));
1185
1186                if (!ll_front_merge_fn(q, req, bio))
1187                        break;
1188
1189                blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
1190
1191                bio->bi_next = req->bio;
1192                req->bio = bio;
1193
1194                /*
1195                 * may not be valid. if the low level driver said
1196                 * it didn't need a bounce buffer then it better
1197                 * not touch req->buffer either...
1198                 */
1199                req->buffer = bio_data(bio);
1200                req->current_nr_sectors = bio_cur_sectors(bio);
1201                req->hard_cur_sectors = req->current_nr_sectors;
1202                req->sector = req->hard_sector = bio->bi_sector;
1203                req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1204                req->ioprio = ioprio_best(req->ioprio, prio);
1205                if (!blk_rq_cpu_valid(req))
1206                        req->cpu = bio->bi_comp_cpu;
1207                drive_stat_acct(req, 0);
1208                if (!attempt_front_merge(q, req))
1209                        elv_merged_request(q, req, el_ret);
1210                goto out;
1211
1212        /* ELV_NO_MERGE: elevator says don't/can't merge. */
1213        default:
1214                ;
1215        }
1216
1217get_rq:
1218        /*
1219         * This sync check and mask will be re-done in init_request_from_bio(),
1220         * but we need to set it earlier to expose the sync flag to the
1221         * rq allocator and io schedulers.
1222         */
1223        rw_flags = bio_data_dir(bio);
1224        if (sync)
1225                rw_flags |= REQ_RW_SYNC;
1226
1227        /*
1228         * Grab a free request. This is might sleep but can not fail.
1229         * Returns with the queue unlocked.
1230         */
1231        req = get_request_wait(q, rw_flags, bio);
1232
1233        /*
1234         * After dropping the lock and possibly sleeping here, our request
1235         * may now be mergeable after it had proven unmergeable (above).
1236         * We don't worry about that case for efficiency. It won't happen
1237         * often, and the elevators are able to handle it.
1238         */
1239        init_request_from_bio(req, bio);
1240
1241        spin_lock_irq(q->queue_lock);
1242        if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1243            bio_flagged(bio, BIO_CPU_AFFINE))
1244                req->cpu = blk_cpu_to_group(smp_processor_id());
1245        if (elv_queue_empty(q))
1246                blk_plug_device(q);
1247        add_request(q, req);
1248out:
1249        if (sync)
1250                __generic_unplug_device(q);
1251        spin_unlock_irq(q->queue_lock);
1252        return 0;
1253
1254end_io:
1255        bio_endio(bio, err);
1256        return 0;
1257}
1258
1259/*
1260 * If bio->bi_dev is a partition, remap the location
1261 */
1262static inline void blk_partition_remap(struct bio *bio)
1263{
1264        struct block_device *bdev = bio->bi_bdev;
1265
1266        if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1267                struct hd_struct *p = bdev->bd_part;
1268
1269                bio->bi_sector += p->start_sect;
1270                bio->bi_bdev = bdev->bd_contains;
1271
1272                blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1273                                    bdev->bd_dev, bio->bi_sector,
1274                                    bio->bi_sector - p->start_sect);
1275        }
1276}
1277
1278static void handle_bad_sector(struct bio *bio)
1279{
1280        char b[BDEVNAME_SIZE];
1281
1282        printk(KERN_INFO "attempt to access beyond end of device\n");
1283        printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1284                        bdevname(bio->bi_bdev, b),
1285                        bio->bi_rw,
1286                        (unsigned long long)bio->bi_sector + bio_sectors(bio),
1287                        (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1288
1289        set_bit(BIO_EOF, &bio->bi_flags);
1290}
1291
1292#ifdef CONFIG_FAIL_MAKE_REQUEST
1293
1294static DECLARE_FAULT_ATTR(fail_make_request);
1295
1296static int __init setup_fail_make_request(char *str)
1297{
1298        return setup_fault_attr(&fail_make_request, str);
1299}
1300__setup("fail_make_request=", setup_fail_make_request);
1301
1302static int should_fail_request(struct bio *bio)
1303{
1304        struct hd_struct *part = bio->bi_bdev->bd_part;
1305
1306        if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1307                return should_fail(&fail_make_request, bio->bi_size);
1308
1309        return 0;
1310}
1311
1312static int __init fail_make_request_debugfs(void)
1313{
1314        return init_fault_attr_dentries(&fail_make_request,
1315                                        "fail_make_request");
1316}
1317
1318late_initcall(fail_make_request_debugfs);
1319
1320#else /* CONFIG_FAIL_MAKE_REQUEST */
1321
1322static inline int should_fail_request(struct bio *bio)
1323{
1324        return 0;
1325}
1326
1327#endif /* CONFIG_FAIL_MAKE_REQUEST */
1328
1329/*
1330 * Check whether this bio extends beyond the end of the device.
1331 */
1332static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1333{
1334        sector_t maxsector;
1335
1336        if (!nr_sectors)
1337                return 0;
1338
1339        /* Test device or partition size, when known. */
1340        maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1341        if (maxsector) {
1342                sector_t sector = bio->bi_sector;
1343
1344                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1345                        /*
1346                         * This may well happen - the kernel calls bread()
1347                         * without checking the size of the device, e.g., when
1348                         * mounting a device.
1349                         */
1350                        handle_bad_sector(bio);
1351                        return 1;
1352                }
1353        }
1354
1355        return 0;
1356}
1357
1358/**
1359 * generic_make_request - hand a buffer to its device driver for I/O
1360 * @bio:  The bio describing the location in memory and on the device.
1361 *
1362 * generic_make_request() is used to make I/O requests of block
1363 * devices. It is passed a &struct bio, which describes the I/O that needs
1364 * to be done.
1365 *
1366 * generic_make_request() does not return any status.  The
1367 * success/failure status of the request, along with notification of
1368 * completion, is delivered asynchronously through the bio->bi_end_io
1369 * function described (one day) else where.
1370 *
1371 * The caller of generic_make_request must make sure that bi_io_vec
1372 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1373 * set to describe the device address, and the
1374 * bi_end_io and optionally bi_private are set to describe how
1375 * completion notification should be signaled.
1376 *
1377 * generic_make_request and the drivers it calls may use bi_next if this
1378 * bio happens to be merged with someone else, and may change bi_dev and
1379 * bi_sector for remaps as it sees fit.  So the values of these fields
1380 * should NOT be depended on after the call to generic_make_request.
1381 */
1382static inline void __generic_make_request(struct bio *bio)
1383{
1384        struct request_queue *q;
1385        sector_t old_sector;
1386        int ret, nr_sectors = bio_sectors(bio);
1387        dev_t old_dev;
1388        int err = -EIO;
1389
1390        might_sleep();
1391
1392        if (bio_check_eod(bio, nr_sectors))
1393                goto end_io;
1394
1395        /*
1396         * Resolve the mapping until finished. (drivers are
1397         * still free to implement/resolve their own stacking
1398         * by explicitly returning 0)
1399         *
1400         * NOTE: we don't repeat the blk_size check for each new device.
1401         * Stacking drivers are expected to know what they are doing.
1402         */
1403        old_sector = -1;
1404        old_dev = 0;
1405        do {
1406                char b[BDEVNAME_SIZE];
1407
1408                q = bdev_get_queue(bio->bi_bdev);
1409                if (!q) {
1410                        printk(KERN_ERR
1411                               "generic_make_request: Trying to access "
1412                                "nonexistent block-device %s (%Lu)\n",
1413                                bdevname(bio->bi_bdev, b),
1414                                (long long) bio->bi_sector);
1415end_io:
1416                        bio_endio(bio, err);
1417                        break;
1418                }
1419
1420                if (unlikely(nr_sectors > q->max_hw_sectors)) {
1421                        printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1422                                bdevname(bio->bi_bdev, b),
1423                                bio_sectors(bio),
1424                                q->max_hw_sectors);
1425                        goto end_io;
1426                }
1427
1428                if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1429                        goto end_io;
1430
1431                if (should_fail_request(bio))
1432                        goto end_io;
1433
1434                /*
1435                 * If this device has partitions, remap block n
1436                 * of partition p to block n+start(p) of the disk.
1437                 */
1438                blk_partition_remap(bio);
1439
1440                if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1441                        goto end_io;
1442
1443                if (old_sector != -1)
1444                        blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
1445                                            old_sector);
1446
1447                blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1448
1449                old_sector = bio->bi_sector;
1450                old_dev = bio->bi_bdev->bd_dev;
1451
1452                if (bio_check_eod(bio, nr_sectors))
1453                        goto end_io;
1454                if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) ||
1455                    (bio_discard(bio) && !q->prepare_discard_fn)) {
1456                        err = -EOPNOTSUPP;
1457                        goto end_io;
1458                }
1459
1460                ret = q->make_request_fn(q, bio);
1461        } while (ret);
1462}
1463
1464/*
1465 * We only want one ->make_request_fn to be active at a time,
1466 * else stack usage with stacked devices could be a problem.
1467 * So use current->bio_{list,tail} to keep a list of requests
1468 * submited by a make_request_fn function.
1469 * current->bio_tail is also used as a flag to say if
1470 * generic_make_request is currently active in this task or not.
1471 * If it is NULL, then no make_request is active.  If it is non-NULL,
1472 * then a make_request is active, and new requests should be added
1473 * at the tail
1474 */
1475void generic_make_request(struct bio *bio)
1476{
1477        if (current->bio_tail) {
1478                /* make_request is active */
1479                *(current->bio_tail) = bio;
1480                bio->bi_next = NULL;
1481                current->bio_tail = &bio->bi_next;
1482                return;
1483        }
1484        /* following loop may be a bit non-obvious, and so deserves some
1485         * explanation.
1486         * Before entering the loop, bio->bi_next is NULL (as all callers
1487         * ensure that) so we have a list with a single bio.
1488         * We pretend that we have just taken it off a longer list, so
1489         * we assign bio_list to the next (which is NULL) and bio_tail
1490         * to &bio_list, thus initialising the bio_list of new bios to be
1491         * added.  __generic_make_request may indeed add some more bios
1492         * through a recursive call to generic_make_request.  If it
1493         * did, we find a non-NULL value in bio_list and re-enter the loop
1494         * from the top.  In this case we really did just take the bio
1495         * of the top of the list (no pretending) and so fixup bio_list and
1496         * bio_tail or bi_next, and call into __generic_make_request again.
1497         *
1498         * The loop was structured like this to make only one call to
1499         * __generic_make_request (which is important as it is large and
1500         * inlined) and to keep the structure simple.
1501         */
1502        BUG_ON(bio->bi_next);
1503        do {
1504                current->bio_list = bio->bi_next;
1505                if (bio->bi_next == NULL)
1506                        current->bio_tail = &current->bio_list;
1507                else
1508                        bio->bi_next = NULL;
1509                __generic_make_request(bio);
1510                bio = current->bio_list;
1511        } while (bio);
1512        current->bio_tail = NULL; /* deactivate */
1513}
1514EXPORT_SYMBOL(generic_make_request);
1515
1516/**
1517 * submit_bio - submit a bio to the block device layer for I/O
1518 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1519 * @bio: The &struct bio which describes the I/O
1520 *
1521 * submit_bio() is very similar in purpose to generic_make_request(), and
1522 * uses that function to do most of the work. Both are fairly rough
1523 * interfaces; @bio must be presetup and ready for I/O.
1524 *
1525 */
1526void submit_bio(int rw, struct bio *bio)
1527{
1528        int count = bio_sectors(bio);
1529
1530        bio->bi_rw |= rw;
1531
1532        /*
1533         * If it's a regular read/write or a barrier with data attached,
1534         * go through the normal accounting stuff before submission.
1535         */
1536        if (bio_has_data(bio)) {
1537                if (rw & WRITE) {
1538                        count_vm_events(PGPGOUT, count);
1539                } else {
1540                        task_io_account_read(bio->bi_size);
1541                        count_vm_events(PGPGIN, count);
1542                }
1543
1544                if (unlikely(block_dump)) {
1545                        char b[BDEVNAME_SIZE];
1546                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1547                        current->comm, task_pid_nr(current),
1548                                (rw & WRITE) ? "WRITE" : "READ",
1549                                (unsigned long long)bio->bi_sector,
1550                                bdevname(bio->bi_bdev, b));
1551                }
1552        }
1553
1554        generic_make_request(bio);
1555}
1556EXPORT_SYMBOL(submit_bio);
1557
1558/**
1559 * blk_rq_check_limits - Helper function to check a request for the queue limit
1560 * @q:  the queue
1561 * @rq: the request being checked
1562 *
1563 * Description:
1564 *    @rq may have been made based on weaker limitations of upper-level queues
1565 *    in request stacking drivers, and it may violate the limitation of @q.
1566 *    Since the block layer and the underlying device driver trust @rq
1567 *    after it is inserted to @q, it should be checked against @q before
1568 *    the insertion using this generic function.
1569 *
1570 *    This function should also be useful for request stacking drivers
1571 *    in some cases below, so export this fuction.
1572 *    Request stacking drivers like request-based dm may change the queue
1573 *    limits while requests are in the queue (e.g. dm's table swapping).
1574 *    Such request stacking drivers should check those requests agaist
1575 *    the new queue limits again when they dispatch those requests,
1576 *    although such checkings are also done against the old queue limits
1577 *    when submitting requests.
1578 */
1579int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1580{
1581        if (rq->nr_sectors > q->max_sectors ||
1582            rq->data_len > q->max_hw_sectors << 9) {
1583                printk(KERN_ERR "%s: over max size limit.\n", __func__);
1584                return -EIO;
1585        }
1586
1587        /*
1588         * queue's settings related to segment counting like q->bounce_pfn
1589         * may differ from that of other stacking queues.
1590         * Recalculate it to check the request correctly on this queue's
1591         * limitation.
1592         */
1593        blk_recalc_rq_segments(rq);
1594        if (rq->nr_phys_segments > q->max_phys_segments ||
1595            rq->nr_phys_segments > q->max_hw_segments) {
1596                printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1597                return -EIO;
1598        }
1599
1600        return 0;
1601}
1602EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1603
1604/**
1605 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1606 * @q:  the queue to submit the request
1607 * @rq: the request being queued
1608 */
1609int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1610{
1611        unsigned long flags;
1612
1613        if (blk_rq_check_limits(q, rq))
1614                return -EIO;
1615
1616#ifdef CONFIG_FAIL_MAKE_REQUEST
1617        if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1618            should_fail(&fail_make_request, blk_rq_bytes(rq)))
1619                return -EIO;
1620#endif
1621
1622        spin_lock_irqsave(q->queue_lock, flags);
1623
1624        /*
1625         * Submitting request must be dequeued before calling this function
1626         * because it will be linked to another request_queue
1627         */
1628        BUG_ON(blk_queued_rq(rq));
1629
1630        drive_stat_acct(rq, 1);
1631        __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1632
1633        spin_unlock_irqrestore(q->queue_lock, flags);
1634
1635        return 0;
1636}
1637EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1638
1639/**
1640 * blkdev_dequeue_request - dequeue request and start timeout timer
1641 * @req: request to dequeue
1642 *
1643 * Dequeue @req and start timeout timer on it.  This hands off the
1644 * request to the driver.
1645 *
1646 * Block internal functions which don't want to start timer should
1647 * call elv_dequeue_request().
1648 */
1649void blkdev_dequeue_request(struct request *req)
1650{
1651        elv_dequeue_request(req->q, req);
1652
1653        /*
1654         * We are now handing the request to the hardware, add the
1655         * timeout handler.
1656         */
1657        blk_add_timer(req);
1658}
1659EXPORT_SYMBOL(blkdev_dequeue_request);
1660
1661/**
1662 * __end_that_request_first - end I/O on a request
1663 * @req:      the request being processed
1664 * @error:    %0 for success, < %0 for error
1665 * @nr_bytes: number of bytes to complete
1666 *
1667 * Description:
1668 *     Ends I/O on a number of bytes attached to @req, and sets it up
1669 *     for the next range of segments (if any) in the cluster.
1670 *
1671 * Return:
1672 *     %0 - we are done with this request, call end_that_request_last()
1673 *     %1 - still buffers pending for this request
1674 **/
1675static int __end_that_request_first(struct request *req, int error,
1676                                    int nr_bytes)
1677{
1678        int total_bytes, bio_nbytes, next_idx = 0;
1679        struct bio *bio;
1680
1681        blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1682
1683        /*
1684         * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1685         * sense key with us all the way through
1686         */
1687        if (!blk_pc_request(req))
1688                req->errors = 0;
1689
1690        if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1691                printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1692                                req->rq_disk ? req->rq_disk->disk_name : "?",
1693                                (unsigned long long)req->sector);
1694        }
1695
1696        if (blk_fs_request(req) && req->rq_disk) {
1697                const int rw = rq_data_dir(req);
1698                struct hd_struct *part;
1699                int cpu;
1700
1701                cpu = part_stat_lock();
1702                part = disk_map_sector_rcu(req->rq_disk, req->sector);
1703                part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9);
1704                part_stat_unlock();
1705        }
1706
1707        total_bytes = bio_nbytes = 0;
1708        while ((bio = req->bio) != NULL) {
1709                int nbytes;
1710
1711                /*
1712                 * For an empty barrier request, the low level driver must
1713                 * store a potential error location in ->sector. We pass
1714                 * that back up in ->bi_sector.
1715                 */
1716                if (blk_empty_barrier(req))
1717                        bio->bi_sector = req->sector;
1718
1719                if (nr_bytes >= bio->bi_size) {
1720                        req->bio = bio->bi_next;
1721                        nbytes = bio->bi_size;
1722                        req_bio_endio(req, bio, nbytes, error);
1723                        next_idx = 0;
1724                        bio_nbytes = 0;
1725                } else {
1726                        int idx = bio->bi_idx + next_idx;
1727
1728                        if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1729                                blk_dump_rq_flags(req, "__end_that");
1730                                printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1731                                       __func__, bio->bi_idx, bio->bi_vcnt);
1732                                break;
1733                        }
1734
1735                        nbytes = bio_iovec_idx(bio, idx)->bv_len;
1736                        BIO_BUG_ON(nbytes > bio->bi_size);
1737
1738                        /*
1739                         * not a complete bvec done
1740                         */
1741                        if (unlikely(nbytes > nr_bytes)) {
1742                                bio_nbytes += nr_bytes;
1743                                total_bytes += nr_bytes;
1744                                break;
1745                        }
1746
1747                        /*
1748                         * advance to the next vector
1749                         */
1750                        next_idx++;
1751                        bio_nbytes += nbytes;
1752                }
1753
1754                total_bytes += nbytes;
1755                nr_bytes -= nbytes;
1756
1757                bio = req->bio;
1758                if (bio) {
1759                        /*
1760                         * end more in this run, or just return 'not-done'
1761                         */
1762                        if (unlikely(nr_bytes <= 0))
1763                                break;
1764                }
1765        }
1766
1767        /*
1768         * completely done
1769         */
1770        if (!req->bio)
1771                return 0;
1772
1773        /*
1774         * if the request wasn't completed, update state
1775         */
1776        if (bio_nbytes) {
1777                req_bio_endio(req, bio, bio_nbytes, error);
1778                bio->bi_idx += next_idx;
1779                bio_iovec(bio)->bv_offset += nr_bytes;
1780                bio_iovec(bio)->bv_len -= nr_bytes;
1781        }
1782
1783        blk_recalc_rq_sectors(req, total_bytes >> 9);
1784        blk_recalc_rq_segments(req);
1785        return 1;
1786}
1787
1788/*
1789 * queue lock must be held
1790 */
1791static void end_that_request_last(struct request *req, int error)
1792{
1793        struct gendisk *disk = req->rq_disk;
1794
1795        if (blk_rq_tagged(req))
1796                blk_queue_end_tag(req->q, req);
1797
1798        if (blk_queued_rq(req))
1799                elv_dequeue_request(req->q, req);
1800
1801        if (unlikely(laptop_mode) && blk_fs_request(req))
1802                laptop_io_completion();
1803
1804        blk_delete_timer(req);
1805
1806        /*
1807         * Account IO completion.  bar_rq isn't accounted as a normal
1808         * IO on queueing nor completion.  Accounting the containing
1809         * request is enough.
1810         */
1811        if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
1812                unsigned long duration = jiffies - req->start_time;
1813                const int rw = rq_data_dir(req);
1814                struct hd_struct *part;
1815                int cpu;
1816
1817                cpu = part_stat_lock();
1818                part = disk_map_sector_rcu(disk, req->sector);
1819
1820                part_stat_inc(cpu, part, ios[rw]);
1821                part_stat_add(cpu, part, ticks[rw], duration);
1822                part_round_stats(cpu, part);
1823                part_dec_in_flight(part);
1824
1825                part_stat_unlock();
1826        }
1827
1828        if (req->end_io)
1829                req->end_io(req, error);
1830        else {
1831                if (blk_bidi_rq(req))
1832                        __blk_put_request(req->next_rq->q, req->next_rq);
1833
1834                __blk_put_request(req->q, req);
1835        }
1836}
1837
1838/**
1839 * blk_rq_bytes - Returns bytes left to complete in the entire request
1840 * @rq: the request being processed
1841 **/
1842unsigned int blk_rq_bytes(struct request *rq)
1843{
1844        if (blk_fs_request(rq))
1845                return rq->hard_nr_sectors << 9;
1846
1847        return rq->data_len;
1848}
1849EXPORT_SYMBOL_GPL(blk_rq_bytes);
1850
1851/**
1852 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1853 * @rq: the request being processed
1854 **/
1855unsigned int blk_rq_cur_bytes(struct request *rq)
1856{
1857        if (blk_fs_request(rq))
1858                return rq->current_nr_sectors << 9;
1859
1860        if (rq->bio)
1861                return rq->bio->bi_size;
1862
1863        return rq->data_len;
1864}
1865EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
1866
1867/**
1868 * end_request - end I/O on the current segment of the request
1869 * @req:        the request being processed
1870 * @uptodate:   error value or %0/%1 uptodate flag
1871 *
1872 * Description:
1873 *     Ends I/O on the current segment of a request. If that is the only
1874 *     remaining segment, the request is also completed and freed.
1875 *
1876 *     This is a remnant of how older block drivers handled I/O completions.
1877 *     Modern drivers typically end I/O on the full request in one go, unless
1878 *     they have a residual value to account for. For that case this function
1879 *     isn't really useful, unless the residual just happens to be the
1880 *     full current segment. In other words, don't use this function in new
1881 *     code. Use blk_end_request() or __blk_end_request() to end a request.
1882 **/
1883void end_request(struct request *req, int uptodate)
1884{
1885        int error = 0;
1886
1887        if (uptodate <= 0)
1888                error = uptodate ? uptodate : -EIO;
1889
1890        __blk_end_request(req, error, req->hard_cur_sectors << 9);
1891}
1892EXPORT_SYMBOL(end_request);
1893
1894static int end_that_request_data(struct request *rq, int error,
1895                                 unsigned int nr_bytes, unsigned int bidi_bytes)
1896{
1897        if (rq->bio) {
1898                if (__end_that_request_first(rq, error, nr_bytes))
1899                        return 1;
1900
1901                /* Bidi request must be completed as a whole */
1902                if (blk_bidi_rq(rq) &&
1903                    __end_that_request_first(rq->next_rq, error, bidi_bytes))
1904                        return 1;
1905        }
1906
1907        return 0;
1908}
1909
1910/**
1911 * blk_end_io - Generic end_io function to complete a request.
1912 * @rq:           the request being processed
1913 * @error:        %0 for success, < %0 for error
1914 * @nr_bytes:     number of bytes to complete @rq
1915 * @bidi_bytes:   number of bytes to complete @rq->next_rq
1916 * @drv_callback: function called between completion of bios in the request
1917 *                and completion of the request.
1918 *                If the callback returns non %0, this helper returns without
1919 *                completion of the request.
1920 *
1921 * Description:
1922 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1923 *     If @rq has leftover, sets it up for the next range of segments.
1924 *
1925 * Return:
1926 *     %0 - we are done with this request
1927 *     %1 - this request is not freed yet, it still has pending buffers.
1928 **/
1929static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1930                      unsigned int bidi_bytes,
1931                      int (drv_callback)(struct request *))
1932{
1933        struct request_queue *q = rq->q;
1934        unsigned long flags = 0UL;
1935
1936        if (end_that_request_data(rq, error, nr_bytes, bidi_bytes))
1937                return 1;
1938
1939        /* Special feature for tricky drivers */
1940        if (drv_callback && drv_callback(rq))
1941                return 1;
1942
1943        add_disk_randomness(rq->rq_disk);
1944
1945        spin_lock_irqsave(q->queue_lock, flags);
1946        end_that_request_last(rq, error);
1947        spin_unlock_irqrestore(q->queue_lock, flags);
1948
1949        return 0;
1950}
1951
1952/**
1953 * blk_end_request - Helper function for drivers to complete the request.
1954 * @rq:       the request being processed
1955 * @error:    %0 for success, < %0 for error
1956 * @nr_bytes: number of bytes to complete
1957 *
1958 * Description:
1959 *     Ends I/O on a number of bytes attached to @rq.
1960 *     If @rq has leftover, sets it up for the next range of segments.
1961 *
1962 * Return:
1963 *     %0 - we are done with this request
1964 *     %1 - still buffers pending for this request
1965 **/
1966int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1967{
1968        return blk_end_io(rq, error, nr_bytes, 0, NULL);
1969}
1970EXPORT_SYMBOL_GPL(blk_end_request);
1971
1972/**
1973 * __blk_end_request - Helper function for drivers to complete the request.
1974 * @rq:       the request being processed
1975 * @error:    %0 for success, < %0 for error
1976 * @nr_bytes: number of bytes to complete
1977 *
1978 * Description:
1979 *     Must be called with queue lock held unlike blk_end_request().
1980 *
1981 * Return:
1982 *     %0 - we are done with this request
1983 *     %1 - still buffers pending for this request
1984 **/
1985int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1986{
1987        if (rq->bio && __end_that_request_first(rq, error, nr_bytes))
1988                return 1;
1989
1990        add_disk_randomness(rq->rq_disk);
1991
1992        end_that_request_last(rq, error);
1993
1994        return 0;
1995}
1996EXPORT_SYMBOL_GPL(__blk_end_request);
1997
1998/**
1999 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
2000 * @rq:         the bidi request being processed
2001 * @error:      %0 for success, < %0 for error
2002 * @nr_bytes:   number of bytes to complete @rq
2003 * @bidi_bytes: number of bytes to complete @rq->next_rq
2004 *
2005 * Description:
2006 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2007 *
2008 * Return:
2009 *     %0 - we are done with this request
2010 *     %1 - still buffers pending for this request
2011 **/
2012int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
2013                         unsigned int bidi_bytes)
2014{
2015        return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
2016}
2017EXPORT_SYMBOL_GPL(blk_end_bidi_request);
2018
2019/**
2020 * blk_update_request - Special helper function for request stacking drivers
2021 * @rq:           the request being processed
2022 * @error:        %0 for success, < %0 for error
2023 * @nr_bytes:     number of bytes to complete @rq
2024 *
2025 * Description:
2026 *     Ends I/O on a number of bytes attached to @rq, but doesn't complete
2027 *     the request structure even if @rq doesn't have leftover.
2028 *     If @rq has leftover, sets it up for the next range of segments.
2029 *
2030 *     This special helper function is only for request stacking drivers
2031 *     (e.g. request-based dm) so that they can handle partial completion.
2032 *     Actual device drivers should use blk_end_request instead.
2033 */
2034void blk_update_request(struct request *rq, int error, unsigned int nr_bytes)
2035{
2036        if (!end_that_request_data(rq, error, nr_bytes, 0)) {
2037                /*
2038                 * These members are not updated in end_that_request_data()
2039                 * when all bios are completed.
2040                 * Update them so that the request stacking driver can find
2041                 * how many bytes remain in the request later.
2042                 */
2043                rq->nr_sectors = rq->hard_nr_sectors = 0;
2044                rq->current_nr_sectors = rq->hard_cur_sectors = 0;
2045        }
2046}
2047EXPORT_SYMBOL_GPL(blk_update_request);
2048
2049/**
2050 * blk_end_request_callback - Special helper function for tricky drivers
2051 * @rq:           the request being processed
2052 * @error:        %0 for success, < %0 for error
2053 * @nr_bytes:     number of bytes to complete
2054 * @drv_callback: function called between completion of bios in the request
2055 *                and completion of the request.
2056 *                If the callback returns non %0, this helper returns without
2057 *                completion of the request.
2058 *
2059 * Description:
2060 *     Ends I/O on a number of bytes attached to @rq.
2061 *     If @rq has leftover, sets it up for the next range of segments.
2062 *
2063 *     This special helper function is used only for existing tricky drivers.
2064 *     (e.g. cdrom_newpc_intr() of ide-cd)
2065 *     This interface will be removed when such drivers are rewritten.
2066 *     Don't use this interface in other places anymore.
2067 *
2068 * Return:
2069 *     %0 - we are done with this request
2070 *     %1 - this request is not freed yet.
2071 *          this request still has pending buffers or
2072 *          the driver doesn't want to finish this request yet.
2073 **/
2074int blk_end_request_callback(struct request *rq, int error,
2075                             unsigned int nr_bytes,
2076                             int (drv_callback)(struct request *))
2077{
2078        return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
2079}
2080EXPORT_SYMBOL_GPL(blk_end_request_callback);
2081
2082void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2083                     struct bio *bio)
2084{
2085        /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2086           we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2087        rq->cmd_flags |= (bio->bi_rw & 3);
2088
2089        if (bio_has_data(bio)) {
2090                rq->nr_phys_segments = bio_phys_segments(q, bio);
2091                rq->buffer = bio_data(bio);
2092        }
2093        rq->current_nr_sectors = bio_cur_sectors(bio);
2094        rq->hard_cur_sectors = rq->current_nr_sectors;
2095        rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2096        rq->data_len = bio->bi_size;
2097
2098        rq->bio = rq->biotail = bio;
2099
2100        if (bio->bi_bdev)
2101                rq->rq_disk = bio->bi_bdev->bd_disk;
2102}
2103
2104/**
2105 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2106 * @q : the queue of the device being checked
2107 *
2108 * Description:
2109 *    Check if underlying low-level drivers of a device are busy.
2110 *    If the drivers want to export their busy state, they must set own
2111 *    exporting function using blk_queue_lld_busy() first.
2112 *
2113 *    Basically, this function is used only by request stacking drivers
2114 *    to stop dispatching requests to underlying devices when underlying
2115 *    devices are busy.  This behavior helps more I/O merging on the queue
2116 *    of the request stacking driver and prevents I/O throughput regression
2117 *    on burst I/O load.
2118 *
2119 * Return:
2120 *    0 - Not busy (The request stacking driver should dispatch request)
2121 *    1 - Busy (The request stacking driver should stop dispatching request)
2122 */
2123int blk_lld_busy(struct request_queue *q)
2124{
2125        if (q->lld_busy_fn)
2126                return q->lld_busy_fn(q);
2127
2128        return 0;
2129}
2130EXPORT_SYMBOL_GPL(blk_lld_busy);
2131
2132int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2133{
2134        return queue_work(kblockd_workqueue, work);
2135}
2136EXPORT_SYMBOL(kblockd_schedule_work);
2137
2138void kblockd_flush_work(struct work_struct *work)
2139{
2140        cancel_work_sync(work);
2141}
2142EXPORT_SYMBOL(kblockd_flush_work);
2143
2144int __init blk_dev_init(void)
2145{
2146        kblockd_workqueue = create_workqueue("kblockd");
2147        if (!kblockd_workqueue)
2148                panic("Failed to create kblockd\n");
2149
2150        request_cachep = kmem_cache_create("blkdev_requests",
2151                        sizeof(struct request), 0, SLAB_PANIC, NULL);
2152
2153        blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2154                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2155
2156        return 0;
2157}
2158
2159
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.