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/highmem.h>
  22#include <linux/mm.h>
  23#include <linux/kernel_stat.h>
  24#include <linux/string.h>
  25#include <linux/init.h>
  26#include <linux/completion.h>
  27#include <linux/slab.h>
  28#include <linux/swap.h>
  29#include <linux/writeback.h>
  30#include <linux/task_io_accounting_ops.h>
  31#include <linux/fault-inject.h>
  32#include <linux/list_sort.h>
  33#include <linux/delay.h>
  34#include <linux/ratelimit.h>
  35#include <linux/pm_runtime.h>
  36#include <linux/blk-cgroup.h>
  37#include <linux/debugfs.h>
  38#include <linux/bpf.h>
  39
  40#define CREATE_TRACE_POINTS
  41#include <trace/events/block.h>
  42
  43#include "blk.h"
  44#include "blk-mq.h"
  45#include "blk-mq-sched.h"
  46#include "blk-pm.h"
  47#include "blk-rq-qos.h"
  48
  49#ifdef CONFIG_DEBUG_FS
  50struct dentry *blk_debugfs_root;
  51#endif
  52
  53EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
  54EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
  55EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
  56EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
  57EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
  58
  59DEFINE_IDA(blk_queue_ida);
  60
  61/*
  62 * For queue allocation
  63 */
  64struct kmem_cache *blk_requestq_cachep;
  65
  66/*
  67 * Controlling structure to kblockd
  68 */
  69static struct workqueue_struct *kblockd_workqueue;
  70
  71/**
  72 * blk_queue_flag_set - atomically set a queue flag
  73 * @flag: flag to be set
  74 * @q: request queue
  75 */
  76void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
  77{
  78        set_bit(flag, &q->queue_flags);
  79}
  80EXPORT_SYMBOL(blk_queue_flag_set);
  81
  82/**
  83 * blk_queue_flag_clear - atomically clear a queue flag
  84 * @flag: flag to be cleared
  85 * @q: request queue
  86 */
  87void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
  88{
  89        clear_bit(flag, &q->queue_flags);
  90}
  91EXPORT_SYMBOL(blk_queue_flag_clear);
  92
  93/**
  94 * blk_queue_flag_test_and_set - atomically test and set a queue flag
  95 * @flag: flag to be set
  96 * @q: request queue
  97 *
  98 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
  99 * the flag was already set.
 100 */
 101bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
 102{
 103        return test_and_set_bit(flag, &q->queue_flags);
 104}
 105EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
 106
 107void blk_rq_init(struct request_queue *q, struct request *rq)
 108{
 109        memset(rq, 0, sizeof(*rq));
 110
 111        INIT_LIST_HEAD(&rq->queuelist);
 112        rq->q = q;
 113        rq->__sector = (sector_t) -1;
 114        INIT_HLIST_NODE(&rq->hash);
 115        RB_CLEAR_NODE(&rq->rb_node);
 116        rq->tag = -1;
 117        rq->internal_tag = -1;
 118        rq->start_time_ns = ktime_get_ns();
 119        rq->part = NULL;
 120        refcount_set(&rq->ref, 1);
 121}
 122EXPORT_SYMBOL(blk_rq_init);
 123
 124#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
 125static const char *const blk_op_name[] = {
 126        REQ_OP_NAME(READ),
 127        REQ_OP_NAME(WRITE),
 128        REQ_OP_NAME(FLUSH),
 129        REQ_OP_NAME(DISCARD),
 130        REQ_OP_NAME(SECURE_ERASE),
 131        REQ_OP_NAME(ZONE_RESET),
 132        REQ_OP_NAME(WRITE_SAME),
 133        REQ_OP_NAME(WRITE_ZEROES),
 134        REQ_OP_NAME(SCSI_IN),
 135        REQ_OP_NAME(SCSI_OUT),
 136        REQ_OP_NAME(DRV_IN),
 137        REQ_OP_NAME(DRV_OUT),
 138};
 139#undef REQ_OP_NAME
 140
 141/**
 142 * blk_op_str - Return string XXX in the REQ_OP_XXX.
 143 * @op: REQ_OP_XXX.
 144 *
 145 * Description: Centralize block layer function to convert REQ_OP_XXX into
 146 * string format. Useful in the debugging and tracing bio or request. For
 147 * invalid REQ_OP_XXX it returns string "UNKNOWN".
 148 */
 149inline const char *blk_op_str(unsigned int op)
 150{
 151        const char *op_str = "UNKNOWN";
 152
 153        if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
 154                op_str = blk_op_name[op];
 155
 156        return op_str;
 157}
 158EXPORT_SYMBOL_GPL(blk_op_str);
 159
 160static const struct {
 161        int             errno;
 162        const char      *name;
 163} blk_errors[] = {
 164        [BLK_STS_OK]            = { 0,          "" },
 165        [BLK_STS_NOTSUPP]       = { -EOPNOTSUPP, "operation not supported" },
 166        [BLK_STS_TIMEOUT]       = { -ETIMEDOUT, "timeout" },
 167        [BLK_STS_NOSPC]         = { -ENOSPC,    "critical space allocation" },
 168        [BLK_STS_TRANSPORT]     = { -ENOLINK,   "recoverable transport" },
 169        [BLK_STS_TARGET]        = { -EREMOTEIO, "critical target" },
 170        [BLK_STS_NEXUS]         = { -EBADE,     "critical nexus" },
 171        [BLK_STS_MEDIUM]        = { -ENODATA,   "critical medium" },
 172        [BLK_STS_PROTECTION]    = { -EILSEQ,    "protection" },
 173        [BLK_STS_RESOURCE]      = { -ENOMEM,    "kernel resource" },
 174        [BLK_STS_DEV_RESOURCE]  = { -EBUSY,     "device resource" },
 175        [BLK_STS_AGAIN]         = { -EAGAIN,    "nonblocking retry" },
 176
 177        /* device mapper special case, should not leak out: */
 178        [BLK_STS_DM_REQUEUE]    = { -EREMCHG, "dm internal retry" },
 179
 180        /* everything else not covered above: */
 181        [BLK_STS_IOERR]         = { -EIO,       "I/O" },
 182};
 183
 184blk_status_t errno_to_blk_status(int errno)
 185{
 186        int i;
 187
 188        for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
 189                if (blk_errors[i].errno == errno)
 190                        return (__force blk_status_t)i;
 191        }
 192
 193        return BLK_STS_IOERR;
 194}
 195EXPORT_SYMBOL_GPL(errno_to_blk_status);
 196
 197int blk_status_to_errno(blk_status_t status)
 198{
 199        int idx = (__force int)status;
 200
 201        if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
 202                return -EIO;
 203        return blk_errors[idx].errno;
 204}
 205EXPORT_SYMBOL_GPL(blk_status_to_errno);
 206
 207static void print_req_error(struct request *req, blk_status_t status,
 208                const char *caller)
 209{
 210        int idx = (__force int)status;
 211
 212        if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
 213                return;
 214
 215        printk_ratelimited(KERN_ERR
 216                "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
 217                "phys_seg %u prio class %u\n",
 218                caller, blk_errors[idx].name,
 219                req->rq_disk ? req->rq_disk->disk_name : "?",
 220                blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
 221                req->cmd_flags & ~REQ_OP_MASK,
 222                req->nr_phys_segments,
 223                IOPRIO_PRIO_CLASS(req->ioprio));
 224}
 225
 226static void req_bio_endio(struct request *rq, struct bio *bio,
 227                          unsigned int nbytes, blk_status_t error)
 228{
 229        if (error)
 230                bio->bi_status = error;
 231
 232        if (unlikely(rq->rq_flags & RQF_QUIET))
 233                bio_set_flag(bio, BIO_QUIET);
 234
 235        bio_advance(bio, nbytes);
 236
 237        /* don't actually finish bio if it's part of flush sequence */
 238        if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
 239                bio_endio(bio);
 240}
 241
 242void blk_dump_rq_flags(struct request *rq, char *msg)
 243{
 244        printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
 245                rq->rq_disk ? rq->rq_disk->disk_name : "?",
 246                (unsigned long long) rq->cmd_flags);
 247
 248        printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
 249               (unsigned long long)blk_rq_pos(rq),
 250               blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
 251        printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
 252               rq->bio, rq->biotail, blk_rq_bytes(rq));
 253}
 254EXPORT_SYMBOL(blk_dump_rq_flags);
 255
 256/**
 257 * blk_sync_queue - cancel any pending callbacks on a queue
 258 * @q: the queue
 259 *
 260 * Description:
 261 *     The block layer may perform asynchronous callback activity
 262 *     on a queue, such as calling the unplug function after a timeout.
 263 *     A block device may call blk_sync_queue to ensure that any
 264 *     such activity is cancelled, thus allowing it to release resources
 265 *     that the callbacks might use. The caller must already have made sure
 266 *     that its ->make_request_fn will not re-add plugging prior to calling
 267 *     this function.
 268 *
 269 *     This function does not cancel any asynchronous activity arising
 270 *     out of elevator or throttling code. That would require elevator_exit()
 271 *     and blkcg_exit_queue() to be called with queue lock initialized.
 272 *
 273 */
 274void blk_sync_queue(struct request_queue *q)
 275{
 276        del_timer_sync(&q->timeout);
 277        cancel_work_sync(&q->timeout_work);
 278}
 279EXPORT_SYMBOL(blk_sync_queue);
 280
 281/**
 282 * blk_set_pm_only - increment pm_only counter
 283 * @q: request queue pointer
 284 */
 285void blk_set_pm_only(struct request_queue *q)
 286{
 287        atomic_inc(&q->pm_only);
 288}
 289EXPORT_SYMBOL_GPL(blk_set_pm_only);
 290
 291void blk_clear_pm_only(struct request_queue *q)
 292{
 293        int pm_only;
 294
 295        pm_only = atomic_dec_return(&q->pm_only);
 296        WARN_ON_ONCE(pm_only < 0);
 297        if (pm_only == 0)
 298                wake_up_all(&q->mq_freeze_wq);
 299}
 300EXPORT_SYMBOL_GPL(blk_clear_pm_only);
 301
 302void blk_put_queue(struct request_queue *q)
 303{
 304        kobject_put(&q->kobj);
 305}
 306EXPORT_SYMBOL(blk_put_queue);
 307
 308void blk_set_queue_dying(struct request_queue *q)
 309{
 310        blk_queue_flag_set(QUEUE_FLAG_DYING, q);
 311
 312        /*
 313         * When queue DYING flag is set, we need to block new req
 314         * entering queue, so we call blk_freeze_queue_start() to
 315         * prevent I/O from crossing blk_queue_enter().
 316         */
 317        blk_freeze_queue_start(q);
 318
 319        if (queue_is_mq(q))
 320                blk_mq_wake_waiters(q);
 321
 322        /* Make blk_queue_enter() reexamine the DYING flag. */
 323        wake_up_all(&q->mq_freeze_wq);
 324}
 325EXPORT_SYMBOL_GPL(blk_set_queue_dying);
 326
 327/**
 328 * blk_cleanup_queue - shutdown a request queue
 329 * @q: request queue to shutdown
 330 *
 331 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
 332 * put it.  All future requests will be failed immediately with -ENODEV.
 333 */
 334void blk_cleanup_queue(struct request_queue *q)
 335{
 336        /* mark @q DYING, no new request or merges will be allowed afterwards */
 337        mutex_lock(&q->sysfs_lock);
 338        blk_set_queue_dying(q);
 339
 340        blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 341        blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 342        blk_queue_flag_set(QUEUE_FLAG_DYING, q);
 343        mutex_unlock(&q->sysfs_lock);
 344
 345        /*
 346         * Drain all requests queued before DYING marking. Set DEAD flag to
 347         * prevent that q->request_fn() gets invoked after draining finished.
 348         */
 349        blk_freeze_queue(q);
 350
 351        rq_qos_exit(q);
 352
 353        blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
 354
 355        /* for synchronous bio-based driver finish in-flight integrity i/o */
 356        blk_flush_integrity();
 357
 358        /* @q won't process any more request, flush async actions */
 359        del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
 360        blk_sync_queue(q);
 361
 362        if (queue_is_mq(q))
 363                blk_mq_exit_queue(q);
 364
 365        /*
 366         * In theory, request pool of sched_tags belongs to request queue.
 367         * However, the current implementation requires tag_set for freeing
 368         * requests, so free the pool now.
 369         *
 370         * Queue has become frozen, there can't be any in-queue requests, so
 371         * it is safe to free requests now.
 372         */
 373        mutex_lock(&q->sysfs_lock);
 374        if (q->elevator)
 375                blk_mq_sched_free_requests(q);
 376        mutex_unlock(&q->sysfs_lock);
 377
 378        percpu_ref_exit(&q->q_usage_counter);
 379
 380        /* @q is and will stay empty, shutdown and put */
 381        blk_put_queue(q);
 382}
 383EXPORT_SYMBOL(blk_cleanup_queue);
 384
 385struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
 386{
 387        return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
 388}
 389EXPORT_SYMBOL(blk_alloc_queue);
 390
 391/**
 392 * blk_queue_enter() - try to increase q->q_usage_counter
 393 * @q: request queue pointer
 394 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
 395 */
 396int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 397{
 398        const bool pm = flags & BLK_MQ_REQ_PREEMPT;
 399
 400        while (true) {
 401                bool success = false;
 402
 403                rcu_read_lock();
 404                if (percpu_ref_tryget_live(&q->q_usage_counter)) {
 405                        /*
 406                         * The code that increments the pm_only counter is
 407                         * responsible for ensuring that that counter is
 408                         * globally visible before the queue is unfrozen.
 409                         */
 410                        if (pm || !blk_queue_pm_only(q)) {
 411                                success = true;
 412                        } else {
 413                                percpu_ref_put(&q->q_usage_counter);
 414                        }
 415                }
 416                rcu_read_unlock();
 417
 418                if (success)
 419                        return 0;
 420
 421                if (flags & BLK_MQ_REQ_NOWAIT)
 422                        return -EBUSY;
 423
 424                /*
 425                 * read pair of barrier in blk_freeze_queue_start(),
 426                 * we need to order reading __PERCPU_REF_DEAD flag of
 427                 * .q_usage_counter and reading .mq_freeze_depth or
 428                 * queue dying flag, otherwise the following wait may
 429                 * never return if the two reads are reordered.
 430                 */
 431                smp_rmb();
 432
 433                wait_event(q->mq_freeze_wq,
 434                           (!q->mq_freeze_depth &&
 435                            (pm || (blk_pm_request_resume(q),
 436                                    !blk_queue_pm_only(q)))) ||
 437                           blk_queue_dying(q));
 438                if (blk_queue_dying(q))
 439                        return -ENODEV;
 440        }
 441}
 442
 443void blk_queue_exit(struct request_queue *q)
 444{
 445        percpu_ref_put(&q->q_usage_counter);
 446}
 447
 448static void blk_queue_usage_counter_release(struct percpu_ref *ref)
 449{
 450        struct request_queue *q =
 451                container_of(ref, struct request_queue, q_usage_counter);
 452
 453        wake_up_all(&q->mq_freeze_wq);
 454}
 455
 456static void blk_rq_timed_out_timer(struct timer_list *t)
 457{
 458        struct request_queue *q = from_timer(q, t, timeout);
 459
 460        kblockd_schedule_work(&q->timeout_work);
 461}
 462
 463static void blk_timeout_work(struct work_struct *work)
 464{
 465}
 466
 467/**
 468 * blk_alloc_queue_node - allocate a request queue
 469 * @gfp_mask: memory allocation flags
 470 * @node_id: NUMA node to allocate memory from
 471 */
 472struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 473{
 474        struct request_queue *q;
 475        int ret;
 476
 477        q = kmem_cache_alloc_node(blk_requestq_cachep,
 478                                gfp_mask | __GFP_ZERO, node_id);
 479        if (!q)
 480                return NULL;
 481
 482        INIT_LIST_HEAD(&q->queue_head);
 483        q->last_merge = NULL;
 484
 485        q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
 486        if (q->id < 0)
 487                goto fail_q;
 488
 489        ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
 490        if (ret)
 491                goto fail_id;
 492
 493        q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
 494        if (!q->backing_dev_info)
 495                goto fail_split;
 496
 497        q->stats = blk_alloc_queue_stats();
 498        if (!q->stats)
 499                goto fail_stats;
 500
 501        q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
 502        q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
 503        q->backing_dev_info->name = "block";
 504        q->node = node_id;
 505
 506        timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
 507                    laptop_mode_timer_fn, 0);
 508        timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
 509        INIT_WORK(&q->timeout_work, blk_timeout_work);
 510        INIT_LIST_HEAD(&q->icq_list);
 511#ifdef CONFIG_BLK_CGROUP
 512        INIT_LIST_HEAD(&q->blkg_list);
 513#endif
 514
 515        kobject_init(&q->kobj, &blk_queue_ktype);
 516
 517#ifdef CONFIG_BLK_DEV_IO_TRACE
 518        mutex_init(&q->blk_trace_mutex);
 519#endif
 520        mutex_init(&q->sysfs_lock);
 521        spin_lock_init(&q->queue_lock);
 522
 523        init_waitqueue_head(&q->mq_freeze_wq);
 524        mutex_init(&q->mq_freeze_lock);
 525
 526        /*
 527         * Init percpu_ref in atomic mode so that it's faster to shutdown.
 528         * See blk_register_queue() for details.
 529         */
 530        if (percpu_ref_init(&q->q_usage_counter,
 531                                blk_queue_usage_counter_release,
 532                                PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
 533                goto fail_bdi;
 534
 535        if (blkcg_init_queue(q))
 536                goto fail_ref;
 537
 538        return q;
 539
 540fail_ref:
 541        percpu_ref_exit(&q->q_usage_counter);
 542fail_bdi:
 543        blk_free_queue_stats(q->stats);
 544fail_stats:
 545        bdi_put(q->backing_dev_info);
 546fail_split:
 547        bioset_exit(&q->bio_split);
 548fail_id:
 549        ida_simple_remove(&blk_queue_ida, q->id);
 550fail_q:
 551        kmem_cache_free(blk_requestq_cachep, q);
 552        return NULL;
 553}
 554EXPORT_SYMBOL(blk_alloc_queue_node);
 555
 556bool blk_get_queue(struct request_queue *q)
 557{
 558        if (likely(!blk_queue_dying(q))) {
 559                __blk_get_queue(q);
 560                return true;
 561        }
 562
 563        return false;
 564}
 565EXPORT_SYMBOL(blk_get_queue);
 566
 567/**
 568 * blk_get_request - allocate a request
 569 * @q: request queue to allocate a request for
 570 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
 571 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
 572 */
 573struct request *blk_get_request(struct request_queue *q, unsigned int op,
 574                                blk_mq_req_flags_t flags)
 575{
 576        struct request *req;
 577
 578        WARN_ON_ONCE(op & REQ_NOWAIT);
 579        WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
 580
 581        req = blk_mq_alloc_request(q, op, flags);
 582        if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
 583                q->mq_ops->initialize_rq_fn(req);
 584
 585        return req;
 586}
 587EXPORT_SYMBOL(blk_get_request);
 588
 589void blk_put_request(struct request *req)
 590{
 591        blk_mq_free_request(req);
 592}
 593EXPORT_SYMBOL(blk_put_request);
 594
 595bool bio_attempt_back_merge(struct request *req, struct bio *bio,
 596                unsigned int nr_segs)
 597{
 598        const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
 599
 600        if (!ll_back_merge_fn(req, bio, nr_segs))
 601                return false;
 602
 603        trace_block_bio_backmerge(req->q, req, bio);
 604
 605        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
 606                blk_rq_set_mixed_merge(req);
 607
 608        req->biotail->bi_next = bio;
 609        req->biotail = bio;
 610        req->__data_len += bio->bi_iter.bi_size;
 611
 612        blk_account_io_start(req, false);
 613        return true;
 614}
 615
 616bool bio_attempt_front_merge(struct request *req, struct bio *bio,
 617                unsigned int nr_segs)
 618{
 619        const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
 620
 621        if (!ll_front_merge_fn(req, bio, nr_segs))
 622                return false;
 623
 624        trace_block_bio_frontmerge(req->q, req, bio);
 625
 626        if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
 627                blk_rq_set_mixed_merge(req);
 628
 629        bio->bi_next = req->bio;
 630        req->bio = bio;
 631
 632        req->__sector = bio->bi_iter.bi_sector;
 633        req->__data_len += bio->bi_iter.bi_size;
 634
 635        blk_account_io_start(req, false);
 636        return true;
 637}
 638
 639bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
 640                struct bio *bio)
 641{
 642        unsigned short segments = blk_rq_nr_discard_segments(req);
 643
 644        if (segments >= queue_max_discard_segments(q))
 645                goto no_merge;
 646        if (blk_rq_sectors(req) + bio_sectors(bio) >
 647            blk_rq_get_max_sectors(req, blk_rq_pos(req)))
 648                goto no_merge;
 649
 650        req->biotail->bi_next = bio;
 651        req->biotail = bio;
 652        req->__data_len += bio->bi_iter.bi_size;
 653        req->nr_phys_segments = segments + 1;
 654
 655        blk_account_io_start(req, false);
 656        return true;
 657no_merge:
 658        req_set_nomerge(q, req);
 659        return false;
 660}
 661
 662/**
 663 * blk_attempt_plug_merge - try to merge with %current's plugged list
 664 * @q: request_queue new bio is being queued at
 665 * @bio: new bio being queued
 666 * @nr_segs: number of segments in @bio
 667 * @same_queue_rq: pointer to &struct request that gets filled in when
 668 * another request associated with @q is found on the plug list
 669 * (optional, may be %NULL)
 670 *
 671 * Determine whether @bio being queued on @q can be merged with a request
 672 * on %current's plugged list.  Returns %true if merge was successful,
 673 * otherwise %false.
 674 *
 675 * Plugging coalesces IOs from the same issuer for the same purpose without
 676 * going through @q->queue_lock.  As such it's more of an issuing mechanism
 677 * than scheduling, and the request, while may have elvpriv data, is not
 678 * added on the elevator at this point.  In addition, we don't have
 679 * reliable access to the elevator outside queue lock.  Only check basic
 680 * merging parameters without querying the elevator.
 681 *
 682 * Caller must ensure !blk_queue_nomerges(q) beforehand.
 683 */
 684bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
 685                unsigned int nr_segs, struct request **same_queue_rq)
 686{
 687        struct blk_plug *plug;
 688        struct request *rq;
 689        struct list_head *plug_list;
 690
 691        plug = blk_mq_plug(q, bio);
 692        if (!plug)
 693                return false;
 694
 695        plug_list = &plug->mq_list;
 696
 697        list_for_each_entry_reverse(rq, plug_list, queuelist) {
 698                bool merged = false;
 699
 700                if (rq->q == q && same_queue_rq) {
 701                        /*
 702                         * Only blk-mq multiple hardware queues case checks the
 703                         * rq in the same queue, there should be only one such
 704                         * rq in a queue
 705                         **/
 706                        *same_queue_rq = rq;
 707                }
 708
 709                if (rq->q != q || !blk_rq_merge_ok(rq, bio))
 710                        continue;
 711
 712                switch (blk_try_merge(rq, bio)) {
 713                case ELEVATOR_BACK_MERGE:
 714                        merged = bio_attempt_back_merge(rq, bio, nr_segs);
 715                        break;
 716                case ELEVATOR_FRONT_MERGE:
 717                        merged = bio_attempt_front_merge(rq, bio, nr_segs);
 718                        break;
 719                case ELEVATOR_DISCARD_MERGE:
 720                        merged = bio_attempt_discard_merge(q, rq, bio);
 721                        break;
 722                default:
 723                        break;
 724                }
 725
 726                if (merged)
 727                        return true;
 728        }
 729
 730        return false;
 731}
 732
 733static void handle_bad_sector(struct bio *bio, sector_t maxsector)
 734{
 735        char b[BDEVNAME_SIZE];
 736
 737        printk(KERN_INFO "attempt to access beyond end of device\n");
 738        printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
 739                        bio_devname(bio, b), bio->bi_opf,
 740                        (unsigned long long)bio_end_sector(bio),
 741                        (long long)maxsector);
 742}
 743
 744#ifdef CONFIG_FAIL_MAKE_REQUEST
 745
 746static DECLARE_FAULT_ATTR(fail_make_request);
 747
 748static int __init setup_fail_make_request(char *str)
 749{
 750        return setup_fault_attr(&fail_make_request, str);
 751}
 752__setup("fail_make_request=", setup_fail_make_request);
 753
 754static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
 755{
 756        return part->make_it_fail && should_fail(&fail_make_request, bytes);
 757}
 758
 759static int __init fail_make_request_debugfs(void)
 760{
 761        struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
 762                                                NULL, &fail_make_request);
 763
 764        return PTR_ERR_OR_ZERO(dir);
 765}
 766
 767late_initcall(fail_make_request_debugfs);
 768
 769#else /* CONFIG_FAIL_MAKE_REQUEST */
 770
 771static inline bool should_fail_request(struct hd_struct *part,
 772                                        unsigned int bytes)
 773{
 774        return false;
 775}
 776
 777#endif /* CONFIG_FAIL_MAKE_REQUEST */
 778
 779static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
 780{
 781        const int op = bio_op(bio);
 782
 783        if (part->policy && op_is_write(op)) {
 784                char b[BDEVNAME_SIZE];
 785
 786                if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
 787                        return false;
 788
 789                WARN_ONCE(1,
 790                       "generic_make_request: Trying to write "
 791                        "to read-only block-device %s (partno %d)\n",
 792                        bio_devname(bio, b), part->partno);
 793                /* Older lvm-tools actually trigger this */
 794                return false;
 795        }
 796
 797        return false;
 798}
 799
 800static noinline int should_fail_bio(struct bio *bio)
 801{
 802        if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
 803                return -EIO;
 804        return 0;
 805}
 806ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
 807
 808/*
 809 * Check whether this bio extends beyond the end of the device or partition.
 810 * This may well happen - the kernel calls bread() without checking the size of
 811 * the device, e.g., when mounting a file system.
 812 */
 813static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
 814{
 815        unsigned int nr_sectors = bio_sectors(bio);
 816
 817        if (nr_sectors && maxsector &&
 818            (nr_sectors > maxsector ||
 819             bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
 820                handle_bad_sector(bio, maxsector);
 821                return -EIO;
 822        }
 823        return 0;
 824}
 825
 826/*
 827 * Remap block n of partition p to block n+start(p) of the disk.
 828 */
 829static inline int blk_partition_remap(struct bio *bio)
 830{
 831        struct hd_struct *p;
 832        int ret = -EIO;
 833
 834        rcu_read_lock();
 835        p = __disk_get_part(bio->bi_disk, bio->bi_partno);
 836        if (unlikely(!p))
 837                goto out;
 838        if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
 839                goto out;
 840        if (unlikely(bio_check_ro(bio, p)))
 841                goto out;
 842
 843        /*
 844         * Zone reset does not include bi_size so bio_sectors() is always 0.
 845         * Include a test for the reset op code and perform the remap if needed.
 846         */
 847        if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
 848                if (bio_check_eod(bio, part_nr_sects_read(p)))
 849                        goto out;
 850                bio->bi_iter.bi_sector += p->start_sect;
 851                trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
 852                                      bio->bi_iter.bi_sector - p->start_sect);
 853        }
 854        bio->bi_partno = 0;
 855        ret = 0;
 856out:
 857        rcu_read_unlock();
 858        return ret;
 859}
 860
 861static noinline_for_stack bool
 862generic_make_request_checks(struct bio *bio)
 863{
 864        struct request_queue *q;
 865        int nr_sectors = bio_sectors(bio);
 866        blk_status_t status = BLK_STS_IOERR;
 867        char b[BDEVNAME_SIZE];
 868
 869        might_sleep();
 870
 871        q = bio->bi_disk->queue;
 872        if (unlikely(!q)) {
 873                printk(KERN_ERR
 874                       "generic_make_request: Trying to access "
 875                        "nonexistent block-device %s (%Lu)\n",
 876                        bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
 877                goto end_io;
 878        }
 879
 880        /*
 881         * For a REQ_NOWAIT based request, return -EOPNOTSUPP
 882         * if queue is not a request based queue.
 883         */
 884        if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
 885                goto not_supported;
 886
 887        if (should_fail_bio(bio))
 888                goto end_io;
 889
 890        if (bio->bi_partno) {
 891                if (unlikely(blk_partition_remap(bio)))
 892                        goto end_io;
 893        } else {
 894                if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
 895                        goto end_io;
 896                if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
 897                        goto end_io;
 898        }
 899
 900        /*
 901         * Filter flush bio's early so that make_request based
 902         * drivers without flush support don't have to worry
 903         * about them.
 904         */
 905        if (op_is_flush(bio->bi_opf) &&
 906            !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
 907                bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
 908                if (!nr_sectors) {
 909                        status = BLK_STS_OK;
 910                        goto end_io;
 911                }
 912        }
 913
 914        if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
 915                bio->bi_opf &= ~REQ_HIPRI;
 916
 917        switch (bio_op(bio)) {
 918        case REQ_OP_DISCARD:
 919                if (!blk_queue_discard(q))
 920                        goto not_supported;
 921                break;
 922        case REQ_OP_SECURE_ERASE:
 923                if (!blk_queue_secure_erase(q))
 924                        goto not_supported;
 925                break;
 926        case REQ_OP_WRITE_SAME:
 927                if (!q->limits.max_write_same_sectors)
 928                        goto not_supported;
 929                break;
 930        case REQ_OP_ZONE_RESET:
 931                if (!blk_queue_is_zoned(q))
 932                        goto not_supported;
 933                break;
 934        case REQ_OP_WRITE_ZEROES:
 935                if (!q->limits.max_write_zeroes_sectors)
 936                        goto not_supported;
 937                break;
 938        default:
 939                break;
 940        }
 941
 942        /*
 943         * Various block parts want %current->io_context and lazy ioc
 944         * allocation ends up trading a lot of pain for a small amount of
 945         * memory.  Just allocate it upfront.  This may fail and block
 946         * layer knows how to live with it.
 947         */
 948        create_io_context(GFP_ATOMIC, q->node);
 949
 950        if (!blkcg_bio_issue_check(q, bio))
 951                return false;
 952
 953        if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
 954                trace_block_bio_queue(q, bio);
 955                /* Now that enqueuing has been traced, we need to trace
 956                 * completion as well.
 957                 */
 958                bio_set_flag(bio, BIO_TRACE_COMPLETION);
 959        }
 960        return true;
 961
 962not_supported:
 963        status = BLK_STS_NOTSUPP;
 964end_io:
 965        bio->bi_status = status;
 966        bio_endio(bio);
 967        return false;
 968}
 969
 970/**
 971 * generic_make_request - hand a buffer to its device driver for I/O
 972 * @bio:  The bio describing the location in memory and on the device.
 973 *
 974 * generic_make_request() is used to make I/O requests of block
 975 * devices. It is passed a &struct bio, which describes the I/O that needs
 976 * to be done.
 977 *
 978 * generic_make_request() does not return any status.  The
 979 * success/failure status of the request, along with notification of
 980 * completion, is delivered asynchronously through the bio->bi_end_io
 981 * function described (one day) else where.
 982 *
 983 * The caller of generic_make_request must make sure that bi_io_vec
 984 * are set to describe the memory buffer, and that bi_dev and bi_sector are
 985 * set to describe the device address, and the
 986 * bi_end_io and optionally bi_private are set to describe how
 987 * completion notification should be signaled.
 988 *
 989 * generic_make_request and the drivers it calls may use bi_next if this
 990 * bio happens to be merged with someone else, and may resubmit the bio to
 991 * a lower device by calling into generic_make_request recursively, which
 992 * means the bio should NOT be touched after the call to ->make_request_fn.
 993 */
 994blk_qc_t generic_make_request(struct bio *bio)
 995{
 996        /*
 997         * bio_list_on_stack[0] contains bios submitted by the current
 998         * make_request_fn.
 999         * bio_list_on_stack[1] contains bios that were submitted before
1000         * the current make_request_fn, but that haven't been processed
1001         * yet.
1002         */
1003        struct bio_list bio_list_on_stack[2];
1004        blk_qc_t ret = BLK_QC_T_NONE;
1005
1006        if (!generic_make_request_checks(bio))
1007                goto out;
1008
1009        /*
1010         * We only want one ->make_request_fn to be active at a time, else
1011         * stack usage with stacked devices could be a problem.  So use
1012         * current->bio_list to keep a list of requests submited by a
1013         * make_request_fn function.  current->bio_list is also used as a
1014         * flag to say if generic_make_request is currently active in this
1015         * task or not.  If it is NULL, then no make_request is active.  If
1016         * it is non-NULL, then a make_request is active, and new requests
1017         * should be added at the tail
1018         */
1019        if (current->bio_list) {
1020                bio_list_add(&current->bio_list[0], bio);
1021                goto out;
1022        }
1023
1024        /* following loop may be a bit non-obvious, and so deserves some
1025         * explanation.
1026         * Before entering the loop, bio->bi_next is NULL (as all callers
1027         * ensure that) so we have a list with a single bio.
1028         * We pretend that we have just taken it off a longer list, so
1029         * we assign bio_list to a pointer to the bio_list_on_stack,
1030         * thus initialising the bio_list of new bios to be
1031         * added.  ->make_request() may indeed add some more bios
1032         * through a recursive call to generic_make_request.  If it
1033         * did, we find a non-NULL value in bio_list and re-enter the loop
1034         * from the top.  In this case we really did just take the bio
1035         * of the top of the list (no pretending) and so remove it from
1036         * bio_list, and call into ->make_request() again.
1037         */
1038        BUG_ON(bio->bi_next);
1039        bio_list_init(&bio_list_on_stack[0]);
1040        current->bio_list = bio_list_on_stack;
1041        do {
1042                struct request_queue *q = bio->bi_disk->queue;
1043                blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1044                        BLK_MQ_REQ_NOWAIT : 0;
1045
1046                if (likely(blk_queue_enter(q, flags) == 0)) {
1047                        struct bio_list lower, same;
1048
1049                        /* Create a fresh bio_list for all subordinate requests */
1050                        bio_list_on_stack[1] = bio_list_on_stack[0];
1051                        bio_list_init(&bio_list_on_stack[0]);
1052                        ret = q->make_request_fn(q, bio);
1053
1054                        blk_queue_exit(q);
1055
1056                        /* sort new bios into those for a lower level
1057                         * and those for the same level
1058                         */
1059                        bio_list_init(&lower);
1060                        bio_list_init(&same);
1061                        while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1062                                if (q == bio->bi_disk->queue)
1063                                        bio_list_add(&same, bio);
1064                                else
1065                                        bio_list_add(&lower, bio);
1066                        /* now assemble so we handle the lowest level first */
1067                        bio_list_merge(&bio_list_on_stack[0], &lower);
1068                        bio_list_merge(&bio_list_on_stack[0], &same);
1069                        bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1070                } else {
1071                        if (unlikely(!blk_queue_dying(q) &&
1072                                        (bio->bi_opf & REQ_NOWAIT)))
1073                                bio_wouldblock_error(bio);
1074                        else
1075                                bio_io_error(bio);
1076                }
1077                bio = bio_list_pop(&bio_list_on_stack[0]);
1078        } while (bio);
1079        current->bio_list = NULL; /* deactivate */
1080
1081out:
1082        return ret;
1083}
1084EXPORT_SYMBOL(generic_make_request);
1085
1086/**
1087 * direct_make_request - hand a buffer directly to its device driver for I/O
1088 * @bio:  The bio describing the location in memory and on the device.
1089 *
1090 * This function behaves like generic_make_request(), but does not protect
1091 * against recursion.  Must only be used if the called driver is known
1092 * to not call generic_make_request (or direct_make_request) again from
1093 * its make_request function.  (Calling direct_make_request again from
1094 * a workqueue is perfectly fine as that doesn't recurse).
1095 */
1096blk_qc_t direct_make_request(struct bio *bio)
1097{
1098        struct request_queue *q = bio->bi_disk->queue;
1099        bool nowait = bio->bi_opf & REQ_NOWAIT;
1100        blk_qc_t ret;
1101
1102        if (!generic_make_request_checks(bio))
1103                return BLK_QC_T_NONE;
1104
1105        if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1106                if (nowait && !blk_queue_dying(q))
1107                        bio->bi_status = BLK_STS_AGAIN;
1108                else
1109                        bio->bi_status = BLK_STS_IOERR;
1110                bio_endio(bio);
1111                return BLK_QC_T_NONE;
1112        }
1113
1114        ret = q->make_request_fn(q, bio);
1115        blk_queue_exit(q);
1116        return ret;
1117}
1118EXPORT_SYMBOL_GPL(direct_make_request);
1119
1120/**
1121 * submit_bio - submit a bio to the block device layer for I/O
1122 * @bio: The &struct bio which describes the I/O
1123 *
1124 * submit_bio() is very similar in purpose to generic_make_request(), and
1125 * uses that function to do most of the work. Both are fairly rough
1126 * interfaces; @bio must be presetup and ready for I/O.
1127 *
1128 */
1129blk_qc_t submit_bio(struct bio *bio)
1130{
1131        if (blkcg_punt_bio_submit(bio))
1132                return BLK_QC_T_NONE;
1133
1134        /*
1135         * If it's a regular read/write or a barrier with data attached,
1136         * go through the normal accounting stuff before submission.
1137         */
1138        if (bio_has_data(bio)) {
1139                unsigned int count;
1140
1141                if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1142                        count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1143                else
1144                        count = bio_sectors(bio);
1145
1146                if (op_is_write(bio_op(bio))) {
1147                        count_vm_events(PGPGOUT, count);
1148                } else {
1149                        task_io_account_read(bio->bi_iter.bi_size);
1150                        count_vm_events(PGPGIN, count);
1151                }
1152
1153                if (unlikely(block_dump)) {
1154                        char b[BDEVNAME_SIZE];
1155                        printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1156                        current->comm, task_pid_nr(current),
1157                                op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1158                                (unsigned long long)bio->bi_iter.bi_sector,
1159                                bio_devname(bio, b), count);
1160                }
1161        }
1162
1163        return generic_make_request(bio);
1164}
1165EXPORT_SYMBOL(submit_bio);
1166
1167/**
1168 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1169 *                              for new the queue limits
1170 * @q:  the queue
1171 * @rq: the request being checked
1172 *
1173 * Description:
1174 *    @rq may have been made based on weaker limitations of upper-level queues
1175 *    in request stacking drivers, and it may violate the limitation of @q.
1176 *    Since the block layer and the underlying device driver trust @rq
1177 *    after it is inserted to @q, it should be checked against @q before
1178 *    the insertion using this generic function.
1179 *
1180 *    Request stacking drivers like request-based dm may change the queue
1181 *    limits when retrying requests on other queues. Those requests need
1182 *    to be checked against the new queue limits again during dispatch.
1183 */
1184static int blk_cloned_rq_check_limits(struct request_queue *q,
1185                                      struct request *rq)
1186{
1187        if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1188                printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1189                        __func__, blk_rq_sectors(rq),
1190                        blk_queue_get_max_sectors(q, req_op(rq)));
1191                return -EIO;
1192        }
1193
1194        /*
1195         * queue's settings related to segment counting like q->bounce_pfn
1196         * may differ from that of other stacking queues.
1197         * Recalculate it to check the request correctly on this queue's
1198         * limitation.
1199         */
1200        rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1201        if (rq->nr_phys_segments > queue_max_segments(q)) {
1202                printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1203                        __func__, rq->nr_phys_segments, queue_max_segments(q));
1204                return -EIO;
1205        }
1206
1207        return 0;
1208}
1209
1210/**
1211 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1212 * @q:  the queue to submit the request
1213 * @rq: the request being queued
1214 */
1215blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1216{
1217        if (blk_cloned_rq_check_limits(q, rq))
1218                return BLK_STS_IOERR;
1219
1220        if (rq->rq_disk &&
1221            should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1222                return BLK_STS_IOERR;
1223
1224        if (blk_queue_io_stat(q))
1225                blk_account_io_start(rq, true);
1226
1227        /*
1228         * Since we have a scheduler attached on the top device,
1229         * bypass a potential scheduler on the bottom device for
1230         * insert.
1231         */
1232        return blk_mq_request_issue_directly(rq, true);
1233}
1234EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1235
1236/**
1237 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1238 * @rq: request to examine
1239 *
1240 * Description:
1241 *     A request could be merge of IOs which require different failure
1242 *     handling.  This function determines the number of bytes which
1243 *     can be failed from the beginning of the request without
1244 *     crossing into area which need to be retried further.
1245 *
1246 * Return:
1247 *     The number of bytes to fail.
1248 */
1249unsigned int blk_rq_err_bytes(const struct request *rq)
1250{
1251        unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1252        unsigned int bytes = 0;
1253        struct bio *bio;
1254
1255        if (!(rq->rq_flags & RQF_MIXED_MERGE))
1256                return blk_rq_bytes(rq);
1257
1258        /*
1259         * Currently the only 'mixing' which can happen is between
1260         * different fastfail types.  We can safely fail portions
1261         * which have all the failfast bits that the first one has -
1262         * the ones which are at least as eager to fail as the first
1263         * one.
1264         */
1265        for (bio = rq->bio; bio; bio = bio->bi_next) {
1266                if ((bio->bi_opf & ff) != ff)
1267                        break;
1268                bytes += bio->bi_iter.bi_size;
1269        }
1270
1271        /* this could lead to infinite loop */
1272        BUG_ON(blk_rq_bytes(rq) && !bytes);
1273        return bytes;
1274}
1275EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1276
1277void blk_account_io_completion(struct request *req, unsigned int bytes)
1278{
1279        if (blk_do_io_stat(req)) {
1280                const int sgrp = op_stat_group(req_op(req));
1281                struct hd_struct *part;
1282
1283                part_stat_lock();
1284                part = req->part;
1285                part_stat_add(part, sectors[sgrp], bytes >> 9);
1286                part_stat_unlock();
1287        }
1288}
1289
1290void blk_account_io_done(struct request *req, u64 now)
1291{
1292        /*
1293         * Account IO completion.  flush_rq isn't accounted as a
1294         * normal IO on queueing nor completion.  Accounting the
1295         * containing request is enough.
1296         */
1297        if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1298                const int sgrp = op_stat_group(req_op(req));
1299                struct hd_struct *part;
1300
1301                part_stat_lock();
1302                part = req->part;
1303
1304                update_io_ticks(part, jiffies);
1305                part_stat_inc(part, ios[sgrp]);
1306                part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1307                part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1308                part_dec_in_flight(req->q, part, rq_data_dir(req));
1309
1310                hd_struct_put(part);
1311                part_stat_unlock();
1312        }
1313}
1314
1315void blk_account_io_start(struct request *rq, bool new_io)
1316{
1317        struct hd_struct *part;
1318        int rw = rq_data_dir(rq);
1319
1320        if (!blk_do_io_stat(rq))
1321                return;
1322
1323        part_stat_lock();
1324
1325        if (!new_io) {
1326                part = rq->part;
1327                part_stat_inc(part, merges[rw]);
1328        } else {
1329                part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1330                if (!hd_struct_try_get(part)) {
1331                        /*
1332                         * The partition is already being removed,
1333                         * the request will be accounted on the disk only
1334                         *
1335                         * We take a reference on disk->part0 although that
1336                         * partition will never be deleted, so we can treat
1337                         * it as any other partition.
1338                         */
1339                        part = &rq->rq_disk->part0;
1340                        hd_struct_get(part);
1341                }
1342                part_inc_in_flight(rq->q, part, rw);
1343                rq->part = part;
1344        }
1345
1346        update_io_ticks(part, jiffies);
1347
1348        part_stat_unlock();
1349}
1350
1351/*
1352 * Steal bios from a request and add them to a bio list.
1353 * The request must not have been partially completed before.
1354 */
1355void blk_steal_bios(struct bio_list *list, struct request *rq)
1356{
1357        if (rq->bio) {
1358                if (list->tail)
1359                        list->tail->bi_next = rq->bio;
1360                else
1361                        list->head = rq->bio;
1362                list->tail = rq->biotail;
1363
1364                rq->bio = NULL;
1365                rq->biotail = NULL;
1366        }
1367
1368        rq->__data_len = 0;
1369}
1370EXPORT_SYMBOL_GPL(blk_steal_bios);
1371
1372/**
1373 * blk_update_request - Special helper function for request stacking drivers
1374 * @req:      the request being processed
1375 * @error:    block status code
1376 * @nr_bytes: number of bytes to complete @req
1377 *
1378 * Description:
1379 *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1380 *     the request structure even if @req doesn't have leftover.
1381 *     If @req has leftover, sets it up for the next range of segments.
1382 *
1383 *     This special helper function is only for request stacking drivers
1384 *     (e.g. request-based dm) so that they can handle partial completion.
1385 *     Actual device drivers should use blk_mq_end_request instead.
1386 *
1387 *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1388 *     %false return from this function.
1389 *
1390 * Note:
1391 *      The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1392 *      blk_rq_bytes() and in blk_update_request().
1393 *
1394 * Return:
1395 *     %false - this request doesn't have any more data
1396 *     %true  - this request has more data
1397 **/
1398bool blk_update_request(struct request *req, blk_status_t error,
1399                unsigned int nr_bytes)
1400{
1401        int total_bytes;
1402
1403        trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1404
1405        if (!req->bio)
1406                return false;
1407
1408        if (unlikely(error && !blk_rq_is_passthrough(req) &&
1409                     !(req->rq_flags & RQF_QUIET)))
1410                print_req_error(req, error, __func__);
1411
1412        blk_account_io_completion(req, nr_bytes);
1413
1414        total_bytes = 0;
1415        while (req->bio) {
1416                struct bio *bio = req->bio;
1417                unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1418
1419                if (bio_bytes == bio->bi_iter.bi_size)
1420                        req->bio = bio->bi_next;
1421
1422                /* Completion has already been traced */
1423                bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1424                req_bio_endio(req, bio, bio_bytes, error);
1425
1426                total_bytes += bio_bytes;
1427                nr_bytes -= bio_bytes;
1428
1429                if (!nr_bytes)
1430                        break;
1431        }
1432
1433        /*
1434         * completely done
1435         */
1436        if (!req->bio) {
1437                /*
1438                 * Reset counters so that the request stacking driver
1439                 * can find how many bytes remain in the request
1440                 * later.
1441                 */
1442                req->__data_len = 0;
1443                return false;
1444        }
1445
1446        req->__data_len -= total_bytes;
1447
1448        /* update sector only for requests with clear definition of sector */
1449        if (!blk_rq_is_passthrough(req))
1450                req->__sector += total_bytes >> 9;
1451
1452        /* mixed attributes always follow the first bio */
1453        if (req->rq_flags & RQF_MIXED_MERGE) {
1454                req->cmd_flags &= ~REQ_FAILFAST_MASK;
1455                req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1456        }
1457
1458        if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1459                /*
1460                 * If total number of sectors is less than the first segment
1461                 * size, something has gone terribly wrong.
1462                 */
1463                if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1464                        blk_dump_rq_flags(req, "request botched");
1465                        req->__data_len = blk_rq_cur_bytes(req);
1466                }
1467
1468                /* recalculate the number of segments */
1469                req->nr_phys_segments = blk_recalc_rq_segments(req);
1470        }
1471
1472        return true;
1473}
1474EXPORT_SYMBOL_GPL(blk_update_request);
1475
1476#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1477/**
1478 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1479 * @rq: the request to be flushed
1480 *
1481 * Description:
1482 *     Flush all pages in @rq.
1483 */
1484void rq_flush_dcache_pages(struct request *rq)
1485{
1486        struct req_iterator iter;
1487        struct bio_vec bvec;
1488
1489        rq_for_each_segment(bvec, rq, iter)
1490                flush_dcache_page(bvec.bv_page);
1491}
1492EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1493#endif
1494
1495/**
1496 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1497 * @q : the queue of the device being checked
1498 *
1499 * Description:
1500 *    Check if underlying low-level drivers of a device are busy.
1501 *    If the drivers want to export their busy state, they must set own
1502 *    exporting function using blk_queue_lld_busy() first.
1503 *
1504 *    Basically, this function is used only by request stacking drivers
1505 *    to stop dispatching requests to underlying devices when underlying
1506 *    devices are busy.  This behavior helps more I/O merging on the queue
1507 *    of the request stacking driver and prevents I/O throughput regression
1508 *    on burst I/O load.
1509 *
1510 * Return:
1511 *    0 - Not busy (The request stacking driver should dispatch request)
1512 *    1 - Busy (The request stacking driver should stop dispatching request)
1513 */
1514int blk_lld_busy(struct request_queue *q)
1515{
1516        if (queue_is_mq(q) && q->mq_ops->busy)
1517                return q->mq_ops->busy(q);
1518
1519        return 0;
1520}
1521EXPORT_SYMBOL_GPL(blk_lld_busy);
1522
1523/**
1524 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1525 * @rq: the clone request to be cleaned up
1526 *
1527 * Description:
1528 *     Free all bios in @rq for a cloned request.
1529 */
1530void blk_rq_unprep_clone(struct request *rq)
1531{
1532        struct bio *bio;
1533
1534        while ((bio = rq->bio) != NULL) {
1535                rq->bio = bio->bi_next;
1536
1537                bio_put(bio);
1538        }
1539}
1540EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1541
1542/*
1543 * Copy attributes of the original request to the clone request.
1544 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1545 */
1546static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1547{
1548        dst->__sector = blk_rq_pos(src);
1549        dst->__data_len = blk_rq_bytes(src);
1550        if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1551                dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1552                dst->special_vec = src->special_vec;
1553        }
1554        dst->nr_phys_segments = src->nr_phys_segments;
1555        dst->ioprio = src->ioprio;
1556        dst->extra_len = src->extra_len;
1557}
1558
1559/**
1560 * blk_rq_prep_clone - Helper function to setup clone request
1561 * @rq: the request to be setup
1562 * @rq_src: original request to be cloned
1563 * @bs: bio_set that bios for clone are allocated from
1564 * @gfp_mask: memory allocation mask for bio
1565 * @bio_ctr: setup function to be called for each clone bio.
1566 *           Returns %0 for success, non %0 for failure.
1567 * @data: private data to be passed to @bio_ctr
1568 *
1569 * Description:
1570 *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1571 *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1572 *     are not copied, and copying such parts is the caller's responsibility.
1573 *     Also, pages which the original bios are pointing to are not copied
1574 *     and the cloned bios just point same pages.
1575 *     So cloned bios must be completed before original bios, which means
1576 *     the caller must complete @rq before @rq_src.
1577 */
1578int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1579                      struct bio_set *bs, gfp_t gfp_mask,
1580                      int (*bio_ctr)(struct bio *, struct bio *, void *),
1581                      void *data)
1582{
1583        struct bio *bio, *bio_src;
1584
1585        if (!bs)
1586                bs = &fs_bio_set;
1587
1588        __rq_for_each_bio(bio_src, rq_src) {
1589                bio = bio_clone_fast(bio_src, gfp_mask, bs);
1590                if (!bio)
1591                        goto free_and_out;
1592
1593                if (bio_ctr && bio_ctr(bio, bio_src, data))
1594                        goto free_and_out;
1595
1596                if (rq->bio) {
1597                        rq->biotail->bi_next = bio;
1598                        rq->biotail = bio;
1599                } else
1600                        rq->bio = rq->biotail = bio;
1601        }
1602
1603        __blk_rq_prep_clone(rq, rq_src);
1604
1605        return 0;
1606
1607free_and_out:
1608        if (bio)
1609                bio_put(bio);
1610        blk_rq_unprep_clone(rq);
1611
1612        return -ENOMEM;
1613}
1614EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1615
1616int kblockd_schedule_work(struct work_struct *work)
1617{
1618        return queue_work(kblockd_workqueue, work);
1619}
1620EXPORT_SYMBOL(kblockd_schedule_work);
1621
1622int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1623{
1624        return queue_work_on(cpu, kblockd_workqueue, work);
1625}
1626EXPORT_SYMBOL(kblockd_schedule_work_on);
1627
1628int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1629                                unsigned long delay)
1630{
1631        return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1632}
1633EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1634
1635/**
1636 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1637 * @plug:       The &struct blk_plug that needs to be initialized
1638 *
1639 * Description:
1640 *   blk_start_plug() indicates to the block layer an intent by the caller
1641 *   to submit multiple I/O requests in a batch.  The block layer may use
1642 *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1643 *   is called.  However, the block layer may choose to submit requests
1644 *   before a call to blk_finish_plug() if the number of queued I/Os
1645 *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1646 *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1647 *   the task schedules (see below).
1648 *
1649 *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1650 *   pending I/O should the task end up blocking between blk_start_plug() and
1651 *   blk_finish_plug(). This is important from a performance perspective, but
1652 *   also ensures that we don't deadlock. For instance, if the task is blocking
1653 *   for a memory allocation, memory reclaim could end up wanting to free a
1654 *   page belonging to that request that is currently residing in our private
1655 *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1656 *   this kind of deadlock.
1657 */
1658void blk_start_plug(struct blk_plug *plug)
1659{
1660        struct task_struct *tsk = current;
1661
1662        /*
1663         * If this is a nested plug, don't actually assign it.
1664         */
1665        if (tsk->plug)
1666                return;
1667
1668        INIT_LIST_HEAD(&plug->mq_list);
1669        INIT_LIST_HEAD(&plug->cb_list);
1670        plug->rq_count = 0;
1671        plug->multiple_queues = false;
1672
1673        /*
1674         * Store ordering should not be needed here, since a potential
1675         * preempt will imply a full memory barrier
1676         */
1677        tsk->plug = plug;
1678}
1679EXPORT_SYMBOL(blk_start_plug);
1680
1681static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1682{
1683        LIST_HEAD(callbacks);
1684
1685        while (!list_empty(&plug->cb_list)) {
1686                list_splice_init(&plug->cb_list, &callbacks);
1687
1688                while (!list_empty(&callbacks)) {
1689                        struct blk_plug_cb *cb = list_first_entry(&callbacks,
1690                                                          struct blk_plug_cb,
1691                                                          list);
1692                        list_del(&cb->list);
1693                        cb->callback(cb, from_schedule);
1694                }
1695        }
1696}
1697
1698struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1699                                      int size)
1700{
1701        struct blk_plug *plug = current->plug;
1702        struct blk_plug_cb *cb;
1703
1704        if (!plug)
1705                return NULL;
1706
1707        list_for_each_entry(cb, &plug->cb_list, list)
1708                if (cb->callback == unplug && cb->data == data)
1709                        return cb;
1710
1711        /* Not currently on the callback list */
1712        BUG_ON(size < sizeof(*cb));
1713        cb = kzalloc(size, GFP_ATOMIC);
1714        if (cb) {
1715                cb->data = data;
1716                cb->callback = unplug;
1717                list_add(&cb->list, &plug->cb_list);
1718        }
1719        return cb;
1720}
1721EXPORT_SYMBOL(blk_check_plugged);
1722
1723void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1724{
1725        flush_plug_callbacks(plug, from_schedule);
1726
1727        if (!list_empty(&plug->mq_list))
1728                blk_mq_flush_plug_list(plug, from_schedule);
1729}
1730
1731/**
1732 * blk_finish_plug - mark the end of a batch of submitted I/O
1733 * @plug:       The &struct blk_plug passed to blk_start_plug()
1734 *
1735 * Description:
1736 * Indicate that a batch of I/O submissions is complete.  This function
1737 * must be paired with an initial call to blk_start_plug().  The intent
1738 * is to allow the block layer to optimize I/O submission.  See the
1739 * documentation for blk_start_plug() for more information.
1740 */
1741void blk_finish_plug(struct blk_plug *plug)
1742{
1743        if (plug != current->plug)
1744                return;
1745        blk_flush_plug_list(plug, false);
1746
1747        current->plug = NULL;
1748}
1749EXPORT_SYMBOL(blk_finish_plug);
1750
1751int __init blk_dev_init(void)
1752{
1753        BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1754        BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1755                        FIELD_SIZEOF(struct request, cmd_flags));
1756        BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1757                        FIELD_SIZEOF(struct bio, bi_opf));
1758
1759        /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1760        kblockd_workqueue = alloc_workqueue("kblockd",
1761                                            WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1762        if (!kblockd_workqueue)
1763                panic("Failed to create kblockd\n");
1764
1765        blk_requestq_cachep = kmem_cache_create("request_queue",
1766                        sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1767
1768#ifdef CONFIG_DEBUG_FS
1769        blk_debugfs_root = debugfs_create_dir("block", NULL);
1770#endif
1771
1772        return 0;
1773}
1774