linux/drivers/scsi/scsi_lib.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 1999 Eric Youngdale
   4 * Copyright (C) 2014 Christoph Hellwig
   5 *
   6 *  SCSI queueing library.
   7 *      Initial versions: Eric Youngdale (eric@andante.org).
   8 *                        Based upon conversations with large numbers
   9 *                        of people at Linux Expo.
  10 */
  11
  12#include <linux/bio.h>
  13#include <linux/bitops.h>
  14#include <linux/blkdev.h>
  15#include <linux/completion.h>
  16#include <linux/kernel.h>
  17#include <linux/export.h>
  18#include <linux/init.h>
  19#include <linux/pci.h>
  20#include <linux/delay.h>
  21#include <linux/hardirq.h>
  22#include <linux/scatterlist.h>
  23#include <linux/blk-mq.h>
  24#include <linux/ratelimit.h>
  25#include <asm/unaligned.h>
  26
  27#include <scsi/scsi.h>
  28#include <scsi/scsi_cmnd.h>
  29#include <scsi/scsi_dbg.h>
  30#include <scsi/scsi_device.h>
  31#include <scsi/scsi_driver.h>
  32#include <scsi/scsi_eh.h>
  33#include <scsi/scsi_host.h>
  34#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
  35#include <scsi/scsi_dh.h>
  36
  37#include <trace/events/scsi.h>
  38
  39#include "scsi_debugfs.h"
  40#include "scsi_priv.h"
  41#include "scsi_logging.h"
  42
  43/*
  44 * Size of integrity metadata is usually small, 1 inline sg should
  45 * cover normal cases.
  46 */
  47#ifdef CONFIG_ARCH_NO_SG_CHAIN
  48#define  SCSI_INLINE_PROT_SG_CNT  0
  49#define  SCSI_INLINE_SG_CNT  0
  50#else
  51#define  SCSI_INLINE_PROT_SG_CNT  1
  52#define  SCSI_INLINE_SG_CNT  2
  53#endif
  54
  55static struct kmem_cache *scsi_sense_cache;
  56static DEFINE_MUTEX(scsi_sense_cache_mutex);
  57
  58static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
  59
  60int scsi_init_sense_cache(struct Scsi_Host *shost)
  61{
  62        int ret = 0;
  63
  64        mutex_lock(&scsi_sense_cache_mutex);
  65        if (!scsi_sense_cache) {
  66                scsi_sense_cache =
  67                        kmem_cache_create_usercopy("scsi_sense_cache",
  68                                SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
  69                                0, SCSI_SENSE_BUFFERSIZE, NULL);
  70                if (!scsi_sense_cache)
  71                        ret = -ENOMEM;
  72        }
  73        mutex_unlock(&scsi_sense_cache_mutex);
  74        return ret;
  75}
  76
  77/*
  78 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
  79 * not change behaviour from the previous unplug mechanism, experimentation
  80 * may prove this needs changing.
  81 */
  82#define SCSI_QUEUE_DELAY        3
  83
  84static void
  85scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
  86{
  87        struct Scsi_Host *host = cmd->device->host;
  88        struct scsi_device *device = cmd->device;
  89        struct scsi_target *starget = scsi_target(device);
  90
  91        /*
  92         * Set the appropriate busy bit for the device/host.
  93         *
  94         * If the host/device isn't busy, assume that something actually
  95         * completed, and that we should be able to queue a command now.
  96         *
  97         * Note that the prior mid-layer assumption that any host could
  98         * always queue at least one command is now broken.  The mid-layer
  99         * will implement a user specifiable stall (see
 100         * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
 101         * if a command is requeued with no other commands outstanding
 102         * either for the device or for the host.
 103         */
 104        switch (reason) {
 105        case SCSI_MLQUEUE_HOST_BUSY:
 106                atomic_set(&host->host_blocked, host->max_host_blocked);
 107                break;
 108        case SCSI_MLQUEUE_DEVICE_BUSY:
 109        case SCSI_MLQUEUE_EH_RETRY:
 110                atomic_set(&device->device_blocked,
 111                           device->max_device_blocked);
 112                break;
 113        case SCSI_MLQUEUE_TARGET_BUSY:
 114                atomic_set(&starget->target_blocked,
 115                           starget->max_target_blocked);
 116                break;
 117        }
 118}
 119
 120static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
 121{
 122        struct request *rq = scsi_cmd_to_rq(cmd);
 123
 124        if (rq->rq_flags & RQF_DONTPREP) {
 125                rq->rq_flags &= ~RQF_DONTPREP;
 126                scsi_mq_uninit_cmd(cmd);
 127        } else {
 128                WARN_ON_ONCE(true);
 129        }
 130        blk_mq_requeue_request(rq, true);
 131}
 132
 133/**
 134 * __scsi_queue_insert - private queue insertion
 135 * @cmd: The SCSI command being requeued
 136 * @reason:  The reason for the requeue
 137 * @unbusy: Whether the queue should be unbusied
 138 *
 139 * This is a private queue insertion.  The public interface
 140 * scsi_queue_insert() always assumes the queue should be unbusied
 141 * because it's always called before the completion.  This function is
 142 * for a requeue after completion, which should only occur in this
 143 * file.
 144 */
 145static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
 146{
 147        struct scsi_device *device = cmd->device;
 148
 149        SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
 150                "Inserting command %p into mlqueue\n", cmd));
 151
 152        scsi_set_blocked(cmd, reason);
 153
 154        /*
 155         * Decrement the counters, since these commands are no longer
 156         * active on the host/device.
 157         */
 158        if (unbusy)
 159                scsi_device_unbusy(device, cmd);
 160
 161        /*
 162         * Requeue this command.  It will go before all other commands
 163         * that are already in the queue. Schedule requeue work under
 164         * lock such that the kblockd_schedule_work() call happens
 165         * before blk_cleanup_queue() finishes.
 166         */
 167        cmd->result = 0;
 168
 169        blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true);
 170}
 171
 172/**
 173 * scsi_queue_insert - Reinsert a command in the queue.
 174 * @cmd:    command that we are adding to queue.
 175 * @reason: why we are inserting command to queue.
 176 *
 177 * We do this for one of two cases. Either the host is busy and it cannot accept
 178 * any more commands for the time being, or the device returned QUEUE_FULL and
 179 * can accept no more commands.
 180 *
 181 * Context: This could be called either from an interrupt context or a normal
 182 * process context.
 183 */
 184void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
 185{
 186        __scsi_queue_insert(cmd, reason, true);
 187}
 188
 189
 190/**
 191 * __scsi_execute - insert request and wait for the result
 192 * @sdev:       scsi device
 193 * @cmd:        scsi command
 194 * @data_direction: data direction
 195 * @buffer:     data buffer
 196 * @bufflen:    len of buffer
 197 * @sense:      optional sense buffer
 198 * @sshdr:      optional decoded sense header
 199 * @timeout:    request timeout in HZ
 200 * @retries:    number of times to retry request
 201 * @flags:      flags for ->cmd_flags
 202 * @rq_flags:   flags for ->rq_flags
 203 * @resid:      optional residual length
 204 *
 205 * Returns the scsi_cmnd result field if a command was executed, or a negative
 206 * Linux error code if we didn't get that far.
 207 */
 208int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
 209                 int data_direction, void *buffer, unsigned bufflen,
 210                 unsigned char *sense, struct scsi_sense_hdr *sshdr,
 211                 int timeout, int retries, u64 flags, req_flags_t rq_flags,
 212                 int *resid)
 213{
 214        struct request *req;
 215        struct scsi_request *rq;
 216        int ret;
 217
 218        req = blk_get_request(sdev->request_queue,
 219                        data_direction == DMA_TO_DEVICE ?
 220                        REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
 221                        rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
 222        if (IS_ERR(req))
 223                return PTR_ERR(req);
 224
 225        rq = scsi_req(req);
 226
 227        if (bufflen) {
 228                ret = blk_rq_map_kern(sdev->request_queue, req,
 229                                      buffer, bufflen, GFP_NOIO);
 230                if (ret)
 231                        goto out;
 232        }
 233        rq->cmd_len = COMMAND_SIZE(cmd[0]);
 234        memcpy(rq->cmd, cmd, rq->cmd_len);
 235        rq->retries = retries;
 236        req->timeout = timeout;
 237        req->cmd_flags |= flags;
 238        req->rq_flags |= rq_flags | RQF_QUIET;
 239
 240        /*
 241         * head injection *required* here otherwise quiesce won't work
 242         */
 243        blk_execute_rq(NULL, req, 1);
 244
 245        /*
 246         * Some devices (USB mass-storage in particular) may transfer
 247         * garbage data together with a residue indicating that the data
 248         * is invalid.  Prevent the garbage from being misinterpreted
 249         * and prevent security leaks by zeroing out the excess data.
 250         */
 251        if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
 252                memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
 253
 254        if (resid)
 255                *resid = rq->resid_len;
 256        if (sense && rq->sense_len)
 257                memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
 258        if (sshdr)
 259                scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
 260        ret = rq->result;
 261 out:
 262        blk_put_request(req);
 263
 264        return ret;
 265}
 266EXPORT_SYMBOL(__scsi_execute);
 267
 268/*
 269 * Wake up the error handler if necessary. Avoid as follows that the error
 270 * handler is not woken up if host in-flight requests number ==
 271 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
 272 * with an RCU read lock in this function to ensure that this function in
 273 * its entirety either finishes before scsi_eh_scmd_add() increases the
 274 * host_failed counter or that it notices the shost state change made by
 275 * scsi_eh_scmd_add().
 276 */
 277static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
 278{
 279        unsigned long flags;
 280
 281        rcu_read_lock();
 282        __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
 283        if (unlikely(scsi_host_in_recovery(shost))) {
 284                spin_lock_irqsave(shost->host_lock, flags);
 285                if (shost->host_failed || shost->host_eh_scheduled)
 286                        scsi_eh_wakeup(shost);
 287                spin_unlock_irqrestore(shost->host_lock, flags);
 288        }
 289        rcu_read_unlock();
 290}
 291
 292void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
 293{
 294        struct Scsi_Host *shost = sdev->host;
 295        struct scsi_target *starget = scsi_target(sdev);
 296
 297        scsi_dec_host_busy(shost, cmd);
 298
 299        if (starget->can_queue > 0)
 300                atomic_dec(&starget->target_busy);
 301
 302        sbitmap_put(&sdev->budget_map, cmd->budget_token);
 303        cmd->budget_token = -1;
 304}
 305
 306static void scsi_kick_queue(struct request_queue *q)
 307{
 308        blk_mq_run_hw_queues(q, false);
 309}
 310
 311/*
 312 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
 313 * and call blk_run_queue for all the scsi_devices on the target -
 314 * including current_sdev first.
 315 *
 316 * Called with *no* scsi locks held.
 317 */
 318static void scsi_single_lun_run(struct scsi_device *current_sdev)
 319{
 320        struct Scsi_Host *shost = current_sdev->host;
 321        struct scsi_device *sdev, *tmp;
 322        struct scsi_target *starget = scsi_target(current_sdev);
 323        unsigned long flags;
 324
 325        spin_lock_irqsave(shost->host_lock, flags);
 326        starget->starget_sdev_user = NULL;
 327        spin_unlock_irqrestore(shost->host_lock, flags);
 328
 329        /*
 330         * Call blk_run_queue for all LUNs on the target, starting with
 331         * current_sdev. We race with others (to set starget_sdev_user),
 332         * but in most cases, we will be first. Ideally, each LU on the
 333         * target would get some limited time or requests on the target.
 334         */
 335        scsi_kick_queue(current_sdev->request_queue);
 336
 337        spin_lock_irqsave(shost->host_lock, flags);
 338        if (starget->starget_sdev_user)
 339                goto out;
 340        list_for_each_entry_safe(sdev, tmp, &starget->devices,
 341                        same_target_siblings) {
 342                if (sdev == current_sdev)
 343                        continue;
 344                if (scsi_device_get(sdev))
 345                        continue;
 346
 347                spin_unlock_irqrestore(shost->host_lock, flags);
 348                scsi_kick_queue(sdev->request_queue);
 349                spin_lock_irqsave(shost->host_lock, flags);
 350
 351                scsi_device_put(sdev);
 352        }
 353 out:
 354        spin_unlock_irqrestore(shost->host_lock, flags);
 355}
 356
 357static inline bool scsi_device_is_busy(struct scsi_device *sdev)
 358{
 359        if (scsi_device_busy(sdev) >= sdev->queue_depth)
 360                return true;
 361        if (atomic_read(&sdev->device_blocked) > 0)
 362                return true;
 363        return false;
 364}
 365
 366static inline bool scsi_target_is_busy(struct scsi_target *starget)
 367{
 368        if (starget->can_queue > 0) {
 369                if (atomic_read(&starget->target_busy) >= starget->can_queue)
 370                        return true;
 371                if (atomic_read(&starget->target_blocked) > 0)
 372                        return true;
 373        }
 374        return false;
 375}
 376
 377static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
 378{
 379        if (atomic_read(&shost->host_blocked) > 0)
 380                return true;
 381        if (shost->host_self_blocked)
 382                return true;
 383        return false;
 384}
 385
 386static void scsi_starved_list_run(struct Scsi_Host *shost)
 387{
 388        LIST_HEAD(starved_list);
 389        struct scsi_device *sdev;
 390        unsigned long flags;
 391
 392        spin_lock_irqsave(shost->host_lock, flags);
 393        list_splice_init(&shost->starved_list, &starved_list);
 394
 395        while (!list_empty(&starved_list)) {
 396                struct request_queue *slq;
 397
 398                /*
 399                 * As long as shost is accepting commands and we have
 400                 * starved queues, call blk_run_queue. scsi_request_fn
 401                 * drops the queue_lock and can add us back to the
 402                 * starved_list.
 403                 *
 404                 * host_lock protects the starved_list and starved_entry.
 405                 * scsi_request_fn must get the host_lock before checking
 406                 * or modifying starved_list or starved_entry.
 407                 */
 408                if (scsi_host_is_busy(shost))
 409                        break;
 410
 411                sdev = list_entry(starved_list.next,
 412                                  struct scsi_device, starved_entry);
 413                list_del_init(&sdev->starved_entry);
 414                if (scsi_target_is_busy(scsi_target(sdev))) {
 415                        list_move_tail(&sdev->starved_entry,
 416                                       &shost->starved_list);
 417                        continue;
 418                }
 419
 420                /*
 421                 * Once we drop the host lock, a racing scsi_remove_device()
 422                 * call may remove the sdev from the starved list and destroy
 423                 * it and the queue.  Mitigate by taking a reference to the
 424                 * queue and never touching the sdev again after we drop the
 425                 * host lock.  Note: if __scsi_remove_device() invokes
 426                 * blk_cleanup_queue() before the queue is run from this
 427                 * function then blk_run_queue() will return immediately since
 428                 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
 429                 */
 430                slq = sdev->request_queue;
 431                if (!blk_get_queue(slq))
 432                        continue;
 433                spin_unlock_irqrestore(shost->host_lock, flags);
 434
 435                scsi_kick_queue(slq);
 436                blk_put_queue(slq);
 437
 438                spin_lock_irqsave(shost->host_lock, flags);
 439        }
 440        /* put any unprocessed entries back */
 441        list_splice(&starved_list, &shost->starved_list);
 442        spin_unlock_irqrestore(shost->host_lock, flags);
 443}
 444
 445/**
 446 * scsi_run_queue - Select a proper request queue to serve next.
 447 * @q:  last request's queue
 448 *
 449 * The previous command was completely finished, start a new one if possible.
 450 */
 451static void scsi_run_queue(struct request_queue *q)
 452{
 453        struct scsi_device *sdev = q->queuedata;
 454
 455        if (scsi_target(sdev)->single_lun)
 456                scsi_single_lun_run(sdev);
 457        if (!list_empty(&sdev->host->starved_list))
 458                scsi_starved_list_run(sdev->host);
 459
 460        blk_mq_run_hw_queues(q, false);
 461}
 462
 463void scsi_requeue_run_queue(struct work_struct *work)
 464{
 465        struct scsi_device *sdev;
 466        struct request_queue *q;
 467
 468        sdev = container_of(work, struct scsi_device, requeue_work);
 469        q = sdev->request_queue;
 470        scsi_run_queue(q);
 471}
 472
 473void scsi_run_host_queues(struct Scsi_Host *shost)
 474{
 475        struct scsi_device *sdev;
 476
 477        shost_for_each_device(sdev, shost)
 478                scsi_run_queue(sdev->request_queue);
 479}
 480
 481static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
 482{
 483        if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
 484                struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
 485
 486                if (drv->uninit_command)
 487                        drv->uninit_command(cmd);
 488        }
 489}
 490
 491void scsi_free_sgtables(struct scsi_cmnd *cmd)
 492{
 493        if (cmd->sdb.table.nents)
 494                sg_free_table_chained(&cmd->sdb.table,
 495                                SCSI_INLINE_SG_CNT);
 496        if (scsi_prot_sg_count(cmd))
 497                sg_free_table_chained(&cmd->prot_sdb->table,
 498                                SCSI_INLINE_PROT_SG_CNT);
 499}
 500EXPORT_SYMBOL_GPL(scsi_free_sgtables);
 501
 502static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
 503{
 504        scsi_free_sgtables(cmd);
 505        scsi_uninit_cmd(cmd);
 506}
 507
 508static void scsi_run_queue_async(struct scsi_device *sdev)
 509{
 510        if (scsi_target(sdev)->single_lun ||
 511            !list_empty(&sdev->host->starved_list)) {
 512                kblockd_schedule_work(&sdev->requeue_work);
 513        } else {
 514                /*
 515                 * smp_mb() present in sbitmap_queue_clear() or implied in
 516                 * .end_io is for ordering writing .device_busy in
 517                 * scsi_device_unbusy() and reading sdev->restarts.
 518                 */
 519                int old = atomic_read(&sdev->restarts);
 520
 521                /*
 522                 * ->restarts has to be kept as non-zero if new budget
 523                 *  contention occurs.
 524                 *
 525                 *  No need to run queue when either another re-run
 526                 *  queue wins in updating ->restarts or a new budget
 527                 *  contention occurs.
 528                 */
 529                if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
 530                        blk_mq_run_hw_queues(sdev->request_queue, true);
 531        }
 532}
 533
 534/* Returns false when no more bytes to process, true if there are more */
 535static bool scsi_end_request(struct request *req, blk_status_t error,
 536                unsigned int bytes)
 537{
 538        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
 539        struct scsi_device *sdev = cmd->device;
 540        struct request_queue *q = sdev->request_queue;
 541
 542        if (blk_update_request(req, error, bytes))
 543                return true;
 544
 545        if (blk_queue_add_random(q))
 546                add_disk_randomness(req->rq_disk);
 547
 548        if (!blk_rq_is_passthrough(req)) {
 549                WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
 550                cmd->flags &= ~SCMD_INITIALIZED;
 551        }
 552
 553        /*
 554         * Calling rcu_barrier() is not necessary here because the
 555         * SCSI error handler guarantees that the function called by
 556         * call_rcu() has been called before scsi_end_request() is
 557         * called.
 558         */
 559        destroy_rcu_head(&cmd->rcu);
 560
 561        /*
 562         * In the MQ case the command gets freed by __blk_mq_end_request,
 563         * so we have to do all cleanup that depends on it earlier.
 564         *
 565         * We also can't kick the queues from irq context, so we
 566         * will have to defer it to a workqueue.
 567         */
 568        scsi_mq_uninit_cmd(cmd);
 569
 570        /*
 571         * queue is still alive, so grab the ref for preventing it
 572         * from being cleaned up during running queue.
 573         */
 574        percpu_ref_get(&q->q_usage_counter);
 575
 576        __blk_mq_end_request(req, error);
 577
 578        scsi_run_queue_async(sdev);
 579
 580        percpu_ref_put(&q->q_usage_counter);
 581        return false;
 582}
 583
 584/**
 585 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
 586 * @cmd:        SCSI command
 587 * @result:     scsi error code
 588 *
 589 * Translate a SCSI result code into a blk_status_t value. May reset the host
 590 * byte of @cmd->result.
 591 */
 592static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
 593{
 594        switch (host_byte(result)) {
 595        case DID_OK:
 596                if (scsi_status_is_good(result))
 597                        return BLK_STS_OK;
 598                return BLK_STS_IOERR;
 599        case DID_TRANSPORT_FAILFAST:
 600        case DID_TRANSPORT_MARGINAL:
 601                return BLK_STS_TRANSPORT;
 602        case DID_TARGET_FAILURE:
 603                set_host_byte(cmd, DID_OK);
 604                return BLK_STS_TARGET;
 605        case DID_NEXUS_FAILURE:
 606                set_host_byte(cmd, DID_OK);
 607                return BLK_STS_NEXUS;
 608        case DID_ALLOC_FAILURE:
 609                set_host_byte(cmd, DID_OK);
 610                return BLK_STS_NOSPC;
 611        case DID_MEDIUM_ERROR:
 612                set_host_byte(cmd, DID_OK);
 613                return BLK_STS_MEDIUM;
 614        default:
 615                return BLK_STS_IOERR;
 616        }
 617}
 618
 619/* Helper for scsi_io_completion() when "reprep" action required. */
 620static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
 621                                      struct request_queue *q)
 622{
 623        /* A new command will be prepared and issued. */
 624        scsi_mq_requeue_cmd(cmd);
 625}
 626
 627static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
 628{
 629        struct request *req = scsi_cmd_to_rq(cmd);
 630        unsigned long wait_for;
 631
 632        if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
 633                return false;
 634
 635        wait_for = (cmd->allowed + 1) * req->timeout;
 636        if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
 637                scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
 638                            wait_for/HZ);
 639                return true;
 640        }
 641        return false;
 642}
 643
 644/* Helper for scsi_io_completion() when special action required. */
 645static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
 646{
 647        struct request_queue *q = cmd->device->request_queue;
 648        struct request *req = scsi_cmd_to_rq(cmd);
 649        int level = 0;
 650        enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
 651              ACTION_DELAYED_RETRY} action;
 652        struct scsi_sense_hdr sshdr;
 653        bool sense_valid;
 654        bool sense_current = true;      /* false implies "deferred sense" */
 655        blk_status_t blk_stat;
 656
 657        sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
 658        if (sense_valid)
 659                sense_current = !scsi_sense_is_deferred(&sshdr);
 660
 661        blk_stat = scsi_result_to_blk_status(cmd, result);
 662
 663        if (host_byte(result) == DID_RESET) {
 664                /* Third party bus reset or reset for error recovery
 665                 * reasons.  Just retry the command and see what
 666                 * happens.
 667                 */
 668                action = ACTION_RETRY;
 669        } else if (sense_valid && sense_current) {
 670                switch (sshdr.sense_key) {
 671                case UNIT_ATTENTION:
 672                        if (cmd->device->removable) {
 673                                /* Detected disc change.  Set a bit
 674                                 * and quietly refuse further access.
 675                                 */
 676                                cmd->device->changed = 1;
 677                                action = ACTION_FAIL;
 678                        } else {
 679                                /* Must have been a power glitch, or a
 680                                 * bus reset.  Could not have been a
 681                                 * media change, so we just retry the
 682                                 * command and see what happens.
 683                                 */
 684                                action = ACTION_RETRY;
 685                        }
 686                        break;
 687                case ILLEGAL_REQUEST:
 688                        /* If we had an ILLEGAL REQUEST returned, then
 689                         * we may have performed an unsupported
 690                         * command.  The only thing this should be
 691                         * would be a ten byte read where only a six
 692                         * byte read was supported.  Also, on a system
 693                         * where READ CAPACITY failed, we may have
 694                         * read past the end of the disk.
 695                         */
 696                        if ((cmd->device->use_10_for_rw &&
 697                            sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
 698                            (cmd->cmnd[0] == READ_10 ||
 699                             cmd->cmnd[0] == WRITE_10)) {
 700                                /* This will issue a new 6-byte command. */
 701                                cmd->device->use_10_for_rw = 0;
 702                                action = ACTION_REPREP;
 703                        } else if (sshdr.asc == 0x10) /* DIX */ {
 704                                action = ACTION_FAIL;
 705                                blk_stat = BLK_STS_PROTECTION;
 706                        /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
 707                        } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
 708                                action = ACTION_FAIL;
 709                                blk_stat = BLK_STS_TARGET;
 710                        } else
 711                                action = ACTION_FAIL;
 712                        break;
 713                case ABORTED_COMMAND:
 714                        action = ACTION_FAIL;
 715                        if (sshdr.asc == 0x10) /* DIF */
 716                                blk_stat = BLK_STS_PROTECTION;
 717                        break;
 718                case NOT_READY:
 719                        /* If the device is in the process of becoming
 720                         * ready, or has a temporary blockage, retry.
 721                         */
 722                        if (sshdr.asc == 0x04) {
 723                                switch (sshdr.ascq) {
 724                                case 0x01: /* becoming ready */
 725                                case 0x04: /* format in progress */
 726                                case 0x05: /* rebuild in progress */
 727                                case 0x06: /* recalculation in progress */
 728                                case 0x07: /* operation in progress */
 729                                case 0x08: /* Long write in progress */
 730                                case 0x09: /* self test in progress */
 731                                case 0x11: /* notify (enable spinup) required */
 732                                case 0x14: /* space allocation in progress */
 733                                case 0x1a: /* start stop unit in progress */
 734                                case 0x1b: /* sanitize in progress */
 735                                case 0x1d: /* configuration in progress */
 736                                case 0x24: /* depopulation in progress */
 737                                        action = ACTION_DELAYED_RETRY;
 738                                        break;
 739                                case 0x0a: /* ALUA state transition */
 740                                        blk_stat = BLK_STS_AGAIN;
 741                                        fallthrough;
 742                                default:
 743                                        action = ACTION_FAIL;
 744                                        break;
 745                                }
 746                        } else
 747                                action = ACTION_FAIL;
 748                        break;
 749                case VOLUME_OVERFLOW:
 750                        /* See SSC3rXX or current. */
 751                        action = ACTION_FAIL;
 752                        break;
 753                case DATA_PROTECT:
 754                        action = ACTION_FAIL;
 755                        if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
 756                            (sshdr.asc == 0x55 &&
 757                             (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
 758                                /* Insufficient zone resources */
 759                                blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
 760                        }
 761                        break;
 762                default:
 763                        action = ACTION_FAIL;
 764                        break;
 765                }
 766        } else
 767                action = ACTION_FAIL;
 768
 769        if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
 770                action = ACTION_FAIL;
 771
 772        switch (action) {
 773        case ACTION_FAIL:
 774                /* Give up and fail the remainder of the request */
 775                if (!(req->rq_flags & RQF_QUIET)) {
 776                        static DEFINE_RATELIMIT_STATE(_rs,
 777                                        DEFAULT_RATELIMIT_INTERVAL,
 778                                        DEFAULT_RATELIMIT_BURST);
 779
 780                        if (unlikely(scsi_logging_level))
 781                                level =
 782                                     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
 783                                                    SCSI_LOG_MLCOMPLETE_BITS);
 784
 785                        /*
 786                         * if logging is enabled the failure will be printed
 787                         * in scsi_log_completion(), so avoid duplicate messages
 788                         */
 789                        if (!level && __ratelimit(&_rs)) {
 790                                scsi_print_result(cmd, NULL, FAILED);
 791                                if (sense_valid)
 792                                        scsi_print_sense(cmd);
 793                                scsi_print_command(cmd);
 794                        }
 795                }
 796                if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
 797                        return;
 798                fallthrough;
 799        case ACTION_REPREP:
 800                scsi_io_completion_reprep(cmd, q);
 801                break;
 802        case ACTION_RETRY:
 803                /* Retry the same command immediately */
 804                __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
 805                break;
 806        case ACTION_DELAYED_RETRY:
 807                /* Retry the same command after a delay */
 808                __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
 809                break;
 810        }
 811}
 812
 813/*
 814 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
 815 * new result that may suppress further error checking. Also modifies
 816 * *blk_statp in some cases.
 817 */
 818static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
 819                                        blk_status_t *blk_statp)
 820{
 821        bool sense_valid;
 822        bool sense_current = true;      /* false implies "deferred sense" */
 823        struct request *req = scsi_cmd_to_rq(cmd);
 824        struct scsi_sense_hdr sshdr;
 825
 826        sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
 827        if (sense_valid)
 828                sense_current = !scsi_sense_is_deferred(&sshdr);
 829
 830        if (blk_rq_is_passthrough(req)) {
 831                if (sense_valid) {
 832                        /*
 833                         * SG_IO wants current and deferred errors
 834                         */
 835                        scsi_req(req)->sense_len =
 836                                min(8 + cmd->sense_buffer[7],
 837                                    SCSI_SENSE_BUFFERSIZE);
 838                }
 839                if (sense_current)
 840                        *blk_statp = scsi_result_to_blk_status(cmd, result);
 841        } else if (blk_rq_bytes(req) == 0 && sense_current) {
 842                /*
 843                 * Flush commands do not transfers any data, and thus cannot use
 844                 * good_bytes != blk_rq_bytes(req) as the signal for an error.
 845                 * This sets *blk_statp explicitly for the problem case.
 846                 */
 847                *blk_statp = scsi_result_to_blk_status(cmd, result);
 848        }
 849        /*
 850         * Recovered errors need reporting, but they're always treated as
 851         * success, so fiddle the result code here.  For passthrough requests
 852         * we already took a copy of the original into sreq->result which
 853         * is what gets returned to the user
 854         */
 855        if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
 856                bool do_print = true;
 857                /*
 858                 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
 859                 * skip print since caller wants ATA registers. Only occurs
 860                 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
 861                 */
 862                if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
 863                        do_print = false;
 864                else if (req->rq_flags & RQF_QUIET)
 865                        do_print = false;
 866                if (do_print)
 867                        scsi_print_sense(cmd);
 868                result = 0;
 869                /* for passthrough, *blk_statp may be set */
 870                *blk_statp = BLK_STS_OK;
 871        }
 872        /*
 873         * Another corner case: the SCSI status byte is non-zero but 'good'.
 874         * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
 875         * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
 876         * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
 877         * intermediate statuses (both obsolete in SAM-4) as good.
 878         */
 879        if ((result & 0xff) && scsi_status_is_good(result)) {
 880                result = 0;
 881                *blk_statp = BLK_STS_OK;
 882        }
 883        return result;
 884}
 885
 886/**
 887 * scsi_io_completion - Completion processing for SCSI commands.
 888 * @cmd:        command that is finished.
 889 * @good_bytes: number of processed bytes.
 890 *
 891 * We will finish off the specified number of sectors. If we are done, the
 892 * command block will be released and the queue function will be goosed. If we
 893 * are not done then we have to figure out what to do next:
 894 *
 895 *   a) We can call scsi_io_completion_reprep().  The request will be
 896 *      unprepared and put back on the queue.  Then a new command will
 897 *      be created for it.  This should be used if we made forward
 898 *      progress, or if we want to switch from READ(10) to READ(6) for
 899 *      example.
 900 *
 901 *   b) We can call scsi_io_completion_action().  The request will be
 902 *      put back on the queue and retried using the same command as
 903 *      before, possibly after a delay.
 904 *
 905 *   c) We can call scsi_end_request() with blk_stat other than
 906 *      BLK_STS_OK, to fail the remainder of the request.
 907 */
 908void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 909{
 910        int result = cmd->result;
 911        struct request_queue *q = cmd->device->request_queue;
 912        struct request *req = scsi_cmd_to_rq(cmd);
 913        blk_status_t blk_stat = BLK_STS_OK;
 914
 915        if (unlikely(result))   /* a nz result may or may not be an error */
 916                result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
 917
 918        if (unlikely(blk_rq_is_passthrough(req))) {
 919                /*
 920                 * scsi_result_to_blk_status may have reset the host_byte
 921                 */
 922                scsi_req(req)->result = cmd->result;
 923        }
 924
 925        /*
 926         * Next deal with any sectors which we were able to correctly
 927         * handle.
 928         */
 929        SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
 930                "%u sectors total, %d bytes done.\n",
 931                blk_rq_sectors(req), good_bytes));
 932
 933        /*
 934         * Failed, zero length commands always need to drop down
 935         * to retry code. Fast path should return in this block.
 936         */
 937        if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
 938                if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
 939                        return; /* no bytes remaining */
 940        }
 941
 942        /* Kill remainder if no retries. */
 943        if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
 944                if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
 945                        WARN_ONCE(true,
 946                            "Bytes remaining after failed, no-retry command");
 947                return;
 948        }
 949
 950        /*
 951         * If there had been no error, but we have leftover bytes in the
 952         * requeues just queue the command up again.
 953         */
 954        if (likely(result == 0))
 955                scsi_io_completion_reprep(cmd, q);
 956        else
 957                scsi_io_completion_action(cmd, result);
 958}
 959
 960static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
 961                struct request *rq)
 962{
 963        return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
 964               !op_is_write(req_op(rq)) &&
 965               sdev->host->hostt->dma_need_drain(rq);
 966}
 967
 968/**
 969 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
 970 * @cmd: SCSI command data structure to initialize.
 971 *
 972 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
 973 * for @cmd.
 974 *
 975 * Returns:
 976 * * BLK_STS_OK       - on success
 977 * * BLK_STS_RESOURCE - if the failure is retryable
 978 * * BLK_STS_IOERR    - if the failure is fatal
 979 */
 980blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
 981{
 982        struct scsi_device *sdev = cmd->device;
 983        struct request *rq = scsi_cmd_to_rq(cmd);
 984        unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
 985        struct scatterlist *last_sg = NULL;
 986        blk_status_t ret;
 987        bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
 988        int count;
 989
 990        if (WARN_ON_ONCE(!nr_segs))
 991                return BLK_STS_IOERR;
 992
 993        /*
 994         * Make sure there is space for the drain.  The driver must adjust
 995         * max_hw_segments to be prepared for this.
 996         */
 997        if (need_drain)
 998                nr_segs++;
 999
1000        /*
1001         * If sg table allocation fails, requeue request later.
1002         */
1003        if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1004                        cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1005                return BLK_STS_RESOURCE;
1006
1007        /*
1008         * Next, walk the list, and fill in the addresses and sizes of
1009         * each segment.
1010         */
1011        count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1012
1013        if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1014                unsigned int pad_len =
1015                        (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1016
1017                last_sg->length += pad_len;
1018                cmd->extra_len += pad_len;
1019        }
1020
1021        if (need_drain) {
1022                sg_unmark_end(last_sg);
1023                last_sg = sg_next(last_sg);
1024                sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1025                sg_mark_end(last_sg);
1026
1027                cmd->extra_len += sdev->dma_drain_len;
1028                count++;
1029        }
1030
1031        BUG_ON(count > cmd->sdb.table.nents);
1032        cmd->sdb.table.nents = count;
1033        cmd->sdb.length = blk_rq_payload_bytes(rq);
1034
1035        if (blk_integrity_rq(rq)) {
1036                struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1037                int ivecs;
1038
1039                if (WARN_ON_ONCE(!prot_sdb)) {
1040                        /*
1041                         * This can happen if someone (e.g. multipath)
1042                         * queues a command to a device on an adapter
1043                         * that does not support DIX.
1044                         */
1045                        ret = BLK_STS_IOERR;
1046                        goto out_free_sgtables;
1047                }
1048
1049                ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1050
1051                if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1052                                prot_sdb->table.sgl,
1053                                SCSI_INLINE_PROT_SG_CNT)) {
1054                        ret = BLK_STS_RESOURCE;
1055                        goto out_free_sgtables;
1056                }
1057
1058                count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1059                                                prot_sdb->table.sgl);
1060                BUG_ON(count > ivecs);
1061                BUG_ON(count > queue_max_integrity_segments(rq->q));
1062
1063                cmd->prot_sdb = prot_sdb;
1064                cmd->prot_sdb->table.nents = count;
1065        }
1066
1067        return BLK_STS_OK;
1068out_free_sgtables:
1069        scsi_free_sgtables(cmd);
1070        return ret;
1071}
1072EXPORT_SYMBOL(scsi_alloc_sgtables);
1073
1074/**
1075 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1076 * @rq: Request associated with the SCSI command to be initialized.
1077 *
1078 * This function initializes the members of struct scsi_cmnd that must be
1079 * initialized before request processing starts and that won't be
1080 * reinitialized if a SCSI command is requeued.
1081 *
1082 * Called from inside blk_get_request() for pass-through requests and from
1083 * inside scsi_init_command() for filesystem requests.
1084 */
1085static void scsi_initialize_rq(struct request *rq)
1086{
1087        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1088        struct scsi_request *req = &cmd->req;
1089
1090        memset(req->__cmd, 0, sizeof(req->__cmd));
1091        req->cmd = req->__cmd;
1092        req->cmd_len = BLK_MAX_CDB;
1093        req->sense_len = 0;
1094
1095        init_rcu_head(&cmd->rcu);
1096        cmd->jiffies_at_alloc = jiffies;
1097        cmd->retries = 0;
1098}
1099
1100/*
1101 * Only called when the request isn't completed by SCSI, and not freed by
1102 * SCSI
1103 */
1104static void scsi_cleanup_rq(struct request *rq)
1105{
1106        if (rq->rq_flags & RQF_DONTPREP) {
1107                scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1108                rq->rq_flags &= ~RQF_DONTPREP;
1109        }
1110}
1111
1112/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1113void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1114{
1115        void *buf = cmd->sense_buffer;
1116        void *prot = cmd->prot_sdb;
1117        struct request *rq = scsi_cmd_to_rq(cmd);
1118        unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1119        unsigned long jiffies_at_alloc;
1120        int retries, to_clear;
1121        bool in_flight;
1122        int budget_token = cmd->budget_token;
1123
1124        if (!blk_rq_is_passthrough(rq) && !(flags & SCMD_INITIALIZED)) {
1125                flags |= SCMD_INITIALIZED;
1126                scsi_initialize_rq(rq);
1127        }
1128
1129        jiffies_at_alloc = cmd->jiffies_at_alloc;
1130        retries = cmd->retries;
1131        in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1132        /*
1133         * Zero out the cmd, except for the embedded scsi_request. Only clear
1134         * the driver-private command data if the LLD does not supply a
1135         * function to initialize that data.
1136         */
1137        to_clear = sizeof(*cmd) - sizeof(cmd->req);
1138        if (!dev->host->hostt->init_cmd_priv)
1139                to_clear += dev->host->hostt->cmd_size;
1140        memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1141
1142        cmd->device = dev;
1143        cmd->sense_buffer = buf;
1144        cmd->prot_sdb = prot;
1145        cmd->flags = flags;
1146        INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1147        cmd->jiffies_at_alloc = jiffies_at_alloc;
1148        cmd->retries = retries;
1149        if (in_flight)
1150                __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1151        cmd->budget_token = budget_token;
1152
1153}
1154
1155static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1156                struct request *req)
1157{
1158        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1159
1160        /*
1161         * Passthrough requests may transfer data, in which case they must
1162         * a bio attached to them.  Or they might contain a SCSI command
1163         * that does not transfer data, in which case they may optionally
1164         * submit a request without an attached bio.
1165         */
1166        if (req->bio) {
1167                blk_status_t ret = scsi_alloc_sgtables(cmd);
1168                if (unlikely(ret != BLK_STS_OK))
1169                        return ret;
1170        } else {
1171                BUG_ON(blk_rq_bytes(req));
1172
1173                memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1174        }
1175
1176        cmd->cmd_len = scsi_req(req)->cmd_len;
1177        if (cmd->cmd_len == 0)
1178                cmd->cmd_len = scsi_command_size(cmd->cmnd);
1179        cmd->cmnd = scsi_req(req)->cmd;
1180        cmd->transfersize = blk_rq_bytes(req);
1181        cmd->allowed = scsi_req(req)->retries;
1182        return BLK_STS_OK;
1183}
1184
1185static blk_status_t
1186scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1187{
1188        switch (sdev->sdev_state) {
1189        case SDEV_CREATED:
1190                return BLK_STS_OK;
1191        case SDEV_OFFLINE:
1192        case SDEV_TRANSPORT_OFFLINE:
1193                /*
1194                 * If the device is offline we refuse to process any
1195                 * commands.  The device must be brought online
1196                 * before trying any recovery commands.
1197                 */
1198                if (!sdev->offline_already) {
1199                        sdev->offline_already = true;
1200                        sdev_printk(KERN_ERR, sdev,
1201                                    "rejecting I/O to offline device\n");
1202                }
1203                return BLK_STS_IOERR;
1204        case SDEV_DEL:
1205                /*
1206                 * If the device is fully deleted, we refuse to
1207                 * process any commands as well.
1208                 */
1209                sdev_printk(KERN_ERR, sdev,
1210                            "rejecting I/O to dead device\n");
1211                return BLK_STS_IOERR;
1212        case SDEV_BLOCK:
1213        case SDEV_CREATED_BLOCK:
1214                return BLK_STS_RESOURCE;
1215        case SDEV_QUIESCE:
1216                /*
1217                 * If the device is blocked we only accept power management
1218                 * commands.
1219                 */
1220                if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1221                        return BLK_STS_RESOURCE;
1222                return BLK_STS_OK;
1223        default:
1224                /*
1225                 * For any other not fully online state we only allow
1226                 * power management commands.
1227                 */
1228                if (req && !(req->rq_flags & RQF_PM))
1229                        return BLK_STS_IOERR;
1230                return BLK_STS_OK;
1231        }
1232}
1233
1234/*
1235 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1236 * and return the token else return -1.
1237 */
1238static inline int scsi_dev_queue_ready(struct request_queue *q,
1239                                  struct scsi_device *sdev)
1240{
1241        int token;
1242
1243        token = sbitmap_get(&sdev->budget_map);
1244        if (atomic_read(&sdev->device_blocked)) {
1245                if (token < 0)
1246                        goto out;
1247
1248                if (scsi_device_busy(sdev) > 1)
1249                        goto out_dec;
1250
1251                /*
1252                 * unblock after device_blocked iterates to zero
1253                 */
1254                if (atomic_dec_return(&sdev->device_blocked) > 0)
1255                        goto out_dec;
1256                SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1257                                   "unblocking device at zero depth\n"));
1258        }
1259
1260        return token;
1261out_dec:
1262        if (token >= 0)
1263                sbitmap_put(&sdev->budget_map, token);
1264out:
1265        return -1;
1266}
1267
1268/*
1269 * scsi_target_queue_ready: checks if there we can send commands to target
1270 * @sdev: scsi device on starget to check.
1271 */
1272static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1273                                           struct scsi_device *sdev)
1274{
1275        struct scsi_target *starget = scsi_target(sdev);
1276        unsigned int busy;
1277
1278        if (starget->single_lun) {
1279                spin_lock_irq(shost->host_lock);
1280                if (starget->starget_sdev_user &&
1281                    starget->starget_sdev_user != sdev) {
1282                        spin_unlock_irq(shost->host_lock);
1283                        return 0;
1284                }
1285                starget->starget_sdev_user = sdev;
1286                spin_unlock_irq(shost->host_lock);
1287        }
1288
1289        if (starget->can_queue <= 0)
1290                return 1;
1291
1292        busy = atomic_inc_return(&starget->target_busy) - 1;
1293        if (atomic_read(&starget->target_blocked) > 0) {
1294                if (busy)
1295                        goto starved;
1296
1297                /*
1298                 * unblock after target_blocked iterates to zero
1299                 */
1300                if (atomic_dec_return(&starget->target_blocked) > 0)
1301                        goto out_dec;
1302
1303                SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1304                                 "unblocking target at zero depth\n"));
1305        }
1306
1307        if (busy >= starget->can_queue)
1308                goto starved;
1309
1310        return 1;
1311
1312starved:
1313        spin_lock_irq(shost->host_lock);
1314        list_move_tail(&sdev->starved_entry, &shost->starved_list);
1315        spin_unlock_irq(shost->host_lock);
1316out_dec:
1317        if (starget->can_queue > 0)
1318                atomic_dec(&starget->target_busy);
1319        return 0;
1320}
1321
1322/*
1323 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1324 * return 0. We must end up running the queue again whenever 0 is
1325 * returned, else IO can hang.
1326 */
1327static inline int scsi_host_queue_ready(struct request_queue *q,
1328                                   struct Scsi_Host *shost,
1329                                   struct scsi_device *sdev,
1330                                   struct scsi_cmnd *cmd)
1331{
1332        if (scsi_host_in_recovery(shost))
1333                return 0;
1334
1335        if (atomic_read(&shost->host_blocked) > 0) {
1336                if (scsi_host_busy(shost) > 0)
1337                        goto starved;
1338
1339                /*
1340                 * unblock after host_blocked iterates to zero
1341                 */
1342                if (atomic_dec_return(&shost->host_blocked) > 0)
1343                        goto out_dec;
1344
1345                SCSI_LOG_MLQUEUE(3,
1346                        shost_printk(KERN_INFO, shost,
1347                                     "unblocking host at zero depth\n"));
1348        }
1349
1350        if (shost->host_self_blocked)
1351                goto starved;
1352
1353        /* We're OK to process the command, so we can't be starved */
1354        if (!list_empty(&sdev->starved_entry)) {
1355                spin_lock_irq(shost->host_lock);
1356                if (!list_empty(&sdev->starved_entry))
1357                        list_del_init(&sdev->starved_entry);
1358                spin_unlock_irq(shost->host_lock);
1359        }
1360
1361        __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1362
1363        return 1;
1364
1365starved:
1366        spin_lock_irq(shost->host_lock);
1367        if (list_empty(&sdev->starved_entry))
1368                list_add_tail(&sdev->starved_entry, &shost->starved_list);
1369        spin_unlock_irq(shost->host_lock);
1370out_dec:
1371        scsi_dec_host_busy(shost, cmd);
1372        return 0;
1373}
1374
1375/*
1376 * Busy state exporting function for request stacking drivers.
1377 *
1378 * For efficiency, no lock is taken to check the busy state of
1379 * shost/starget/sdev, since the returned value is not guaranteed and
1380 * may be changed after request stacking drivers call the function,
1381 * regardless of taking lock or not.
1382 *
1383 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1384 * needs to return 'not busy'. Otherwise, request stacking drivers
1385 * may hold requests forever.
1386 */
1387static bool scsi_mq_lld_busy(struct request_queue *q)
1388{
1389        struct scsi_device *sdev = q->queuedata;
1390        struct Scsi_Host *shost;
1391
1392        if (blk_queue_dying(q))
1393                return false;
1394
1395        shost = sdev->host;
1396
1397        /*
1398         * Ignore host/starget busy state.
1399         * Since block layer does not have a concept of fairness across
1400         * multiple queues, congestion of host/starget needs to be handled
1401         * in SCSI layer.
1402         */
1403        if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1404                return true;
1405
1406        return false;
1407}
1408
1409/*
1410 * Block layer request completion callback. May be called from interrupt
1411 * context.
1412 */
1413static void scsi_complete(struct request *rq)
1414{
1415        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1416        enum scsi_disposition disposition;
1417
1418        INIT_LIST_HEAD(&cmd->eh_entry);
1419
1420        atomic_inc(&cmd->device->iodone_cnt);
1421        if (cmd->result)
1422                atomic_inc(&cmd->device->ioerr_cnt);
1423
1424        disposition = scsi_decide_disposition(cmd);
1425        if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1426                disposition = SUCCESS;
1427
1428        scsi_log_completion(cmd, disposition);
1429
1430        switch (disposition) {
1431        case SUCCESS:
1432                scsi_finish_command(cmd);
1433                break;
1434        case NEEDS_RETRY:
1435                scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1436                break;
1437        case ADD_TO_MLQUEUE:
1438                scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1439                break;
1440        default:
1441                scsi_eh_scmd_add(cmd);
1442                break;
1443        }
1444}
1445
1446/**
1447 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1448 * @cmd: command block we are dispatching.
1449 *
1450 * Return: nonzero return request was rejected and device's queue needs to be
1451 * plugged.
1452 */
1453static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1454{
1455        struct Scsi_Host *host = cmd->device->host;
1456        int rtn = 0;
1457
1458        atomic_inc(&cmd->device->iorequest_cnt);
1459
1460        /* check if the device is still usable */
1461        if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1462                /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1463                 * returns an immediate error upwards, and signals
1464                 * that the device is no longer present */
1465                cmd->result = DID_NO_CONNECT << 16;
1466                goto done;
1467        }
1468
1469        /* Check to see if the scsi lld made this device blocked. */
1470        if (unlikely(scsi_device_blocked(cmd->device))) {
1471                /*
1472                 * in blocked state, the command is just put back on
1473                 * the device queue.  The suspend state has already
1474                 * blocked the queue so future requests should not
1475                 * occur until the device transitions out of the
1476                 * suspend state.
1477                 */
1478                SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1479                        "queuecommand : device blocked\n"));
1480                return SCSI_MLQUEUE_DEVICE_BUSY;
1481        }
1482
1483        /* Store the LUN value in cmnd, if needed. */
1484        if (cmd->device->lun_in_cdb)
1485                cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1486                               (cmd->device->lun << 5 & 0xe0);
1487
1488        scsi_log_send(cmd);
1489
1490        /*
1491         * Before we queue this command, check if the command
1492         * length exceeds what the host adapter can handle.
1493         */
1494        if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1495                SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1496                               "queuecommand : command too long. "
1497                               "cdb_size=%d host->max_cmd_len=%d\n",
1498                               cmd->cmd_len, cmd->device->host->max_cmd_len));
1499                cmd->result = (DID_ABORT << 16);
1500                goto done;
1501        }
1502
1503        if (unlikely(host->shost_state == SHOST_DEL)) {
1504                cmd->result = (DID_NO_CONNECT << 16);
1505                goto done;
1506
1507        }
1508
1509        trace_scsi_dispatch_cmd_start(cmd);
1510        rtn = host->hostt->queuecommand(host, cmd);
1511        if (rtn) {
1512                trace_scsi_dispatch_cmd_error(cmd, rtn);
1513                if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1514                    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1515                        rtn = SCSI_MLQUEUE_HOST_BUSY;
1516
1517                SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1518                        "queuecommand : request rejected\n"));
1519        }
1520
1521        return rtn;
1522 done:
1523        cmd->scsi_done(cmd);
1524        return 0;
1525}
1526
1527/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1528static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1529{
1530        return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1531                sizeof(struct scatterlist);
1532}
1533
1534static blk_status_t scsi_prepare_cmd(struct request *req)
1535{
1536        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1537        struct scsi_device *sdev = req->q->queuedata;
1538        struct Scsi_Host *shost = sdev->host;
1539        struct scatterlist *sg;
1540
1541        scsi_init_command(sdev, cmd);
1542
1543        cmd->prot_op = SCSI_PROT_NORMAL;
1544        if (blk_rq_bytes(req))
1545                cmd->sc_data_direction = rq_dma_dir(req);
1546        else
1547                cmd->sc_data_direction = DMA_NONE;
1548
1549        sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1550        cmd->sdb.table.sgl = sg;
1551
1552        if (scsi_host_get_prot(shost)) {
1553                memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1554
1555                cmd->prot_sdb->table.sgl =
1556                        (struct scatterlist *)(cmd->prot_sdb + 1);
1557        }
1558
1559        /*
1560         * Special handling for passthrough commands, which don't go to the ULP
1561         * at all:
1562         */
1563        if (blk_rq_is_passthrough(req))
1564                return scsi_setup_scsi_cmnd(sdev, req);
1565
1566        if (sdev->handler && sdev->handler->prep_fn) {
1567                blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1568
1569                if (ret != BLK_STS_OK)
1570                        return ret;
1571        }
1572
1573        cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1574        memset(cmd->cmnd, 0, BLK_MAX_CDB);
1575        return scsi_cmd_to_driver(cmd)->init_command(cmd);
1576}
1577
1578static void scsi_mq_done(struct scsi_cmnd *cmd)
1579{
1580        if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1581                return;
1582        if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1583                return;
1584        trace_scsi_dispatch_cmd_done(cmd);
1585        blk_mq_complete_request(scsi_cmd_to_rq(cmd));
1586}
1587
1588static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1589{
1590        struct scsi_device *sdev = q->queuedata;
1591
1592        sbitmap_put(&sdev->budget_map, budget_token);
1593}
1594
1595static int scsi_mq_get_budget(struct request_queue *q)
1596{
1597        struct scsi_device *sdev = q->queuedata;
1598        int token = scsi_dev_queue_ready(q, sdev);
1599
1600        if (token >= 0)
1601                return token;
1602
1603        atomic_inc(&sdev->restarts);
1604
1605        /*
1606         * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1607         * .restarts must be incremented before .device_busy is read because the
1608         * code in scsi_run_queue_async() depends on the order of these operations.
1609         */
1610        smp_mb__after_atomic();
1611
1612        /*
1613         * If all in-flight requests originated from this LUN are completed
1614         * before reading .device_busy, sdev->device_busy will be observed as
1615         * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1616         * soon. Otherwise, completion of one of these requests will observe
1617         * the .restarts flag, and the request queue will be run for handling
1618         * this request, see scsi_end_request().
1619         */
1620        if (unlikely(scsi_device_busy(sdev) == 0 &&
1621                                !scsi_device_blocked(sdev)))
1622                blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1623        return -1;
1624}
1625
1626static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1627{
1628        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1629
1630        cmd->budget_token = token;
1631}
1632
1633static int scsi_mq_get_rq_budget_token(struct request *req)
1634{
1635        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1636
1637        return cmd->budget_token;
1638}
1639
1640static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1641                         const struct blk_mq_queue_data *bd)
1642{
1643        struct request *req = bd->rq;
1644        struct request_queue *q = req->q;
1645        struct scsi_device *sdev = q->queuedata;
1646        struct Scsi_Host *shost = sdev->host;
1647        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1648        blk_status_t ret;
1649        int reason;
1650
1651        WARN_ON_ONCE(cmd->budget_token < 0);
1652
1653        /*
1654         * If the device is not in running state we will reject some or all
1655         * commands.
1656         */
1657        if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1658                ret = scsi_device_state_check(sdev, req);
1659                if (ret != BLK_STS_OK)
1660                        goto out_put_budget;
1661        }
1662
1663        ret = BLK_STS_RESOURCE;
1664        if (!scsi_target_queue_ready(shost, sdev))
1665                goto out_put_budget;
1666        if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1667                goto out_dec_target_busy;
1668
1669        if (!(req->rq_flags & RQF_DONTPREP)) {
1670                ret = scsi_prepare_cmd(req);
1671                if (ret != BLK_STS_OK)
1672                        goto out_dec_host_busy;
1673                req->rq_flags |= RQF_DONTPREP;
1674        } else {
1675                clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1676        }
1677
1678        cmd->flags &= SCMD_PRESERVED_FLAGS;
1679        if (sdev->simple_tags)
1680                cmd->flags |= SCMD_TAGGED;
1681        if (bd->last)
1682                cmd->flags |= SCMD_LAST;
1683
1684        scsi_set_resid(cmd, 0);
1685        memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1686        cmd->scsi_done = scsi_mq_done;
1687
1688        blk_mq_start_request(req);
1689        reason = scsi_dispatch_cmd(cmd);
1690        if (reason) {
1691                scsi_set_blocked(cmd, reason);
1692                ret = BLK_STS_RESOURCE;
1693                goto out_dec_host_busy;
1694        }
1695
1696        return BLK_STS_OK;
1697
1698out_dec_host_busy:
1699        scsi_dec_host_busy(shost, cmd);
1700out_dec_target_busy:
1701        if (scsi_target(sdev)->can_queue > 0)
1702                atomic_dec(&scsi_target(sdev)->target_busy);
1703out_put_budget:
1704        scsi_mq_put_budget(q, cmd->budget_token);
1705        cmd->budget_token = -1;
1706        switch (ret) {
1707        case BLK_STS_OK:
1708                break;
1709        case BLK_STS_RESOURCE:
1710        case BLK_STS_ZONE_RESOURCE:
1711                if (scsi_device_blocked(sdev))
1712                        ret = BLK_STS_DEV_RESOURCE;
1713                break;
1714        case BLK_STS_AGAIN:
1715                scsi_req(req)->result = DID_BUS_BUSY << 16;
1716                if (req->rq_flags & RQF_DONTPREP)
1717                        scsi_mq_uninit_cmd(cmd);
1718                break;
1719        default:
1720                if (unlikely(!scsi_device_online(sdev)))
1721                        scsi_req(req)->result = DID_NO_CONNECT << 16;
1722                else
1723                        scsi_req(req)->result = DID_ERROR << 16;
1724                /*
1725                 * Make sure to release all allocated resources when
1726                 * we hit an error, as we will never see this command
1727                 * again.
1728                 */
1729                if (req->rq_flags & RQF_DONTPREP)
1730                        scsi_mq_uninit_cmd(cmd);
1731                scsi_run_queue_async(sdev);
1732                break;
1733        }
1734        return ret;
1735}
1736
1737static enum blk_eh_timer_return scsi_timeout(struct request *req,
1738                bool reserved)
1739{
1740        if (reserved)
1741                return BLK_EH_RESET_TIMER;
1742        return scsi_times_out(req);
1743}
1744
1745static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1746                                unsigned int hctx_idx, unsigned int numa_node)
1747{
1748        struct Scsi_Host *shost = set->driver_data;
1749        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1750        struct scatterlist *sg;
1751        int ret = 0;
1752
1753        cmd->sense_buffer =
1754                kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1755        if (!cmd->sense_buffer)
1756                return -ENOMEM;
1757        cmd->req.sense = cmd->sense_buffer;
1758
1759        if (scsi_host_get_prot(shost)) {
1760                sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1761                        shost->hostt->cmd_size;
1762                cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1763        }
1764
1765        if (shost->hostt->init_cmd_priv) {
1766                ret = shost->hostt->init_cmd_priv(shost, cmd);
1767                if (ret < 0)
1768                        kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1769        }
1770
1771        return ret;
1772}
1773
1774static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1775                                 unsigned int hctx_idx)
1776{
1777        struct Scsi_Host *shost = set->driver_data;
1778        struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1779
1780        if (shost->hostt->exit_cmd_priv)
1781                shost->hostt->exit_cmd_priv(shost, cmd);
1782        kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1783}
1784
1785
1786static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx)
1787{
1788        struct Scsi_Host *shost = hctx->driver_data;
1789
1790        if (shost->hostt->mq_poll)
1791                return shost->hostt->mq_poll(shost, hctx->queue_num);
1792
1793        return 0;
1794}
1795
1796static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1797                          unsigned int hctx_idx)
1798{
1799        struct Scsi_Host *shost = data;
1800
1801        hctx->driver_data = shost;
1802        return 0;
1803}
1804
1805static int scsi_map_queues(struct blk_mq_tag_set *set)
1806{
1807        struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1808
1809        if (shost->hostt->map_queues)
1810                return shost->hostt->map_queues(shost);
1811        return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1812}
1813
1814void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1815{
1816        struct device *dev = shost->dma_dev;
1817
1818        /*
1819         * this limit is imposed by hardware restrictions
1820         */
1821        blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1822                                        SG_MAX_SEGMENTS));
1823
1824        if (scsi_host_prot_dma(shost)) {
1825                shost->sg_prot_tablesize =
1826                        min_not_zero(shost->sg_prot_tablesize,
1827                                     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1828                BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1829                blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1830        }
1831
1832        if (dev->dma_mask) {
1833                shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1834                                dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1835        }
1836        blk_queue_max_hw_sectors(q, shost->max_sectors);
1837        blk_queue_segment_boundary(q, shost->dma_boundary);
1838        dma_set_seg_boundary(dev, shost->dma_boundary);
1839
1840        blk_queue_max_segment_size(q, shost->max_segment_size);
1841        blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1842        dma_set_max_seg_size(dev, queue_max_segment_size(q));
1843
1844        /*
1845         * Set a reasonable default alignment:  The larger of 32-byte (dword),
1846         * which is a common minimum for HBAs, and the minimum DMA alignment,
1847         * which is set by the platform.
1848         *
1849         * Devices that require a bigger alignment can increase it later.
1850         */
1851        blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1852}
1853EXPORT_SYMBOL_GPL(__scsi_init_queue);
1854
1855static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1856        .get_budget     = scsi_mq_get_budget,
1857        .put_budget     = scsi_mq_put_budget,
1858        .queue_rq       = scsi_queue_rq,
1859        .complete       = scsi_complete,
1860        .timeout        = scsi_timeout,
1861#ifdef CONFIG_BLK_DEBUG_FS
1862        .show_rq        = scsi_show_rq,
1863#endif
1864        .init_request   = scsi_mq_init_request,
1865        .exit_request   = scsi_mq_exit_request,
1866        .initialize_rq_fn = scsi_initialize_rq,
1867        .cleanup_rq     = scsi_cleanup_rq,
1868        .busy           = scsi_mq_lld_busy,
1869        .map_queues     = scsi_map_queues,
1870        .init_hctx      = scsi_init_hctx,
1871        .poll           = scsi_mq_poll,
1872        .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1873        .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1874};
1875
1876
1877static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1878{
1879        struct Scsi_Host *shost = hctx->driver_data;
1880
1881        shost->hostt->commit_rqs(shost, hctx->queue_num);
1882}
1883
1884static const struct blk_mq_ops scsi_mq_ops = {
1885        .get_budget     = scsi_mq_get_budget,
1886        .put_budget     = scsi_mq_put_budget,
1887        .queue_rq       = scsi_queue_rq,
1888        .commit_rqs     = scsi_commit_rqs,
1889        .complete       = scsi_complete,
1890        .timeout        = scsi_timeout,
1891#ifdef CONFIG_BLK_DEBUG_FS
1892        .show_rq        = scsi_show_rq,
1893#endif
1894        .init_request   = scsi_mq_init_request,
1895        .exit_request   = scsi_mq_exit_request,
1896        .initialize_rq_fn = scsi_initialize_rq,
1897        .cleanup_rq     = scsi_cleanup_rq,
1898        .busy           = scsi_mq_lld_busy,
1899        .map_queues     = scsi_map_queues,
1900        .init_hctx      = scsi_init_hctx,
1901        .poll           = scsi_mq_poll,
1902        .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1903        .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1904};
1905
1906int scsi_mq_setup_tags(struct Scsi_Host *shost)
1907{
1908        unsigned int cmd_size, sgl_size;
1909        struct blk_mq_tag_set *tag_set = &shost->tag_set;
1910
1911        sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1912                                scsi_mq_inline_sgl_size(shost));
1913        cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1914        if (scsi_host_get_prot(shost))
1915                cmd_size += sizeof(struct scsi_data_buffer) +
1916                        sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1917
1918        memset(tag_set, 0, sizeof(*tag_set));
1919        if (shost->hostt->commit_rqs)
1920                tag_set->ops = &scsi_mq_ops;
1921        else
1922                tag_set->ops = &scsi_mq_ops_no_commit;
1923        tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1924        tag_set->nr_maps = shost->nr_maps ? : 1;
1925        tag_set->queue_depth = shost->can_queue;
1926        tag_set->cmd_size = cmd_size;
1927        tag_set->numa_node = NUMA_NO_NODE;
1928        tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1929        tag_set->flags |=
1930                BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1931        tag_set->driver_data = shost;
1932        if (shost->host_tagset)
1933                tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1934
1935        return blk_mq_alloc_tag_set(tag_set);
1936}
1937
1938void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1939{
1940        blk_mq_free_tag_set(&shost->tag_set);
1941}
1942
1943/**
1944 * scsi_device_from_queue - return sdev associated with a request_queue
1945 * @q: The request queue to return the sdev from
1946 *
1947 * Return the sdev associated with a request queue or NULL if the
1948 * request_queue does not reference a SCSI device.
1949 */
1950struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1951{
1952        struct scsi_device *sdev = NULL;
1953
1954        if (q->mq_ops == &scsi_mq_ops_no_commit ||
1955            q->mq_ops == &scsi_mq_ops)
1956                sdev = q->queuedata;
1957        if (!sdev || !get_device(&sdev->sdev_gendev))
1958                sdev = NULL;
1959
1960        return sdev;
1961}
1962
1963/**
1964 * scsi_block_requests - Utility function used by low-level drivers to prevent
1965 * further commands from being queued to the device.
1966 * @shost:  host in question
1967 *
1968 * There is no timer nor any other means by which the requests get unblocked
1969 * other than the low-level driver calling scsi_unblock_requests().
1970 */
1971void scsi_block_requests(struct Scsi_Host *shost)
1972{
1973        shost->host_self_blocked = 1;
1974}
1975EXPORT_SYMBOL(scsi_block_requests);
1976
1977/**
1978 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1979 * further commands to be queued to the device.
1980 * @shost:  host in question
1981 *
1982 * There is no timer nor any other means by which the requests get unblocked
1983 * other than the low-level driver calling scsi_unblock_requests(). This is done
1984 * as an API function so that changes to the internals of the scsi mid-layer
1985 * won't require wholesale changes to drivers that use this feature.
1986 */
1987void scsi_unblock_requests(struct Scsi_Host *shost)
1988{
1989        shost->host_self_blocked = 0;
1990        scsi_run_host_queues(shost);
1991}
1992EXPORT_SYMBOL(scsi_unblock_requests);
1993
1994void scsi_exit_queue(void)
1995{
1996        kmem_cache_destroy(scsi_sense_cache);
1997}
1998
1999/**
2000 *      scsi_mode_select - issue a mode select
2001 *      @sdev:  SCSI device to be queried
2002 *      @pf:    Page format bit (1 == standard, 0 == vendor specific)
2003 *      @sp:    Save page bit (0 == don't save, 1 == save)
2004 *      @modepage: mode page being requested
2005 *      @buffer: request buffer (may not be smaller than eight bytes)
2006 *      @len:   length of request buffer.
2007 *      @timeout: command timeout
2008 *      @retries: number of retries before failing
2009 *      @data: returns a structure abstracting the mode header data
2010 *      @sshdr: place to put sense data (or NULL if no sense to be collected).
2011 *              must be SCSI_SENSE_BUFFERSIZE big.
2012 *
2013 *      Returns zero if successful; negative error number or scsi
2014 *      status on error
2015 *
2016 */
2017int
2018scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2019                 unsigned char *buffer, int len, int timeout, int retries,
2020                 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2021{
2022        unsigned char cmd[10];
2023        unsigned char *real_buffer;
2024        int ret;
2025
2026        memset(cmd, 0, sizeof(cmd));
2027        cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2028
2029        if (sdev->use_10_for_ms) {
2030                if (len > 65535)
2031                        return -EINVAL;
2032                real_buffer = kmalloc(8 + len, GFP_KERNEL);
2033                if (!real_buffer)
2034                        return -ENOMEM;
2035                memcpy(real_buffer + 8, buffer, len);
2036                len += 8;
2037                real_buffer[0] = 0;
2038                real_buffer[1] = 0;
2039                real_buffer[2] = data->medium_type;
2040                real_buffer[3] = data->device_specific;
2041                real_buffer[4] = data->longlba ? 0x01 : 0;
2042                real_buffer[5] = 0;
2043                real_buffer[6] = data->block_descriptor_length >> 8;
2044                real_buffer[7] = data->block_descriptor_length;
2045
2046                cmd[0] = MODE_SELECT_10;
2047                cmd[7] = len >> 8;
2048                cmd[8] = len;
2049        } else {
2050                if (len > 255 || data->block_descriptor_length > 255 ||
2051                    data->longlba)
2052                        return -EINVAL;
2053
2054                real_buffer = kmalloc(4 + len, GFP_KERNEL);
2055                if (!real_buffer)
2056                        return -ENOMEM;
2057                memcpy(real_buffer + 4, buffer, len);
2058                len += 4;
2059                real_buffer[0] = 0;
2060                real_buffer[1] = data->medium_type;
2061                real_buffer[2] = data->device_specific;
2062                real_buffer[3] = data->block_descriptor_length;
2063
2064                cmd[0] = MODE_SELECT;
2065                cmd[4] = len;
2066        }
2067
2068        ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2069                               sshdr, timeout, retries, NULL);
2070        kfree(real_buffer);
2071        return ret;
2072}
2073EXPORT_SYMBOL_GPL(scsi_mode_select);
2074
2075/**
2076 *      scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2077 *      @sdev:  SCSI device to be queried
2078 *      @dbd:   set if mode sense will allow block descriptors to be returned
2079 *      @modepage: mode page being requested
2080 *      @buffer: request buffer (may not be smaller than eight bytes)
2081 *      @len:   length of request buffer.
2082 *      @timeout: command timeout
2083 *      @retries: number of retries before failing
2084 *      @data: returns a structure abstracting the mode header data
2085 *      @sshdr: place to put sense data (or NULL if no sense to be collected).
2086 *              must be SCSI_SENSE_BUFFERSIZE big.
2087 *
2088 *      Returns zero if successful, or a negative error number on failure
2089 */
2090int
2091scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2092                  unsigned char *buffer, int len, int timeout, int retries,
2093                  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2094{
2095        unsigned char cmd[12];
2096        int use_10_for_ms;
2097        int header_length;
2098        int result, retry_count = retries;
2099        struct scsi_sense_hdr my_sshdr;
2100
2101        memset(data, 0, sizeof(*data));
2102        memset(&cmd[0], 0, 12);
2103
2104        dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2105        cmd[1] = dbd & 0x18;    /* allows DBD and LLBA bits */
2106        cmd[2] = modepage;
2107
2108        /* caller might not be interested in sense, but we need it */
2109        if (!sshdr)
2110                sshdr = &my_sshdr;
2111
2112 retry:
2113        use_10_for_ms = sdev->use_10_for_ms;
2114
2115        if (use_10_for_ms) {
2116                if (len < 8)
2117                        len = 8;
2118
2119                cmd[0] = MODE_SENSE_10;
2120                cmd[8] = len;
2121                header_length = 8;
2122        } else {
2123                if (len < 4)
2124                        len = 4;
2125
2126                cmd[0] = MODE_SENSE;
2127                cmd[4] = len;
2128                header_length = 4;
2129        }
2130
2131        memset(buffer, 0, len);
2132
2133        result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2134                                  sshdr, timeout, retries, NULL);
2135        if (result < 0)
2136                return result;
2137
2138        /* This code looks awful: what it's doing is making sure an
2139         * ILLEGAL REQUEST sense return identifies the actual command
2140         * byte as the problem.  MODE_SENSE commands can return
2141         * ILLEGAL REQUEST if the code page isn't supported */
2142
2143        if (!scsi_status_is_good(result)) {
2144                if (scsi_sense_valid(sshdr)) {
2145                        if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2146                            (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2147                                /*
2148                                 * Invalid command operation code
2149                                 */
2150                                if (use_10_for_ms) {
2151                                        sdev->use_10_for_ms = 0;
2152                                        goto retry;
2153                                }
2154                        }
2155                        if (scsi_status_is_check_condition(result) &&
2156                            sshdr->sense_key == UNIT_ATTENTION &&
2157                            retry_count) {
2158                                retry_count--;
2159                                goto retry;
2160                        }
2161                }
2162                return -EIO;
2163        }
2164        if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2165                     (modepage == 6 || modepage == 8))) {
2166                /* Initio breakage? */
2167                header_length = 0;
2168                data->length = 13;
2169                data->medium_type = 0;
2170                data->device_specific = 0;
2171                data->longlba = 0;
2172                data->block_descriptor_length = 0;
2173        } else if (use_10_for_ms) {
2174                data->length = buffer[0]*256 + buffer[1] + 2;
2175                data->medium_type = buffer[2];
2176                data->device_specific = buffer[3];
2177                data->longlba = buffer[4] & 0x01;
2178                data->block_descriptor_length = buffer[6]*256
2179                        + buffer[7];
2180        } else {
2181                data->length = buffer[0] + 1;
2182                data->medium_type = buffer[1];
2183                data->device_specific = buffer[2];
2184                data->block_descriptor_length = buffer[3];
2185        }
2186        data->header_length = header_length;
2187
2188        return 0;
2189}
2190EXPORT_SYMBOL(scsi_mode_sense);
2191
2192/**
2193 *      scsi_test_unit_ready - test if unit is ready
2194 *      @sdev:  scsi device to change the state of.
2195 *      @timeout: command timeout
2196 *      @retries: number of retries before failing
2197 *      @sshdr: outpout pointer for decoded sense information.
2198 *
2199 *      Returns zero if unsuccessful or an error if TUR failed.  For
2200 *      removable media, UNIT_ATTENTION sets ->changed flag.
2201 **/
2202int
2203scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2204                     struct scsi_sense_hdr *sshdr)
2205{
2206        char cmd[] = {
2207                TEST_UNIT_READY, 0, 0, 0, 0, 0,
2208        };
2209        int result;
2210
2211        /* try to eat the UNIT_ATTENTION if there are enough retries */
2212        do {
2213                result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2214                                          timeout, 1, NULL);
2215                if (sdev->removable && scsi_sense_valid(sshdr) &&
2216                    sshdr->sense_key == UNIT_ATTENTION)
2217                        sdev->changed = 1;
2218        } while (scsi_sense_valid(sshdr) &&
2219                 sshdr->sense_key == UNIT_ATTENTION && --retries);
2220
2221        return result;
2222}
2223EXPORT_SYMBOL(scsi_test_unit_ready);
2224
2225/**
2226 *      scsi_device_set_state - Take the given device through the device state model.
2227 *      @sdev:  scsi device to change the state of.
2228 *      @state: state to change to.
2229 *
2230 *      Returns zero if successful or an error if the requested
2231 *      transition is illegal.
2232 */
2233int
2234scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2235{
2236        enum scsi_device_state oldstate = sdev->sdev_state;
2237
2238        if (state == oldstate)
2239                return 0;
2240
2241        switch (state) {
2242        case SDEV_CREATED:
2243                switch (oldstate) {
2244                case SDEV_CREATED_BLOCK:
2245                        break;
2246                default:
2247                        goto illegal;
2248                }
2249                break;
2250
2251        case SDEV_RUNNING:
2252                switch (oldstate) {
2253                case SDEV_CREATED:
2254                case SDEV_OFFLINE:
2255                case SDEV_TRANSPORT_OFFLINE:
2256                case SDEV_QUIESCE:
2257                case SDEV_BLOCK:
2258                        break;
2259                default:
2260                        goto illegal;
2261                }
2262                break;
2263
2264        case SDEV_QUIESCE:
2265                switch (oldstate) {
2266                case SDEV_RUNNING:
2267                case SDEV_OFFLINE:
2268                case SDEV_TRANSPORT_OFFLINE:
2269                        break;
2270                default:
2271                        goto illegal;
2272                }
2273                break;
2274
2275        case SDEV_OFFLINE:
2276        case SDEV_TRANSPORT_OFFLINE:
2277                switch (oldstate) {
2278                case SDEV_CREATED:
2279                case SDEV_RUNNING:
2280                case SDEV_QUIESCE:
2281                case SDEV_BLOCK:
2282                        break;
2283                default:
2284                        goto illegal;
2285                }
2286                break;
2287
2288        case SDEV_BLOCK:
2289                switch (oldstate) {
2290                case SDEV_RUNNING:
2291                case SDEV_CREATED_BLOCK:
2292                case SDEV_QUIESCE:
2293                case SDEV_OFFLINE:
2294                        break;
2295                default:
2296                        goto illegal;
2297                }
2298                break;
2299
2300        case SDEV_CREATED_BLOCK:
2301                switch (oldstate) {
2302                case SDEV_CREATED:
2303                        break;
2304                default:
2305                        goto illegal;
2306                }
2307                break;
2308
2309        case SDEV_CANCEL:
2310                switch (oldstate) {
2311                case SDEV_CREATED:
2312                case SDEV_RUNNING:
2313                case SDEV_QUIESCE:
2314                case SDEV_OFFLINE:
2315                case SDEV_TRANSPORT_OFFLINE:
2316                        break;
2317                default:
2318                        goto illegal;
2319                }
2320                break;
2321
2322        case SDEV_DEL:
2323                switch (oldstate) {
2324                case SDEV_CREATED:
2325                case SDEV_RUNNING:
2326                case SDEV_OFFLINE:
2327                case SDEV_TRANSPORT_OFFLINE:
2328                case SDEV_CANCEL:
2329                case SDEV_BLOCK:
2330                case SDEV_CREATED_BLOCK:
2331                        break;
2332                default:
2333                        goto illegal;
2334                }
2335                break;
2336
2337        }
2338        sdev->offline_already = false;
2339        sdev->sdev_state = state;
2340        return 0;
2341
2342 illegal:
2343        SCSI_LOG_ERROR_RECOVERY(1,
2344                                sdev_printk(KERN_ERR, sdev,
2345                                            "Illegal state transition %s->%s",
2346                                            scsi_device_state_name(oldstate),
2347                                            scsi_device_state_name(state))
2348                                );
2349        return -EINVAL;
2350}
2351EXPORT_SYMBOL(scsi_device_set_state);
2352
2353/**
2354 *      scsi_evt_emit - emit a single SCSI device uevent
2355 *      @sdev: associated SCSI device
2356 *      @evt: event to emit
2357 *
2358 *      Send a single uevent (scsi_event) to the associated scsi_device.
2359 */
2360static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2361{
2362        int idx = 0;
2363        char *envp[3];
2364
2365        switch (evt->evt_type) {
2366        case SDEV_EVT_MEDIA_CHANGE:
2367                envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2368                break;
2369        case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2370                scsi_rescan_device(&sdev->sdev_gendev);
2371                envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2372                break;
2373        case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2374                envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2375                break;
2376        case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2377               envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2378                break;
2379        case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2380                envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2381                break;
2382        case SDEV_EVT_LUN_CHANGE_REPORTED:
2383                envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2384                break;
2385        case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2386                envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2387                break;
2388        case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2389                envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2390                break;
2391        default:
2392                /* do nothing */
2393                break;
2394        }
2395
2396        envp[idx++] = NULL;
2397
2398        kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2399}
2400
2401/**
2402 *      scsi_evt_thread - send a uevent for each scsi event
2403 *      @work: work struct for scsi_device
2404 *
2405 *      Dispatch queued events to their associated scsi_device kobjects
2406 *      as uevents.
2407 */
2408void scsi_evt_thread(struct work_struct *work)
2409{
2410        struct scsi_device *sdev;
2411        enum scsi_device_event evt_type;
2412        LIST_HEAD(event_list);
2413
2414        sdev = container_of(work, struct scsi_device, event_work);
2415
2416        for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2417                if (test_and_clear_bit(evt_type, sdev->pending_events))
2418                        sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2419
2420        while (1) {
2421                struct scsi_event *evt;
2422                struct list_head *this, *tmp;
2423                unsigned long flags;
2424
2425                spin_lock_irqsave(&sdev->list_lock, flags);
2426                list_splice_init(&sdev->event_list, &event_list);
2427                spin_unlock_irqrestore(&sdev->list_lock, flags);
2428
2429                if (list_empty(&event_list))
2430                        break;
2431
2432                list_for_each_safe(this, tmp, &event_list) {
2433                        evt = list_entry(this, struct scsi_event, node);
2434                        list_del(&evt->node);
2435                        scsi_evt_emit(sdev, evt);
2436                        kfree(evt);
2437                }
2438        }
2439}
2440
2441/**
2442 *      sdev_evt_send - send asserted event to uevent thread
2443 *      @sdev: scsi_device event occurred on
2444 *      @evt: event to send
2445 *
2446 *      Assert scsi device event asynchronously.
2447 */
2448void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2449{
2450        unsigned long flags;
2451
2452#if 0
2453        /* FIXME: currently this check eliminates all media change events
2454         * for polled devices.  Need to update to discriminate between AN
2455         * and polled events */
2456        if (!test_bit(evt->evt_type, sdev->supported_events)) {
2457                kfree(evt);
2458                return;
2459        }
2460#endif
2461
2462        spin_lock_irqsave(&sdev->list_lock, flags);
2463        list_add_tail(&evt->node, &sdev->event_list);
2464        schedule_work(&sdev->event_work);
2465        spin_unlock_irqrestore(&sdev->list_lock, flags);
2466}
2467EXPORT_SYMBOL_GPL(sdev_evt_send);
2468
2469/**
2470 *      sdev_evt_alloc - allocate a new scsi event
2471 *      @evt_type: type of event to allocate
2472 *      @gfpflags: GFP flags for allocation
2473 *
2474 *      Allocates and returns a new scsi_event.
2475 */
2476struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2477                                  gfp_t gfpflags)
2478{
2479        struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2480        if (!evt)
2481                return NULL;
2482
2483        evt->evt_type = evt_type;
2484        INIT_LIST_HEAD(&evt->node);
2485
2486        /* evt_type-specific initialization, if any */
2487        switch (evt_type) {
2488        case SDEV_EVT_MEDIA_CHANGE:
2489        case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2490        case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2491        case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2492        case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2493        case SDEV_EVT_LUN_CHANGE_REPORTED:
2494        case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2495        case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2496        default:
2497                /* do nothing */
2498                break;
2499        }
2500
2501        return evt;
2502}
2503EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2504
2505/**
2506 *      sdev_evt_send_simple - send asserted event to uevent thread
2507 *      @sdev: scsi_device event occurred on
2508 *      @evt_type: type of event to send
2509 *      @gfpflags: GFP flags for allocation
2510 *
2511 *      Assert scsi device event asynchronously, given an event type.
2512 */
2513void sdev_evt_send_simple(struct scsi_device *sdev,
2514                          enum scsi_device_event evt_type, gfp_t gfpflags)
2515{
2516        struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2517        if (!evt) {
2518                sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2519                            evt_type);
2520                return;
2521        }
2522
2523        sdev_evt_send(sdev, evt);
2524}
2525EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2526
2527/**
2528 *      scsi_device_quiesce - Block all commands except power management.
2529 *      @sdev:  scsi device to quiesce.
2530 *
2531 *      This works by trying to transition to the SDEV_QUIESCE state
2532 *      (which must be a legal transition).  When the device is in this
2533 *      state, only power management requests will be accepted, all others will
2534 *      be deferred.
2535 *
2536 *      Must be called with user context, may sleep.
2537 *
2538 *      Returns zero if unsuccessful or an error if not.
2539 */
2540int
2541scsi_device_quiesce(struct scsi_device *sdev)
2542{
2543        struct request_queue *q = sdev->request_queue;
2544        int err;
2545
2546        /*
2547         * It is allowed to call scsi_device_quiesce() multiple times from
2548         * the same context but concurrent scsi_device_quiesce() calls are
2549         * not allowed.
2550         */
2551        WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2552
2553        if (sdev->quiesced_by == current)
2554                return 0;
2555
2556        blk_set_pm_only(q);
2557
2558        blk_mq_freeze_queue(q);
2559        /*
2560         * Ensure that the effect of blk_set_pm_only() will be visible
2561         * for percpu_ref_tryget() callers that occur after the queue
2562         * unfreeze even if the queue was already frozen before this function
2563         * was called. See also https://lwn.net/Articles/573497/.
2564         */
2565        synchronize_rcu();
2566        blk_mq_unfreeze_queue(q);
2567
2568        mutex_lock(&sdev->state_mutex);
2569        err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2570        if (err == 0)
2571                sdev->quiesced_by = current;
2572        else
2573                blk_clear_pm_only(q);
2574        mutex_unlock(&sdev->state_mutex);
2575
2576        return err;
2577}
2578EXPORT_SYMBOL(scsi_device_quiesce);
2579
2580/**
2581 *      scsi_device_resume - Restart user issued commands to a quiesced device.
2582 *      @sdev:  scsi device to resume.
2583 *
2584 *      Moves the device from quiesced back to running and restarts the
2585 *      queues.
2586 *
2587 *      Must be called with user context, may sleep.
2588 */
2589void scsi_device_resume(struct scsi_device *sdev)
2590{
2591        /* check if the device state was mutated prior to resume, and if
2592         * so assume the state is being managed elsewhere (for example
2593         * device deleted during suspend)
2594         */
2595        mutex_lock(&sdev->state_mutex);
2596        if (sdev->sdev_state == SDEV_QUIESCE)
2597                scsi_device_set_state(sdev, SDEV_RUNNING);
2598        if (sdev->quiesced_by) {
2599                sdev->quiesced_by = NULL;
2600                blk_clear_pm_only(sdev->request_queue);
2601        }
2602        mutex_unlock(&sdev->state_mutex);
2603}
2604EXPORT_SYMBOL(scsi_device_resume);
2605
2606static void
2607device_quiesce_fn(struct scsi_device *sdev, void *data)
2608{
2609        scsi_device_quiesce(sdev);
2610}
2611
2612void
2613scsi_target_quiesce(struct scsi_target *starget)
2614{
2615        starget_for_each_device(starget, NULL, device_quiesce_fn);
2616}
2617EXPORT_SYMBOL(scsi_target_quiesce);
2618
2619static void
2620device_resume_fn(struct scsi_device *sdev, void *data)
2621{
2622        scsi_device_resume(sdev);
2623}
2624
2625void
2626scsi_target_resume(struct scsi_target *starget)
2627{
2628        starget_for_each_device(starget, NULL, device_resume_fn);
2629}
2630EXPORT_SYMBOL(scsi_target_resume);
2631
2632/**
2633 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2634 * @sdev: device to block
2635 *
2636 * Pause SCSI command processing on the specified device. Does not sleep.
2637 *
2638 * Returns zero if successful or a negative error code upon failure.
2639 *
2640 * Notes:
2641 * This routine transitions the device to the SDEV_BLOCK state (which must be
2642 * a legal transition). When the device is in this state, command processing
2643 * is paused until the device leaves the SDEV_BLOCK state. See also
2644 * scsi_internal_device_unblock_nowait().
2645 */
2646int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2647{
2648        struct request_queue *q = sdev->request_queue;
2649        int err = 0;
2650
2651        err = scsi_device_set_state(sdev, SDEV_BLOCK);
2652        if (err) {
2653                err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2654
2655                if (err)
2656                        return err;
2657        }
2658
2659        /*
2660         * The device has transitioned to SDEV_BLOCK.  Stop the
2661         * block layer from calling the midlayer with this device's
2662         * request queue.
2663         */
2664        blk_mq_quiesce_queue_nowait(q);
2665        return 0;
2666}
2667EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2668
2669/**
2670 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2671 * @sdev: device to block
2672 *
2673 * Pause SCSI command processing on the specified device and wait until all
2674 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2675 *
2676 * Returns zero if successful or a negative error code upon failure.
2677 *
2678 * Note:
2679 * This routine transitions the device to the SDEV_BLOCK state (which must be
2680 * a legal transition). When the device is in this state, command processing
2681 * is paused until the device leaves the SDEV_BLOCK state. See also
2682 * scsi_internal_device_unblock().
2683 */
2684static int scsi_internal_device_block(struct scsi_device *sdev)
2685{
2686        struct request_queue *q = sdev->request_queue;
2687        int err;
2688
2689        mutex_lock(&sdev->state_mutex);
2690        err = scsi_internal_device_block_nowait(sdev);
2691        if (err == 0)
2692                blk_mq_quiesce_queue(q);
2693        mutex_unlock(&sdev->state_mutex);
2694
2695        return err;
2696}
2697
2698void scsi_start_queue(struct scsi_device *sdev)
2699{
2700        struct request_queue *q = sdev->request_queue;
2701
2702        blk_mq_unquiesce_queue(q);
2703}
2704
2705/**
2706 * scsi_internal_device_unblock_nowait - resume a device after a block request
2707 * @sdev:       device to resume
2708 * @new_state:  state to set the device to after unblocking
2709 *
2710 * Restart the device queue for a previously suspended SCSI device. Does not
2711 * sleep.
2712 *
2713 * Returns zero if successful or a negative error code upon failure.
2714 *
2715 * Notes:
2716 * This routine transitions the device to the SDEV_RUNNING state or to one of
2717 * the offline states (which must be a legal transition) allowing the midlayer
2718 * to goose the queue for this device.
2719 */
2720int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2721                                        enum scsi_device_state new_state)
2722{
2723        switch (new_state) {
2724        case SDEV_RUNNING:
2725        case SDEV_TRANSPORT_OFFLINE:
2726                break;
2727        default:
2728                return -EINVAL;
2729        }
2730
2731        /*
2732         * Try to transition the scsi device to SDEV_RUNNING or one of the
2733         * offlined states and goose the device queue if successful.
2734         */
2735        switch (sdev->sdev_state) {
2736        case SDEV_BLOCK:
2737        case SDEV_TRANSPORT_OFFLINE:
2738                sdev->sdev_state = new_state;
2739                break;
2740        case SDEV_CREATED_BLOCK:
2741                if (new_state == SDEV_TRANSPORT_OFFLINE ||
2742                    new_state == SDEV_OFFLINE)
2743                        sdev->sdev_state = new_state;
2744                else
2745                        sdev->sdev_state = SDEV_CREATED;
2746                break;
2747        case SDEV_CANCEL:
2748        case SDEV_OFFLINE:
2749                break;
2750        default:
2751                return -EINVAL;
2752        }
2753        scsi_start_queue(sdev);
2754
2755        return 0;
2756}
2757EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2758
2759/**
2760 * scsi_internal_device_unblock - resume a device after a block request
2761 * @sdev:       device to resume
2762 * @new_state:  state to set the device to after unblocking
2763 *
2764 * Restart the device queue for a previously suspended SCSI device. May sleep.
2765 *
2766 * Returns zero if successful or a negative error code upon failure.
2767 *
2768 * Notes:
2769 * This routine transitions the device to the SDEV_RUNNING state or to one of
2770 * the offline states (which must be a legal transition) allowing the midlayer
2771 * to goose the queue for this device.
2772 */
2773static int scsi_internal_device_unblock(struct scsi_device *sdev,
2774                                        enum scsi_device_state new_state)
2775{
2776        int ret;
2777
2778        mutex_lock(&sdev->state_mutex);
2779        ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2780        mutex_unlock(&sdev->state_mutex);
2781
2782        return ret;
2783}
2784
2785static void
2786device_block(struct scsi_device *sdev, void *data)
2787{
2788        int ret;
2789
2790        ret = scsi_internal_device_block(sdev);
2791
2792        WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2793                  dev_name(&sdev->sdev_gendev), ret);
2794}
2795
2796static int
2797target_block(struct device *dev, void *data)
2798{
2799        if (scsi_is_target_device(dev))
2800                starget_for_each_device(to_scsi_target(dev), NULL,
2801                                        device_block);
2802        return 0;
2803}
2804
2805void
2806scsi_target_block(struct device *dev)
2807{
2808        if (scsi_is_target_device(dev))
2809                starget_for_each_device(to_scsi_target(dev), NULL,
2810                                        device_block);
2811        else
2812                device_for_each_child(dev, NULL, target_block);
2813}
2814EXPORT_SYMBOL_GPL(scsi_target_block);
2815
2816static void
2817device_unblock(struct scsi_device *sdev, void *data)
2818{
2819        scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2820}
2821
2822static int
2823target_unblock(struct device *dev, void *data)
2824{
2825        if (scsi_is_target_device(dev))
2826                starget_for_each_device(to_scsi_target(dev), data,
2827                                        device_unblock);
2828        return 0;
2829}
2830
2831void
2832scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2833{
2834        if (scsi_is_target_device(dev))
2835                starget_for_each_device(to_scsi_target(dev), &new_state,
2836                                        device_unblock);
2837        else
2838                device_for_each_child(dev, &new_state, target_unblock);
2839}
2840EXPORT_SYMBOL_GPL(scsi_target_unblock);
2841
2842int
2843scsi_host_block(struct Scsi_Host *shost)
2844{
2845        struct scsi_device *sdev;
2846        int ret = 0;
2847
2848        /*
2849         * Call scsi_internal_device_block_nowait so we can avoid
2850         * calling synchronize_rcu() for each LUN.
2851         */
2852        shost_for_each_device(sdev, shost) {
2853                mutex_lock(&sdev->state_mutex);
2854                ret = scsi_internal_device_block_nowait(sdev);
2855                mutex_unlock(&sdev->state_mutex);
2856                if (ret) {
2857                        scsi_device_put(sdev);
2858                        break;
2859                }
2860        }
2861
2862        /*
2863         * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2864         * calling synchronize_rcu() once is enough.
2865         */
2866        WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2867
2868        if (!ret)
2869                synchronize_rcu();
2870
2871        return ret;
2872}
2873EXPORT_SYMBOL_GPL(scsi_host_block);
2874
2875int
2876scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2877{
2878        struct scsi_device *sdev;
2879        int ret = 0;
2880
2881        shost_for_each_device(sdev, shost) {
2882                ret = scsi_internal_device_unblock(sdev, new_state);
2883                if (ret) {
2884                        scsi_device_put(sdev);
2885                        break;
2886                }
2887        }
2888        return ret;
2889}
2890EXPORT_SYMBOL_GPL(scsi_host_unblock);
2891
2892/**
2893 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2894 * @sgl:        scatter-gather list
2895 * @sg_count:   number of segments in sg
2896 * @offset:     offset in bytes into sg, on return offset into the mapped area
2897 * @len:        bytes to map, on return number of bytes mapped
2898 *
2899 * Returns virtual address of the start of the mapped page
2900 */
2901void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2902                          size_t *offset, size_t *len)
2903{
2904        int i;
2905        size_t sg_len = 0, len_complete = 0;
2906        struct scatterlist *sg;
2907        struct page *page;
2908
2909        WARN_ON(!irqs_disabled());
2910
2911        for_each_sg(sgl, sg, sg_count, i) {
2912                len_complete = sg_len; /* Complete sg-entries */
2913                sg_len += sg->length;
2914                if (sg_len > *offset)
2915                        break;
2916        }
2917
2918        if (unlikely(i == sg_count)) {
2919                printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2920                        "elements %d\n",
2921                       __func__, sg_len, *offset, sg_count);
2922                WARN_ON(1);
2923                return NULL;
2924        }
2925
2926        /* Offset starting from the beginning of first page in this sg-entry */
2927        *offset = *offset - len_complete + sg->offset;
2928
2929        /* Assumption: contiguous pages can be accessed as "page + i" */
2930        page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2931        *offset &= ~PAGE_MASK;
2932
2933        /* Bytes in this sg-entry from *offset to the end of the page */
2934        sg_len = PAGE_SIZE - *offset;
2935        if (*len > sg_len)
2936                *len = sg_len;
2937
2938        return kmap_atomic(page);
2939}
2940EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2941
2942/**
2943 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2944 * @virt:       virtual address to be unmapped
2945 */
2946void scsi_kunmap_atomic_sg(void *virt)
2947{
2948        kunmap_atomic(virt);
2949}
2950EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2951
2952void sdev_disable_disk_events(struct scsi_device *sdev)
2953{
2954        atomic_inc(&sdev->disk_events_disable_depth);
2955}
2956EXPORT_SYMBOL(sdev_disable_disk_events);
2957
2958void sdev_enable_disk_events(struct scsi_device *sdev)
2959{
2960        if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2961                return;
2962        atomic_dec(&sdev->disk_events_disable_depth);
2963}
2964EXPORT_SYMBOL(sdev_enable_disk_events);
2965
2966static unsigned char designator_prio(const unsigned char *d)
2967{
2968        if (d[1] & 0x30)
2969                /* not associated with LUN */
2970                return 0;
2971
2972        if (d[3] == 0)
2973                /* invalid length */
2974                return 0;
2975
2976        /*
2977         * Order of preference for lun descriptor:
2978         * - SCSI name string
2979         * - NAA IEEE Registered Extended
2980         * - EUI-64 based 16-byte
2981         * - EUI-64 based 12-byte
2982         * - NAA IEEE Registered
2983         * - NAA IEEE Extended
2984         * - EUI-64 based 8-byte
2985         * - SCSI name string (truncated)
2986         * - T10 Vendor ID
2987         * as longer descriptors reduce the likelyhood
2988         * of identification clashes.
2989         */
2990
2991        switch (d[1] & 0xf) {
2992        case 8:
2993                /* SCSI name string, variable-length UTF-8 */
2994                return 9;
2995        case 3:
2996                switch (d[4] >> 4) {
2997                case 6:
2998                        /* NAA registered extended */
2999                        return 8;
3000                case 5:
3001                        /* NAA registered */
3002                        return 5;
3003                case 4:
3004                        /* NAA extended */
3005                        return 4;
3006                case 3:
3007                        /* NAA locally assigned */
3008                        return 1;
3009                default:
3010                        break;
3011                }
3012                break;
3013        case 2:
3014                switch (d[3]) {
3015                case 16:
3016                        /* EUI64-based, 16 byte */
3017                        return 7;
3018                case 12:
3019                        /* EUI64-based, 12 byte */
3020                        return 6;
3021                case 8:
3022                        /* EUI64-based, 8 byte */
3023                        return 3;
3024                default:
3025                        break;
3026                }
3027                break;
3028        case 1:
3029                /* T10 vendor ID */
3030                return 1;
3031        default:
3032                break;
3033        }
3034
3035        return 0;
3036}
3037
3038/**
3039 * scsi_vpd_lun_id - return a unique device identification
3040 * @sdev: SCSI device
3041 * @id:   buffer for the identification
3042 * @id_len:  length of the buffer
3043 *
3044 * Copies a unique device identification into @id based
3045 * on the information in the VPD page 0x83 of the device.
3046 * The string will be formatted as a SCSI name string.
3047 *
3048 * Returns the length of the identification or error on failure.
3049 * If the identifier is longer than the supplied buffer the actual
3050 * identifier length is returned and the buffer is not zero-padded.
3051 */
3052int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3053{
3054        u8 cur_id_prio = 0;
3055        u8 cur_id_size = 0;
3056        const unsigned char *d, *cur_id_str;
3057        const struct scsi_vpd *vpd_pg83;
3058        int id_size = -EINVAL;
3059
3060        rcu_read_lock();
3061        vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3062        if (!vpd_pg83) {
3063                rcu_read_unlock();
3064                return -ENXIO;
3065        }
3066
3067        /* The id string must be at least 20 bytes + terminating NULL byte */
3068        if (id_len < 21) {
3069                rcu_read_unlock();
3070                return -EINVAL;
3071        }
3072
3073        memset(id, 0, id_len);
3074        for (d = vpd_pg83->data + 4;
3075             d < vpd_pg83->data + vpd_pg83->len;
3076             d += d[3] + 4) {
3077                u8 prio = designator_prio(d);
3078
3079                if (prio == 0 || cur_id_prio > prio)
3080                        continue;
3081
3082                switch (d[1] & 0xf) {
3083                case 0x1:
3084                        /* T10 Vendor ID */
3085                        if (cur_id_size > d[3])
3086                                break;
3087                        cur_id_prio = prio;
3088                        cur_id_size = d[3];
3089                        if (cur_id_size + 4 > id_len)
3090                                cur_id_size = id_len - 4;
3091                        cur_id_str = d + 4;
3092                        id_size = snprintf(id, id_len, "t10.%*pE",
3093                                           cur_id_size, cur_id_str);
3094                        break;
3095                case 0x2:
3096                        /* EUI-64 */
3097                        cur_id_prio = prio;
3098                        cur_id_size = d[3];
3099                        cur_id_str = d + 4;
3100                        switch (cur_id_size) {
3101                        case 8:
3102                                id_size = snprintf(id, id_len,
3103                                                   "eui.%8phN",
3104                                                   cur_id_str);
3105                                break;
3106                        case 12:
3107                                id_size = snprintf(id, id_len,
3108                                                   "eui.%12phN",
3109                                                   cur_id_str);
3110                                break;
3111                        case 16:
3112                                id_size = snprintf(id, id_len,
3113                                                   "eui.%16phN",
3114                                                   cur_id_str);
3115                                break;
3116                        default:
3117                                break;
3118                        }
3119                        break;
3120                case 0x3:
3121                        /* NAA */
3122                        cur_id_prio = prio;
3123                        cur_id_size = d[3];
3124                        cur_id_str = d + 4;
3125                        switch (cur_id_size) {
3126                        case 8:
3127                                id_size = snprintf(id, id_len,
3128                                                   "naa.%8phN",
3129                                                   cur_id_str);
3130                                break;
3131                        case 16:
3132                                id_size = snprintf(id, id_len,
3133                                                   "naa.%16phN",
3134                                                   cur_id_str);
3135                                break;
3136                        default:
3137                                break;
3138                        }
3139                        break;
3140                case 0x8:
3141                        /* SCSI name string */
3142                        if (cur_id_size > d[3])
3143                                break;
3144                        /* Prefer others for truncated descriptor */
3145                        if (d[3] > id_len) {
3146                                prio = 2;
3147                                if (cur_id_prio > prio)
3148                                        break;
3149                        }
3150                        cur_id_prio = prio;
3151                        cur_id_size = id_size = d[3];
3152                        cur_id_str = d + 4;
3153                        if (cur_id_size >= id_len)
3154                                cur_id_size = id_len - 1;
3155                        memcpy(id, cur_id_str, cur_id_size);
3156                        break;
3157                default:
3158                        break;
3159                }
3160        }
3161        rcu_read_unlock();
3162
3163        return id_size;
3164}
3165EXPORT_SYMBOL(scsi_vpd_lun_id);
3166
3167/*
3168 * scsi_vpd_tpg_id - return a target port group identifier
3169 * @sdev: SCSI device
3170 *
3171 * Returns the Target Port Group identifier from the information
3172 * froom VPD page 0x83 of the device.
3173 *
3174 * Returns the identifier or error on failure.
3175 */
3176int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3177{
3178        const unsigned char *d;
3179        const struct scsi_vpd *vpd_pg83;
3180        int group_id = -EAGAIN, rel_port = -1;
3181
3182        rcu_read_lock();
3183        vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3184        if (!vpd_pg83) {
3185                rcu_read_unlock();
3186                return -ENXIO;
3187        }
3188
3189        d = vpd_pg83->data + 4;
3190        while (d < vpd_pg83->data + vpd_pg83->len) {
3191                switch (d[1] & 0xf) {
3192                case 0x4:
3193                        /* Relative target port */
3194                        rel_port = get_unaligned_be16(&d[6]);
3195                        break;
3196                case 0x5:
3197                        /* Target port group */
3198                        group_id = get_unaligned_be16(&d[6]);
3199                        break;
3200                default:
3201                        break;
3202                }
3203                d += d[3] + 4;
3204        }
3205        rcu_read_unlock();
3206
3207        if (group_id >= 0 && rel_id && rel_port != -1)
3208                *rel_id = rel_port;
3209
3210        return group_id;
3211}
3212EXPORT_SYMBOL(scsi_vpd_tpg_id);
3213
3214/**
3215 * scsi_build_sense - build sense data for a command
3216 * @scmd:       scsi command for which the sense should be formatted
3217 * @desc:       Sense format (non-zero == descriptor format,
3218 *              0 == fixed format)
3219 * @key:        Sense key
3220 * @asc:        Additional sense code
3221 * @ascq:       Additional sense code qualifier
3222 *
3223 **/
3224void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3225{
3226        scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3227        scmd->result = SAM_STAT_CHECK_CONDITION;
3228}
3229EXPORT_SYMBOL_GPL(scsi_build_sense);
3230