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/blk-mq.h>
  20#include <linux/highmem.h>
  21#include <linux/mm.h>
  22#include <linux/kernel_stat.h>
  23#include <linux/string.h>
  24#include <linux/init.h>
  25#include <linux/completion.h>
  26#include <linux/slab.h>
  27#include <linux/swap.h>
  28#include <linux/writeback.h>
  29#include <linux/task_io_accounting_ops.h>
  30#include <linux/fault-inject.h>
  31#include <linux/list_sort.h>
  32#include <linux/delay.h>
  33#include <linux/ratelimit.h>
  34#include <linux/pm_runtime.h>
  35#include <linux/blk-cgroup.h>
  36
  37#define CREATE_TRACE_POINTS
  38#include <trace/events/block.h>
  39
  40#include "blk.h"
  41#include "blk-mq.h"
  42
  43EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
  44EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
  45EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
  46EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
  47EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
  48
  49DEFINE_IDA(blk_queue_ida);
  50
  51/*
  52 * For the allocated request tables
  53 */
  54struct kmem_cache *request_cachep;
  55
  56/*
  57 * For queue allocation
  58 */
  59struct kmem_cache *blk_requestq_cachep;
  60
  61/*
  62 * Controlling structure to kblockd
  63 */
  64static struct workqueue_struct *kblockd_workqueue;
  65
  66static void blk_clear_congested(struct request_list *rl, int sync)
  67{
  68#ifdef CONFIG_CGROUP_WRITEBACK
  69        clear_wb_congested(rl->blkg->wb_congested, sync);
  70#else
  71        /*
  72         * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
  73         * flip its congestion state for events on other blkcgs.
  74         */
  75        if (rl == &rl->q->root_rl)
  76                clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
  77#endif
  78}
  79
  80static void blk_set_congested(struct request_list *rl, int sync)
  81{
  82#ifdef CONFIG_CGROUP_WRITEBACK
  83        set_wb_congested(rl->blkg->wb_congested, sync);
  84#else
  85        /* see blk_clear_congested() */
  86        if (rl == &rl->q->root_rl)
  87                set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
  88#endif
  89}
  90
  91void blk_queue_congestion_threshold(struct request_queue *q)
  92{
  93        int nr;
  94
  95        nr = q->nr_requests - (q->nr_requests / 8) + 1;
  96        if (nr > q->nr_requests)
  97                nr = q->nr_requests;
  98        q->nr_congestion_on = nr;
  99
 100        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
 101        if (nr < 1)
 102                nr = 1;
 103        q->nr_congestion_off = nr;
 104}
 105
 106/**
 107 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
 108 * @bdev:       device
 109 *
 110 * Locates the passed device's request queue and returns the address of its
 111 * backing_dev_info.  This function can only be called if @bdev is opened
 112 * and the return value is never NULL.
 113 */
 114struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
 115{
 116        struct request_queue *q = bdev_get_queue(bdev);
 117
 118        return &q->backing_dev_info;
 119}
 120EXPORT_SYMBOL(blk_get_backing_dev_info);
 121
 122void blk_rq_init(struct request_queue *q, struct request *rq)
 123{
 124        memset(rq, 0, sizeof(*rq));
 125
 126        INIT_LIST_HEAD(&rq->queuelist);
 127        INIT_LIST_HEAD(&rq->timeout_list);
 128        rq->cpu = -1;
 129        rq->q = q;
 130        rq->__sector = (sector_t) -1;
 131        INIT_HLIST_NODE(&rq->hash);
 132        RB_CLEAR_NODE(&rq->rb_node);
 133        rq->cmd = rq->__cmd;
 134        rq->cmd_len = BLK_MAX_CDB;
 135        rq->tag = -1;
 136        rq->start_time = jiffies;
 137        set_start_time_ns(rq);
 138        rq->part = NULL;
 139}
 140EXPORT_SYMBOL(blk_rq_init);
 141
 142static void req_bio_endio(struct request *rq, struct bio *bio,
 143                          unsigned int nbytes, int error)
 144{
 145        if (error)
 146                bio->bi_error = error;
 147
 148        if (unlikely(rq->cmd_flags & REQ_QUIET))
 149                bio_set_flag(bio, BIO_QUIET);
 150
 151        bio_advance(bio, nbytes);
 152
 153        /* don't actually finish bio if it's part of flush sequence */
 154        if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
 155                bio_endio(bio);
 156}
 157
 158void blk_dump_rq_flags(struct request *rq, char *msg)
 159{
 160        int bit;
 161
 162        printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
 163                rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
 164                (unsigned long long) rq->cmd_flags);
 165
 166        printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
 167               (unsigned long long)blk_rq_pos(rq),
 168               blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
 169        printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
 170               rq->bio, rq->biotail, blk_rq_bytes(rq));
 171
 172        if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 173                printk(KERN_INFO "  cdb: ");
 174                for (bit = 0; bit < BLK_MAX_CDB; bit++)
 175                        printk("%02x ", rq->cmd[bit]);
 176                printk("\n");
 177        }
 178}
 179EXPORT_SYMBOL(blk_dump_rq_flags);
 180
 181static void blk_delay_work(struct work_struct *work)
 182{
 183        struct request_queue *q;
 184
 185        q = container_of(work, struct request_queue, delay_work.work);
 186        spin_lock_irq(q->queue_lock);
 187        __blk_run_queue(q);
 188        spin_unlock_irq(q->queue_lock);
 189}
 190
 191/**
 192 * blk_delay_queue - restart queueing after defined interval
 193 * @q:          The &struct request_queue in question
 194 * @msecs:      Delay in msecs
 195 *
 196 * Description:
 197 *   Sometimes queueing needs to be postponed for a little while, to allow
 198 *   resources to come back. This function will make sure that queueing is
 199 *   restarted around the specified time. Queue lock must be held.
 200 */
 201void blk_delay_queue(struct request_queue *q, unsigned long msecs)
 202{
 203        if (likely(!blk_queue_dead(q)))
 204                queue_delayed_work(kblockd_workqueue, &q->delay_work,
 205                                   msecs_to_jiffies(msecs));
 206}
 207EXPORT_SYMBOL(blk_delay_queue);
 208
 209/**
 210 * blk_start_queue_async - asynchronously restart a previously stopped queue
 211 * @q:    The &struct request_queue in question
 212 *
 213 * Description:
 214 *   blk_start_queue_async() will clear the stop flag on the queue, and
 215 *   ensure that the request_fn for the queue is run from an async
 216 *   context.
 217 **/
 218void blk_start_queue_async(struct request_queue *q)
 219{
 220        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 221        blk_run_queue_async(q);
 222}
 223EXPORT_SYMBOL(blk_start_queue_async);
 224
 225/**
 226 * blk_start_queue - restart a previously stopped queue
 227 * @q:    The &struct request_queue in question
 228 *
 229 * Description:
 230 *   blk_start_queue() will clear the stop flag on the queue, and call
 231 *   the request_fn for the queue if it was in a stopped state when
 232 *   entered. Also see blk_stop_queue(). Queue lock must be held.
 233 **/
 234void blk_start_queue(struct request_queue *q)
 235{
 236        WARN_ON(!irqs_disabled());
 237
 238        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 239        __blk_run_queue(q);
 240}
 241EXPORT_SYMBOL(blk_start_queue);
 242
 243/**
 244 * blk_stop_queue - stop a queue
 245 * @q:    The &struct request_queue in question
 246 *
 247 * Description:
 248 *   The Linux block layer assumes that a block driver will consume all
 249 *   entries on the request queue when the request_fn strategy is called.
 250 *   Often this will not happen, because of hardware limitations (queue
 251 *   depth settings). If a device driver gets a 'queue full' response,
 252 *   or if it simply chooses not to queue more I/O at one point, it can
 253 *   call this function to prevent the request_fn from being called until
 254 *   the driver has signalled it's ready to go again. This happens by calling
 255 *   blk_start_queue() to restart queue operations. Queue lock must be held.
 256 **/
 257void blk_stop_queue(struct request_queue *q)
 258{
 259        cancel_delayed_work(&q->delay_work);
 260        queue_flag_set(QUEUE_FLAG_STOPPED, q);
 261}
 262EXPORT_SYMBOL(blk_stop_queue);
 263
 264/**
 265 * blk_sync_queue - cancel any pending callbacks on a queue
 266 * @q: the queue
 267 *
 268 * Description:
 269 *     The block layer may perform asynchronous callback activity
 270 *     on a queue, such as calling the unplug function after a timeout.
 271 *     A block device may call blk_sync_queue to ensure that any
 272 *     such activity is cancelled, thus allowing it to release resources
 273 *     that the callbacks might use. The caller must already have made sure
 274 *     that its ->make_request_fn will not re-add plugging prior to calling
 275 *     this function.
 276 *
 277 *     This function does not cancel any asynchronous activity arising
 278 *     out of elevator or throttling code. That would require elevator_exit()
 279 *     and blkcg_exit_queue() to be called with queue lock initialized.
 280 *
 281 */
 282void blk_sync_queue(struct request_queue *q)
 283{
 284        del_timer_sync(&q->timeout);
 285
 286        if (q->mq_ops) {
 287                struct blk_mq_hw_ctx *hctx;
 288                int i;
 289
 290                queue_for_each_hw_ctx(q, hctx, i) {
 291                        cancel_work_sync(&hctx->run_work);
 292                        cancel_delayed_work_sync(&hctx->delay_work);
 293                }
 294        } else {
 295                cancel_delayed_work_sync(&q->delay_work);
 296        }
 297}
 298EXPORT_SYMBOL(blk_sync_queue);
 299
 300/**
 301 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
 302 * @q:  The queue to run
 303 *
 304 * Description:
 305 *    Invoke request handling on a queue if there are any pending requests.
 306 *    May be used to restart request handling after a request has completed.
 307 *    This variant runs the queue whether or not the queue has been
 308 *    stopped. Must be called with the queue lock held and interrupts
 309 *    disabled. See also @blk_run_queue.
 310 */
 311inline void __blk_run_queue_uncond(struct request_queue *q)
 312{
 313        if (unlikely(blk_queue_dead(q)))
 314                return;
 315
 316        /*
 317         * Some request_fn implementations, e.g. scsi_request_fn(), unlock
 318         * the queue lock internally. As a result multiple threads may be
 319         * running such a request function concurrently. Keep track of the
 320         * number of active request_fn invocations such that blk_drain_queue()
 321         * can wait until all these request_fn calls have finished.
 322         */
 323        q->request_fn_active++;
 324        q->request_fn(q);
 325        q->request_fn_active--;
 326}
 327EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
 328
 329/**
 330 * __blk_run_queue - run a single device queue
 331 * @q:  The queue to run
 332 *
 333 * Description:
 334 *    See @blk_run_queue. This variant must be called with the queue lock
 335 *    held and interrupts disabled.
 336 */
 337void __blk_run_queue(struct request_queue *q)
 338{
 339        if (unlikely(blk_queue_stopped(q)))
 340                return;
 341
 342        __blk_run_queue_uncond(q);
 343}
 344EXPORT_SYMBOL(__blk_run_queue);
 345
 346/**
 347 * blk_run_queue_async - run a single device queue in workqueue context
 348 * @q:  The queue to run
 349 *
 350 * Description:
 351 *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
 352 *    of us. The caller must hold the queue lock.
 353 */
 354void blk_run_queue_async(struct request_queue *q)
 355{
 356        if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
 357                mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
 358}
 359EXPORT_SYMBOL(blk_run_queue_async);
 360
 361/**
 362 * blk_run_queue - run a single device queue
 363 * @q: The queue to run
 364 *
 365 * Description:
 366 *    Invoke request handling on this queue, if it has pending work to do.
 367 *    May be used to restart queueing when a request has completed.
 368 */
 369void blk_run_queue(struct request_queue *q)
 370{
 371        unsigned long flags;
 372
 373        spin_lock_irqsave(q->queue_lock, flags);
 374        __blk_run_queue(q);
 375        spin_unlock_irqrestore(q->queue_lock, flags);
 376}
 377EXPORT_SYMBOL(blk_run_queue);
 378
 379void blk_put_queue(struct request_queue *q)
 380{
 381        kobject_put(&q->kobj);
 382}
 383EXPORT_SYMBOL(blk_put_queue);
 384
 385/**
 386 * __blk_drain_queue - drain requests from request_queue
 387 * @q: queue to drain
 388 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
 389 *
 390 * Drain requests from @q.  If @drain_all is set, all requests are drained.
 391 * If not, only ELVPRIV requests are drained.  The caller is responsible
 392 * for ensuring that no new requests which need to be drained are queued.
 393 */
 394static void __blk_drain_queue(struct request_queue *q, bool drain_all)
 395        __releases(q->queue_lock)
 396        __acquires(q->queue_lock)
 397{
 398        int i;
 399
 400        lockdep_assert_held(q->queue_lock);
 401
 402        while (true) {
 403                bool drain = false;
 404
 405                /*
 406                 * The caller might be trying to drain @q before its
 407                 * elevator is initialized.
 408                 */
 409                if (q->elevator)
 410                        elv_drain_elevator(q);
 411
 412                blkcg_drain_queue(q);
 413
 414                /*
 415                 * This function might be called on a queue which failed
 416                 * driver init after queue creation or is not yet fully
 417                 * active yet.  Some drivers (e.g. fd and loop) get unhappy
 418                 * in such cases.  Kick queue iff dispatch queue has
 419                 * something on it and @q has request_fn set.
 420                 */
 421                if (!list_empty(&q->queue_head) && q->request_fn)
 422                        __blk_run_queue(q);
 423
 424                drain |= q->nr_rqs_elvpriv;
 425                drain |= q->request_fn_active;
 426
 427                /*
 428                 * Unfortunately, requests are queued at and tracked from
 429                 * multiple places and there's no single counter which can
 430                 * be drained.  Check all the queues and counters.
 431                 */
 432                if (drain_all) {
 433                        struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
 434                        drain |= !list_empty(&q->queue_head);
 435                        for (i = 0; i < 2; i++) {
 436                                drain |= q->nr_rqs[i];
 437                                drain |= q->in_flight[i];
 438                                if (fq)
 439                                    drain |= !list_empty(&fq->flush_queue[i]);
 440                        }
 441                }
 442
 443                if (!drain)
 444                        break;
 445
 446                spin_unlock_irq(q->queue_lock);
 447
 448                msleep(10);
 449
 450                spin_lock_irq(q->queue_lock);
 451        }
 452
 453        /*
 454         * With queue marked dead, any woken up waiter will fail the
 455         * allocation path, so the wakeup chaining is lost and we're
 456         * left with hung waiters. We need to wake up those waiters.
 457         */
 458        if (q->request_fn) {
 459                struct request_list *rl;
 460
 461                blk_queue_for_each_rl(rl, q)
 462                        for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
 463                                wake_up_all(&rl->wait[i]);
 464        }
 465}
 466
 467/**
 468 * blk_queue_bypass_start - enter queue bypass mode
 469 * @q: queue of interest
 470 *
 471 * In bypass mode, only the dispatch FIFO queue of @q is used.  This
 472 * function makes @q enter bypass mode and drains all requests which were
 473 * throttled or issued before.  On return, it's guaranteed that no request
 474 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
 475 * inside queue or RCU read lock.
 476 */
 477void blk_queue_bypass_start(struct request_queue *q)
 478{
 479        spin_lock_irq(q->queue_lock);
 480        q->bypass_depth++;
 481        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 482        spin_unlock_irq(q->queue_lock);
 483
 484        /*
 485         * Queues start drained.  Skip actual draining till init is
 486         * complete.  This avoids lenghty delays during queue init which
 487         * can happen many times during boot.
 488         */
 489        if (blk_queue_init_done(q)) {
 490                spin_lock_irq(q->queue_lock);
 491                __blk_drain_queue(q, false);
 492                spin_unlock_irq(q->queue_lock);
 493
 494                /* ensure blk_queue_bypass() is %true inside RCU read lock */
 495                synchronize_rcu();
 496        }
 497}
 498EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
 499
 500/**
 501 * blk_queue_bypass_end - leave queue bypass mode
 502 * @q: queue of interest
 503 *
 504 * Leave bypass mode and restore the normal queueing behavior.
 505 */
 506void blk_queue_bypass_end(struct request_queue *q)
 507{
 508        spin_lock_irq(q->queue_lock);
 509        if (!--q->bypass_depth)
 510                queue_flag_clear(QUEUE_FLAG_BYPASS, q);
 511        WARN_ON_ONCE(q->bypass_depth < 0);
 512        spin_unlock_irq(q->queue_lock);
 513}
 514EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
 515
 516void blk_set_queue_dying(struct request_queue *q)
 517{
 518        spin_lock_irq(q->queue_lock);
 519        queue_flag_set(QUEUE_FLAG_DYING, q);
 520        spin_unlock_irq(q->queue_lock);
 521
 522        if (q->mq_ops)
 523                blk_mq_wake_waiters(q);
 524        else {
 525                struct request_list *rl;
 526
 527                blk_queue_for_each_rl(rl, q) {
 528                        if (rl->rq_pool) {
 529                                wake_up(&rl->wait[BLK_RW_SYNC]);
 530                                wake_up(&rl->wait[BLK_RW_ASYNC]);
 531                        }
 532                }
 533        }
 534}
 535EXPORT_SYMBOL_GPL(blk_set_queue_dying);
 536
 537/**
 538 * blk_cleanup_queue - shutdown a request queue
 539 * @q: request queue to shutdown
 540 *
 541 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
 542 * put it.  All future requests will be failed immediately with -ENODEV.
 543 */
 544void blk_cleanup_queue(struct request_queue *q)
 545{
 546        spinlock_t *lock = q->queue_lock;
 547
 548        /* mark @q DYING, no new request or merges will be allowed afterwards */
 549        mutex_lock(&q->sysfs_lock);
 550        blk_set_queue_dying(q);
 551        spin_lock_irq(lock);
 552
 553        /*
 554         * A dying queue is permanently in bypass mode till released.  Note
 555         * that, unlike blk_queue_bypass_start(), we aren't performing
 556         * synchronize_rcu() after entering bypass mode to avoid the delay
 557         * as some drivers create and destroy a lot of queues while
 558         * probing.  This is still safe because blk_release_queue() will be
 559         * called only after the queue refcnt drops to zero and nothing,
 560         * RCU or not, would be traversing the queue by then.
 561         */
 562        q->bypass_depth++;
 563        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 564
 565        queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 566        queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 567        queue_flag_set(QUEUE_FLAG_DYING, q);
 568        spin_unlock_irq(lock);
 569        mutex_unlock(&q->sysfs_lock);
 570
 571        /*
 572         * Drain all requests queued before DYING marking. Set DEAD flag to
 573         * prevent that q->request_fn() gets invoked after draining finished.
 574         */
 575        blk_freeze_queue(q);
 576        spin_lock_irq(lock);
 577        if (!q->mq_ops)
 578                __blk_drain_queue(q, true);
 579        queue_flag_set(QUEUE_FLAG_DEAD, q);
 580        spin_unlock_irq(lock);
 581
 582        /* for synchronous bio-based driver finish in-flight integrity i/o */
 583        blk_flush_integrity();
 584
 585        /* @q won't process any more request, flush async actions */
 586        del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
 587        blk_sync_queue(q);
 588
 589        if (q->mq_ops)
 590                blk_mq_free_queue(q);
 591        percpu_ref_exit(&q->q_usage_counter);
 592
 593        spin_lock_irq(lock);
 594        if (q->queue_lock != &q->__queue_lock)
 595                q->queue_lock = &q->__queue_lock;
 596        spin_unlock_irq(lock);
 597
 598        bdi_unregister(&q->backing_dev_info);
 599
 600        /* @q is and will stay empty, shutdown and put */
 601        blk_put_queue(q);
 602}
 603EXPORT_SYMBOL(blk_cleanup_queue);
 604
 605/* Allocate memory local to the request queue */
 606static void *alloc_request_struct(gfp_t gfp_mask, void *data)
 607{
 608        int nid = (int)(long)data;
 609        return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
 610}
 611
 612static void free_request_struct(void *element, void *unused)
 613{
 614        kmem_cache_free(request_cachep, element);
 615}
 616
 617int blk_init_rl(struct request_list *rl, struct request_queue *q,
 618                gfp_t gfp_mask)
 619{
 620        if (unlikely(rl->rq_pool))
 621                return 0;
 622
 623        rl->q = q;
 624        rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
 625        rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
 626        init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
 627        init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
 628
 629        rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
 630                                          free_request_struct,
 631                                          (void *)(long)q->node, gfp_mask,
 632                                          q->node);
 633        if (!rl->rq_pool)
 634                return -ENOMEM;
 635
 636        return 0;
 637}
 638
 639void blk_exit_rl(struct request_list *rl)
 640{
 641        if (rl->rq_pool)
 642                mempool_destroy(rl->rq_pool);
 643}
 644
 645struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 646{
 647        return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
 648}
 649EXPORT_SYMBOL(blk_alloc_queue);
 650
 651int blk_queue_enter(struct request_queue *q, bool nowait)
 652{
 653        while (true) {
 654                int ret;
 655
 656                if (percpu_ref_tryget_live(&q->q_usage_counter))
 657                        return 0;
 658
 659                if (nowait)
 660                        return -EBUSY;
 661
 662                ret = wait_event_interruptible(q->mq_freeze_wq,
 663                                !atomic_read(&q->mq_freeze_depth) ||
 664                                blk_queue_dying(q));
 665                if (blk_queue_dying(q))
 666                        return -ENODEV;
 667                if (ret)
 668                        return ret;
 669        }
 670}
 671
 672void blk_queue_exit(struct request_queue *q)
 673{
 674        percpu_ref_put(&q->q_usage_counter);
 675}
 676
 677static void blk_queue_usage_counter_release(struct percpu_ref *ref)
 678{
 679        struct request_queue *q =
 680                container_of(ref, struct request_queue, q_usage_counter);
 681
 682        wake_up_all(&q->mq_freeze_wq);
 683}
 684
 685static void blk_rq_timed_out_timer(unsigned long data)
 686{
 687        struct request_queue *q = (struct request_queue *)data;
 688
 689        kblockd_schedule_work(&q->timeout_work);
 690}
 691
 692struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 693{
 694        struct request_queue *q;
 695        int err;
 696
 697        q = kmem_cache_alloc_node(blk_requestq_cachep,
 698                                gfp_mask | __GFP_ZERO, node_id);
 699        if (!q)
 700                return NULL;
 701
 702        q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
 703        if (q->id < 0)
 704                goto fail_q;
 705
 706        q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
 707        if (!q->bio_split)
 708                goto fail_id;
 709
 710        q->backing_dev_info.ra_pages =
 711                        (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
 712        q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
 713        q->backing_dev_info.name = "block";
 714        q->node = node_id;
 715
 716        err = bdi_init(&q->backing_dev_info);
 717        if (err)
 718                goto fail_split;
 719
 720        setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
 721                    laptop_mode_timer_fn, (unsigned long) q);
 722        setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
 723        INIT_LIST_HEAD(&q->queue_head);
 724        INIT_LIST_HEAD(&q->timeout_list);
 725        INIT_LIST_HEAD(&q->icq_list);
 726#ifdef CONFIG_BLK_CGROUP
 727        INIT_LIST_HEAD(&q->blkg_list);
 728#endif
 729        INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
 730
 731        kobject_init(&q->kobj, &blk_queue_ktype);
 732
 733        mutex_init(&q->sysfs_lock);
 734        spin_lock_init(&q->__queue_lock);
 735
 736        /*
 737         * By default initialize queue_lock to internal lock and driver can
 738         * override it later if need be.
 739         */
 740        q->queue_lock = &q->__queue_lock;
 741
 742        /*
 743         * A queue starts its life with bypass turned on to avoid
 744         * unnecessary bypass on/off overhead and nasty surprises during
 745         * init.  The initial bypass will be finished when the queue is
 746         * registered by blk_register_queue().
 747         */
 748        q->bypass_depth = 1;
 749        __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
 750
 751        init_waitqueue_head(&q->mq_freeze_wq);
 752
 753        /*
 754         * Init percpu_ref in atomic mode so that it's faster to shutdown.
 755         * See blk_register_queue() for details.
 756         */
 757        if (percpu_ref_init(&q->q_usage_counter,
 758                                blk_queue_usage_counter_release,
 759                                PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
 760                goto fail_bdi;
 761
 762        if (blkcg_init_queue(q))
 763                goto fail_ref;
 764
 765        return q;
 766
 767fail_ref:
 768        percpu_ref_exit(&q->q_usage_counter);
 769fail_bdi:
 770        bdi_destroy(&q->backing_dev_info);
 771fail_split:
 772        bioset_free(q->bio_split);
 773fail_id:
 774        ida_simple_remove(&blk_queue_ida, q->id);
 775fail_q:
 776        kmem_cache_free(blk_requestq_cachep, q);
 777        return NULL;
 778}
 779EXPORT_SYMBOL(blk_alloc_queue_node);
 780
 781/**
 782 * blk_init_queue  - prepare a request queue for use with a block device
 783 * @rfn:  The function to be called to process requests that have been
 784 *        placed on the queue.
 785 * @lock: Request queue spin lock
 786 *
 787 * Description:
 788 *    If a block device wishes to use the standard request handling procedures,
 789 *    which sorts requests and coalesces adjacent requests, then it must
 790 *    call blk_init_queue().  The function @rfn will be called when there
 791 *    are requests on the queue that need to be processed.  If the device
 792 *    supports plugging, then @rfn may not be called immediately when requests
 793 *    are available on the queue, but may be called at some time later instead.
 794 *    Plugged queues are generally unplugged when a buffer belonging to one
 795 *    of the requests on the queue is needed, or due to memory pressure.
 796 *
 797 *    @rfn is not required, or even expected, to remove all requests off the
 798 *    queue, but only as many as it can handle at a time.  If it does leave
 799 *    requests on the queue, it is responsible for arranging that the requests
 800 *    get dealt with eventually.
 801 *
 802 *    The queue spin lock must be held while manipulating the requests on the
 803 *    request queue; this lock will be taken also from interrupt context, so irq
 804 *    disabling is needed for it.
 805 *
 806 *    Function returns a pointer to the initialized request queue, or %NULL if
 807 *    it didn't succeed.
 808 *
 809 * Note:
 810 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
 811 *    when the block device is deactivated (such as at module unload).
 812 **/
 813
 814struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
 815{
 816        return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
 817}
 818EXPORT_SYMBOL(blk_init_queue);
 819
 820struct request_queue *
 821blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
 822{
 823        struct request_queue *uninit_q, *q;
 824
 825        uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
 826        if (!uninit_q)
 827                return NULL;
 828
 829        q = blk_init_allocated_queue(uninit_q, rfn, lock);
 830        if (!q)
 831                blk_cleanup_queue(uninit_q);
 832
 833        return q;
 834}
 835EXPORT_SYMBOL(blk_init_queue_node);
 836
 837static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
 838
 839struct request_queue *
 840blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
 841                         spinlock_t *lock)
 842{
 843        if (!q)
 844                return NULL;
 845
 846        q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
 847        if (!q->fq)
 848                return NULL;
 849
 850        if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
 851                goto fail;
 852
 853        INIT_WORK(&q->timeout_work, blk_timeout_work);
 854        q->request_fn           = rfn;
 855        q->prep_rq_fn           = NULL;
 856        q->unprep_rq_fn         = NULL;
 857        q->queue_flags          |= QUEUE_FLAG_DEFAULT;
 858
 859        /* Override internal queue lock with supplied lock pointer */
 860        if (lock)
 861                q->queue_lock           = lock;
 862
 863        /*
 864         * This also sets hw/phys segments, boundary and size
 865         */
 866        blk_queue_make_request(q, blk_queue_bio);
 867
 868        q->sg_reserved_size = INT_MAX;
 869
 870        /* Protect q->elevator from elevator_change */
 871        mutex_lock(&q->sysfs_lock);
 872
 873        /* init elevator */
 874        if (elevator_init(q, NULL)) {
 875                mutex_unlock(&q->sysfs_lock);
 876                goto fail;
 877        }
 878
 879        mutex_unlock(&q->sysfs_lock);
 880
 881        return q;
 882
 883fail:
 884        blk_free_flush_queue(q->fq);
 885        return NULL;
 886}
 887EXPORT_SYMBOL(blk_init_allocated_queue);
 888
 889bool blk_get_queue(struct request_queue *q)
 890{
 891        if (likely(!blk_queue_dying(q))) {
 892                __blk_get_queue(q);
 893                return true;
 894        }
 895
 896        return false;
 897}
 898EXPORT_SYMBOL(blk_get_queue);
 899
 900static inline void blk_free_request(struct request_list *rl, struct request *rq)
 901{
 902        if (rq->cmd_flags & REQ_ELVPRIV) {
 903                elv_put_request(rl->q, rq);
 904                if (rq->elv.icq)
 905                        put_io_context(rq->elv.icq->ioc);
 906        }
 907
 908        mempool_free(rq, rl->rq_pool);
 909}
 910
 911/*
 912 * ioc_batching returns true if the ioc is a valid batching request and
 913 * should be given priority access to a request.
 914 */
 915static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
 916{
 917        if (!ioc)
 918                return 0;
 919
 920        /*
 921         * Make sure the process is able to allocate at least 1 request
 922         * even if the batch times out, otherwise we could theoretically
 923         * lose wakeups.
 924         */
 925        return ioc->nr_batch_requests == q->nr_batching ||
 926                (ioc->nr_batch_requests > 0
 927                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
 928}
 929
 930/*
 931 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
 932 * will cause the process to be a "batcher" on all queues in the system. This
 933 * is the behaviour we want though - once it gets a wakeup it should be given
 934 * a nice run.
 935 */
 936static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
 937{
 938        if (!ioc || ioc_batching(q, ioc))
 939                return;
 940
 941        ioc->nr_batch_requests = q->nr_batching;
 942        ioc->last_waited = jiffies;
 943}
 944
 945static void __freed_request(struct request_list *rl, int sync)
 946{
 947        struct request_queue *q = rl->q;
 948
 949        if (rl->count[sync] < queue_congestion_off_threshold(q))
 950                blk_clear_congested(rl, sync);
 951
 952        if (rl->count[sync] + 1 <= q->nr_requests) {
 953                if (waitqueue_active(&rl->wait[sync]))
 954                        wake_up(&rl->wait[sync]);
 955
 956                blk_clear_rl_full(rl, sync);
 957        }
 958}
 959
 960/*
 961 * A request has just been released.  Account for it, update the full and
 962 * congestion status, wake up any waiters.   Called under q->queue_lock.
 963 */
 964static void freed_request(struct request_list *rl, int op, unsigned int flags)
 965{
 966        struct request_queue *q = rl->q;
 967        int sync = rw_is_sync(op, flags);
 968
 969        q->nr_rqs[sync]--;
 970        rl->count[sync]--;
 971        if (flags & REQ_ELVPRIV)
 972                q->nr_rqs_elvpriv--;
 973
 974        __freed_request(rl, sync);
 975
 976        if (unlikely(rl->starved[sync ^ 1]))
 977                __freed_request(rl, sync ^ 1);
 978}
 979
 980int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
 981{
 982        struct request_list *rl;
 983        int on_thresh, off_thresh;
 984
 985        spin_lock_irq(q->queue_lock);
 986        q->nr_requests = nr;
 987        blk_queue_congestion_threshold(q);
 988        on_thresh = queue_congestion_on_threshold(q);
 989        off_thresh = queue_congestion_off_threshold(q);
 990
 991        blk_queue_for_each_rl(rl, q) {
 992                if (rl->count[BLK_RW_SYNC] >= on_thresh)
 993                        blk_set_congested(rl, BLK_RW_SYNC);
 994                else if (rl->count[BLK_RW_SYNC] < off_thresh)
 995                        blk_clear_congested(rl, BLK_RW_SYNC);
 996
 997                if (rl->count[BLK_RW_ASYNC] >= on_thresh)
 998                        blk_set_congested(rl, BLK_RW_ASYNC);
 999                else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1000                        blk_clear_congested(rl, BLK_RW_ASYNC);
1001
1002                if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1003                        blk_set_rl_full(rl, BLK_RW_SYNC);
1004                } else {
1005                        blk_clear_rl_full(rl, BLK_RW_SYNC);
1006                        wake_up(&rl->wait[BLK_RW_SYNC]);
1007                }
1008
1009                if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1010                        blk_set_rl_full(rl, BLK_RW_ASYNC);
1011                } else {
1012                        blk_clear_rl_full(rl, BLK_RW_ASYNC);
1013                        wake_up(&rl->wait[BLK_RW_ASYNC]);
1014                }
1015        }
1016
1017        spin_unlock_irq(q->queue_lock);
1018        return 0;
1019}
1020
1021/*
1022 * Determine if elevator data should be initialized when allocating the
1023 * request associated with @bio.
1024 */
1025static bool blk_rq_should_init_elevator(struct bio *bio)
1026{
1027        if (!bio)
1028                return true;
1029
1030        /*
1031         * Flush requests do not use the elevator so skip initialization.
1032         * This allows a request to share the flush and elevator data.
1033         */
1034        if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA))
1035                return false;
1036
1037        return true;
1038}
1039
1040/**
1041 * rq_ioc - determine io_context for request allocation
1042 * @bio: request being allocated is for this bio (can be %NULL)
1043 *
1044 * Determine io_context to use for request allocation for @bio.  May return
1045 * %NULL if %current->io_context doesn't exist.
1046 */
1047static struct io_context *rq_ioc(struct bio *bio)
1048{
1049#ifdef CONFIG_BLK_CGROUP
1050        if (bio && bio->bi_ioc)
1051                return bio->bi_ioc;
1052#endif
1053        return current->io_context;
1054}
1055
1056/**
1057 * __get_request - get a free request
1058 * @rl: request list to allocate from
1059 * @op: REQ_OP_READ/REQ_OP_WRITE
1060 * @op_flags: rq_flag_bits
1061 * @bio: bio to allocate request for (can be %NULL)
1062 * @gfp_mask: allocation mask
1063 *
1064 * Get a free request from @q.  This function may fail under memory
1065 * pressure or if @q is dead.
1066 *
1067 * Must be called with @q->queue_lock held and,
1068 * Returns ERR_PTR on failure, with @q->queue_lock held.
1069 * Returns request pointer on success, with @q->queue_lock *not held*.
1070 */
1071static struct request *__get_request(struct request_list *rl, int op,
1072                                     int op_flags, struct bio *bio,
1073                                     gfp_t gfp_mask)
1074{
1075        struct request_queue *q = rl->q;
1076        struct request *rq;
1077        struct elevator_type *et = q->elevator->type;
1078        struct io_context *ioc = rq_ioc(bio);
1079        struct io_cq *icq = NULL;
1080        const bool is_sync = rw_is_sync(op, op_flags) != 0;
1081        int may_queue;
1082
1083        if (unlikely(blk_queue_dying(q)))
1084                return ERR_PTR(-ENODEV);
1085
1086        may_queue = elv_may_queue(q, op, op_flags);
1087        if (may_queue == ELV_MQUEUE_NO)
1088                goto rq_starved;
1089
1090        if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1091                if (rl->count[is_sync]+1 >= q->nr_requests) {
1092                        /*
1093                         * The queue will fill after this allocation, so set
1094                         * it as full, and mark this process as "batching".
1095                         * This process will be allowed to complete a batch of
1096                         * requests, others will be blocked.
1097                         */
1098                        if (!blk_rl_full(rl, is_sync)) {
1099                                ioc_set_batching(q, ioc);
1100                                blk_set_rl_full(rl, is_sync);
1101                        } else {
1102                                if (may_queue != ELV_MQUEUE_MUST
1103                                                && !ioc_batching(q, ioc)) {
1104                                        /*
1105                                         * The queue is full and the allocating
1106                                         * process is not a "batcher", and not
1107                                         * exempted by the IO scheduler
1108                                         */
1109                                        return ERR_PTR(-ENOMEM);
1110                                }
1111                        }
1112                }
1113                blk_set_congested(rl, is_sync);
1114        }
1115
1116        /*
1117         * Only allow batching queuers to allocate up to 50% over the defined
1118         * limit of requests, otherwise we could have thousands of requests
1119         * allocated with any setting of ->nr_requests
1120         */
1121        if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1122                return ERR_PTR(-ENOMEM);
1123
1124        q->nr_rqs[is_sync]++;
1125        rl->count[is_sync]++;
1126        rl->starved[is_sync] = 0;
1127
1128        /*
1129         * Decide whether the new request will be managed by elevator.  If
1130         * so, mark @op_flags and increment elvpriv.  Non-zero elvpriv will
1131         * prevent the current elevator from being destroyed until the new
1132         * request is freed.  This guarantees icq's won't be destroyed and
1133         * makes creating new ones safe.
1134         *
1135         * Also, lookup icq while holding queue_lock.  If it doesn't exist,
1136         * it will be created after releasing queue_lock.
1137         */
1138        if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1139                op_flags |= REQ_ELVPRIV;
1140                q->nr_rqs_elvpriv++;
1141                if (et->icq_cache && ioc)
1142                        icq = ioc_lookup_icq(ioc, q);
1143        }
1144
1145        if (blk_queue_io_stat(q))
1146                op_flags |= REQ_IO_STAT;
1147        spin_unlock_irq(q->queue_lock);
1148
1149        /* allocate and init request */
1150        rq = mempool_alloc(rl->rq_pool, gfp_mask);
1151        if (!rq)
1152                goto fail_alloc;
1153
1154        blk_rq_init(q, rq);
1155        blk_rq_set_rl(rq, rl);
1156        req_set_op_attrs(rq, op, op_flags | REQ_ALLOCED);
1157
1158        /* init elvpriv */
1159        if (op_flags & REQ_ELVPRIV) {
1160                if (unlikely(et->icq_cache && !icq)) {
1161                        if (ioc)
1162                                icq = ioc_create_icq(ioc, q, gfp_mask);
1163                        if (!icq)
1164                                goto fail_elvpriv;
1165                }
1166
1167                rq->elv.icq = icq;
1168                if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1169                        goto fail_elvpriv;
1170
1171                /* @rq->elv.icq holds io_context until @rq is freed */
1172                if (icq)
1173                        get_io_context(icq->ioc);
1174        }
1175out:
1176        /*
1177         * ioc may be NULL here, and ioc_batching will be false. That's
1178         * OK, if the queue is under the request limit then requests need
1179         * not count toward the nr_batch_requests limit. There will always
1180         * be some limit enforced by BLK_BATCH_TIME.
1181         */
1182        if (ioc_batching(q, ioc))
1183                ioc->nr_batch_requests--;
1184
1185        trace_block_getrq(q, bio, op);
1186        return rq;
1187
1188fail_elvpriv:
1189        /*
1190         * elvpriv init failed.  ioc, icq and elvpriv aren't mempool backed
1191         * and may fail indefinitely under memory pressure and thus
1192         * shouldn't stall IO.  Treat this request as !elvpriv.  This will
1193         * disturb iosched and blkcg but weird is bettern than dead.
1194         */
1195        printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1196                           __func__, dev_name(q->backing_dev_info.dev));
1197
1198        rq->cmd_flags &= ~REQ_ELVPRIV;
1199        rq->elv.icq = NULL;
1200
1201        spin_lock_irq(q->queue_lock);
1202        q->nr_rqs_elvpriv--;
1203        spin_unlock_irq(q->queue_lock);
1204        goto out;
1205
1206fail_alloc:
1207        /*
1208         * Allocation failed presumably due to memory. Undo anything we
1209         * might have messed up.
1210         *
1211         * Allocating task should really be put onto the front of the wait
1212         * queue, but this is pretty rare.
1213         */
1214        spin_lock_irq(q->queue_lock);
1215        freed_request(rl, op, op_flags);
1216
1217        /*
1218         * in the very unlikely event that allocation failed and no
1219         * requests for this direction was pending, mark us starved so that
1220         * freeing of a request in the other direction will notice
1221         * us. another possible fix would be to split the rq mempool into
1222         * READ and WRITE
1223         */
1224rq_starved:
1225        if (unlikely(rl->count[is_sync] == 0))
1226                rl->starved[is_sync] = 1;
1227        return ERR_PTR(-ENOMEM);
1228}
1229
1230/**
1231 * get_request - get a free request
1232 * @q: request_queue to allocate request from
1233 * @op: REQ_OP_READ/REQ_OP_WRITE
1234 * @op_flags: rq_flag_bits
1235 * @bio: bio to allocate request for (can be %NULL)
1236 * @gfp_mask: allocation mask
1237 *
1238 * Get a free request from @q.  If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1239 * this function keeps retrying under memory pressure and fails iff @q is dead.
1240 *
1241 * Must be called with @q->queue_lock held and,
1242 * Returns ERR_PTR on failure, with @q->queue_lock held.
1243 * Returns request pointer on success, with @q->queue_lock *not held*.
1244 */
1245static struct request *get_request(struct request_queue *q, int op,
1246                                   int op_flags, struct bio *bio,
1247                                   gfp_t gfp_mask)
1248{
1249        const bool is_sync = rw_is_sync(op, op_flags) != 0;
1250        DEFINE_WAIT(wait);
1251        struct request_list *rl;
1252        struct request *rq;
1253
1254        rl = blk_get_rl(q, bio);        /* transferred to @rq on success */
1255retry:
1256        rq = __get_request(rl, op, op_flags, bio, gfp_mask);
1257        if (!IS_ERR(rq))
1258                return rq;
1259
1260        if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1261                blk_put_rl(rl);
1262                return rq;
1263        }
1264
1265        /* wait on @rl and retry */
1266        prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1267                                  TASK_UNINTERRUPTIBLE);
1268
1269        trace_block_sleeprq(q, bio, op);
1270
1271        spin_unlock_irq(q->queue_lock);
1272        io_schedule();
1273
1274        /*
1275         * After sleeping, we become a "batching" process and will be able
1276         * to allocate at least one request, and up to a big batch of them
1277         * for a small period time.  See ioc_batching, ioc_set_batching
1278         */
1279        ioc_set_batching(q, current->io_context);
1280
1281        spin_lock_irq(q->queue_lock);
1282        finish_wait(&rl->wait[is_sync], &wait);
1283
1284        goto retry;
1285}
1286
1287static struct request *blk_old_get_request(struct request_queue *q, int rw,
1288                gfp_t gfp_mask)
1289{
1290        struct request *rq;
1291
1292        BUG_ON(rw != READ && rw != WRITE);
1293
1294        /* create ioc upfront */
1295        create_io_context(gfp_mask, q->node);
1296
1297        spin_lock_irq(q->queue_lock);
1298        rq = get_request(q, rw, 0, NULL, gfp_mask);
1299        if (IS_ERR(rq)) {
1300                spin_unlock_irq(q->queue_lock);
1301                return rq;
1302        }
1303
1304        /* q->queue_lock is unlocked at this point */
1305        rq->__data_len = 0;
1306        rq->__sector = (sector_t) -1;
1307        rq->bio = rq->biotail = NULL;
1308        return rq;
1309}
1310
1311struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1312{
1313        if (q->mq_ops)
1314                return blk_mq_alloc_request(q, rw,
1315                        (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1316                                0 : BLK_MQ_REQ_NOWAIT);
1317        else
1318                return blk_old_get_request(q, rw, gfp_mask);
1319}
1320EXPORT_SYMBOL(blk_get_request);
1321
1322/**
1323 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1324 * @rq:         request to be initialized
1325 *
1326 */
1327void blk_rq_set_block_pc(struct request *rq)
1328{
1329        rq->cmd_type = REQ_TYPE_BLOCK_PC;
1330        memset(rq->__cmd, 0, sizeof(rq->__cmd));
1331}
1332EXPORT_SYMBOL(blk_rq_set_block_pc);
1333
1334/**
1335 * blk_requeue_request - put a request back on queue
1336 * @q:          request queue where request should be inserted
1337 * @rq:         request to be inserted
1338 *
1339 * Description:
1340 *    Drivers often keep queueing requests until the hardware cannot accept
1341 *    more, when that condition happens we need to put the request back
1342 *    on the queue. Must be called with queue lock held.
1343 */
1344void blk_requeue_request(struct request_queue *q, struct request *rq)
1345{
1346        blk_delete_timer(rq);
1347        blk_clear_rq_complete(rq);
1348        trace_block_rq_requeue(q, rq);
1349
1350        if (rq->cmd_flags & REQ_QUEUED)
1351                blk_queue_end_tag(q, rq);
1352
1353        BUG_ON(blk_queued_rq(rq));
1354
1355        elv_requeue_request(q, rq);
1356}
1357EXPORT_SYMBOL(blk_requeue_request);
1358
1359static void add_acct_request(struct request_queue *q, struct request *rq,
1360                             int where)
1361{
1362        blk_account_io_start(rq, true);
1363        __elv_add_request(q, rq, where);
1364}
1365
1366static void part_round_stats_single(int cpu, struct hd_struct *part,
1367                                    unsigned long now)
1368{
1369        int inflight;
1370
1371        if (now == part->stamp)
1372                return;
1373
1374        inflight = part_in_flight(part);
1375        if (inflight) {
1376                __part_stat_add(cpu, part, time_in_queue,
1377                                inflight * (now - part->stamp));
1378                __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1379        }
1380        part->stamp = now;
1381}
1382
1383/**
1384 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1385 * @cpu: cpu number for stats access
1386 * @part: target partition
1387 *
1388 * The average IO queue length and utilisation statistics are maintained
1389 * by observing the current state of the queue length and the amount of
1390 * time it has been in this state for.
1391 *
1392 * Normally, that accounting is done on IO completion, but that can result
1393 * in more than a second's worth of IO being accounted for within any one
1394 * second, leading to >100% utilisation.  To deal with that, we call this
1395 * function to do a round-off before returning the results when reading
1396 * /proc/diskstats.  This accounts immediately for all queue usage up to
1397 * the current jiffies and restarts the counters again.
1398 */
1399void part_round_stats(int cpu, struct hd_struct *part)
1400{
1401        unsigned long now = jiffies;
1402
1403        if (part->partno)
1404                part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1405        part_round_stats_single(cpu, part, now);
1406}
1407EXPORT_SYMBOL_GPL(part_round_stats);
1408
1409#ifdef CONFIG_PM
1410static void blk_pm_put_request(struct request *rq)
1411{
1412        if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1413                pm_runtime_mark_last_busy(rq->q->dev);
1414}
1415#else
1416static inline void blk_pm_put_request(struct request *rq) {}
1417#endif
1418
1419/*
1420 * queue lock must be held
1421 */
1422void __blk_put_request(struct request_queue *q, struct request *req)
1423{
1424        if (unlikely(!q))
1425                return;
1426
1427        if (q->mq_ops) {
1428                blk_mq_free_request(req);
1429                return;
1430        }
1431
1432        blk_pm_put_request(req);
1433
1434        elv_completed_request(q, req);
1435
1436        /* this is a bio leak */
1437        WARN_ON(req->bio != NULL);
1438
1439        /*
1440         * Request may not have originated from ll_rw_blk. if not,
1441         * it didn't come out of our reserved rq pools
1442         */
1443        if (req->cmd_flags & REQ_ALLOCED) {
1444                unsigned int flags = req->cmd_flags;
1445                int op = req_op(req);
1446                struct request_list *rl = blk_rq_rl(req);
1447
1448                BUG_ON(!list_empty(&req->queuelist));
1449                BUG_ON(ELV_ON_HASH(req));
1450
1451                blk_free_request(rl, req);
1452                freed_request(rl, op, flags);
1453                blk_put_rl(rl);
1454        }
1455}
1456EXPORT_SYMBOL_GPL(__blk_put_request);
1457
1458void blk_put_request(struct request *req)
1459{
1460        struct request_queue *q = req->q;
1461
1462        if (q->mq_ops)
1463                blk_mq_free_request(req);
1464        else {
1465                unsigned long flags;
1466
1467                spin_lock_irqsave(q->queue_lock, flags);
1468                __blk_put_request(q, req);
1469                spin_unlock_irqrestore(q->queue_lock, flags);
1470        }
1471}
1472EXPORT_SYMBOL(blk_put_request);
1473
1474/**
1475 * blk_add_request_payload - add a payload to a request
1476 * @rq: request to update
1477 * @page: page backing the payload
1478 * @offset: offset in page
1479 * @len: length of the payload.
1480 *
1481 * This allows to later add a payload to an already submitted request by
1482 * a block driver.  The driver needs to take care of freeing the payload
1483 * itself.
1484 *
1485 * Note that this is a quite horrible hack and nothing but handling of
1486 * discard requests should ever use it.
1487 */
1488void blk_add_request_payload(struct request *rq, struct page *page,
1489                int offset, unsigned int len)
1490{
1491        struct bio *bio = rq->bio;
1492
1493        bio->bi_io_vec->bv_page = page;
1494        bio->bi_io_vec->bv_offset = offset;
1495        bio->bi_io_vec->bv_len = len;
1496
1497        bio->bi_iter.bi_size = len;
1498        bio->bi_vcnt = 1;
1499        bio->bi_phys_segments = 1;
1500
1501        rq->__data_len = rq->resid_len = len;
1502        rq->nr_phys_segments = 1;
1503}
1504EXPORT_SYMBOL_GPL(blk_add_request_payload);
1505
1506bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1507                            struct bio *bio)
1508{
1509        const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1510
1511        if (!ll_back_merge_fn(q, req, bio))
1512                return false;
1513
1514        trace_block_bio_backmerge(q, req, bio);
1515
1516        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1517                blk_rq_set_mixed_merge(req);
1518
1519        req->biotail->bi_next = bio;
1520        req->biotail = bio;
1521        req->__data_len += bio->bi_iter.bi_size;
1522        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1523
1524        blk_account_io_start(req, false);
1525        return true;
1526}
1527
1528bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1529                             struct bio *bio)
1530{
1531        const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1532
1533        if (!ll_front_merge_fn(q, req, bio))
1534                return false;
1535
1536        trace_block_bio_frontmerge(q, req, bio);
1537
1538        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1539                blk_rq_set_mixed_merge(req);
1540
1541        bio->bi_next = req->bio;
1542        req->bio = bio;
1543
1544        req->__sector = bio->bi_iter.bi_sector;
1545        req->__data_len += bio->bi_iter.bi_size;
1546        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1547
1548        blk_account_io_start(req, false);
1549        return true;
1550}
1551
1552/**
1553 * blk_attempt_plug_merge - try to merge with %current's plugged list
1554 * @q: request_queue new bio is being queued at
1555 * @bio: new bio being queued
1556 * @request_count: out parameter for number of traversed plugged requests
1557 * @same_queue_rq: pointer to &struct request that gets filled in when
1558 * another request associated with @q is found on the plug list
1559 * (optional, may be %NULL)
1560 *
1561 * Determine whether @bio being queued on @q can be merged with a request
1562 * on %current's plugged list.  Returns %true if merge was successful,
1563 * otherwise %false.
1564 *
1565 * Plugging coalesces IOs from the same issuer for the same purpose without
1566 * going through @q->queue_lock.  As such it's more of an issuing mechanism
1567 * than scheduling, and the request, while may have elvpriv data, is not
1568 * added on the elevator at this point.  In addition, we don't have
1569 * reliable access to the elevator outside queue lock.  Only check basic
1570 * merging parameters without querying the elevator.
1571 *
1572 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1573 */
1574bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1575                            unsigned int *request_count,
1576                            struct request **same_queue_rq)
1577{
1578        struct blk_plug *plug;
1579        struct request *rq;
1580        bool ret = false;
1581        struct list_head *plug_list;
1582
1583        plug = current->plug;
1584        if (!plug)
1585                goto out;
1586        *request_count = 0;
1587
1588        if (q->mq_ops)
1589                plug_list = &plug->mq_list;
1590        else
1591                plug_list = &plug->list;
1592
1593        list_for_each_entry_reverse(rq, plug_list, queuelist) {
1594                int el_ret;
1595
1596                if (rq->q == q) {
1597                        (*request_count)++;
1598                        /*
1599                         * Only blk-mq multiple hardware queues case checks the
1600                         * rq in the same queue, there should be only one such
1601                         * rq in a queue
1602                         **/
1603                        if (same_queue_rq)
1604                                *same_queue_rq = rq;
1605                }
1606
1607                if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1608                        continue;
1609
1610                el_ret = blk_try_merge(rq, bio);
1611                if (el_ret == ELEVATOR_BACK_MERGE) {
1612                        ret = bio_attempt_back_merge(q, rq, bio);
1613                        if (ret)
1614                                break;
1615                } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1616                        ret = bio_attempt_front_merge(q, rq, bio);
1617                        if (ret)
1618                                break;
1619                }
1620        }
1621out:
1622        return ret;
1623}
1624
1625unsigned int blk_plug_queued_count(struct request_queue *q)
1626{
1627        struct blk_plug *plug;
1628        struct request *rq;
1629        struct list_head *plug_list;
1630        unsigned int ret = 0;
1631
1632        plug = current->plug;
1633        if (!plug)
1634                goto out;
1635
1636        if (q->mq_ops)
1637                plug_list = &plug->mq_list;
1638        else
1639                plug_list = &plug->list;
1640
1641        list_for_each_entry(rq, plug_list, queuelist) {
1642                if (rq->q == q)
1643                        ret++;
1644        }
1645out:
1646        return ret;
1647}
1648
1649void init_request_from_bio(struct request *req, struct bio *bio)
1650{
1651        req->cmd_type = REQ_TYPE_FS;
1652
1653        req->cmd_flags |= bio->bi_opf & REQ_COMMON_MASK;
1654        if (bio->bi_opf & REQ_RAHEAD)
1655                req->cmd_flags |= REQ_FAILFAST_MASK;
1656
1657        req->errors = 0;
1658        req->__sector = bio->bi_iter.bi_sector;
1659        req->ioprio = bio_prio(bio);
1660        blk_rq_bio_prep(req->q, req, bio);
1661}
1662
1663static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1664{
1665        const bool sync = !!(bio->bi_opf & REQ_SYNC);
1666        struct blk_plug *plug;
1667        int el_ret, rw_flags = 0, where = ELEVATOR_INSERT_SORT;
1668        struct request *req;
1669        unsigned int request_count = 0;
1670
1671        /*
1672         * low level driver can indicate that it wants pages above a
1673         * certain limit bounced to low memory (ie for highmem, or even
1674         * ISA dma in theory)
1675         */
1676        blk_queue_bounce(q, &bio);
1677
1678        blk_queue_split(q, &bio, q->bio_split);
1679
1680        if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1681                bio->bi_error = -EIO;
1682                bio_endio(bio);
1683                return BLK_QC_T_NONE;
1684        }
1685
1686        if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) {
1687                spin_lock_irq(q->queue_lock);
1688                where = ELEVATOR_INSERT_FLUSH;
1689                goto get_rq;
1690        }
1691
1692        /*
1693         * Check if we can merge with the plugged list before grabbing
1694         * any locks.
1695         */
1696        if (!blk_queue_nomerges(q)) {
1697                if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1698                        return BLK_QC_T_NONE;
1699        } else
1700                request_count = blk_plug_queued_count(q);
1701
1702        spin_lock_irq(q->queue_lock);
1703
1704        el_ret = elv_merge(q, &req, bio);
1705        if (el_ret == ELEVATOR_BACK_MERGE) {
1706                if (bio_attempt_back_merge(q, req, bio)) {
1707                        elv_bio_merged(q, req, bio);
1708                        if (!attempt_back_merge(q, req))
1709                                elv_merged_request(q, req, el_ret);
1710                        goto out_unlock;
1711                }
1712        } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1713                if (bio_attempt_front_merge(q, req, bio)) {
1714                        elv_bio_merged(q, req, bio);
1715                        if (!attempt_front_merge(q, req))
1716                                elv_merged_request(q, req, el_ret);
1717                        goto out_unlock;
1718                }
1719        }
1720
1721get_rq:
1722        /*
1723         * This sync check and mask will be re-done in init_request_from_bio(),
1724         * but we need to set it earlier to expose the sync flag to the
1725         * rq allocator and io schedulers.
1726         */
1727        if (sync)
1728                rw_flags |= REQ_SYNC;
1729
1730        /*
1731         * Add in META/PRIO flags, if set, before we get to the IO scheduler
1732         */
1733        rw_flags |= (bio->bi_opf & (REQ_META | REQ_PRIO));
1734
1735        /*
1736         * Grab a free request. This is might sleep but can not fail.
1737         * Returns with the queue unlocked.
1738         */
1739        req = get_request(q, bio_data_dir(bio), rw_flags, bio, GFP_NOIO);
1740        if (IS_ERR(req)) {
1741                bio->bi_error = PTR_ERR(req);
1742                bio_endio(bio);
1743                goto out_unlock;
1744        }
1745
1746        /*
1747         * After dropping the lock and possibly sleeping here, our request
1748         * may now be mergeable after it had proven unmergeable (above).
1749         * We don't worry about that case for efficiency. It won't happen
1750         * often, and the elevators are able to handle it.
1751         */
1752        init_request_from_bio(req, bio);
1753
1754        if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1755                req->cpu = raw_smp_processor_id();
1756
1757        plug = current->plug;
1758        if (plug) {
1759                /*
1760                 * If this is the first request added after a plug, fire
1761                 * of a plug trace.
1762                 */
1763                if (!request_count)
1764                        trace_block_plug(q);
1765                else {
1766                        if (request_count >= BLK_MAX_REQUEST_COUNT) {
1767                                blk_flush_plug_list(plug, false);
1768                                trace_block_plug(q);
1769                        }
1770                }
1771                list_add_tail(&req->queuelist, &plug->list);
1772                blk_account_io_start(req, true);
1773        } else {
1774                spin_lock_irq(q->queue_lock);
1775                add_acct_request(q, req, where);
1776                __blk_run_queue(q);
1777out_unlock:
1778                spin_unlock_irq(q->queue_lock);
1779        }
1780
1781        return BLK_QC_T_NONE;
1782}
1783
1784/*
1785 * If bio->bi_dev is a partition, remap the location
1786 */
1787static inline void blk_partition_remap(struct bio *bio)
1788{
1789        struct block_device *bdev = bio->bi_bdev;
1790
1791        if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1792                struct hd_struct *p = bdev->bd_part;
1793
1794                bio->bi_iter.bi_sector += p->start_sect;
1795                bio->bi_bdev = bdev->bd_contains;
1796
1797                trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1798                                      bdev->bd_dev,
1799                                      bio->bi_iter.bi_sector - p->start_sect);
1800        }
1801}
1802
1803static void handle_bad_sector(struct bio *bio)
1804{
1805        char b[BDEVNAME_SIZE];
1806
1807        printk(KERN_INFO "attempt to access beyond end of device\n");
1808        printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1809                        bdevname(bio->bi_bdev, b),
1810                        bio->bi_opf,
1811                        (unsigned long long)bio_end_sector(bio),
1812                        (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1813}
1814
1815#ifdef CONFIG_FAIL_MAKE_REQUEST
1816
1817static DECLARE_FAULT_ATTR(fail_make_request);
1818
1819static int __init setup_fail_make_request(char *str)
1820{
1821        return setup_fault_attr(&fail_make_request, str);
1822}
1823__setup("fail_make_request=", setup_fail_make_request);
1824
1825static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1826{
1827        return part->make_it_fail && should_fail(&fail_make_request, bytes);
1828}
1829
1830static int __init fail_make_request_debugfs(void)
1831{
1832        struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1833                                                NULL, &fail_make_request);
1834
1835        return PTR_ERR_OR_ZERO(dir);
1836}
1837
1838late_initcall(fail_make_request_debugfs);
1839
1840#else /* CONFIG_FAIL_MAKE_REQUEST */
1841
1842static inline bool should_fail_request(struct hd_struct *part,
1843                                        unsigned int bytes)
1844{
1845        return false;
1846}
1847
1848#endif /* CONFIG_FAIL_MAKE_REQUEST */
1849
1850/*
1851 * Check whether this bio extends beyond the end of the device.
1852 */
1853static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1854{
1855        sector_t maxsector;
1856
1857        if (!nr_sectors)
1858                return 0;
1859
1860        /* Test device or partition size, when known. */
1861        maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1862        if (maxsector) {
1863                sector_t sector = bio->bi_iter.bi_sector;
1864
1865                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1866                        /*
1867                         * This may well happen - the kernel calls bread()
1868                         * without checking the size of the device, e.g., when
1869                         * mounting a device.
1870                         */
1871                        handle_bad_sector(bio);
1872                        return 1;
1873                }
1874        }
1875
1876        return 0;
1877}
1878
1879static noinline_for_stack bool
1880generic_make_request_checks(struct bio *bio)
1881{
1882        struct request_queue *q;
1883        int nr_sectors = bio_sectors(bio);
1884        int err = -EIO;
1885        char b[BDEVNAME_SIZE];
1886        struct hd_struct *part;
1887
1888        might_sleep();
1889
1890        if (bio_check_eod(bio, nr_sectors))
1891                goto end_io;
1892
1893        q = bdev_get_queue(bio->bi_bdev);
1894        if (unlikely(!q)) {
1895                printk(KERN_ERR
1896                       "generic_make_request: Trying to access "
1897                        "nonexistent block-device %s (%Lu)\n",
1898                        bdevname(bio->bi_bdev, b),
1899                        (long long) bio->bi_iter.bi_sector);
1900                goto end_io;
1901        }
1902
1903        part = bio->bi_bdev->bd_part;
1904        if (should_fail_request(part, bio->bi_iter.bi_size) ||
1905            should_fail_request(&part_to_disk(part)->part0,
1906                                bio->bi_iter.bi_size))
1907                goto end_io;
1908
1909        /*
1910         * If this device has partitions, remap block n
1911         * of partition p to block n+start(p) of the disk.
1912         */
1913        blk_partition_remap(bio);
1914
1915        if (bio_check_eod(bio, nr_sectors))
1916                goto end_io;
1917
1918        /*
1919         * Filter flush bio's early so that make_request based
1920         * drivers without flush support don't have to worry
1921         * about them.
1922         */
1923        if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
1924            !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1925                bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1926                if (!nr_sectors) {
1927                        err = 0;
1928                        goto end_io;
1929                }
1930        }
1931
1932        switch (bio_op(bio)) {
1933        case REQ_OP_DISCARD:
1934                if (!blk_queue_discard(q))
1935                        goto not_supported;
1936                break;
1937        case REQ_OP_SECURE_ERASE:
1938                if (!blk_queue_secure_erase(q))
1939                        goto not_supported;
1940                break;
1941        case REQ_OP_WRITE_SAME:
1942                if (!bdev_write_same(bio->bi_bdev))
1943                        goto not_supported;
1944                break;
1945        default:
1946                break;
1947        }
1948
1949        /*
1950         * Various block parts want %current->io_context and lazy ioc
1951         * allocation ends up trading a lot of pain for a small amount of
1952         * memory.  Just allocate it upfront.  This may fail and block
1953         * layer knows how to live with it.
1954         */
1955        create_io_context(GFP_ATOMIC, q->node);
1956
1957        if (!blkcg_bio_issue_check(q, bio))
1958                return false;
1959
1960        trace_block_bio_queue(q, bio);
1961        return true;
1962
1963not_supported:
1964        err = -EOPNOTSUPP;
1965end_io:
1966        bio->bi_error = err;
1967        bio_endio(bio);
1968        return false;
1969}
1970
1971/**
1972 * generic_make_request - hand a buffer to its device driver for I/O
1973 * @bio:  The bio describing the location in memory and on the device.
1974 *
1975 * generic_make_request() is used to make I/O requests of block
1976 * devices. It is passed a &struct bio, which describes the I/O that needs
1977 * to be done.
1978 *
1979 * generic_make_request() does not return any status.  The
1980 * success/failure status of the request, along with notification of
1981 * completion, is delivered asynchronously through the bio->bi_end_io
1982 * function described (one day) else where.
1983 *
1984 * The caller of generic_make_request must make sure that bi_io_vec
1985 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1986 * set to describe the device address, and the
1987 * bi_end_io and optionally bi_private are set to describe how
1988 * completion notification should be signaled.
1989 *
1990 * generic_make_request and the drivers it calls may use bi_next if this
1991 * bio happens to be merged with someone else, and may resubmit the bio to
1992 * a lower device by calling into generic_make_request recursively, which
1993 * means the bio should NOT be touched after the call to ->make_request_fn.
1994 */
1995blk_qc_t generic_make_request(struct bio *bio)
1996{
1997        struct bio_list bio_list_on_stack;
1998        blk_qc_t ret = BLK_QC_T_NONE;
1999
2000        if (!generic_make_request_checks(bio))
2001                goto out;
2002
2003        /*
2004         * We only want one ->make_request_fn to be active at a time, else
2005         * stack usage with stacked devices could be a problem.  So use
2006         * current->bio_list to keep a list of requests submited by a
2007         * make_request_fn function.  current->bio_list is also used as a
2008         * flag to say if generic_make_request is currently active in this
2009         * task or not.  If it is NULL, then no make_request is active.  If
2010         * it is non-NULL, then a make_request is active, and new requests
2011         * should be added at the tail
2012         */
2013        if (current->bio_list) {
2014                bio_list_add(current->bio_list, bio);
2015                goto out;
2016        }
2017
2018        /* following loop may be a bit non-obvious, and so deserves some
2019         * explanation.
2020         * Before entering the loop, bio->bi_next is NULL (as all callers
2021         * ensure that) so we have a list with a single bio.
2022         * We pretend that we have just taken it off a longer list, so
2023         * we assign bio_list to a pointer to the bio_list_on_stack,
2024         * thus initialising the bio_list of new bios to be
2025         * added.  ->make_request() may indeed add some more bios
2026         * through a recursive call to generic_make_request.  If it
2027         * did, we find a non-NULL value in bio_list and re-enter the loop
2028         * from the top.  In this case we really did just take the bio
2029         * of the top of the list (no pretending) and so remove it from
2030         * bio_list, and call into ->make_request() again.
2031         */
2032        BUG_ON(bio->bi_next);
2033        bio_list_init(&bio_list_on_stack);
2034        current->bio_list = &bio_list_on_stack;
2035        do {
2036                struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2037
2038                if (likely(blk_queue_enter(q, false) == 0)) {
2039                        ret = q->make_request_fn(q, bio);
2040
2041                        blk_queue_exit(q);
2042
2043                        bio = bio_list_pop(current->bio_list);
2044                } else {
2045                        struct bio *bio_next = bio_list_pop(current->bio_list);
2046
2047                        bio_io_error(bio);
2048                        bio = bio_next;
2049                }
2050        } while (bio);
2051        current->bio_list = NULL; /* deactivate */
2052
2053out:
2054        return ret;
2055}
2056EXPORT_SYMBOL(generic_make_request);
2057
2058/**
2059 * submit_bio - submit a bio to the block device layer for I/O
2060 * @bio: The &struct bio which describes the I/O
2061 *
2062 * submit_bio() is very similar in purpose to generic_make_request(), and
2063 * uses that function to do most of the work. Both are fairly rough
2064 * interfaces; @bio must be presetup and ready for I/O.
2065 *
2066 */
2067blk_qc_t submit_bio(struct bio *bio)
2068{
2069        /*
2070         * If it's a regular read/write or a barrier with data attached,
2071         * go through the normal accounting stuff before submission.
2072         */
2073        if (bio_has_data(bio)) {
2074                unsigned int count;
2075
2076                if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2077                        count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2078                else
2079                        count = bio_sectors(bio);
2080
2081                if (op_is_write(bio_op(bio))) {
2082                        count_vm_events(PGPGOUT, count);
2083                } else {
2084                        task_io_account_read(bio->bi_iter.bi_size);
2085                        count_vm_events(PGPGIN, count);
2086                }
2087
2088                if (unlikely(block_dump)) {
2089                        char b[BDEVNAME_SIZE];
2090                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2091                        current->comm, task_pid_nr(current),
2092                                op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2093                                (unsigned long long)bio->bi_iter.bi_sector,
2094                                bdevname(bio->bi_bdev, b),
2095                                count);
2096                }
2097        }
2098
2099        return generic_make_request(bio);
2100}
2101EXPORT_SYMBOL(submit_bio);
2102
2103/**
2104 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2105 *                              for new the queue limits
2106 * @q:  the queue
2107 * @rq: the request being checked
2108 *
2109 * Description:
2110 *    @rq may have been made based on weaker limitations of upper-level queues
2111 *    in request stacking drivers, and it may violate the limitation of @q.
2112 *    Since the block layer and the underlying device driver trust @rq
2113 *    after it is inserted to @q, it should be checked against @q before
2114 *    the insertion using this generic function.
2115 *
2116 *    Request stacking drivers like request-based dm may change the queue
2117 *    limits when retrying requests on other queues. Those requests need
2118 *    to be checked against the new queue limits again during dispatch.
2119 */
2120static int blk_cloned_rq_check_limits(struct request_queue *q,
2121                                      struct request *rq)
2122{
2123        if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2124                printk(KERN_ERR "%s: over max size limit.\n", __func__);
2125                return -EIO;
2126        }
2127
2128        /*
2129         * queue's settings related to segment counting like q->bounce_pfn
2130         * may differ from that of other stacking queues.
2131         * Recalculate it to check the request correctly on this queue's
2132         * limitation.
2133         */
2134        blk_recalc_rq_segments(rq);
2135        if (rq->nr_phys_segments > queue_max_segments(q)) {
2136                printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2137                return -EIO;
2138        }
2139
2140        return 0;
2141}
2142
2143/**
2144 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2145 * @q:  the queue to submit the request
2146 * @rq: the request being queued
2147 */
2148int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2149{
2150        unsigned long flags;
2151        int where = ELEVATOR_INSERT_BACK;
2152
2153        if (blk_cloned_rq_check_limits(q, rq))
2154                return -EIO;
2155
2156        if (rq->rq_disk &&
2157            should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2158                return -EIO;
2159
2160        if (q->mq_ops) {
2161                if (blk_queue_io_stat(q))
2162                        blk_account_io_start(rq, true);
2163                blk_mq_insert_request(rq, false, true, false);
2164                return 0;
2165        }
2166
2167        spin_lock_irqsave(q->queue_lock, flags);
2168        if (unlikely(blk_queue_dying(q))) {
2169                spin_unlock_irqrestore(q->queue_lock, flags);
2170                return -ENODEV;
2171        }
2172
2173        /*
2174         * Submitting request must be dequeued before calling this function
2175         * because it will be linked to another request_queue
2176         */
2177        BUG_ON(blk_queued_rq(rq));
2178
2179        if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
2180                where = ELEVATOR_INSERT_FLUSH;
2181
2182        add_acct_request(q, rq, where);
2183        if (where == ELEVATOR_INSERT_FLUSH)
2184                __blk_run_queue(q);
2185        spin_unlock_irqrestore(q->queue_lock, flags);
2186
2187        return 0;
2188}
2189EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2190
2191/**
2192 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2193 * @rq: request to examine
2194 *
2195 * Description:
2196 *     A request could be merge of IOs which require different failure
2197 *     handling.  This function determines the number of bytes which
2198 *     can be failed from the beginning of the request without
2199 *     crossing into area which need to be retried further.
2200 *
2201 * Return:
2202 *     The number of bytes to fail.
2203 *
2204 * Context:
2205 *     queue_lock must be held.
2206 */
2207unsigned int blk_rq_err_bytes(const struct request *rq)
2208{
2209        unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2210        unsigned int bytes = 0;
2211        struct bio *bio;
2212
2213        if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2214                return blk_rq_bytes(rq);
2215
2216        /*
2217         * Currently the only 'mixing' which can happen is between
2218         * different fastfail types.  We can safely fail portions
2219         * which have all the failfast bits that the first one has -
2220         * the ones which are at least as eager to fail as the first
2221         * one.
2222         */
2223        for (bio = rq->bio; bio; bio = bio->bi_next) {
2224                if ((bio->bi_opf & ff) != ff)
2225                        break;
2226                bytes += bio->bi_iter.bi_size;
2227        }
2228
2229        /* this could lead to infinite loop */
2230        BUG_ON(blk_rq_bytes(rq) && !bytes);
2231        return bytes;
2232}
2233EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2234
2235void blk_account_io_completion(struct request *req, unsigned int bytes)
2236{
2237        if (blk_do_io_stat(req)) {
2238                const int rw = rq_data_dir(req);
2239                struct hd_struct *part;
2240                int cpu;
2241
2242                cpu = part_stat_lock();
2243                part = req->part;
2244                part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2245                part_stat_unlock();
2246        }
2247}
2248
2249void blk_account_io_done(struct request *req)
2250{
2251        /*
2252         * Account IO completion.  flush_rq isn't accounted as a
2253         * normal IO on queueing nor completion.  Accounting the
2254         * containing request is enough.
2255         */
2256        if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2257                unsigned long duration = jiffies - req->start_time;
2258                const int rw = rq_data_dir(req);
2259                struct hd_struct *part;
2260                int cpu;
2261
2262                cpu = part_stat_lock();
2263                part = req->part;
2264
2265                part_stat_inc(cpu, part, ios[rw]);
2266                part_stat_add(cpu, part, ticks[rw], duration);
2267                part_round_stats(cpu, part);
2268                part_dec_in_flight(part, rw);
2269
2270                hd_struct_put(part);
2271                part_stat_unlock();
2272        }
2273}
2274
2275#ifdef CONFIG_PM
2276/*
2277 * Don't process normal requests when queue is suspended
2278 * or in the process of suspending/resuming
2279 */
2280static struct request *blk_pm_peek_request(struct request_queue *q,
2281                                           struct request *rq)
2282{
2283        if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2284            (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2285                return NULL;
2286        else
2287                return rq;
2288}
2289#else
2290static inline struct request *blk_pm_peek_request(struct request_queue *q,
2291                                                  struct request *rq)
2292{
2293        return rq;
2294}
2295#endif
2296
2297void blk_account_io_start(struct request *rq, bool new_io)
2298{
2299        struct hd_struct *part;
2300        int rw = rq_data_dir(rq);
2301        int cpu;
2302
2303        if (!blk_do_io_stat(rq))
2304                return;
2305
2306        cpu = part_stat_lock();
2307
2308        if (!new_io) {
2309                part = rq->part;
2310                part_stat_inc(cpu, part, merges[rw]);
2311        } else {
2312                part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2313                if (!hd_struct_try_get(part)) {
2314                        /*
2315                         * The partition is already being removed,
2316                         * the request will be accounted on the disk only
2317                         *
2318                         * We take a reference on disk->part0 although that
2319                         * partition will never be deleted, so we can treat
2320                         * it as any other partition.
2321                         */
2322                        part = &rq->rq_disk->part0;
2323                        hd_struct_get(part);
2324                }
2325                part_round_stats(cpu, part);
2326                part_inc_in_flight(part, rw);
2327                rq->part = part;
2328        }
2329
2330        part_stat_unlock();
2331}
2332
2333/**
2334 * blk_peek_request - peek at the top of a request queue
2335 * @q: request queue to peek at
2336 *
2337 * Description:
2338 *     Return the request at the top of @q.  The returned request
2339 *     should be started using blk_start_request() before LLD starts
2340 *     processing it.
2341 *
2342 * Return:
2343 *     Pointer to the request at the top of @q if available.  Null
2344 *     otherwise.
2345 *
2346 * Context:
2347 *     queue_lock must be held.
2348 */
2349struct request *blk_peek_request(struct request_queue *q)
2350{
2351        struct request *rq;
2352        int ret;
2353
2354        while ((rq = __elv_next_request(q)) != NULL) {
2355
2356                rq = blk_pm_peek_request(q, rq);
2357                if (!rq)
2358                        break;
2359
2360                if (!(rq->cmd_flags & REQ_STARTED)) {
2361                        /*
2362                         * This is the first time the device driver
2363                         * sees this request (possibly after
2364                         * requeueing).  Notify IO scheduler.
2365                         */
2366                        if (rq->cmd_flags & REQ_SORTED)
2367                                elv_activate_rq(q, rq);
2368
2369                        /*
2370                         * just mark as started even if we don't start
2371                         * it, a request that has been delayed should
2372                         * not be passed by new incoming requests
2373                         */
2374                        rq->cmd_flags |= REQ_STARTED;
2375                        trace_block_rq_issue(q, rq);
2376                }
2377
2378                if (!q->boundary_rq || q->boundary_rq == rq) {
2379                        q->end_sector = rq_end_sector(rq);
2380                        q->boundary_rq = NULL;
2381                }
2382
2383                if (rq->cmd_flags & REQ_DONTPREP)
2384                        break;
2385
2386                if (q->dma_drain_size && blk_rq_bytes(rq)) {
2387                        /*
2388                         * make sure space for the drain appears we
2389                         * know we can do this because max_hw_segments
2390                         * has been adjusted to be one fewer than the
2391                         * device can handle
2392                         */
2393                        rq->nr_phys_segments++;
2394                }
2395
2396                if (!q->prep_rq_fn)
2397                        break;
2398
2399                ret = q->prep_rq_fn(q, rq);
2400                if (ret == BLKPREP_OK) {
2401                        break;
2402                } else if (ret == BLKPREP_DEFER) {
2403                        /*
2404                         * the request may have been (partially) prepped.
2405                         * we need to keep this request in the front to
2406                         * avoid resource deadlock.  REQ_STARTED will
2407                         * prevent other fs requests from passing this one.
2408                         */
2409                        if (q->dma_drain_size && blk_rq_bytes(rq) &&
2410                            !(rq->cmd_flags & REQ_DONTPREP)) {
2411                                /*
2412                                 * remove the space for the drain we added
2413                                 * so that we don't add it again
2414                                 */
2415                                --rq->nr_phys_segments;
2416                        }
2417
2418                        rq = NULL;
2419                        break;
2420                } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2421                        int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2422
2423                        rq->cmd_flags |= REQ_QUIET;
2424                        /*
2425                         * Mark this request as started so we don't trigger
2426                         * any debug logic in the end I/O path.
2427                         */
2428                        blk_start_request(rq);
2429                        __blk_end_request_all(rq, err);
2430                } else {
2431                        printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2432                        break;
2433                }
2434        }
2435
2436        return rq;
2437}
2438EXPORT_SYMBOL(blk_peek_request);
2439
2440void blk_dequeue_request(struct request *rq)
2441{
2442        struct request_queue *q = rq->q;
2443
2444        BUG_ON(list_empty(&rq->queuelist));
2445        BUG_ON(ELV_ON_HASH(rq));
2446
2447        list_del_init(&rq->queuelist);
2448
2449        /*
2450         * the time frame between a request being removed from the lists
2451         * and to it is freed is accounted as io that is in progress at
2452         * the driver side.
2453         */
2454        if (blk_account_rq(rq)) {
2455                q->in_flight[rq_is_sync(rq)]++;
2456                set_io_start_time_ns(rq);
2457        }
2458}
2459
2460/**
2461 * blk_start_request - start request processing on the driver
2462 * @req: request to dequeue
2463 *
2464 * Description:
2465 *     Dequeue @req and start timeout timer on it.  This hands off the
2466 *     request to the driver.
2467 *
2468 *     Block internal functions which don't want to start timer should
2469 *     call blk_dequeue_request().
2470 *
2471 * Context:
2472 *     queue_lock must be held.
2473 */
2474void blk_start_request(struct request *req)
2475{
2476        blk_dequeue_request(req);
2477
2478        /*
2479         * We are now handing the request to the hardware, initialize
2480         * resid_len to full count and add the timeout handler.
2481         */
2482        req->resid_len = blk_rq_bytes(req);
2483        if (unlikely(blk_bidi_rq(req)))
2484                req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2485
2486        BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2487        blk_add_timer(req);
2488}
2489EXPORT_SYMBOL(blk_start_request);
2490
2491/**
2492 * blk_fetch_request - fetch a request from a request queue
2493 * @q: request queue to fetch a request from
2494 *
2495 * Description:
2496 *     Return the request at the top of @q.  The request is started on
2497 *     return and LLD can start processing it immediately.
2498 *
2499 * Return:
2500 *     Pointer to the request at the top of @q if available.  Null
2501 *     otherwise.
2502 *
2503 * Context:
2504 *     queue_lock must be held.
2505 */
2506struct request *blk_fetch_request(struct request_queue *q)
2507{
2508        struct request *rq;
2509
2510        rq = blk_peek_request(q);
2511        if (rq)
2512                blk_start_request(rq);
2513        return rq;
2514}
2515EXPORT_SYMBOL(blk_fetch_request);
2516
2517/**
2518 * blk_update_request - Special helper function for request stacking drivers
2519 * @req:      the request being processed
2520 * @error:    %0 for success, < %0 for error
2521 * @nr_bytes: number of bytes to complete @req
2522 *
2523 * Description:
2524 *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2525 *     the request structure even if @req doesn't have leftover.
2526 *     If @req has leftover, sets it up for the next range of segments.
2527 *
2528 *     This special helper function is only for request stacking drivers
2529 *     (e.g. request-based dm) so that they can handle partial completion.
2530 *     Actual device drivers should use blk_end_request instead.
2531 *
2532 *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2533 *     %false return from this function.
2534 *
2535 * Return:
2536 *     %false - this request doesn't have any more data
2537 *     %true  - this request has more data
2538 **/
2539bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2540{
2541        int total_bytes;
2542
2543        trace_block_rq_complete(req->q, req, nr_bytes);
2544
2545        if (!req->bio)
2546                return false;
2547
2548        /*
2549         * For fs requests, rq is just carrier of independent bio's
2550         * and each partial completion should be handled separately.
2551         * Reset per-request error on each partial completion.
2552         *
2553         * TODO: tj: This is too subtle.  It would be better to let
2554         * low level drivers do what they see fit.
2555         */
2556        if (req->cmd_type == REQ_TYPE_FS)
2557                req->errors = 0;
2558
2559        if (error && req->cmd_type == REQ_TYPE_FS &&
2560            !(req->cmd_flags & REQ_QUIET)) {
2561                char *error_type;
2562
2563                switch (error) {
2564                case -ENOLINK:
2565                        error_type = "recoverable transport";
2566                        break;
2567                case -EREMOTEIO:
2568                        error_type = "critical target";
2569                        break;
2570                case -EBADE:
2571                        error_type = "critical nexus";
2572                        break;
2573                case -ETIMEDOUT:
2574                        error_type = "timeout";
2575                        break;
2576                case -ENOSPC:
2577                        error_type = "critical space allocation";
2578                        break;
2579                case -ENODATA:
2580                        error_type = "critical medium";
2581                        break;
2582                case -EIO:
2583                default:
2584                        error_type = "I/O";
2585                        break;
2586                }
2587                printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2588                                   __func__, error_type, req->rq_disk ?
2589                                   req->rq_disk->disk_name : "?",
2590                                   (unsigned long long)blk_rq_pos(req));
2591
2592        }
2593
2594        blk_account_io_completion(req, nr_bytes);
2595
2596        total_bytes = 0;
2597        while (req->bio) {
2598                struct bio *bio = req->bio;
2599                unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2600
2601                if (bio_bytes == bio->bi_iter.bi_size)
2602                        req->bio = bio->bi_next;
2603
2604                req_bio_endio(req, bio, bio_bytes, error);
2605
2606                total_bytes += bio_bytes;
2607                nr_bytes -= bio_bytes;
2608
2609                if (!nr_bytes)
2610                        break;
2611        }
2612
2613        /*
2614         * completely done
2615         */
2616        if (!req->bio) {
2617                /*
2618                 * Reset counters so that the request stacking driver
2619                 * can find how many bytes remain in the request
2620                 * later.
2621                 */
2622                req->__data_len = 0;
2623                return false;
2624        }
2625
2626        req->__data_len -= total_bytes;
2627
2628        /* update sector only for requests with clear definition of sector */
2629        if (req->cmd_type == REQ_TYPE_FS)
2630                req->__sector += total_bytes >> 9;
2631
2632        /* mixed attributes always follow the first bio */
2633        if (req->cmd_flags & REQ_MIXED_MERGE) {
2634                req->cmd_flags &= ~REQ_FAILFAST_MASK;
2635                req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2636        }
2637
2638        /*
2639         * If total number of sectors is less than the first segment
2640         * size, something has gone terribly wrong.
2641         */
2642        if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2643                blk_dump_rq_flags(req, "request botched");
2644                req->__data_len = blk_rq_cur_bytes(req);
2645        }
2646
2647        /* recalculate the number of segments */
2648        blk_recalc_rq_segments(req);
2649
2650        return true;
2651}
2652EXPORT_SYMBOL_GPL(blk_update_request);
2653
2654static bool blk_update_bidi_request(struct request *rq, int error,
2655                                    unsigned int nr_bytes,
2656                                    unsigned int bidi_bytes)
2657{
2658        if (blk_update_request(rq, error, nr_bytes))
2659                return true;
2660
2661        /* Bidi request must be completed as a whole */
2662        if (unlikely(blk_bidi_rq(rq)) &&
2663            blk_update_request(rq->next_rq, error, bidi_bytes))
2664                return true;
2665
2666        if (blk_queue_add_random(rq->q))
2667                add_disk_randomness(rq->rq_disk);
2668
2669        return false;
2670}
2671
2672/**
2673 * blk_unprep_request - unprepare a request
2674 * @req:        the request
2675 *
2676 * This function makes a request ready for complete resubmission (or
2677 * completion).  It happens only after all error handling is complete,
2678 * so represents the appropriate moment to deallocate any resources
2679 * that were allocated to the request in the prep_rq_fn.  The queue
2680 * lock is held when calling this.
2681 */
2682void blk_unprep_request(struct request *req)
2683{
2684        struct request_queue *q = req->q;
2685
2686        req->cmd_flags &= ~REQ_DONTPREP;
2687        if (q->unprep_rq_fn)
2688                q->unprep_rq_fn(q, req);
2689}
2690EXPORT_SYMBOL_GPL(blk_unprep_request);
2691
2692/*
2693 * queue lock must be held
2694 */
2695void blk_finish_request(struct request *req, int error)
2696{
2697        if (req->cmd_flags & REQ_QUEUED)
2698                blk_queue_end_tag(req->q, req);
2699
2700        BUG_ON(blk_queued_rq(req));
2701
2702        if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2703                laptop_io_completion(&req->q->backing_dev_info);
2704
2705        blk_delete_timer(req);
2706
2707        if (req->cmd_flags & REQ_DONTPREP)
2708                blk_unprep_request(req);
2709
2710        blk_account_io_done(req);
2711
2712        if (req->end_io)
2713                req->end_io(req, error);
2714        else {
2715                if (blk_bidi_rq(req))
2716                        __blk_put_request(req->next_rq->q, req->next_rq);
2717
2718                __blk_put_request(req->q, req);
2719        }
2720}
2721EXPORT_SYMBOL(blk_finish_request);
2722
2723/**
2724 * blk_end_bidi_request - Complete a bidi request
2725 * @rq:         the request to complete
2726 * @error:      %0 for success, < %0 for error
2727 * @nr_bytes:   number of bytes to complete @rq
2728 * @bidi_bytes: number of bytes to complete @rq->next_rq
2729 *
2730 * Description:
2731 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2732 *     Drivers that supports bidi can safely call this member for any
2733 *     type of request, bidi or uni.  In the later case @bidi_bytes is
2734 *     just ignored.
2735 *
2736 * Return:
2737 *     %false - we are done with this request
2738 *     %true  - still buffers pending for this request
2739 **/
2740static bool blk_end_bidi_request(struct request *rq, int error,
2741                                 unsigned int nr_bytes, unsigned int bidi_bytes)
2742{
2743        struct request_queue *q = rq->q;
2744        unsigned long flags;
2745
2746        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2747                return true;
2748
2749        spin_lock_irqsave(q->queue_lock, flags);
2750        blk_finish_request(rq, error);
2751        spin_unlock_irqrestore(q->queue_lock, flags);
2752
2753        return false;
2754}
2755
2756/**
2757 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2758 * @rq:         the request to complete
2759 * @error:      %0 for success, < %0 for error
2760 * @nr_bytes:   number of bytes to complete @rq
2761 * @bidi_bytes: number of bytes to complete @rq->next_rq
2762 *
2763 * Description:
2764 *     Identical to blk_end_bidi_request() except that queue lock is
2765 *     assumed to be locked on entry and remains so on return.
2766 *
2767 * Return:
2768 *     %false - we are done with this request
2769 *     %true  - still buffers pending for this request
2770 **/
2771bool __blk_end_bidi_request(struct request *rq, int error,
2772                                   unsigned int nr_bytes, unsigned int bidi_bytes)
2773{
2774        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2775                return true;
2776
2777        blk_finish_request(rq, error);
2778
2779        return false;
2780}
2781
2782/**
2783 * blk_end_request - Helper function for drivers to complete the request.
2784 * @rq:       the request being processed
2785 * @error:    %0 for success, < %0 for error
2786 * @nr_bytes: number of bytes to complete
2787 *
2788 * Description:
2789 *     Ends I/O on a number of bytes attached to @rq.
2790 *     If @rq has leftover, sets it up for the next range of segments.
2791 *
2792 * Return:
2793 *     %false - we are done with this request
2794 *     %true  - still buffers pending for this request
2795 **/
2796bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2797{
2798        return blk_end_bidi_request(rq, error, nr_bytes, 0);
2799}
2800EXPORT_SYMBOL(blk_end_request);
2801
2802/**
2803 * blk_end_request_all - Helper function for drives to finish the request.
2804 * @rq: the request to finish
2805 * @error: %0 for success, < %0 for error
2806 *
2807 * Description:
2808 *     Completely finish @rq.
2809 */
2810void blk_end_request_all(struct request *rq, int error)
2811{
2812        bool pending;
2813        unsigned int bidi_bytes = 0;
2814
2815        if (unlikely(blk_bidi_rq(rq)))
2816                bidi_bytes = blk_rq_bytes(rq->next_rq);
2817
2818        pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2819        BUG_ON(pending);
2820}
2821EXPORT_SYMBOL(blk_end_request_all);
2822
2823/**
2824 * blk_end_request_cur - Helper function to finish the current request chunk.
2825 * @rq: the request to finish the current chunk for
2826 * @error: %0 for success, < %0 for error
2827 *
2828 * Description:
2829 *     Complete the current consecutively mapped chunk from @rq.
2830 *
2831 * Return:
2832 *     %false - we are done with this request
2833 *     %true  - still buffers pending for this request
2834 */
2835bool blk_end_request_cur(struct request *rq, int error)
2836{
2837        return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2838}
2839EXPORT_SYMBOL(blk_end_request_cur);
2840
2841/**
2842 * blk_end_request_err - Finish a request till the next failure boundary.
2843 * @rq: the request to finish till the next failure boundary for
2844 * @error: must be negative errno
2845 *
2846 * Description:
2847 *     Complete @rq till the next failure boundary.
2848 *
2849 * Return:
2850 *     %false - we are done with this request
2851 *     %true  - still buffers pending for this request
2852 */
2853bool blk_end_request_err(struct request *rq, int error)
2854{
2855        WARN_ON(error >= 0);
2856        return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2857}
2858EXPORT_SYMBOL_GPL(blk_end_request_err);
2859
2860/**
2861 * __blk_end_request - Helper function for drivers to complete the request.
2862 * @rq:       the request being processed
2863 * @error:    %0 for success, < %0 for error
2864 * @nr_bytes: number of bytes to complete
2865 *
2866 * Description:
2867 *     Must be called with queue lock held unlike blk_end_request().
2868 *
2869 * Return:
2870 *     %false - we are done with this request
2871 *     %true  - still buffers pending for this request
2872 **/
2873bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2874{
2875        return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2876}
2877EXPORT_SYMBOL(__blk_end_request);
2878
2879/**
2880 * __blk_end_request_all - Helper function for drives to finish the request.
2881 * @rq: the request to finish
2882 * @error: %0 for success, < %0 for error
2883 *
2884 * Description:
2885 *     Completely finish @rq.  Must be called with queue lock held.
2886 */
2887void __blk_end_request_all(struct request *rq, int error)
2888{
2889        bool pending;
2890        unsigned int bidi_bytes = 0;
2891
2892        if (unlikely(blk_bidi_rq(rq)))
2893                bidi_bytes = blk_rq_bytes(rq->next_rq);
2894
2895        pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2896        BUG_ON(pending);
2897}
2898EXPORT_SYMBOL(__blk_end_request_all);
2899
2900/**
2901 * __blk_end_request_cur - Helper function to finish the current request chunk.
2902 * @rq: the request to finish the current chunk for
2903 * @error: %0 for success, < %0 for error
2904 *
2905 * Description:
2906 *     Complete the current consecutively mapped chunk from @rq.  Must
2907 *     be called with queue lock held.
2908 *
2909 * Return:
2910 *     %false - we are done with this request
2911 *     %true  - still buffers pending for this request
2912 */
2913bool __blk_end_request_cur(struct request *rq, int error)
2914{
2915        return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2916}
2917EXPORT_SYMBOL(__blk_end_request_cur);
2918
2919/**
2920 * __blk_end_request_err - Finish a request till the next failure boundary.
2921 * @rq: the request to finish till the next failure boundary for
2922 * @error: must be negative errno
2923 *
2924 * Description:
2925 *     Complete @rq till the next failure boundary.  Must be called
2926 *     with queue lock held.
2927 *
2928 * Return:
2929 *     %false - we are done with this request
2930 *     %true  - still buffers pending for this request
2931 */
2932bool __blk_end_request_err(struct request *rq, int error)
2933{
2934        WARN_ON(error >= 0);
2935        return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2936}
2937EXPORT_SYMBOL_GPL(__blk_end_request_err);
2938
2939void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2940                     struct bio *bio)
2941{
2942        req_set_op(rq, bio_op(bio));
2943
2944        if (bio_has_data(bio))
2945                rq->nr_phys_segments = bio_phys_segments(q, bio);
2946
2947        rq->__data_len = bio->bi_iter.bi_size;
2948        rq->bio = rq->biotail = bio;
2949
2950        if (bio->bi_bdev)
2951                rq->rq_disk = bio->bi_bdev->bd_disk;
2952}
2953
2954#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2955/**
2956 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2957 * @rq: the request to be flushed
2958 *
2959 * Description:
2960 *     Flush all pages in @rq.
2961 */
2962void rq_flush_dcache_pages(struct request *rq)
2963{
2964        struct req_iterator iter;
2965        struct bio_vec bvec;
2966
2967        rq_for_each_segment(bvec, rq, iter)
2968                flush_dcache_page(bvec.bv_page);
2969}
2970EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2971#endif
2972
2973/**
2974 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2975 * @q : the queue of the device being checked
2976 *
2977 * Description:
2978 *    Check if underlying low-level drivers of a device are busy.
2979 *    If the drivers want to export their busy state, they must set own
2980 *    exporting function using blk_queue_lld_busy() first.
2981 *
2982 *    Basically, this function is used only by request stacking drivers
2983 *    to stop dispatching requests to underlying devices when underlying
2984 *    devices are busy.  This behavior helps more I/O merging on the queue
2985 *    of the request stacking driver and prevents I/O throughput regression
2986 *    on burst I/O load.
2987 *
2988 * Return:
2989 *    0 - Not busy (The request stacking driver should dispatch request)
2990 *    1 - Busy (The request stacking driver should stop dispatching request)
2991 */
2992int blk_lld_busy(struct request_queue *q)
2993{
2994        if (q->lld_busy_fn)
2995                return q->lld_busy_fn(q);
2996
2997        return 0;
2998}
2999EXPORT_SYMBOL_GPL(blk_lld_busy);
3000
3001/**
3002 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3003 * @rq: the clone request to be cleaned up
3004 *
3005 * Description:
3006 *     Free all bios in @rq for a cloned request.
3007 */
3008void blk_rq_unprep_clone(struct request *rq)
3009{
3010        struct bio *bio;
3011
3012        while ((bio = rq->bio) != NULL) {
3013                rq->bio = bio->bi_next;
3014
3015                bio_put(bio);
3016        }
3017}
3018EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3019
3020/*
3021 * Copy attributes of the original request to the clone request.
3022 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3023 */
3024static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3025{
3026        dst->cpu = src->cpu;
3027        req_set_op_attrs(dst, req_op(src),
3028                         (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE);
3029        dst->cmd_type = src->cmd_type;
3030        dst->__sector = blk_rq_pos(src);
3031        dst->__data_len = blk_rq_bytes(src);
3032        dst->nr_phys_segments = src->nr_phys_segments;
3033        dst->ioprio = src->ioprio;
3034        dst->extra_len = src->extra_len;
3035}
3036
3037/**
3038 * blk_rq_prep_clone - Helper function to setup clone request
3039 * @rq: the request to be setup
3040 * @rq_src: original request to be cloned
3041 * @bs: bio_set that bios for clone are allocated from
3042 * @gfp_mask: memory allocation mask for bio
3043 * @bio_ctr: setup function to be called for each clone bio.
3044 *           Returns %0 for success, non %0 for failure.
3045 * @data: private data to be passed to @bio_ctr
3046 *
3047 * Description:
3048 *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3049 *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3050 *     are not copied, and copying such parts is the caller's responsibility.
3051 *     Also, pages which the original bios are pointing to are not copied
3052 *     and the cloned bios just point same pages.
3053 *     So cloned bios must be completed before original bios, which means
3054 *     the caller must complete @rq before @rq_src.
3055 */
3056int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3057                      struct bio_set *bs, gfp_t gfp_mask,
3058                      int (*bio_ctr)(struct bio *, struct bio *, void *),
3059                      void *data)
3060{
3061        struct bio *bio, *bio_src;
3062
3063        if (!bs)
3064                bs = fs_bio_set;
3065
3066        __rq_for_each_bio(bio_src, rq_src) {
3067                bio = bio_clone_fast(bio_src, gfp_mask, bs);
3068                if (!bio)
3069                        goto free_and_out;
3070
3071                if (bio_ctr && bio_ctr(bio, bio_src, data))
3072                        goto free_and_out;
3073
3074                if (rq->bio) {
3075                        rq->biotail->bi_next = bio;
3076                        rq->biotail = bio;
3077                } else
3078                        rq->bio = rq->biotail = bio;
3079        }
3080
3081        __blk_rq_prep_clone(rq, rq_src);
3082
3083        return 0;
3084
3085free_and_out:
3086        if (bio)
3087                bio_put(bio);
3088        blk_rq_unprep_clone(rq);
3089
3090        return -ENOMEM;
3091}
3092EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3093
3094int kblockd_schedule_work(struct work_struct *work)
3095{
3096        return queue_work(kblockd_workqueue, work);
3097}
3098EXPORT_SYMBOL(kblockd_schedule_work);
3099
3100int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3101{
3102        return queue_work_on(cpu, kblockd_workqueue, work);
3103}
3104EXPORT_SYMBOL(kblockd_schedule_work_on);
3105
3106int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3107                                  unsigned long delay)
3108{
3109        return queue_delayed_work(kblockd_workqueue, dwork, delay);
3110}
3111EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3112
3113int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3114                                     unsigned long delay)
3115{
3116        return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3117}
3118EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3119
3120/**
3121 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3122 * @plug:       The &struct blk_plug that needs to be initialized
3123 *
3124 * Description:
3125 *   Tracking blk_plug inside the task_struct will help with auto-flushing the
3126 *   pending I/O should the task end up blocking between blk_start_plug() and
3127 *   blk_finish_plug(). This is important from a performance perspective, but
3128 *   also ensures that we don't deadlock. For instance, if the task is blocking
3129 *   for a memory allocation, memory reclaim could end up wanting to free a
3130 *   page belonging to that request that is currently residing in our private
3131 *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
3132 *   this kind of deadlock.
3133 */
3134void blk_start_plug(struct blk_plug *plug)
3135{
3136        struct task_struct *tsk = current;
3137
3138        /*
3139         * If this is a nested plug, don't actually assign it.
3140         */
3141        if (tsk->plug)
3142                return;
3143
3144        INIT_LIST_HEAD(&plug->list);
3145        INIT_LIST_HEAD(&plug->mq_list);
3146        INIT_LIST_HEAD(&plug->cb_list);
3147        /*
3148         * Store ordering should not be needed here, since a potential
3149         * preempt will imply a full memory barrier
3150         */
3151        tsk->plug = plug;
3152}
3153EXPORT_SYMBOL(blk_start_plug);
3154
3155static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3156{
3157        struct request *rqa = container_of(a, struct request, queuelist);
3158        struct request *rqb = container_of(b, struct request, queuelist);
3159
3160        return !(rqa->q < rqb->q ||
3161                (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3162}
3163
3164/*
3165 * If 'from_schedule' is true, then postpone the dispatch of requests
3166 * until a safe kblockd context. We due this to avoid accidental big
3167 * additional stack usage in driver dispatch, in places where the originally
3168 * plugger did not intend it.
3169 */
3170static void queue_unplugged(struct request_queue *q, unsigned int depth,
3171                            bool from_schedule)
3172        __releases(q->queue_lock)
3173{
3174        trace_block_unplug(q, depth, !from_schedule);
3175
3176        if (from_schedule)
3177                blk_run_queue_async(q);
3178        else
3179                __blk_run_queue(q);
3180        spin_unlock(q->queue_lock);
3181}
3182
3183static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3184{
3185        LIST_HEAD(callbacks);
3186
3187        while (!list_empty(&plug->cb_list)) {
3188                list_splice_init(&plug->cb_list, &callbacks);
3189
3190                while (!list_empty(&callbacks)) {
3191                        struct blk_plug_cb *cb = list_first_entry(&callbacks,
3192                                                          struct blk_plug_cb,
3193                                                          list);
3194                        list_del(&cb->list);
3195                        cb->callback(cb, from_schedule);
3196                }
3197        }
3198}
3199
3200struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3201                                      int size)
3202{
3203        struct blk_plug *plug = current->plug;
3204        struct blk_plug_cb *cb;
3205
3206        if (!plug)
3207                return NULL;
3208
3209        list_for_each_entry(cb, &plug->cb_list, list)
3210                if (cb->callback == unplug && cb->data == data)
3211                        return cb;
3212
3213        /* Not currently on the callback list */
3214        BUG_ON(size < sizeof(*cb));
3215        cb = kzalloc(size, GFP_ATOMIC);
3216        if (cb) {
3217                cb->data = data;
3218                cb->callback = unplug;
3219                list_add(&cb->list, &plug->cb_list);
3220        }
3221        return cb;
3222}
3223EXPORT_SYMBOL(blk_check_plugged);
3224
3225void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3226{
3227        struct request_queue *q;
3228        unsigned long flags;
3229        struct request *rq;
3230        LIST_HEAD(list);
3231        unsigned int depth;
3232
3233        flush_plug_callbacks(plug, from_schedule);
3234
3235        if (!list_empty(&plug->mq_list))
3236                blk_mq_flush_plug_list(plug, from_schedule);
3237
3238        if (list_empty(&plug->list))
3239                return;
3240
3241        list_splice_init(&plug->list, &list);
3242
3243        list_sort(NULL, &list, plug_rq_cmp);
3244
3245        q = NULL;
3246        depth = 0;
3247
3248        /*
3249         * Save and disable interrupts here, to avoid doing it for every
3250         * queue lock we have to take.
3251         */
3252        local_irq_save(flags);
3253        while (!list_empty(&list)) {
3254                rq = list_entry_rq(list.next);
3255                list_del_init(&rq->queuelist);
3256                BUG_ON(!rq->q);
3257                if (rq->q != q) {
3258                        /*
3259                         * This drops the queue lock
3260                         */
3261                        if (q)
3262                                queue_unplugged(q, depth, from_schedule);
3263                        q = rq->q;
3264                        depth = 0;
3265                        spin_lock(q->queue_lock);
3266                }
3267
3268                /*
3269                 * Short-circuit if @q is dead
3270                 */
3271                if (unlikely(blk_queue_dying(q))) {
3272                        __blk_end_request_all(rq, -ENODEV);
3273                        continue;
3274                }
3275
3276                /*
3277                 * rq is already accounted, so use raw insert
3278                 */
3279                if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
3280                        __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3281                else
3282                        __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3283
3284                depth++;
3285        }
3286
3287        /*
3288         * This drops the queue lock
3289         */
3290        if (q)
3291                queue_unplugged(q, depth, from_schedule);
3292
3293        local_irq_restore(flags);
3294}
3295
3296void blk_finish_plug(struct blk_plug *plug)
3297{
3298        if (plug != current->plug)
3299                return;
3300        blk_flush_plug_list(plug, false);
3301
3302        current->plug = NULL;
3303}
3304EXPORT_SYMBOL(blk_finish_plug);
3305
3306bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3307{
3308        struct blk_plug *plug;
3309        long state;
3310        unsigned int queue_num;
3311        struct blk_mq_hw_ctx *hctx;
3312
3313        if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3314            !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3315                return false;
3316
3317        queue_num = blk_qc_t_to_queue_num(cookie);
3318        hctx = q->queue_hw_ctx[queue_num];
3319        hctx->poll_considered++;
3320
3321        plug = current->plug;
3322        if (plug)
3323                blk_flush_plug_list(plug, false);
3324
3325        state = current->state;
3326        while (!need_resched()) {
3327                int ret;
3328
3329                hctx->poll_invoked++;
3330
3331                ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3332                if (ret > 0) {
3333                        hctx->poll_success++;
3334                        set_current_state(TASK_RUNNING);
3335                        return true;
3336                }
3337
3338                if (signal_pending_state(state, current))
3339                        set_current_state(TASK_RUNNING);
3340
3341                if (current->state == TASK_RUNNING)
3342                        return true;
3343                if (ret < 0)
3344                        break;
3345                cpu_relax();
3346        }
3347
3348        return false;
3349}
3350EXPORT_SYMBOL_GPL(blk_poll);
3351
3352#ifdef CONFIG_PM
3353/**
3354 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3355 * @q: the queue of the device
3356 * @dev: the device the queue belongs to
3357 *
3358 * Description:
3359 *    Initialize runtime-PM-related fields for @q and start auto suspend for
3360 *    @dev. Drivers that want to take advantage of request-based runtime PM
3361 *    should call this function after @dev has been initialized, and its
3362 *    request queue @q has been allocated, and runtime PM for it can not happen
3363 *    yet(either due to disabled/forbidden or its usage_count > 0). In most
3364 *    cases, driver should call this function before any I/O has taken place.
3365 *
3366 *    This function takes care of setting up using auto suspend for the device,
3367 *    the autosuspend delay is set to -1 to make runtime suspend impossible
3368 *    until an updated value is either set by user or by driver. Drivers do
3369 *    not need to touch other autosuspend settings.
3370 *
3371 *    The block layer runtime PM is request based, so only works for drivers
3372 *    that use request as their IO unit instead of those directly use bio's.
3373 */
3374void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3375{
3376        q->dev = dev;
3377        q->rpm_status = RPM_ACTIVE;
3378        pm_runtime_set_autosuspend_delay(q->dev, -1);
3379        pm_runtime_use_autosuspend(q->dev);
3380}
3381EXPORT_SYMBOL(blk_pm_runtime_init);
3382
3383/**
3384 * blk_pre_runtime_suspend - Pre runtime suspend check
3385 * @q: the queue of the device
3386 *
3387 * Description:
3388 *    This function will check if runtime suspend is allowed for the device
3389 *    by examining if there are any requests pending in the queue. If there
3390 *    are requests pending, the device can not be runtime suspended; otherwise,
3391 *    the queue's status will be updated to SUSPENDING and the driver can
3392 *    proceed to suspend the device.
3393 *
3394 *    For the not allowed case, we mark last busy for the device so that
3395 *    runtime PM core will try to autosuspend it some time later.
3396 *
3397 *    This function should be called near the start of the device's
3398 *    runtime_suspend callback.
3399 *
3400 * Return:
3401 *    0         - OK to runtime suspend the device
3402 *    -EBUSY    - Device should not be runtime suspended
3403 */
3404int blk_pre_runtime_suspend(struct request_queue *q)
3405{
3406        int ret = 0;
3407
3408        if (!q->dev)
3409                return ret;
3410
3411        spin_lock_irq(q->queue_lock);
3412        if (q->nr_pending) {
3413                ret = -EBUSY;
3414                pm_runtime_mark_last_busy(q->dev);
3415        } else {
3416                q->rpm_status = RPM_SUSPENDING;
3417        }
3418        spin_unlock_irq(q->queue_lock);
3419        return ret;
3420}
3421EXPORT_SYMBOL(blk_pre_runtime_suspend);
3422
3423/**
3424 * blk_post_runtime_suspend - Post runtime suspend processing
3425 * @q: the queue of the device
3426 * @err: return value of the device's runtime_suspend function
3427 *
3428 * Description:
3429 *    Update the queue's runtime status according to the return value of the
3430 *    device's runtime suspend function and mark last busy for the device so
3431 *    that PM core will try to auto suspend the device at a later time.
3432 *
3433 *    This function should be called near the end of the device's
3434 *    runtime_suspend callback.
3435 */
3436void blk_post_runtime_suspend(struct request_queue *q, int err)
3437{
3438        if (!q->dev)
3439                return;
3440
3441        spin_lock_irq(q->queue_lock);
3442        if (!err) {
3443                q->rpm_status = RPM_SUSPENDED;
3444        } else {
3445                q->rpm_status = RPM_ACTIVE;
3446                pm_runtime_mark_last_busy(q->dev);
3447        }
3448        spin_unlock_irq(q->queue_lock);
3449}
3450EXPORT_SYMBOL(blk_post_runtime_suspend);
3451
3452/**
3453 * blk_pre_runtime_resume - Pre runtime resume processing
3454 * @q: the queue of the device
3455 *
3456 * Description:
3457 *    Update the queue's runtime status to RESUMING in preparation for the
3458 *    runtime resume of the device.
3459 *
3460 *    This function should be called near the start of the device's
3461 *    runtime_resume callback.
3462 */
3463void blk_pre_runtime_resume(struct request_queue *q)
3464{
3465        if (!q->dev)
3466                return;
3467
3468        spin_lock_irq(q->queue_lock);
3469        q->rpm_status = RPM_RESUMING;
3470        spin_unlock_irq(q->queue_lock);
3471}
3472EXPORT_SYMBOL(blk_pre_runtime_resume);
3473
3474/**
3475 * blk_post_runtime_resume - Post runtime resume processing
3476 * @q: the queue of the device
3477 * @err: return value of the device's runtime_resume function
3478 *
3479 * Description:
3480 *    Update the queue's runtime status according to the return value of the
3481 *    device's runtime_resume function. If it is successfully resumed, process
3482 *    the requests that are queued into the device's queue when it is resuming
3483 *    and then mark last busy and initiate autosuspend for it.
3484 *
3485 *    This function should be called near the end of the device's
3486 *    runtime_resume callback.
3487 */
3488void blk_post_runtime_resume(struct request_queue *q, int err)
3489{
3490        if (!q->dev)
3491                return;
3492
3493        spin_lock_irq(q->queue_lock);
3494        if (!err) {
3495                q->rpm_status = RPM_ACTIVE;
3496                __blk_run_queue(q);
3497                pm_runtime_mark_last_busy(q->dev);
3498                pm_request_autosuspend(q->dev);
3499        } else {
3500                q->rpm_status = RPM_SUSPENDED;
3501        }
3502        spin_unlock_irq(q->queue_lock);
3503}
3504EXPORT_SYMBOL(blk_post_runtime_resume);
3505
3506/**
3507 * blk_set_runtime_active - Force runtime status of the queue to be active
3508 * @q: the queue of the device
3509 *
3510 * If the device is left runtime suspended during system suspend the resume
3511 * hook typically resumes the device and corrects runtime status
3512 * accordingly. However, that does not affect the queue runtime PM status
3513 * which is still "suspended". This prevents processing requests from the
3514 * queue.
3515 *
3516 * This function can be used in driver's resume hook to correct queue
3517 * runtime PM status and re-enable peeking requests from the queue. It
3518 * should be called before first request is added to the queue.
3519 */
3520void blk_set_runtime_active(struct request_queue *q)
3521{
3522        spin_lock_irq(q->queue_lock);
3523        q->rpm_status = RPM_ACTIVE;
3524        pm_runtime_mark_last_busy(q->dev);
3525        pm_request_autosuspend(q->dev);
3526        spin_unlock_irq(q->queue_lock);
3527}
3528EXPORT_SYMBOL(blk_set_runtime_active);
3529#endif
3530
3531int __init blk_dev_init(void)
3532{
3533        BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3534                        FIELD_SIZEOF(struct request, cmd_flags));
3535
3536        /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3537        kblockd_workqueue = alloc_workqueue("kblockd",
3538                                            WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3539        if (!kblockd_workqueue)
3540                panic("Failed to create kblockd\n");
3541
3542        request_cachep = kmem_cache_create("blkdev_requests",
3543                        sizeof(struct request), 0, SLAB_PANIC, NULL);
3544
3545        blk_requestq_cachep = kmem_cache_create("request_queue",
3546                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3547
3548        return 0;
3549}
3550