linux/block/blk-core.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 1991, 1992 Linus Torvalds
   4 * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
   5 * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
   6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
   7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
   8 *      -  July2000
   9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
  10 */
  11
  12/*
  13 * This handles all read/write requests to block devices
  14 */
  15#include <linux/kernel.h>
  16#include <linux/module.h>
  17#include <linux/backing-dev.h>
  18#include <linux/bio.h>
  19#include <linux/blkdev.h>
  20#include <linux/blk-mq.h>
  21#include <linux/blk-pm.h>
  22#include <linux/highmem.h>
  23#include <linux/mm.h>
  24#include <linux/pagemap.h>
  25#include <linux/kernel_stat.h>
  26#include <linux/string.h>
  27#include <linux/init.h>
  28#include <linux/completion.h>
  29#include <linux/slab.h>
  30#include <linux/swap.h>
  31#include <linux/writeback.h>
  32#include <linux/task_io_accounting_ops.h>
  33#include <linux/fault-inject.h>
  34#include <linux/list_sort.h>
  35#include <linux/delay.h>
  36#include <linux/ratelimit.h>
  37#include <linux/pm_runtime.h>
  38#include <linux/blk-cgroup.h>
  39#include <linux/t10-pi.h>
  40#include <linux/debugfs.h>
  41#include <linux/bpf.h>
  42#include <linux/psi.h>
  43#include <linux/sched/sysctl.h>
  44#include <linux/blk-crypto.h>
  45
  46#define CREATE_TRACE_POINTS
  47#include <trace/events/block.h>
  48
  49#include "blk.h"
  50#include "blk-mq.h"
  51#include "blk-mq-sched.h"
  52#include "blk-pm.h"
  53#include "blk-rq-qos.h"
  54
  55struct dentry *blk_debugfs_root;
  56
  57EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
  58EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
  59EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
  60EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
  61EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
  62EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
  63
  64DEFINE_IDA(blk_queue_ida);
  65
  66/*
  67 * For queue allocation
  68 */
  69struct kmem_cache *blk_requestq_cachep;
  70
  71/*
  72 * Controlling structure to kblockd
  73 */
  74static struct workqueue_struct *kblockd_workqueue;
  75
  76/**
  77 * blk_queue_flag_set - atomically set a queue flag
  78 * @flag: flag to be set
  79 * @q: request queue
  80 */
  81void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
  82{
  83        set_bit(flag, &q->queue_flags);
  84}
  85EXPORT_SYMBOL(blk_queue_flag_set);
  86
  87/**
  88 * blk_queue_flag_clear - atomically clear a queue flag
  89 * @flag: flag to be cleared
  90 * @q: request queue
  91 */
  92void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
  93{
  94        clear_bit(flag, &q->queue_flags);
  95}
  96EXPORT_SYMBOL(blk_queue_flag_clear);
  97
  98/**
  99 * blk_queue_flag_test_and_set - atomically test and set a queue flag
 100 * @flag: flag to be set
 101 * @q: request queue
 102 *
 103 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
 104 * the flag was already set.
 105 */
 106bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
 107{
 108        return test_and_set_bit(flag, &q->queue_flags);
 109}
 110EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
 111
 112void blk_rq_init(struct request_queue *q, struct request *rq)
 113{
 114        memset(rq, 0, sizeof(*rq));
 115
 116        INIT_LIST_HEAD(&rq->queuelist);
 117        rq->q = q;
 118        rq->__sector = (sector_t) -1;
 119        INIT_HLIST_NODE(&rq->hash);
 120        RB_CLEAR_NODE(&rq->rb_node);
 121        rq->tag = BLK_MQ_NO_TAG;
 122        rq->internal_tag = BLK_MQ_NO_TAG;
 123        rq->start_time_ns = ktime_get_ns();
 124        rq->part = NULL;
 125        blk_crypto_rq_set_defaults(rq);
 126}
 127EXPORT_SYMBOL(blk_rq_init);
 128
 129#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
 130static const char *const blk_op_name[] = {
 131        REQ_OP_NAME(READ),
 132        REQ_OP_NAME(WRITE),
 133        REQ_OP_NAME(FLUSH),
 134        REQ_OP_NAME(DISCARD),
 135        REQ_OP_NAME(SECURE_ERASE),
 136        REQ_OP_NAME(ZONE_RESET),
 137        REQ_OP_NAME(ZONE_RESET_ALL),
 138        REQ_OP_NAME(ZONE_OPEN),
 139        REQ_OP_NAME(ZONE_CLOSE),
 140        REQ_OP_NAME(ZONE_FINISH),
 141        REQ_OP_NAME(ZONE_APPEND),
 142        REQ_OP_NAME(WRITE_SAME),
 143        REQ_OP_NAME(WRITE_ZEROES),
 144        REQ_OP_NAME(DRV_IN),
 145        REQ_OP_NAME(DRV_OUT),
 146};
 147#undef REQ_OP_NAME
 148
 149/**
 150 * blk_op_str - Return string XXX in the REQ_OP_XXX.
 151 * @op: REQ_OP_XXX.
 152 *
 153 * Description: Centralize block layer function to convert REQ_OP_XXX into
 154 * string format. Useful in the debugging and tracing bio or request. For
 155 * invalid REQ_OP_XXX it returns string "UNKNOWN".
 156 */
 157inline const char *blk_op_str(unsigned int op)
 158{
 159        const char *op_str = "UNKNOWN";
 160
 161        if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
 162                op_str = blk_op_name[op];
 163
 164        return op_str;
 165}
 166EXPORT_SYMBOL_GPL(blk_op_str);
 167
 168static const struct {
 169        int             errno;
 170        const char      *name;
 171} blk_errors[] = {
 172        [BLK_STS_OK]            = { 0,          "" },
 173        [BLK_STS_NOTSUPP]       = { -EOPNOTSUPP, "operation not supported" },
 174        [BLK_STS_TIMEOUT]       = { -ETIMEDOUT, "timeout" },
 175        [BLK_STS_NOSPC]         = { -ENOSPC,    "critical space allocation" },
 176        [BLK_STS_TRANSPORT]     = { -ENOLINK,   "recoverable transport" },
 177        [BLK_STS_TARGET]        = { -EREMOTEIO, "critical target" },
 178        [BLK_STS_NEXUS]         = { -EBADE,     "critical nexus" },
 179        [BLK_STS_MEDIUM]        = { -ENODATA,   "critical medium" },
 180        [BLK_STS_PROTECTION]    = { -EILSEQ,    "protection" },
 181        [BLK_STS_RESOURCE]      = { -ENOMEM,    "kernel resource" },
 182        [BLK_STS_DEV_RESOURCE]  = { -EBUSY,     "device resource" },
 183        [BLK_STS_AGAIN]         = { -EAGAIN,    "nonblocking retry" },
 184
 185        /* device mapper special case, should not leak out: */
 186        [BLK_STS_DM_REQUEUE]    = { -EREMCHG, "dm internal retry" },
 187
 188        /* zone device specific errors */
 189        [BLK_STS_ZONE_OPEN_RESOURCE]    = { -ETOOMANYREFS, "open zones exceeded" },
 190        [BLK_STS_ZONE_ACTIVE_RESOURCE]  = { -EOVERFLOW, "active zones exceeded" },
 191
 192        /* everything else not covered above: */
 193        [BLK_STS_IOERR]         = { -EIO,       "I/O" },
 194};
 195
 196blk_status_t errno_to_blk_status(int errno)
 197{
 198        int i;
 199
 200        for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
 201                if (blk_errors[i].errno == errno)
 202                        return (__force blk_status_t)i;
 203        }
 204
 205        return BLK_STS_IOERR;
 206}
 207EXPORT_SYMBOL_GPL(errno_to_blk_status);
 208
 209int blk_status_to_errno(blk_status_t status)
 210{
 211        int idx = (__force int)status;
 212
 213        if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
 214                return -EIO;
 215        return blk_errors[idx].errno;
 216}
 217EXPORT_SYMBOL_GPL(blk_status_to_errno);
 218
 219static void print_req_error(struct request *req, blk_status_t status,
 220                const char *caller)
 221{
 222        int idx = (__force int)status;
 223
 224        if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
 225                return;
 226
 227        printk_ratelimited(KERN_ERR
 228                "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
 229                "phys_seg %u prio class %u\n",
 230                caller, blk_errors[idx].name,
 231                req->rq_disk ? req->rq_disk->disk_name : "?",
 232                blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
 233                req->cmd_flags & ~REQ_OP_MASK,
 234                req->nr_phys_segments,
 235                IOPRIO_PRIO_CLASS(req->ioprio));
 236}
 237
 238static void req_bio_endio(struct request *rq, struct bio *bio,
 239                          unsigned int nbytes, blk_status_t error)
 240{
 241        if (error)
 242                bio->bi_status = error;
 243
 244        if (unlikely(rq->rq_flags & RQF_QUIET))
 245                bio_set_flag(bio, BIO_QUIET);
 246
 247        bio_advance(bio, nbytes);
 248
 249        if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
 250                /*
 251                 * Partial zone append completions cannot be supported as the
 252                 * BIO fragments may end up not being written sequentially.
 253                 */
 254                if (bio->bi_iter.bi_size)
 255                        bio->bi_status = BLK_STS_IOERR;
 256                else
 257                        bio->bi_iter.bi_sector = rq->__sector;
 258        }
 259
 260        /* don't actually finish bio if it's part of flush sequence */
 261        if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
 262                bio_endio(bio);
 263}
 264
 265void blk_dump_rq_flags(struct request *rq, char *msg)
 266{
 267        printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
 268                rq->rq_disk ? rq->rq_disk->disk_name : "?",
 269                (unsigned long long) rq->cmd_flags);
 270
 271        printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
 272               (unsigned long long)blk_rq_pos(rq),
 273               blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
 274        printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
 275               rq->bio, rq->biotail, blk_rq_bytes(rq));
 276}
 277EXPORT_SYMBOL(blk_dump_rq_flags);
 278
 279/**
 280 * blk_sync_queue - cancel any pending callbacks on a queue
 281 * @q: the queue
 282 *
 283 * Description:
 284 *     The block layer may perform asynchronous callback activity
 285 *     on a queue, such as calling the unplug function after a timeout.
 286 *     A block device may call blk_sync_queue to ensure that any
 287 *     such activity is cancelled, thus allowing it to release resources
 288 *     that the callbacks might use. The caller must already have made sure
 289 *     that its ->submit_bio will not re-add plugging prior to calling
 290 *     this function.
 291 *
 292 *     This function does not cancel any asynchronous activity arising
 293 *     out of elevator or throttling code. That would require elevator_exit()
 294 *     and blkcg_exit_queue() to be called with queue lock initialized.
 295 *
 296 */
 297void blk_sync_queue(struct request_queue *q)
 298{
 299        del_timer_sync(&q->timeout);
 300        cancel_work_sync(&q->timeout_work);
 301}
 302EXPORT_SYMBOL(blk_sync_queue);
 303
 304/**
 305 * blk_set_pm_only - increment pm_only counter
 306 * @q: request queue pointer
 307 */
 308void blk_set_pm_only(struct request_queue *q)
 309{
 310        atomic_inc(&q->pm_only);
 311}
 312EXPORT_SYMBOL_GPL(blk_set_pm_only);
 313
 314void blk_clear_pm_only(struct request_queue *q)
 315{
 316        int pm_only;
 317
 318        pm_only = atomic_dec_return(&q->pm_only);
 319        WARN_ON_ONCE(pm_only < 0);
 320        if (pm_only == 0)
 321                wake_up_all(&q->mq_freeze_wq);
 322}
 323EXPORT_SYMBOL_GPL(blk_clear_pm_only);
 324
 325/**
 326 * blk_put_queue - decrement the request_queue refcount
 327 * @q: the request_queue structure to decrement the refcount for
 328 *
 329 * Decrements the refcount of the request_queue kobject. When this reaches 0
 330 * we'll have blk_release_queue() called.
 331 *
 332 * Context: Any context, but the last reference must not be dropped from
 333 *          atomic context.
 334 */
 335void blk_put_queue(struct request_queue *q)
 336{
 337        kobject_put(&q->kobj);
 338}
 339EXPORT_SYMBOL(blk_put_queue);
 340
 341void blk_set_queue_dying(struct request_queue *q)
 342{
 343        blk_queue_flag_set(QUEUE_FLAG_DYING, q);
 344
 345        /*
 346         * When queue DYING flag is set, we need to block new req
 347         * entering queue, so we call blk_freeze_queue_start() to
 348         * prevent I/O from crossing blk_queue_enter().
 349         */
 350        blk_freeze_queue_start(q);
 351
 352        if (queue_is_mq(q))
 353                blk_mq_wake_waiters(q);
 354
 355        /* Make blk_queue_enter() reexamine the DYING flag. */
 356        wake_up_all(&q->mq_freeze_wq);
 357}
 358EXPORT_SYMBOL_GPL(blk_set_queue_dying);
 359
 360/**
 361 * blk_cleanup_queue - shutdown a request queue
 362 * @q: request queue to shutdown
 363 *
 364 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
 365 * put it.  All future requests will be failed immediately with -ENODEV.
 366 *
 367 * Context: can sleep
 368 */
 369void blk_cleanup_queue(struct request_queue *q)
 370{
 371        /* cannot be called from atomic context */
 372        might_sleep();
 373
 374        WARN_ON_ONCE(blk_queue_registered(q));
 375
 376        /* mark @q DYING, no new request or merges will be allowed afterwards */
 377        blk_set_queue_dying(q);
 378
 379        blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 380        blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 381
 382        /*
 383         * Drain all requests queued before DYING marking. Set DEAD flag to
 384         * prevent that blk_mq_run_hw_queues() accesses the hardware queues
 385         * after draining finished.
 386         */
 387        blk_freeze_queue(q);
 388
 389        rq_qos_exit(q);
 390
 391        blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
 392
 393        /* for synchronous bio-based driver finish in-flight integrity i/o */
 394        blk_flush_integrity();
 395
 396        /* @q won't process any more request, flush async actions */
 397        del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
 398        blk_sync_queue(q);
 399
 400        if (queue_is_mq(q))
 401                blk_mq_exit_queue(q);
 402
 403        /*
 404         * In theory, request pool of sched_tags belongs to request queue.
 405         * However, the current implementation requires tag_set for freeing
 406         * requests, so free the pool now.
 407         *
 408         * Queue has become frozen, there can't be any in-queue requests, so
 409         * it is safe to free requests now.
 410         */
 411        mutex_lock(&q->sysfs_lock);
 412        if (q->elevator)
 413                blk_mq_sched_free_requests(q);
 414        mutex_unlock(&q->sysfs_lock);
 415
 416        percpu_ref_exit(&q->q_usage_counter);
 417
 418        /* @q is and will stay empty, shutdown and put */
 419        blk_put_queue(q);
 420}
 421EXPORT_SYMBOL(blk_cleanup_queue);
 422
 423/**
 424 * blk_queue_enter() - try to increase q->q_usage_counter
 425 * @q: request queue pointer
 426 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
 427 */
 428int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 429{
 430        const bool pm = flags & BLK_MQ_REQ_PM;
 431
 432        while (true) {
 433                bool success = false;
 434
 435                rcu_read_lock();
 436                if (percpu_ref_tryget_live(&q->q_usage_counter)) {
 437                        /*
 438                         * The code that increments the pm_only counter is
 439                         * responsible for ensuring that that counter is
 440                         * globally visible before the queue is unfrozen.
 441                         */
 442                        if ((pm && queue_rpm_status(q) != RPM_SUSPENDED) ||
 443                            !blk_queue_pm_only(q)) {
 444                                success = true;
 445                        } else {
 446                                percpu_ref_put(&q->q_usage_counter);
 447                        }
 448                }
 449                rcu_read_unlock();
 450
 451                if (success)
 452                        return 0;
 453
 454                if (flags & BLK_MQ_REQ_NOWAIT)
 455                        return -EBUSY;
 456
 457                /*
 458                 * read pair of barrier in blk_freeze_queue_start(),
 459                 * we need to order reading __PERCPU_REF_DEAD flag of
 460                 * .q_usage_counter and reading .mq_freeze_depth or
 461                 * queue dying flag, otherwise the following wait may
 462                 * never return if the two reads are reordered.
 463                 */
 464                smp_rmb();
 465
 466                wait_event(q->mq_freeze_wq,
 467                           (!q->mq_freeze_depth &&
 468                            blk_pm_resume_queue(pm, q)) ||
 469                           blk_queue_dying(q));
 470                if (blk_queue_dying(q))
 471                        return -ENODEV;
 472        }
 473}
 474
 475static inline int bio_queue_enter(struct bio *bio)
 476{
 477        struct request_queue *q = bio->bi_bdev->bd_disk->queue;
 478        bool nowait = bio->bi_opf & REQ_NOWAIT;
 479        int ret;
 480
 481        ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
 482        if (unlikely(ret)) {
 483                if (nowait && !blk_queue_dying(q))
 484                        bio_wouldblock_error(bio);
 485                else
 486                        bio_io_error(bio);
 487        }
 488
 489        return ret;
 490}
 491
 492void blk_queue_exit(struct request_queue *q)
 493{
 494        percpu_ref_put(&q->q_usage_counter);
 495}
 496
 497static void blk_queue_usage_counter_release(struct percpu_ref *ref)
 498{
 499        struct request_queue *q =
 500                container_of(ref, struct request_queue, q_usage_counter);
 501
 502        wake_up_all(&q->mq_freeze_wq);
 503}
 504
 505static void blk_rq_timed_out_timer(struct timer_list *t)
 506{
 507        struct request_queue *q = from_timer(q, t, timeout);
 508
 509        kblockd_schedule_work(&q->timeout_work);
 510}
 511
 512static void blk_timeout_work(struct work_struct *work)
 513{
 514}
 515
 516struct request_queue *blk_alloc_queue(int node_id)
 517{
 518        struct request_queue *q;
 519        int ret;
 520
 521        q = kmem_cache_alloc_node(blk_requestq_cachep,
 522                                GFP_KERNEL | __GFP_ZERO, node_id);
 523        if (!q)
 524                return NULL;
 525
 526        q->last_merge = NULL;
 527
 528        q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
 529        if (q->id < 0)
 530                goto fail_q;
 531
 532        ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
 533        if (ret)
 534                goto fail_id;
 535
 536        q->backing_dev_info = bdi_alloc(node_id);
 537        if (!q->backing_dev_info)
 538                goto fail_split;
 539
 540        q->stats = blk_alloc_queue_stats();
 541        if (!q->stats)
 542                goto fail_stats;
 543
 544        q->node = node_id;
 545
 546        atomic_set(&q->nr_active_requests_shared_sbitmap, 0);
 547
 548        timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
 549                    laptop_mode_timer_fn, 0);
 550        timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
 551        INIT_WORK(&q->timeout_work, blk_timeout_work);
 552        INIT_LIST_HEAD(&q->icq_list);
 553#ifdef CONFIG_BLK_CGROUP
 554        INIT_LIST_HEAD(&q->blkg_list);
 555#endif
 556
 557        kobject_init(&q->kobj, &blk_queue_ktype);
 558
 559        mutex_init(&q->debugfs_mutex);
 560        mutex_init(&q->sysfs_lock);
 561        mutex_init(&q->sysfs_dir_lock);
 562        spin_lock_init(&q->queue_lock);
 563
 564        init_waitqueue_head(&q->mq_freeze_wq);
 565        mutex_init(&q->mq_freeze_lock);
 566
 567        /*
 568         * Init percpu_ref in atomic mode so that it's faster to shutdown.
 569         * See blk_register_queue() for details.
 570         */
 571        if (percpu_ref_init(&q->q_usage_counter,
 572                                blk_queue_usage_counter_release,
 573                                PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
 574                goto fail_bdi;
 575
 576        if (blkcg_init_queue(q))
 577                goto fail_ref;
 578
 579        blk_queue_dma_alignment(q, 511);
 580        blk_set_default_limits(&q->limits);
 581        q->nr_requests = BLKDEV_MAX_RQ;
 582
 583        return q;
 584
 585fail_ref:
 586        percpu_ref_exit(&q->q_usage_counter);
 587fail_bdi:
 588        blk_free_queue_stats(q->stats);
 589fail_stats:
 590        bdi_put(q->backing_dev_info);
 591fail_split:
 592        bioset_exit(&q->bio_split);
 593fail_id:
 594        ida_simple_remove(&blk_queue_ida, q->id);
 595fail_q:
 596        kmem_cache_free(blk_requestq_cachep, q);
 597        return NULL;
 598}
 599
 600/**
 601 * blk_get_queue - increment the request_queue refcount
 602 * @q: the request_queue structure to increment the refcount for
 603 *
 604 * Increment the refcount of the request_queue kobject.
 605 *
 606 * Context: Any context.
 607 */
 608bool blk_get_queue(struct request_queue *q)
 609{
 610        if (likely(!blk_queue_dying(q))) {
 611                __blk_get_queue(q);
 612                return true;
 613        }
 614
 615        return false;
 616}
 617EXPORT_SYMBOL(blk_get_queue);
 618
 619/**
 620 * blk_get_request - allocate a request
 621 * @q: request queue to allocate a request for
 622 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
 623 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
 624 */
 625struct request *blk_get_request(struct request_queue *q, unsigned int op,
 626                                blk_mq_req_flags_t flags)
 627{
 628        struct request *req;
 629
 630        WARN_ON_ONCE(op & REQ_NOWAIT);
 631        WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM));
 632
 633        req = blk_mq_alloc_request(q, op, flags);
 634        if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
 635                q->mq_ops->initialize_rq_fn(req);
 636
 637        return req;
 638}
 639EXPORT_SYMBOL(blk_get_request);
 640
 641void blk_put_request(struct request *req)
 642{
 643        blk_mq_free_request(req);
 644}
 645EXPORT_SYMBOL(blk_put_request);
 646
 647static void handle_bad_sector(struct bio *bio, sector_t maxsector)
 648{
 649        char b[BDEVNAME_SIZE];
 650
 651        pr_info_ratelimited("attempt to access beyond end of device\n"
 652                            "%s: rw=%d, want=%llu, limit=%llu\n",
 653                            bio_devname(bio, b), bio->bi_opf,
 654                            bio_end_sector(bio), maxsector);
 655}
 656
 657#ifdef CONFIG_FAIL_MAKE_REQUEST
 658
 659static DECLARE_FAULT_ATTR(fail_make_request);
 660
 661static int __init setup_fail_make_request(char *str)
 662{
 663        return setup_fault_attr(&fail_make_request, str);
 664}
 665__setup("fail_make_request=", setup_fail_make_request);
 666
 667static bool should_fail_request(struct block_device *part, unsigned int bytes)
 668{
 669        return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
 670}
 671
 672static int __init fail_make_request_debugfs(void)
 673{
 674        struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
 675                                                NULL, &fail_make_request);
 676
 677        return PTR_ERR_OR_ZERO(dir);
 678}
 679
 680late_initcall(fail_make_request_debugfs);
 681
 682#else /* CONFIG_FAIL_MAKE_REQUEST */
 683
 684static inline bool should_fail_request(struct block_device *part,
 685                                        unsigned int bytes)
 686{
 687        return false;
 688}
 689
 690#endif /* CONFIG_FAIL_MAKE_REQUEST */
 691
 692static inline bool bio_check_ro(struct bio *bio)
 693{
 694        if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
 695                char b[BDEVNAME_SIZE];
 696
 697                if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
 698                        return false;
 699
 700                WARN_ONCE(1,
 701                       "Trying to write to read-only block-device %s (partno %d)\n",
 702                        bio_devname(bio, b), bio->bi_bdev->bd_partno);
 703                /* Older lvm-tools actually trigger this */
 704                return false;
 705        }
 706
 707        return false;
 708}
 709
 710static noinline int should_fail_bio(struct bio *bio)
 711{
 712        if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
 713                return -EIO;
 714        return 0;
 715}
 716ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
 717
 718/*
 719 * Check whether this bio extends beyond the end of the device or partition.
 720 * This may well happen - the kernel calls bread() without checking the size of
 721 * the device, e.g., when mounting a file system.
 722 */
 723static inline int bio_check_eod(struct bio *bio)
 724{
 725        sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
 726        unsigned int nr_sectors = bio_sectors(bio);
 727
 728        if (nr_sectors && maxsector &&
 729            (nr_sectors > maxsector ||
 730             bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
 731                handle_bad_sector(bio, maxsector);
 732                return -EIO;
 733        }
 734        return 0;
 735}
 736
 737/*
 738 * Remap block n of partition p to block n+start(p) of the disk.
 739 */
 740static int blk_partition_remap(struct bio *bio)
 741{
 742        struct block_device *p = bio->bi_bdev;
 743
 744        if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
 745                return -EIO;
 746        if (bio_sectors(bio)) {
 747                bio->bi_iter.bi_sector += p->bd_start_sect;
 748                trace_block_bio_remap(bio, p->bd_dev,
 749                                      bio->bi_iter.bi_sector -
 750                                      p->bd_start_sect);
 751        }
 752        bio_set_flag(bio, BIO_REMAPPED);
 753        return 0;
 754}
 755
 756/*
 757 * Check write append to a zoned block device.
 758 */
 759static inline blk_status_t blk_check_zone_append(struct request_queue *q,
 760                                                 struct bio *bio)
 761{
 762        sector_t pos = bio->bi_iter.bi_sector;
 763        int nr_sectors = bio_sectors(bio);
 764
 765        /* Only applicable to zoned block devices */
 766        if (!blk_queue_is_zoned(q))
 767                return BLK_STS_NOTSUPP;
 768
 769        /* The bio sector must point to the start of a sequential zone */
 770        if (pos & (blk_queue_zone_sectors(q) - 1) ||
 771            !blk_queue_zone_is_seq(q, pos))
 772                return BLK_STS_IOERR;
 773
 774        /*
 775         * Not allowed to cross zone boundaries. Otherwise, the BIO will be
 776         * split and could result in non-contiguous sectors being written in
 777         * different zones.
 778         */
 779        if (nr_sectors > q->limits.chunk_sectors)
 780                return BLK_STS_IOERR;
 781
 782        /* Make sure the BIO is small enough and will not get split */
 783        if (nr_sectors > q->limits.max_zone_append_sectors)
 784                return BLK_STS_IOERR;
 785
 786        bio->bi_opf |= REQ_NOMERGE;
 787
 788        return BLK_STS_OK;
 789}
 790
 791static noinline_for_stack bool submit_bio_checks(struct bio *bio)
 792{
 793        struct block_device *bdev = bio->bi_bdev;
 794        struct request_queue *q = bdev->bd_disk->queue;
 795        blk_status_t status = BLK_STS_IOERR;
 796        struct blk_plug *plug;
 797
 798        might_sleep();
 799
 800        plug = blk_mq_plug(q, bio);
 801        if (plug && plug->nowait)
 802                bio->bi_opf |= REQ_NOWAIT;
 803
 804        /*
 805         * For a REQ_NOWAIT based request, return -EOPNOTSUPP
 806         * if queue does not support NOWAIT.
 807         */
 808        if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
 809                goto not_supported;
 810
 811        if (should_fail_bio(bio))
 812                goto end_io;
 813        if (unlikely(bio_check_ro(bio)))
 814                goto end_io;
 815        if (!bio_flagged(bio, BIO_REMAPPED)) {
 816                if (unlikely(bio_check_eod(bio)))
 817                        goto end_io;
 818                if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
 819                        goto end_io;
 820        }
 821
 822        /*
 823         * Filter flush bio's early so that bio based drivers without flush
 824         * support don't have to worry about them.
 825         */
 826        if (op_is_flush(bio->bi_opf) &&
 827            !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
 828                bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
 829                if (!bio_sectors(bio)) {
 830                        status = BLK_STS_OK;
 831                        goto end_io;
 832                }
 833        }
 834
 835        if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
 836                bio->bi_opf &= ~REQ_HIPRI;
 837
 838        switch (bio_op(bio)) {
 839        case REQ_OP_DISCARD:
 840                if (!blk_queue_discard(q))
 841                        goto not_supported;
 842                break;
 843        case REQ_OP_SECURE_ERASE:
 844                if (!blk_queue_secure_erase(q))
 845                        goto not_supported;
 846                break;
 847        case REQ_OP_WRITE_SAME:
 848                if (!q->limits.max_write_same_sectors)
 849                        goto not_supported;
 850                break;
 851        case REQ_OP_ZONE_APPEND:
 852                status = blk_check_zone_append(q, bio);
 853                if (status != BLK_STS_OK)
 854                        goto end_io;
 855                break;
 856        case REQ_OP_ZONE_RESET:
 857        case REQ_OP_ZONE_OPEN:
 858        case REQ_OP_ZONE_CLOSE:
 859        case REQ_OP_ZONE_FINISH:
 860                if (!blk_queue_is_zoned(q))
 861                        goto not_supported;
 862                break;
 863        case REQ_OP_ZONE_RESET_ALL:
 864                if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
 865                        goto not_supported;
 866                break;
 867        case REQ_OP_WRITE_ZEROES:
 868                if (!q->limits.max_write_zeroes_sectors)
 869                        goto not_supported;
 870                break;
 871        default:
 872                break;
 873        }
 874
 875        /*
 876         * Various block parts want %current->io_context, so allocate it up
 877         * front rather than dealing with lots of pain to allocate it only
 878         * where needed. This may fail and the block layer knows how to live
 879         * with it.
 880         */
 881        if (unlikely(!current->io_context))
 882                create_task_io_context(current, GFP_ATOMIC, q->node);
 883
 884        if (blk_throtl_bio(bio)) {
 885                blkcg_bio_issue_init(bio);
 886                return false;
 887        }
 888
 889        blk_cgroup_bio_start(bio);
 890        blkcg_bio_issue_init(bio);
 891
 892        if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
 893                trace_block_bio_queue(bio);
 894                /* Now that enqueuing has been traced, we need to trace
 895                 * completion as well.
 896                 */
 897                bio_set_flag(bio, BIO_TRACE_COMPLETION);
 898        }
 899        return true;
 900
 901not_supported:
 902        status = BLK_STS_NOTSUPP;
 903end_io:
 904        bio->bi_status = status;
 905        bio_endio(bio);
 906        return false;
 907}
 908
 909static blk_qc_t __submit_bio(struct bio *bio)
 910{
 911        struct gendisk *disk = bio->bi_bdev->bd_disk;
 912        blk_qc_t ret = BLK_QC_T_NONE;
 913
 914        if (blk_crypto_bio_prep(&bio)) {
 915                if (!disk->fops->submit_bio)
 916                        return blk_mq_submit_bio(bio);
 917                ret = disk->fops->submit_bio(bio);
 918        }
 919        blk_queue_exit(disk->queue);
 920        return ret;
 921}
 922
 923/*
 924 * The loop in this function may be a bit non-obvious, and so deserves some
 925 * explanation:
 926 *
 927 *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
 928 *    that), so we have a list with a single bio.
 929 *  - We pretend that we have just taken it off a longer list, so we assign
 930 *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
 931 *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
 932 *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
 933 *    non-NULL value in bio_list and re-enter the loop from the top.
 934 *  - In this case we really did just take the bio of the top of the list (no
 935 *    pretending) and so remove it from bio_list, and call into ->submit_bio()
 936 *    again.
 937 *
 938 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
 939 * bio_list_on_stack[1] contains bios that were submitted before the current
 940 *      ->submit_bio_bio, but that haven't been processed yet.
 941 */
 942static blk_qc_t __submit_bio_noacct(struct bio *bio)
 943{
 944        struct bio_list bio_list_on_stack[2];
 945        blk_qc_t ret = BLK_QC_T_NONE;
 946
 947        BUG_ON(bio->bi_next);
 948
 949        bio_list_init(&bio_list_on_stack[0]);
 950        current->bio_list = bio_list_on_stack;
 951
 952        do {
 953                struct request_queue *q = bio->bi_bdev->bd_disk->queue;
 954                struct bio_list lower, same;
 955
 956                if (unlikely(bio_queue_enter(bio) != 0))
 957                        continue;
 958
 959                /*
 960                 * Create a fresh bio_list for all subordinate requests.
 961                 */
 962                bio_list_on_stack[1] = bio_list_on_stack[0];
 963                bio_list_init(&bio_list_on_stack[0]);
 964
 965                ret = __submit_bio(bio);
 966
 967                /*
 968                 * Sort new bios into those for a lower level and those for the
 969                 * same level.
 970                 */
 971                bio_list_init(&lower);
 972                bio_list_init(&same);
 973                while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
 974                        if (q == bio->bi_bdev->bd_disk->queue)
 975                                bio_list_add(&same, bio);
 976                        else
 977                                bio_list_add(&lower, bio);
 978
 979                /*
 980                 * Now assemble so we handle the lowest level first.
 981                 */
 982                bio_list_merge(&bio_list_on_stack[0], &lower);
 983                bio_list_merge(&bio_list_on_stack[0], &same);
 984                bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
 985        } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
 986
 987        current->bio_list = NULL;
 988        return ret;
 989}
 990
 991static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
 992{
 993        struct bio_list bio_list[2] = { };
 994        blk_qc_t ret = BLK_QC_T_NONE;
 995
 996        current->bio_list = bio_list;
 997
 998        do {
 999                struct gendisk *disk = bio->bi_bdev->bd_disk;
1000
1001                if (unlikely(bio_queue_enter(bio) != 0))
1002                        continue;
1003
1004                if (!blk_crypto_bio_prep(&bio)) {
1005                        blk_queue_exit(disk->queue);
1006                        ret = BLK_QC_T_NONE;
1007                        continue;
1008                }
1009
1010                ret = blk_mq_submit_bio(bio);
1011        } while ((bio = bio_list_pop(&bio_list[0])));
1012
1013        current->bio_list = NULL;
1014        return ret;
1015}
1016
1017/**
1018 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1019 * @bio:  The bio describing the location in memory and on the device.
1020 *
1021 * This is a version of submit_bio() that shall only be used for I/O that is
1022 * resubmitted to lower level drivers by stacking block drivers.  All file
1023 * systems and other upper level users of the block layer should use
1024 * submit_bio() instead.
1025 */
1026blk_qc_t submit_bio_noacct(struct bio *bio)
1027{
1028        if (!submit_bio_checks(bio))
1029                return BLK_QC_T_NONE;
1030
1031        /*
1032         * We only want one ->submit_bio to be active at a time, else stack
1033         * usage with stacked devices could be a problem.  Use current->bio_list
1034         * to collect a list of requests submited by a ->submit_bio method while
1035         * it is active, and then process them after it returned.
1036         */
1037        if (current->bio_list) {
1038                bio_list_add(&current->bio_list[0], bio);
1039                return BLK_QC_T_NONE;
1040        }
1041
1042        if (!bio->bi_bdev->bd_disk->fops->submit_bio)
1043                return __submit_bio_noacct_mq(bio);
1044        return __submit_bio_noacct(bio);
1045}
1046EXPORT_SYMBOL(submit_bio_noacct);
1047
1048/**
1049 * submit_bio - submit a bio to the block device layer for I/O
1050 * @bio: The &struct bio which describes the I/O
1051 *
1052 * submit_bio() is used to submit I/O requests to block devices.  It is passed a
1053 * fully set up &struct bio that describes the I/O that needs to be done.  The
1054 * bio will be send to the device described by the bi_bdev field.
1055 *
1056 * The success/failure status of the request, along with notification of
1057 * completion, is delivered asynchronously through the ->bi_end_io() callback
1058 * in @bio.  The bio must NOT be touched by thecaller until ->bi_end_io() has
1059 * been called.
1060 */
1061blk_qc_t submit_bio(struct bio *bio)
1062{
1063        if (blkcg_punt_bio_submit(bio))
1064                return BLK_QC_T_NONE;
1065
1066        /*
1067         * If it's a regular read/write or a barrier with data attached,
1068         * go through the normal accounting stuff before submission.
1069         */
1070        if (bio_has_data(bio)) {
1071                unsigned int count;
1072
1073                if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1074                        count = queue_logical_block_size(
1075                                        bio->bi_bdev->bd_disk->queue) >> 9;
1076                else
1077                        count = bio_sectors(bio);
1078
1079                if (op_is_write(bio_op(bio))) {
1080                        count_vm_events(PGPGOUT, count);
1081                } else {
1082                        task_io_account_read(bio->bi_iter.bi_size);
1083                        count_vm_events(PGPGIN, count);
1084                }
1085        }
1086
1087        /*
1088         * If we're reading data that is part of the userspace workingset, count
1089         * submission time as memory stall.  When the device is congested, or
1090         * the submitting cgroup IO-throttled, submission can be a significant
1091         * part of overall IO time.
1092         */
1093        if (unlikely(bio_op(bio) == REQ_OP_READ &&
1094            bio_flagged(bio, BIO_WORKINGSET))) {
1095                unsigned long pflags;
1096                blk_qc_t ret;
1097
1098                psi_memstall_enter(&pflags);
1099                ret = submit_bio_noacct(bio);
1100                psi_memstall_leave(&pflags);
1101
1102                return ret;
1103        }
1104
1105        return submit_bio_noacct(bio);
1106}
1107EXPORT_SYMBOL(submit_bio);
1108
1109/**
1110 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1111 *                              for the new queue limits
1112 * @q:  the queue
1113 * @rq: the request being checked
1114 *
1115 * Description:
1116 *    @rq may have been made based on weaker limitations of upper-level queues
1117 *    in request stacking drivers, and it may violate the limitation of @q.
1118 *    Since the block layer and the underlying device driver trust @rq
1119 *    after it is inserted to @q, it should be checked against @q before
1120 *    the insertion using this generic function.
1121 *
1122 *    Request stacking drivers like request-based dm may change the queue
1123 *    limits when retrying requests on other queues. Those requests need
1124 *    to be checked against the new queue limits again during dispatch.
1125 */
1126static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q,
1127                                      struct request *rq)
1128{
1129        unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
1130
1131        if (blk_rq_sectors(rq) > max_sectors) {
1132                /*
1133                 * SCSI device does not have a good way to return if
1134                 * Write Same/Zero is actually supported. If a device rejects
1135                 * a non-read/write command (discard, write same,etc.) the
1136                 * low-level device driver will set the relevant queue limit to
1137                 * 0 to prevent blk-lib from issuing more of the offending
1138                 * operations. Commands queued prior to the queue limit being
1139                 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1140                 * errors being propagated to upper layers.
1141                 */
1142                if (max_sectors == 0)
1143                        return BLK_STS_NOTSUPP;
1144
1145                printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1146                        __func__, blk_rq_sectors(rq), max_sectors);
1147                return BLK_STS_IOERR;
1148        }
1149
1150        /*
1151         * The queue settings related to segment counting may differ from the
1152         * original queue.
1153         */
1154        rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1155        if (rq->nr_phys_segments > queue_max_segments(q)) {
1156                printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1157                        __func__, rq->nr_phys_segments, queue_max_segments(q));
1158                return BLK_STS_IOERR;
1159        }
1160
1161        return BLK_STS_OK;
1162}
1163
1164/**
1165 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1166 * @q:  the queue to submit the request
1167 * @rq: the request being queued
1168 */
1169blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1170{
1171        blk_status_t ret;
1172
1173        ret = blk_cloned_rq_check_limits(q, rq);
1174        if (ret != BLK_STS_OK)
1175                return ret;
1176
1177        if (rq->rq_disk &&
1178            should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq)))
1179                return BLK_STS_IOERR;
1180
1181        if (blk_crypto_insert_cloned_request(rq))
1182                return BLK_STS_IOERR;
1183
1184        if (blk_queue_io_stat(q))
1185                blk_account_io_start(rq);
1186
1187        /*
1188         * Since we have a scheduler attached on the top device,
1189         * bypass a potential scheduler on the bottom device for
1190         * insert.
1191         */
1192        return blk_mq_request_issue_directly(rq, true);
1193}
1194EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1195
1196/**
1197 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1198 * @rq: request to examine
1199 *
1200 * Description:
1201 *     A request could be merge of IOs which require different failure
1202 *     handling.  This function determines the number of bytes which
1203 *     can be failed from the beginning of the request without
1204 *     crossing into area which need to be retried further.
1205 *
1206 * Return:
1207 *     The number of bytes to fail.
1208 */
1209unsigned int blk_rq_err_bytes(const struct request *rq)
1210{
1211        unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1212        unsigned int bytes = 0;
1213        struct bio *bio;
1214
1215        if (!(rq->rq_flags & RQF_MIXED_MERGE))
1216                return blk_rq_bytes(rq);
1217
1218        /*
1219         * Currently the only 'mixing' which can happen is between
1220         * different fastfail types.  We can safely fail portions
1221         * which have all the failfast bits that the first one has -
1222         * the ones which are at least as eager to fail as the first
1223         * one.
1224         */
1225        for (bio = rq->bio; bio; bio = bio->bi_next) {
1226                if ((bio->bi_opf & ff) != ff)
1227                        break;
1228                bytes += bio->bi_iter.bi_size;
1229        }
1230
1231        /* this could lead to infinite loop */
1232        BUG_ON(blk_rq_bytes(rq) && !bytes);
1233        return bytes;
1234}
1235EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1236
1237static void update_io_ticks(struct block_device *part, unsigned long now,
1238                bool end)
1239{
1240        unsigned long stamp;
1241again:
1242        stamp = READ_ONCE(part->bd_stamp);
1243        if (unlikely(time_after(now, stamp))) {
1244                if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1245                        __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1246        }
1247        if (part->bd_partno) {
1248                part = bdev_whole(part);
1249                goto again;
1250        }
1251}
1252
1253static void blk_account_io_completion(struct request *req, unsigned int bytes)
1254{
1255        if (req->part && blk_do_io_stat(req)) {
1256                const int sgrp = op_stat_group(req_op(req));
1257
1258                part_stat_lock();
1259                part_stat_add(req->part, sectors[sgrp], bytes >> 9);
1260                part_stat_unlock();
1261        }
1262}
1263
1264void blk_account_io_done(struct request *req, u64 now)
1265{
1266        /*
1267         * Account IO completion.  flush_rq isn't accounted as a
1268         * normal IO on queueing nor completion.  Accounting the
1269         * containing request is enough.
1270         */
1271        if (req->part && blk_do_io_stat(req) &&
1272            !(req->rq_flags & RQF_FLUSH_SEQ)) {
1273                const int sgrp = op_stat_group(req_op(req));
1274
1275                part_stat_lock();
1276                update_io_ticks(req->part, jiffies, true);
1277                part_stat_inc(req->part, ios[sgrp]);
1278                part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns);
1279                part_stat_unlock();
1280        }
1281}
1282
1283void blk_account_io_start(struct request *rq)
1284{
1285        if (!blk_do_io_stat(rq))
1286                return;
1287
1288        /* passthrough requests can hold bios that do not have ->bi_bdev set */
1289        if (rq->bio && rq->bio->bi_bdev)
1290                rq->part = rq->bio->bi_bdev;
1291        else
1292                rq->part = rq->rq_disk->part0;
1293
1294        part_stat_lock();
1295        update_io_ticks(rq->part, jiffies, false);
1296        part_stat_unlock();
1297}
1298
1299static unsigned long __part_start_io_acct(struct block_device *part,
1300                                          unsigned int sectors, unsigned int op)
1301{
1302        const int sgrp = op_stat_group(op);
1303        unsigned long now = READ_ONCE(jiffies);
1304
1305        part_stat_lock();
1306        update_io_ticks(part, now, false);
1307        part_stat_inc(part, ios[sgrp]);
1308        part_stat_add(part, sectors[sgrp], sectors);
1309        part_stat_local_inc(part, in_flight[op_is_write(op)]);
1310        part_stat_unlock();
1311
1312        return now;
1313}
1314
1315/**
1316 * bio_start_io_acct - start I/O accounting for bio based drivers
1317 * @bio:        bio to start account for
1318 *
1319 * Returns the start time that should be passed back to bio_end_io_acct().
1320 */
1321unsigned long bio_start_io_acct(struct bio *bio)
1322{
1323        return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), bio_op(bio));
1324}
1325EXPORT_SYMBOL_GPL(bio_start_io_acct);
1326
1327unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1328                                 unsigned int op)
1329{
1330        return __part_start_io_acct(disk->part0, sectors, op);
1331}
1332EXPORT_SYMBOL(disk_start_io_acct);
1333
1334static void __part_end_io_acct(struct block_device *part, unsigned int op,
1335                               unsigned long start_time)
1336{
1337        const int sgrp = op_stat_group(op);
1338        unsigned long now = READ_ONCE(jiffies);
1339        unsigned long duration = now - start_time;
1340
1341        part_stat_lock();
1342        update_io_ticks(part, now, true);
1343        part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1344        part_stat_local_dec(part, in_flight[op_is_write(op)]);
1345        part_stat_unlock();
1346}
1347
1348void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1349                struct block_device *orig_bdev)
1350{
1351        __part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1352}
1353EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1354
1355void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1356                      unsigned long start_time)
1357{
1358        __part_end_io_acct(disk->part0, op, start_time);
1359}
1360EXPORT_SYMBOL(disk_end_io_acct);
1361
1362/*
1363 * Steal bios from a request and add them to a bio list.
1364 * The request must not have been partially completed before.
1365 */
1366void blk_steal_bios(struct bio_list *list, struct request *rq)
1367{
1368        if (rq->bio) {
1369                if (list->tail)
1370                        list->tail->bi_next = rq->bio;
1371                else
1372                        list->head = rq->bio;
1373                list->tail = rq->biotail;
1374
1375                rq->bio = NULL;
1376                rq->biotail = NULL;
1377        }
1378
1379        rq->__data_len = 0;
1380}
1381EXPORT_SYMBOL_GPL(blk_steal_bios);
1382
1383/**
1384 * blk_update_request - Complete multiple bytes without completing the request
1385 * @req:      the request being processed
1386 * @error:    block status code
1387 * @nr_bytes: number of bytes to complete for @req
1388 *
1389 * Description:
1390 *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1391 *     the request structure even if @req doesn't have leftover.
1392 *     If @req has leftover, sets it up for the next range of segments.
1393 *
1394 *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1395 *     %false return from this function.
1396 *
1397 * Note:
1398 *      The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in this function
1399 *      except in the consistency check at the end of this function.
1400 *
1401 * Return:
1402 *     %false - this request doesn't have any more data
1403 *     %true  - this request has more data
1404 **/
1405bool blk_update_request(struct request *req, blk_status_t error,
1406                unsigned int nr_bytes)
1407{
1408        int total_bytes;
1409
1410        trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1411
1412        if (!req->bio)
1413                return false;
1414
1415#ifdef CONFIG_BLK_DEV_INTEGRITY
1416        if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1417            error == BLK_STS_OK)
1418                req->q->integrity.profile->complete_fn(req, nr_bytes);
1419#endif
1420
1421        if (unlikely(error && !blk_rq_is_passthrough(req) &&
1422                     !(req->rq_flags & RQF_QUIET)))
1423                print_req_error(req, error, __func__);
1424
1425        blk_account_io_completion(req, nr_bytes);
1426
1427        total_bytes = 0;
1428        while (req->bio) {
1429                struct bio *bio = req->bio;
1430                unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1431
1432                if (bio_bytes == bio->bi_iter.bi_size)
1433                        req->bio = bio->bi_next;
1434
1435                /* Completion has already been traced */
1436                bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1437                req_bio_endio(req, bio, bio_bytes, error);
1438
1439                total_bytes += bio_bytes;
1440                nr_bytes -= bio_bytes;
1441
1442                if (!nr_bytes)
1443                        break;
1444        }
1445
1446        /*
1447         * completely done
1448         */
1449        if (!req->bio) {
1450                /*
1451                 * Reset counters so that the request stacking driver
1452                 * can find how many bytes remain in the request
1453                 * later.
1454                 */
1455                req->__data_len = 0;
1456                return false;
1457        }
1458
1459        req->__data_len -= total_bytes;
1460
1461        /* update sector only for requests with clear definition of sector */
1462        if (!blk_rq_is_passthrough(req))
1463                req->__sector += total_bytes >> 9;
1464
1465        /* mixed attributes always follow the first bio */
1466        if (req->rq_flags & RQF_MIXED_MERGE) {
1467                req->cmd_flags &= ~REQ_FAILFAST_MASK;
1468                req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1469        }
1470
1471        if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1472                /*
1473                 * If total number of sectors is less than the first segment
1474                 * size, something has gone terribly wrong.
1475                 */
1476                if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1477                        blk_dump_rq_flags(req, "request botched");
1478                        req->__data_len = blk_rq_cur_bytes(req);
1479                }
1480
1481                /* recalculate the number of segments */
1482                req->nr_phys_segments = blk_recalc_rq_segments(req);
1483        }
1484
1485        return true;
1486}
1487EXPORT_SYMBOL_GPL(blk_update_request);
1488
1489#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1490/**
1491 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1492 * @rq: the request to be flushed
1493 *
1494 * Description:
1495 *     Flush all pages in @rq.
1496 */
1497void rq_flush_dcache_pages(struct request *rq)
1498{
1499        struct req_iterator iter;
1500        struct bio_vec bvec;
1501
1502        rq_for_each_segment(bvec, rq, iter)
1503                flush_dcache_page(bvec.bv_page);
1504}
1505EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1506#endif
1507
1508/**
1509 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1510 * @q : the queue of the device being checked
1511 *
1512 * Description:
1513 *    Check if underlying low-level drivers of a device are busy.
1514 *    If the drivers want to export their busy state, they must set own
1515 *    exporting function using blk_queue_lld_busy() first.
1516 *
1517 *    Basically, this function is used only by request stacking drivers
1518 *    to stop dispatching requests to underlying devices when underlying
1519 *    devices are busy.  This behavior helps more I/O merging on the queue
1520 *    of the request stacking driver and prevents I/O throughput regression
1521 *    on burst I/O load.
1522 *
1523 * Return:
1524 *    0 - Not busy (The request stacking driver should dispatch request)
1525 *    1 - Busy (The request stacking driver should stop dispatching request)
1526 */
1527int blk_lld_busy(struct request_queue *q)
1528{
1529        if (queue_is_mq(q) && q->mq_ops->busy)
1530                return q->mq_ops->busy(q);
1531
1532        return 0;
1533}
1534EXPORT_SYMBOL_GPL(blk_lld_busy);
1535
1536/**
1537 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1538 * @rq: the clone request to be cleaned up
1539 *
1540 * Description:
1541 *     Free all bios in @rq for a cloned request.
1542 */
1543void blk_rq_unprep_clone(struct request *rq)
1544{
1545        struct bio *bio;
1546
1547        while ((bio = rq->bio) != NULL) {
1548                rq->bio = bio->bi_next;
1549
1550                bio_put(bio);
1551        }
1552}
1553EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1554
1555/**
1556 * blk_rq_prep_clone - Helper function to setup clone request
1557 * @rq: the request to be setup
1558 * @rq_src: original request to be cloned
1559 * @bs: bio_set that bios for clone are allocated from
1560 * @gfp_mask: memory allocation mask for bio
1561 * @bio_ctr: setup function to be called for each clone bio.
1562 *           Returns %0 for success, non %0 for failure.
1563 * @data: private data to be passed to @bio_ctr
1564 *
1565 * Description:
1566 *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1567 *     Also, pages which the original bios are pointing to are not copied
1568 *     and the cloned bios just point same pages.
1569 *     So cloned bios must be completed before original bios, which means
1570 *     the caller must complete @rq before @rq_src.
1571 */
1572int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1573                      struct bio_set *bs, gfp_t gfp_mask,
1574                      int (*bio_ctr)(struct bio *, struct bio *, void *),
1575                      void *data)
1576{
1577        struct bio *bio, *bio_src;
1578
1579        if (!bs)
1580                bs = &fs_bio_set;
1581
1582        __rq_for_each_bio(bio_src, rq_src) {
1583                bio = bio_clone_fast(bio_src, gfp_mask, bs);
1584                if (!bio)
1585                        goto free_and_out;
1586
1587                if (bio_ctr && bio_ctr(bio, bio_src, data))
1588                        goto free_and_out;
1589
1590                if (rq->bio) {
1591                        rq->biotail->bi_next = bio;
1592                        rq->biotail = bio;
1593                } else {
1594                        rq->bio = rq->biotail = bio;
1595                }
1596                bio = NULL;
1597        }
1598
1599        /* Copy attributes of the original request to the clone request. */
1600        rq->__sector = blk_rq_pos(rq_src);
1601        rq->__data_len = blk_rq_bytes(rq_src);
1602        if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1603                rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1604                rq->special_vec = rq_src->special_vec;
1605        }
1606        rq->nr_phys_segments = rq_src->nr_phys_segments;
1607        rq->ioprio = rq_src->ioprio;
1608
1609        if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0)
1610                goto free_and_out;
1611
1612        return 0;
1613
1614free_and_out:
1615        if (bio)
1616                bio_put(bio);
1617        blk_rq_unprep_clone(rq);
1618
1619        return -ENOMEM;
1620}
1621EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1622
1623int kblockd_schedule_work(struct work_struct *work)
1624{
1625        return queue_work(kblockd_workqueue, work);
1626}
1627EXPORT_SYMBOL(kblockd_schedule_work);
1628
1629int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1630                                unsigned long delay)
1631{
1632        return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1633}
1634EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1635
1636/**
1637 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1638 * @plug:       The &struct blk_plug that needs to be initialized
1639 *
1640 * Description:
1641 *   blk_start_plug() indicates to the block layer an intent by the caller
1642 *   to submit multiple I/O requests in a batch.  The block layer may use
1643 *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1644 *   is called.  However, the block layer may choose to submit requests
1645 *   before a call to blk_finish_plug() if the number of queued I/Os
1646 *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1647 *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1648 *   the task schedules (see below).
1649 *
1650 *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1651 *   pending I/O should the task end up blocking between blk_start_plug() and
1652 *   blk_finish_plug(). This is important from a performance perspective, but
1653 *   also ensures that we don't deadlock. For instance, if the task is blocking
1654 *   for a memory allocation, memory reclaim could end up wanting to free a
1655 *   page belonging to that request that is currently residing in our private
1656 *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1657 *   this kind of deadlock.
1658 */
1659void blk_start_plug(struct blk_plug *plug)
1660{
1661        struct task_struct *tsk = current;
1662
1663        /*
1664         * If this is a nested plug, don't actually assign it.
1665         */
1666        if (tsk->plug)
1667                return;
1668
1669        INIT_LIST_HEAD(&plug->mq_list);
1670        INIT_LIST_HEAD(&plug->cb_list);
1671        plug->rq_count = 0;
1672        plug->multiple_queues = false;
1673        plug->nowait = false;
1674
1675        /*
1676         * Store ordering should not be needed here, since a potential
1677         * preempt will imply a full memory barrier
1678         */
1679        tsk->plug = plug;
1680}
1681EXPORT_SYMBOL(blk_start_plug);
1682
1683static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1684{
1685        LIST_HEAD(callbacks);
1686
1687        while (!list_empty(&plug->cb_list)) {
1688                list_splice_init(&plug->cb_list, &callbacks);
1689
1690                while (!list_empty(&callbacks)) {
1691                        struct blk_plug_cb *cb = list_first_entry(&callbacks,
1692                                                          struct blk_plug_cb,
1693                                                          list);
1694                        list_del(&cb->list);
1695                        cb->callback(cb, from_schedule);
1696                }
1697        }
1698}
1699
1700struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1701                                      int size)
1702{
1703        struct blk_plug *plug = current->plug;
1704        struct blk_plug_cb *cb;
1705
1706        if (!plug)
1707                return NULL;
1708
1709        list_for_each_entry(cb, &plug->cb_list, list)
1710                if (cb->callback == unplug && cb->data == data)
1711                        return cb;
1712
1713        /* Not currently on the callback list */
1714        BUG_ON(size < sizeof(*cb));
1715        cb = kzalloc(size, GFP_ATOMIC);
1716        if (cb) {
1717                cb->data = data;
1718                cb->callback = unplug;
1719                list_add(&cb->list, &plug->cb_list);
1720        }
1721        return cb;
1722}
1723EXPORT_SYMBOL(blk_check_plugged);
1724
1725void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1726{
1727        flush_plug_callbacks(plug, from_schedule);
1728
1729        if (!list_empty(&plug->mq_list))
1730                blk_mq_flush_plug_list(plug, from_schedule);
1731}
1732
1733/**
1734 * blk_finish_plug - mark the end of a batch of submitted I/O
1735 * @plug:       The &struct blk_plug passed to blk_start_plug()
1736 *
1737 * Description:
1738 * Indicate that a batch of I/O submissions is complete.  This function
1739 * must be paired with an initial call to blk_start_plug().  The intent
1740 * is to allow the block layer to optimize I/O submission.  See the
1741 * documentation for blk_start_plug() for more information.
1742 */
1743void blk_finish_plug(struct blk_plug *plug)
1744{
1745        if (plug != current->plug)
1746                return;
1747        blk_flush_plug_list(plug, false);
1748
1749        current->plug = NULL;
1750}
1751EXPORT_SYMBOL(blk_finish_plug);
1752
1753void blk_io_schedule(void)
1754{
1755        /* Prevent hang_check timer from firing at us during very long I/O */
1756        unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1757
1758        if (timeout)
1759                io_schedule_timeout(timeout);
1760        else
1761                io_schedule();
1762}
1763EXPORT_SYMBOL_GPL(blk_io_schedule);
1764
1765int __init blk_dev_init(void)
1766{
1767        BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1768        BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1769                        sizeof_field(struct request, cmd_flags));
1770        BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1771                        sizeof_field(struct bio, bi_opf));
1772
1773        /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1774        kblockd_workqueue = alloc_workqueue("kblockd",
1775                                            WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1776        if (!kblockd_workqueue)
1777                panic("Failed to create kblockd\n");
1778
1779        blk_requestq_cachep = kmem_cache_create("request_queue",
1780                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1781
1782        blk_debugfs_root = debugfs_create_dir("block", NULL);
1783
1784        return 0;
1785}
1786