linux/block/blk-core.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 1991, 1992 Linus Torvalds
   3 * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
   4 * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
   5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
   6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
   7 *      -  July2000
   8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
   9 */
  10
  11/*
  12 * This handles all read/write requests to block devices
  13 */
  14#include <linux/kernel.h>
  15#include <linux/module.h>
  16#include <linux/backing-dev.h>
  17#include <linux/bio.h>
  18#include <linux/blkdev.h>
  19#include <linux/highmem.h>
  20#include <linux/mm.h>
  21#include <linux/kernel_stat.h>
  22#include <linux/string.h>
  23#include <linux/init.h>
  24#include <linux/completion.h>
  25#include <linux/slab.h>
  26#include <linux/swap.h>
  27#include <linux/writeback.h>
  28#include <linux/task_io_accounting_ops.h>
  29#include <linux/fault-inject.h>
  30#include <linux/list_sort.h>
  31#include <linux/delay.h>
  32#include <linux/ratelimit.h>
  33#include <linux/pm_runtime.h>
  34
  35#define CREATE_TRACE_POINTS
  36#include <trace/events/block.h>
  37
  38#include "blk.h"
  39#include "blk-cgroup.h"
  40
  41EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
  42EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
  43EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
  44EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
  45
  46DEFINE_IDA(blk_queue_ida);
  47
  48/*
  49 * For the allocated request tables
  50 */
  51static struct kmem_cache *request_cachep;
  52
  53/*
  54 * For queue allocation
  55 */
  56struct kmem_cache *blk_requestq_cachep;
  57
  58/*
  59 * Controlling structure to kblockd
  60 */
  61static struct workqueue_struct *kblockd_workqueue;
  62
  63static void drive_stat_acct(struct request *rq, int new_io)
  64{
  65        struct hd_struct *part;
  66        int rw = rq_data_dir(rq);
  67        int cpu;
  68
  69        if (!blk_do_io_stat(rq))
  70                return;
  71
  72        cpu = part_stat_lock();
  73
  74        if (!new_io) {
  75                part = rq->part;
  76                part_stat_inc(cpu, part, merges[rw]);
  77        } else {
  78                part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
  79                if (!hd_struct_try_get(part)) {
  80                        /*
  81                         * The partition is already being removed,
  82                         * the request will be accounted on the disk only
  83                         *
  84                         * We take a reference on disk->part0 although that
  85                         * partition will never be deleted, so we can treat
  86                         * it as any other partition.
  87                         */
  88                        part = &rq->rq_disk->part0;
  89                        hd_struct_get(part);
  90                }
  91                part_round_stats(cpu, part);
  92                part_inc_in_flight(part, rw);
  93                rq->part = part;
  94        }
  95
  96        part_stat_unlock();
  97}
  98
  99void blk_queue_congestion_threshold(struct request_queue *q)
 100{
 101        int nr;
 102
 103        nr = q->nr_requests - (q->nr_requests / 8) + 1;
 104        if (nr > q->nr_requests)
 105                nr = q->nr_requests;
 106        q->nr_congestion_on = nr;
 107
 108        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
 109        if (nr < 1)
 110                nr = 1;
 111        q->nr_congestion_off = nr;
 112}
 113
 114/**
 115 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
 116 * @bdev:       device
 117 *
 118 * Locates the passed device's request queue and returns the address of its
 119 * backing_dev_info
 120 *
 121 * Will return NULL if the request queue cannot be located.
 122 */
 123struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
 124{
 125        struct backing_dev_info *ret = NULL;
 126        struct request_queue *q = bdev_get_queue(bdev);
 127
 128        if (q)
 129                ret = &q->backing_dev_info;
 130        return ret;
 131}
 132EXPORT_SYMBOL(blk_get_backing_dev_info);
 133
 134void blk_rq_init(struct request_queue *q, struct request *rq)
 135{
 136        memset(rq, 0, sizeof(*rq));
 137
 138        INIT_LIST_HEAD(&rq->queuelist);
 139        INIT_LIST_HEAD(&rq->timeout_list);
 140        rq->cpu = -1;
 141        rq->q = q;
 142        rq->__sector = (sector_t) -1;
 143        INIT_HLIST_NODE(&rq->hash);
 144        RB_CLEAR_NODE(&rq->rb_node);
 145        rq->cmd = rq->__cmd;
 146        rq->cmd_len = BLK_MAX_CDB;
 147        rq->tag = -1;
 148        rq->ref_count = 1;
 149        rq->start_time = jiffies;
 150        set_start_time_ns(rq);
 151        rq->part = NULL;
 152}
 153EXPORT_SYMBOL(blk_rq_init);
 154
 155static void req_bio_endio(struct request *rq, struct bio *bio,
 156                          unsigned int nbytes, int error)
 157{
 158        if (error)
 159                clear_bit(BIO_UPTODATE, &bio->bi_flags);
 160        else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 161                error = -EIO;
 162
 163        if (unlikely(rq->cmd_flags & REQ_QUIET))
 164                set_bit(BIO_QUIET, &bio->bi_flags);
 165
 166        bio_advance(bio, nbytes);
 167
 168        /* don't actually finish bio if it's part of flush sequence */
 169        if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
 170                bio_endio(bio, error);
 171}
 172
 173void blk_dump_rq_flags(struct request *rq, char *msg)
 174{
 175        int bit;
 176
 177        printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
 178                rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
 179                rq->cmd_flags);
 180
 181        printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
 182               (unsigned long long)blk_rq_pos(rq),
 183               blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
 184        printk(KERN_INFO "  bio %p, biotail %p, buffer %p, len %u\n",
 185               rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
 186
 187        if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 188                printk(KERN_INFO "  cdb: ");
 189                for (bit = 0; bit < BLK_MAX_CDB; bit++)
 190                        printk("%02x ", rq->cmd[bit]);
 191                printk("\n");
 192        }
 193}
 194EXPORT_SYMBOL(blk_dump_rq_flags);
 195
 196static void blk_delay_work(struct work_struct *work)
 197{
 198        struct request_queue *q;
 199
 200        q = container_of(work, struct request_queue, delay_work.work);
 201        spin_lock_irq(q->queue_lock);
 202        __blk_run_queue(q);
 203        spin_unlock_irq(q->queue_lock);
 204}
 205
 206/**
 207 * blk_delay_queue - restart queueing after defined interval
 208 * @q:          The &struct request_queue in question
 209 * @msecs:      Delay in msecs
 210 *
 211 * Description:
 212 *   Sometimes queueing needs to be postponed for a little while, to allow
 213 *   resources to come back. This function will make sure that queueing is
 214 *   restarted around the specified time. Queue lock must be held.
 215 */
 216void blk_delay_queue(struct request_queue *q, unsigned long msecs)
 217{
 218        if (likely(!blk_queue_dead(q)))
 219                queue_delayed_work(kblockd_workqueue, &q->delay_work,
 220                                   msecs_to_jiffies(msecs));
 221}
 222EXPORT_SYMBOL(blk_delay_queue);
 223
 224/**
 225 * blk_start_queue - restart a previously stopped queue
 226 * @q:    The &struct request_queue in question
 227 *
 228 * Description:
 229 *   blk_start_queue() will clear the stop flag on the queue, and call
 230 *   the request_fn for the queue if it was in a stopped state when
 231 *   entered. Also see blk_stop_queue(). Queue lock must be held.
 232 **/
 233void blk_start_queue(struct request_queue *q)
 234{
 235        WARN_ON(!irqs_disabled());
 236
 237        queue_flag_clear(QUEUE_FLAG_STOPPED, q);
 238        __blk_run_queue(q);
 239}
 240EXPORT_SYMBOL(blk_start_queue);
 241
 242/**
 243 * blk_stop_queue - stop a queue
 244 * @q:    The &struct request_queue in question
 245 *
 246 * Description:
 247 *   The Linux block layer assumes that a block driver will consume all
 248 *   entries on the request queue when the request_fn strategy is called.
 249 *   Often this will not happen, because of hardware limitations (queue
 250 *   depth settings). If a device driver gets a 'queue full' response,
 251 *   or if it simply chooses not to queue more I/O at one point, it can
 252 *   call this function to prevent the request_fn from being called until
 253 *   the driver has signalled it's ready to go again. This happens by calling
 254 *   blk_start_queue() to restart queue operations. Queue lock must be held.
 255 **/
 256void blk_stop_queue(struct request_queue *q)
 257{
 258        cancel_delayed_work(&q->delay_work);
 259        queue_flag_set(QUEUE_FLAG_STOPPED, q);
 260}
 261EXPORT_SYMBOL(blk_stop_queue);
 262
 263/**
 264 * blk_sync_queue - cancel any pending callbacks on a queue
 265 * @q: the queue
 266 *
 267 * Description:
 268 *     The block layer may perform asynchronous callback activity
 269 *     on a queue, such as calling the unplug function after a timeout.
 270 *     A block device may call blk_sync_queue to ensure that any
 271 *     such activity is cancelled, thus allowing it to release resources
 272 *     that the callbacks might use. The caller must already have made sure
 273 *     that its ->make_request_fn will not re-add plugging prior to calling
 274 *     this function.
 275 *
 276 *     This function does not cancel any asynchronous activity arising
 277 *     out of elevator or throttling code. That would require elevaotor_exit()
 278 *     and blkcg_exit_queue() to be called with queue lock initialized.
 279 *
 280 */
 281void blk_sync_queue(struct request_queue *q)
 282{
 283        del_timer_sync(&q->timeout);
 284        cancel_delayed_work_sync(&q->delay_work);
 285}
 286EXPORT_SYMBOL(blk_sync_queue);
 287
 288/**
 289 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
 290 * @q:  The queue to run
 291 *
 292 * Description:
 293 *    Invoke request handling on a queue if there are any pending requests.
 294 *    May be used to restart request handling after a request has completed.
 295 *    This variant runs the queue whether or not the queue has been
 296 *    stopped. Must be called with the queue lock held and interrupts
 297 *    disabled. See also @blk_run_queue.
 298 */
 299inline void __blk_run_queue_uncond(struct request_queue *q)
 300{
 301        if (unlikely(blk_queue_dead(q)))
 302                return;
 303
 304        /*
 305         * Some request_fn implementations, e.g. scsi_request_fn(), unlock
 306         * the queue lock internally. As a result multiple threads may be
 307         * running such a request function concurrently. Keep track of the
 308         * number of active request_fn invocations such that blk_drain_queue()
 309         * can wait until all these request_fn calls have finished.
 310         */
 311        q->request_fn_active++;
 312        q->request_fn(q);
 313        q->request_fn_active--;
 314}
 315
 316/**
 317 * __blk_run_queue - run a single device queue
 318 * @q:  The queue to run
 319 *
 320 * Description:
 321 *    See @blk_run_queue. This variant must be called with the queue lock
 322 *    held and interrupts disabled.
 323 */
 324void __blk_run_queue(struct request_queue *q)
 325{
 326        if (unlikely(blk_queue_stopped(q)))
 327                return;
 328
 329        __blk_run_queue_uncond(q);
 330}
 331EXPORT_SYMBOL(__blk_run_queue);
 332
 333/**
 334 * blk_run_queue_async - run a single device queue in workqueue context
 335 * @q:  The queue to run
 336 *
 337 * Description:
 338 *    Tells kblockd to perform the equivalent of @blk_run_queue on behalf
 339 *    of us. The caller must hold the queue lock.
 340 */
 341void blk_run_queue_async(struct request_queue *q)
 342{
 343        if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
 344                mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
 345}
 346EXPORT_SYMBOL(blk_run_queue_async);
 347
 348/**
 349 * blk_run_queue - run a single device queue
 350 * @q: The queue to run
 351 *
 352 * Description:
 353 *    Invoke request handling on this queue, if it has pending work to do.
 354 *    May be used to restart queueing when a request has completed.
 355 */
 356void blk_run_queue(struct request_queue *q)
 357{
 358        unsigned long flags;
 359
 360        spin_lock_irqsave(q->queue_lock, flags);
 361        __blk_run_queue(q);
 362        spin_unlock_irqrestore(q->queue_lock, flags);
 363}
 364EXPORT_SYMBOL(blk_run_queue);
 365
 366void blk_put_queue(struct request_queue *q)
 367{
 368        kobject_put(&q->kobj);
 369}
 370EXPORT_SYMBOL(blk_put_queue);
 371
 372/**
 373 * __blk_drain_queue - drain requests from request_queue
 374 * @q: queue to drain
 375 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
 376 *
 377 * Drain requests from @q.  If @drain_all is set, all requests are drained.
 378 * If not, only ELVPRIV requests are drained.  The caller is responsible
 379 * for ensuring that no new requests which need to be drained are queued.
 380 */
 381static void __blk_drain_queue(struct request_queue *q, bool drain_all)
 382        __releases(q->queue_lock)
 383        __acquires(q->queue_lock)
 384{
 385        int i;
 386
 387        lockdep_assert_held(q->queue_lock);
 388
 389        while (true) {
 390                bool drain = false;
 391
 392                /*
 393                 * The caller might be trying to drain @q before its
 394                 * elevator is initialized.
 395                 */
 396                if (q->elevator)
 397                        elv_drain_elevator(q);
 398
 399                blkcg_drain_queue(q);
 400
 401                /*
 402                 * This function might be called on a queue which failed
 403                 * driver init after queue creation or is not yet fully
 404                 * active yet.  Some drivers (e.g. fd and loop) get unhappy
 405                 * in such cases.  Kick queue iff dispatch queue has
 406                 * something on it and @q has request_fn set.
 407                 */
 408                if (!list_empty(&q->queue_head) && q->request_fn)
 409                        __blk_run_queue(q);
 410
 411                drain |= q->nr_rqs_elvpriv;
 412                drain |= q->request_fn_active;
 413
 414                /*
 415                 * Unfortunately, requests are queued at and tracked from
 416                 * multiple places and there's no single counter which can
 417                 * be drained.  Check all the queues and counters.
 418                 */
 419                if (drain_all) {
 420                        drain |= !list_empty(&q->queue_head);
 421                        for (i = 0; i < 2; i++) {
 422                                drain |= q->nr_rqs[i];
 423                                drain |= q->in_flight[i];
 424                                drain |= !list_empty(&q->flush_queue[i]);
 425                        }
 426                }
 427
 428                if (!drain)
 429                        break;
 430
 431                spin_unlock_irq(q->queue_lock);
 432
 433                msleep(10);
 434
 435                spin_lock_irq(q->queue_lock);
 436        }
 437
 438        /*
 439         * With queue marked dead, any woken up waiter will fail the
 440         * allocation path, so the wakeup chaining is lost and we're
 441         * left with hung waiters. We need to wake up those waiters.
 442         */
 443        if (q->request_fn) {
 444                struct request_list *rl;
 445
 446                blk_queue_for_each_rl(rl, q)
 447                        for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
 448                                wake_up_all(&rl->wait[i]);
 449        }
 450}
 451
 452/**
 453 * blk_queue_bypass_start - enter queue bypass mode
 454 * @q: queue of interest
 455 *
 456 * In bypass mode, only the dispatch FIFO queue of @q is used.  This
 457 * function makes @q enter bypass mode and drains all requests which were
 458 * throttled or issued before.  On return, it's guaranteed that no request
 459 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
 460 * inside queue or RCU read lock.
 461 */
 462void blk_queue_bypass_start(struct request_queue *q)
 463{
 464        bool drain;
 465
 466        spin_lock_irq(q->queue_lock);
 467        drain = !q->bypass_depth++;
 468        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 469        spin_unlock_irq(q->queue_lock);
 470
 471        if (drain) {
 472                spin_lock_irq(q->queue_lock);
 473                __blk_drain_queue(q, false);
 474                spin_unlock_irq(q->queue_lock);
 475
 476                /* ensure blk_queue_bypass() is %true inside RCU read lock */
 477                synchronize_rcu();
 478        }
 479}
 480EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
 481
 482/**
 483 * blk_queue_bypass_end - leave queue bypass mode
 484 * @q: queue of interest
 485 *
 486 * Leave bypass mode and restore the normal queueing behavior.
 487 */
 488void blk_queue_bypass_end(struct request_queue *q)
 489{
 490        spin_lock_irq(q->queue_lock);
 491        if (!--q->bypass_depth)
 492                queue_flag_clear(QUEUE_FLAG_BYPASS, q);
 493        WARN_ON_ONCE(q->bypass_depth < 0);
 494        spin_unlock_irq(q->queue_lock);
 495}
 496EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
 497
 498/**
 499 * blk_cleanup_queue - shutdown a request queue
 500 * @q: request queue to shutdown
 501 *
 502 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
 503 * put it.  All future requests will be failed immediately with -ENODEV.
 504 */
 505void blk_cleanup_queue(struct request_queue *q)
 506{
 507        spinlock_t *lock = q->queue_lock;
 508
 509        /* mark @q DYING, no new request or merges will be allowed afterwards */
 510        mutex_lock(&q->sysfs_lock);
 511        queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
 512        spin_lock_irq(lock);
 513
 514        /*
 515         * A dying queue is permanently in bypass mode till released.  Note
 516         * that, unlike blk_queue_bypass_start(), we aren't performing
 517         * synchronize_rcu() after entering bypass mode to avoid the delay
 518         * as some drivers create and destroy a lot of queues while
 519         * probing.  This is still safe because blk_release_queue() will be
 520         * called only after the queue refcnt drops to zero and nothing,
 521         * RCU or not, would be traversing the queue by then.
 522         */
 523        q->bypass_depth++;
 524        queue_flag_set(QUEUE_FLAG_BYPASS, q);
 525
 526        queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 527        queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 528        queue_flag_set(QUEUE_FLAG_DYING, q);
 529        spin_unlock_irq(lock);
 530        mutex_unlock(&q->sysfs_lock);
 531
 532        /*
 533         * Drain all requests queued before DYING marking. Set DEAD flag to
 534         * prevent that q->request_fn() gets invoked after draining finished.
 535         */
 536        spin_lock_irq(lock);
 537        __blk_drain_queue(q, true);
 538        queue_flag_set(QUEUE_FLAG_DEAD, q);
 539        spin_unlock_irq(lock);
 540
 541        /* @q won't process any more request, flush async actions */
 542        del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
 543        blk_sync_queue(q);
 544
 545        spin_lock_irq(lock);
 546        if (q->queue_lock != &q->__queue_lock)
 547                q->queue_lock = &q->__queue_lock;
 548        spin_unlock_irq(lock);
 549
 550        /* @q is and will stay empty, shutdown and put */
 551        blk_put_queue(q);
 552}
 553EXPORT_SYMBOL(blk_cleanup_queue);
 554
 555int blk_init_rl(struct request_list *rl, struct request_queue *q,
 556                gfp_t gfp_mask)
 557{
 558        if (unlikely(rl->rq_pool))
 559                return 0;
 560
 561        rl->q = q;
 562        rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
 563        rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
 564        init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
 565        init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
 566
 567        rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
 568                                          mempool_free_slab, request_cachep,
 569                                          gfp_mask, q->node);
 570        if (!rl->rq_pool)
 571                return -ENOMEM;
 572
 573        return 0;
 574}
 575
 576void blk_exit_rl(struct request_list *rl)
 577{
 578        if (rl->rq_pool)
 579                mempool_destroy(rl->rq_pool);
 580}
 581
 582struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 583{
 584        return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
 585}
 586EXPORT_SYMBOL(blk_alloc_queue);
 587
 588struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 589{
 590        struct request_queue *q;
 591        int err;
 592
 593        q = kmem_cache_alloc_node(blk_requestq_cachep,
 594                                gfp_mask | __GFP_ZERO, node_id);
 595        if (!q)
 596                return NULL;
 597
 598        q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
 599        if (q->id < 0)
 600                goto fail_q;
 601
 602        q->backing_dev_info.ra_pages =
 603                        (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 604        q->backing_dev_info.state = 0;
 605        q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
 606        q->backing_dev_info.name = "block";
 607        q->node = node_id;
 608
 609        err = bdi_init(&q->backing_dev_info);
 610        if (err)
 611                goto fail_id;
 612
 613        setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
 614                    laptop_mode_timer_fn, (unsigned long) q);
 615        setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
 616        INIT_LIST_HEAD(&q->queue_head);
 617        INIT_LIST_HEAD(&q->timeout_list);
 618        INIT_LIST_HEAD(&q->icq_list);
 619#ifdef CONFIG_BLK_CGROUP
 620        INIT_LIST_HEAD(&q->blkg_list);
 621#endif
 622        INIT_LIST_HEAD(&q->flush_queue[0]);
 623        INIT_LIST_HEAD(&q->flush_queue[1]);
 624        INIT_LIST_HEAD(&q->flush_data_in_flight);
 625        INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
 626
 627        kobject_init(&q->kobj, &blk_queue_ktype);
 628
 629        mutex_init(&q->sysfs_lock);
 630        spin_lock_init(&q->__queue_lock);
 631
 632        /*
 633         * By default initialize queue_lock to internal lock and driver can
 634         * override it later if need be.
 635         */
 636        q->queue_lock = &q->__queue_lock;
 637
 638        /*
 639         * A queue starts its life with bypass turned on to avoid
 640         * unnecessary bypass on/off overhead and nasty surprises during
 641         * init.  The initial bypass will be finished when the queue is
 642         * registered by blk_register_queue().
 643         */
 644        q->bypass_depth = 1;
 645        __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
 646
 647        if (blkcg_init_queue(q))
 648                goto fail_id;
 649
 650        return q;
 651
 652fail_id:
 653        ida_simple_remove(&blk_queue_ida, q->id);
 654fail_q:
 655        kmem_cache_free(blk_requestq_cachep, q);
 656        return NULL;
 657}
 658EXPORT_SYMBOL(blk_alloc_queue_node);
 659
 660/**
 661 * blk_init_queue  - prepare a request queue for use with a block device
 662 * @rfn:  The function to be called to process requests that have been
 663 *        placed on the queue.
 664 * @lock: Request queue spin lock
 665 *
 666 * Description:
 667 *    If a block device wishes to use the standard request handling procedures,
 668 *    which sorts requests and coalesces adjacent requests, then it must
 669 *    call blk_init_queue().  The function @rfn will be called when there
 670 *    are requests on the queue that need to be processed.  If the device
 671 *    supports plugging, then @rfn may not be called immediately when requests
 672 *    are available on the queue, but may be called at some time later instead.
 673 *    Plugged queues are generally unplugged when a buffer belonging to one
 674 *    of the requests on the queue is needed, or due to memory pressure.
 675 *
 676 *    @rfn is not required, or even expected, to remove all requests off the
 677 *    queue, but only as many as it can handle at a time.  If it does leave
 678 *    requests on the queue, it is responsible for arranging that the requests
 679 *    get dealt with eventually.
 680 *
 681 *    The queue spin lock must be held while manipulating the requests on the
 682 *    request queue; this lock will be taken also from interrupt context, so irq
 683 *    disabling is needed for it.
 684 *
 685 *    Function returns a pointer to the initialized request queue, or %NULL if
 686 *    it didn't succeed.
 687 *
 688 * Note:
 689 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
 690 *    when the block device is deactivated (such as at module unload).
 691 **/
 692
 693struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
 694{
 695        return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
 696}
 697EXPORT_SYMBOL(blk_init_queue);
 698
 699struct request_queue *
 700blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
 701{
 702        struct request_queue *uninit_q, *q;
 703
 704        uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
 705        if (!uninit_q)
 706                return NULL;
 707
 708        q = blk_init_allocated_queue(uninit_q, rfn, lock);
 709        if (!q)
 710                blk_cleanup_queue(uninit_q);
 711
 712        return q;
 713}
 714EXPORT_SYMBOL(blk_init_queue_node);
 715
 716struct request_queue *
 717blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
 718                         spinlock_t *lock)
 719{
 720        if (!q)
 721                return NULL;
 722
 723        if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
 724                return NULL;
 725
 726        q->request_fn           = rfn;
 727        q->prep_rq_fn           = NULL;
 728        q->unprep_rq_fn         = NULL;
 729        q->queue_flags          |= QUEUE_FLAG_DEFAULT;
 730
 731        /* Override internal queue lock with supplied lock pointer */
 732        if (lock)
 733                q->queue_lock           = lock;
 734
 735        /*
 736         * This also sets hw/phys segments, boundary and size
 737         */
 738        blk_queue_make_request(q, blk_queue_bio);
 739
 740        q->sg_reserved_size = INT_MAX;
 741
 742        /* init elevator */
 743        if (elevator_init(q, NULL))
 744                return NULL;
 745        return q;
 746}
 747EXPORT_SYMBOL(blk_init_allocated_queue);
 748
 749bool blk_get_queue(struct request_queue *q)
 750{
 751        if (likely(!blk_queue_dying(q))) {
 752                __blk_get_queue(q);
 753                return true;
 754        }
 755
 756        return false;
 757}
 758EXPORT_SYMBOL(blk_get_queue);
 759
 760static inline void blk_free_request(struct request_list *rl, struct request *rq)
 761{
 762        if (rq->cmd_flags & REQ_ELVPRIV) {
 763                elv_put_request(rl->q, rq);
 764                if (rq->elv.icq)
 765                        put_io_context(rq->elv.icq->ioc);
 766        }
 767
 768        mempool_free(rq, rl->rq_pool);
 769}
 770
 771/*
 772 * ioc_batching returns true if the ioc is a valid batching request and
 773 * should be given priority access to a request.
 774 */
 775static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
 776{
 777        if (!ioc)
 778                return 0;
 779
 780        /*
 781         * Make sure the process is able to allocate at least 1 request
 782         * even if the batch times out, otherwise we could theoretically
 783         * lose wakeups.
 784         */
 785        return ioc->nr_batch_requests == q->nr_batching ||
 786                (ioc->nr_batch_requests > 0
 787                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
 788}
 789
 790/*
 791 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
 792 * will cause the process to be a "batcher" on all queues in the system. This
 793 * is the behaviour we want though - once it gets a wakeup it should be given
 794 * a nice run.
 795 */
 796static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
 797{
 798        if (!ioc || ioc_batching(q, ioc))
 799                return;
 800
 801        ioc->nr_batch_requests = q->nr_batching;
 802        ioc->last_waited = jiffies;
 803}
 804
 805static void __freed_request(struct request_list *rl, int sync)
 806{
 807        struct request_queue *q = rl->q;
 808
 809        /*
 810         * bdi isn't aware of blkcg yet.  As all async IOs end up root
 811         * blkcg anyway, just use root blkcg state.
 812         */
 813        if (rl == &q->root_rl &&
 814            rl->count[sync] < queue_congestion_off_threshold(q))
 815                blk_clear_queue_congested(q, sync);
 816
 817        if (rl->count[sync] + 1 <= q->nr_requests) {
 818                if (waitqueue_active(&rl->wait[sync]))
 819                        wake_up(&rl->wait[sync]);
 820
 821                blk_clear_rl_full(rl, sync);
 822        }
 823}
 824
 825/*
 826 * A request has just been released.  Account for it, update the full and
 827 * congestion status, wake up any waiters.   Called under q->queue_lock.
 828 */
 829static void freed_request(struct request_list *rl, unsigned int flags)
 830{
 831        struct request_queue *q = rl->q;
 832        int sync = rw_is_sync(flags);
 833
 834        q->nr_rqs[sync]--;
 835        rl->count[sync]--;
 836        if (flags & REQ_ELVPRIV)
 837                q->nr_rqs_elvpriv--;
 838
 839        __freed_request(rl, sync);
 840
 841        if (unlikely(rl->starved[sync ^ 1]))
 842                __freed_request(rl, sync ^ 1);
 843}
 844
 845/*
 846 * Determine if elevator data should be initialized when allocating the
 847 * request associated with @bio.
 848 */
 849static bool blk_rq_should_init_elevator(struct bio *bio)
 850{
 851        if (!bio)
 852                return true;
 853
 854        /*
 855         * Flush requests do not use the elevator so skip initialization.
 856         * This allows a request to share the flush and elevator data.
 857         */
 858        if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
 859                return false;
 860
 861        return true;
 862}
 863
 864/**
 865 * rq_ioc - determine io_context for request allocation
 866 * @bio: request being allocated is for this bio (can be %NULL)
 867 *
 868 * Determine io_context to use for request allocation for @bio.  May return
 869 * %NULL if %current->io_context doesn't exist.
 870 */
 871static struct io_context *rq_ioc(struct bio *bio)
 872{
 873#ifdef CONFIG_BLK_CGROUP
 874        if (bio && bio->bi_ioc)
 875                return bio->bi_ioc;
 876#endif
 877        return current->io_context;
 878}
 879
 880/**
 881 * __get_request - get a free request
 882 * @rl: request list to allocate from
 883 * @rw_flags: RW and SYNC flags
 884 * @bio: bio to allocate request for (can be %NULL)
 885 * @gfp_mask: allocation mask
 886 *
 887 * Get a free request from @q.  This function may fail under memory
 888 * pressure or if @q is dead.
 889 *
 890 * Must be callled with @q->queue_lock held and,
 891 * Returns %NULL on failure, with @q->queue_lock held.
 892 * Returns !%NULL on success, with @q->queue_lock *not held*.
 893 */
 894static struct request *__get_request(struct request_list *rl, int rw_flags,
 895                                     struct bio *bio, gfp_t gfp_mask)
 896{
 897        struct request_queue *q = rl->q;
 898        struct request *rq;
 899        struct elevator_type *et = q->elevator->type;
 900        struct io_context *ioc = rq_ioc(bio);
 901        struct io_cq *icq = NULL;
 902        const bool is_sync = rw_is_sync(rw_flags) != 0;
 903        int may_queue;
 904
 905        if (unlikely(blk_queue_dying(q)))
 906                return NULL;
 907
 908        may_queue = elv_may_queue(q, rw_flags);
 909        if (may_queue == ELV_MQUEUE_NO)
 910                goto rq_starved;
 911
 912        if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
 913                if (rl->count[is_sync]+1 >= q->nr_requests) {
 914                        /*
 915                         * The queue will fill after this allocation, so set
 916                         * it as full, and mark this process as "batching".
 917                         * This process will be allowed to complete a batch of
 918                         * requests, others will be blocked.
 919                         */
 920                        if (!blk_rl_full(rl, is_sync)) {
 921                                ioc_set_batching(q, ioc);
 922                                blk_set_rl_full(rl, is_sync);
 923                        } else {
 924                                if (may_queue != ELV_MQUEUE_MUST
 925                                                && !ioc_batching(q, ioc)) {
 926                                        /*
 927                                         * The queue is full and the allocating
 928                                         * process is not a "batcher", and not
 929                                         * exempted by the IO scheduler
 930                                         */
 931                                        return NULL;
 932                                }
 933                        }
 934                }
 935                /*
 936                 * bdi isn't aware of blkcg yet.  As all async IOs end up
 937                 * root blkcg anyway, just use root blkcg state.
 938                 */
 939                if (rl == &q->root_rl)
 940                        blk_set_queue_congested(q, is_sync);
 941        }
 942
 943        /*
 944         * Only allow batching queuers to allocate up to 50% over the defined
 945         * limit of requests, otherwise we could have thousands of requests
 946         * allocated with any setting of ->nr_requests
 947         */
 948        if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
 949                return NULL;
 950
 951        q->nr_rqs[is_sync]++;
 952        rl->count[is_sync]++;
 953        rl->starved[is_sync] = 0;
 954
 955        /*
 956         * Decide whether the new request will be managed by elevator.  If
 957         * so, mark @rw_flags and increment elvpriv.  Non-zero elvpriv will
 958         * prevent the current elevator from being destroyed until the new
 959         * request is freed.  This guarantees icq's won't be destroyed and
 960         * makes creating new ones safe.
 961         *
 962         * Also, lookup icq while holding queue_lock.  If it doesn't exist,
 963         * it will be created after releasing queue_lock.
 964         */
 965        if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
 966                rw_flags |= REQ_ELVPRIV;
 967                q->nr_rqs_elvpriv++;
 968                if (et->icq_cache && ioc)
 969                        icq = ioc_lookup_icq(ioc, q);
 970        }
 971
 972        if (blk_queue_io_stat(q))
 973                rw_flags |= REQ_IO_STAT;
 974        spin_unlock_irq(q->queue_lock);
 975
 976        /* allocate and init request */
 977        rq = mempool_alloc(rl->rq_pool, gfp_mask);
 978        if (!rq)
 979                goto fail_alloc;
 980
 981        blk_rq_init(q, rq);
 982        blk_rq_set_rl(rq, rl);
 983        rq->cmd_flags = rw_flags | REQ_ALLOCED;
 984
 985        /* init elvpriv */
 986        if (rw_flags & REQ_ELVPRIV) {
 987                if (unlikely(et->icq_cache && !icq)) {
 988                        if (ioc)
 989                                icq = ioc_create_icq(ioc, q, gfp_mask);
 990                        if (!icq)
 991                                goto fail_elvpriv;
 992                }
 993
 994                rq->elv.icq = icq;
 995                if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
 996                        goto fail_elvpriv;
 997
 998                /* @rq->elv.icq holds io_context until @rq is freed */
 999                if (icq)
1000                        get_io_context(icq->ioc);
1001        }
1002out:
1003        /*
1004         * ioc may be NULL here, and ioc_batching will be false. That's
1005         * OK, if the queue is under the request limit then requests need
1006         * not count toward the nr_batch_requests limit. There will always
1007         * be some limit enforced by BLK_BATCH_TIME.
1008         */
1009        if (ioc_batching(q, ioc))
1010                ioc->nr_batch_requests--;
1011
1012        trace_block_getrq(q, bio, rw_flags & 1);
1013        return rq;
1014
1015fail_elvpriv:
1016        /*
1017         * elvpriv init failed.  ioc, icq and elvpriv aren't mempool backed
1018         * and may fail indefinitely under memory pressure and thus
1019         * shouldn't stall IO.  Treat this request as !elvpriv.  This will
1020         * disturb iosched and blkcg but weird is bettern than dead.
1021         */
1022        printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1023                           dev_name(q->backing_dev_info.dev));
1024
1025        rq->cmd_flags &= ~REQ_ELVPRIV;
1026        rq->elv.icq = NULL;
1027
1028        spin_lock_irq(q->queue_lock);
1029        q->nr_rqs_elvpriv--;
1030        spin_unlock_irq(q->queue_lock);
1031        goto out;
1032
1033fail_alloc:
1034        /*
1035         * Allocation failed presumably due to memory. Undo anything we
1036         * might have messed up.
1037         *
1038         * Allocating task should really be put onto the front of the wait
1039         * queue, but this is pretty rare.
1040         */
1041        spin_lock_irq(q->queue_lock);
1042        freed_request(rl, rw_flags);
1043
1044        /*
1045         * in the very unlikely event that allocation failed and no
1046         * requests for this direction was pending, mark us starved so that
1047         * freeing of a request in the other direction will notice
1048         * us. another possible fix would be to split the rq mempool into
1049         * READ and WRITE
1050         */
1051rq_starved:
1052        if (unlikely(rl->count[is_sync] == 0))
1053                rl->starved[is_sync] = 1;
1054        return NULL;
1055}
1056
1057/**
1058 * get_request - get a free request
1059 * @q: request_queue to allocate request from
1060 * @rw_flags: RW and SYNC flags
1061 * @bio: bio to allocate request for (can be %NULL)
1062 * @gfp_mask: allocation mask
1063 *
1064 * Get a free request from @q.  If %__GFP_WAIT is set in @gfp_mask, this
1065 * function keeps retrying under memory pressure and fails iff @q is dead.
1066 *
1067 * Must be callled with @q->queue_lock held and,
1068 * Returns %NULL on failure, with @q->queue_lock held.
1069 * Returns !%NULL on success, with @q->queue_lock *not held*.
1070 */
1071static struct request *get_request(struct request_queue *q, int rw_flags,
1072                                   struct bio *bio, gfp_t gfp_mask)
1073{
1074        const bool is_sync = rw_is_sync(rw_flags) != 0;
1075        DEFINE_WAIT(wait);
1076        struct request_list *rl;
1077        struct request *rq;
1078
1079        rl = blk_get_rl(q, bio);        /* transferred to @rq on success */
1080retry:
1081        rq = __get_request(rl, rw_flags, bio, gfp_mask);
1082        if (rq)
1083                return rq;
1084
1085        if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1086                blk_put_rl(rl);
1087                return NULL;
1088        }
1089
1090        /* wait on @rl and retry */
1091        prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1092                                  TASK_UNINTERRUPTIBLE);
1093
1094        trace_block_sleeprq(q, bio, rw_flags & 1);
1095
1096        spin_unlock_irq(q->queue_lock);
1097        io_schedule();
1098
1099        /*
1100         * After sleeping, we become a "batching" process and will be able
1101         * to allocate at least one request, and up to a big batch of them
1102         * for a small period time.  See ioc_batching, ioc_set_batching
1103         */
1104        ioc_set_batching(q, current->io_context);
1105
1106        spin_lock_irq(q->queue_lock);
1107        finish_wait(&rl->wait[is_sync], &wait);
1108
1109        goto retry;
1110}
1111
1112struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1113{
1114        struct request *rq;
1115
1116        BUG_ON(rw != READ && rw != WRITE);
1117
1118        /* create ioc upfront */
1119        create_io_context(gfp_mask, q->node);
1120
1121        spin_lock_irq(q->queue_lock);
1122        rq = get_request(q, rw, NULL, gfp_mask);
1123        if (!rq)
1124                spin_unlock_irq(q->queue_lock);
1125        /* q->queue_lock is unlocked at this point */
1126
1127        return rq;
1128}
1129EXPORT_SYMBOL(blk_get_request);
1130
1131/**
1132 * blk_make_request - given a bio, allocate a corresponding struct request.
1133 * @q: target request queue
1134 * @bio:  The bio describing the memory mappings that will be submitted for IO.
1135 *        It may be a chained-bio properly constructed by block/bio layer.
1136 * @gfp_mask: gfp flags to be used for memory allocation
1137 *
1138 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1139 * type commands. Where the struct request needs to be farther initialized by
1140 * the caller. It is passed a &struct bio, which describes the memory info of
1141 * the I/O transfer.
1142 *
1143 * The caller of blk_make_request must make sure that bi_io_vec
1144 * are set to describe the memory buffers. That bio_data_dir() will return
1145 * the needed direction of the request. (And all bio's in the passed bio-chain
1146 * are properly set accordingly)
1147 *
1148 * If called under none-sleepable conditions, mapped bio buffers must not
1149 * need bouncing, by calling the appropriate masked or flagged allocator,
1150 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1151 * BUG.
1152 *
1153 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1154 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1155 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1156 * completion of a bio that hasn't been submitted yet, thus resulting in a
1157 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1158 * of bio_alloc(), as that avoids the mempool deadlock.
1159 * If possible a big IO should be split into smaller parts when allocation
1160 * fails. Partial allocation should not be an error, or you risk a live-lock.
1161 */
1162struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1163                                 gfp_t gfp_mask)
1164{
1165        struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1166
1167        if (unlikely(!rq))
1168                return ERR_PTR(-ENOMEM);
1169
1170        for_each_bio(bio) {
1171                struct bio *bounce_bio = bio;
1172                int ret;
1173
1174                blk_queue_bounce(q, &bounce_bio);
1175                ret = blk_rq_append_bio(q, rq, bounce_bio);
1176                if (unlikely(ret)) {
1177                        blk_put_request(rq);
1178                        return ERR_PTR(ret);
1179                }
1180        }
1181
1182        return rq;
1183}
1184EXPORT_SYMBOL(blk_make_request);
1185
1186/**
1187 * blk_requeue_request - put a request back on queue
1188 * @q:          request queue where request should be inserted
1189 * @rq:         request to be inserted
1190 *
1191 * Description:
1192 *    Drivers often keep queueing requests until the hardware cannot accept
1193 *    more, when that condition happens we need to put the request back
1194 *    on the queue. Must be called with queue lock held.
1195 */
1196void blk_requeue_request(struct request_queue *q, struct request *rq)
1197{
1198        blk_delete_timer(rq);
1199        blk_clear_rq_complete(rq);
1200        trace_block_rq_requeue(q, rq);
1201
1202        if (blk_rq_tagged(rq))
1203                blk_queue_end_tag(q, rq);
1204
1205        BUG_ON(blk_queued_rq(rq));
1206
1207        elv_requeue_request(q, rq);
1208}
1209EXPORT_SYMBOL(blk_requeue_request);
1210
1211static void add_acct_request(struct request_queue *q, struct request *rq,
1212                             int where)
1213{
1214        drive_stat_acct(rq, 1);
1215        __elv_add_request(q, rq, where);
1216}
1217
1218static void part_round_stats_single(int cpu, struct hd_struct *part,
1219                                    unsigned long now)
1220{
1221        if (now == part->stamp)
1222                return;
1223
1224        if (part_in_flight(part)) {
1225                __part_stat_add(cpu, part, time_in_queue,
1226                                part_in_flight(part) * (now - part->stamp));
1227                __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1228        }
1229        part->stamp = now;
1230}
1231
1232/**
1233 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1234 * @cpu: cpu number for stats access
1235 * @part: target partition
1236 *
1237 * The average IO queue length and utilisation statistics are maintained
1238 * by observing the current state of the queue length and the amount of
1239 * time it has been in this state for.
1240 *
1241 * Normally, that accounting is done on IO completion, but that can result
1242 * in more than a second's worth of IO being accounted for within any one
1243 * second, leading to >100% utilisation.  To deal with that, we call this
1244 * function to do a round-off before returning the results when reading
1245 * /proc/diskstats.  This accounts immediately for all queue usage up to
1246 * the current jiffies and restarts the counters again.
1247 */
1248void part_round_stats(int cpu, struct hd_struct *part)
1249{
1250        unsigned long now = jiffies;
1251
1252        if (part->partno)
1253                part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1254        part_round_stats_single(cpu, part, now);
1255}
1256EXPORT_SYMBOL_GPL(part_round_stats);
1257
1258#ifdef CONFIG_PM_RUNTIME
1259static void blk_pm_put_request(struct request *rq)
1260{
1261        if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1262                pm_runtime_mark_last_busy(rq->q->dev);
1263}
1264#else
1265static inline void blk_pm_put_request(struct request *rq) {}
1266#endif
1267
1268/*
1269 * queue lock must be held
1270 */
1271void __blk_put_request(struct request_queue *q, struct request *req)
1272{
1273        if (unlikely(!q))
1274                return;
1275        if (unlikely(--req->ref_count))
1276                return;
1277
1278        blk_pm_put_request(req);
1279
1280        elv_completed_request(q, req);
1281
1282        /* this is a bio leak */
1283        WARN_ON(req->bio != NULL);
1284
1285        /*
1286         * Request may not have originated from ll_rw_blk. if not,
1287         * it didn't come out of our reserved rq pools
1288         */
1289        if (req->cmd_flags & REQ_ALLOCED) {
1290                unsigned int flags = req->cmd_flags;
1291                struct request_list *rl = blk_rq_rl(req);
1292
1293                BUG_ON(!list_empty(&req->queuelist));
1294                BUG_ON(!hlist_unhashed(&req->hash));
1295
1296                blk_free_request(rl, req);
1297                freed_request(rl, flags);
1298                blk_put_rl(rl);
1299        }
1300}
1301EXPORT_SYMBOL_GPL(__blk_put_request);
1302
1303void blk_put_request(struct request *req)
1304{
1305        unsigned long flags;
1306        struct request_queue *q = req->q;
1307
1308        spin_lock_irqsave(q->queue_lock, flags);
1309        __blk_put_request(q, req);
1310        spin_unlock_irqrestore(q->queue_lock, flags);
1311}
1312EXPORT_SYMBOL(blk_put_request);
1313
1314/**
1315 * blk_add_request_payload - add a payload to a request
1316 * @rq: request to update
1317 * @page: page backing the payload
1318 * @len: length of the payload.
1319 *
1320 * This allows to later add a payload to an already submitted request by
1321 * a block driver.  The driver needs to take care of freeing the payload
1322 * itself.
1323 *
1324 * Note that this is a quite horrible hack and nothing but handling of
1325 * discard requests should ever use it.
1326 */
1327void blk_add_request_payload(struct request *rq, struct page *page,
1328                unsigned int len)
1329{
1330        struct bio *bio = rq->bio;
1331
1332        bio->bi_io_vec->bv_page = page;
1333        bio->bi_io_vec->bv_offset = 0;
1334        bio->bi_io_vec->bv_len = len;
1335
1336        bio->bi_size = len;
1337        bio->bi_vcnt = 1;
1338        bio->bi_phys_segments = 1;
1339
1340        rq->__data_len = rq->resid_len = len;
1341        rq->nr_phys_segments = 1;
1342        rq->buffer = bio_data(bio);
1343}
1344EXPORT_SYMBOL_GPL(blk_add_request_payload);
1345
1346static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1347                                   struct bio *bio)
1348{
1349        const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1350
1351        if (!ll_back_merge_fn(q, req, bio))
1352                return false;
1353
1354        trace_block_bio_backmerge(q, req, bio);
1355
1356        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1357                blk_rq_set_mixed_merge(req);
1358
1359        req->biotail->bi_next = bio;
1360        req->biotail = bio;
1361        req->__data_len += bio->bi_size;
1362        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1363
1364        drive_stat_acct(req, 0);
1365        return true;
1366}
1367
1368static bool bio_attempt_front_merge(struct request_queue *q,
1369                                    struct request *req, struct bio *bio)
1370{
1371        const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1372
1373        if (!ll_front_merge_fn(q, req, bio))
1374                return false;
1375
1376        trace_block_bio_frontmerge(q, req, bio);
1377
1378        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1379                blk_rq_set_mixed_merge(req);
1380
1381        bio->bi_next = req->bio;
1382        req->bio = bio;
1383
1384        /*
1385         * may not be valid. if the low level driver said
1386         * it didn't need a bounce buffer then it better
1387         * not touch req->buffer either...
1388         */
1389        req->buffer = bio_data(bio);
1390        req->__sector = bio->bi_sector;
1391        req->__data_len += bio->bi_size;
1392        req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1393
1394        drive_stat_acct(req, 0);
1395        return true;
1396}
1397
1398/**
1399 * attempt_plug_merge - try to merge with %current's plugged list
1400 * @q: request_queue new bio is being queued at
1401 * @bio: new bio being queued
1402 * @request_count: out parameter for number of traversed plugged requests
1403 *
1404 * Determine whether @bio being queued on @q can be merged with a request
1405 * on %current's plugged list.  Returns %true if merge was successful,
1406 * otherwise %false.
1407 *
1408 * Plugging coalesces IOs from the same issuer for the same purpose without
1409 * going through @q->queue_lock.  As such it's more of an issuing mechanism
1410 * than scheduling, and the request, while may have elvpriv data, is not
1411 * added on the elevator at this point.  In addition, we don't have
1412 * reliable access to the elevator outside queue lock.  Only check basic
1413 * merging parameters without querying the elevator.
1414 */
1415static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1416                               unsigned int *request_count)
1417{
1418        struct blk_plug *plug;
1419        struct request *rq;
1420        bool ret = false;
1421
1422        plug = current->plug;
1423        if (!plug)
1424                goto out;
1425        *request_count = 0;
1426
1427        list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1428                int el_ret;
1429
1430                if (rq->q == q)
1431                        (*request_count)++;
1432
1433                if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1434                        continue;
1435
1436                el_ret = blk_try_merge(rq, bio);
1437                if (el_ret == ELEVATOR_BACK_MERGE) {
1438                        ret = bio_attempt_back_merge(q, rq, bio);
1439                        if (ret)
1440                                break;
1441                } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1442                        ret = bio_attempt_front_merge(q, rq, bio);
1443                        if (ret)
1444                                break;
1445                }
1446        }
1447out:
1448        return ret;
1449}
1450
1451void init_request_from_bio(struct request *req, struct bio *bio)
1452{
1453        req->cmd_type = REQ_TYPE_FS;
1454
1455        req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1456        if (bio->bi_rw & REQ_RAHEAD)
1457                req->cmd_flags |= REQ_FAILFAST_MASK;
1458
1459        req->errors = 0;
1460        req->__sector = bio->bi_sector;
1461        req->ioprio = bio_prio(bio);
1462        blk_rq_bio_prep(req->q, req, bio);
1463}
1464
1465void blk_queue_bio(struct request_queue *q, struct bio *bio)
1466{
1467        const bool sync = !!(bio->bi_rw & REQ_SYNC);
1468        struct blk_plug *plug;
1469        int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1470        struct request *req;
1471        unsigned int request_count = 0;
1472
1473        /*
1474         * low level driver can indicate that it wants pages above a
1475         * certain limit bounced to low memory (ie for highmem, or even
1476         * ISA dma in theory)
1477         */
1478        blk_queue_bounce(q, &bio);
1479
1480        if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1481                bio_endio(bio, -EIO);
1482                return;
1483        }
1484
1485        if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1486                spin_lock_irq(q->queue_lock);
1487                where = ELEVATOR_INSERT_FLUSH;
1488                goto get_rq;
1489        }
1490
1491        /*
1492         * Check if we can merge with the plugged list before grabbing
1493         * any locks.
1494         */
1495        if (attempt_plug_merge(q, bio, &request_count))
1496                return;
1497
1498        spin_lock_irq(q->queue_lock);
1499
1500        el_ret = elv_merge(q, &req, bio);
1501        if (el_ret == ELEVATOR_BACK_MERGE) {
1502                if (bio_attempt_back_merge(q, req, bio)) {
1503                        elv_bio_merged(q, req, bio);
1504                        if (!attempt_back_merge(q, req))
1505                                elv_merged_request(q, req, el_ret);
1506                        goto out_unlock;
1507                }
1508        } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1509                if (bio_attempt_front_merge(q, req, bio)) {
1510                        elv_bio_merged(q, req, bio);
1511                        if (!attempt_front_merge(q, req))
1512                                elv_merged_request(q, req, el_ret);
1513                        goto out_unlock;
1514                }
1515        }
1516
1517get_rq:
1518        /*
1519         * This sync check and mask will be re-done in init_request_from_bio(),
1520         * but we need to set it earlier to expose the sync flag to the
1521         * rq allocator and io schedulers.
1522         */
1523        rw_flags = bio_data_dir(bio);
1524        if (sync)
1525                rw_flags |= REQ_SYNC;
1526
1527        /*
1528         * Grab a free request. This is might sleep but can not fail.
1529         * Returns with the queue unlocked.
1530         */
1531        req = get_request(q, rw_flags, bio, GFP_NOIO);
1532        if (unlikely(!req)) {
1533                bio_endio(bio, -ENODEV);        /* @q is dead */
1534                goto out_unlock;
1535        }
1536
1537        /*
1538         * After dropping the lock and possibly sleeping here, our request
1539         * may now be mergeable after it had proven unmergeable (above).
1540         * We don't worry about that case for efficiency. It won't happen
1541         * often, and the elevators are able to handle it.
1542         */
1543        init_request_from_bio(req, bio);
1544
1545        if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1546                req->cpu = raw_smp_processor_id();
1547
1548        plug = current->plug;
1549        if (plug) {
1550                /*
1551                 * If this is the first request added after a plug, fire
1552                 * of a plug trace. If others have been added before, check
1553                 * if we have multiple devices in this plug. If so, make a
1554                 * note to sort the list before dispatch.
1555                 */
1556                if (list_empty(&plug->list))
1557                        trace_block_plug(q);
1558                else {
1559                        if (request_count >= BLK_MAX_REQUEST_COUNT) {
1560                                blk_flush_plug_list(plug, false);
1561                                trace_block_plug(q);
1562                        }
1563                }
1564                list_add_tail(&req->queuelist, &plug->list);
1565                drive_stat_acct(req, 1);
1566        } else {
1567                spin_lock_irq(q->queue_lock);
1568                add_acct_request(q, req, where);
1569                __blk_run_queue(q);
1570out_unlock:
1571                spin_unlock_irq(q->queue_lock);
1572        }
1573}
1574EXPORT_SYMBOL_GPL(blk_queue_bio);       /* for device mapper only */
1575
1576/*
1577 * If bio->bi_dev is a partition, remap the location
1578 */
1579static inline void blk_partition_remap(struct bio *bio)
1580{
1581        struct block_device *bdev = bio->bi_bdev;
1582
1583        if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1584                struct hd_struct *p = bdev->bd_part;
1585
1586                bio->bi_sector += p->start_sect;
1587                bio->bi_bdev = bdev->bd_contains;
1588
1589                trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1590                                      bdev->bd_dev,
1591                                      bio->bi_sector - p->start_sect);
1592        }
1593}
1594
1595static void handle_bad_sector(struct bio *bio)
1596{
1597        char b[BDEVNAME_SIZE];
1598
1599        printk(KERN_INFO "attempt to access beyond end of device\n");
1600        printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1601                        bdevname(bio->bi_bdev, b),
1602                        bio->bi_rw,
1603                        (unsigned long long)bio_end_sector(bio),
1604                        (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1605
1606        set_bit(BIO_EOF, &bio->bi_flags);
1607}
1608
1609#ifdef CONFIG_FAIL_MAKE_REQUEST
1610
1611static DECLARE_FAULT_ATTR(fail_make_request);
1612
1613static int __init setup_fail_make_request(char *str)
1614{
1615        return setup_fault_attr(&fail_make_request, str);
1616}
1617__setup("fail_make_request=", setup_fail_make_request);
1618
1619static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1620{
1621        return part->make_it_fail && should_fail(&fail_make_request, bytes);
1622}
1623
1624static int __init fail_make_request_debugfs(void)
1625{
1626        struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1627                                                NULL, &fail_make_request);
1628
1629        return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1630}
1631
1632late_initcall(fail_make_request_debugfs);
1633
1634#else /* CONFIG_FAIL_MAKE_REQUEST */
1635
1636static inline bool should_fail_request(struct hd_struct *part,
1637                                        unsigned int bytes)
1638{
1639        return false;
1640}
1641
1642#endif /* CONFIG_FAIL_MAKE_REQUEST */
1643
1644/*
1645 * Check whether this bio extends beyond the end of the device.
1646 */
1647static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1648{
1649        sector_t maxsector;
1650
1651        if (!nr_sectors)
1652                return 0;
1653
1654        /* Test device or partition size, when known. */
1655        maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1656        if (maxsector) {
1657                sector_t sector = bio->bi_sector;
1658
1659                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1660                        /*
1661                         * This may well happen - the kernel calls bread()
1662                         * without checking the size of the device, e.g., when
1663                         * mounting a device.
1664                         */
1665                        handle_bad_sector(bio);
1666                        return 1;
1667                }
1668        }
1669
1670        return 0;
1671}
1672
1673static noinline_for_stack bool
1674generic_make_request_checks(struct bio *bio)
1675{
1676        struct request_queue *q;
1677        int nr_sectors = bio_sectors(bio);
1678        int err = -EIO;
1679        char b[BDEVNAME_SIZE];
1680        struct hd_struct *part;
1681
1682        might_sleep();
1683
1684        if (bio_check_eod(bio, nr_sectors))
1685                goto end_io;
1686
1687        q = bdev_get_queue(bio->bi_bdev);
1688        if (unlikely(!q)) {
1689                printk(KERN_ERR
1690                       "generic_make_request: Trying to access "
1691                        "nonexistent block-device %s (%Lu)\n",
1692                        bdevname(bio->bi_bdev, b),
1693                        (long long) bio->bi_sector);
1694                goto end_io;
1695        }
1696
1697        if (likely(bio_is_rw(bio) &&
1698                   nr_sectors > queue_max_hw_sectors(q))) {
1699                printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1700                       bdevname(bio->bi_bdev, b),
1701                       bio_sectors(bio),
1702                       queue_max_hw_sectors(q));
1703                goto end_io;
1704        }
1705
1706        part = bio->bi_bdev->bd_part;
1707        if (should_fail_request(part, bio->bi_size) ||
1708            should_fail_request(&part_to_disk(part)->part0,
1709                                bio->bi_size))
1710                goto end_io;
1711
1712        /*
1713         * If this device has partitions, remap block n
1714         * of partition p to block n+start(p) of the disk.
1715         */
1716        blk_partition_remap(bio);
1717
1718        if (bio_check_eod(bio, nr_sectors))
1719                goto end_io;
1720
1721        /*
1722         * Filter flush bio's early so that make_request based
1723         * drivers without flush support don't have to worry
1724         * about them.
1725         */
1726        if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1727                bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1728                if (!nr_sectors) {
1729                        err = 0;
1730                        goto end_io;
1731                }
1732        }
1733
1734        if ((bio->bi_rw & REQ_DISCARD) &&
1735            (!blk_queue_discard(q) ||
1736             ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1737                err = -EOPNOTSUPP;
1738                goto end_io;
1739        }
1740
1741        if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1742                err = -EOPNOTSUPP;
1743                goto end_io;
1744        }
1745
1746        /*
1747         * Various block parts want %current->io_context and lazy ioc
1748         * allocation ends up trading a lot of pain for a small amount of
1749         * memory.  Just allocate it upfront.  This may fail and block
1750         * layer knows how to live with it.
1751         */
1752        create_io_context(GFP_ATOMIC, q->node);
1753
1754        if (blk_throtl_bio(q, bio))
1755                return false;   /* throttled, will be resubmitted later */
1756
1757        trace_block_bio_queue(q, bio);
1758        return true;
1759
1760end_io:
1761        bio_endio(bio, err);
1762        return false;
1763}
1764
1765/**
1766 * generic_make_request - hand a buffer to its device driver for I/O
1767 * @bio:  The bio describing the location in memory and on the device.
1768 *
1769 * generic_make_request() is used to make I/O requests of block
1770 * devices. It is passed a &struct bio, which describes the I/O that needs
1771 * to be done.
1772 *
1773 * generic_make_request() does not return any status.  The
1774 * success/failure status of the request, along with notification of
1775 * completion, is delivered asynchronously through the bio->bi_end_io
1776 * function described (one day) else where.
1777 *
1778 * The caller of generic_make_request must make sure that bi_io_vec
1779 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1780 * set to describe the device address, and the
1781 * bi_end_io and optionally bi_private are set to describe how
1782 * completion notification should be signaled.
1783 *
1784 * generic_make_request and the drivers it calls may use bi_next if this
1785 * bio happens to be merged with someone else, and may resubmit the bio to
1786 * a lower device by calling into generic_make_request recursively, which
1787 * means the bio should NOT be touched after the call to ->make_request_fn.
1788 */
1789void generic_make_request(struct bio *bio)
1790{
1791        struct bio_list bio_list_on_stack;
1792
1793        if (!generic_make_request_checks(bio))
1794                return;
1795
1796        /*
1797         * We only want one ->make_request_fn to be active at a time, else
1798         * stack usage with stacked devices could be a problem.  So use
1799         * current->bio_list to keep a list of requests submited by a
1800         * make_request_fn function.  current->bio_list is also used as a
1801         * flag to say if generic_make_request is currently active in this
1802         * task or not.  If it is NULL, then no make_request is active.  If
1803         * it is non-NULL, then a make_request is active, and new requests
1804         * should be added at the tail
1805         */
1806        if (current->bio_list) {
1807                bio_list_add(current->bio_list, bio);
1808                return;
1809        }
1810
1811        /* following loop may be a bit non-obvious, and so deserves some
1812         * explanation.
1813         * Before entering the loop, bio->bi_next is NULL (as all callers
1814         * ensure that) so we have a list with a single bio.
1815         * We pretend that we have just taken it off a longer list, so
1816         * we assign bio_list to a pointer to the bio_list_on_stack,
1817         * thus initialising the bio_list of new bios to be
1818         * added.  ->make_request() may indeed add some more bios
1819         * through a recursive call to generic_make_request.  If it
1820         * did, we find a non-NULL value in bio_list and re-enter the loop
1821         * from the top.  In this case we really did just take the bio
1822         * of the top of the list (no pretending) and so remove it from
1823         * bio_list, and call into ->make_request() again.
1824         */
1825        BUG_ON(bio->bi_next);
1826        bio_list_init(&bio_list_on_stack);
1827        current->bio_list = &bio_list_on_stack;
1828        do {
1829                struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1830
1831                q->make_request_fn(q, bio);
1832
1833                bio = bio_list_pop(current->bio_list);
1834        } while (bio);
1835        current->bio_list = NULL; /* deactivate */
1836}
1837EXPORT_SYMBOL(generic_make_request);
1838
1839/**
1840 * submit_bio - submit a bio to the block device layer for I/O
1841 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1842 * @bio: The &struct bio which describes the I/O
1843 *
1844 * submit_bio() is very similar in purpose to generic_make_request(), and
1845 * uses that function to do most of the work. Both are fairly rough
1846 * interfaces; @bio must be presetup and ready for I/O.
1847 *
1848 */
1849void submit_bio(int rw, struct bio *bio)
1850{
1851        bio->bi_rw |= rw;
1852
1853        /*
1854         * If it's a regular read/write or a barrier with data attached,
1855         * go through the normal accounting stuff before submission.
1856         */
1857        if (bio_has_data(bio)) {
1858                unsigned int count;
1859
1860                if (unlikely(rw & REQ_WRITE_SAME))
1861                        count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1862                else
1863                        count = bio_sectors(bio);
1864
1865                if (rw & WRITE) {
1866                        count_vm_events(PGPGOUT, count);
1867                } else {
1868                        task_io_account_read(bio->bi_size);
1869                        count_vm_events(PGPGIN, count);
1870                }
1871
1872                if (unlikely(block_dump)) {
1873                        char b[BDEVNAME_SIZE];
1874                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1875                        current->comm, task_pid_nr(current),
1876                                (rw & WRITE) ? "WRITE" : "READ",
1877                                (unsigned long long)bio->bi_sector,
1878                                bdevname(bio->bi_bdev, b),
1879                                count);
1880                }
1881        }
1882
1883        generic_make_request(bio);
1884}
1885EXPORT_SYMBOL(submit_bio);
1886
1887/**
1888 * blk_rq_check_limits - Helper function to check a request for the queue limit
1889 * @q:  the queue
1890 * @rq: the request being checked
1891 *
1892 * Description:
1893 *    @rq may have been made based on weaker limitations of upper-level queues
1894 *    in request stacking drivers, and it may violate the limitation of @q.
1895 *    Since the block layer and the underlying device driver trust @rq
1896 *    after it is inserted to @q, it should be checked against @q before
1897 *    the insertion using this generic function.
1898 *
1899 *    This function should also be useful for request stacking drivers
1900 *    in some cases below, so export this function.
1901 *    Request stacking drivers like request-based dm may change the queue
1902 *    limits while requests are in the queue (e.g. dm's table swapping).
1903 *    Such request stacking drivers should check those requests agaist
1904 *    the new queue limits again when they dispatch those requests,
1905 *    although such checkings are also done against the old queue limits
1906 *    when submitting requests.
1907 */
1908int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1909{
1910        if (!rq_mergeable(rq))
1911                return 0;
1912
1913        if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1914                printk(KERN_ERR "%s: over max size limit.\n", __func__);
1915                return -EIO;
1916        }
1917
1918        /*
1919         * queue's settings related to segment counting like q->bounce_pfn
1920         * may differ from that of other stacking queues.
1921         * Recalculate it to check the request correctly on this queue's
1922         * limitation.
1923         */
1924        blk_recalc_rq_segments(rq);
1925        if (rq->nr_phys_segments > queue_max_segments(q)) {
1926                printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1927                return -EIO;
1928        }
1929
1930        return 0;
1931}
1932EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1933
1934/**
1935 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1936 * @q:  the queue to submit the request
1937 * @rq: the request being queued
1938 */
1939int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1940{
1941        unsigned long flags;
1942        int where = ELEVATOR_INSERT_BACK;
1943
1944        if (blk_rq_check_limits(q, rq))
1945                return -EIO;
1946
1947        if (rq->rq_disk &&
1948            should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1949                return -EIO;
1950
1951        spin_lock_irqsave(q->queue_lock, flags);
1952        if (unlikely(blk_queue_dying(q))) {
1953                spin_unlock_irqrestore(q->queue_lock, flags);
1954                return -ENODEV;
1955        }
1956
1957        /*
1958         * Submitting request must be dequeued before calling this function
1959         * because it will be linked to another request_queue
1960         */
1961        BUG_ON(blk_queued_rq(rq));
1962
1963        if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1964                where = ELEVATOR_INSERT_FLUSH;
1965
1966        add_acct_request(q, rq, where);
1967        if (where == ELEVATOR_INSERT_FLUSH)
1968                __blk_run_queue(q);
1969        spin_unlock_irqrestore(q->queue_lock, flags);
1970
1971        return 0;
1972}
1973EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1974
1975/**
1976 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1977 * @rq: request to examine
1978 *
1979 * Description:
1980 *     A request could be merge of IOs which require different failure
1981 *     handling.  This function determines the number of bytes which
1982 *     can be failed from the beginning of the request without
1983 *     crossing into area which need to be retried further.
1984 *
1985 * Return:
1986 *     The number of bytes to fail.
1987 *
1988 * Context:
1989 *     queue_lock must be held.
1990 */
1991unsigned int blk_rq_err_bytes(const struct request *rq)
1992{
1993        unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1994        unsigned int bytes = 0;
1995        struct bio *bio;
1996
1997        if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1998                return blk_rq_bytes(rq);
1999
2000        /*
2001         * Currently the only 'mixing' which can happen is between
2002         * different fastfail types.  We can safely fail portions
2003         * which have all the failfast bits that the first one has -
2004         * the ones which are at least as eager to fail as the first
2005         * one.
2006         */
2007        for (bio = rq->bio; bio; bio = bio->bi_next) {
2008                if ((bio->bi_rw & ff) != ff)
2009                        break;
2010                bytes += bio->bi_size;
2011        }
2012
2013        /* this could lead to infinite loop */
2014        BUG_ON(blk_rq_bytes(rq) && !bytes);
2015        return bytes;
2016}
2017EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2018
2019static void blk_account_io_completion(struct request *req, unsigned int bytes)
2020{
2021        if (blk_do_io_stat(req)) {
2022                const int rw = rq_data_dir(req);
2023                struct hd_struct *part;
2024                int cpu;
2025
2026                cpu = part_stat_lock();
2027                part = req->part;
2028                part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2029                part_stat_unlock();
2030        }
2031}
2032
2033static void blk_account_io_done(struct request *req)
2034{
2035        /*
2036         * Account IO completion.  flush_rq isn't accounted as a
2037         * normal IO on queueing nor completion.  Accounting the
2038         * containing request is enough.
2039         */
2040        if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2041                unsigned long duration = jiffies - req->start_time;
2042                const int rw = rq_data_dir(req);
2043                struct hd_struct *part;
2044                int cpu;
2045
2046                cpu = part_stat_lock();
2047                part = req->part;
2048
2049                part_stat_inc(cpu, part, ios[rw]);
2050                part_stat_add(cpu, part, ticks[rw], duration);
2051                part_round_stats(cpu, part);
2052                part_dec_in_flight(part, rw);
2053
2054                hd_struct_put(part);
2055                part_stat_unlock();
2056        }
2057}
2058
2059#ifdef CONFIG_PM_RUNTIME
2060/*
2061 * Don't process normal requests when queue is suspended
2062 * or in the process of suspending/resuming
2063 */
2064static struct request *blk_pm_peek_request(struct request_queue *q,
2065                                           struct request *rq)
2066{
2067        if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2068            (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2069                return NULL;
2070        else
2071                return rq;
2072}
2073#else
2074static inline struct request *blk_pm_peek_request(struct request_queue *q,
2075                                                  struct request *rq)
2076{
2077        return rq;
2078}
2079#endif
2080
2081/**
2082 * blk_peek_request - peek at the top of a request queue
2083 * @q: request queue to peek at
2084 *
2085 * Description:
2086 *     Return the request at the top of @q.  The returned request
2087 *     should be started using blk_start_request() before LLD starts
2088 *     processing it.
2089 *
2090 * Return:
2091 *     Pointer to the request at the top of @q if available.  Null
2092 *     otherwise.
2093 *
2094 * Context:
2095 *     queue_lock must be held.
2096 */
2097struct request *blk_peek_request(struct request_queue *q)
2098{
2099        struct request *rq;
2100        int ret;
2101
2102        while ((rq = __elv_next_request(q)) != NULL) {
2103
2104                rq = blk_pm_peek_request(q, rq);
2105                if (!rq)
2106                        break;
2107
2108                if (!(rq->cmd_flags & REQ_STARTED)) {
2109                        /*
2110                         * This is the first time the device driver
2111                         * sees this request (possibly after
2112                         * requeueing).  Notify IO scheduler.
2113                         */
2114                        if (rq->cmd_flags & REQ_SORTED)
2115                                elv_activate_rq(q, rq);
2116
2117                        /*
2118                         * just mark as started even if we don't start
2119                         * it, a request that has been delayed should
2120                         * not be passed by new incoming requests
2121                         */
2122                        rq->cmd_flags |= REQ_STARTED;
2123                        trace_block_rq_issue(q, rq);
2124                }
2125
2126                if (!q->boundary_rq || q->boundary_rq == rq) {
2127                        q->end_sector = rq_end_sector(rq);
2128                        q->boundary_rq = NULL;
2129                }
2130
2131                if (rq->cmd_flags & REQ_DONTPREP)
2132                        break;
2133
2134                if (q->dma_drain_size && blk_rq_bytes(rq)) {
2135                        /*
2136                         * make sure space for the drain appears we
2137                         * know we can do this because max_hw_segments
2138                         * has been adjusted to be one fewer than the
2139                         * device can handle
2140                         */
2141                        rq->nr_phys_segments++;
2142                }
2143
2144                if (!q->prep_rq_fn)
2145                        break;
2146
2147                ret = q->prep_rq_fn(q, rq);
2148                if (ret == BLKPREP_OK) {
2149                        break;
2150                } else if (ret == BLKPREP_DEFER) {
2151                        /*
2152                         * the request may have been (partially) prepped.
2153                         * we need to keep this request in the front to
2154                         * avoid resource deadlock.  REQ_STARTED will
2155                         * prevent other fs requests from passing this one.
2156                         */
2157                        if (q->dma_drain_size && blk_rq_bytes(rq) &&
2158                            !(rq->cmd_flags & REQ_DONTPREP)) {
2159                                /*
2160                                 * remove the space for the drain we added
2161                                 * so that we don't add it again
2162                                 */
2163                                --rq->nr_phys_segments;
2164                        }
2165
2166                        rq = NULL;
2167                        break;
2168                } else if (ret == BLKPREP_KILL) {
2169                        rq->cmd_flags |= REQ_QUIET;
2170                        /*
2171                         * Mark this request as started so we don't trigger
2172                         * any debug logic in the end I/O path.
2173                         */
2174                        blk_start_request(rq);
2175                        __blk_end_request_all(rq, -EIO);
2176                } else {
2177                        printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2178                        break;
2179                }
2180        }
2181
2182        return rq;
2183}
2184EXPORT_SYMBOL(blk_peek_request);
2185
2186void blk_dequeue_request(struct request *rq)
2187{
2188        struct request_queue *q = rq->q;
2189
2190        BUG_ON(list_empty(&rq->queuelist));
2191        BUG_ON(ELV_ON_HASH(rq));
2192
2193        list_del_init(&rq->queuelist);
2194
2195        /*
2196         * the time frame between a request being removed from the lists
2197         * and to it is freed is accounted as io that is in progress at
2198         * the driver side.
2199         */
2200        if (blk_account_rq(rq)) {
2201                q->in_flight[rq_is_sync(rq)]++;
2202                set_io_start_time_ns(rq);
2203        }
2204}
2205
2206/**
2207 * blk_start_request - start request processing on the driver
2208 * @req: request to dequeue
2209 *
2210 * Description:
2211 *     Dequeue @req and start timeout timer on it.  This hands off the
2212 *     request to the driver.
2213 *
2214 *     Block internal functions which don't want to start timer should
2215 *     call blk_dequeue_request().
2216 *
2217 * Context:
2218 *     queue_lock must be held.
2219 */
2220void blk_start_request(struct request *req)
2221{
2222        blk_dequeue_request(req);
2223
2224        /*
2225         * We are now handing the request to the hardware, initialize
2226         * resid_len to full count and add the timeout handler.
2227         */
2228        req->resid_len = blk_rq_bytes(req);
2229        if (unlikely(blk_bidi_rq(req)))
2230                req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2231
2232        blk_add_timer(req);
2233}
2234EXPORT_SYMBOL(blk_start_request);
2235
2236/**
2237 * blk_fetch_request - fetch a request from a request queue
2238 * @q: request queue to fetch a request from
2239 *
2240 * Description:
2241 *     Return the request at the top of @q.  The request is started on
2242 *     return and LLD can start processing it immediately.
2243 *
2244 * Return:
2245 *     Pointer to the request at the top of @q if available.  Null
2246 *     otherwise.
2247 *
2248 * Context:
2249 *     queue_lock must be held.
2250 */
2251struct request *blk_fetch_request(struct request_queue *q)
2252{
2253        struct request *rq;
2254
2255        rq = blk_peek_request(q);
2256        if (rq)
2257                blk_start_request(rq);
2258        return rq;
2259}
2260EXPORT_SYMBOL(blk_fetch_request);
2261
2262/**
2263 * blk_update_request - Special helper function for request stacking drivers
2264 * @req:      the request being processed
2265 * @error:    %0 for success, < %0 for error
2266 * @nr_bytes: number of bytes to complete @req
2267 *
2268 * Description:
2269 *     Ends I/O on a number of bytes attached to @req, but doesn't complete
2270 *     the request structure even if @req doesn't have leftover.
2271 *     If @req has leftover, sets it up for the next range of segments.
2272 *
2273 *     This special helper function is only for request stacking drivers
2274 *     (e.g. request-based dm) so that they can handle partial completion.
2275 *     Actual device drivers should use blk_end_request instead.
2276 *
2277 *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2278 *     %false return from this function.
2279 *
2280 * Return:
2281 *     %false - this request doesn't have any more data
2282 *     %true  - this request has more data
2283 **/
2284bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2285{
2286        int total_bytes;
2287
2288        if (!req->bio)
2289                return false;
2290
2291        trace_block_rq_complete(req->q, req);
2292
2293        /*
2294         * For fs requests, rq is just carrier of independent bio's
2295         * and each partial completion should be handled separately.
2296         * Reset per-request error on each partial completion.
2297         *
2298         * TODO: tj: This is too subtle.  It would be better to let
2299         * low level drivers do what they see fit.
2300         */
2301        if (req->cmd_type == REQ_TYPE_FS)
2302                req->errors = 0;
2303
2304        if (error && req->cmd_type == REQ_TYPE_FS &&
2305            !(req->cmd_flags & REQ_QUIET)) {
2306                char *error_type;
2307
2308                switch (error) {
2309                case -ENOLINK:
2310                        error_type = "recoverable transport";
2311                        break;
2312                case -EREMOTEIO:
2313                        error_type = "critical target";
2314                        break;
2315                case -EBADE:
2316                        error_type = "critical nexus";
2317                        break;
2318                case -EIO:
2319                default:
2320                        error_type = "I/O";
2321                        break;
2322                }
2323                printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2324                                   error_type, req->rq_disk ?
2325                                   req->rq_disk->disk_name : "?",
2326                                   (unsigned long long)blk_rq_pos(req));
2327
2328        }
2329
2330        blk_account_io_completion(req, nr_bytes);
2331
2332        total_bytes = 0;
2333        while (req->bio) {
2334                struct bio *bio = req->bio;
2335                unsigned bio_bytes = min(bio->bi_size, nr_bytes);
2336
2337                if (bio_bytes == bio->bi_size)
2338                        req->bio = bio->bi_next;
2339
2340                req_bio_endio(req, bio, bio_bytes, error);
2341
2342                total_bytes += bio_bytes;
2343                nr_bytes -= bio_bytes;
2344
2345                if (!nr_bytes)
2346                        break;
2347        }
2348
2349        /*
2350         * completely done
2351         */
2352        if (!req->bio) {
2353                /*
2354                 * Reset counters so that the request stacking driver
2355                 * can find how many bytes remain in the request
2356                 * later.
2357                 */
2358                req->__data_len = 0;
2359                return false;
2360        }
2361
2362        req->__data_len -= total_bytes;
2363        req->buffer = bio_data(req->bio);
2364
2365        /* update sector only for requests with clear definition of sector */
2366        if (req->cmd_type == REQ_TYPE_FS)
2367                req->__sector += total_bytes >> 9;
2368
2369        /* mixed attributes always follow the first bio */
2370        if (req->cmd_flags & REQ_MIXED_MERGE) {
2371                req->cmd_flags &= ~REQ_FAILFAST_MASK;
2372                req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2373        }
2374
2375        /*
2376         * If total number of sectors is less than the first segment
2377         * size, something has gone terribly wrong.
2378         */
2379        if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2380                blk_dump_rq_flags(req, "request botched");
2381                req->__data_len = blk_rq_cur_bytes(req);
2382        }
2383
2384        /* recalculate the number of segments */
2385        blk_recalc_rq_segments(req);
2386
2387        return true;
2388}
2389EXPORT_SYMBOL_GPL(blk_update_request);
2390
2391static bool blk_update_bidi_request(struct request *rq, int error,
2392                                    unsigned int nr_bytes,
2393                                    unsigned int bidi_bytes)
2394{
2395        if (blk_update_request(rq, error, nr_bytes))
2396                return true;
2397
2398        /* Bidi request must be completed as a whole */
2399        if (unlikely(blk_bidi_rq(rq)) &&
2400            blk_update_request(rq->next_rq, error, bidi_bytes))
2401                return true;
2402
2403        if (blk_queue_add_random(rq->q))
2404                add_disk_randomness(rq->rq_disk);
2405
2406        return false;
2407}
2408
2409/**
2410 * blk_unprep_request - unprepare a request
2411 * @req:        the request
2412 *
2413 * This function makes a request ready for complete resubmission (or
2414 * completion).  It happens only after all error handling is complete,
2415 * so represents the appropriate moment to deallocate any resources
2416 * that were allocated to the request in the prep_rq_fn.  The queue
2417 * lock is held when calling this.
2418 */
2419void blk_unprep_request(struct request *req)
2420{
2421        struct request_queue *q = req->q;
2422
2423        req->cmd_flags &= ~REQ_DONTPREP;
2424        if (q->unprep_rq_fn)
2425                q->unprep_rq_fn(q, req);
2426}
2427EXPORT_SYMBOL_GPL(blk_unprep_request);
2428
2429/*
2430 * queue lock must be held
2431 */
2432static void blk_finish_request(struct request *req, int error)
2433{
2434        if (blk_rq_tagged(req))
2435                blk_queue_end_tag(req->q, req);
2436
2437        BUG_ON(blk_queued_rq(req));
2438
2439        if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2440                laptop_io_completion(&req->q->backing_dev_info);
2441
2442        blk_delete_timer(req);
2443
2444        if (req->cmd_flags & REQ_DONTPREP)
2445                blk_unprep_request(req);
2446
2447
2448        blk_account_io_done(req);
2449
2450        if (req->end_io)
2451                req->end_io(req, error);
2452        else {
2453                if (blk_bidi_rq(req))
2454                        __blk_put_request(req->next_rq->q, req->next_rq);
2455
2456                __blk_put_request(req->q, req);
2457        }
2458}
2459
2460/**
2461 * blk_end_bidi_request - Complete a bidi request
2462 * @rq:         the request to complete
2463 * @error:      %0 for success, < %0 for error
2464 * @nr_bytes:   number of bytes to complete @rq
2465 * @bidi_bytes: number of bytes to complete @rq->next_rq
2466 *
2467 * Description:
2468 *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2469 *     Drivers that supports bidi can safely call this member for any
2470 *     type of request, bidi or uni.  In the later case @bidi_bytes is
2471 *     just ignored.
2472 *
2473 * Return:
2474 *     %false - we are done with this request
2475 *     %true  - still buffers pending for this request
2476 **/
2477static bool blk_end_bidi_request(struct request *rq, int error,
2478                                 unsigned int nr_bytes, unsigned int bidi_bytes)
2479{
2480        struct request_queue *q = rq->q;
2481        unsigned long flags;
2482
2483        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2484                return true;
2485
2486        spin_lock_irqsave(q->queue_lock, flags);
2487        blk_finish_request(rq, error);
2488        spin_unlock_irqrestore(q->queue_lock, flags);
2489
2490        return false;
2491}
2492
2493/**
2494 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2495 * @rq:         the request to complete
2496 * @error:      %0 for success, < %0 for error
2497 * @nr_bytes:   number of bytes to complete @rq
2498 * @bidi_bytes: number of bytes to complete @rq->next_rq
2499 *
2500 * Description:
2501 *     Identical to blk_end_bidi_request() except that queue lock is
2502 *     assumed to be locked on entry and remains so on return.
2503 *
2504 * Return:
2505 *     %false - we are done with this request
2506 *     %true  - still buffers pending for this request
2507 **/
2508bool __blk_end_bidi_request(struct request *rq, int error,
2509                                   unsigned int nr_bytes, unsigned int bidi_bytes)
2510{
2511        if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2512                return true;
2513
2514        blk_finish_request(rq, error);
2515
2516        return false;
2517}
2518
2519/**
2520 * blk_end_request - Helper function for drivers to complete the request.
2521 * @rq:       the request being processed
2522 * @error:    %0 for success, < %0 for error
2523 * @nr_bytes: number of bytes to complete
2524 *
2525 * Description:
2526 *     Ends I/O on a number of bytes attached to @rq.
2527 *     If @rq has leftover, sets it up for the next range of segments.
2528 *
2529 * Return:
2530 *     %false - we are done with this request
2531 *     %true  - still buffers pending for this request
2532 **/
2533bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2534{
2535        return blk_end_bidi_request(rq, error, nr_bytes, 0);
2536}
2537EXPORT_SYMBOL(blk_end_request);
2538
2539/**
2540 * blk_end_request_all - Helper function for drives to finish the request.
2541 * @rq: the request to finish
2542 * @error: %0 for success, < %0 for error
2543 *
2544 * Description:
2545 *     Completely finish @rq.
2546 */
2547void blk_end_request_all(struct request *rq, int error)
2548{
2549        bool pending;
2550        unsigned int bidi_bytes = 0;
2551
2552        if (unlikely(blk_bidi_rq(rq)))
2553                bidi_bytes = blk_rq_bytes(rq->next_rq);
2554
2555        pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2556        BUG_ON(pending);
2557}
2558EXPORT_SYMBOL(blk_end_request_all);
2559
2560/**
2561 * blk_end_request_cur - Helper function to finish the current request chunk.
2562 * @rq: the request to finish the current chunk for
2563 * @error: %0 for success, < %0 for error
2564 *
2565 * Description:
2566 *     Complete the current consecutively mapped chunk from @rq.
2567 *
2568 * Return:
2569 *     %false - we are done with this request
2570 *     %true  - still buffers pending for this request
2571 */
2572bool blk_end_request_cur(struct request *rq, int error)
2573{
2574        return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2575}
2576EXPORT_SYMBOL(blk_end_request_cur);
2577
2578/**
2579 * blk_end_request_err - Finish a request till the next failure boundary.
2580 * @rq: the request to finish till the next failure boundary for
2581 * @error: must be negative errno
2582 *
2583 * Description:
2584 *     Complete @rq till the next failure boundary.
2585 *
2586 * Return:
2587 *     %false - we are done with this request
2588 *     %true  - still buffers pending for this request
2589 */
2590bool blk_end_request_err(struct request *rq, int error)
2591{
2592        WARN_ON(error >= 0);
2593        return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2594}
2595EXPORT_SYMBOL_GPL(blk_end_request_err);
2596
2597/**
2598 * __blk_end_request - Helper function for drivers to complete the request.
2599 * @rq:       the request being processed
2600 * @error:    %0 for success, < %0 for error
2601 * @nr_bytes: number of bytes to complete
2602 *
2603 * Description:
2604 *     Must be called with queue lock held unlike blk_end_request().
2605 *
2606 * Return:
2607 *     %false - we are done with this request
2608 *     %true  - still buffers pending for this request
2609 **/
2610bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2611{
2612        return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2613}
2614EXPORT_SYMBOL(__blk_end_request);
2615
2616/**
2617 * __blk_end_request_all - Helper function for drives to finish the request.
2618 * @rq: the request to finish
2619 * @error: %0 for success, < %0 for error
2620 *
2621 * Description:
2622 *     Completely finish @rq.  Must be called with queue lock held.
2623 */
2624void __blk_end_request_all(struct request *rq, int error)
2625{
2626        bool pending;
2627        unsigned int bidi_bytes = 0;
2628
2629        if (unlikely(blk_bidi_rq(rq)))
2630                bidi_bytes = blk_rq_bytes(rq->next_rq);
2631
2632        pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2633        BUG_ON(pending);
2634}
2635EXPORT_SYMBOL(__blk_end_request_all);
2636
2637/**
2638 * __blk_end_request_cur - Helper function to finish the current request chunk.
2639 * @rq: the request to finish the current chunk for
2640 * @error: %0 for success, < %0 for error
2641 *
2642 * Description:
2643 *     Complete the current consecutively mapped chunk from @rq.  Must
2644 *     be called with queue lock held.
2645 *
2646 * Return:
2647 *     %false - we are done with this request
2648 *     %true  - still buffers pending for this request
2649 */
2650bool __blk_end_request_cur(struct request *rq, int error)
2651{
2652        return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2653}
2654EXPORT_SYMBOL(__blk_end_request_cur);
2655
2656/**
2657 * __blk_end_request_err - Finish a request till the next failure boundary.
2658 * @rq: the request to finish till the next failure boundary for
2659 * @error: must be negative errno
2660 *
2661 * Description:
2662 *     Complete @rq till the next failure boundary.  Must be called
2663 *     with queue lock held.
2664 *
2665 * Return:
2666 *     %false - we are done with this request
2667 *     %true  - still buffers pending for this request
2668 */
2669bool __blk_end_request_err(struct request *rq, int error)
2670{
2671        WARN_ON(error >= 0);
2672        return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2673}
2674EXPORT_SYMBOL_GPL(__blk_end_request_err);
2675
2676void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2677                     struct bio *bio)
2678{
2679        /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2680        rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2681
2682        if (bio_has_data(bio)) {
2683                rq->nr_phys_segments = bio_phys_segments(q, bio);
2684                rq->buffer = bio_data(bio);
2685        }
2686        rq->__data_len = bio->bi_size;
2687        rq->bio = rq->biotail = bio;
2688
2689        if (bio->bi_bdev)
2690                rq->rq_disk = bio->bi_bdev->bd_disk;
2691}
2692
2693#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2694/**
2695 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2696 * @rq: the request to be flushed
2697 *
2698 * Description:
2699 *     Flush all pages in @rq.
2700 */
2701void rq_flush_dcache_pages(struct request *rq)
2702{
2703        struct req_iterator iter;
2704        struct bio_vec *bvec;
2705
2706        rq_for_each_segment(bvec, rq, iter)
2707                flush_dcache_page(bvec->bv_page);
2708}
2709EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2710#endif
2711
2712/**
2713 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2714 * @q : the queue of the device being checked
2715 *
2716 * Description:
2717 *    Check if underlying low-level drivers of a device are busy.
2718 *    If the drivers want to export their busy state, they must set own
2719 *    exporting function using blk_queue_lld_busy() first.
2720 *
2721 *    Basically, this function is used only by request stacking drivers
2722 *    to stop dispatching requests to underlying devices when underlying
2723 *    devices are busy.  This behavior helps more I/O merging on the queue
2724 *    of the request stacking driver and prevents I/O throughput regression
2725 *    on burst I/O load.
2726 *
2727 * Return:
2728 *    0 - Not busy (The request stacking driver should dispatch request)
2729 *    1 - Busy (The request stacking driver should stop dispatching request)
2730 */
2731int blk_lld_busy(struct request_queue *q)
2732{
2733        if (q->lld_busy_fn)
2734                return q->lld_busy_fn(q);
2735
2736        return 0;
2737}
2738EXPORT_SYMBOL_GPL(blk_lld_busy);
2739
2740/**
2741 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2742 * @rq: the clone request to be cleaned up
2743 *
2744 * Description:
2745 *     Free all bios in @rq for a cloned request.
2746 */
2747void blk_rq_unprep_clone(struct request *rq)
2748{
2749        struct bio *bio;
2750
2751        while ((bio = rq->bio) != NULL) {
2752                rq->bio = bio->bi_next;
2753
2754                bio_put(bio);
2755        }
2756}
2757EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2758
2759/*
2760 * Copy attributes of the original request to the clone request.
2761 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2762 */
2763static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2764{
2765        dst->cpu = src->cpu;
2766        dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2767        dst->cmd_type = src->cmd_type;
2768        dst->__sector = blk_rq_pos(src);
2769        dst->__data_len = blk_rq_bytes(src);
2770        dst->nr_phys_segments = src->nr_phys_segments;
2771        dst->ioprio = src->ioprio;
2772        dst->extra_len = src->extra_len;
2773}
2774
2775/**
2776 * blk_rq_prep_clone - Helper function to setup clone request
2777 * @rq: the request to be setup
2778 * @rq_src: original request to be cloned
2779 * @bs: bio_set that bios for clone are allocated from
2780 * @gfp_mask: memory allocation mask for bio
2781 * @bio_ctr: setup function to be called for each clone bio.
2782 *           Returns %0 for success, non %0 for failure.
2783 * @data: private data to be passed to @bio_ctr
2784 *
2785 * Description:
2786 *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2787 *     The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2788 *     are not copied, and copying such parts is the caller's responsibility.
2789 *     Also, pages which the original bios are pointing to are not copied
2790 *     and the cloned bios just point same pages.
2791 *     So cloned bios must be completed before original bios, which means
2792 *     the caller must complete @rq before @rq_src.
2793 */
2794int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2795                      struct bio_set *bs, gfp_t gfp_mask,
2796                      int (*bio_ctr)(struct bio *, struct bio *, void *),
2797                      void *data)
2798{
2799        struct bio *bio, *bio_src;
2800
2801        if (!bs)
2802                bs = fs_bio_set;
2803
2804        blk_rq_init(NULL, rq);
2805
2806        __rq_for_each_bio(bio_src, rq_src) {
2807                bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2808                if (!bio)
2809                        goto free_and_out;
2810
2811                if (bio_ctr && bio_ctr(bio, bio_src, data))
2812                        goto free_and_out;
2813
2814                if (rq->bio) {
2815                        rq->biotail->bi_next = bio;
2816                        rq->biotail = bio;
2817                } else
2818                        rq->bio = rq->biotail = bio;
2819        }
2820
2821        __blk_rq_prep_clone(rq, rq_src);
2822
2823        return 0;
2824
2825free_and_out:
2826        if (bio)
2827                bio_put(bio);
2828        blk_rq_unprep_clone(rq);
2829
2830        return -ENOMEM;
2831}
2832EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2833
2834int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2835{
2836        return queue_work(kblockd_workqueue, work);
2837}
2838EXPORT_SYMBOL(kblockd_schedule_work);
2839
2840int kblockd_schedule_delayed_work(struct request_queue *q,
2841                        struct delayed_work *dwork, unsigned long delay)
2842{
2843        return queue_delayed_work(kblockd_workqueue, dwork, delay);
2844}
2845EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2846
2847#define PLUG_MAGIC      0x91827364
2848
2849/**
2850 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2851 * @plug:       The &struct blk_plug that needs to be initialized
2852 *
2853 * Description:
2854 *   Tracking blk_plug inside the task_struct will help with auto-flushing the
2855 *   pending I/O should the task end up blocking between blk_start_plug() and
2856 *   blk_finish_plug(). This is important from a performance perspective, but
2857 *   also ensures that we don't deadlock. For instance, if the task is blocking
2858 *   for a memory allocation, memory reclaim could end up wanting to free a
2859 *   page belonging to that request that is currently residing in our private
2860 *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
2861 *   this kind of deadlock.
2862 */
2863void blk_start_plug(struct blk_plug *plug)
2864{
2865        struct task_struct *tsk = current;
2866
2867        plug->magic = PLUG_MAGIC;
2868        INIT_LIST_HEAD(&plug->list);
2869        INIT_LIST_HEAD(&plug->cb_list);
2870
2871        /*
2872         * If this is a nested plug, don't actually assign it. It will be
2873         * flushed on its own.
2874         */
2875        if (!tsk->plug) {
2876                /*
2877                 * Store ordering should not be needed here, since a potential
2878                 * preempt will imply a full memory barrier
2879                 */
2880                tsk->plug = plug;
2881        }
2882}
2883EXPORT_SYMBOL(blk_start_plug);
2884
2885static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2886{
2887        struct request *rqa = container_of(a, struct request, queuelist);
2888        struct request *rqb = container_of(b, struct request, queuelist);
2889
2890        return !(rqa->q < rqb->q ||
2891                (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2892}
2893
2894/*
2895 * If 'from_schedule' is true, then postpone the dispatch of requests
2896 * until a safe kblockd context. We due this to avoid accidental big
2897 * additional stack usage in driver dispatch, in places where the originally
2898 * plugger did not intend it.
2899 */
2900static void queue_unplugged(struct request_queue *q, unsigned int depth,
2901                            bool from_schedule)
2902        __releases(q->queue_lock)
2903{
2904        trace_block_unplug(q, depth, !from_schedule);
2905
2906        if (from_schedule)
2907                blk_run_queue_async(q);
2908        else
2909                __blk_run_queue(q);
2910        spin_unlock(q->queue_lock);
2911}
2912
2913static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2914{
2915        LIST_HEAD(callbacks);
2916
2917        while (!list_empty(&plug->cb_list)) {
2918                list_splice_init(&plug->cb_list, &callbacks);
2919
2920                while (!list_empty(&callbacks)) {
2921                        struct blk_plug_cb *cb = list_first_entry(&callbacks,
2922                                                          struct blk_plug_cb,
2923                                                          list);
2924                        list_del(&cb->list);
2925                        cb->callback(cb, from_schedule);
2926                }
2927        }
2928}
2929
2930struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2931                                      int size)
2932{
2933        struct blk_plug *plug = current->plug;
2934        struct blk_plug_cb *cb;
2935
2936        if (!plug)
2937                return NULL;
2938
2939        list_for_each_entry(cb, &plug->cb_list, list)
2940                if (cb->callback == unplug && cb->data == data)
2941                        return cb;
2942
2943        /* Not currently on the callback list */
2944        BUG_ON(size < sizeof(*cb));
2945        cb = kzalloc(size, GFP_ATOMIC);
2946        if (cb) {
2947                cb->data = data;
2948                cb->callback = unplug;
2949                list_add(&cb->list, &plug->cb_list);
2950        }
2951        return cb;
2952}
2953EXPORT_SYMBOL(blk_check_plugged);
2954
2955void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2956{
2957        struct request_queue *q;
2958        unsigned long flags;
2959        struct request *rq;
2960        LIST_HEAD(list);
2961        unsigned int depth;
2962
2963        BUG_ON(plug->magic != PLUG_MAGIC);
2964
2965        flush_plug_callbacks(plug, from_schedule);
2966        if (list_empty(&plug->list))
2967                return;
2968
2969        list_splice_init(&plug->list, &list);
2970
2971        list_sort(NULL, &list, plug_rq_cmp);
2972
2973        q = NULL;
2974        depth = 0;
2975
2976        /*
2977         * Save and disable interrupts here, to avoid doing it for every
2978         * queue lock we have to take.
2979         */
2980        local_irq_save(flags);
2981        while (!list_empty(&list)) {
2982                rq = list_entry_rq(list.next);
2983                list_del_init(&rq->queuelist);
2984                BUG_ON(!rq->q);
2985                if (rq->q != q) {
2986                        /*
2987                         * This drops the queue lock
2988                         */
2989                        if (q)
2990                                queue_unplugged(q, depth, from_schedule);
2991                        q = rq->q;
2992                        depth = 0;
2993                        spin_lock(q->queue_lock);
2994                }
2995
2996                /*
2997                 * Short-circuit if @q is dead
2998                 */
2999                if (unlikely(blk_queue_dying(q))) {
3000                        __blk_end_request_all(rq, -ENODEV);
3001                        continue;
3002                }
3003
3004                /*
3005                 * rq is already accounted, so use raw insert
3006                 */
3007                if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3008                        __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3009                else
3010                        __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3011
3012                depth++;
3013        }
3014
3015        /*
3016         * This drops the queue lock
3017         */
3018        if (q)
3019                queue_unplugged(q, depth, from_schedule);
3020
3021        local_irq_restore(flags);
3022}
3023
3024void blk_finish_plug(struct blk_plug *plug)
3025{
3026        blk_flush_plug_list(plug, false);
3027
3028        if (plug == current->plug)
3029                current->plug = NULL;
3030}
3031EXPORT_SYMBOL(blk_finish_plug);
3032
3033#ifdef CONFIG_PM_RUNTIME
3034/**
3035 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3036 * @q: the queue of the device
3037 * @dev: the device the queue belongs to
3038 *
3039 * Description:
3040 *    Initialize runtime-PM-related fields for @q and start auto suspend for
3041 *    @dev. Drivers that want to take advantage of request-based runtime PM
3042 *    should call this function after @dev has been initialized, and its
3043 *    request queue @q has been allocated, and runtime PM for it can not happen
3044 *    yet(either due to disabled/forbidden or its usage_count > 0). In most
3045 *    cases, driver should call this function before any I/O has taken place.
3046 *
3047 *    This function takes care of setting up using auto suspend for the device,
3048 *    the autosuspend delay is set to -1 to make runtime suspend impossible
3049 *    until an updated value is either set by user or by driver. Drivers do
3050 *    not need to touch other autosuspend settings.
3051 *
3052 *    The block layer runtime PM is request based, so only works for drivers
3053 *    that use request as their IO unit instead of those directly use bio's.
3054 */
3055void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3056{
3057        q->dev = dev;
3058        q->rpm_status = RPM_ACTIVE;
3059        pm_runtime_set_autosuspend_delay(q->dev, -1);
3060        pm_runtime_use_autosuspend(q->dev);
3061}
3062EXPORT_SYMBOL(blk_pm_runtime_init);
3063
3064/**
3065 * blk_pre_runtime_suspend - Pre runtime suspend check
3066 * @q: the queue of the device
3067 *
3068 * Description:
3069 *    This function will check if runtime suspend is allowed for the device
3070 *    by examining if there are any requests pending in the queue. If there
3071 *    are requests pending, the device can not be runtime suspended; otherwise,
3072 *    the queue's status will be updated to SUSPENDING and the driver can
3073 *    proceed to suspend the device.
3074 *
3075 *    For the not allowed case, we mark last busy for the device so that
3076 *    runtime PM core will try to autosuspend it some time later.
3077 *
3078 *    This function should be called near the start of the device's
3079 *    runtime_suspend callback.
3080 *
3081 * Return:
3082 *    0         - OK to runtime suspend the device
3083 *    -EBUSY    - Device should not be runtime suspended
3084 */
3085int blk_pre_runtime_suspend(struct request_queue *q)
3086{
3087        int ret = 0;
3088
3089        spin_lock_irq(q->queue_lock);
3090        if (q->nr_pending) {
3091                ret = -EBUSY;
3092                pm_runtime_mark_last_busy(q->dev);
3093        } else {
3094                q->rpm_status = RPM_SUSPENDING;
3095        }
3096        spin_unlock_irq(q->queue_lock);
3097        return ret;
3098}
3099EXPORT_SYMBOL(blk_pre_runtime_suspend);
3100
3101/**
3102 * blk_post_runtime_suspend - Post runtime suspend processing
3103 * @q: the queue of the device
3104 * @err: return value of the device's runtime_suspend function
3105 *
3106 * Description:
3107 *    Update the queue's runtime status according to the return value of the
3108 *    device's runtime suspend function and mark last busy for the device so
3109 *    that PM core will try to auto suspend the device at a later time.
3110 *
3111 *    This function should be called near the end of the device's
3112 *    runtime_suspend callback.
3113 */
3114void blk_post_runtime_suspend(struct request_queue *q, int err)
3115{
3116        spin_lock_irq(q->queue_lock);
3117        if (!err) {
3118                q->rpm_status = RPM_SUSPENDED;
3119        } else {
3120                q->rpm_status = RPM_ACTIVE;
3121                pm_runtime_mark_last_busy(q->dev);
3122        }
3123        spin_unlock_irq(q->queue_lock);
3124}
3125EXPORT_SYMBOL(blk_post_runtime_suspend);
3126
3127/**
3128 * blk_pre_runtime_resume - Pre runtime resume processing
3129 * @q: the queue of the device
3130 *
3131 * Description:
3132 *    Update the queue's runtime status to RESUMING in preparation for the
3133 *    runtime resume of the device.
3134 *
3135 *    This function should be called near the start of the device's
3136 *    runtime_resume callback.
3137 */
3138void blk_pre_runtime_resume(struct request_queue *q)
3139{
3140        spin_lock_irq(q->queue_lock);
3141        q->rpm_status = RPM_RESUMING;
3142        spin_unlock_irq(q->queue_lock);
3143}
3144EXPORT_SYMBOL(blk_pre_runtime_resume);
3145
3146/**
3147 * blk_post_runtime_resume - Post runtime resume processing
3148 * @q: the queue of the device
3149 * @err: return value of the device's runtime_resume function
3150 *
3151 * Description:
3152 *    Update the queue's runtime status according to the return value of the
3153 *    device's runtime_resume function. If it is successfully resumed, process
3154 *    the requests that are queued into the device's queue when it is resuming
3155 *    and then mark last busy and initiate autosuspend for it.
3156 *
3157 *    This function should be called near the end of the device's
3158 *    runtime_resume callback.
3159 */
3160void blk_post_runtime_resume(struct request_queue *q, int err)
3161{
3162        spin_lock_irq(q->queue_lock);
3163        if (!err) {
3164                q->rpm_status = RPM_ACTIVE;
3165                __blk_run_queue(q);
3166                pm_runtime_mark_last_busy(q->dev);
3167                pm_request_autosuspend(q->dev);
3168        } else {
3169                q->rpm_status = RPM_SUSPENDED;
3170        }
3171        spin_unlock_irq(q->queue_lock);
3172}
3173EXPORT_SYMBOL(blk_post_runtime_resume);
3174#endif
3175
3176int __init blk_dev_init(void)
3177{
3178        BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3179                        sizeof(((struct request *)0)->cmd_flags));
3180
3181        /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3182        kblockd_workqueue = alloc_workqueue("kblockd",
3183                                            WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3184        if (!kblockd_workqueue)
3185                panic("Failed to create kblockd\n");
3186
3187        request_cachep = kmem_cache_create("blkdev_requests",
3188                        sizeof(struct request), 0, SLAB_PANIC, NULL);
3189
3190        blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3191                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3192
3193        return 0;
3194}
3195
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.