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