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#include <linux/debugfs.h>
  37#include <linux/bpf.h>
  38
  39#define CREATE_TRACE_POINTS
  40#include <trace/events/block.h>
  41
  42#include "blk.h"
  43#include "blk-mq.h"
  44#include "blk-mq-sched.h"
  45#include "blk-pm.h"
  46#include "blk-rq-qos.h"
  47
  48#ifdef CONFIG_DEBUG_FS
  49struct dentry *blk_debugfs_root;
  50#endif
  51
  52EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
  53EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
  54EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
  55EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
  56EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
  57
  58DEFINE_IDA(blk_queue_ida);
  59
  60/*
  61 * For the allocated request tables
  62 */
  63struct kmem_cache *request_cachep;
  64
  65/*
  66 * For queue allocation
  67 */
  68struct kmem_cache *blk_requestq_cachep;
  69
  70/*
  71 * Controlling structure to kblockd
  72 */
  73static struct workqueue_struct *kblockd_workqueue;
  74
  75/**
  76 * blk_queue_flag_set - atomically set a queue flag
  77 * @flag: flag to be set
  78 * @q: request queue
  79 */
  80void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
  81{
  82        unsigned long flags;
  83
  84        spin_lock_irqsave(q->queue_lock, flags);
  85        queue_flag_set(flag, q);
  86        spin_unlock_irqrestore(q->queue_lock, flags);
  87}
  88EXPORT_SYMBOL(blk_queue_flag_set);
  89
  90/**
  91 * blk_queue_flag_clear - atomically clear a queue flag
  92 * @flag: flag to be cleared
  93 * @q: request queue
  94 */
  95void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
  96{
  97        unsigned long flags;
  98
  99        spin_lock_irqsave(q->queue_lock, flags);
 100        queue_flag_clear(flag, q);
 101        spin_unlock_irqrestore(q->queue_lock, flags);
 102}
 103EXPORT_SYMBOL(blk_queue_flag_clear);
 104
 105/**
 106 * blk_queue_flag_test_and_set - atomically test and set a queue flag
 107 * @flag: flag to be set
 108 * @q: request queue
 109 *
 110 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
 111 * the flag was already set.
 112 */
 113bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
 114{
 115        unsigned long flags;
 116        bool res;
 117
 118        spin_lock_irqsave(q->queue_lock, flags);
 119        res = queue_flag_test_and_set(flag, q);
 120        spin_unlock_irqrestore(q->queue_lock, flags);
 121
 122        return res;
 123}
 124EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
 125
 126/**
 127 * blk_queue_flag_test_and_clear - atomically test and clear a queue flag
 128 * @flag: flag to be cleared
 129 * @q: request queue
 130 *
 131 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
 132 * the flag was set.
 133 */
 134bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q)
 135{
 136        unsigned long flags;
 137        bool res;
 138
 139        spin_lock_irqsave(q->queue_lock, flags);
 140        res = queue_flag_test_and_clear(flag, q);
 141        spin_unlock_irqrestore(q->queue_lock, flags);
 142
 143        return res;
 144}
 145EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_clear);
 146
 147static void blk_clear_congested(struct request_list *rl, int sync)
 148{
 149#ifdef CONFIG_CGROUP_WRITEBACK
 150        clear_wb_congested(rl->blkg->wb_congested, sync);
 151#else
 152        /*
 153         * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
 154         * flip its congestion state for events on other blkcgs.
 155         */
 156        if (rl == &rl->q->root_rl)
 157                clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
 158#endif
 159}
 160
 161static void blk_set_congested(struct request_list *rl, int sync)
 162{
 163#ifdef CONFIG_CGROUP_WRITEBACK
 164        set_wb_congested(rl->blkg->wb_congested, sync);
 165#else
 166        /* see blk_clear_congested() */
 167        if (rl == &rl->q->root_rl)
 168                set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
 169#endif
 170}
 171
 172void blk_queue_congestion_threshold(struct request_queue *q)
 173{
 174        int nr;
 175
 176        nr = q->nr_requests - (q->nr_requests / 8) + 1;
 177        if (nr > q->nr_requests)
 178                nr = q->nr_requests;
 179        q->nr_congestion_on = nr;
 180
 181        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
 182        if (nr < 1)
 183                nr = 1;
 184        q->nr_congestion_off = nr;
 185}
 186
 187void blk_rq_init(struct request_queue *q, struct request *rq)
 188{
 189        memset(rq, 0, sizeof(*rq));
 190
 191        INIT_LIST_HEAD(&rq->queuelist);
 192        INIT_LIST_HEAD(&rq->timeout_list);
 193        rq->cpu = -1;
 194        rq->q = q;
 195        rq->__sector = (sector_t) -1;
 196        INIT_HLIST_NODE(&rq->hash);
 197        RB_CLEAR_NODE(&rq->rb_node);
 198        rq->tag = -1;
 199        rq->internal_tag = -1;
 200        rq->start_time_ns = ktime_get_ns();
 201        rq->part = NULL;
 202}
 203EXPORT_SYMBOL(blk_rq_init);
 204
 205static const struct {
 206        int             errno;
 207        const char      *name;
 208} blk_errors[] = {
 209        [BLK_STS_OK]            = { 0,          "" },
 210        [BLK_STS_NOTSUPP]       = { -EOPNOTSUPP, "operation not supported" },
 211        [BLK_STS_TIMEOUT]       = { -ETIMEDOUT, "timeout" },
 212        [BLK_STS_NOSPC]         = { -ENOSPC,    "critical space allocation" },
 213        [BLK_STS_TRANSPORT]     = { -ENOLINK,   "recoverable transport" },
 214        [BLK_STS_TARGET]        = { -EREMOTEIO, "critical target" },
 215        [BLK_STS_NEXUS]         = { -EBADE,     "critical nexus" },
 216        [BLK_STS_MEDIUM]        = { -ENODATA,   "critical medium" },
 217        [BLK_STS_PROTECTION]    = { -EILSEQ,    "protection" },
 218        [BLK_STS_RESOURCE]      = { -ENOMEM,    "kernel resource" },
 219        [BLK_STS_DEV_RESOURCE]  = { -EBUSY,     "device resource" },
 220        [BLK_STS_AGAIN]         = { -EAGAIN,    "nonblocking retry" },
 221
 222        /* device mapper special case, should not leak out: */
 223        [BLK_STS_DM_REQUEUE]    = { -EREMCHG, "dm internal retry" },
 224
 225        /* everything else not covered above: */
 226        [BLK_STS_IOERR]         = { -EIO,       "I/O" },
 227};
 228
 229blk_status_t errno_to_blk_status(int errno)
 230{
 231        int i;
 232
 233        for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
 234                if (blk_errors[i].errno == errno)
 235                        return (__force blk_status_t)i;
 236        }
 237
 238        return BLK_STS_IOERR;
 239}
 240EXPORT_SYMBOL_GPL(errno_to_blk_status);
 241
 242int blk_status_to_errno(blk_status_t status)
 243{
 244        int idx = (__force int)status;
 245
 246        if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
 247                return -EIO;
 248        return blk_errors[idx].errno;
 249}
 250EXPORT_SYMBOL_GPL(blk_status_to_errno);
 251
 252static void print_req_error(struct request *req, blk_status_t status)
 253{
 254        int idx = (__force int)status;
 255
 256        if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
 257                return;
 258
 259        printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
 260                           __func__, blk_errors[idx].name, req->rq_disk ?
 261                           req->rq_disk->disk_name : "?",
 262                           (unsigned long long)blk_rq_pos(req));
 263}
 264
 265static void req_bio_endio(struct request *rq, struct bio *bio,
 266                          unsigned int nbytes, blk_status_t error)
 267{
 268        if (error)
 269                bio->bi_status = error;
 270
 271        if (unlikely(rq->rq_flags & RQF_QUIET))
 272                bio_set_flag(bio, BIO_QUIET);
 273
 274        bio_advance(bio, nbytes);
 275
 276        /* don't actually finish bio if it's part of flush sequence */
 277        if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
 278                bio_endio(bio);
 279}
 280
 281void blk_dump_rq_flags(struct request *rq, char *msg)
 282{
 283        printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
 284                rq->rq_disk ? rq->rq_disk->disk_name : "?",
 285                (unsigned long long) rq->cmd_flags);
 286
 287        printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
 288               (unsigned long long)blk_rq_pos(rq),
 289               blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
 290        printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
 291               rq->bio, rq->biotail, blk_rq_bytes(rq));
 292}
 293EXPORT_SYMBOL(blk_dump_rq_flags);
 294
 295static void blk_delay_work(struct work_struct *work)
 296{
 297        struct request_queue *q;
 298
 299        q = container_of(work, struct request_queue, delay_work.work);
 300        spin_lock_irq(q->queue_lock);
 301        __blk_run_queue(q);
 302        spin_unlock_irq(q->queue_lock);
 303}
 304
 305/**
 306 * blk_delay_queue - restart queueing after defined interval
 307 * @q:          The &struct request_queue in question
 308 * @msecs:      Delay in msecs
 309 *
 310 * Description:
 311 *   Sometimes queueing needs to be postponed for a little while, to allow
 312 *   resources to come back. This function will make sure that queueing is
 313 *   restarted around the specified time.
 314 */
 315void blk_delay_queue(struct request_queue *q, unsigned long msecs)
 316{
 317        lockdep_assert_held(q->queue_lock);
 318        WARN_ON_ONCE(q->mq_ops);
 319
 320        if (likely(!blk_queue_dead(q)))
 321                queue_delayed_work(kblockd_workqueue, &q->delay_work,
 322                                   msecs_to_jiffies(msecs));
 323}
 324EXPORT_SYMBOL(blk_delay_queue);
 325
 326/**
 327 * blk_start_queue_async - asynchronously restart a previously stopped queue
 328 * @q:    The &struct request_queue in question
 329 *
 330 * Description:
 331 *   blk_start_queue_async() will clear the stop flag on the queue, and
 332 *   ensure that the request_fn for the queue is run from an async
 333 *   context.
 334 **/
 335void blk_start_queue_async(struct request_queue *q)
 336{
 337        lockdep_assert_held(q->queue_lock);
 338        WARN_ON_ONCE(q->mq_ops);
 339
 340        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 341        blk_run_queue_async(q);
 342}
 343EXPORT_SYMBOL(blk_start_queue_async);
 344
 345/**
 346 * blk_start_queue - restart a previously stopped queue
 347 * @q:    The &struct request_queue in question
 348 *
 349 * Description:
 350 *   blk_start_queue() will clear the stop flag on the queue, and call
 351 *   the request_fn for the queue if it was in a stopped state when
 352 *   entered. Also see blk_stop_queue().
 353 **/
 354void blk_start_queue(struct request_queue *q)
 355{
 356        lockdep_assert_held(q->queue_lock);
 357        WARN_ON_ONCE(q->mq_ops);
 358
 359        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 360        __blk_run_queue(q);
 361}
 362EXPORT_SYMBOL(blk_start_queue);
 363
 364/**
 365 * blk_stop_queue - stop a queue
 366 * @q:    The &struct request_queue in question
 367 *
 368 * Description:
 369 *   The Linux block layer assumes that a block driver will consume all
 370 *   entries on the request queue when the request_fn strategy is called.
 371 *   Often this will not happen, because of hardware limitations (queue
 372 *   depth settings). If a device driver gets a 'queue full' response,
 373 *   or if it simply chooses not to queue more I/O at one point, it can
 374 *   call this function to prevent the request_fn from being called until
 375 *   the driver has signalled it's ready to go again. This happens by calling
 376 *   blk_start_queue() to restart queue operations.
 377 **/
 378void blk_stop_queue(struct request_queue *q)
 379{
 380        lockdep_assert_held(q->queue_lock);
 381        WARN_ON_ONCE(q->mq_ops);
 382
 383        cancel_delayed_work(&q->delay_work);
 384        queue_flag_set(QUEUE_FLAG_STOPPED, q);
 385}
 386EXPORT_SYMBOL(blk_stop_queue);
 387
 388/**
 389 * blk_sync_queue - cancel any pending callbacks on a queue
 390 * @q: the queue
 391 *
 392 * Description:
 393 *     The block layer may perform asynchronous callback activity
 394 *     on a queue, such as calling the unplug function after a timeout.
 395 *     A block device may call blk_sync_queue to ensure that any
 396 *     such activity is cancelled, thus allowing it to release resources
 397 *     that the callbacks might use. The caller must already have made sure
 398 *     that its ->make_request_fn will not re-add plugging prior to calling
 399 *     this function.
 400 *
 401 *     This function does not cancel any asynchronous activity arising
 402 *     out of elevator or throttling code. That would require elevator_exit()
 403 *     and blkcg_exit_queue() to be called with queue lock initialized.
 404 *
 405 */
 406void blk_sync_queue(struct request_queue *q)
 407{
 408        del_timer_sync(&q->timeout);
 409        cancel_work_sync(&q->timeout_work);
 410
 411        if (q->mq_ops) {
 412                struct blk_mq_hw_ctx *hctx;
 413                int i;
 414
 415                cancel_delayed_work_sync(&q->requeue_work);
 416                queue_for_each_hw_ctx(q, hctx, i)
 417                        cancel_delayed_work_sync(&hctx->run_work);
 418        } else {
 419                cancel_delayed_work_sync(&q->delay_work);
 420        }
 421}
 422EXPORT_SYMBOL(blk_sync_queue);
 423
 424/**
 425 * blk_set_pm_only - increment pm_only counter
 426 * @q: request queue pointer
 427 */
 428void blk_set_pm_only(struct request_queue *q)
 429{
 430        atomic_inc(&q->pm_only);
 431}
 432EXPORT_SYMBOL_GPL(blk_set_pm_only);
 433
 434void blk_clear_pm_only(struct request_queue *q)
 435{
 436        int pm_only;
 437
 438        pm_only = atomic_dec_return(&q->pm_only);
 439        WARN_ON_ONCE(pm_only < 0);
 440        if (pm_only == 0)
 441                wake_up_all(&q->mq_freeze_wq);
 442}
 443EXPORT_SYMBOL_GPL(blk_clear_pm_only);
 444
 445/**
 446 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
 447 * @q:  The queue to run
 448 *
 449 * Description:
 450 *    Invoke request handling on a queue if there are any pending requests.
 451 *    May be used to restart request handling after a request has completed.
 452 *    This variant runs the queue whether or not the queue has been
 453 *    stopped. Must be called with the queue lock held and interrupts
 454 *    disabled. See also @blk_run_queue.
 455 */
 456inline void __blk_run_queue_uncond(struct request_queue *q)
 457{
 458        lockdep_assert_held(q->queue_lock);
 459        WARN_ON_ONCE(q->mq_ops);
 460
 461        if (unlikely(blk_queue_dead(q)))
 462                return;
 463
 464        /*
 465         * Some request_fn implementations, e.g. scsi_request_fn(), unlock
 466         * the queue lock internally. As a result multiple threads may be
 467         * running such a request function concurrently. Keep track of the
 468         * number of active request_fn invocations such that blk_drain_queue()
 469         * can wait until all these request_fn calls have finished.
 470         */
 471        q->request_fn_active++;
 472        q->request_fn(q);
 473        q->request_fn_active--;
 474}
 475EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
 476
 477/**
 478 * __blk_run_queue - run a single device queue
 479 * @q:  The queue to run
 480 *
 481 * Description:
 482 *    See @blk_run_queue.
 483 */
 484void __blk_run_queue(struct request_queue *q)
 485{
 486        lockdep_assert_held(q->queue_lock);
 487        WARN_ON_ONCE(q->mq_ops);
 488
 489        if (unlikely(blk_queue_stopped(q)))
 490                return;
 491
 492        __blk_run_queue_uncond(q);
 493}
 494EXPORT_SYMBOL(__blk_run_queue);
 495
 496/**
 497 * blk_run_queue_async - run a single device queue in workqueue context
 498 * @q:  The queue to run
 499 *
 500 * Description:
 501 *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
 502 *    of us.
 503 *
 504 * Note:
 505 *    Since it is not allowed to run q->delay_work after blk_cleanup_queue()
 506 *    has canceled q->delay_work, callers must hold the queue lock to avoid
 507 *    race conditions between blk_cleanup_queue() and blk_run_queue_async().
 508 */
 509void blk_run_queue_async(struct request_queue *q)
 510{
 511        lockdep_assert_held(q->queue_lock);
 512        WARN_ON_ONCE(q->mq_ops);
 513
 514        if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
 515                mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
 516}
 517EXPORT_SYMBOL(blk_run_queue_async);
 518
 519/**
 520 * blk_run_queue - run a single device queue
 521 * @q: The queue to run
 522 *
 523 * Description:
 524 *    Invoke request handling on this queue, if it has pending work to do.
 525 *    May be used to restart queueing when a request has completed.
 526 */
 527void blk_run_queue(struct request_queue *q)
 528{
 529        unsigned long flags;
 530
 531        WARN_ON_ONCE(q->mq_ops);
 532
 533        spin_lock_irqsave(q->queue_lock, flags);
 534        __blk_run_queue(q);
 535        spin_unlock_irqrestore(q->queue_lock, flags);
 536}
 537EXPORT_SYMBOL(blk_run_queue);
 538
 539void blk_put_queue(struct request_queue *q)
 540{
 541        kobject_put(&q->kobj);
 542}
 543EXPORT_SYMBOL(blk_put_queue);
 544
 545/**
 546 * __blk_drain_queue - drain requests from request_queue
 547 * @q: queue to drain
 548 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
 549 *
 550 * Drain requests from @q.  If @drain_all is set, all requests are drained.
 551 * If not, only ELVPRIV requests are drained.  The caller is responsible
 552 * for ensuring that no new requests which need to be drained are queued.
 553 */
 554static void __blk_drain_queue(struct request_queue *q, bool drain_all)
 555        __releases(q->queue_lock)
 556        __acquires(q->queue_lock)
 557{
 558        int i;
 559
 560        lockdep_assert_held(q->queue_lock);
 561        WARN_ON_ONCE(q->mq_ops);
 562
 563        while (true) {
 564                bool drain = false;
 565
 566                /*
 567                 * The caller might be trying to drain @q before its
 568                 * elevator is initialized.
 569                 */
 570                if (q->elevator)
 571                        elv_drain_elevator(q);
 572
 573                blkcg_drain_queue(q);
 574
 575                /*
 576                 * This function might be called on a queue which failed
 577                 * driver init after queue creation or is not yet fully
 578                 * active yet.  Some drivers (e.g. fd and loop) get unhappy
 579                 * in such cases.  Kick queue iff dispatch queue has
 580                 * something on it and @q has request_fn set.
 581                 */
 582                if (!list_empty(&q->queue_head) && q->request_fn)
 583                        __blk_run_queue(q);
 584
 585                drain |= q->nr_rqs_elvpriv;
 586                drain |= q->request_fn_active;
 587
 588                /*
 589                 * Unfortunately, requests are queued at and tracked from
 590                 * multiple places and there's no single counter which can
 591                 * be drained.  Check all the queues and counters.
 592                 */
 593                if (drain_all) {
 594                        struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
 595                        drain |= !list_empty(&q->queue_head);
 596                        for (i = 0; i < 2; i++) {
 597                                drain |= q->nr_rqs[i];
 598                                drain |= q->in_flight[i];
 599                                if (fq)
 600                                    drain |= !list_empty(&fq->flush_queue[i]);
 601                        }
 602                }
 603
 604                if (!drain)
 605                        break;
 606
 607                spin_unlock_irq(q->queue_lock);
 608
 609                msleep(10);
 610
 611                spin_lock_irq(q->queue_lock);
 612        }
 613
 614        /*
 615         * With queue marked dead, any woken up waiter will fail the
 616         * allocation path, so the wakeup chaining is lost and we're
 617         * left with hung waiters. We need to wake up those waiters.
 618         */
 619        if (q->request_fn) {
 620                struct request_list *rl;
 621
 622                blk_queue_for_each_rl(rl, q)
 623                        for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
 624                                wake_up_all(&rl->wait[i]);
 625        }
 626}
 627
 628void blk_drain_queue(struct request_queue *q)
 629{
 630        spin_lock_irq(q->queue_lock);
 631        __blk_drain_queue(q, true);
 632        spin_unlock_irq(q->queue_lock);
 633}
 634
 635/**
 636 * blk_queue_bypass_start - enter queue bypass mode
 637 * @q: queue of interest
 638 *
 639 * In bypass mode, only the dispatch FIFO queue of @q is used.  This
 640 * function makes @q enter bypass mode and drains all requests which were
 641 * throttled or issued before.  On return, it's guaranteed that no request
 642 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
 643 * inside queue or RCU read lock.
 644 */
 645void blk_queue_bypass_start(struct request_queue *q)
 646{
 647        WARN_ON_ONCE(q->mq_ops);
 648
 649        spin_lock_irq(q->queue_lock);
 650        q->bypass_depth++;
 651        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 652        spin_unlock_irq(q->queue_lock);
 653
 654        /*
 655         * Queues start drained.  Skip actual draining till init is
 656         * complete.  This avoids lenghty delays during queue init which
 657         * can happen many times during boot.
 658         */
 659        if (blk_queue_init_done(q)) {
 660                spin_lock_irq(q->queue_lock);
 661                __blk_drain_queue(q, false);
 662                spin_unlock_irq(q->queue_lock);
 663
 664                /* ensure blk_queue_bypass() is %true inside RCU read lock */
 665                synchronize_rcu();
 666        }
 667}
 668EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
 669
 670/**
 671 * blk_queue_bypass_end - leave queue bypass mode
 672 * @q: queue of interest
 673 *
 674 * Leave bypass mode and restore the normal queueing behavior.
 675 *
 676 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
 677 * this function is called for both blk-sq and blk-mq queues.
 678 */
 679void blk_queue_bypass_end(struct request_queue *q)
 680{
 681        spin_lock_irq(q->queue_lock);
 682        if (!--q->bypass_depth)
 683                queue_flag_clear(QUEUE_FLAG_BYPASS, q);
 684        WARN_ON_ONCE(q->bypass_depth < 0);
 685        spin_unlock_irq(q->queue_lock);
 686}
 687EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
 688
 689void blk_set_queue_dying(struct request_queue *q)
 690{
 691        blk_queue_flag_set(QUEUE_FLAG_DYING, q);
 692
 693        /*
 694         * When queue DYING flag is set, we need to block new req
 695         * entering queue, so we call blk_freeze_queue_start() to
 696         * prevent I/O from crossing blk_queue_enter().
 697         */
 698        blk_freeze_queue_start(q);
 699
 700        if (q->mq_ops)
 701                blk_mq_wake_waiters(q);
 702        else {
 703                struct request_list *rl;
 704
 705                spin_lock_irq(q->queue_lock);
 706                blk_queue_for_each_rl(rl, q) {
 707                        if (rl->rq_pool) {
 708                                wake_up_all(&rl->wait[BLK_RW_SYNC]);
 709                                wake_up_all(&rl->wait[BLK_RW_ASYNC]);
 710                        }
 711                }
 712                spin_unlock_irq(q->queue_lock);
 713        }
 714
 715        /* Make blk_queue_enter() reexamine the DYING flag. */
 716        wake_up_all(&q->mq_freeze_wq);
 717}
 718EXPORT_SYMBOL_GPL(blk_set_queue_dying);
 719
 720/* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
 721void blk_exit_queue(struct request_queue *q)
 722{
 723        /*
 724         * Since the I/O scheduler exit code may access cgroup information,
 725         * perform I/O scheduler exit before disassociating from the block
 726         * cgroup controller.
 727         */
 728        if (q->elevator) {
 729                ioc_clear_queue(q);
 730                elevator_exit(q, q->elevator);
 731                q->elevator = NULL;
 732        }
 733
 734        /*
 735         * Remove all references to @q from the block cgroup controller before
 736         * restoring @q->queue_lock to avoid that restoring this pointer causes
 737         * e.g. blkcg_print_blkgs() to crash.
 738         */
 739        blkcg_exit_queue(q);
 740
 741        /*
 742         * Since the cgroup code may dereference the @q->backing_dev_info
 743         * pointer, only decrease its reference count after having removed the
 744         * association with the block cgroup controller.
 745         */
 746        bdi_put(q->backing_dev_info);
 747}
 748
 749/**
 750 * blk_cleanup_queue - shutdown a request queue
 751 * @q: request queue to shutdown
 752 *
 753 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
 754 * put it.  All future requests will be failed immediately with -ENODEV.
 755 */
 756void blk_cleanup_queue(struct request_queue *q)
 757{
 758        spinlock_t *lock = q->queue_lock;
 759
 760        /* mark @q DYING, no new request or merges will be allowed afterwards */
 761        mutex_lock(&q->sysfs_lock);
 762        blk_set_queue_dying(q);
 763        spin_lock_irq(lock);
 764
 765        /*
 766         * A dying queue is permanently in bypass mode till released.  Note
 767         * that, unlike blk_queue_bypass_start(), we aren't performing
 768         * synchronize_rcu() after entering bypass mode to avoid the delay
 769         * as some drivers create and destroy a lot of queues while
 770         * probing.  This is still safe because blk_release_queue() will be
 771         * called only after the queue refcnt drops to zero and nothing,
 772         * RCU or not, would be traversing the queue by then.
 773         */
 774        q->bypass_depth++;
 775        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 776
 777        queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 778        queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 779        queue_flag_set(QUEUE_FLAG_DYING, q);
 780        spin_unlock_irq(lock);
 781        mutex_unlock(&q->sysfs_lock);
 782
 783        /*
 784         * Drain all requests queued before DYING marking. Set DEAD flag to
 785         * prevent that q->request_fn() gets invoked after draining finished.
 786         */
 787        blk_freeze_queue(q);
 788
 789        rq_qos_exit(q);
 790
 791        spin_lock_irq(lock);
 792        queue_flag_set(QUEUE_FLAG_DEAD, q);
 793        spin_unlock_irq(lock);
 794
 795        /*
 796         * make sure all in-progress dispatch are completed because
 797         * blk_freeze_queue() can only complete all requests, and
 798         * dispatch may still be in-progress since we dispatch requests
 799         * from more than one contexts.
 800         *
 801         * We rely on driver to deal with the race in case that queue
 802         * initialization isn't done.
 803         */
 804        if (q->mq_ops && blk_queue_init_done(q))
 805                blk_mq_quiesce_queue(q);
 806
 807        /* for synchronous bio-based driver finish in-flight integrity i/o */
 808        blk_flush_integrity();
 809
 810        /* @q won't process any more request, flush async actions */
 811        del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
 812        blk_sync_queue(q);
 813
 814        /*
 815         * I/O scheduler exit is only safe after the sysfs scheduler attribute
 816         * has been removed.
 817         */
 818        WARN_ON_ONCE(q->kobj.state_in_sysfs);
 819
 820        blk_exit_queue(q);
 821
 822        if (q->mq_ops)
 823                blk_mq_free_queue(q);
 824        percpu_ref_exit(&q->q_usage_counter);
 825
 826        spin_lock_irq(lock);
 827        if (q->queue_lock != &q->__queue_lock)
 828                q->queue_lock = &q->__queue_lock;
 829        spin_unlock_irq(lock);
 830
 831        /* @q is and will stay empty, shutdown and put */
 832        blk_put_queue(q);
 833}
 834EXPORT_SYMBOL(blk_cleanup_queue);
 835
 836/* Allocate memory local to the request queue */
 837static void *alloc_request_simple(gfp_t gfp_mask, void *data)
 838{
 839        struct request_queue *q = data;
 840
 841        return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
 842}
 843
 844static void free_request_simple(void *element, void *data)
 845{
 846        kmem_cache_free(request_cachep, element);
 847}
 848
 849static void *alloc_request_size(gfp_t gfp_mask, void *data)
 850{
 851        struct request_queue *q = data;
 852        struct request *rq;
 853
 854        rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
 855                        q->node);
 856        if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
 857                kfree(rq);
 858                rq = NULL;
 859        }
 860        return rq;
 861}
 862
 863static void free_request_size(void *element, void *data)
 864{
 865        struct request_queue *q = data;
 866
 867        if (q->exit_rq_fn)
 868                q->exit_rq_fn(q, element);
 869        kfree(element);
 870}
 871
 872int blk_init_rl(struct request_list *rl, struct request_queue *q,
 873                gfp_t gfp_mask)
 874{
 875        if (unlikely(rl->rq_pool) || q->mq_ops)
 876                return 0;
 877
 878        rl->q = q;
 879        rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
 880        rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
 881        init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
 882        init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
 883
 884        if (q->cmd_size) {
 885                rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
 886                                alloc_request_size, free_request_size,
 887                                q, gfp_mask, q->node);
 888        } else {
 889                rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
 890                                alloc_request_simple, free_request_simple,
 891                                q, gfp_mask, q->node);
 892        }
 893        if (!rl->rq_pool)
 894                return -ENOMEM;
 895
 896        if (rl != &q->root_rl)
 897                WARN_ON_ONCE(!blk_get_queue(q));
 898
 899        return 0;
 900}
 901
 902void blk_exit_rl(struct request_queue *q, struct request_list *rl)
 903{
 904        if (rl->rq_pool) {
 905                mempool_destroy(rl->rq_pool);
 906                if (rl != &q->root_rl)
 907                        blk_put_queue(q);
 908        }
 909}
 910
 911struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 912{
 913        return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE, NULL);
 914}
 915EXPORT_SYMBOL(blk_alloc_queue);
 916
 917/**
 918 * blk_queue_enter() - try to increase q->q_usage_counter
 919 * @q: request queue pointer
 920 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
 921 */
 922int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 923{
 924        const bool pm = flags & BLK_MQ_REQ_PREEMPT;
 925
 926        while (true) {
 927                bool success = false;
 928
 929                rcu_read_lock();
 930                if (percpu_ref_tryget_live(&q->q_usage_counter)) {
 931                        /*
 932                         * The code that increments the pm_only counter is
 933                         * responsible for ensuring that that counter is
 934                         * globally visible before the queue is unfrozen.
 935                         */
 936                        if (pm || !blk_queue_pm_only(q)) {
 937                                success = true;
 938                        } else {
 939                                percpu_ref_put(&q->q_usage_counter);
 940                        }
 941                }
 942                rcu_read_unlock();
 943
 944                if (success)
 945                        return 0;
 946
 947                if (flags & BLK_MQ_REQ_NOWAIT)
 948                        return -EBUSY;
 949
 950                /*
 951                 * read pair of barrier in blk_freeze_queue_start(),
 952                 * we need to order reading __PERCPU_REF_DEAD flag of
 953                 * .q_usage_counter and reading .mq_freeze_depth or
 954                 * queue dying flag, otherwise the following wait may
 955                 * never return if the two reads are reordered.
 956                 */
 957                smp_rmb();
 958
 959                wait_event(q->mq_freeze_wq,
 960                           (atomic_read(&q->mq_freeze_depth) == 0 &&
 961                            (pm || (blk_pm_request_resume(q),
 962                                    !blk_queue_pm_only(q)))) ||
 963                           blk_queue_dying(q));
 964                if (blk_queue_dying(q))
 965                        return -ENODEV;
 966        }
 967}
 968
 969void blk_queue_exit(struct request_queue *q)
 970{
 971        percpu_ref_put(&q->q_usage_counter);
 972}
 973
 974static void blk_queue_usage_counter_release(struct percpu_ref *ref)
 975{
 976        struct request_queue *q =
 977                container_of(ref, struct request_queue, q_usage_counter);
 978
 979        wake_up_all(&q->mq_freeze_wq);
 980}
 981
 982static void blk_rq_timed_out_timer(struct timer_list *t)
 983{
 984        struct request_queue *q = from_timer(q, t, timeout);
 985
 986        kblockd_schedule_work(&q->timeout_work);
 987}
 988
 989/**
 990 * blk_alloc_queue_node - allocate a request queue
 991 * @gfp_mask: memory allocation flags
 992 * @node_id: NUMA node to allocate memory from
 993 * @lock: For legacy queues, pointer to a spinlock that will be used to e.g.
 994 *        serialize calls to the legacy .request_fn() callback. Ignored for
 995 *        blk-mq request queues.
 996 *
 997 * Note: pass the queue lock as the third argument to this function instead of
 998 * setting the queue lock pointer explicitly to avoid triggering a sporadic
 999 * crash in the blkcg code. This function namely calls blkcg_init_queue() and
1000 * the queue lock pointer must be set before blkcg_init_queue() is called.
1001 */
1002struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
1003                                           spinlock_t *lock)
1004{
1005        struct request_queue *q;
1006        int ret;
1007
1008        q = kmem_cache_alloc_node(blk_requestq_cachep,
1009                                gfp_mask | __GFP_ZERO, node_id);
1010        if (!q)
1011                return NULL;
1012
1013        INIT_LIST_HEAD(&q->queue_head);
1014        q->last_merge = NULL;
1015        q->end_sector = 0;
1016        q->boundary_rq = NULL;
1017
1018        q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
1019        if (q->id < 0)
1020                goto fail_q;
1021
1022        ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1023        if (ret)
1024                goto fail_id;
1025
1026        q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1027        if (!q->backing_dev_info)
1028                goto fail_split;
1029
1030        q->stats = blk_alloc_queue_stats();
1031        if (!q->stats)
1032                goto fail_stats;
1033
1034        q->backing_dev_info->ra_pages =
1035                        (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1036        q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1037        q->backing_dev_info->name = "block";
1038        q->node = node_id;
1039
1040        timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1041                    laptop_mode_timer_fn, 0);
1042        timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
1043        INIT_WORK(&q->timeout_work, NULL);
1044        INIT_LIST_HEAD(&q->timeout_list);
1045        INIT_LIST_HEAD(&q->icq_list);
1046#ifdef CONFIG_BLK_CGROUP
1047        INIT_LIST_HEAD(&q->blkg_list);
1048#endif
1049        INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
1050
1051        kobject_init(&q->kobj, &blk_queue_ktype);
1052
1053#ifdef CONFIG_BLK_DEV_IO_TRACE
1054        mutex_init(&q->blk_trace_mutex);
1055#endif
1056        mutex_init(&q->sysfs_lock);
1057        spin_lock_init(&q->__queue_lock);
1058
1059        q->queue_lock = lock ? : &q->__queue_lock;
1060
1061        /*
1062         * A queue starts its life with bypass turned on to avoid
1063         * unnecessary bypass on/off overhead and nasty surprises during
1064         * init.  The initial bypass will be finished when the queue is
1065         * registered by blk_register_queue().
1066         */
1067        q->bypass_depth = 1;
1068        queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
1069
1070        init_waitqueue_head(&q->mq_freeze_wq);
1071
1072        /*
1073         * Init percpu_ref in atomic mode so that it's faster to shutdown.
1074         * See blk_register_queue() for details.
1075         */
1076        if (percpu_ref_init(&q->q_usage_counter,
1077                                blk_queue_usage_counter_release,
1078                                PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
1079                goto fail_bdi;
1080
1081        if (blkcg_init_queue(q))
1082                goto fail_ref;
1083
1084        return q;
1085
1086fail_ref:
1087        percpu_ref_exit(&q->q_usage_counter);
1088fail_bdi:
1089        blk_free_queue_stats(q->stats);
1090fail_stats:
1091        bdi_put(q->backing_dev_info);
1092fail_split:
1093        bioset_exit(&q->bio_split);
1094fail_id:
1095        ida_simple_remove(&blk_queue_ida, q->id);
1096fail_q:
1097        kmem_cache_free(blk_requestq_cachep, q);
1098        return NULL;
1099}
1100EXPORT_SYMBOL(blk_alloc_queue_node);
1101
1102/**
1103 * blk_init_queue  - prepare a request queue for use with a block device
1104 * @rfn:  The function to be called to process requests that have been
1105 *        placed on the queue.
1106 * @lock: Request queue spin lock
1107 *
1108 * Description:
1109 *    If a block device wishes to use the standard request handling procedures,
1110 *    which sorts requests and coalesces adjacent requests, then it must
1111 *    call blk_init_queue().  The function @rfn will be called when there
1112 *    are requests on the queue that need to be processed.  If the device
1113 *    supports plugging, then @rfn may not be called immediately when requests
1114 *    are available on the queue, but may be called at some time later instead.
1115 *    Plugged queues are generally unplugged when a buffer belonging to one
1116 *    of the requests on the queue is needed, or due to memory pressure.
1117 *
1118 *    @rfn is not required, or even expected, to remove all requests off the
1119 *    queue, but only as many as it can handle at a time.  If it does leave
1120 *    requests on the queue, it is responsible for arranging that the requests
1121 *    get dealt with eventually.
1122 *
1123 *    The queue spin lock must be held while manipulating the requests on the
1124 *    request queue; this lock will be taken also from interrupt context, so irq
1125 *    disabling is needed for it.
1126 *
1127 *    Function returns a pointer to the initialized request queue, or %NULL if
1128 *    it didn't succeed.
1129 *
1130 * Note:
1131 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
1132 *    when the block device is deactivated (such as at module unload).
1133 **/
1134
1135struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1136{
1137        return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1138}
1139EXPORT_SYMBOL(blk_init_queue);
1140
1141struct request_queue *
1142blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1143{
1144        struct request_queue *q;
1145
1146        q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
1147        if (!q)
1148                return NULL;
1149
1150        q->request_fn = rfn;
1151        if (blk_init_allocated_queue(q) < 0) {
1152                blk_cleanup_queue(q);
1153                return NULL;
1154        }
1155
1156        return q;
1157}
1158EXPORT_SYMBOL(blk_init_queue_node);
1159
1160static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1161
1162
1163int blk_init_allocated_queue(struct request_queue *q)
1164{
1165        WARN_ON_ONCE(q->mq_ops);
1166
1167        q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size, GFP_KERNEL);
1168        if (!q->fq)
1169                return -ENOMEM;
1170
1171        if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1172                goto out_free_flush_queue;
1173
1174        if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1175                goto out_exit_flush_rq;
1176
1177        INIT_WORK(&q->timeout_work, blk_timeout_work);
1178        q->queue_flags          |= QUEUE_FLAG_DEFAULT;
1179
1180        /*
1181         * This also sets hw/phys segments, boundary and size
1182         */
1183        blk_queue_make_request(q, blk_queue_bio);
1184
1185        q->sg_reserved_size = INT_MAX;
1186
1187        if (elevator_init(q))
1188                goto out_exit_flush_rq;
1189        return 0;
1190
1191out_exit_flush_rq:
1192        if (q->exit_rq_fn)
1193                q->exit_rq_fn(q, q->fq->flush_rq);
1194out_free_flush_queue:
1195        blk_free_flush_queue(q->fq);
1196        q->fq = NULL;
1197        return -ENOMEM;
1198}
1199EXPORT_SYMBOL(blk_init_allocated_queue);
1200
1201bool blk_get_queue(struct request_queue *q)
1202{
1203        if (likely(!blk_queue_dying(q))) {
1204                __blk_get_queue(q);
1205                return true;
1206        }
1207
1208        return false;
1209}
1210EXPORT_SYMBOL(blk_get_queue);
1211
1212static inline void blk_free_request(struct request_list *rl, struct request *rq)
1213{
1214        if (rq->rq_flags & RQF_ELVPRIV) {
1215                elv_put_request(rl->q, rq);
1216                if (rq->elv.icq)
1217                        put_io_context(rq->elv.icq->ioc);
1218        }
1219
1220        mempool_free(rq, rl->rq_pool);
1221}
1222
1223/*
1224 * ioc_batching returns true if the ioc is a valid batching request and
1225 * should be given priority access to a request.
1226 */
1227static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1228{
1229        if (!ioc)
1230                return 0;
1231
1232        /*
1233         * Make sure the process is able to allocate at least 1 request
1234         * even if the batch times out, otherwise we could theoretically
1235         * lose wakeups.
1236         */
1237        return ioc->nr_batch_requests == q->nr_batching ||
1238                (ioc->nr_batch_requests > 0
1239                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1240}
1241
1242/*
1243 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1244 * will cause the process to be a "batcher" on all queues in the system. This
1245 * is the behaviour we want though - once it gets a wakeup it should be given
1246 * a nice run.
1247 */
1248static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1249{
1250        if (!ioc || ioc_batching(q, ioc))
1251                return;
1252
1253        ioc->nr_batch_requests = q->nr_batching;
1254        ioc->last_waited = jiffies;
1255}
1256
1257static void __freed_request(struct request_list *rl, int sync)
1258{
1259        struct request_queue *q = rl->q;
1260
1261        if (rl->count[sync] < queue_congestion_off_threshold(q))
1262                blk_clear_congested(rl, sync);
1263
1264        if (rl->count[sync] + 1 <= q->nr_requests) {
1265                if (waitqueue_active(&rl->wait[sync]))
1266                        wake_up(&rl->wait[sync]);
1267
1268                blk_clear_rl_full(rl, sync);
1269        }
1270}
1271
1272/*
1273 * A request has just been released.  Account for it, update the full and
1274 * congestion status, wake up any waiters.   Called under q->queue_lock.
1275 */
1276static void freed_request(struct request_list *rl, bool sync,
1277                req_flags_t rq_flags)
1278{
1279        struct request_queue *q = rl->q;
1280
1281        q->nr_rqs[sync]--;
1282        rl->count[sync]--;
1283        if (rq_flags & RQF_ELVPRIV)
1284                q->nr_rqs_elvpriv--;
1285
1286        __freed_request(rl, sync);
1287
1288        if (unlikely(rl->starved[sync ^ 1]))
1289                __freed_request(rl, sync ^ 1);
1290}
1291
1292int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1293{
1294        struct request_list *rl;
1295        int on_thresh, off_thresh;
1296
1297        WARN_ON_ONCE(q->mq_ops);
1298
1299        spin_lock_irq(q->queue_lock);
1300        q->nr_requests = nr;
1301        blk_queue_congestion_threshold(q);
1302        on_thresh = queue_congestion_on_threshold(q);
1303        off_thresh = queue_congestion_off_threshold(q);
1304
1305        blk_queue_for_each_rl(rl, q) {
1306                if (rl->count[BLK_RW_SYNC] >= on_thresh)
1307                        blk_set_congested(rl, BLK_RW_SYNC);
1308                else if (rl->count[BLK_RW_SYNC] < off_thresh)
1309                        blk_clear_congested(rl, BLK_RW_SYNC);
1310
1311                if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1312                        blk_set_congested(rl, BLK_RW_ASYNC);
1313                else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1314                        blk_clear_congested(rl, BLK_RW_ASYNC);
1315
1316                if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1317                        blk_set_rl_full(rl, BLK_RW_SYNC);
1318                } else {
1319                        blk_clear_rl_full(rl, BLK_RW_SYNC);
1320                        wake_up(&rl->wait[BLK_RW_SYNC]);
1321                }
1322
1323                if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1324                        blk_set_rl_full(rl, BLK_RW_ASYNC);
1325                } else {
1326                        blk_clear_rl_full(rl, BLK_RW_ASYNC);
1327                        wake_up(&rl->wait[BLK_RW_ASYNC]);
1328                }
1329        }
1330
1331        spin_unlock_irq(q->queue_lock);
1332        return 0;
1333}
1334
1335/**
1336 * __get_request - get a free request
1337 * @rl: request list to allocate from
1338 * @op: operation and flags
1339 * @bio: bio to allocate request for (can be %NULL)
1340 * @flags: BLQ_MQ_REQ_* flags
1341 * @gfp_mask: allocator flags
1342 *
1343 * Get a free request from @q.  This function may fail under memory
1344 * pressure or if @q is dead.
1345 *
1346 * Must be called with @q->queue_lock held and,
1347 * Returns ERR_PTR on failure, with @q->queue_lock held.
1348 * Returns request pointer on success, with @q->queue_lock *not held*.
1349 */
1350static struct request *__get_request(struct request_list *rl, unsigned int op,
1351                struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp_mask)
1352{
1353        struct request_queue *q = rl->q;
1354        struct request *rq;
1355        struct elevator_type *et = q->elevator->type;
1356        struct io_context *ioc = rq_ioc(bio);
1357        struct io_cq *icq = NULL;
1358        const bool is_sync = op_is_sync(op);
1359        int may_queue;
1360        req_flags_t rq_flags = RQF_ALLOCED;
1361
1362        lockdep_assert_held(q->queue_lock);
1363
1364        if (unlikely(blk_queue_dying(q)))
1365                return ERR_PTR(-ENODEV);
1366
1367        may_queue = elv_may_queue(q, op);
1368        if (may_queue == ELV_MQUEUE_NO)
1369                goto rq_starved;
1370
1371        if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1372                if (rl->count[is_sync]+1 >= q->nr_requests) {
1373                        /*
1374                         * The queue will fill after this allocation, so set
1375                         * it as full, and mark this process as "batching".
1376                         * This process will be allowed to complete a batch of
1377                         * requests, others will be blocked.
1378                         */
1379                        if (!blk_rl_full(rl, is_sync)) {
1380                                ioc_set_batching(q, ioc);
1381                                blk_set_rl_full(rl, is_sync);
1382                        } else {
1383                                if (may_queue != ELV_MQUEUE_MUST
1384                                                && !ioc_batching(q, ioc)) {
1385                                        /*
1386                                         * The queue is full and the allocating
1387                                         * process is not a "batcher", and not
1388                                         * exempted by the IO scheduler
1389                                         */
1390                                        return ERR_PTR(-ENOMEM);
1391                                }
1392                        }
1393                }
1394                blk_set_congested(rl, is_sync);
1395        }
1396
1397        /*
1398         * Only allow batching queuers to allocate up to 50% over the defined
1399         * limit of requests, otherwise we could have thousands of requests
1400         * allocated with any setting of ->nr_requests
1401         */
1402        if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1403                return ERR_PTR(-ENOMEM);
1404
1405        q->nr_rqs[is_sync]++;
1406        rl->count[is_sync]++;
1407        rl->starved[is_sync] = 0;
1408
1409        /*
1410         * Decide whether the new request will be managed by elevator.  If
1411         * so, mark @rq_flags and increment elvpriv.  Non-zero elvpriv will
1412         * prevent the current elevator from being destroyed until the new
1413         * request is freed.  This guarantees icq's won't be destroyed and
1414         * makes creating new ones safe.
1415         *
1416         * Flush requests do not use the elevator so skip initialization.
1417         * This allows a request to share the flush and elevator data.
1418         *
1419         * Also, lookup icq while holding queue_lock.  If it doesn't exist,
1420         * it will be created after releasing queue_lock.
1421         */
1422        if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1423                rq_flags |= RQF_ELVPRIV;
1424                q->nr_rqs_elvpriv++;
1425                if (et->icq_cache && ioc)
1426                        icq = ioc_lookup_icq(ioc, q);
1427        }
1428
1429        if (blk_queue_io_stat(q))
1430                rq_flags |= RQF_IO_STAT;
1431        spin_unlock_irq(q->queue_lock);
1432
1433        /* allocate and init request */
1434        rq = mempool_alloc(rl->rq_pool, gfp_mask);
1435        if (!rq)
1436                goto fail_alloc;
1437
1438        blk_rq_init(q, rq);
1439        blk_rq_set_rl(rq, rl);
1440        rq->cmd_flags = op;
1441        rq->rq_flags = rq_flags;
1442        if (flags & BLK_MQ_REQ_PREEMPT)
1443                rq->rq_flags |= RQF_PREEMPT;
1444
1445        /* init elvpriv */
1446        if (rq_flags & RQF_ELVPRIV) {
1447                if (unlikely(et->icq_cache && !icq)) {
1448                        if (ioc)
1449                                icq = ioc_create_icq(ioc, q, gfp_mask);
1450                        if (!icq)
1451                                goto fail_elvpriv;
1452                }
1453
1454                rq->elv.icq = icq;
1455                if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1456                        goto fail_elvpriv;
1457
1458                /* @rq->elv.icq holds io_context until @rq is freed */
1459                if (icq)
1460                        get_io_context(icq->ioc);
1461        }
1462out:
1463        /*
1464         * ioc may be NULL here, and ioc_batching will be false. That's
1465         * OK, if the queue is under the request limit then requests need
1466         * not count toward the nr_batch_requests limit. There will always
1467         * be some limit enforced by BLK_BATCH_TIME.
1468         */
1469        if (ioc_batching(q, ioc))
1470                ioc->nr_batch_requests--;
1471
1472        trace_block_getrq(q, bio, op);
1473        return rq;
1474
1475fail_elvpriv:
1476        /*
1477         * elvpriv init failed.  ioc, icq and elvpriv aren't mempool backed
1478         * and may fail indefinitely under memory pressure and thus
1479         * shouldn't stall IO.  Treat this request as !elvpriv.  This will
1480         * disturb iosched and blkcg but weird is bettern than dead.
1481         */
1482        printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1483                           __func__, dev_name(q->backing_dev_info->dev));
1484
1485        rq->rq_flags &= ~RQF_ELVPRIV;
1486        rq->elv.icq = NULL;
1487
1488        spin_lock_irq(q->queue_lock);
1489        q->nr_rqs_elvpriv--;
1490        spin_unlock_irq(q->queue_lock);
1491        goto out;
1492
1493fail_alloc:
1494        /*
1495         * Allocation failed presumably due to memory. Undo anything we
1496         * might have messed up.
1497         *
1498         * Allocating task should really be put onto the front of the wait
1499         * queue, but this is pretty rare.
1500         */
1501        spin_lock_irq(q->queue_lock);
1502        freed_request(rl, is_sync, rq_flags);
1503
1504        /*
1505         * in the very unlikely event that allocation failed and no
1506         * requests for this direction was pending, mark us starved so that
1507         * freeing of a request in the other direction will notice
1508         * us. another possible fix would be to split the rq mempool into
1509         * READ and WRITE
1510         */
1511rq_starved:
1512        if (unlikely(rl->count[is_sync] == 0))
1513                rl->starved[is_sync] = 1;
1514        return ERR_PTR(-ENOMEM);
1515}
1516
1517/**
1518 * get_request - get a free request
1519 * @q: request_queue to allocate request from
1520 * @op: operation and flags
1521 * @bio: bio to allocate request for (can be %NULL)
1522 * @flags: BLK_MQ_REQ_* flags.
1523 * @gfp: allocator flags
1524 *
1525 * Get a free request from @q.  If %BLK_MQ_REQ_NOWAIT is set in @flags,
1526 * this function keeps retrying under memory pressure and fails iff @q is dead.
1527 *
1528 * Must be called with @q->queue_lock held and,
1529 * Returns ERR_PTR on failure, with @q->queue_lock held.
1530 * Returns request pointer on success, with @q->queue_lock *not held*.
1531 */
1532static struct request *get_request(struct request_queue *q, unsigned int op,
1533                struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp)
1534{
1535        const bool is_sync = op_is_sync(op);
1536        DEFINE_WAIT(wait);
1537        struct request_list *rl;
1538        struct request *rq;
1539
1540        lockdep_assert_held(q->queue_lock);
1541        WARN_ON_ONCE(q->mq_ops);
1542
1543        rl = blk_get_rl(q, bio);        /* transferred to @rq on success */
1544retry:
1545        rq = __get_request(rl, op, bio, flags, gfp);
1546        if (!IS_ERR(rq))
1547                return rq;
1548
1549        if (op & REQ_NOWAIT) {
1550                blk_put_rl(rl);
1551                return ERR_PTR(-EAGAIN);
1552        }
1553
1554        if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1555                blk_put_rl(rl);
1556                return rq;
1557        }
1558
1559        /* wait on @rl and retry */
1560        prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1561                                  TASK_UNINTERRUPTIBLE);
1562
1563        trace_block_sleeprq(q, bio, op);
1564
1565        spin_unlock_irq(q->queue_lock);
1566        io_schedule();
1567
1568        /*
1569         * After sleeping, we become a "batching" process and will be able
1570         * to allocate at least one request, and up to a big batch of them
1571         * for a small period time.  See ioc_batching, ioc_set_batching
1572         */
1573        ioc_set_batching(q, current->io_context);
1574
1575        spin_lock_irq(q->queue_lock);
1576        finish_wait(&rl->wait[is_sync], &wait);
1577
1578        goto retry;
1579}
1580
1581/* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1582static struct request *blk_old_get_request(struct request_queue *q,
1583                                unsigned int op, blk_mq_req_flags_t flags)
1584{
1585        struct request *rq;
1586        gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : GFP_NOIO;
1587        int ret = 0;
1588
1589        WARN_ON_ONCE(q->mq_ops);
1590
1591        /* create ioc upfront */
1592        create_io_context(gfp_mask, q->node);
1593
1594        ret = blk_queue_enter(q, flags);
1595        if (ret)
1596                return ERR_PTR(ret);
1597        spin_lock_irq(q->queue_lock);
1598        rq = get_request(q, op, NULL, flags, gfp_mask);
1599        if (IS_ERR(rq)) {
1600                spin_unlock_irq(q->queue_lock);
1601                blk_queue_exit(q);
1602                return rq;
1603        }
1604
1605        /* q->queue_lock is unlocked at this point */
1606        rq->__data_len = 0;
1607        rq->__sector = (sector_t) -1;
1608        rq->bio = rq->biotail = NULL;
1609        return rq;
1610}
1611
1612/**
1613 * blk_get_request - allocate a request
1614 * @q: request queue to allocate a request for
1615 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1616 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1617 */
1618struct request *blk_get_request(struct request_queue *q, unsigned int op,
1619                                blk_mq_req_flags_t flags)
1620{
1621        struct request *req;
1622
1623        WARN_ON_ONCE(op & REQ_NOWAIT);
1624        WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1625
1626        if (q->mq_ops) {
1627                req = blk_mq_alloc_request(q, op, flags);
1628                if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1629                        q->mq_ops->initialize_rq_fn(req);
1630        } else {
1631                req = blk_old_get_request(q, op, flags);
1632                if (!IS_ERR(req) && q->initialize_rq_fn)
1633                        q->initialize_rq_fn(req);
1634        }
1635
1636        return req;
1637}
1638EXPORT_SYMBOL(blk_get_request);
1639
1640/**
1641 * blk_requeue_request - put a request back on queue
1642 * @q:          request queue where request should be inserted
1643 * @rq:         request to be inserted
1644 *
1645 * Description:
1646 *    Drivers often keep queueing requests until the hardware cannot accept
1647 *    more, when that condition happens we need to put the request back
1648 *    on the queue. Must be called with queue lock held.
1649 */
1650void blk_requeue_request(struct request_queue *q, struct request *rq)
1651{
1652        lockdep_assert_held(q->queue_lock);
1653        WARN_ON_ONCE(q->mq_ops);
1654
1655        blk_delete_timer(rq);
1656        blk_clear_rq_complete(rq);
1657        trace_block_rq_requeue(q, rq);
1658        rq_qos_requeue(q, rq);
1659
1660        if (rq->rq_flags & RQF_QUEUED)
1661                blk_queue_end_tag(q, rq);
1662
1663        BUG_ON(blk_queued_rq(rq));
1664
1665        elv_requeue_request(q, rq);
1666}
1667EXPORT_SYMBOL(blk_requeue_request);
1668
1669static void add_acct_request(struct request_queue *q, struct request *rq,
1670                             int where)
1671{
1672        blk_account_io_start(rq, true);
1673        __elv_add_request(q, rq, where);
1674}
1675
1676static void part_round_stats_single(struct request_queue *q, int cpu,
1677                                    struct hd_struct *part, unsigned long now,
1678                                    unsigned int inflight)
1679{
1680        if (inflight) {
1681                __part_stat_add(cpu, part, time_in_queue,
1682                                inflight * (now - part->stamp));
1683                __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1684        }
1685        part->stamp = now;
1686}
1687
1688/**
1689 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1690 * @q: target block queue
1691 * @cpu: cpu number for stats access
1692 * @part: target partition
1693 *
1694 * The average IO queue length and utilisation statistics are maintained
1695 * by observing the current state of the queue length and the amount of
1696 * time it has been in this state for.
1697 *
1698 * Normally, that accounting is done on IO completion, but that can result
1699 * in more than a second's worth of IO being accounted for within any one
1700 * second, leading to >100% utilisation.  To deal with that, we call this
1701 * function to do a round-off before returning the results when reading
1702 * /proc/diskstats.  This accounts immediately for all queue usage up to
1703 * the current jiffies and restarts the counters again.
1704 */
1705void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1706{
1707        struct hd_struct *part2 = NULL;
1708        unsigned long now = jiffies;
1709        unsigned int inflight[2];
1710        int stats = 0;
1711
1712        if (part->stamp != now)
1713                stats |= 1;
1714
1715        if (part->partno) {
1716                part2 = &part_to_disk(part)->part0;
1717                if (part2->stamp != now)
1718                        stats |= 2;
1719        }
1720
1721        if (!stats)
1722                return;
1723
1724        part_in_flight(q, part, inflight);
1725
1726        if (stats & 2)
1727                part_round_stats_single(q, cpu, part2, now, inflight[1]);
1728        if (stats & 1)
1729                part_round_stats_single(q, cpu, part, now, inflight[0]);
1730}
1731EXPORT_SYMBOL_GPL(part_round_stats);
1732
1733void __blk_put_request(struct request_queue *q, struct request *req)
1734{
1735        req_flags_t rq_flags = req->rq_flags;
1736
1737        if (unlikely(!q))
1738                return;
1739
1740        if (q->mq_ops) {
1741                blk_mq_free_request(req);
1742                return;
1743        }
1744
1745        lockdep_assert_held(q->queue_lock);
1746
1747        blk_req_zone_write_unlock(req);
1748        blk_pm_put_request(req);
1749        blk_pm_mark_last_busy(req);
1750
1751        elv_completed_request(q, req);
1752
1753        /* this is a bio leak */
1754        WARN_ON(req->bio != NULL);
1755
1756        rq_qos_done(q, req);
1757
1758        /*
1759         * Request may not have originated from ll_rw_blk. if not,
1760         * it didn't come out of our reserved rq pools
1761         */
1762        if (rq_flags & RQF_ALLOCED) {
1763                struct request_list *rl = blk_rq_rl(req);
1764                bool sync = op_is_sync(req->cmd_flags);
1765
1766                BUG_ON(!list_empty(&req->queuelist));
1767                BUG_ON(ELV_ON_HASH(req));
1768
1769                blk_free_request(rl, req);
1770                freed_request(rl, sync, rq_flags);
1771                blk_put_rl(rl);
1772                blk_queue_exit(q);
1773        }
1774}
1775EXPORT_SYMBOL_GPL(__blk_put_request);
1776
1777void blk_put_request(struct request *req)
1778{
1779        struct request_queue *q = req->q;
1780
1781        if (q->mq_ops)
1782                blk_mq_free_request(req);
1783        else {
1784                unsigned long flags;
1785
1786                spin_lock_irqsave(q->queue_lock, flags);
1787                __blk_put_request(q, req);
1788                spin_unlock_irqrestore(q->queue_lock, flags);
1789        }
1790}
1791EXPORT_SYMBOL(blk_put_request);
1792
1793bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1794                            struct bio *bio)
1795{
1796        const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1797
1798        if (!ll_back_merge_fn(q, req, bio))
1799                return false;
1800
1801        trace_block_bio_backmerge(q, req, bio);
1802
1803        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1804                blk_rq_set_mixed_merge(req);
1805
1806        req->biotail->bi_next = bio;
1807        req->biotail = bio;
1808        req->__data_len += bio->bi_iter.bi_size;
1809        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1810
1811        blk_account_io_start(req, false);
1812        return true;
1813}
1814
1815bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1816                             struct bio *bio)
1817{
1818        const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1819
1820        if (!ll_front_merge_fn(q, req, bio))
1821                return false;
1822
1823        trace_block_bio_frontmerge(q, req, bio);
1824
1825        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1826                blk_rq_set_mixed_merge(req);
1827
1828        bio->bi_next = req->bio;
1829        req->bio = bio;
1830
1831        req->__sector = bio->bi_iter.bi_sector;
1832        req->__data_len += bio->bi_iter.bi_size;
1833        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1834
1835        blk_account_io_start(req, false);
1836        return true;
1837}
1838
1839bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1840                struct bio *bio)
1841{
1842        unsigned short segments = blk_rq_nr_discard_segments(req);
1843
1844        if (segments >= queue_max_discard_segments(q))
1845                goto no_merge;
1846        if (blk_rq_sectors(req) + bio_sectors(bio) >
1847            blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1848                goto no_merge;
1849
1850        req->biotail->bi_next = bio;
1851        req->biotail = bio;
1852        req->__data_len += bio->bi_iter.bi_size;
1853        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1854        req->nr_phys_segments = segments + 1;
1855
1856        blk_account_io_start(req, false);
1857        return true;
1858no_merge:
1859        req_set_nomerge(q, req);
1860        return false;
1861}
1862
1863/**
1864 * blk_attempt_plug_merge - try to merge with %current's plugged list
1865 * @q: request_queue new bio is being queued at
1866 * @bio: new bio being queued
1867 * @request_count: out parameter for number of traversed plugged requests
1868 * @same_queue_rq: pointer to &struct request that gets filled in when
1869 * another request associated with @q is found on the plug list
1870 * (optional, may be %NULL)
1871 *
1872 * Determine whether @bio being queued on @q can be merged with a request
1873 * on %current's plugged list.  Returns %true if merge was successful,
1874 * otherwise %false.
1875 *
1876 * Plugging coalesces IOs from the same issuer for the same purpose without
1877 * going through @q->queue_lock.  As such it's more of an issuing mechanism
1878 * than scheduling, and the request, while may have elvpriv data, is not
1879 * added on the elevator at this point.  In addition, we don't have
1880 * reliable access to the elevator outside queue lock.  Only check basic
1881 * merging parameters without querying the elevator.
1882 *
1883 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1884 */
1885bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1886                            unsigned int *request_count,
1887                            struct request **same_queue_rq)
1888{
1889        struct blk_plug *plug;
1890        struct request *rq;
1891        struct list_head *plug_list;
1892
1893        plug = current->plug;
1894        if (!plug)
1895                return false;
1896        *request_count = 0;
1897
1898        if (q->mq_ops)
1899                plug_list = &plug->mq_list;
1900        else
1901                plug_list = &plug->list;
1902
1903        list_for_each_entry_reverse(rq, plug_list, queuelist) {
1904                bool merged = false;
1905
1906                if (rq->q == q) {
1907                        (*request_count)++;
1908                        /*
1909                         * Only blk-mq multiple hardware queues case checks the
1910                         * rq in the same queue, there should be only one such
1911                         * rq in a queue
1912                         **/
1913                        if (same_queue_rq)
1914                                *same_queue_rq = rq;
1915                }
1916
1917                if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1918                        continue;
1919
1920                switch (blk_try_merge(rq, bio)) {
1921                case ELEVATOR_BACK_MERGE:
1922                        merged = bio_attempt_back_merge(q, rq, bio);
1923                        break;
1924                case ELEVATOR_FRONT_MERGE:
1925                        merged = bio_attempt_front_merge(q, rq, bio);
1926                        break;
1927                case ELEVATOR_DISCARD_MERGE:
1928                        merged = bio_attempt_discard_merge(q, rq, bio);
1929                        break;
1930                default:
1931                        break;
1932                }
1933
1934                if (merged)
1935                        return true;
1936        }
1937
1938        return false;
1939}
1940
1941unsigned int blk_plug_queued_count(struct request_queue *q)
1942{
1943        struct blk_plug *plug;
1944        struct request *rq;
1945        struct list_head *plug_list;
1946        unsigned int ret = 0;
1947
1948        plug = current->plug;
1949        if (!plug)
1950                goto out;
1951
1952        if (q->mq_ops)
1953                plug_list = &plug->mq_list;
1954        else
1955                plug_list = &plug->list;
1956
1957        list_for_each_entry(rq, plug_list, queuelist) {
1958                if (rq->q == q)
1959                        ret++;
1960        }
1961out:
1962        return ret;
1963}
1964
1965void blk_init_request_from_bio(struct request *req, struct bio *bio)
1966{
1967        struct io_context *ioc = rq_ioc(bio);
1968
1969        if (bio->bi_opf & REQ_RAHEAD)
1970                req->cmd_flags |= REQ_FAILFAST_MASK;
1971
1972        req->__sector = bio->bi_iter.bi_sector;
1973        if (ioprio_valid(bio_prio(bio)))
1974                req->ioprio = bio_prio(bio);
1975        else if (ioc)
1976                req->ioprio = ioc->ioprio;
1977        else
1978                req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1979        req->write_hint = bio->bi_write_hint;
1980        blk_rq_bio_prep(req->q, req, bio);
1981}
1982EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1983
1984static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1985{
1986        struct blk_plug *plug;
1987        int where = ELEVATOR_INSERT_SORT;
1988        struct request *req, *free;
1989        unsigned int request_count = 0;
1990
1991        /*
1992         * low level driver can indicate that it wants pages above a
1993         * certain limit bounced to low memory (ie for highmem, or even
1994         * ISA dma in theory)
1995         */
1996        blk_queue_bounce(q, &bio);
1997
1998        blk_queue_split(q, &bio);
1999
2000        if (!bio_integrity_prep(bio))
2001                return BLK_QC_T_NONE;
2002
2003        if (op_is_flush(bio->bi_opf)) {
2004                spin_lock_irq(q->queue_lock);
2005                where = ELEVATOR_INSERT_FLUSH;
2006                goto get_rq;
2007        }
2008
2009        /*
2010         * Check if we can merge with the plugged list before grabbing
2011         * any locks.
2012         */
2013        if (!blk_queue_nomerges(q)) {
2014                if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
2015                        return BLK_QC_T_NONE;
2016        } else
2017                request_count = blk_plug_queued_count(q);
2018
2019        spin_lock_irq(q->queue_lock);
2020
2021        switch (elv_merge(q, &req, bio)) {
2022        case ELEVATOR_BACK_MERGE:
2023                if (!bio_attempt_back_merge(q, req, bio))
2024                        break;
2025                elv_bio_merged(q, req, bio);
2026                free = attempt_back_merge(q, req);
2027                if (free)
2028                        __blk_put_request(q, free);
2029                else
2030                        elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
2031                goto out_unlock;
2032        case ELEVATOR_FRONT_MERGE:
2033                if (!bio_attempt_front_merge(q, req, bio))
2034                        break;
2035                elv_bio_merged(q, req, bio);
2036                free = attempt_front_merge(q, req);
2037                if (free)
2038                        __blk_put_request(q, free);
2039                else
2040                        elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
2041                goto out_unlock;
2042        default:
2043                break;
2044        }
2045
2046get_rq:
2047        rq_qos_throttle(q, bio, q->queue_lock);
2048
2049        /*
2050         * Grab a free request. This is might sleep but can not fail.
2051         * Returns with the queue unlocked.
2052         */
2053        blk_queue_enter_live(q);
2054        req = get_request(q, bio->bi_opf, bio, 0, GFP_NOIO);
2055        if (IS_ERR(req)) {
2056                blk_queue_exit(q);
2057                rq_qos_cleanup(q, bio);
2058                if (PTR_ERR(req) == -ENOMEM)
2059                        bio->bi_status = BLK_STS_RESOURCE;
2060                else
2061                        bio->bi_status = BLK_STS_IOERR;
2062                bio_endio(bio);
2063                goto out_unlock;
2064        }
2065
2066        rq_qos_track(q, req, bio);
2067
2068        /*
2069         * After dropping the lock and possibly sleeping here, our request
2070         * may now be mergeable after it had proven unmergeable (above).
2071         * We don't worry about that case for efficiency. It won't happen
2072         * often, and the elevators are able to handle it.
2073         */
2074        blk_init_request_from_bio(req, bio);
2075
2076        if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
2077                req->cpu = raw_smp_processor_id();
2078
2079        plug = current->plug;
2080        if (plug) {
2081                /*
2082                 * If this is the first request added after a plug, fire
2083                 * of a plug trace.
2084                 *
2085                 * @request_count may become stale because of schedule
2086                 * out, so check plug list again.
2087                 */
2088                if (!request_count || list_empty(&plug->list))
2089                        trace_block_plug(q);
2090                else {
2091                        struct request *last = list_entry_rq(plug->list.prev);
2092                        if (request_count >= BLK_MAX_REQUEST_COUNT ||
2093                            blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
2094                                blk_flush_plug_list(plug, false);
2095                                trace_block_plug(q);
2096                        }
2097                }
2098                list_add_tail(&req->queuelist, &plug->list);
2099                blk_account_io_start(req, true);
2100        } else {
2101                spin_lock_irq(q->queue_lock);
2102                add_acct_request(q, req, where);
2103                __blk_run_queue(q);
2104out_unlock:
2105                spin_unlock_irq(q->queue_lock);
2106        }
2107
2108        return BLK_QC_T_NONE;
2109}
2110
2111static void handle_bad_sector(struct bio *bio, sector_t maxsector)
2112{
2113        char b[BDEVNAME_SIZE];
2114
2115        printk(KERN_INFO "attempt to access beyond end of device\n");
2116        printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
2117                        bio_devname(bio, b), bio->bi_opf,
2118                        (unsigned long long)bio_end_sector(bio),
2119                        (long long)maxsector);
2120}
2121
2122#ifdef CONFIG_FAIL_MAKE_REQUEST
2123
2124static DECLARE_FAULT_ATTR(fail_make_request);
2125
2126static int __init setup_fail_make_request(char *str)
2127{
2128        return setup_fault_attr(&fail_make_request, str);
2129}
2130__setup("fail_make_request=", setup_fail_make_request);
2131
2132static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
2133{
2134        return part->make_it_fail && should_fail(&fail_make_request, bytes);
2135}
2136
2137static int __init fail_make_request_debugfs(void)
2138{
2139        struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2140                                                NULL, &fail_make_request);
2141
2142        return PTR_ERR_OR_ZERO(dir);
2143}
2144
2145late_initcall(fail_make_request_debugfs);
2146
2147#else /* CONFIG_FAIL_MAKE_REQUEST */
2148
2149static inline bool should_fail_request(struct hd_struct *part,
2150                                        unsigned int bytes)
2151{
2152        return false;
2153}
2154
2155#endif /* CONFIG_FAIL_MAKE_REQUEST */
2156
2157static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2158{
2159        const int op = bio_op(bio);
2160
2161        if (part->policy && op_is_write(op)) {
2162                char b[BDEVNAME_SIZE];
2163
2164                if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
2165                        return false;
2166
2167                WARN_ONCE(1,
2168                       "generic_make_request: Trying to write "
2169                        "to read-only block-device %s (partno %d)\n",
2170                        bio_devname(bio, b), part->partno);
2171                /* Older lvm-tools actually trigger this */
2172                return false;
2173        }
2174
2175        return false;
2176}
2177
2178static noinline int should_fail_bio(struct bio *bio)
2179{
2180        if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2181                return -EIO;
2182        return 0;
2183}
2184ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2185
2186/*
2187 * Check whether this bio extends beyond the end of the device or partition.
2188 * This may well happen - the kernel calls bread() without checking the size of
2189 * the device, e.g., when mounting a file system.
2190 */
2191static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2192{
2193        unsigned int nr_sectors = bio_sectors(bio);
2194
2195        if (nr_sectors && maxsector &&
2196            (nr_sectors > maxsector ||
2197             bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2198                handle_bad_sector(bio, maxsector);
2199                return -EIO;
2200        }
2201        return 0;
2202}
2203
2204/*
2205 * Remap block n of partition p to block n+start(p) of the disk.
2206 */
2207static inline int blk_partition_remap(struct bio *bio)
2208{
2209        struct hd_struct *p;
2210        int ret = -EIO;
2211
2212        rcu_read_lock();
2213        p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2214        if (unlikely(!p))
2215                goto out;
2216        if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2217                goto out;
2218        if (unlikely(bio_check_ro(bio, p)))
2219                goto out;
2220
2221        /*
2222         * Zone reset does not include bi_size so bio_sectors() is always 0.
2223         * Include a test for the reset op code and perform the remap if needed.
2224         */
2225        if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2226                if (bio_check_eod(bio, part_nr_sects_read(p)))
2227                        goto out;
2228                bio->bi_iter.bi_sector += p->start_sect;
2229                trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2230                                      bio->bi_iter.bi_sector - p->start_sect);
2231        }
2232        bio->bi_partno = 0;
2233        ret = 0;
2234out:
2235        rcu_read_unlock();
2236        return ret;
2237}
2238
2239static noinline_for_stack bool
2240generic_make_request_checks(struct bio *bio)
2241{
2242        struct request_queue *q;
2243        int nr_sectors = bio_sectors(bio);
2244        blk_status_t status = BLK_STS_IOERR;
2245        char b[BDEVNAME_SIZE];
2246
2247        might_sleep();
2248
2249        q = bio->bi_disk->queue;
2250        if (unlikely(!q)) {
2251                printk(KERN_ERR
2252                       "generic_make_request: Trying to access "
2253                        "nonexistent block-device %s (%Lu)\n",
2254                        bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2255                goto end_io;
2256        }
2257
2258        /*
2259         * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2260         * if queue is not a request based queue.
2261         */
2262        if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2263                goto not_supported;
2264
2265        if (should_fail_bio(bio))
2266                goto end_io;
2267
2268        if (bio->bi_partno) {
2269                if (unlikely(blk_partition_remap(bio)))
2270                        goto end_io;
2271        } else {
2272                if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2273                        goto end_io;
2274                if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
2275                        goto end_io;
2276        }
2277
2278        /*
2279         * Filter flush bio's early so that make_request based
2280         * drivers without flush support don't have to worry
2281         * about them.
2282         */
2283        if (op_is_flush(bio->bi_opf) &&
2284            !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2285                bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2286                if (!nr_sectors) {
2287                        status = BLK_STS_OK;
2288                        goto end_io;
2289                }
2290        }
2291
2292        switch (bio_op(bio)) {
2293        case REQ_OP_DISCARD:
2294                if (!blk_queue_discard(q))
2295                        goto not_supported;
2296                break;
2297        case REQ_OP_SECURE_ERASE:
2298                if (!blk_queue_secure_erase(q))
2299                        goto not_supported;
2300                break;
2301        case REQ_OP_WRITE_SAME:
2302                if (!q->limits.max_write_same_sectors)
2303                        goto not_supported;
2304                break;
2305        case REQ_OP_ZONE_RESET:
2306                if (!blk_queue_is_zoned(q))
2307                        goto not_supported;
2308                break;
2309        case REQ_OP_WRITE_ZEROES:
2310                if (!q->limits.max_write_zeroes_sectors)
2311                        goto not_supported;
2312                break;
2313        default:
2314                break;
2315        }
2316
2317        /*
2318         * Various block parts want %current->io_context and lazy ioc
2319         * allocation ends up trading a lot of pain for a small amount of
2320         * memory.  Just allocate it upfront.  This may fail and block
2321         * layer knows how to live with it.
2322         */
2323        create_io_context(GFP_ATOMIC, q->node);
2324
2325        if (!blkcg_bio_issue_check(q, bio))
2326                return false;
2327
2328        if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2329                trace_block_bio_queue(q, bio);
2330                /* Now that enqueuing has been traced, we need to trace
2331                 * completion as well.
2332                 */
2333                bio_set_flag(bio, BIO_TRACE_COMPLETION);
2334        }
2335        return true;
2336
2337not_supported:
2338        status = BLK_STS_NOTSUPP;
2339end_io:
2340        bio->bi_status = status;
2341        bio_endio(bio);
2342        return false;
2343}
2344
2345/**
2346 * generic_make_request - hand a buffer to its device driver for I/O
2347 * @bio:  The bio describing the location in memory and on the device.
2348 *
2349 * generic_make_request() is used to make I/O requests of block
2350 * devices. It is passed a &struct bio, which describes the I/O that needs
2351 * to be done.
2352 *
2353 * generic_make_request() does not return any status.  The
2354 * success/failure status of the request, along with notification of
2355 * completion, is delivered asynchronously through the bio->bi_end_io
2356 * function described (one day) else where.
2357 *
2358 * The caller of generic_make_request must make sure that bi_io_vec
2359 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2360 * set to describe the device address, and the
2361 * bi_end_io and optionally bi_private are set to describe how
2362 * completion notification should be signaled.
2363 *
2364 * generic_make_request and the drivers it calls may use bi_next if this
2365 * bio happens to be merged with someone else, and may resubmit the bio to
2366 * a lower device by calling into generic_make_request recursively, which
2367 * means the bio should NOT be touched after the call to ->make_request_fn.
2368 */
2369blk_qc_t generic_make_request(struct bio *bio)
2370{
2371        /*
2372         * bio_list_on_stack[0] contains bios submitted by the current
2373         * make_request_fn.
2374         * bio_list_on_stack[1] contains bios that were submitted before
2375         * the current make_request_fn, but that haven't been processed
2376         * yet.
2377         */
2378        struct bio_list bio_list_on_stack[2];
2379        blk_mq_req_flags_t flags = 0;
2380        struct request_queue *q = bio->bi_disk->queue;
2381        blk_qc_t ret = BLK_QC_T_NONE;
2382
2383        if (bio->bi_opf & REQ_NOWAIT)
2384                flags = BLK_MQ_REQ_NOWAIT;
2385        if (bio_flagged(bio, BIO_QUEUE_ENTERED))
2386                blk_queue_enter_live(q);
2387        else if (blk_queue_enter(q, flags) < 0) {
2388                if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
2389                        bio_wouldblock_error(bio);
2390                else
2391                        bio_io_error(bio);
2392                return ret;
2393        }
2394
2395        if (!generic_make_request_checks(bio))
2396                goto out;
2397
2398        /*
2399         * We only want one ->make_request_fn to be active at a time, else
2400         * stack usage with stacked devices could be a problem.  So use
2401         * current->bio_list to keep a list of requests submited by a
2402         * make_request_fn function.  current->bio_list is also used as a
2403         * flag to say if generic_make_request is currently active in this
2404         * task or not.  If it is NULL, then no make_request is active.  If
2405         * it is non-NULL, then a make_request is active, and new requests
2406         * should be added at the tail
2407         */
2408        if (current->bio_list) {
2409                bio_list_add(&current->bio_list[0], bio);
2410                goto out;
2411        }
2412
2413        /* following loop may be a bit non-obvious, and so deserves some
2414         * explanation.
2415         * Before entering the loop, bio->bi_next is NULL (as all callers
2416         * ensure that) so we have a list with a single bio.
2417         * We pretend that we have just taken it off a longer list, so
2418         * we assign bio_list to a pointer to the bio_list_on_stack,
2419         * thus initialising the bio_list of new bios to be
2420         * added.  ->make_request() may indeed add some more bios
2421         * through a recursive call to generic_make_request.  If it
2422         * did, we find a non-NULL value in bio_list and re-enter the loop
2423         * from the top.  In this case we really did just take the bio
2424         * of the top of the list (no pretending) and so remove it from
2425         * bio_list, and call into ->make_request() again.
2426         */
2427        BUG_ON(bio->bi_next);
2428        bio_list_init(&bio_list_on_stack[0]);
2429        current->bio_list = bio_list_on_stack;
2430        do {
2431                bool enter_succeeded = true;
2432
2433                if (unlikely(q != bio->bi_disk->queue)) {
2434                        if (q)
2435                                blk_queue_exit(q);
2436                        q = bio->bi_disk->queue;
2437                        flags = 0;
2438                        if (bio->bi_opf & REQ_NOWAIT)
2439                                flags = BLK_MQ_REQ_NOWAIT;
2440                        if (blk_queue_enter(q, flags) < 0) {
2441                                enter_succeeded = false;
2442                                q = NULL;
2443                        }
2444                }
2445
2446                if (enter_succeeded) {
2447                        struct bio_list lower, same;
2448
2449                        /* Create a fresh bio_list for all subordinate requests */
2450                        bio_list_on_stack[1] = bio_list_on_stack[0];
2451                        bio_list_init(&bio_list_on_stack[0]);
2452                        ret = q->make_request_fn(q, bio);
2453
2454                        /* sort new bios into those for a lower level
2455                         * and those for the same level
2456                         */
2457                        bio_list_init(&lower);
2458                        bio_list_init(&same);
2459                        while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2460                                if (q == bio->bi_disk->queue)
2461                                        bio_list_add(&same, bio);
2462                                else
2463                                        bio_list_add(&lower, bio);
2464                        /* now assemble so we handle the lowest level first */
2465                        bio_list_merge(&bio_list_on_stack[0], &lower);
2466                        bio_list_merge(&bio_list_on_stack[0], &same);
2467                        bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2468                } else {
2469                        if (unlikely(!blk_queue_dying(q) &&
2470                                        (bio->bi_opf & REQ_NOWAIT)))
2471                                bio_wouldblock_error(bio);
2472                        else
2473                                bio_io_error(bio);
2474                }
2475                bio = bio_list_pop(&bio_list_on_stack[0]);
2476        } while (bio);
2477        current->bio_list = NULL; /* deactivate */
2478
2479out:
2480        if (q)
2481                blk_queue_exit(q);
2482        return ret;
2483}
2484EXPORT_SYMBOL(generic_make_request);
2485
2486/**
2487 * direct_make_request - hand a buffer directly to its device driver for I/O
2488 * @bio:  The bio describing the location in memory and on the device.
2489 *
2490 * This function behaves like generic_make_request(), but does not protect
2491 * against recursion.  Must only be used if the called driver is known
2492 * to not call generic_make_request (or direct_make_request) again from
2493 * its make_request function.  (Calling direct_make_request again from
2494 * a workqueue is perfectly fine as that doesn't recurse).
2495 */
2496blk_qc_t direct_make_request(struct bio *bio)
2497{
2498        struct request_queue *q = bio->bi_disk->queue;
2499        bool nowait = bio->bi_opf & REQ_NOWAIT;
2500        blk_qc_t ret;
2501
2502        if (!generic_make_request_checks(bio))
2503                return BLK_QC_T_NONE;
2504
2505        if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2506                if (nowait && !blk_queue_dying(q))
2507                        bio->bi_status = BLK_STS_AGAIN;
2508                else
2509                        bio->bi_status = BLK_STS_IOERR;
2510                bio_endio(bio);
2511                return BLK_QC_T_NONE;
2512        }
2513
2514        ret = q->make_request_fn(q, bio);
2515        blk_queue_exit(q);
2516        return ret;
2517}
2518EXPORT_SYMBOL_GPL(direct_make_request);
2519
2520/**
2521 * submit_bio - submit a bio to the block device layer for I/O
2522 * @bio: The &struct bio which describes the I/O
2523 *
2524 * submit_bio() is very similar in purpose to generic_make_request(), and
2525 * uses that function to do most of the work. Both are fairly rough
2526 * interfaces; @bio must be presetup and ready for I/O.
2527 *
2528 */
2529blk_qc_t submit_bio(struct bio *bio)
2530{
2531        /*
2532         * If it's a regular read/write or a barrier with data attached,
2533         * go through the normal accounting stuff before submission.
2534         */
2535        if (bio_has_data(bio)) {
2536                unsigned int count;
2537
2538                if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2539                        count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
2540                else
2541                        count = bio_sectors(bio);
2542
2543                if (op_is_write(bio_op(bio))) {
2544                        count_vm_events(PGPGOUT, count);
2545                } else {
2546                        task_io_account_read(bio->bi_iter.bi_size);
2547                        count_vm_events(PGPGIN, count);
2548                }
2549
2550                if (unlikely(block_dump)) {
2551                        char b[BDEVNAME_SIZE];
2552                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2553                        current->comm, task_pid_nr(current),
2554                                op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2555                                (unsigned long long)bio->bi_iter.bi_sector,
2556                                bio_devname(bio, b), count);
2557                }
2558        }
2559
2560        return generic_make_request(bio);
2561}
2562EXPORT_SYMBOL(submit_bio);
2563
2564bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2565{
2566        if (!q->poll_fn || !blk_qc_t_valid(cookie))
2567                return false;
2568
2569        if (current->plug)
2570                blk_flush_plug_list(current->plug, false);
2571        return q->poll_fn(q, cookie);
2572}
2573EXPORT_SYMBOL_GPL(blk_poll);
2574
2575/**
2576 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2577 *                              for new the queue limits
2578 * @q:  the queue
2579 * @rq: the request being checked
2580 *
2581 * Description:
2582 *    @rq may have been made based on weaker limitations of upper-level queues
2583 *    in request stacking drivers, and it may violate the limitation of @q.
2584 *    Since the block layer and the underlying device driver trust @rq
2585 *    after it is inserted to @q, it should be checked against @q before
2586 *    the insertion using this generic function.
2587 *
2588 *    Request stacking drivers like request-based dm may change the queue
2589 *    limits when retrying requests on other queues. Those requests need
2590 *    to be checked against the new queue limits again during dispatch.
2591 */
2592static int blk_cloned_rq_check_limits(struct request_queue *q,
2593                                      struct request *rq)
2594{
2595        if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2596                printk(KERN_ERR "%s: over max size limit.\n", __func__);
2597                return -EIO;
2598        }
2599
2600        /*
2601         * queue's settings related to segment counting like q->bounce_pfn
2602         * may differ from that of other stacking queues.
2603         * Recalculate it to check the request correctly on this queue's
2604         * limitation.
2605         */
2606        blk_recalc_rq_segments(rq);
2607        if (rq->nr_phys_segments > queue_max_segments(q)) {
2608                printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2609                return -EIO;
2610        }
2611
2612        return 0;
2613}
2614
2615/**
2616 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2617 * @q:  the queue to submit the request
2618 * @rq: the request being queued
2619 */
2620blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2621{
2622        unsigned long flags;
2623        int where = ELEVATOR_INSERT_BACK;
2624
2625        if (blk_cloned_rq_check_limits(q, rq))
2626                return BLK_STS_IOERR;
2627
2628        if (rq->rq_disk &&
2629            should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2630                return BLK_STS_IOERR;
2631
2632        if (q->mq_ops) {
2633                if (blk_queue_io_stat(q))
2634                        blk_account_io_start(rq, true);
2635                /*
2636                 * Since we have a scheduler attached on the top device,
2637                 * bypass a potential scheduler on the bottom device for
2638                 * insert.
2639                 */
2640                return blk_mq_request_issue_directly(rq);
2641        }
2642
2643        spin_lock_irqsave(q->queue_lock, flags);
2644        if (unlikely(blk_queue_dying(q))) {
2645                spin_unlock_irqrestore(q->queue_lock, flags);
2646                return BLK_STS_IOERR;
2647        }
2648
2649        /*
2650         * Submitting request must be dequeued before calling this function
2651         * because it will be linked to another request_queue
2652         */
2653        BUG_ON(blk_queued_rq(rq));
2654
2655        if (op_is_flush(rq->cmd_flags))
2656                where = ELEVATOR_INSERT_FLUSH;
2657
2658        add_acct_request(q, rq, where);
2659        if (where == ELEVATOR_INSERT_FLUSH)
2660                __blk_run_queue(q);
2661        spin_unlock_irqrestore(q->queue_lock, flags);
2662
2663        return BLK_STS_OK;
2664}
2665EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2666
2667/**
2668 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2669 * @rq: request to examine
2670 *
2671 * Description:
2672 *     A request could be merge of IOs which require different failure
2673 *     handling.  This function determines the number of bytes which
2674 *     can be failed from the beginning of the request without
2675 *     crossing into area which need to be retried further.
2676 *
2677 * Return:
2678 *     The number of bytes to fail.
2679 */
2680unsigned int blk_rq_err_bytes(const struct request *rq)
2681{
2682        unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2683        unsigned int bytes = 0;
2684        struct bio *bio;
2685
2686        if (!(rq->rq_flags & RQF_MIXED_MERGE))
2687                return blk_rq_bytes(rq);
2688
2689        /*
2690         * Currently the only 'mixing' which can happen is between
2691         * different fastfail types.  We can safely fail portions
2692         * which have all the failfast bits that the first one has -
2693         * the ones which are at least as eager to fail as the first
2694         * one.
2695         */
2696        for (bio = rq->bio; bio; bio = bio->bi_next) {
2697                if ((bio->bi_opf & ff) != ff)
2698                        break;
2699                bytes += bio->bi_iter.bi_size;
2700        }
2701
2702        /* this could lead to infinite loop */
2703        BUG_ON(blk_rq_bytes(rq) && !bytes);
2704        return bytes;
2705}
2706EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2707
2708void blk_account_io_completion(struct request *req, unsigned int bytes)
2709{
2710        if (blk_do_io_stat(req)) {
2711                const int sgrp = op_stat_group(req_op(req));
2712                struct hd_struct *part;
2713                int cpu;
2714
2715                cpu = part_stat_lock();
2716                part = req->part;
2717                part_stat_add(cpu, part, sectors[sgrp], bytes >> 9);
2718                part_stat_unlock();
2719        }
2720}
2721
2722void blk_account_io_done(struct request *req, u64 now)
2723{
2724        /*
2725         * Account IO completion.  flush_rq isn't accounted as a
2726         * normal IO on queueing nor completion.  Accounting the
2727         * containing request is enough.
2728         */
2729        if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2730                const int sgrp = op_stat_group(req_op(req));
2731                struct hd_struct *part;
2732                int cpu;
2733
2734                cpu = part_stat_lock();
2735                part = req->part;
2736
2737                part_stat_inc(cpu, part, ios[sgrp]);
2738                part_stat_add(cpu, part, nsecs[sgrp], now - req->start_time_ns);
2739                part_round_stats(req->q, cpu, part);
2740                part_dec_in_flight(req->q, part, rq_data_dir(req));
2741
2742                hd_struct_put(part);
2743                part_stat_unlock();
2744        }
2745}
2746
2747void blk_account_io_start(struct request *rq, bool new_io)
2748{
2749        struct hd_struct *part;
2750        int rw = rq_data_dir(rq);
2751        int cpu;
2752
2753        if (!blk_do_io_stat(rq))
2754                return;
2755
2756        cpu = part_stat_lock();
2757
2758        if (!new_io) {
2759                part = rq->part;
2760                part_stat_inc(cpu, part, merges[rw]);
2761        } else {
2762                part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2763                if (!hd_struct_try_get(part)) {
2764                        /*
2765                         * The partition is already being removed,
2766                         * the request will be accounted on the disk only
2767                         *
2768                         * We take a reference on disk->part0 although that
2769                         * partition will never be deleted, so we can treat
2770                         * it as any other partition.
2771                         */
2772                        part = &rq->rq_disk->part0;
2773                        hd_struct_get(part);
2774                }
2775                part_round_stats(rq->q, cpu, part);
2776                part_inc_in_flight(rq->q, part, rw);
2777                rq->part = part;
2778        }
2779
2780        part_stat_unlock();
2781}
2782
2783static struct request *elv_next_request(struct request_queue *q)
2784{
2785        struct request *rq;
2786        struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2787
2788        WARN_ON_ONCE(q->mq_ops);
2789
2790        while (1) {
2791                list_for_each_entry(rq, &q->queue_head, queuelist) {
2792#ifdef CONFIG_PM
2793                        /*
2794                         * If a request gets queued in state RPM_SUSPENDED
2795                         * then that's a kernel bug.
2796                         */
2797                        WARN_ON_ONCE(q->rpm_status == RPM_SUSPENDED);
2798#endif
2799                        return rq;
2800                }
2801
2802                /*
2803                 * Flush request is running and flush request isn't queueable
2804                 * in the drive, we can hold the queue till flush request is
2805                 * finished. Even we don't do this, driver can't dispatch next
2806                 * requests and will requeue them. And this can improve
2807                 * throughput too. For example, we have request flush1, write1,
2808                 * flush 2. flush1 is dispatched, then queue is hold, write1
2809                 * isn't inserted to queue. After flush1 is finished, flush2
2810                 * will be dispatched. Since disk cache is already clean,
2811                 * flush2 will be finished very soon, so looks like flush2 is
2812                 * folded to flush1.
2813                 * Since the queue is hold, a flag is set to indicate the queue
2814                 * should be restarted later. Please see flush_end_io() for
2815                 * details.
2816                 */
2817                if (fq->flush_pending_idx != fq->flush_running_idx &&
2818                                !queue_flush_queueable(q)) {
2819                        fq->flush_queue_delayed = 1;
2820                        return NULL;
2821                }
2822                if (unlikely(blk_queue_bypass(q)) ||
2823                    !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2824                        return NULL;
2825        }
2826}
2827
2828/**
2829 * blk_peek_request - peek at the top of a request queue
2830 * @q: request queue to peek at
2831 *
2832 * Description:
2833 *     Return the request at the top of @q.  The returned request
2834 *     should be started using blk_start_request() before LLD starts
2835 *     processing it.
2836 *
2837 * Return:
2838 *     Pointer to the request at the top of @q if available.  Null
2839 *     otherwise.
2840 */
2841struct request *blk_peek_request(struct request_queue *q)
2842{
2843        struct request *rq;
2844        int ret;
2845
2846        lockdep_assert_held(q->queue_lock);
2847        WARN_ON_ONCE(q->mq_ops);
2848
2849        while ((rq = elv_next_request(q)) != NULL) {
2850                if (!(rq->rq_flags & RQF_STARTED)) {
2851                        /*
2852                         * This is the first time the device driver
2853                         * sees this request (possibly after
2854                         * requeueing).  Notify IO scheduler.
2855                         */
2856                        if (rq->rq_flags & RQF_SORTED)
2857                                elv_activate_rq(q, rq);
2858
2859                        /*
2860                         * just mark as started even if we don't start
2861                         * it, a request that has been delayed should
2862                         * not be passed by new incoming requests
2863                         */
2864                        rq->rq_flags |= RQF_STARTED;
2865                        trace_block_rq_issue(q, rq);
2866                }
2867
2868                if (!q->boundary_rq || q->boundary_rq == rq) {
2869                        q->end_sector = rq_end_sector(rq);
2870                        q->boundary_rq = NULL;
2871                }
2872
2873                if (rq->rq_flags & RQF_DONTPREP)
2874                        break;
2875
2876                if (q->dma_drain_size && blk_rq_bytes(rq)) {
2877                        /*
2878                         * make sure space for the drain appears we
2879                         * know we can do this because max_hw_segments
2880                         * has been adjusted to be one fewer than the
2881                         * device can handle
2882                         */
2883                        rq->nr_phys_segments++;
2884                }
2885
2886                if (!q->prep_rq_fn)
2887                        break;
2888
2889                ret = q->prep_rq_fn(q, rq);
2890                if (ret == BLKPREP_OK) {
2891                        break;
2892                } else if (ret == BLKPREP_DEFER) {
2893                        /*
2894                         * the request may have been (partially) prepped.
2895                         * we need to keep this request in the front to
2896                         * avoid resource deadlock.  RQF_STARTED will
2897                         * prevent other fs requests from passing this one.
2898                         */
2899                        if (q->dma_drain_size && blk_rq_bytes(rq) &&
2900                            !(rq->rq_flags & RQF_DONTPREP)) {
2901                                /*
2902                                 * remove the space for the drain we added
2903                                 * so that we don't add it again
2904                                 */
2905                                --rq->nr_phys_segments;
2906                        }
2907
2908                        rq = NULL;
2909                        break;
2910                } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2911                        rq->rq_flags |= RQF_QUIET;
2912                        /*
2913                         * Mark this request as started so we don't trigger
2914                         * any debug logic in the end I/O path.
2915                         */
2916                        blk_start_request(rq);
2917                        __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2918                                        BLK_STS_TARGET : BLK_STS_IOERR);
2919                } else {
2920                        printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2921                        break;
2922                }
2923        }
2924
2925        return rq;
2926}
2927EXPORT_SYMBOL(blk_peek_request);
2928
2929static void blk_dequeue_request(struct request *rq)
2930{
2931        struct request_queue *q = rq->q;
2932
2933        BUG_ON(list_empty(&rq->queuelist));
2934        BUG_ON(ELV_ON_HASH(rq));
2935
2936        list_del_init(&rq->queuelist);
2937
2938        /*
2939         * the time frame between a request being removed from the lists
2940         * and to it is freed is accounted as io that is in progress at
2941         * the driver side.
2942         */
2943        if (blk_account_rq(rq))
2944                q->in_flight[rq_is_sync(rq)]++;
2945}
2946
2947/**
2948 * blk_start_request - start request processing on the driver
2949 * @req: request to dequeue
2950 *
2951 * Description:
2952 *     Dequeue @req and start timeout timer on it.  This hands off the
2953 *     request to the driver.
2954 */
2955void blk_start_request(struct request *req)
2956{
2957        lockdep_assert_held(req->q->queue_lock);
2958        WARN_ON_ONCE(req->q->mq_ops);
2959
2960        blk_dequeue_request(req);
2961
2962        if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2963                req->io_start_time_ns = ktime_get_ns();
2964#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2965                req->throtl_size = blk_rq_sectors(req);
2966#endif
2967                req->rq_flags |= RQF_STATS;
2968                rq_qos_issue(req->q, req);
2969        }
2970
2971        BUG_ON(blk_rq_is_complete(req));
2972        blk_add_timer(req);
2973}
2974EXPORT_SYMBOL(blk_start_request);
2975
2976/**
2977 * blk_fetch_request - fetch a request from a request queue
2978 * @q: request queue to fetch a request from
2979 *
2980 * Description:
2981 *     Return the request at the top of @q.  The request is started on
2982 *     return and LLD can start processing it immediately.
2983 *
2984 * Return:
2985 *     Pointer to the request at the top of @q if available.  Null
2986 *     otherwise.
2987 */
2988struct request *blk_fetch_request(struct request_queue *q)
2989{
2990        struct request *rq;
2991
2992        lockdep_assert_held(q->queue_lock);
2993        WARN_ON_ONCE(q->mq_ops);
2994
2995        rq = blk_peek_request(q);
2996        if (rq)
2997                blk_start_request(rq);
2998        return rq;
2999}
3000EXPORT_SYMBOL(blk_fetch_request);
3001
3002/*
3003 * Steal bios from a request and add them to a bio list.
3004 * The request must not have been partially completed before.
3005 */
3006void blk_steal_bios(struct bio_list *list, struct request *rq)
3007{
3008        if (rq->bio) {
3009                if (list->tail)
3010                        list->tail->bi_next = rq->bio;
3011                else
3012                        list->head = rq->bio;
3013                list->tail = rq->biotail;
3014
3015                rq->bio = NULL;
3016                rq->biotail = NULL;
3017        }
3018
3019        rq->__data_len = 0;
3020}
3021EXPORT_SYMBOL_GPL(blk_steal_bios);
3022
3023/**
3024 * blk_update_request - Special helper function for request stacking drivers
3025 * @req:      the request being processed
3026 * @error:    block status code
3027 * @nr_bytes: number of bytes to complete @req
3028 *
3029 * Description:
3030 *     Ends I/O on a number of bytes attached to @req, but doesn't complete
3031 *     the request structure even if @req doesn't have leftover.
3032 *     If @req has leftover, sets it up for the next range of segments.
3033 *
3034 *     This special helper function is only for request stacking drivers
3035 *     (e.g. request-based dm) so that they can handle partial completion.
3036 *     Actual device drivers should use blk_end_request instead.
3037 *
3038 *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3039 *     %false return from this function.
3040 *
3041 * Note:
3042 *      The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
3043 *      blk_rq_bytes() and in blk_update_request().
3044 *
3045 * Return:
3046 *     %false - this request doesn't have any more data
3047 *     %true  - this request has more data
3048 **/
3049bool blk_update_request(struct request *req, blk_status_t error,
3050                unsigned int nr_bytes)
3051{
3052        int total_bytes;
3053
3054        trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
3055
3056        if (!req->bio)
3057                return false;
3058
3059        if (unlikely(error && !blk_rq_is_passthrough(req) &&
3060                     !(req->rq_flags & RQF_QUIET)))
3061                print_req_error(req, error);
3062
3063        blk_account_io_completion(req, nr_bytes);
3064
3065        total_bytes = 0;
3066        while (req->bio) {
3067                struct bio *bio = req->bio;
3068                unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3069
3070                if (bio_bytes == bio->bi_iter.bi_size)
3071                        req->bio = bio->bi_next;
3072
3073                /* Completion has already been traced */
3074                bio_clear_flag(bio, BIO_TRACE_COMPLETION);
3075                req_bio_endio(req, bio, bio_bytes, error);
3076
3077                total_bytes += bio_bytes;
3078                nr_bytes -= bio_bytes;
3079
3080                if (!nr_bytes)
3081                        break;
3082        }
3083
3084        /*
3085         * completely done
3086         */
3087        if (!req->bio) {
3088                /*
3089                 * Reset counters so that the request stacking driver
3090                 * can find how many bytes remain in the request
3091                 * later.
3092                 */
3093                req->__data_len = 0;
3094                return false;
3095        }
3096
3097        req->__data_len -= total_bytes;
3098
3099        /* update sector only for requests with clear definition of sector */
3100        if (!blk_rq_is_passthrough(req))
3101                req->__sector += total_bytes >> 9;
3102
3103        /* mixed attributes always follow the first bio */
3104        if (req->rq_flags & RQF_MIXED_MERGE) {
3105                req->cmd_flags &= ~REQ_FAILFAST_MASK;
3106                req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
3107        }
3108
3109        if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3110                /*
3111                 * If total number of sectors is less than the first segment
3112                 * size, something has gone terribly wrong.
3113                 */
3114                if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3115                        blk_dump_rq_flags(req, "request botched");
3116                        req->__data_len = blk_rq_cur_bytes(req);
3117                }
3118
3119                /* recalculate the number of segments */
3120                blk_recalc_rq_segments(req);
3121        }
3122
3123        return true;
3124}
3125EXPORT_SYMBOL_GPL(blk_update_request);
3126
3127static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3128                                    unsigned int nr_bytes,
3129                                    unsigned int bidi_bytes)
3130{
3131        if (blk_update_request(rq, error, nr_bytes))
3132                return true;
3133
3134        /* Bidi request must be completed as a whole */
3135        if (unlikely(blk_bidi_rq(rq)) &&
3136            blk_update_request(rq->next_rq, error, bidi_bytes))
3137                return true;
3138
3139        if (blk_queue_add_random(rq->q))
3140                add_disk_randomness(rq->rq_disk);
3141
3142        return false;
3143}
3144
3145/**
3146 * blk_unprep_request - unprepare a request
3147 * @req:        the request
3148 *
3149 * This function makes a request ready for complete resubmission (or
3150 * completion).  It happens only after all error handling is complete,
3151 * so represents the appropriate moment to deallocate any resources
3152 * that were allocated to the request in the prep_rq_fn.  The queue
3153 * lock is held when calling this.
3154 */
3155void blk_unprep_request(struct request *req)
3156{
3157        struct request_queue *q = req->q;
3158
3159        req->rq_flags &= ~RQF_DONTPREP;
3160        if (q->unprep_rq_fn)
3161                q->unprep_rq_fn(q, req);
3162}
3163EXPORT_SYMBOL_GPL(blk_unprep_request);
3164
3165void blk_finish_request(struct request *req, blk_status_t error)
3166{
3167        struct request_queue *q = req->q;
3168        u64 now = ktime_get_ns();
3169
3170        lockdep_assert_held(req->q->queue_lock);
3171        WARN_ON_ONCE(q->mq_ops);
3172
3173        if (req->rq_flags & RQF_STATS)
3174                blk_stat_add(req, now);
3175
3176        if (req->rq_flags & RQF_QUEUED)
3177                blk_queue_end_tag(q, req);
3178
3179        BUG_ON(blk_queued_rq(req));
3180
3181        if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3182                laptop_io_completion(req->q->backing_dev_info);
3183
3184        blk_delete_timer(req);
3185
3186        if (req->rq_flags & RQF_DONTPREP)
3187                blk_unprep_request(req);
3188
3189        blk_account_io_done(req, now);
3190
3191        if (req->end_io) {
3192                rq_qos_done(q, req);
3193                req->end_io(req, error);
3194        } else {
3195                if (blk_bidi_rq(req))
3196                        __blk_put_request(req->next_rq->q, req->next_rq);
3197
3198                __blk_put_request(q, req);
3199        }
3200}
3201EXPORT_SYMBOL(blk_finish_request);
3202
3203/**
3204 * blk_end_bidi_request - Complete a bidi request
3205 * @rq:         the request to complete
3206 * @error:      block status code
3207 * @nr_bytes:   number of bytes to complete @rq
3208 * @bidi_bytes: number of bytes to complete @rq->next_rq
3209 *
3210 * Description:
3211 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3212 *     Drivers that supports bidi can safely call this member for any
3213 *     type of request, bidi or uni.  In the later case @bidi_bytes is
3214 *     just ignored.
3215 *
3216 * Return:
3217 *     %false - we are done with this request
3218 *     %true  - still buffers pending for this request
3219 **/
3220static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3221                                 unsigned int nr_bytes, unsigned int bidi_bytes)
3222{
3223        struct request_queue *q = rq->q;
3224        unsigned long flags;
3225
3226        WARN_ON_ONCE(q->mq_ops);
3227
3228        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3229                return true;
3230
3231        spin_lock_irqsave(q->queue_lock, flags);
3232        blk_finish_request(rq, error);
3233        spin_unlock_irqrestore(q->queue_lock, flags);
3234
3235        return false;
3236}
3237
3238/**
3239 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3240 * @rq:         the request to complete
3241 * @error:      block status code
3242 * @nr_bytes:   number of bytes to complete @rq
3243 * @bidi_bytes: number of bytes to complete @rq->next_rq
3244 *
3245 * Description:
3246 *     Identical to blk_end_bidi_request() except that queue lock is
3247 *     assumed to be locked on entry and remains so on return.
3248 *
3249 * Return:
3250 *     %false - we are done with this request
3251 *     %true  - still buffers pending for this request
3252 **/
3253static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3254                                   unsigned int nr_bytes, unsigned int bidi_bytes)
3255{
3256        lockdep_assert_held(rq->q->queue_lock);
3257        WARN_ON_ONCE(rq->q->mq_ops);
3258
3259        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3260                return true;
3261
3262        blk_finish_request(rq, error);
3263
3264        return false;
3265}
3266
3267/**
3268 * blk_end_request - Helper function for drivers to complete the request.
3269 * @rq:       the request being processed
3270 * @error:    block status code
3271 * @nr_bytes: number of bytes to complete
3272 *
3273 * Description:
3274 *     Ends I/O on a number of bytes attached to @rq.
3275 *     If @rq has leftover, sets it up for the next range of segments.
3276 *
3277 * Return:
3278 *     %false - we are done with this request
3279 *     %true  - still buffers pending for this request
3280 **/
3281bool blk_end_request(struct request *rq, blk_status_t error,
3282                unsigned int nr_bytes)
3283{
3284        WARN_ON_ONCE(rq->q->mq_ops);
3285        return blk_end_bidi_request(rq, error, nr_bytes, 0);
3286}
3287EXPORT_SYMBOL(blk_end_request);
3288
3289/**
3290 * blk_end_request_all - Helper function for drives to finish the request.
3291 * @rq: the request to finish
3292 * @error: block status code
3293 *
3294 * Description:
3295 *     Completely finish @rq.
3296 */
3297void blk_end_request_all(struct request *rq, blk_status_t error)
3298{
3299        bool pending;
3300        unsigned int bidi_bytes = 0;
3301
3302        if (unlikely(blk_bidi_rq(rq)))
3303                bidi_bytes = blk_rq_bytes(rq->next_rq);
3304
3305        pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3306        BUG_ON(pending);
3307}
3308EXPORT_SYMBOL(blk_end_request_all);
3309
3310/**
3311 * __blk_end_request - Helper function for drivers to complete the request.
3312 * @rq:       the request being processed
3313 * @error:    block status code
3314 * @nr_bytes: number of bytes to complete
3315 *
3316 * Description:
3317 *     Must be called with queue lock held unlike blk_end_request().
3318 *
3319 * Return:
3320 *     %false - we are done with this request
3321 *     %true  - still buffers pending for this request
3322 **/
3323bool __blk_end_request(struct request *rq, blk_status_t error,
3324                unsigned int nr_bytes)
3325{
3326        lockdep_assert_held(rq->q->queue_lock);
3327        WARN_ON_ONCE(rq->q->mq_ops);
3328
3329        return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3330}
3331EXPORT_SYMBOL(__blk_end_request);
3332
3333/**
3334 * __blk_end_request_all - Helper function for drives to finish the request.
3335 * @rq: the request to finish
3336 * @error:    block status code
3337 *
3338 * Description:
3339 *     Completely finish @rq.  Must be called with queue lock held.
3340 */
3341void __blk_end_request_all(struct request *rq, blk_status_t error)
3342{
3343        bool pending;
3344        unsigned int bidi_bytes = 0;
3345
3346        lockdep_assert_held(rq->q->queue_lock);
3347        WARN_ON_ONCE(rq->q->mq_ops);
3348
3349        if (unlikely(blk_bidi_rq(rq)))
3350                bidi_bytes = blk_rq_bytes(rq->next_rq);
3351
3352        pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3353        BUG_ON(pending);
3354}
3355EXPORT_SYMBOL(__blk_end_request_all);
3356
3357/**
3358 * __blk_end_request_cur - Helper function to finish the current request chunk.
3359 * @rq: the request to finish the current chunk for
3360 * @error:    block status code
3361 *
3362 * Description:
3363 *     Complete the current consecutively mapped chunk from @rq.  Must
3364 *     be called with queue lock held.
3365 *
3366 * Return:
3367 *     %false - we are done with this request
3368 *     %true  - still buffers pending for this request
3369 */
3370bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3371{
3372        return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3373}
3374EXPORT_SYMBOL(__blk_end_request_cur);
3375
3376void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3377                     struct bio *bio)
3378{
3379        if (bio_has_data(bio))
3380                rq->nr_phys_segments = bio_phys_segments(q, bio);
3381        else if (bio_op(bio) == REQ_OP_DISCARD)
3382                rq->nr_phys_segments = 1;
3383
3384        rq->__data_len = bio->bi_iter.bi_size;
3385        rq->bio = rq->biotail = bio;
3386
3387        if (bio->bi_disk)
3388                rq->rq_disk = bio->bi_disk;
3389}
3390
3391#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3392/**
3393 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3394 * @rq: the request to be flushed
3395 *
3396 * Description:
3397 *     Flush all pages in @rq.
3398 */
3399void rq_flush_dcache_pages(struct request *rq)
3400{
3401        struct req_iterator iter;
3402        struct bio_vec bvec;
3403
3404        rq_for_each_segment(bvec, rq, iter)
3405                flush_dcache_page(bvec.bv_page);
3406}
3407EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3408#endif
3409
3410/**
3411 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3412 * @q : the queue of the device being checked
3413 *
3414 * Description:
3415 *    Check if underlying low-level drivers of a device are busy.
3416 *    If the drivers want to export their busy state, they must set own
3417 *    exporting function using blk_queue_lld_busy() first.
3418 *
3419 *    Basically, this function is used only by request stacking drivers
3420 *    to stop dispatching requests to underlying devices when underlying
3421 *    devices are busy.  This behavior helps more I/O merging on the queue
3422 *    of the request stacking driver and prevents I/O throughput regression
3423 *    on burst I/O load.
3424 *
3425 * Return:
3426 *    0 - Not busy (The request stacking driver should dispatch request)
3427 *    1 - Busy (The request stacking driver should stop dispatching request)
3428 */
3429int blk_lld_busy(struct request_queue *q)
3430{
3431        if (q->lld_busy_fn)
3432                return q->lld_busy_fn(q);
3433
3434        return 0;
3435}
3436EXPORT_SYMBOL_GPL(blk_lld_busy);
3437
3438/**
3439 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3440 * @rq: the clone request to be cleaned up
3441 *
3442 * Description:
3443 *     Free all bios in @rq for a cloned request.
3444 */
3445void blk_rq_unprep_clone(struct request *rq)
3446{
3447        struct bio *bio;
3448
3449        while ((bio = rq->bio) != NULL) {
3450                rq->bio = bio->bi_next;
3451
3452                bio_put(bio);
3453        }
3454}
3455EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3456
3457/*
3458 * Copy attributes of the original request to the clone request.
3459 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3460 */
3461static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3462{
3463        dst->cpu = src->cpu;
3464        dst->__sector = blk_rq_pos(src);
3465        dst->__data_len = blk_rq_bytes(src);
3466        if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
3467                dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
3468                dst->special_vec = src->special_vec;
3469        }
3470        dst->nr_phys_segments = src->nr_phys_segments;
3471        dst->ioprio = src->ioprio;
3472        dst->extra_len = src->extra_len;
3473}
3474
3475/**
3476 * blk_rq_prep_clone - Helper function to setup clone request
3477 * @rq: the request to be setup
3478 * @rq_src: original request to be cloned
3479 * @bs: bio_set that bios for clone are allocated from
3480 * @gfp_mask: memory allocation mask for bio
3481 * @bio_ctr: setup function to be called for each clone bio.
3482 *           Returns %0 for success, non %0 for failure.
3483 * @data: private data to be passed to @bio_ctr
3484 *
3485 * Description:
3486 *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3487 *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3488 *     are not copied, and copying such parts is the caller's responsibility.
3489 *     Also, pages which the original bios are pointing to are not copied
3490 *     and the cloned bios just point same pages.
3491 *     So cloned bios must be completed before original bios, which means
3492 *     the caller must complete @rq before @rq_src.
3493 */
3494int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3495                      struct bio_set *bs, gfp_t gfp_mask,
3496                      int (*bio_ctr)(struct bio *, struct bio *, void *),
3497                      void *data)
3498{
3499        struct bio *bio, *bio_src;
3500
3501        if (!bs)
3502                bs = &fs_bio_set;
3503
3504        __rq_for_each_bio(bio_src, rq_src) {
3505                bio = bio_clone_fast(bio_src, gfp_mask, bs);
3506                if (!bio)
3507                        goto free_and_out;
3508
3509                if (bio_ctr && bio_ctr(bio, bio_src, data))
3510                        goto free_and_out;
3511
3512                if (rq->bio) {
3513                        rq->biotail->bi_next = bio;
3514                        rq->biotail = bio;
3515                } else
3516                        rq->bio = rq->biotail = bio;
3517        }
3518
3519        __blk_rq_prep_clone(rq, rq_src);
3520
3521        return 0;
3522
3523free_and_out:
3524        if (bio)
3525                bio_put(bio);
3526        blk_rq_unprep_clone(rq);
3527
3528        return -ENOMEM;
3529}
3530EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3531
3532int kblockd_schedule_work(struct work_struct *work)
3533{
3534        return queue_work(kblockd_workqueue, work);
3535}
3536EXPORT_SYMBOL(kblockd_schedule_work);
3537
3538int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3539{
3540        return queue_work_on(cpu, kblockd_workqueue, work);
3541}
3542EXPORT_SYMBOL(kblockd_schedule_work_on);
3543
3544int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3545                                unsigned long delay)
3546{
3547        return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3548}
3549EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3550
3551/**
3552 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3553 * @plug:       The &struct blk_plug that needs to be initialized
3554 *
3555 * Description:
3556 *   Tracking blk_plug inside the task_struct will help with auto-flushing the
3557 *   pending I/O should the task end up blocking between blk_start_plug() and
3558 *   blk_finish_plug(). This is important from a performance perspective, but
3559 *   also ensures that we don't deadlock. For instance, if the task is blocking
3560 *   for a memory allocation, memory reclaim could end up wanting to free a
3561 *   page belonging to that request that is currently residing in our private
3562 *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
3563 *   this kind of deadlock.
3564 */
3565void blk_start_plug(struct blk_plug *plug)
3566{
3567        struct task_struct *tsk = current;
3568
3569        /*
3570         * If this is a nested plug, don't actually assign it.
3571         */
3572        if (tsk->plug)
3573                return;
3574
3575        INIT_LIST_HEAD(&plug->list);
3576        INIT_LIST_HEAD(&plug->mq_list);
3577        INIT_LIST_HEAD(&plug->cb_list);
3578        /*
3579         * Store ordering should not be needed here, since a potential
3580         * preempt will imply a full memory barrier
3581         */
3582        tsk->plug = plug;
3583}
3584EXPORT_SYMBOL(blk_start_plug);
3585
3586static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3587{
3588        struct request *rqa = container_of(a, struct request, queuelist);
3589        struct request *rqb = container_of(b, struct request, queuelist);
3590
3591        return !(rqa->q < rqb->q ||
3592                (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3593}
3594
3595/*
3596 * If 'from_schedule' is true, then postpone the dispatch of requests
3597 * until a safe kblockd context. We due this to avoid accidental big
3598 * additional stack usage in driver dispatch, in places where the originally
3599 * plugger did not intend it.
3600 */
3601static void queue_unplugged(struct request_queue *q, unsigned int depth,
3602                            bool from_schedule)
3603        __releases(q->queue_lock)
3604{
3605        lockdep_assert_held(q->queue_lock);
3606
3607        trace_block_unplug(q, depth, !from_schedule);
3608
3609        if (from_schedule)
3610                blk_run_queue_async(q);
3611        else
3612                __blk_run_queue(q);
3613        spin_unlock_irq(q->queue_lock);
3614}
3615
3616static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3617{
3618        LIST_HEAD(callbacks);
3619
3620        while (!list_empty(&plug->cb_list)) {
3621                list_splice_init(&plug->cb_list, &callbacks);
3622
3623                while (!list_empty(&callbacks)) {
3624                        struct blk_plug_cb *cb = list_first_entry(&callbacks,
3625                                                          struct blk_plug_cb,
3626                                                          list);
3627                        list_del(&cb->list);
3628                        cb->callback(cb, from_schedule);
3629                }
3630        }
3631}
3632
3633struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3634                                      int size)
3635{
3636        struct blk_plug *plug = current->plug;
3637        struct blk_plug_cb *cb;
3638
3639        if (!plug)
3640                return NULL;
3641
3642        list_for_each_entry(cb, &plug->cb_list, list)
3643                if (cb->callback == unplug && cb->data == data)
3644                        return cb;
3645
3646        /* Not currently on the callback list */
3647        BUG_ON(size < sizeof(*cb));
3648        cb = kzalloc(size, GFP_ATOMIC);
3649        if (cb) {
3650                cb->data = data;
3651                cb->callback = unplug;
3652                list_add(&cb->list, &plug->cb_list);
3653        }
3654        return cb;
3655}
3656EXPORT_SYMBOL(blk_check_plugged);
3657
3658void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3659{
3660        struct request_queue *q;
3661        struct request *rq;
3662        LIST_HEAD(list);
3663        unsigned int depth;
3664
3665        flush_plug_callbacks(plug, from_schedule);
3666
3667        if (!list_empty(&plug->mq_list))
3668                blk_mq_flush_plug_list(plug, from_schedule);
3669
3670        if (list_empty(&plug->list))
3671                return;
3672
3673        list_splice_init(&plug->list, &list);
3674
3675        list_sort(NULL, &list, plug_rq_cmp);
3676
3677        q = NULL;
3678        depth = 0;
3679
3680        while (!list_empty(&list)) {
3681                rq = list_entry_rq(list.next);
3682                list_del_init(&rq->queuelist);
3683                BUG_ON(!rq->q);
3684                if (rq->q != q) {
3685                        /*
3686                         * This drops the queue lock
3687                         */
3688                        if (q)
3689                                queue_unplugged(q, depth, from_schedule);
3690                        q = rq->q;
3691                        depth = 0;
3692                        spin_lock_irq(q->queue_lock);
3693                }
3694
3695                /*
3696                 * Short-circuit if @q is dead
3697                 */
3698                if (unlikely(blk_queue_dying(q))) {
3699                        __blk_end_request_all(rq, BLK_STS_IOERR);
3700                        continue;
3701                }
3702
3703                /*
3704                 * rq is already accounted, so use raw insert
3705                 */
3706                if (op_is_flush(rq->cmd_flags))
3707                        __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3708                else
3709                        __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3710
3711                depth++;
3712        }
3713
3714        /*
3715         * This drops the queue lock
3716         */
3717        if (q)
3718                queue_unplugged(q, depth, from_schedule);
3719}
3720
3721void blk_finish_plug(struct blk_plug *plug)
3722{
3723        if (plug != current->plug)
3724                return;
3725        blk_flush_plug_list(plug, false);
3726
3727        current->plug = NULL;
3728}
3729EXPORT_SYMBOL(blk_finish_plug);
3730
3731int __init blk_dev_init(void)
3732{
3733        BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3734        BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3735                        FIELD_SIZEOF(struct request, cmd_flags));
3736        BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3737                        FIELD_SIZEOF(struct bio, bi_opf));
3738
3739        /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3740        kblockd_workqueue = alloc_workqueue("kblockd",
3741                                            WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3742        if (!kblockd_workqueue)
3743                panic("Failed to create kblockd\n");
3744
3745        request_cachep = kmem_cache_create("blkdev_requests",
3746                        sizeof(struct request), 0, SLAB_PANIC, NULL);
3747
3748        blk_requestq_cachep = kmem_cache_create("request_queue",
3749                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3750
3751#ifdef CONFIG_DEBUG_FS
3752        blk_debugfs_root = debugfs_create_dir("block", NULL);
3753#endif
3754
3755        return 0;
3756}
3757