linux/block/blk-core.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 1991, 1992 Linus Torvalds
   3 * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
   4 * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
   5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
   6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
   7 *      -  July2000
   8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
   9 */
  10
  11/*
  12 * This handles all read/write requests to block devices
  13 */
  14#include <linux/kernel.h>
  15#include <linux/module.h>
  16#include <linux/backing-dev.h>
  17#include <linux/bio.h>
  18#include <linux/blkdev.h>
  19#include <linux/highmem.h>
  20#include <linux/mm.h>
  21#include <linux/kernel_stat.h>
  22#include <linux/string.h>
  23#include <linux/init.h>
  24#include <linux/completion.h>
  25#include <linux/slab.h>
  26#include <linux/swap.h>
  27#include <linux/writeback.h>
  28#include <linux/task_io_accounting_ops.h>
  29#include <linux/fault-inject.h>
  30#include <linux/list_sort.h>
  31#include <linux/delay.h>
  32#include <linux/ratelimit.h>
  33
  34#define CREATE_TRACE_POINTS
  35#include <trace/events/block.h>
  36
  37#include "blk.h"
  38#include "blk-cgroup.h"
  39
  40EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
  41EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
  42EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
  43EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
  44
  45DEFINE_IDA(blk_queue_ida);
  46
  47/*
  48 * For the allocated request tables
  49 */
  50static struct kmem_cache *request_cachep;
  51
  52/*
  53 * For queue allocation
  54 */
  55struct kmem_cache *blk_requestq_cachep;
  56
  57/*
  58 * Controlling structure to kblockd
  59 */
  60static struct workqueue_struct *kblockd_workqueue;
  61
  62static void drive_stat_acct(struct request *rq, int new_io)
  63{
  64        struct hd_struct *part;
  65        int rw = rq_data_dir(rq);
  66        int cpu;
  67
  68        if (!blk_do_io_stat(rq))
  69                return;
  70
  71        cpu = part_stat_lock();
  72
  73        if (!new_io) {
  74                part = rq->part;
  75                part_stat_inc(cpu, part, merges[rw]);
  76        } else {
  77                part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
  78                if (!hd_struct_try_get(part)) {
  79                        /*
  80                         * The partition is already being removed,
  81                         * the request will be accounted on the disk only
  82                         *
  83                         * We take a reference on disk->part0 although that
  84                         * partition will never be deleted, so we can treat
  85                         * it as any other partition.
  86                         */
  87                        part = &rq->rq_disk->part0;
  88                        hd_struct_get(part);
  89                }
  90                part_round_stats(cpu, part);
  91                part_inc_in_flight(part, rw);
  92                rq->part = part;
  93        }
  94
  95        part_stat_unlock();
  96}
  97
  98void blk_queue_congestion_threshold(struct request_queue *q)
  99{
 100        int nr;
 101
 102        nr = q->nr_requests - (q->nr_requests / 8) + 1;
 103        if (nr > q->nr_requests)
 104                nr = q->nr_requests;
 105        q->nr_congestion_on = nr;
 106
 107        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
 108        if (nr < 1)
 109                nr = 1;
 110        q->nr_congestion_off = nr;
 111}
 112
 113/**
 114 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
 115 * @bdev:       device
 116 *
 117 * Locates the passed device's request queue and returns the address of its
 118 * backing_dev_info
 119 *
 120 * Will return NULL if the request queue cannot be located.
 121 */
 122struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
 123{
 124        struct backing_dev_info *ret = NULL;
 125        struct request_queue *q = bdev_get_queue(bdev);
 126
 127        if (q)
 128                ret = &q->backing_dev_info;
 129        return ret;
 130}
 131EXPORT_SYMBOL(blk_get_backing_dev_info);
 132
 133void blk_rq_init(struct request_queue *q, struct request *rq)
 134{
 135        memset(rq, 0, sizeof(*rq));
 136
 137        INIT_LIST_HEAD(&rq->queuelist);
 138        INIT_LIST_HEAD(&rq->timeout_list);
 139        rq->cpu = -1;
 140        rq->q = q;
 141        rq->__sector = (sector_t) -1;
 142        INIT_HLIST_NODE(&rq->hash);
 143        RB_CLEAR_NODE(&rq->rb_node);
 144        rq->cmd = rq->__cmd;
 145        rq->cmd_len = BLK_MAX_CDB;
 146        rq->tag = -1;
 147        rq->ref_count = 1;
 148        rq->start_time = jiffies;
 149        set_start_time_ns(rq);
 150        rq->part = NULL;
 151}
 152EXPORT_SYMBOL(blk_rq_init);
 153
 154static void req_bio_endio(struct request *rq, struct bio *bio,
 155                          unsigned int nbytes, int error)
 156{
 157        if (error)
 158                clear_bit(BIO_UPTODATE, &bio->bi_flags);
 159        else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 160                error = -EIO;
 161
 162        if (unlikely(nbytes > bio->bi_size)) {
 163                printk(KERN_ERR "%s: want %u bytes done, %u left\n",
 164                       __func__, nbytes, bio->bi_size);
 165                nbytes = bio->bi_size;
 166        }
 167
 168        if (unlikely(rq->cmd_flags & REQ_QUIET))
 169                set_bit(BIO_QUIET, &bio->bi_flags);
 170
 171        bio->bi_size -= nbytes;
 172        bio->bi_sector += (nbytes >> 9);
 173
 174        if (bio_integrity(bio))
 175                bio_integrity_advance(bio, nbytes);
 176
 177        /* don't actually finish bio if it's part of flush sequence */
 178        if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
 179                bio_endio(bio, error);
 180}
 181
 182void blk_dump_rq_flags(struct request *rq, char *msg)
 183{
 184        int bit;
 185
 186        printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
 187                rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
 188                rq->cmd_flags);
 189
 190        printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
 191               (unsigned long long)blk_rq_pos(rq),
 192               blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
 193        printk(KERN_INFO "  bio %p, biotail %p, buffer %p, len %u\n",
 194               rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
 195
 196        if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 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
 205static void blk_delay_work(struct work_struct *work)
 206{
 207        struct request_queue *q;
 208
 209        q = container_of(work, struct request_queue, delay_work.work);
 210        spin_lock_irq(q->queue_lock);
 211        __blk_run_queue(q);
 212        spin_unlock_irq(q->queue_lock);
 213}
 214
 215/**
 216 * blk_delay_queue - restart queueing after defined interval
 217 * @q:          The &struct request_queue in question
 218 * @msecs:      Delay in msecs
 219 *
 220 * Description:
 221 *   Sometimes queueing needs to be postponed for a little while, to allow
 222 *   resources to come back. This function will make sure that queueing is
 223 *   restarted around the specified time. Queue lock must be held.
 224 */
 225void blk_delay_queue(struct request_queue *q, unsigned long msecs)
 226{
 227        if (likely(!blk_queue_dead(q)))
 228                queue_delayed_work(kblockd_workqueue, &q->delay_work,
 229                                   msecs_to_jiffies(msecs));
 230}
 231EXPORT_SYMBOL(blk_delay_queue);
 232
 233/**
 234 * blk_start_queue - restart a previously stopped queue
 235 * @q:    The &struct request_queue in question
 236 *
 237 * Description:
 238 *   blk_start_queue() will clear the stop flag on the queue, and call
 239 *   the request_fn for the queue if it was in a stopped state when
 240 *   entered. Also see blk_stop_queue(). Queue lock must be held.
 241 **/
 242void blk_start_queue(struct request_queue *q)
 243{
 244        WARN_ON(!irqs_disabled());
 245
 246        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 247        __blk_run_queue(q);
 248}
 249EXPORT_SYMBOL(blk_start_queue);
 250
 251/**
 252 * blk_stop_queue - stop a queue
 253 * @q:    The &struct request_queue in question
 254 *
 255 * Description:
 256 *   The Linux block layer assumes that a block driver will consume all
 257 *   entries on the request queue when the request_fn strategy is called.
 258 *   Often this will not happen, because of hardware limitations (queue
 259 *   depth settings). If a device driver gets a 'queue full' response,
 260 *   or if it simply chooses not to queue more I/O at one point, it can
 261 *   call this function to prevent the request_fn from being called until
 262 *   the driver has signalled it's ready to go again. This happens by calling
 263 *   blk_start_queue() to restart queue operations. Queue lock must be held.
 264 **/
 265void blk_stop_queue(struct request_queue *q)
 266{
 267        cancel_delayed_work(&q->delay_work);
 268        queue_flag_set(QUEUE_FLAG_STOPPED, q);
 269}
 270EXPORT_SYMBOL(blk_stop_queue);
 271
 272/**
 273 * blk_sync_queue - cancel any pending callbacks on a queue
 274 * @q: the queue
 275 *
 276 * Description:
 277 *     The block layer may perform asynchronous callback activity
 278 *     on a queue, such as calling the unplug function after a timeout.
 279 *     A block device may call blk_sync_queue to ensure that any
 280 *     such activity is cancelled, thus allowing it to release resources
 281 *     that the callbacks might use. The caller must already have made sure
 282 *     that its ->make_request_fn will not re-add plugging prior to calling
 283 *     this function.
 284 *
 285 *     This function does not cancel any asynchronous activity arising
 286 *     out of elevator or throttling code. That would require elevaotor_exit()
 287 *     and blkcg_exit_queue() to be called with queue lock initialized.
 288 *
 289 */
 290void blk_sync_queue(struct request_queue *q)
 291{
 292        del_timer_sync(&q->timeout);
 293        cancel_delayed_work_sync(&q->delay_work);
 294}
 295EXPORT_SYMBOL(blk_sync_queue);
 296
 297/**
 298 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
 299 * @q:  The queue to run
 300 *
 301 * Description:
 302 *    Invoke request handling on a queue if there are any pending requests.
 303 *    May be used to restart request handling after a request has completed.
 304 *    This variant runs the queue whether or not the queue has been
 305 *    stopped. Must be called with the queue lock held and interrupts
 306 *    disabled. See also @blk_run_queue.
 307 */
 308inline void __blk_run_queue_uncond(struct request_queue *q)
 309{
 310        if (unlikely(blk_queue_dead(q)))
 311                return;
 312
 313        /*
 314         * Some request_fn implementations, e.g. scsi_request_fn(), unlock
 315         * the queue lock internally. As a result multiple threads may be
 316         * running such a request function concurrently. Keep track of the
 317         * number of active request_fn invocations such that blk_drain_queue()
 318         * can wait until all these request_fn calls have finished.
 319         */
 320        q->request_fn_active++;
 321        q->request_fn(q);
 322        q->request_fn_active--;
 323}
 324
 325/**
 326 * __blk_run_queue - run a single device queue
 327 * @q:  The queue to run
 328 *
 329 * Description:
 330 *    See @blk_run_queue. This variant must be called with the queue lock
 331 *    held and interrupts disabled.
 332 */
 333void __blk_run_queue(struct request_queue *q)
 334{
 335        if (unlikely(blk_queue_stopped(q)))
 336                return;
 337
 338        __blk_run_queue_uncond(q);
 339}
 340EXPORT_SYMBOL(__blk_run_queue);
 341
 342/**
 343 * blk_run_queue_async - run a single device queue in workqueue context
 344 * @q:  The queue to run
 345 *
 346 * Description:
 347 *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
 348 *    of us. The caller must hold the queue lock.
 349 */
 350void blk_run_queue_async(struct request_queue *q)
 351{
 352        if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
 353                mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
 354}
 355EXPORT_SYMBOL(blk_run_queue_async);
 356
 357/**
 358 * blk_run_queue - run a single device queue
 359 * @q: The queue to run
 360 *
 361 * Description:
 362 *    Invoke request handling on this queue, if it has pending work to do.
 363 *    May be used to restart queueing when a request has completed.
 364 */
 365void blk_run_queue(struct request_queue *q)
 366{
 367        unsigned long flags;
 368
 369        spin_lock_irqsave(q->queue_lock, flags);
 370        __blk_run_queue(q);
 371        spin_unlock_irqrestore(q->queue_lock, flags);
 372}
 373EXPORT_SYMBOL(blk_run_queue);
 374
 375void blk_put_queue(struct request_queue *q)
 376{
 377        kobject_put(&q->kobj);
 378}
 379EXPORT_SYMBOL(blk_put_queue);
 380
 381/**
 382 * __blk_drain_queue - drain requests from request_queue
 383 * @q: queue to drain
 384 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
 385 *
 386 * Drain requests from @q.  If @drain_all is set, all requests are drained.
 387 * If not, only ELVPRIV requests are drained.  The caller is responsible
 388 * for ensuring that no new requests which need to be drained are queued.
 389 */
 390static void __blk_drain_queue(struct request_queue *q, bool drain_all)
 391        __releases(q->queue_lock)
 392        __acquires(q->queue_lock)
 393{
 394        int i;
 395
 396        lockdep_assert_held(q->queue_lock);
 397
 398        while (true) {
 399                bool drain = false;
 400
 401                /*
 402                 * The caller might be trying to drain @q before its
 403                 * elevator is initialized.
 404                 */
 405                if (q->elevator)
 406                        elv_drain_elevator(q);
 407
 408                blkcg_drain_queue(q);
 409
 410                /*
 411                 * This function might be called on a queue which failed
 412                 * driver init after queue creation or is not yet fully
 413                 * active yet.  Some drivers (e.g. fd and loop) get unhappy
 414                 * in such cases.  Kick queue iff dispatch queue has
 415                 * something on it and @q has request_fn set.
 416                 */
 417                if (!list_empty(&q->queue_head) && q->request_fn)
 418                        __blk_run_queue(q);
 419
 420                drain |= q->nr_rqs_elvpriv;
 421                drain |= q->request_fn_active;
 422
 423                /*
 424                 * Unfortunately, requests are queued at and tracked from
 425                 * multiple places and there's no single counter which can
 426                 * be drained.  Check all the queues and counters.
 427                 */
 428                if (drain_all) {
 429                        drain |= !list_empty(&q->queue_head);
 430                        for (i = 0; i < 2; i++) {
 431                                drain |= q->nr_rqs[i];
 432                                drain |= q->in_flight[i];
 433                                drain |= !list_empty(&q->flush_queue[i]);
 434                        }
 435                }
 436
 437                if (!drain)
 438                        break;
 439
 440                spin_unlock_irq(q->queue_lock);
 441
 442                msleep(10);
 443
 444                spin_lock_irq(q->queue_lock);
 445        }
 446
 447        /*
 448         * With queue marked dead, any woken up waiter will fail the
 449         * allocation path, so the wakeup chaining is lost and we're
 450         * left with hung waiters. We need to wake up those waiters.
 451         */
 452        if (q->request_fn) {
 453                struct request_list *rl;
 454
 455                blk_queue_for_each_rl(rl, q)
 456                        for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
 457                                wake_up_all(&rl->wait[i]);
 458        }
 459}
 460
 461/**
 462 * blk_queue_bypass_start - enter queue bypass mode
 463 * @q: queue of interest
 464 *
 465 * In bypass mode, only the dispatch FIFO queue of @q is used.  This
 466 * function makes @q enter bypass mode and drains all requests which were
 467 * throttled or issued before.  On return, it's guaranteed that no request
 468 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
 469 * inside queue or RCU read lock.
 470 */
 471void blk_queue_bypass_start(struct request_queue *q)
 472{
 473        bool drain;
 474
 475        spin_lock_irq(q->queue_lock);
 476        drain = !q->bypass_depth++;
 477        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 478        spin_unlock_irq(q->queue_lock);
 479
 480        if (drain) {
 481                spin_lock_irq(q->queue_lock);
 482                __blk_drain_queue(q, false);
 483                spin_unlock_irq(q->queue_lock);
 484
 485                /* ensure blk_queue_bypass() is %true inside RCU read lock */
 486                synchronize_rcu();
 487        }
 488}
 489EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
 490
 491/**
 492 * blk_queue_bypass_end - leave queue bypass mode
 493 * @q: queue of interest
 494 *
 495 * Leave bypass mode and restore the normal queueing behavior.
 496 */
 497void blk_queue_bypass_end(struct request_queue *q)
 498{
 499        spin_lock_irq(q->queue_lock);
 500        if (!--q->bypass_depth)
 501                queue_flag_clear(QUEUE_FLAG_BYPASS, q);
 502        WARN_ON_ONCE(q->bypass_depth < 0);
 503        spin_unlock_irq(q->queue_lock);
 504}
 505EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
 506
 507/**
 508 * blk_cleanup_queue - shutdown a request queue
 509 * @q: request queue to shutdown
 510 *
 511 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
 512 * put it.  All future requests will be failed immediately with -ENODEV.
 513 */
 514void blk_cleanup_queue(struct request_queue *q)
 515{
 516        spinlock_t *lock = q->queue_lock;
 517
 518        /* mark @q DYING, no new request or merges will be allowed afterwards */
 519        mutex_lock(&q->sysfs_lock);
 520        queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
 521        spin_lock_irq(lock);
 522
 523        /*
 524         * A dying queue is permanently in bypass mode till released.  Note
 525         * that, unlike blk_queue_bypass_start(), we aren't performing
 526         * synchronize_rcu() after entering bypass mode to avoid the delay
 527         * as some drivers create and destroy a lot of queues while
 528         * probing.  This is still safe because blk_release_queue() will be
 529         * called only after the queue refcnt drops to zero and nothing,
 530         * RCU or not, would be traversing the queue by then.
 531         */
 532        q->bypass_depth++;
 533        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 534
 535        queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 536        queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 537        queue_flag_set(QUEUE_FLAG_DYING, q);
 538        spin_unlock_irq(lock);
 539        mutex_unlock(&q->sysfs_lock);
 540
 541        /*
 542         * Drain all requests queued before DYING marking. Set DEAD flag to
 543         * prevent that q->request_fn() gets invoked after draining finished.
 544         */
 545        spin_lock_irq(lock);
 546        __blk_drain_queue(q, true);
 547        queue_flag_set(QUEUE_FLAG_DEAD, q);
 548        spin_unlock_irq(lock);
 549
 550        /* @q won't process any more request, flush async actions */
 551        del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
 552        blk_sync_queue(q);
 553
 554        spin_lock_irq(lock);
 555        if (q->queue_lock != &q->__queue_lock)
 556                q->queue_lock = &q->__queue_lock;
 557        spin_unlock_irq(lock);
 558
 559        /* @q is and will stay empty, shutdown and put */
 560        blk_put_queue(q);
 561}
 562EXPORT_SYMBOL(blk_cleanup_queue);
 563
 564int blk_init_rl(struct request_list *rl, struct request_queue *q,
 565                gfp_t gfp_mask)
 566{
 567        if (unlikely(rl->rq_pool))
 568                return 0;
 569
 570        rl->q = q;
 571        rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
 572        rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
 573        init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
 574        init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
 575
 576        rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
 577                                          mempool_free_slab, request_cachep,
 578                                          gfp_mask, q->node);
 579        if (!rl->rq_pool)
 580                return -ENOMEM;
 581
 582        return 0;
 583}
 584
 585void blk_exit_rl(struct request_list *rl)
 586{
 587        if (rl->rq_pool)
 588                mempool_destroy(rl->rq_pool);
 589}
 590
 591struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 592{
 593        return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
 594}
 595EXPORT_SYMBOL(blk_alloc_queue);
 596
 597struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 598{
 599        struct request_queue *q;
 600        int err;
 601
 602        q = kmem_cache_alloc_node(blk_requestq_cachep,
 603                                gfp_mask | __GFP_ZERO, node_id);
 604        if (!q)
 605                return NULL;
 606
 607        q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
 608        if (q->id < 0)
 609                goto fail_q;
 610
 611        q->backing_dev_info.ra_pages =
 612                        (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 613        q->backing_dev_info.state = 0;
 614        q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
 615        q->backing_dev_info.name = "block";
 616        q->node = node_id;
 617
 618        err = bdi_init(&q->backing_dev_info);
 619        if (err)
 620                goto fail_id;
 621
 622        setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
 623                    laptop_mode_timer_fn, (unsigned long) q);
 624        setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
 625        INIT_LIST_HEAD(&q->queue_head);
 626        INIT_LIST_HEAD(&q->timeout_list);
 627        INIT_LIST_HEAD(&q->icq_list);
 628#ifdef CONFIG_BLK_CGROUP
 629        INIT_LIST_HEAD(&q->blkg_list);
 630#endif
 631        INIT_LIST_HEAD(&q->flush_queue[0]);
 632        INIT_LIST_HEAD(&q->flush_queue[1]);
 633        INIT_LIST_HEAD(&q->flush_data_in_flight);
 634        INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
 635
 636        kobject_init(&q->kobj, &blk_queue_ktype);
 637
 638        mutex_init(&q->sysfs_lock);
 639        spin_lock_init(&q->__queue_lock);
 640
 641        /*
 642         * By default initialize queue_lock to internal lock and driver can
 643         * override it later if need be.
 644         */
 645        q->queue_lock = &q->__queue_lock;
 646
 647        /*
 648         * A queue starts its life with bypass turned on to avoid
 649         * unnecessary bypass on/off overhead and nasty surprises during
 650         * init.  The initial bypass will be finished when the queue is
 651         * registered by blk_register_queue().
 652         */
 653        q->bypass_depth = 1;
 654        __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
 655
 656        if (blkcg_init_queue(q))
 657                goto fail_id;
 658
 659        return q;
 660
 661fail_id:
 662        ida_simple_remove(&blk_queue_ida, q->id);
 663fail_q:
 664        kmem_cache_free(blk_requestq_cachep, q);
 665        return NULL;
 666}
 667EXPORT_SYMBOL(blk_alloc_queue_node);
 668
 669/**
 670 * blk_init_queue  - prepare a request queue for use with a block device
 671 * @rfn:  The function to be called to process requests that have been
 672 *        placed on the queue.
 673 * @lock: Request queue spin lock
 674 *
 675 * Description:
 676 *    If a block device wishes to use the standard request handling procedures,
 677 *    which sorts requests and coalesces adjacent requests, then it must
 678 *    call blk_init_queue().  The function @rfn will be called when there
 679 *    are requests on the queue that need to be processed.  If the device
 680 *    supports plugging, then @rfn may not be called immediately when requests
 681 *    are available on the queue, but may be called at some time later instead.
 682 *    Plugged queues are generally unplugged when a buffer belonging to one
 683 *    of the requests on the queue is needed, or due to memory pressure.
 684 *
 685 *    @rfn is not required, or even expected, to remove all requests off the
 686 *    queue, but only as many as it can handle at a time.  If it does leave
 687 *    requests on the queue, it is responsible for arranging that the requests
 688 *    get dealt with eventually.
 689 *
 690 *    The queue spin lock must be held while manipulating the requests on the
 691 *    request queue; this lock will be taken also from interrupt context, so irq
 692 *    disabling is needed for it.
 693 *
 694 *    Function returns a pointer to the initialized request queue, or %NULL if
 695 *    it didn't succeed.
 696 *
 697 * Note:
 698 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
 699 *    when the block device is deactivated (such as at module unload).
 700 **/
 701
 702struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
 703{
 704        return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
 705}
 706EXPORT_SYMBOL(blk_init_queue);
 707
 708struct request_queue *
 709blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
 710{
 711        struct request_queue *uninit_q, *q;
 712
 713        uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
 714        if (!uninit_q)
 715                return NULL;
 716
 717        q = blk_init_allocated_queue(uninit_q, rfn, lock);
 718        if (!q)
 719                blk_cleanup_queue(uninit_q);
 720
 721        return q;
 722}
 723EXPORT_SYMBOL(blk_init_queue_node);
 724
 725struct request_queue *
 726blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
 727                         spinlock_t *lock)
 728{
 729        if (!q)
 730                return NULL;
 731
 732        if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
 733                return NULL;
 734
 735        q->request_fn           = rfn;
 736        q->prep_rq_fn           = NULL;
 737        q->unprep_rq_fn         = NULL;
 738        q->queue_flags          |= QUEUE_FLAG_DEFAULT;
 739
 740        /* Override internal queue lock with supplied lock pointer */
 741        if (lock)
 742                q->queue_lock           = lock;
 743
 744        /*
 745         * This also sets hw/phys segments, boundary and size
 746         */
 747        blk_queue_make_request(q, blk_queue_bio);
 748
 749        q->sg_reserved_size = INT_MAX;
 750
 751        /* init elevator */
 752        if (elevator_init(q, NULL))
 753                return NULL;
 754        return q;
 755}
 756EXPORT_SYMBOL(blk_init_allocated_queue);
 757
 758bool blk_get_queue(struct request_queue *q)
 759{
 760        if (likely(!blk_queue_dying(q))) {
 761                __blk_get_queue(q);
 762                return true;
 763        }
 764
 765        return false;
 766}
 767EXPORT_SYMBOL(blk_get_queue);
 768
 769static inline void blk_free_request(struct request_list *rl, struct request *rq)
 770{
 771        if (rq->cmd_flags & REQ_ELVPRIV) {
 772                elv_put_request(rl->q, rq);
 773                if (rq->elv.icq)
 774                        put_io_context(rq->elv.icq->ioc);
 775        }
 776
 777        mempool_free(rq, rl->rq_pool);
 778}
 779
 780/*
 781 * ioc_batching returns true if the ioc is a valid batching request and
 782 * should be given priority access to a request.
 783 */
 784static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
 785{
 786        if (!ioc)
 787                return 0;
 788
 789        /*
 790         * Make sure the process is able to allocate at least 1 request
 791         * even if the batch times out, otherwise we could theoretically
 792         * lose wakeups.
 793         */
 794        return ioc->nr_batch_requests == q->nr_batching ||
 795                (ioc->nr_batch_requests > 0
 796                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
 797}
 798
 799/*
 800 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
 801 * will cause the process to be a "batcher" on all queues in the system. This
 802 * is the behaviour we want though - once it gets a wakeup it should be given
 803 * a nice run.
 804 */
 805static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
 806{
 807        if (!ioc || ioc_batching(q, ioc))
 808                return;
 809
 810        ioc->nr_batch_requests = q->nr_batching;
 811        ioc->last_waited = jiffies;
 812}
 813
 814static void __freed_request(struct request_list *rl, int sync)
 815{
 816        struct request_queue *q = rl->q;
 817
 818        /*
 819         * bdi isn't aware of blkcg yet.  As all async IOs end up root
 820         * blkcg anyway, just use root blkcg state.
 821         */
 822        if (rl == &q->root_rl &&
 823            rl->count[sync] < queue_congestion_off_threshold(q))
 824                blk_clear_queue_congested(q, sync);
 825
 826        if (rl->count[sync] + 1 <= q->nr_requests) {
 827                if (waitqueue_active(&rl->wait[sync]))
 828                        wake_up(&rl->wait[sync]);
 829
 830                blk_clear_rl_full(rl, sync);
 831        }
 832}
 833
 834/*
 835 * A request has just been released.  Account for it, update the full and
 836 * congestion status, wake up any waiters.   Called under q->queue_lock.
 837 */
 838static void freed_request(struct request_list *rl, unsigned int flags)
 839{
 840        struct request_queue *q = rl->q;
 841        int sync = rw_is_sync(flags);
 842
 843        q->nr_rqs[sync]--;
 844        rl->count[sync]--;
 845        if (flags & REQ_ELVPRIV)
 846                q->nr_rqs_elvpriv--;
 847
 848        __freed_request(rl, sync);
 849
 850        if (unlikely(rl->starved[sync ^ 1]))
 851                __freed_request(rl, sync ^ 1);
 852}
 853
 854/*
 855 * Determine if elevator data should be initialized when allocating the
 856 * request associated with @bio.
 857 */
 858static bool blk_rq_should_init_elevator(struct bio *bio)
 859{
 860        if (!bio)
 861                return true;
 862
 863        /*
 864         * Flush requests do not use the elevator so skip initialization.
 865         * This allows a request to share the flush and elevator data.
 866         */
 867        if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
 868                return false;
 869
 870        return true;
 871}
 872
 873/**
 874 * rq_ioc - determine io_context for request allocation
 875 * @bio: request being allocated is for this bio (can be %NULL)
 876 *
 877 * Determine io_context to use for request allocation for @bio.  May return
 878 * %NULL if %current->io_context doesn't exist.
 879 */
 880static struct io_context *rq_ioc(struct bio *bio)
 881{
 882#ifdef CONFIG_BLK_CGROUP
 883        if (bio && bio->bi_ioc)
 884                return bio->bi_ioc;
 885#endif
 886        return current->io_context;
 887}
 888
 889/**
 890 * __get_request - get a free request
 891 * @rl: request list to allocate from
 892 * @rw_flags: RW and SYNC flags
 893 * @bio: bio to allocate request for (can be %NULL)
 894 * @gfp_mask: allocation mask
 895 *
 896 * Get a free request from @q.  This function may fail under memory
 897 * pressure or if @q is dead.
 898 *
 899 * Must be callled with @q->queue_lock held and,
 900 * Returns %NULL on failure, with @q->queue_lock held.
 901 * Returns !%NULL on success, with @q->queue_lock *not held*.
 902 */
 903static struct request *__get_request(struct request_list *rl, int rw_flags,
 904                                     struct bio *bio, gfp_t gfp_mask)
 905{
 906        struct request_queue *q = rl->q;
 907        struct request *rq;
 908        struct elevator_type *et = q->elevator->type;
 909        struct io_context *ioc = rq_ioc(bio);
 910        struct io_cq *icq = NULL;
 911        const bool is_sync = rw_is_sync(rw_flags) != 0;
 912        int may_queue;
 913
 914        if (unlikely(blk_queue_dying(q)))
 915                return NULL;
 916
 917        may_queue = elv_may_queue(q, rw_flags);
 918        if (may_queue == ELV_MQUEUE_NO)
 919                goto rq_starved;
 920
 921        if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
 922                if (rl->count[is_sync]+1 >= q->nr_requests) {
 923                        /*
 924                         * The queue will fill after this allocation, so set
 925                         * it as full, and mark this process as "batching".
 926                         * This process will be allowed to complete a batch of
 927                         * requests, others will be blocked.
 928                         */
 929                        if (!blk_rl_full(rl, is_sync)) {
 930                                ioc_set_batching(q, ioc);
 931                                blk_set_rl_full(rl, is_sync);
 932                        } else {
 933                                if (may_queue != ELV_MQUEUE_MUST
 934                                                && !ioc_batching(q, ioc)) {
 935                                        /*
 936                                         * The queue is full and the allocating
 937                                         * process is not a "batcher", and not
 938                                         * exempted by the IO scheduler
 939                                         */
 940                                        return NULL;
 941                                }
 942                        }
 943                }
 944                /*
 945                 * bdi isn't aware of blkcg yet.  As all async IOs end up
 946                 * root blkcg anyway, just use root blkcg state.
 947                 */
 948                if (rl == &q->root_rl)
 949                        blk_set_queue_congested(q, is_sync);
 950        }
 951
 952        /*
 953         * Only allow batching queuers to allocate up to 50% over the defined
 954         * limit of requests, otherwise we could have thousands of requests
 955         * allocated with any setting of ->nr_requests
 956         */
 957        if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
 958                return NULL;
 959
 960        q->nr_rqs[is_sync]++;
 961        rl->count[is_sync]++;
 962        rl->starved[is_sync] = 0;
 963
 964        /*
 965         * Decide whether the new request will be managed by elevator.  If
 966         * so, mark @rw_flags and increment elvpriv.  Non-zero elvpriv will
 967         * prevent the current elevator from being destroyed until the new
 968         * request is freed.  This guarantees icq's won't be destroyed and
 969         * makes creating new ones safe.
 970         *
 971         * Also, lookup icq while holding queue_lock.  If it doesn't exist,
 972         * it will be created after releasing queue_lock.
 973         */
 974        if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
 975                rw_flags |= REQ_ELVPRIV;
 976                q->nr_rqs_elvpriv++;
 977                if (et->icq_cache && ioc)
 978                        icq = ioc_lookup_icq(ioc, q);
 979        }
 980
 981        if (blk_queue_io_stat(q))
 982                rw_flags |= REQ_IO_STAT;
 983        spin_unlock_irq(q->queue_lock);
 984
 985        /* allocate and init request */
 986        rq = mempool_alloc(rl->rq_pool, gfp_mask);
 987        if (!rq)
 988                goto fail_alloc;
 989
 990        blk_rq_init(q, rq);
 991        blk_rq_set_rl(rq, rl);
 992        rq->cmd_flags = rw_flags | REQ_ALLOCED;
 993
 994        /* init elvpriv */
 995        if (rw_flags & REQ_ELVPRIV) {
 996                if (unlikely(et->icq_cache && !icq)) {
 997                        if (ioc)
 998                                icq = ioc_create_icq(ioc, q, gfp_mask);
 999                        if (!icq)
1000                                goto fail_elvpriv;
1001                }
1002
1003                rq->elv.icq = icq;
1004                if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1005                        goto fail_elvpriv;
1006
1007                /* @rq->elv.icq holds io_context until @rq is freed */
1008                if (icq)
1009                        get_io_context(icq->ioc);
1010        }
1011out:
1012        /*
1013         * ioc may be NULL here, and ioc_batching will be false. That's
1014         * OK, if the queue is under the request limit then requests need
1015         * not count toward the nr_batch_requests limit. There will always
1016         * be some limit enforced by BLK_BATCH_TIME.
1017         */
1018        if (ioc_batching(q, ioc))
1019                ioc->nr_batch_requests--;
1020
1021        trace_block_getrq(q, bio, rw_flags & 1);
1022        return rq;
1023
1024fail_elvpriv:
1025        /*
1026         * elvpriv init failed.  ioc, icq and elvpriv aren't mempool backed
1027         * and may fail indefinitely under memory pressure and thus
1028         * shouldn't stall IO.  Treat this request as !elvpriv.  This will
1029         * disturb iosched and blkcg but weird is bettern than dead.
1030         */
1031        printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1032                           dev_name(q->backing_dev_info.dev));
1033
1034        rq->cmd_flags &= ~REQ_ELVPRIV;
1035        rq->elv.icq = NULL;
1036
1037        spin_lock_irq(q->queue_lock);
1038        q->nr_rqs_elvpriv--;
1039        spin_unlock_irq(q->queue_lock);
1040        goto out;
1041
1042fail_alloc:
1043        /*
1044         * Allocation failed presumably due to memory. Undo anything we
1045         * might have messed up.
1046         *
1047         * Allocating task should really be put onto the front of the wait
1048         * queue, but this is pretty rare.
1049         */
1050        spin_lock_irq(q->queue_lock);
1051        freed_request(rl, rw_flags);
1052
1053        /*
1054         * in the very unlikely event that allocation failed and no
1055         * requests for this direction was pending, mark us starved so that
1056         * freeing of a request in the other direction will notice
1057         * us. another possible fix would be to split the rq mempool into
1058         * READ and WRITE
1059         */
1060rq_starved:
1061        if (unlikely(rl->count[is_sync] == 0))
1062                rl->starved[is_sync] = 1;
1063        return NULL;
1064}
1065
1066/**
1067 * get_request - get a free request
1068 * @q: request_queue to allocate request from
1069 * @rw_flags: RW and SYNC flags
1070 * @bio: bio to allocate request for (can be %NULL)
1071 * @gfp_mask: allocation mask
1072 *
1073 * Get a free request from @q.  If %__GFP_WAIT is set in @gfp_mask, this
1074 * function keeps retrying under memory pressure and fails iff @q is dead.
1075 *
1076 * Must be callled with @q->queue_lock held and,
1077 * Returns %NULL on failure, with @q->queue_lock held.
1078 * Returns !%NULL on success, with @q->queue_lock *not held*.
1079 */
1080static struct request *get_request(struct request_queue *q, int rw_flags,
1081                                   struct bio *bio, gfp_t gfp_mask)
1082{
1083        const bool is_sync = rw_is_sync(rw_flags) != 0;
1084        DEFINE_WAIT(wait);
1085        struct request_list *rl;
1086        struct request *rq;
1087
1088        rl = blk_get_rl(q, bio);        /* transferred to @rq on success */
1089retry:
1090        rq = __get_request(rl, rw_flags, bio, gfp_mask);
1091        if (rq)
1092                return rq;
1093
1094        if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1095                blk_put_rl(rl);
1096                return NULL;
1097        }
1098
1099        /* wait on @rl and retry */
1100        prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1101                                  TASK_UNINTERRUPTIBLE);
1102
1103        trace_block_sleeprq(q, bio, rw_flags & 1);
1104
1105        spin_unlock_irq(q->queue_lock);
1106        io_schedule();
1107
1108        /*
1109         * After sleeping, we become a "batching" process and will be able
1110         * to allocate at least one request, and up to a big batch of them
1111         * for a small period time.  See ioc_batching, ioc_set_batching
1112         */
1113        ioc_set_batching(q, current->io_context);
1114
1115        spin_lock_irq(q->queue_lock);
1116        finish_wait(&rl->wait[is_sync], &wait);
1117
1118        goto retry;
1119}
1120
1121struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1122{
1123        struct request *rq;
1124
1125        BUG_ON(rw != READ && rw != WRITE);
1126
1127        /* create ioc upfront */
1128        create_io_context(gfp_mask, q->node);
1129
1130        spin_lock_irq(q->queue_lock);
1131        rq = get_request(q, rw, NULL, gfp_mask);
1132        if (!rq)
1133                spin_unlock_irq(q->queue_lock);
1134        /* q->queue_lock is unlocked at this point */
1135
1136        return rq;
1137}
1138EXPORT_SYMBOL(blk_get_request);
1139
1140/**
1141 * blk_make_request - given a bio, allocate a corresponding struct request.
1142 * @q: target request queue
1143 * @bio:  The bio describing the memory mappings that will be submitted for IO.
1144 *        It may be a chained-bio properly constructed by block/bio layer.
1145 * @gfp_mask: gfp flags to be used for memory allocation
1146 *
1147 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1148 * type commands. Where the struct request needs to be farther initialized by
1149 * the caller. It is passed a &struct bio, which describes the memory info of
1150 * the I/O transfer.
1151 *
1152 * The caller of blk_make_request must make sure that bi_io_vec
1153 * are set to describe the memory buffers. That bio_data_dir() will return
1154 * the needed direction of the request. (And all bio's in the passed bio-chain
1155 * are properly set accordingly)
1156 *
1157 * If called under none-sleepable conditions, mapped bio buffers must not
1158 * need bouncing, by calling the appropriate masked or flagged allocator,
1159 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1160 * BUG.
1161 *
1162 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1163 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1164 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1165 * completion of a bio that hasn't been submitted yet, thus resulting in a
1166 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1167 * of bio_alloc(), as that avoids the mempool deadlock.
1168 * If possible a big IO should be split into smaller parts when allocation
1169 * fails. Partial allocation should not be an error, or you risk a live-lock.
1170 */
1171struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1172                                 gfp_t gfp_mask)
1173{
1174        struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1175
1176        if (unlikely(!rq))
1177                return ERR_PTR(-ENOMEM);
1178
1179        for_each_bio(bio) {
1180                struct bio *bounce_bio = bio;
1181                int ret;
1182
1183                blk_queue_bounce(q, &bounce_bio);
1184                ret = blk_rq_append_bio(q, rq, bounce_bio);
1185                if (unlikely(ret)) {
1186                        blk_put_request(rq);
1187                        return ERR_PTR(ret);
1188                }
1189        }
1190
1191        return rq;
1192}
1193EXPORT_SYMBOL(blk_make_request);
1194
1195/**
1196 * blk_requeue_request - put a request back on queue
1197 * @q:          request queue where request should be inserted
1198 * @rq:         request to be inserted
1199 *
1200 * Description:
1201 *    Drivers often keep queueing requests until the hardware cannot accept
1202 *    more, when that condition happens we need to put the request back
1203 *    on the queue. Must be called with queue lock held.
1204 */
1205void blk_requeue_request(struct request_queue *q, struct request *rq)
1206{
1207        blk_delete_timer(rq);
1208        blk_clear_rq_complete(rq);
1209        trace_block_rq_requeue(q, rq);
1210
1211        if (blk_rq_tagged(rq))
1212                blk_queue_end_tag(q, rq);
1213
1214        BUG_ON(blk_queued_rq(rq));
1215
1216        elv_requeue_request(q, rq);
1217}
1218EXPORT_SYMBOL(blk_requeue_request);
1219
1220static void add_acct_request(struct request_queue *q, struct request *rq,
1221                             int where)
1222{
1223        drive_stat_acct(rq, 1);
1224        __elv_add_request(q, rq, where);
1225}
1226
1227static void part_round_stats_single(int cpu, struct hd_struct *part,
1228                                    unsigned long now)
1229{
1230        if (now == part->stamp)
1231                return;
1232
1233        if (part_in_flight(part)) {
1234                __part_stat_add(cpu, part, time_in_queue,
1235                                part_in_flight(part) * (now - part->stamp));
1236                __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1237        }
1238        part->stamp = now;
1239}
1240
1241/**
1242 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1243 * @cpu: cpu number for stats access
1244 * @part: target partition
1245 *
1246 * The average IO queue length and utilisation statistics are maintained
1247 * by observing the current state of the queue length and the amount of
1248 * time it has been in this state for.
1249 *
1250 * Normally, that accounting is done on IO completion, but that can result
1251 * in more than a second's worth of IO being accounted for within any one
1252 * second, leading to >100% utilisation.  To deal with that, we call this
1253 * function to do a round-off before returning the results when reading
1254 * /proc/diskstats.  This accounts immediately for all queue usage up to
1255 * the current jiffies and restarts the counters again.
1256 */
1257void part_round_stats(int cpu, struct hd_struct *part)
1258{
1259        unsigned long now = jiffies;
1260
1261        if (part->partno)
1262                part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1263        part_round_stats_single(cpu, part, now);
1264}
1265EXPORT_SYMBOL_GPL(part_round_stats);
1266
1267/*
1268 * queue lock must be held
1269 */
1270void __blk_put_request(struct request_queue *q, struct request *req)
1271{
1272        if (unlikely(!q))
1273                return;
1274        if (unlikely(--req->ref_count))
1275                return;
1276
1277        elv_completed_request(q, req);
1278
1279        /* this is a bio leak */
1280        WARN_ON(req->bio != NULL);
1281
1282        /*
1283         * Request may not have originated from ll_rw_blk. if not,
1284         * it didn't come out of our reserved rq pools
1285         */
1286        if (req->cmd_flags & REQ_ALLOCED) {
1287                unsigned int flags = req->cmd_flags;
1288                struct request_list *rl = blk_rq_rl(req);
1289
1290                BUG_ON(!list_empty(&req->queuelist));
1291                BUG_ON(!hlist_unhashed(&req->hash));
1292
1293                blk_free_request(rl, req);
1294                freed_request(rl, flags);
1295                blk_put_rl(rl);
1296        }
1297}
1298EXPORT_SYMBOL_GPL(__blk_put_request);
1299
1300void blk_put_request(struct request *req)
1301{
1302        unsigned long flags;
1303        struct request_queue *q = req->q;
1304
1305        spin_lock_irqsave(q->queue_lock, flags);
1306        __blk_put_request(q, req);
1307        spin_unlock_irqrestore(q->queue_lock, flags);
1308}
1309EXPORT_SYMBOL(blk_put_request);
1310
1311/**
1312 * blk_add_request_payload - add a payload to a request
1313 * @rq: request to update
1314 * @page: page backing the payload
1315 * @len: length of the payload.
1316 *
1317 * This allows to later add a payload to an already submitted request by
1318 * a block driver.  The driver needs to take care of freeing the payload
1319 * itself.
1320 *
1321 * Note that this is a quite horrible hack and nothing but handling of
1322 * discard requests should ever use it.
1323 */
1324void blk_add_request_payload(struct request *rq, struct page *page,
1325                unsigned int len)
1326{
1327        struct bio *bio = rq->bio;
1328
1329        bio->bi_io_vec->bv_page = page;
1330        bio->bi_io_vec->bv_offset = 0;
1331        bio->bi_io_vec->bv_len = len;
1332
1333        bio->bi_size = len;
1334        bio->bi_vcnt = 1;
1335        bio->bi_phys_segments = 1;
1336
1337        rq->__data_len = rq->resid_len = len;
1338        rq->nr_phys_segments = 1;
1339        rq->buffer = bio_data(bio);
1340}
1341EXPORT_SYMBOL_GPL(blk_add_request_payload);
1342
1343static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1344                                   struct bio *bio)
1345{
1346        const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1347
1348        if (!ll_back_merge_fn(q, req, bio))
1349                return false;
1350
1351        trace_block_bio_backmerge(q, bio);
1352
1353        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1354                blk_rq_set_mixed_merge(req);
1355
1356        req->biotail->bi_next = bio;
1357        req->biotail = bio;
1358        req->__data_len += bio->bi_size;
1359        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1360
1361        drive_stat_acct(req, 0);
1362        return true;
1363}
1364
1365static bool bio_attempt_front_merge(struct request_queue *q,
1366                                    struct request *req, struct bio *bio)
1367{
1368        const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1369
1370        if (!ll_front_merge_fn(q, req, bio))
1371                return false;
1372
1373        trace_block_bio_frontmerge(q, bio);
1374
1375        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1376                blk_rq_set_mixed_merge(req);
1377
1378        bio->bi_next = req->bio;
1379        req->bio = bio;
1380
1381        /*
1382         * may not be valid. if the low level driver said
1383         * it didn't need a bounce buffer then it better
1384         * not touch req->buffer either...
1385         */
1386        req->buffer = bio_data(bio);
1387        req->__sector = bio->bi_sector;
1388        req->__data_len += bio->bi_size;
1389        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1390
1391        drive_stat_acct(req, 0);
1392        return true;
1393}
1394
1395/**
1396 * attempt_plug_merge - try to merge with %current's plugged list
1397 * @q: request_queue new bio is being queued at
1398 * @bio: new bio being queued
1399 * @request_count: out parameter for number of traversed plugged requests
1400 *
1401 * Determine whether @bio being queued on @q can be merged with a request
1402 * on %current's plugged list.  Returns %true if merge was successful,
1403 * otherwise %false.
1404 *
1405 * Plugging coalesces IOs from the same issuer for the same purpose without
1406 * going through @q->queue_lock.  As such it's more of an issuing mechanism
1407 * than scheduling, and the request, while may have elvpriv data, is not
1408 * added on the elevator at this point.  In addition, we don't have
1409 * reliable access to the elevator outside queue lock.  Only check basic
1410 * merging parameters without querying the elevator.
1411 */
1412static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1413                               unsigned int *request_count)
1414{
1415        struct blk_plug *plug;
1416        struct request *rq;
1417        bool ret = false;
1418
1419        plug = current->plug;
1420        if (!plug)
1421                goto out;
1422        *request_count = 0;
1423
1424        list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1425                int el_ret;
1426
1427                if (rq->q == q)
1428                        (*request_count)++;
1429
1430                if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1431                        continue;
1432
1433                el_ret = blk_try_merge(rq, bio);
1434                if (el_ret == ELEVATOR_BACK_MERGE) {
1435                        ret = bio_attempt_back_merge(q, rq, bio);
1436                        if (ret)
1437                                break;
1438                } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1439                        ret = bio_attempt_front_merge(q, rq, bio);
1440                        if (ret)
1441                                break;
1442                }
1443        }
1444out:
1445        return ret;
1446}
1447
1448void init_request_from_bio(struct request *req, struct bio *bio)
1449{
1450        req->cmd_type = REQ_TYPE_FS;
1451
1452        req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1453        if (bio->bi_rw & REQ_RAHEAD)
1454                req->cmd_flags |= REQ_FAILFAST_MASK;
1455
1456        req->errors = 0;
1457        req->__sector = bio->bi_sector;
1458        req->ioprio = bio_prio(bio);
1459        blk_rq_bio_prep(req->q, req, bio);
1460}
1461
1462void blk_queue_bio(struct request_queue *q, struct bio *bio)
1463{
1464        const bool sync = !!(bio->bi_rw & REQ_SYNC);
1465        struct blk_plug *plug;
1466        int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1467        struct request *req;
1468        unsigned int request_count = 0;
1469
1470        /*
1471         * low level driver can indicate that it wants pages above a
1472         * certain limit bounced to low memory (ie for highmem, or even
1473         * ISA dma in theory)
1474         */
1475        blk_queue_bounce(q, &bio);
1476
1477        if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1478                spin_lock_irq(q->queue_lock);
1479                where = ELEVATOR_INSERT_FLUSH;
1480                goto get_rq;
1481        }
1482
1483        /*
1484         * Check if we can merge with the plugged list before grabbing
1485         * any locks.
1486         */
1487        if (attempt_plug_merge(q, bio, &request_count))
1488                return;
1489
1490        spin_lock_irq(q->queue_lock);
1491
1492        el_ret = elv_merge(q, &req, bio);
1493        if (el_ret == ELEVATOR_BACK_MERGE) {
1494                if (bio_attempt_back_merge(q, req, bio)) {
1495                        elv_bio_merged(q, req, bio);
1496                        if (!attempt_back_merge(q, req))
1497                                elv_merged_request(q, req, el_ret);
1498                        goto out_unlock;
1499                }
1500        } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1501                if (bio_attempt_front_merge(q, req, bio)) {
1502                        elv_bio_merged(q, req, bio);
1503                        if (!attempt_front_merge(q, req))
1504                                elv_merged_request(q, req, el_ret);
1505                        goto out_unlock;
1506                }
1507        }
1508
1509get_rq:
1510        /*
1511         * This sync check and mask will be re-done in init_request_from_bio(),
1512         * but we need to set it earlier to expose the sync flag to the
1513         * rq allocator and io schedulers.
1514         */
1515        rw_flags = bio_data_dir(bio);
1516        if (sync)
1517                rw_flags |= REQ_SYNC;
1518
1519        /*
1520         * Grab a free request. This is might sleep but can not fail.
1521         * Returns with the queue unlocked.
1522         */
1523        req = get_request(q, rw_flags, bio, GFP_NOIO);
1524        if (unlikely(!req)) {
1525                bio_endio(bio, -ENODEV);        /* @q is dead */
1526                goto out_unlock;
1527        }
1528
1529        /*
1530         * After dropping the lock and possibly sleeping here, our request
1531         * may now be mergeable after it had proven unmergeable (above).
1532         * We don't worry about that case for efficiency. It won't happen
1533         * often, and the elevators are able to handle it.
1534         */
1535        init_request_from_bio(req, bio);
1536
1537        if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1538                req->cpu = raw_smp_processor_id();
1539
1540        plug = current->plug;
1541        if (plug) {
1542                /*
1543                 * If this is the first request added after a plug, fire
1544                 * of a plug trace. If others have been added before, check
1545                 * if we have multiple devices in this plug. If so, make a
1546                 * note to sort the list before dispatch.
1547                 */
1548                if (list_empty(&plug->list))
1549                        trace_block_plug(q);
1550                else {
1551                        if (!plug->should_sort) {
1552                                struct request *__rq;
1553
1554                                __rq = list_entry_rq(plug->list.prev);
1555                                if (__rq->q != q)
1556                                        plug->should_sort = 1;
1557                        }
1558                        if (request_count >= BLK_MAX_REQUEST_COUNT) {
1559                                blk_flush_plug_list(plug, false);
1560                                trace_block_plug(q);
1561                        }
1562                }
1563                list_add_tail(&req->queuelist, &plug->list);
1564                drive_stat_acct(req, 1);
1565        } else {
1566                spin_lock_irq(q->queue_lock);
1567                add_acct_request(q, req, where);
1568                __blk_run_queue(q);
1569out_unlock:
1570                spin_unlock_irq(q->queue_lock);
1571        }
1572}
1573EXPORT_SYMBOL_GPL(blk_queue_bio);       /* for device mapper only */
1574
1575/*
1576 * If bio->bi_dev is a partition, remap the location
1577 */
1578static inline void blk_partition_remap(struct bio *bio)
1579{
1580        struct block_device *bdev = bio->bi_bdev;
1581
1582        if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1583                struct hd_struct *p = bdev->bd_part;
1584
1585                bio->bi_sector += p->start_sect;
1586                bio->bi_bdev = bdev->bd_contains;
1587
1588                trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1589                                      bdev->bd_dev,
1590                                      bio->bi_sector - p->start_sect);
1591        }
1592}
1593
1594static void handle_bad_sector(struct bio *bio)
1595{
1596        char b[BDEVNAME_SIZE];
1597
1598        printk(KERN_INFO "attempt to access beyond end of device\n");
1599        printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1600                        bdevname(bio->bi_bdev, b),
1601                        bio->bi_rw,
1602                        (unsigned long long)bio->bi_sector + bio_sectors(bio),
1603                        (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1604
1605        set_bit(BIO_EOF, &bio->bi_flags);
1606}
1607
1608#ifdef CONFIG_FAIL_MAKE_REQUEST
1609
1610static DECLARE_FAULT_ATTR(fail_make_request);
1611
1612static int __init setup_fail_make_request(char *str)
1613{
1614        return setup_fault_attr(&fail_make_request, str);
1615}
1616__setup("fail_make_request=", setup_fail_make_request);
1617
1618static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1619{
1620        return part->make_it_fail && should_fail(&fail_make_request, bytes);
1621}
1622
1623static int __init fail_make_request_debugfs(void)
1624{
1625        struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1626                                                NULL, &fail_make_request);
1627
1628        return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1629}
1630
1631late_initcall(fail_make_request_debugfs);
1632
1633#else /* CONFIG_FAIL_MAKE_REQUEST */
1634
1635static inline bool should_fail_request(struct hd_struct *part,
1636                                        unsigned int bytes)
1637{
1638        return false;
1639}
1640
1641#endif /* CONFIG_FAIL_MAKE_REQUEST */
1642
1643/*
1644 * Check whether this bio extends beyond the end of the device.
1645 */
1646static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1647{
1648        sector_t maxsector;
1649
1650        if (!nr_sectors)
1651                return 0;
1652
1653        /* Test device or partition size, when known. */
1654        maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1655        if (maxsector) {
1656                sector_t sector = bio->bi_sector;
1657
1658                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1659                        /*
1660                         * This may well happen - the kernel calls bread()
1661                         * without checking the size of the device, e.g., when
1662                         * mounting a device.
1663                         */
1664                        handle_bad_sector(bio);
1665                        return 1;
1666                }
1667        }
1668
1669        return 0;
1670}
1671
1672static noinline_for_stack bool
1673generic_make_request_checks(struct bio *bio)
1674{
1675        struct request_queue *q;
1676        int nr_sectors = bio_sectors(bio);
1677        int err = -EIO;
1678        char b[BDEVNAME_SIZE];
1679        struct hd_struct *part;
1680
1681        might_sleep();
1682
1683        if (bio_check_eod(bio, nr_sectors))
1684                goto end_io;
1685
1686        q = bdev_get_queue(bio->bi_bdev);
1687        if (unlikely(!q)) {
1688                printk(KERN_ERR
1689                       "generic_make_request: Trying to access "
1690                        "nonexistent block-device %s (%Lu)\n",
1691                        bdevname(bio->bi_bdev, b),
1692                        (long long) bio->bi_sector);
1693                goto end_io;
1694        }
1695
1696        if (likely(bio_is_rw(bio) &&
1697                   nr_sectors > queue_max_hw_sectors(q))) {
1698                printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1699                       bdevname(bio->bi_bdev, b),
1700                       bio_sectors(bio),
1701                       queue_max_hw_sectors(q));
1702                goto end_io;
1703        }
1704
1705        part = bio->bi_bdev->bd_part;
1706        if (should_fail_request(part, bio->bi_size) ||
1707            should_fail_request(&part_to_disk(part)->part0,
1708                                bio->bi_size))
1709                goto end_io;
1710
1711        /*
1712         * If this device has partitions, remap block n
1713         * of partition p to block n+start(p) of the disk.
1714         */
1715        blk_partition_remap(bio);
1716
1717        if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1718                goto end_io;
1719
1720        if (bio_check_eod(bio, nr_sectors))
1721                goto end_io;
1722
1723        /*
1724         * Filter flush bio's early so that make_request based
1725         * drivers without flush support don't have to worry
1726         * about them.
1727         */
1728        if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1729                bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1730                if (!nr_sectors) {
1731                        err = 0;
1732                        goto end_io;
1733                }
1734        }
1735
1736        if ((bio->bi_rw & REQ_DISCARD) &&
1737            (!blk_queue_discard(q) ||
1738             ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1739                err = -EOPNOTSUPP;
1740                goto end_io;
1741        }
1742
1743        if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1744                err = -EOPNOTSUPP;
1745                goto end_io;
1746        }
1747
1748        /*
1749         * Various block parts want %current->io_context and lazy ioc
1750         * allocation ends up trading a lot of pain for a small amount of
1751         * memory.  Just allocate it upfront.  This may fail and block
1752         * layer knows how to live with it.
1753         */
1754        create_io_context(GFP_ATOMIC, q->node);
1755
1756        if (blk_throtl_bio(q, bio))
1757                return false;   /* throttled, will be resubmitted later */
1758
1759        trace_block_bio_queue(q, bio);
1760        return true;
1761
1762end_io:
1763        bio_endio(bio, err);
1764        return false;
1765}
1766
1767/**
1768 * generic_make_request - hand a buffer to its device driver for I/O
1769 * @bio:  The bio describing the location in memory and on the device.
1770 *
1771 * generic_make_request() is used to make I/O requests of block
1772 * devices. It is passed a &struct bio, which describes the I/O that needs
1773 * to be done.
1774 *
1775 * generic_make_request() does not return any status.  The
1776 * success/failure status of the request, along with notification of
1777 * completion, is delivered asynchronously through the bio->bi_end_io
1778 * function described (one day) else where.
1779 *
1780 * The caller of generic_make_request must make sure that bi_io_vec
1781 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1782 * set to describe the device address, and the
1783 * bi_end_io and optionally bi_private are set to describe how
1784 * completion notification should be signaled.
1785 *
1786 * generic_make_request and the drivers it calls may use bi_next if this
1787 * bio happens to be merged with someone else, and may resubmit the bio to
1788 * a lower device by calling into generic_make_request recursively, which
1789 * means the bio should NOT be touched after the call to ->make_request_fn.
1790 */
1791void generic_make_request(struct bio *bio)
1792{
1793        struct bio_list bio_list_on_stack;
1794
1795        if (!generic_make_request_checks(bio))
1796                return;
1797
1798        /*
1799         * We only want one ->make_request_fn to be active at a time, else
1800         * stack usage with stacked devices could be a problem.  So use
1801         * current->bio_list to keep a list of requests submited by a
1802         * make_request_fn function.  current->bio_list is also used as a
1803         * flag to say if generic_make_request is currently active in this
1804         * task or not.  If it is NULL, then no make_request is active.  If
1805         * it is non-NULL, then a make_request is active, and new requests
1806         * should be added at the tail
1807         */
1808        if (current->bio_list) {
1809                bio_list_add(current->bio_list, bio);
1810                return;
1811        }
1812
1813        /* following loop may be a bit non-obvious, and so deserves some
1814         * explanation.
1815         * Before entering the loop, bio->bi_next is NULL (as all callers
1816         * ensure that) so we have a list with a single bio.
1817         * We pretend that we have just taken it off a longer list, so
1818         * we assign bio_list to a pointer to the bio_list_on_stack,
1819         * thus initialising the bio_list of new bios to be
1820         * added.  ->make_request() may indeed add some more bios
1821         * through a recursive call to generic_make_request.  If it
1822         * did, we find a non-NULL value in bio_list and re-enter the loop
1823         * from the top.  In this case we really did just take the bio
1824         * of the top of the list (no pretending) and so remove it from
1825         * bio_list, and call into ->make_request() again.
1826         */
1827        BUG_ON(bio->bi_next);
1828        bio_list_init(&bio_list_on_stack);
1829        current->bio_list = &bio_list_on_stack;
1830        do {
1831                struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1832
1833                q->make_request_fn(q, bio);
1834
1835                bio = bio_list_pop(current->bio_list);
1836        } while (bio);
1837        current->bio_list = NULL; /* deactivate */
1838}
1839EXPORT_SYMBOL(generic_make_request);
1840
1841/**
1842 * submit_bio - submit a bio to the block device layer for I/O
1843 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1844 * @bio: The &struct bio which describes the I/O
1845 *
1846 * submit_bio() is very similar in purpose to generic_make_request(), and
1847 * uses that function to do most of the work. Both are fairly rough
1848 * interfaces; @bio must be presetup and ready for I/O.
1849 *
1850 */
1851void submit_bio(int rw, struct bio *bio)
1852{
1853        bio->bi_rw |= rw;
1854
1855        /*
1856         * If it's a regular read/write or a barrier with data attached,
1857         * go through the normal accounting stuff before submission.
1858         */
1859        if (bio_has_data(bio)) {
1860                unsigned int count;
1861
1862                if (unlikely(rw & REQ_WRITE_SAME))
1863                        count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1864                else
1865                        count = bio_sectors(bio);
1866
1867                if (rw & WRITE) {
1868                        count_vm_events(PGPGOUT, count);
1869                } else {
1870                        task_io_account_read(bio->bi_size);
1871                        count_vm_events(PGPGIN, count);
1872                }
1873
1874                if (unlikely(block_dump)) {
1875                        char b[BDEVNAME_SIZE];
1876                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1877                        current->comm, task_pid_nr(current),
1878                                (rw & WRITE) ? "WRITE" : "READ",
1879                                (unsigned long long)bio->bi_sector,
1880                                bdevname(bio->bi_bdev, b),
1881                                count);
1882                }
1883        }
1884
1885        generic_make_request(bio);
1886}
1887EXPORT_SYMBOL(submit_bio);
1888
1889/**
1890 * blk_rq_check_limits - Helper function to check a request for the queue limit
1891 * @q:  the queue
1892 * @rq: the request being checked
1893 *
1894 * Description:
1895 *    @rq may have been made based on weaker limitations of upper-level queues
1896 *    in request stacking drivers, and it may violate the limitation of @q.
1897 *    Since the block layer and the underlying device driver trust @rq
1898 *    after it is inserted to @q, it should be checked against @q before
1899 *    the insertion using this generic function.
1900 *
1901 *    This function should also be useful for request stacking drivers
1902 *    in some cases below, so export this function.
1903 *    Request stacking drivers like request-based dm may change the queue
1904 *    limits while requests are in the queue (e.g. dm's table swapping).
1905 *    Such request stacking drivers should check those requests agaist
1906 *    the new queue limits again when they dispatch those requests,
1907 *    although such checkings are also done against the old queue limits
1908 *    when submitting requests.
1909 */
1910int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1911{
1912        if (!rq_mergeable(rq))
1913                return 0;
1914
1915        if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1916                printk(KERN_ERR "%s: over max size limit.\n", __func__);
1917                return -EIO;
1918        }
1919
1920        /*
1921         * queue's settings related to segment counting like q->bounce_pfn
1922         * may differ from that of other stacking queues.
1923         * Recalculate it to check the request correctly on this queue's
1924         * limitation.
1925         */
1926        blk_recalc_rq_segments(rq);
1927        if (rq->nr_phys_segments > queue_max_segments(q)) {
1928                printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1929                return -EIO;
1930        }
1931
1932        return 0;
1933}
1934EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1935
1936/**
1937 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1938 * @q:  the queue to submit the request
1939 * @rq: the request being queued
1940 */
1941int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1942{
1943        unsigned long flags;
1944        int where = ELEVATOR_INSERT_BACK;
1945
1946        if (blk_rq_check_limits(q, rq))
1947                return -EIO;
1948
1949        if (rq->rq_disk &&
1950            should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1951                return -EIO;
1952
1953        spin_lock_irqsave(q->queue_lock, flags);
1954        if (unlikely(blk_queue_dying(q))) {
1955                spin_unlock_irqrestore(q->queue_lock, flags);
1956                return -ENODEV;
1957        }
1958
1959        /*
1960         * Submitting request must be dequeued before calling this function
1961         * because it will be linked to another request_queue
1962         */
1963        BUG_ON(blk_queued_rq(rq));
1964
1965        if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1966                where = ELEVATOR_INSERT_FLUSH;
1967
1968        add_acct_request(q, rq, where);
1969        if (where == ELEVATOR_INSERT_FLUSH)
1970                __blk_run_queue(q);
1971        spin_unlock_irqrestore(q->queue_lock, flags);
1972
1973        return 0;
1974}
1975EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1976
1977/**
1978 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1979 * @rq: request to examine
1980 *
1981 * Description:
1982 *     A request could be merge of IOs which require different failure
1983 *     handling.  This function determines the number of bytes which
1984 *     can be failed from the beginning of the request without
1985 *     crossing into area which need to be retried further.
1986 *
1987 * Return:
1988 *     The number of bytes to fail.
1989 *
1990 * Context:
1991 *     queue_lock must be held.
1992 */
1993unsigned int blk_rq_err_bytes(const struct request *rq)
1994{
1995        unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1996        unsigned int bytes = 0;
1997        struct bio *bio;
1998
1999        if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2000                return blk_rq_bytes(rq);
2001
2002        /*
2003         * Currently the only 'mixing' which can happen is between
2004         * different fastfail types.  We can safely fail portions
2005         * which have all the failfast bits that the first one has -
2006         * the ones which are at least as eager to fail as the first
2007         * one.
2008         */
2009        for (bio = rq->bio; bio; bio = bio->bi_next) {
2010                if ((bio->bi_rw & ff) != ff)
2011                        break;
2012                bytes += bio->bi_size;
2013        }
2014
2015        /* this could lead to infinite loop */
2016        BUG_ON(blk_rq_bytes(rq) && !bytes);
2017        return bytes;
2018}
2019EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2020
2021static void blk_account_io_completion(struct request *req, unsigned int bytes)
2022{
2023        if (blk_do_io_stat(req)) {
2024                const int rw = rq_data_dir(req);
2025                struct hd_struct *part;
2026                int cpu;
2027
2028                cpu = part_stat_lock();
2029                part = req->part;
2030                part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2031                part_stat_unlock();
2032        }
2033}
2034
2035static void blk_account_io_done(struct request *req)
2036{
2037        /*
2038         * Account IO completion.  flush_rq isn't accounted as a
2039         * normal IO on queueing nor completion.  Accounting the
2040         * containing request is enough.
2041         */
2042        if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2043                unsigned long duration = jiffies - req->start_time;
2044                const int rw = rq_data_dir(req);
2045                struct hd_struct *part;
2046                int cpu;
2047
2048                cpu = part_stat_lock();
2049                part = req->part;
2050
2051                part_stat_inc(cpu, part, ios[rw]);
2052                part_stat_add(cpu, part, ticks[rw], duration);
2053                part_round_stats(cpu, part);
2054                part_dec_in_flight(part, rw);
2055
2056                hd_struct_put(part);
2057                part_stat_unlock();
2058        }
2059}
2060
2061/**
2062 * blk_peek_request - peek at the top of a request queue
2063 * @q: request queue to peek at
2064 *
2065 * Description:
2066 *     Return the request at the top of @q.  The returned request
2067 *     should be started using blk_start_request() before LLD starts
2068 *     processing it.
2069 *
2070 * Return:
2071 *     Pointer to the request at the top of @q if available.  Null
2072 *     otherwise.
2073 *
2074 * Context:
2075 *     queue_lock must be held.
2076 */
2077struct request *blk_peek_request(struct request_queue *q)
2078{
2079        struct request *rq;
2080        int ret;
2081
2082        while ((rq = __elv_next_request(q)) != NULL) {
2083                if (!(rq->cmd_flags & REQ_STARTED)) {
2084                        /*
2085                         * This is the first time the device driver
2086                         * sees this request (possibly after
2087                         * requeueing).  Notify IO scheduler.
2088                         */
2089                        if (rq->cmd_flags & REQ_SORTED)
2090                                elv_activate_rq(q, rq);
2091
2092                        /*
2093                         * just mark as started even if we don't start
2094                         * it, a request that has been delayed should
2095                         * not be passed by new incoming requests
2096                         */
2097                        rq->cmd_flags |= REQ_STARTED;
2098                        trace_block_rq_issue(q, rq);
2099                }
2100
2101                if (!q->boundary_rq || q->boundary_rq == rq) {
2102                        q->end_sector = rq_end_sector(rq);
2103                        q->boundary_rq = NULL;
2104                }
2105
2106                if (rq->cmd_flags & REQ_DONTPREP)
2107                        break;
2108
2109                if (q->dma_drain_size && blk_rq_bytes(rq)) {
2110                        /*
2111                         * make sure space for the drain appears we
2112                         * know we can do this because max_hw_segments
2113                         * has been adjusted to be one fewer than the
2114                         * device can handle
2115                         */
2116                        rq->nr_phys_segments++;
2117                }
2118
2119                if (!q->prep_rq_fn)
2120                        break;
2121
2122                ret = q->prep_rq_fn(q, rq);
2123                if (ret == BLKPREP_OK) {
2124                        break;
2125                } else if (ret == BLKPREP_DEFER) {
2126                        /*
2127                         * the request may have been (partially) prepped.
2128                         * we need to keep this request in the front to
2129                         * avoid resource deadlock.  REQ_STARTED will
2130                         * prevent other fs requests from passing this one.
2131                         */
2132                        if (q->dma_drain_size && blk_rq_bytes(rq) &&
2133                            !(rq->cmd_flags & REQ_DONTPREP)) {
2134                                /*
2135                                 * remove the space for the drain we added
2136                                 * so that we don't add it again
2137                                 */
2138                                --rq->nr_phys_segments;
2139                        }
2140
2141                        rq = NULL;
2142                        break;
2143                } else if (ret == BLKPREP_KILL) {
2144                        rq->cmd_flags |= REQ_QUIET;
2145                        /*
2146                         * Mark this request as started so we don't trigger
2147                         * any debug logic in the end I/O path.
2148                         */
2149                        blk_start_request(rq);
2150                        __blk_end_request_all(rq, -EIO);
2151                } else {
2152                        printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2153                        break;
2154                }
2155        }
2156
2157        return rq;
2158}
2159EXPORT_SYMBOL(blk_peek_request);
2160
2161void blk_dequeue_request(struct request *rq)
2162{
2163        struct request_queue *q = rq->q;
2164
2165        BUG_ON(list_empty(&rq->queuelist));
2166        BUG_ON(ELV_ON_HASH(rq));
2167
2168        list_del_init(&rq->queuelist);
2169
2170        /*
2171         * the time frame between a request being removed from the lists
2172         * and to it is freed is accounted as io that is in progress at
2173         * the driver side.
2174         */
2175        if (blk_account_rq(rq)) {
2176                q->in_flight[rq_is_sync(rq)]++;
2177                set_io_start_time_ns(rq);
2178        }
2179}
2180
2181/**
2182 * blk_start_request - start request processing on the driver
2183 * @req: request to dequeue
2184 *
2185 * Description:
2186 *     Dequeue @req and start timeout timer on it.  This hands off the
2187 *     request to the driver.
2188 *
2189 *     Block internal functions which don't want to start timer should
2190 *     call blk_dequeue_request().
2191 *
2192 * Context:
2193 *     queue_lock must be held.
2194 */
2195void blk_start_request(struct request *req)
2196{
2197        blk_dequeue_request(req);
2198
2199        /*
2200         * We are now handing the request to the hardware, initialize
2201         * resid_len to full count and add the timeout handler.
2202         */
2203        req->resid_len = blk_rq_bytes(req);
2204        if (unlikely(blk_bidi_rq(req)))
2205                req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2206
2207        blk_add_timer(req);
2208}
2209EXPORT_SYMBOL(blk_start_request);
2210
2211/**
2212 * blk_fetch_request - fetch a request from a request queue
2213 * @q: request queue to fetch a request from
2214 *
2215 * Description:
2216 *     Return the request at the top of @q.  The request is started on
2217 *     return and LLD can start processing it immediately.
2218 *
2219 * Return:
2220 *     Pointer to the request at the top of @q if available.  Null
2221 *     otherwise.
2222 *
2223 * Context:
2224 *     queue_lock must be held.
2225 */
2226struct request *blk_fetch_request(struct request_queue *q)
2227{
2228        struct request *rq;
2229
2230        rq = blk_peek_request(q);
2231        if (rq)
2232                blk_start_request(rq);
2233        return rq;
2234}
2235EXPORT_SYMBOL(blk_fetch_request);
2236
2237/**
2238 * blk_update_request - Special helper function for request stacking drivers
2239 * @req:      the request being processed
2240 * @error:    %0 for success, < %0 for error
2241 * @nr_bytes: number of bytes to complete @req
2242 *
2243 * Description:
2244 *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2245 *     the request structure even if @req doesn't have leftover.
2246 *     If @req has leftover, sets it up for the next range of segments.
2247 *
2248 *     This special helper function is only for request stacking drivers
2249 *     (e.g. request-based dm) so that they can handle partial completion.
2250 *     Actual device drivers should use blk_end_request instead.
2251 *
2252 *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2253 *     %false return from this function.
2254 *
2255 * Return:
2256 *     %false - this request doesn't have any more data
2257 *     %true  - this request has more data
2258 **/
2259bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2260{
2261        int total_bytes, bio_nbytes, next_idx = 0;
2262        struct bio *bio;
2263
2264        if (!req->bio)
2265                return false;
2266
2267        trace_block_rq_complete(req->q, req);
2268
2269        /*
2270         * For fs requests, rq is just carrier of independent bio's
2271         * and each partial completion should be handled separately.
2272         * Reset per-request error on each partial completion.
2273         *
2274         * TODO: tj: This is too subtle.  It would be better to let
2275         * low level drivers do what they see fit.
2276         */
2277        if (req->cmd_type == REQ_TYPE_FS)
2278                req->errors = 0;
2279
2280        if (error && req->cmd_type == REQ_TYPE_FS &&
2281            !(req->cmd_flags & REQ_QUIET)) {
2282                char *error_type;
2283
2284                switch (error) {
2285                case -ENOLINK:
2286                        error_type = "recoverable transport";
2287                        break;
2288                case -EREMOTEIO:
2289                        error_type = "critical target";
2290                        break;
2291                case -EBADE:
2292                        error_type = "critical nexus";
2293                        break;
2294                case -EIO:
2295                default:
2296                        error_type = "I/O";
2297                        break;
2298                }
2299                printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2300                                   error_type, req->rq_disk ?
2301                                   req->rq_disk->disk_name : "?",
2302                                   (unsigned long long)blk_rq_pos(req));
2303
2304        }
2305
2306        blk_account_io_completion(req, nr_bytes);
2307
2308        total_bytes = bio_nbytes = 0;
2309        while ((bio = req->bio) != NULL) {
2310                int nbytes;
2311
2312                if (nr_bytes >= bio->bi_size) {
2313                        req->bio = bio->bi_next;
2314                        nbytes = bio->bi_size;
2315                        req_bio_endio(req, bio, nbytes, error);
2316                        next_idx = 0;
2317                        bio_nbytes = 0;
2318                } else {
2319                        int idx = bio->bi_idx + next_idx;
2320
2321                        if (unlikely(idx >= bio->bi_vcnt)) {
2322                                blk_dump_rq_flags(req, "__end_that");
2323                                printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2324                                       __func__, idx, bio->bi_vcnt);
2325                                break;
2326                        }
2327
2328                        nbytes = bio_iovec_idx(bio, idx)->bv_len;
2329                        BIO_BUG_ON(nbytes > bio->bi_size);
2330
2331                        /*
2332                         * not a complete bvec done
2333                         */
2334                        if (unlikely(nbytes > nr_bytes)) {
2335                                bio_nbytes += nr_bytes;
2336                                total_bytes += nr_bytes;
2337                                break;
2338                        }
2339
2340                        /*
2341                         * advance to the next vector
2342                         */
2343                        next_idx++;
2344                        bio_nbytes += nbytes;
2345                }
2346
2347                total_bytes += nbytes;
2348                nr_bytes -= nbytes;
2349
2350                bio = req->bio;
2351                if (bio) {
2352                        /*
2353                         * end more in this run, or just return 'not-done'
2354                         */
2355                        if (unlikely(nr_bytes <= 0))
2356                                break;
2357                }
2358        }
2359
2360        /*
2361         * completely done
2362         */
2363        if (!req->bio) {
2364                /*
2365                 * Reset counters so that the request stacking driver
2366                 * can find how many bytes remain in the request
2367                 * later.
2368                 */
2369                req->__data_len = 0;
2370                return false;
2371        }
2372
2373        /*
2374         * if the request wasn't completed, update state
2375         */
2376        if (bio_nbytes) {
2377                req_bio_endio(req, bio, bio_nbytes, error);
2378                bio->bi_idx += next_idx;
2379                bio_iovec(bio)->bv_offset += nr_bytes;
2380                bio_iovec(bio)->bv_len -= nr_bytes;
2381        }
2382
2383        req->__data_len -= total_bytes;
2384        req->buffer = bio_data(req->bio);
2385
2386        /* update sector only for requests with clear definition of sector */
2387        if (req->cmd_type == REQ_TYPE_FS)
2388                req->__sector += total_bytes >> 9;
2389
2390        /* mixed attributes always follow the first bio */
2391        if (req->cmd_flags & REQ_MIXED_MERGE) {
2392                req->cmd_flags &= ~REQ_FAILFAST_MASK;
2393                req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2394        }
2395
2396        /*
2397         * If total number of sectors is less than the first segment
2398         * size, something has gone terribly wrong.
2399         */
2400        if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2401                blk_dump_rq_flags(req, "request botched");
2402                req->__data_len = blk_rq_cur_bytes(req);
2403        }
2404
2405        /* recalculate the number of segments */
2406        blk_recalc_rq_segments(req);
2407
2408        return true;
2409}
2410EXPORT_SYMBOL_GPL(blk_update_request);
2411
2412static bool blk_update_bidi_request(struct request *rq, int error,
2413                                    unsigned int nr_bytes,
2414                                    unsigned int bidi_bytes)
2415{
2416        if (blk_update_request(rq, error, nr_bytes))
2417                return true;
2418
2419        /* Bidi request must be completed as a whole */
2420        if (unlikely(blk_bidi_rq(rq)) &&
2421            blk_update_request(rq->next_rq, error, bidi_bytes))
2422                return true;
2423
2424        if (blk_queue_add_random(rq->q))
2425                add_disk_randomness(rq->rq_disk);
2426
2427        return false;
2428}
2429
2430/**
2431 * blk_unprep_request - unprepare a request
2432 * @req:        the request
2433 *
2434 * This function makes a request ready for complete resubmission (or
2435 * completion).  It happens only after all error handling is complete,
2436 * so represents the appropriate moment to deallocate any resources
2437 * that were allocated to the request in the prep_rq_fn.  The queue
2438 * lock is held when calling this.
2439 */
2440void blk_unprep_request(struct request *req)
2441{
2442        struct request_queue *q = req->q;
2443
2444        req->cmd_flags &= ~REQ_DONTPREP;
2445        if (q->unprep_rq_fn)
2446                q->unprep_rq_fn(q, req);
2447}
2448EXPORT_SYMBOL_GPL(blk_unprep_request);
2449
2450/*
2451 * queue lock must be held
2452 */
2453static void blk_finish_request(struct request *req, int error)
2454{
2455        if (blk_rq_tagged(req))
2456                blk_queue_end_tag(req->q, req);
2457
2458        BUG_ON(blk_queued_rq(req));
2459
2460        if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2461                laptop_io_completion(&req->q->backing_dev_info);
2462
2463        blk_delete_timer(req);
2464
2465        if (req->cmd_flags & REQ_DONTPREP)
2466                blk_unprep_request(req);
2467
2468
2469        blk_account_io_done(req);
2470
2471        if (req->end_io)
2472                req->end_io(req, error);
2473        else {
2474                if (blk_bidi_rq(req))
2475                        __blk_put_request(req->next_rq->q, req->next_rq);
2476
2477                __blk_put_request(req->q, req);
2478        }
2479}
2480
2481/**
2482 * blk_end_bidi_request - Complete a bidi request
2483 * @rq:         the request to complete
2484 * @error:      %0 for success, < %0 for error
2485 * @nr_bytes:   number of bytes to complete @rq
2486 * @bidi_bytes: number of bytes to complete @rq->next_rq
2487 *
2488 * Description:
2489 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2490 *     Drivers that supports bidi can safely call this member for any
2491 *     type of request, bidi or uni.  In the later case @bidi_bytes is
2492 *     just ignored.
2493 *
2494 * Return:
2495 *     %false - we are done with this request
2496 *     %true  - still buffers pending for this request
2497 **/
2498static bool blk_end_bidi_request(struct request *rq, int error,
2499                                 unsigned int nr_bytes, unsigned int bidi_bytes)
2500{
2501        struct request_queue *q = rq->q;
2502        unsigned long flags;
2503
2504        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2505                return true;
2506
2507        spin_lock_irqsave(q->queue_lock, flags);
2508        blk_finish_request(rq, error);
2509        spin_unlock_irqrestore(q->queue_lock, flags);
2510
2511        return false;
2512}
2513
2514/**
2515 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2516 * @rq:         the request to complete
2517 * @error:      %0 for success, < %0 for error
2518 * @nr_bytes:   number of bytes to complete @rq
2519 * @bidi_bytes: number of bytes to complete @rq->next_rq
2520 *
2521 * Description:
2522 *     Identical to blk_end_bidi_request() except that queue lock is
2523 *     assumed to be locked on entry and remains so on return.
2524 *
2525 * Return:
2526 *     %false - we are done with this request
2527 *     %true  - still buffers pending for this request
2528 **/
2529bool __blk_end_bidi_request(struct request *rq, int error,
2530                                   unsigned int nr_bytes, unsigned int bidi_bytes)
2531{
2532        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2533                return true;
2534
2535        blk_finish_request(rq, error);
2536
2537        return false;
2538}
2539
2540/**
2541 * blk_end_request - Helper function for drivers to complete the request.
2542 * @rq:       the request being processed
2543 * @error:    %0 for success, < %0 for error
2544 * @nr_bytes: number of bytes to complete
2545 *
2546 * Description:
2547 *     Ends I/O on a number of bytes attached to @rq.
2548 *     If @rq has leftover, sets it up for the next range of segments.
2549 *
2550 * Return:
2551 *     %false - we are done with this request
2552 *     %true  - still buffers pending for this request
2553 **/
2554bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2555{
2556        return blk_end_bidi_request(rq, error, nr_bytes, 0);
2557}
2558EXPORT_SYMBOL(blk_end_request);
2559
2560/**
2561 * blk_end_request_all - Helper function for drives to finish the request.
2562 * @rq: the request to finish
2563 * @error: %0 for success, < %0 for error
2564 *
2565 * Description:
2566 *     Completely finish @rq.
2567 */
2568void blk_end_request_all(struct request *rq, int error)
2569{
2570        bool pending;
2571        unsigned int bidi_bytes = 0;
2572
2573        if (unlikely(blk_bidi_rq(rq)))
2574                bidi_bytes = blk_rq_bytes(rq->next_rq);
2575
2576        pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2577        BUG_ON(pending);
2578}
2579EXPORT_SYMBOL(blk_end_request_all);
2580
2581/**
2582 * blk_end_request_cur - Helper function to finish the current request chunk.
2583 * @rq: the request to finish the current chunk for
2584 * @error: %0 for success, < %0 for error
2585 *
2586 * Description:
2587 *     Complete the current consecutively mapped chunk from @rq.
2588 *
2589 * Return:
2590 *     %false - we are done with this request
2591 *     %true  - still buffers pending for this request
2592 */
2593bool blk_end_request_cur(struct request *rq, int error)
2594{
2595        return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2596}
2597EXPORT_SYMBOL(blk_end_request_cur);
2598
2599/**
2600 * blk_end_request_err - Finish a request till the next failure boundary.
2601 * @rq: the request to finish till the next failure boundary for
2602 * @error: must be negative errno
2603 *
2604 * Description:
2605 *     Complete @rq till the next failure boundary.
2606 *
2607 * Return:
2608 *     %false - we are done with this request
2609 *     %true  - still buffers pending for this request
2610 */
2611bool blk_end_request_err(struct request *rq, int error)
2612{
2613        WARN_ON(error >= 0);
2614        return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2615}
2616EXPORT_SYMBOL_GPL(blk_end_request_err);
2617
2618/**
2619 * __blk_end_request - Helper function for drivers to complete the request.
2620 * @rq:       the request being processed
2621 * @error:    %0 for success, < %0 for error
2622 * @nr_bytes: number of bytes to complete
2623 *
2624 * Description:
2625 *     Must be called with queue lock held unlike blk_end_request().
2626 *
2627 * Return:
2628 *     %false - we are done with this request
2629 *     %true  - still buffers pending for this request
2630 **/
2631bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2632{
2633        return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2634}
2635EXPORT_SYMBOL(__blk_end_request);
2636
2637/**
2638 * __blk_end_request_all - Helper function for drives to finish the request.
2639 * @rq: the request to finish
2640 * @error: %0 for success, < %0 for error
2641 *
2642 * Description:
2643 *     Completely finish @rq.  Must be called with queue lock held.
2644 */
2645void __blk_end_request_all(struct request *rq, int error)
2646{
2647        bool pending;
2648        unsigned int bidi_bytes = 0;
2649
2650        if (unlikely(blk_bidi_rq(rq)))
2651                bidi_bytes = blk_rq_bytes(rq->next_rq);
2652
2653        pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2654        BUG_ON(pending);
2655}
2656EXPORT_SYMBOL(__blk_end_request_all);
2657
2658/**
2659 * __blk_end_request_cur - Helper function to finish the current request chunk.
2660 * @rq: the request to finish the current chunk for
2661 * @error: %0 for success, < %0 for error
2662 *
2663 * Description:
2664 *     Complete the current consecutively mapped chunk from @rq.  Must
2665 *     be called with queue lock held.
2666 *
2667 * Return:
2668 *     %false - we are done with this request
2669 *     %true  - still buffers pending for this request
2670 */
2671bool __blk_end_request_cur(struct request *rq, int error)
2672{
2673        return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2674}
2675EXPORT_SYMBOL(__blk_end_request_cur);
2676
2677/**
2678 * __blk_end_request_err - Finish a request till the next failure boundary.
2679 * @rq: the request to finish till the next failure boundary for
2680 * @error: must be negative errno
2681 *
2682 * Description:
2683 *     Complete @rq till the next failure boundary.  Must be called
2684 *     with queue lock held.
2685 *
2686 * Return:
2687 *     %false - we are done with this request
2688 *     %true  - still buffers pending for this request
2689 */
2690bool __blk_end_request_err(struct request *rq, int error)
2691{
2692        WARN_ON(error >= 0);
2693        return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2694}
2695EXPORT_SYMBOL_GPL(__blk_end_request_err);
2696
2697void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2698                     struct bio *bio)
2699{
2700        /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2701        rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2702
2703        if (bio_has_data(bio)) {
2704                rq->nr_phys_segments = bio_phys_segments(q, bio);
2705                rq->buffer = bio_data(bio);
2706        }
2707        rq->__data_len = bio->bi_size;
2708        rq->bio = rq->biotail = bio;
2709
2710        if (bio->bi_bdev)
2711                rq->rq_disk = bio->bi_bdev->bd_disk;
2712}
2713
2714#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2715/**
2716 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2717 * @rq: the request to be flushed
2718 *
2719 * Description:
2720 *     Flush all pages in @rq.
2721 */
2722void rq_flush_dcache_pages(struct request *rq)
2723{
2724        struct req_iterator iter;
2725        struct bio_vec *bvec;
2726
2727        rq_for_each_segment(bvec, rq, iter)
2728                flush_dcache_page(bvec->bv_page);
2729}
2730EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2731#endif
2732
2733/**
2734 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2735 * @q : the queue of the device being checked
2736 *
2737 * Description:
2738 *    Check if underlying low-level drivers of a device are busy.
2739 *    If the drivers want to export their busy state, they must set own
2740 *    exporting function using blk_queue_lld_busy() first.
2741 *
2742 *    Basically, this function is used only by request stacking drivers
2743 *    to stop dispatching requests to underlying devices when underlying
2744 *    devices are busy.  This behavior helps more I/O merging on the queue
2745 *    of the request stacking driver and prevents I/O throughput regression
2746 *    on burst I/O load.
2747 *
2748 * Return:
2749 *    0 - Not busy (The request stacking driver should dispatch request)
2750 *    1 - Busy (The request stacking driver should stop dispatching request)
2751 */
2752int blk_lld_busy(struct request_queue *q)
2753{
2754        if (q->lld_busy_fn)
2755                return q->lld_busy_fn(q);
2756
2757        return 0;
2758}
2759EXPORT_SYMBOL_GPL(blk_lld_busy);
2760
2761/**
2762 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2763 * @rq: the clone request to be cleaned up
2764 *
2765 * Description:
2766 *     Free all bios in @rq for a cloned request.
2767 */
2768void blk_rq_unprep_clone(struct request *rq)
2769{
2770        struct bio *bio;
2771
2772        while ((bio = rq->bio) != NULL) {
2773                rq->bio = bio->bi_next;
2774
2775                bio_put(bio);
2776        }
2777}
2778EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2779
2780/*
2781 * Copy attributes of the original request to the clone request.
2782 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2783 */
2784static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2785{
2786        dst->cpu = src->cpu;
2787        dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2788        dst->cmd_type = src->cmd_type;
2789        dst->__sector = blk_rq_pos(src);
2790        dst->__data_len = blk_rq_bytes(src);
2791        dst->nr_phys_segments = src->nr_phys_segments;
2792        dst->ioprio = src->ioprio;
2793        dst->extra_len = src->extra_len;
2794}
2795
2796/**
2797 * blk_rq_prep_clone - Helper function to setup clone request
2798 * @rq: the request to be setup
2799 * @rq_src: original request to be cloned
2800 * @bs: bio_set that bios for clone are allocated from
2801 * @gfp_mask: memory allocation mask for bio
2802 * @bio_ctr: setup function to be called for each clone bio.
2803 *           Returns %0 for success, non %0 for failure.
2804 * @data: private data to be passed to @bio_ctr
2805 *
2806 * Description:
2807 *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2808 *     The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2809 *     are not copied, and copying such parts is the caller's responsibility.
2810 *     Also, pages which the original bios are pointing to are not copied
2811 *     and the cloned bios just point same pages.
2812 *     So cloned bios must be completed before original bios, which means
2813 *     the caller must complete @rq before @rq_src.
2814 */
2815int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2816                      struct bio_set *bs, gfp_t gfp_mask,
2817                      int (*bio_ctr)(struct bio *, struct bio *, void *),
2818                      void *data)
2819{
2820        struct bio *bio, *bio_src;
2821
2822        if (!bs)
2823                bs = fs_bio_set;
2824
2825        blk_rq_init(NULL, rq);
2826
2827        __rq_for_each_bio(bio_src, rq_src) {
2828                bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2829                if (!bio)
2830                        goto free_and_out;
2831
2832                if (bio_ctr && bio_ctr(bio, bio_src, data))
2833                        goto free_and_out;
2834
2835                if (rq->bio) {
2836                        rq->biotail->bi_next = bio;
2837                        rq->biotail = bio;
2838                } else
2839                        rq->bio = rq->biotail = bio;
2840        }
2841
2842        __blk_rq_prep_clone(rq, rq_src);
2843
2844        return 0;
2845
2846free_and_out:
2847        if (bio)
2848                bio_put(bio);
2849        blk_rq_unprep_clone(rq);
2850
2851        return -ENOMEM;
2852}
2853EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2854
2855int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2856{
2857        return queue_work(kblockd_workqueue, work);
2858}
2859EXPORT_SYMBOL(kblockd_schedule_work);
2860
2861int kblockd_schedule_delayed_work(struct request_queue *q,
2862                        struct delayed_work *dwork, unsigned long delay)
2863{
2864        return queue_delayed_work(kblockd_workqueue, dwork, delay);
2865}
2866EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2867
2868#define PLUG_MAGIC      0x91827364
2869
2870/**
2871 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2872 * @plug:       The &struct blk_plug that needs to be initialized
2873 *
2874 * Description:
2875 *   Tracking blk_plug inside the task_struct will help with auto-flushing the
2876 *   pending I/O should the task end up blocking between blk_start_plug() and
2877 *   blk_finish_plug(). This is important from a performance perspective, but
2878 *   also ensures that we don't deadlock. For instance, if the task is blocking
2879 *   for a memory allocation, memory reclaim could end up wanting to free a
2880 *   page belonging to that request that is currently residing in our private
2881 *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
2882 *   this kind of deadlock.
2883 */
2884void blk_start_plug(struct blk_plug *plug)
2885{
2886        struct task_struct *tsk = current;
2887
2888        plug->magic = PLUG_MAGIC;
2889        INIT_LIST_HEAD(&plug->list);
2890        INIT_LIST_HEAD(&plug->cb_list);
2891        plug->should_sort = 0;
2892
2893        /*
2894         * If this is a nested plug, don't actually assign it. It will be
2895         * flushed on its own.
2896         */
2897        if (!tsk->plug) {
2898                /*
2899                 * Store ordering should not be needed here, since a potential
2900                 * preempt will imply a full memory barrier
2901                 */
2902                tsk->plug = plug;
2903        }
2904}
2905EXPORT_SYMBOL(blk_start_plug);
2906
2907static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2908{
2909        struct request *rqa = container_of(a, struct request, queuelist);
2910        struct request *rqb = container_of(b, struct request, queuelist);
2911
2912        return !(rqa->q < rqb->q ||
2913                (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2914}
2915
2916/*
2917 * If 'from_schedule' is true, then postpone the dispatch of requests
2918 * until a safe kblockd context. We due this to avoid accidental big
2919 * additional stack usage in driver dispatch, in places where the originally
2920 * plugger did not intend it.
2921 */
2922static void queue_unplugged(struct request_queue *q, unsigned int depth,
2923                            bool from_schedule)
2924        __releases(q->queue_lock)
2925{
2926        trace_block_unplug(q, depth, !from_schedule);
2927
2928        if (from_schedule)
2929                blk_run_queue_async(q);
2930        else
2931                __blk_run_queue(q);
2932        spin_unlock(q->queue_lock);
2933}
2934
2935static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2936{
2937        LIST_HEAD(callbacks);
2938
2939        while (!list_empty(&plug->cb_list)) {
2940                list_splice_init(&plug->cb_list, &callbacks);
2941
2942                while (!list_empty(&callbacks)) {
2943                        struct blk_plug_cb *cb = list_first_entry(&callbacks,
2944                                                          struct blk_plug_cb,
2945                                                          list);
2946                        list_del(&cb->list);
2947                        cb->callback(cb, from_schedule);
2948                }
2949        }
2950}
2951
2952struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2953                                      int size)
2954{
2955        struct blk_plug *plug = current->plug;
2956        struct blk_plug_cb *cb;
2957
2958        if (!plug)
2959                return NULL;
2960
2961        list_for_each_entry(cb, &plug->cb_list, list)
2962                if (cb->callback == unplug && cb->data == data)
2963                        return cb;
2964
2965        /* Not currently on the callback list */
2966        BUG_ON(size < sizeof(*cb));
2967        cb = kzalloc(size, GFP_ATOMIC);
2968        if (cb) {
2969                cb->data = data;
2970                cb->callback = unplug;
2971                list_add(&cb->list, &plug->cb_list);
2972        }
2973        return cb;
2974}
2975EXPORT_SYMBOL(blk_check_plugged);
2976
2977void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2978{
2979        struct request_queue *q;
2980        unsigned long flags;
2981        struct request *rq;
2982        LIST_HEAD(list);
2983        unsigned int depth;
2984
2985        BUG_ON(plug->magic != PLUG_MAGIC);
2986
2987        flush_plug_callbacks(plug, from_schedule);
2988        if (list_empty(&plug->list))
2989                return;
2990
2991        list_splice_init(&plug->list, &list);
2992
2993        if (plug->should_sort) {
2994                list_sort(NULL, &list, plug_rq_cmp);
2995                plug->should_sort = 0;
2996        }
2997
2998        q = NULL;
2999        depth = 0;
3000
3001        /*
3002         * Save and disable interrupts here, to avoid doing it for every
3003         * queue lock we have to take.
3004         */
3005        local_irq_save(flags);
3006        while (!list_empty(&list)) {
3007                rq = list_entry_rq(list.next);
3008                list_del_init(&rq->queuelist);
3009                BUG_ON(!rq->q);
3010                if (rq->q != q) {
3011                        /*
3012                         * This drops the queue lock
3013                         */
3014                        if (q)
3015                                queue_unplugged(q, depth, from_schedule);
3016                        q = rq->q;
3017                        depth = 0;
3018                        spin_lock(q->queue_lock);
3019                }
3020
3021                /*
3022                 * Short-circuit if @q is dead
3023                 */
3024                if (unlikely(blk_queue_dying(q))) {
3025                        __blk_end_request_all(rq, -ENODEV);
3026                        continue;
3027                }
3028
3029                /*
3030                 * rq is already accounted, so use raw insert
3031                 */
3032                if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3033                        __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3034                else
3035                        __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3036
3037                depth++;
3038        }
3039
3040        /*
3041         * This drops the queue lock
3042         */
3043        if (q)
3044                queue_unplugged(q, depth, from_schedule);
3045
3046        local_irq_restore(flags);
3047}
3048
3049void blk_finish_plug(struct blk_plug *plug)
3050{
3051        blk_flush_plug_list(plug, false);
3052
3053        if (plug == current->plug)
3054                current->plug = NULL;
3055}
3056EXPORT_SYMBOL(blk_finish_plug);
3057
3058int __init blk_dev_init(void)
3059{
3060        BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3061                        sizeof(((struct request *)0)->cmd_flags));
3062
3063        /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3064        kblockd_workqueue = alloc_workqueue("kblockd",
3065                                            WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3066        if (!kblockd_workqueue)
3067                panic("Failed to create kblockd\n");
3068
3069        request_cachep = kmem_cache_create("blkdev_requests",
3070                        sizeof(struct request), 0, SLAB_PANIC, NULL);
3071
3072        blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3073                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3074
3075        return 0;
3076}
3077
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.