drivers/scsi/scsi_error.c

   1 /*
   2  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
   3  *
   4  *  SCSI error/timeout handling
   5  *      Initial versions: Eric Youngdale.  Based upon conversations with
   6  *                        Leonard Zubkoff and David Miller at Linux Expo,
   7  *                        ideas originating from all over the place.
   8  *
   9  *      Restructured scsi_unjam_host and associated functions.
  10  *      September 04, 2002 Mike Anderson (andmike@us.ibm.com)
  11  *
  12  *      Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
  13  *      minor cleanups.
  14  *      September 30, 2002 Mike Anderson (andmike@us.ibm.com)
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/gfp.h>
  20 #include <linux/timer.h>
  21 #include <linux/string.h>
  22 #include <linux/kernel.h>
  23 #include <linux/freezer.h>
  24 #include <linux/kthread.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/delay.h>
  28 #include <linux/jiffies.h>
  29
  30 #include <scsi/scsi.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_dbg.h>
  33 #include <scsi/scsi_device.h>
  34 #include <scsi/scsi_driver.h>
  35 #include <scsi/scsi_eh.h>
  36 #include <scsi/scsi_transport.h>
  37 #include <scsi/scsi_host.h>
  38 #include <scsi/scsi_ioctl.h>
  39 #include <scsi/sg.h>
  40
  41 #include "scsi_priv.h"
  42 #include "scsi_logging.h"
  43 #include "scsi_transport_api.h"
  44
  45 #include <trace/events/scsi.h>
  46
  47 static void scsi_eh_done(struct scsi_cmnd *scmd);
  48
  49 /*
  50  * These should *probably* be handled by the host itself.
  51  * Since it is allowed to sleep, it probably should.
  52  */
  53 #define BUS_RESET_SETTLE_TIME   (10)
  54 #define HOST_RESET_SETTLE_TIME  (10)
  55
  56 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
  57 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
  58                                  struct scsi_cmnd *);
  59
  60 /* called with shost->host_lock held */
  61 void scsi_eh_wakeup(struct Scsi_Host *shost)
  62 {
  63         if (atomic_read(&shost->host_busy) == shost->host_failed) {
  64                 trace_scsi_eh_wakeup(shost);
  65                 wake_up_process(shost->ehandler);
  66                 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
  67                         "Waking error handler thread\n"));
  68         }
  69 }
  70
  71 /**
  72  * scsi_schedule_eh - schedule EH for SCSI host
  73  * @shost:      SCSI host to invoke error handling on.
  74  *
  75  * Schedule SCSI EH without scmd.
  76  */
  77 void scsi_schedule_eh(struct Scsi_Host *shost)
  78 {
  79         unsigned long flags;
  80
  81         spin_lock_irqsave(shost->host_lock, flags);
  82
  83         if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
  84             scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
  85                 shost->host_eh_scheduled++;
  86                 scsi_eh_wakeup(shost);
  87         }
  88
  89         spin_unlock_irqrestore(shost->host_lock, flags);
  90 }
  91 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
  92
  93 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
  94 {
  95         if (!shost->last_reset || shost->eh_deadline == -1)
  96                 return 0;
  97
  98         /*
  99          * 32bit accesses are guaranteed to be atomic
 100          * (on all supported architectures), so instead
 101          * of using a spinlock we can as well double check
 102          * if eh_deadline has been set to 'off' during the
 103          * time_before call.
 104          */
 105         if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
 106             shost->eh_deadline > -1)
 107                 return 0;
 108
 109         return 1;
 110 }
 111
 112 /**
 113  * scmd_eh_abort_handler - Handle command aborts
 114  * @work:       command to be aborted.
 115  */
 116 void
 117 scmd_eh_abort_handler(struct work_struct *work)
 118 {
 119         struct scsi_cmnd *scmd =
 120                 container_of(work, struct scsi_cmnd, abort_work.work);
 121         struct scsi_device *sdev = scmd->device;
 122         int rtn;
 123
 124         if (scsi_host_eh_past_deadline(sdev->host)) {
 125                 SCSI_LOG_ERROR_RECOVERY(3,
 126                         scmd_printk(KERN_INFO, scmd,
 127                                     "scmd %p eh timeout, not aborting\n",
 128                                     scmd));
 129         } else {
 130                 SCSI_LOG_ERROR_RECOVERY(3,
 131                         scmd_printk(KERN_INFO, scmd,
 132                                     "aborting command %p\n", scmd));
 133                 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
 134                 if (rtn == SUCCESS) {
 135                         set_host_byte(scmd, DID_TIME_OUT);
 136                         if (scsi_host_eh_past_deadline(sdev->host)) {
 137                                 SCSI_LOG_ERROR_RECOVERY(3,
 138                                         scmd_printk(KERN_INFO, scmd,
 139                                                     "scmd %p eh timeout, "
 140                                                     "not retrying aborted "
 141                                                     "command\n", scmd));
 142                         } else if (!scsi_noretry_cmd(scmd) &&
 143                             (++scmd->retries <= scmd->allowed)) {
 144                                 SCSI_LOG_ERROR_RECOVERY(3,
 145                                         scmd_printk(KERN_WARNING, scmd,
 146                                                     "scmd %p retry "
 147                                                     "aborted command\n", scmd));
 148                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
 149                                 return;
 150                         } else {
 151                                 SCSI_LOG_ERROR_RECOVERY(3,
 152                                         scmd_printk(KERN_WARNING, scmd,
 153                                                     "scmd %p finish "
 154                                                     "aborted command\n", scmd));
 155                                 scsi_finish_command(scmd);
 156                                 return;
 157                         }
 158                 } else {
 159                         SCSI_LOG_ERROR_RECOVERY(3,
 160                                 scmd_printk(KERN_INFO, scmd,
 161                                             "scmd %p abort %s\n", scmd,
 162                                             (rtn == FAST_IO_FAIL) ?
 163                                             "not send" : "failed"));
 164                 }
 165         }
 166
 167         if (!scsi_eh_scmd_add(scmd, 0)) {
 168                 SCSI_LOG_ERROR_RECOVERY(3,
 169                         scmd_printk(KERN_WARNING, scmd,
 170                                     "scmd %p terminate "
 171                                     "aborted command\n", scmd));
 172                 set_host_byte(scmd, DID_TIME_OUT);
 173                 scsi_finish_command(scmd);
 174         }
 175 }
 176
 177 /**
 178  * scsi_abort_command - schedule a command abort
 179  * @scmd:       scmd to abort.
 180  *
 181  * We only need to abort commands after a command timeout
 182  */
 183 static int
 184 scsi_abort_command(struct scsi_cmnd *scmd)
 185 {
 186         struct scsi_device *sdev = scmd->device;
 187         struct Scsi_Host *shost = sdev->host;
 188         unsigned long flags;
 189
 190         if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
 191                 /*
 192                  * Retry after abort failed, escalate to next level.
 193                  */
 194                 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
 195                 SCSI_LOG_ERROR_RECOVERY(3,
 196                         scmd_printk(KERN_INFO, scmd,
 197                                     "scmd %p previous abort failed\n", scmd));
 198                 BUG_ON(delayed_work_pending(&scmd->abort_work));
 199                 return FAILED;
 200         }
 201
 202         /*
 203          * Do not try a command abort if
 204          * SCSI EH has already started.
 205          */
 206         spin_lock_irqsave(shost->host_lock, flags);
 207         if (scsi_host_in_recovery(shost)) {
 208                 spin_unlock_irqrestore(shost->host_lock, flags);
 209                 SCSI_LOG_ERROR_RECOVERY(3,
 210                         scmd_printk(KERN_INFO, scmd,
 211                                     "scmd %p not aborting, host in recovery\n",
 212                                     scmd));
 213                 return FAILED;
 214         }
 215
 216         if (shost->eh_deadline != -1 && !shost->last_reset)
 217                 shost->last_reset = jiffies;
 218         spin_unlock_irqrestore(shost->host_lock, flags);
 219
 220         scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
 221         SCSI_LOG_ERROR_RECOVERY(3,
 222                 scmd_printk(KERN_INFO, scmd,
 223                             "scmd %p abort scheduled\n", scmd));
 224         queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
 225         return SUCCESS;
 226 }
 227
 228 /**
 229  * scsi_eh_scmd_add - add scsi cmd to error handling.
 230  * @scmd:       scmd to run eh on.
 231  * @eh_flag:    optional SCSI_EH flag.
 232  *
 233  * Return value:
 234  *      0 on failure.
 235  */
 236 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
 237 {
 238         struct Scsi_Host *shost = scmd->device->host;
 239         unsigned long flags;
 240         int ret = 0;
 241
 242         if (!shost->ehandler)
 243                 return 0;
 244
 245         spin_lock_irqsave(shost->host_lock, flags);
 246         if (scsi_host_set_state(shost, SHOST_RECOVERY))
 247                 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
 248                         goto out_unlock;
 249
 250         if (shost->eh_deadline != -1 && !shost->last_reset)
 251                 shost->last_reset = jiffies;
 252
 253         ret = 1;
 254         if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
 255                 eh_flag &= ~SCSI_EH_CANCEL_CMD;
 256         scmd->eh_eflags |= eh_flag;
 257         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
 258         shost->host_failed++;
 259         scsi_eh_wakeup(shost);
 260  out_unlock:
 261         spin_unlock_irqrestore(shost->host_lock, flags);
 262         return ret;
 263 }
 264
 265 /**
 266  * scsi_times_out - Timeout function for normal scsi commands.
 267  * @req:        request that is timing out.
 268  *
 269  * Notes:
 270  *     We do not need to lock this.  There is the potential for a race
 271  *     only in that the normal completion handling might run, but if the
 272  *     normal completion function determines that the timer has already
 273  *     fired, then it mustn't do anything.
 274  */
 275 enum blk_eh_timer_return scsi_times_out(struct request *req)
 276 {
 277         struct scsi_cmnd *scmd = req->special;
 278         enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
 279         struct Scsi_Host *host = scmd->device->host;
 280
 281         trace_scsi_dispatch_cmd_timeout(scmd);
 282         scsi_log_completion(scmd, TIMEOUT_ERROR);
 283
 284         if (host->eh_deadline != -1 && !host->last_reset)
 285                 host->last_reset = jiffies;
 286
 287         if (host->transportt->eh_timed_out)
 288                 rtn = host->transportt->eh_timed_out(scmd);
 289         else if (host->hostt->eh_timed_out)
 290                 rtn = host->hostt->eh_timed_out(scmd);
 291
 292         if (rtn == BLK_EH_NOT_HANDLED) {
 293                 if (!host->hostt->no_async_abort &&
 294                     scsi_abort_command(scmd) == SUCCESS)
 295                         return BLK_EH_NOT_HANDLED;
 296
 297                 set_host_byte(scmd, DID_TIME_OUT);
 298                 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
 299                         rtn = BLK_EH_HANDLED;
 300         }
 301
 302         return rtn;
 303 }
 304
 305 /**
 306  * scsi_block_when_processing_errors - Prevent cmds from being queued.
 307  * @sdev:       Device on which we are performing recovery.
 308  *
 309  * Description:
 310  *     We block until the host is out of error recovery, and then check to
 311  *     see whether the host or the device is offline.
 312  *
 313  * Return value:
 314  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
 315  */
 316 int scsi_block_when_processing_errors(struct scsi_device *sdev)
 317 {
 318         int online;
 319
 320         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
 321
 322         online = scsi_device_online(sdev);
 323
 324         SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
 325                 "%s: rtn: %d\n", __func__, online));
 326
 327         return online;
 328 }
 329 EXPORT_SYMBOL(scsi_block_when_processing_errors);
 330
 331 #ifdef CONFIG_SCSI_LOGGING
 332 /**
 333  * scsi_eh_prt_fail_stats - Log info on failures.
 334  * @shost:      scsi host being recovered.
 335  * @work_q:     Queue of scsi cmds to process.
 336  */
 337 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
 338                                           struct list_head *work_q)
 339 {
 340         struct scsi_cmnd *scmd;
 341         struct scsi_device *sdev;
 342         int total_failures = 0;
 343         int cmd_failed = 0;
 344         int cmd_cancel = 0;
 345         int devices_failed = 0;
 346
 347         shost_for_each_device(sdev, shost) {
 348                 list_for_each_entry(scmd, work_q, eh_entry) {
 349                         if (scmd->device == sdev) {
 350                                 ++total_failures;
 351                                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
 352                                         ++cmd_cancel;
 353                                 else
 354                                         ++cmd_failed;
 355                         }
 356                 }
 357
 358                 if (cmd_cancel || cmd_failed) {
 359                         SCSI_LOG_ERROR_RECOVERY(3,
 360                                 shost_printk(KERN_INFO, shost,
 361                                             "%s: cmds failed: %d, cancel: %d\n",
 362                                             __func__, cmd_failed,
 363                                             cmd_cancel));
 364                         cmd_cancel = 0;
 365                         cmd_failed = 0;
 366                         ++devices_failed;
 367                 }
 368         }
 369
 370         SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
 371                                    "Total of %d commands on %d"
 372                                    " devices require eh work\n",
 373                                    total_failures, devices_failed));
 374 }
 375 #endif
 376
 377  /**
 378  * scsi_report_lun_change - Set flag on all *other* devices on the same target
 379  *                          to indicate that a UNIT ATTENTION is expected.
 380  * @sdev:       Device reporting the UNIT ATTENTION
 381  */
 382 static void scsi_report_lun_change(struct scsi_device *sdev)
 383 {
 384         sdev->sdev_target->expecting_lun_change = 1;
 385 }
 386
 387 /**
 388  * scsi_report_sense - Examine scsi sense information and log messages for
 389  *                     certain conditions, also issue uevents for some of them.
 390  * @sdev:       Device reporting the sense code
 391  * @sshdr:      sshdr to be examined
 392  */
 393 static void scsi_report_sense(struct scsi_device *sdev,
 394                               struct scsi_sense_hdr *sshdr)
 395 {
 396         enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;     /* i.e. none */
 397
 398         if (sshdr->sense_key == UNIT_ATTENTION) {
 399                 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
 400                         evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
 401                         sdev_printk(KERN_WARNING, sdev,
 402                                     "Inquiry data has changed");
 403                 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
 404                         evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
 405                         scsi_report_lun_change(sdev);
 406                         sdev_printk(KERN_WARNING, sdev,
 407                                     "Warning! Received an indication that the "
 408                                     "LUN assignments on this target have "
 409                                     "changed. The Linux SCSI layer does not "
 410                                     "automatically remap LUN assignments.\n");
 411                 } else if (sshdr->asc == 0x3f)
 412                         sdev_printk(KERN_WARNING, sdev,
 413                                     "Warning! Received an indication that the "
 414                                     "operating parameters on this target have "
 415                                     "changed. The Linux SCSI layer does not "
 416                                     "automatically adjust these parameters.\n");
 417
 418                 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
 419                         evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
 420                         sdev_printk(KERN_WARNING, sdev,
 421                                     "Warning! Received an indication that the "
 422                                     "LUN reached a thin provisioning soft "
 423                                     "threshold.\n");
 424                 }
 425
 426                 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
 427                         evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
 428                         sdev_printk(KERN_WARNING, sdev,
 429                                     "Mode parameters changed");
 430                 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
 431                         evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
 432                         sdev_printk(KERN_WARNING, sdev,
 433                                     "Capacity data has changed");
 434                 } else if (sshdr->asc == 0x2a)
 435                         sdev_printk(KERN_WARNING, sdev,
 436                                     "Parameters changed");
 437         }
 438
 439         if (evt_type != SDEV_EVT_MAXBITS) {
 440                 set_bit(evt_type, sdev->pending_events);
 441                 schedule_work(&sdev->event_work);
 442         }
 443 }
 444
 445 /**
 446  * scsi_check_sense - Examine scsi cmd sense
 447  * @scmd:       Cmd to have sense checked.
 448  *
 449  * Return value:
 450  *      SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
 451  *
 452  * Notes:
 453  *      When a deferred error is detected the current command has
 454  *      not been executed and needs retrying.
 455  */
 456 static int scsi_check_sense(struct scsi_cmnd *scmd)
 457 {
 458         struct scsi_device *sdev = scmd->device;
 459         struct scsi_sense_hdr sshdr;
 460
 461         if (! scsi_command_normalize_sense(scmd, &sshdr))
 462                 return FAILED;  /* no valid sense data */
 463
 464         scsi_report_sense(sdev, &sshdr);
 465
 466         if (scsi_sense_is_deferred(&sshdr))
 467                 return NEEDS_RETRY;
 468
 469         if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
 470                         sdev->scsi_dh_data->scsi_dh->check_sense) {
 471                 int rc;
 472
 473                 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
 474                 if (rc != SCSI_RETURN_NOT_HANDLED)
 475                         return rc;
 476                 /* handler does not care. Drop down to default handling */
 477         }
 478
 479         if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
 480                 /*
 481                  * nasty: for mid-layer issued TURs, we need to return the
 482                  * actual sense data without any recovery attempt.  For eh
 483                  * issued ones, we need to try to recover and interpret
 484                  */
 485                 return SUCCESS;
 486
 487         /*
 488          * Previous logic looked for FILEMARK, EOM or ILI which are
 489          * mainly associated with tapes and returned SUCCESS.
 490          */
 491         if (sshdr.response_code == 0x70) {
 492                 /* fixed format */
 493                 if (scmd->sense_buffer[2] & 0xe0)
 494                         return SUCCESS;
 495         } else {
 496                 /*
 497                  * descriptor format: look for "stream commands sense data
 498                  * descriptor" (see SSC-3). Assume single sense data
 499                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
 500                  */
 501                 if ((sshdr.additional_length > 3) &&
 502                     (scmd->sense_buffer[8] == 0x4) &&
 503                     (scmd->sense_buffer[11] & 0xe0))
 504                         return SUCCESS;
 505         }
 506
 507         switch (sshdr.sense_key) {
 508         case NO_SENSE:
 509                 return SUCCESS;
 510         case RECOVERED_ERROR:
 511                 return /* soft_error */ SUCCESS;
 512
 513         case ABORTED_COMMAND:
 514                 if (sshdr.asc == 0x10) /* DIF */
 515                         return SUCCESS;
 516
 517                 return NEEDS_RETRY;
 518         case NOT_READY:
 519         case UNIT_ATTENTION:
 520                 /*
 521                  * if we are expecting a cc/ua because of a bus reset that we
 522                  * performed, treat this just as a retry.  otherwise this is
 523                  * information that we should pass up to the upper-level driver
 524                  * so that we can deal with it there.
 525                  */
 526                 if (scmd->device->expecting_cc_ua) {
 527                         /*
 528                          * Because some device does not queue unit
 529                          * attentions correctly, we carefully check
 530                          * additional sense code and qualifier so as
 531                          * not to squash media change unit attention.
 532                          */
 533                         if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
 534                                 scmd->device->expecting_cc_ua = 0;
 535                                 return NEEDS_RETRY;
 536                         }
 537                 }
 538                 /*
 539                  * we might also expect a cc/ua if another LUN on the target
 540                  * reported a UA with an ASC/ASCQ of 3F 0E -
 541                  * REPORTED LUNS DATA HAS CHANGED.
 542                  */
 543                 if (scmd->device->sdev_target->expecting_lun_change &&
 544                     sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
 545                         return NEEDS_RETRY;
 546                 /*
 547                  * if the device is in the process of becoming ready, we
 548                  * should retry.
 549                  */
 550                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
 551                         return NEEDS_RETRY;
 552                 /*
 553                  * if the device is not started, we need to wake
 554                  * the error handler to start the motor
 555                  */
 556                 if (scmd->device->allow_restart &&
 557                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
 558                         return FAILED;
 559                 /*
 560                  * Pass the UA upwards for a determination in the completion
 561                  * functions.
 562                  */
 563                 return SUCCESS;
 564
 565                 /* these are not supported */
 566         case DATA_PROTECT:
 567                 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
 568                         /* Thin provisioning hard threshold reached */
 569                         set_host_byte(scmd, DID_ALLOC_FAILURE);
 570                         return SUCCESS;
 571                 }
 572         case COPY_ABORTED:
 573         case VOLUME_OVERFLOW:
 574         case MISCOMPARE:
 575         case BLANK_CHECK:
 576                 set_host_byte(scmd, DID_TARGET_FAILURE);
 577                 return SUCCESS;
 578
 579         case MEDIUM_ERROR:
 580                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
 581                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
 582                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
 583                         set_host_byte(scmd, DID_MEDIUM_ERROR);
 584                         return SUCCESS;
 585                 }
 586                 return NEEDS_RETRY;
 587
 588         case HARDWARE_ERROR:
 589                 if (scmd->device->retry_hwerror)
 590                         return ADD_TO_MLQUEUE;
 591                 else
 592                         set_host_byte(scmd, DID_TARGET_FAILURE);
 593
 594         case ILLEGAL_REQUEST:
 595                 if (sshdr.asc == 0x20 || /* Invalid command operation code */
 596                     sshdr.asc == 0x21 || /* Logical block address out of range */
 597                     sshdr.asc == 0x24 || /* Invalid field in cdb */
 598                     sshdr.asc == 0x26) { /* Parameter value invalid */
 599                         set_host_byte(scmd, DID_TARGET_FAILURE);
 600                 }
 601                 return SUCCESS;
 602
 603         default:
 604                 return SUCCESS;
 605         }
 606 }
 607
 608 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
 609 {
 610         struct scsi_host_template *sht = sdev->host->hostt;
 611         struct scsi_device *tmp_sdev;
 612
 613         if (!sht->track_queue_depth ||
 614             sdev->queue_depth >= sdev->max_queue_depth)
 615                 return;
 616
 617         if (time_before(jiffies,
 618             sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
 619                 return;
 620
 621         if (time_before(jiffies,
 622             sdev->last_queue_full_time + sdev->queue_ramp_up_period))
 623                 return;
 624
 625         /*
 626          * Walk all devices of a target and do
 627          * ramp up on them.
 628          */
 629         shost_for_each_device(tmp_sdev, sdev->host) {
 630                 if (tmp_sdev->channel != sdev->channel ||
 631                     tmp_sdev->id != sdev->id ||
 632                     tmp_sdev->queue_depth == sdev->max_queue_depth)
 633                         continue;
 634
 635                 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
 636                 sdev->last_queue_ramp_up = jiffies;
 637         }
 638 }
 639
 640 static void scsi_handle_queue_full(struct scsi_device *sdev)
 641 {
 642         struct scsi_host_template *sht = sdev->host->hostt;
 643         struct scsi_device *tmp_sdev;
 644
 645         if (!sht->track_queue_depth)
 646                 return;
 647
 648         shost_for_each_device(tmp_sdev, sdev->host) {
 649                 if (tmp_sdev->channel != sdev->channel ||
 650                     tmp_sdev->id != sdev->id)
 651                         continue;
 652                 /*
 653                  * We do not know the number of commands that were at
 654                  * the device when we got the queue full so we start
 655                  * from the highest possible value and work our way down.
 656                  */
 657                 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
 658         }
 659 }
 660
 661 /**
 662  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
 663  * @scmd:       SCSI cmd to examine.
 664  *
 665  * Notes:
 666  *    This is *only* called when we are examining the status of commands
 667  *    queued during error recovery.  the main difference here is that we
 668  *    don't allow for the possibility of retries here, and we are a lot
 669  *    more restrictive about what we consider acceptable.
 670  */
 671 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
 672 {
 673         /*
 674          * first check the host byte, to see if there is anything in there
 675          * that would indicate what we need to do.
 676          */
 677         if (host_byte(scmd->result) == DID_RESET) {
 678                 /*
 679                  * rats.  we are already in the error handler, so we now
 680                  * get to try and figure out what to do next.  if the sense
 681                  * is valid, we have a pretty good idea of what to do.
 682                  * if not, we mark it as FAILED.
 683                  */
 684                 return scsi_check_sense(scmd);
 685         }
 686         if (host_byte(scmd->result) != DID_OK)
 687                 return FAILED;
 688
 689         /*
 690          * next, check the message byte.
 691          */
 692         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
 693                 return FAILED;
 694
 695         /*
 696          * now, check the status byte to see if this indicates
 697          * anything special.
 698          */
 699         switch (status_byte(scmd->result)) {
 700         case GOOD:
 701                 scsi_handle_queue_ramp_up(scmd->device);
 702         case COMMAND_TERMINATED:
 703                 return SUCCESS;
 704         case CHECK_CONDITION:
 705                 return scsi_check_sense(scmd);
 706         case CONDITION_GOOD:
 707         case INTERMEDIATE_GOOD:
 708         case INTERMEDIATE_C_GOOD:
 709                 /*
 710                  * who knows?  FIXME(eric)
 711                  */
 712                 return SUCCESS;
 713         case RESERVATION_CONFLICT:
 714                 if (scmd->cmnd[0] == TEST_UNIT_READY)
 715                         /* it is a success, we probed the device and
 716                          * found it */
 717                         return SUCCESS;
 718                 /* otherwise, we failed to send the command */
 719                 return FAILED;
 720         case QUEUE_FULL:
 721                 scsi_handle_queue_full(scmd->device);
 722                 /* fall through */
 723         case BUSY:
 724                 return NEEDS_RETRY;
 725         default:
 726                 return FAILED;
 727         }
 728         return FAILED;
 729 }
 730
 731 /**
 732  * scsi_eh_done - Completion function for error handling.
 733  * @scmd:       Cmd that is done.
 734  */
 735 static void scsi_eh_done(struct scsi_cmnd *scmd)
 736 {
 737         struct completion *eh_action;
 738
 739         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
 740                         "%s scmd: %p result: %x\n",
 741                         __func__, scmd, scmd->result));
 742
 743         eh_action = scmd->device->host->eh_action;
 744         if (eh_action)
 745                 complete(eh_action);
 746 }
 747
 748 /**
 749  * scsi_try_host_reset - ask host adapter to reset itself
 750  * @scmd:       SCSI cmd to send host reset.
 751  */
 752 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
 753 {
 754         unsigned long flags;
 755         int rtn;
 756         struct Scsi_Host *host = scmd->device->host;
 757         struct scsi_host_template *hostt = host->hostt;
 758
 759         SCSI_LOG_ERROR_RECOVERY(3,
 760                 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
 761
 762         if (!hostt->eh_host_reset_handler)
 763                 return FAILED;
 764
 765         rtn = hostt->eh_host_reset_handler(scmd);
 766
 767         if (rtn == SUCCESS) {
 768                 if (!hostt->skip_settle_delay)
 769                         ssleep(HOST_RESET_SETTLE_TIME);
 770                 spin_lock_irqsave(host->host_lock, flags);
 771                 scsi_report_bus_reset(host, scmd_channel(scmd));
 772                 spin_unlock_irqrestore(host->host_lock, flags);
 773         }
 774
 775         return rtn;
 776 }
 777
 778 /**
 779  * scsi_try_bus_reset - ask host to perform a bus reset
 780  * @scmd:       SCSI cmd to send bus reset.
 781  */
 782 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
 783 {
 784         unsigned long flags;
 785         int rtn;
 786         struct Scsi_Host *host = scmd->device->host;
 787         struct scsi_host_template *hostt = host->hostt;
 788
 789         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
 790                 "%s: Snd Bus RST\n", __func__));
 791
 792         if (!hostt->eh_bus_reset_handler)
 793                 return FAILED;
 794
 795         rtn = hostt->eh_bus_reset_handler(scmd);
 796
 797         if (rtn == SUCCESS) {
 798                 if (!hostt->skip_settle_delay)
 799                         ssleep(BUS_RESET_SETTLE_TIME);
 800                 spin_lock_irqsave(host->host_lock, flags);
 801                 scsi_report_bus_reset(host, scmd_channel(scmd));
 802                 spin_unlock_irqrestore(host->host_lock, flags);
 803         }
 804
 805         return rtn;
 806 }
 807
 808 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
 809 {
 810         sdev->was_reset = 1;
 811         sdev->expecting_cc_ua = 1;
 812 }
 813
 814 /**
 815  * scsi_try_target_reset - Ask host to perform a target reset
 816  * @scmd:       SCSI cmd used to send a target reset
 817  *
 818  * Notes:
 819  *    There is no timeout for this operation.  if this operation is
 820  *    unreliable for a given host, then the host itself needs to put a
 821  *    timer on it, and set the host back to a consistent state prior to
 822  *    returning.
 823  */
 824 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
 825 {
 826         unsigned long flags;
 827         int rtn;
 828         struct Scsi_Host *host = scmd->device->host;
 829         struct scsi_host_template *hostt = host->hostt;
 830
 831         if (!hostt->eh_target_reset_handler)
 832                 return FAILED;
 833
 834         rtn = hostt->eh_target_reset_handler(scmd);
 835         if (rtn == SUCCESS) {
 836                 spin_lock_irqsave(host->host_lock, flags);
 837                 __starget_for_each_device(scsi_target(scmd->device), NULL,
 838                                           __scsi_report_device_reset);
 839                 spin_unlock_irqrestore(host->host_lock, flags);
 840         }
 841
 842         return rtn;
 843 }
 844
 845 /**
 846  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
 847  * @scmd:       SCSI cmd used to send BDR
 848  *
 849  * Notes:
 850  *    There is no timeout for this operation.  if this operation is
 851  *    unreliable for a given host, then the host itself needs to put a
 852  *    timer on it, and set the host back to a consistent state prior to
 853  *    returning.
 854  */
 855 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
 856 {
 857         int rtn;
 858         struct scsi_host_template *hostt = scmd->device->host->hostt;
 859
 860         if (!hostt->eh_device_reset_handler)
 861                 return FAILED;
 862
 863         rtn = hostt->eh_device_reset_handler(scmd);
 864         if (rtn == SUCCESS)
 865                 __scsi_report_device_reset(scmd->device, NULL);
 866         return rtn;
 867 }
 868
 869 /**
 870  * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
 871  * @scmd:       SCSI cmd used to send a target reset
 872  *
 873  * Return value:
 874  *      SUCCESS, FAILED, or FAST_IO_FAIL
 875  *
 876  * Notes:
 877  *    SUCCESS does not necessarily indicate that the command
 878  *    has been aborted; it only indicates that the LLDDs
 879  *    has cleared all references to that command.
 880  *    LLDDs should return FAILED only if an abort was required
 881  *    but could not be executed. LLDDs should return FAST_IO_FAIL
 882  *    if the device is temporarily unavailable (eg due to a
 883  *    link down on FibreChannel)
 884  */
 885 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
 886                                  struct scsi_cmnd *scmd)
 887 {
 888         if (!hostt->eh_abort_handler)
 889                 return FAILED;
 890
 891         return hostt->eh_abort_handler(scmd);
 892 }
 893
 894 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
 895 {
 896         if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
 897                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
 898                         if (scsi_try_target_reset(scmd) != SUCCESS)
 899                                 if (scsi_try_bus_reset(scmd) != SUCCESS)
 900                                         scsi_try_host_reset(scmd);
 901 }
 902
 903 /**
 904  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
 905  * @scmd:       SCSI command structure to hijack
 906  * @ses:        structure to save restore information
 907  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
 908  * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
 909  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
 910  *
 911  * This function is used to save a scsi command information before re-execution
 912  * as part of the error recovery process.  If @sense_bytes is 0 the command
 913  * sent must be one that does not transfer any data.  If @sense_bytes != 0
 914  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
 915  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
 916  */
 917 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
 918                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
 919 {
 920         struct scsi_device *sdev = scmd->device;
 921
 922         /*
 923          * We need saved copies of a number of fields - this is because
 924          * error handling may need to overwrite these with different values
 925          * to run different commands, and once error handling is complete,
 926          * we will need to restore these values prior to running the actual
 927          * command.
 928          */
 929         ses->cmd_len = scmd->cmd_len;
 930         ses->cmnd = scmd->cmnd;
 931         ses->data_direction = scmd->sc_data_direction;
 932         ses->sdb = scmd->sdb;
 933         ses->next_rq = scmd->request->next_rq;
 934         ses->result = scmd->result;
 935         ses->underflow = scmd->underflow;
 936         ses->prot_op = scmd->prot_op;
 937
 938         scmd->prot_op = SCSI_PROT_NORMAL;
 939         scmd->eh_eflags = 0;
 940         scmd->cmnd = ses->eh_cmnd;
 941         memset(scmd->cmnd, 0, BLK_MAX_CDB);
 942         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
 943         scmd->request->next_rq = NULL;
 944         scmd->result = 0;
 945
 946         if (sense_bytes) {
 947                 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
 948                                          sense_bytes);
 949                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
 950                             scmd->sdb.length);
 951                 scmd->sdb.table.sgl = &ses->sense_sgl;
 952                 scmd->sc_data_direction = DMA_FROM_DEVICE;
 953                 scmd->sdb.table.nents = 1;
 954                 scmd->cmnd[0] = REQUEST_SENSE;
 955                 scmd->cmnd[4] = scmd->sdb.length;
 956                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 957         } else {
 958                 scmd->sc_data_direction = DMA_NONE;
 959                 if (cmnd) {
 960                         BUG_ON(cmnd_size > BLK_MAX_CDB);
 961                         memcpy(scmd->cmnd, cmnd, cmnd_size);
 962                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 963                 }
 964         }
 965
 966         scmd->underflow = 0;
 967
 968         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
 969                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
 970                         (sdev->lun << 5 & 0xe0);
 971
 972         /*
 973          * Zero the sense buffer.  The scsi spec mandates that any
 974          * untransferred sense data should be interpreted as being zero.
 975          */
 976         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 977 }
 978 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
 979
 980 /**
 981  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
 982  * @scmd:       SCSI command structure to restore
 983  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
 984  *
 985  * Undo any damage done by above scsi_eh_prep_cmnd().
 986  */
 987 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
 988 {
 989         /*
 990          * Restore original data
 991          */
 992         scmd->cmd_len = ses->cmd_len;
 993         scmd->cmnd = ses->cmnd;
 994         scmd->sc_data_direction = ses->data_direction;
 995         scmd->sdb = ses->sdb;
 996         scmd->request->next_rq = ses->next_rq;
 997         scmd->result = ses->result;
 998         scmd->underflow = ses->underflow;
 999         scmd->prot_op = ses->prot_op;
1000 }
1001 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1002
1003 /**
1004  * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
1005  * @scmd:       SCSI command structure to hijack
1006  * @cmnd:       CDB to send
1007  * @cmnd_size:  size in bytes of @cmnd
1008  * @timeout:    timeout for this request
1009  * @sense_bytes: size of sense data to copy or 0
1010  *
1011  * This function is used to send a scsi command down to a target device
1012  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1013  *
1014  * Return value:
1015  *    SUCCESS or FAILED or NEEDS_RETRY
1016  */
1017 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1018                              int cmnd_size, int timeout, unsigned sense_bytes)
1019 {
1020         struct scsi_device *sdev = scmd->device;
1021         struct Scsi_Host *shost = sdev->host;
1022         DECLARE_COMPLETION_ONSTACK(done);
1023         unsigned long timeleft = timeout;
1024         struct scsi_eh_save ses;
1025         const unsigned long stall_for = msecs_to_jiffies(100);
1026         int rtn;
1027
1028 retry:
1029         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1030         shost->eh_action = &done;
1031
1032         scsi_log_send(scmd);
1033         scmd->scsi_done = scsi_eh_done;
1034         rtn = shost->hostt->queuecommand(shost, scmd);
1035         if (rtn) {
1036                 if (timeleft > stall_for) {
1037                         scsi_eh_restore_cmnd(scmd, &ses);
1038                         timeleft -= stall_for;
1039                         msleep(jiffies_to_msecs(stall_for));
1040                         goto retry;
1041                 }
1042                 /* signal not to enter either branch of the if () below */
1043                 timeleft = 0;
1044                 rtn = FAILED;
1045         } else {
1046                 timeleft = wait_for_completion_timeout(&done, timeout);
1047                 rtn = SUCCESS;
1048         }
1049
1050         shost->eh_action = NULL;
1051
1052         scsi_log_completion(scmd, rtn);
1053
1054         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1055                         "%s: scmd: %p, timeleft: %ld\n",
1056                         __func__, scmd, timeleft));
1057
1058         /*
1059          * If there is time left scsi_eh_done got called, and we will examine
1060          * the actual status codes to see whether the command actually did
1061          * complete normally, else if we have a zero return and no time left,
1062          * the command must still be pending, so abort it and return FAILED.
1063          * If we never actually managed to issue the command, because
1064          * ->queuecommand() kept returning non zero, use the rtn = FAILED
1065          * value above (so don't execute either branch of the if)
1066          */
1067         if (timeleft) {
1068                 rtn = scsi_eh_completed_normally(scmd);
1069                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1070                         "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1071
1072                 switch (rtn) {
1073                 case SUCCESS:
1074                 case NEEDS_RETRY:
1075                 case FAILED:
1076                         break;
1077                 case ADD_TO_MLQUEUE:
1078                         rtn = NEEDS_RETRY;
1079                         break;
1080                 default:
1081                         rtn = FAILED;
1082                         break;
1083                 }
1084         } else if (rtn != FAILED) {
1085                 scsi_abort_eh_cmnd(scmd);
1086                 rtn = FAILED;
1087         }
1088
1089         scsi_eh_restore_cmnd(scmd, &ses);
1090
1091         return rtn;
1092 }
1093
1094 /**
1095  * scsi_request_sense - Request sense data from a particular target.
1096  * @scmd:       SCSI cmd for request sense.
1097  *
1098  * Notes:
1099  *    Some hosts automatically obtain this information, others require
1100  *    that we obtain it on our own. This function will *not* return until
1101  *    the command either times out, or it completes.
1102  */
1103 static int scsi_request_sense(struct scsi_cmnd *scmd)
1104 {
1105         return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1106 }
1107
1108 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1109 {
1110         if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1111                 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1112                 if (sdrv->eh_action)
1113                         rtn = sdrv->eh_action(scmd, rtn);
1114         }
1115         return rtn;
1116 }
1117
1118 /**
1119  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1120  * @scmd:       Original SCSI cmd that eh has finished.
1121  * @done_q:     Queue for processed commands.
1122  *
1123  * Notes:
1124  *    We don't want to use the normal command completion while we are are
1125  *    still handling errors - it may cause other commands to be queued,
1126  *    and that would disturb what we are doing.  Thus we really want to
1127  *    keep a list of pending commands for final completion, and once we
1128  *    are ready to leave error handling we handle completion for real.
1129  */
1130 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1131 {
1132         scmd->device->host->host_failed--;
1133         scmd->eh_eflags = 0;
1134         list_move_tail(&scmd->eh_entry, done_q);
1135 }
1136 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1137
1138 /**
1139  * scsi_eh_get_sense - Get device sense data.
1140  * @work_q:     Queue of commands to process.
1141  * @done_q:     Queue of processed commands.
1142  *
1143  * Description:
1144  *    See if we need to request sense information.  if so, then get it
1145  *    now, so we have a better idea of what to do.
1146  *
1147  * Notes:
1148  *    This has the unfortunate side effect that if a shost adapter does
1149  *    not automatically request sense information, we end up shutting
1150  *    it down before we request it.
1151  *
1152  *    All drivers should request sense information internally these days,
1153  *    so for now all I have to say is tough noogies if you end up in here.
1154  *
1155  *    XXX: Long term this code should go away, but that needs an audit of
1156  *         all LLDDs first.
1157  */
1158 int scsi_eh_get_sense(struct list_head *work_q,
1159                       struct list_head *done_q)
1160 {
1161         struct scsi_cmnd *scmd, *next;
1162         struct Scsi_Host *shost;
1163         int rtn;
1164
1165         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1166                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1167                     SCSI_SENSE_VALID(scmd))
1168                         continue;
1169
1170                 shost = scmd->device->host;
1171                 if (scsi_host_eh_past_deadline(shost)) {
1172                         SCSI_LOG_ERROR_RECOVERY(3,
1173                                 scmd_printk(KERN_INFO, scmd,
1174                                             "%s: skip request sense, past eh deadline\n",
1175                                              current->comm));
1176                         break;
1177                 }
1178                 if (status_byte(scmd->result) != CHECK_CONDITION)
1179                         /*
1180                          * don't request sense if there's no check condition
1181                          * status because the error we're processing isn't one
1182                          * that has a sense code (and some devices get
1183                          * confused by sense requests out of the blue)
1184                          */
1185                         continue;
1186
1187                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1188                                                   "%s: requesting sense\n",
1189                                                   current->comm));
1190                 rtn = scsi_request_sense(scmd);
1191                 if (rtn != SUCCESS)
1192                         continue;
1193
1194                 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1195                         "sense requested for %p result %x\n",
1196                         scmd, scmd->result));
1197                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1198
1199                 rtn = scsi_decide_disposition(scmd);
1200
1201                 /*
1202                  * if the result was normal, then just pass it along to the
1203                  * upper level.
1204                  */
1205                 if (rtn == SUCCESS)
1206                         /* we don't want this command reissued, just
1207                          * finished with the sense data, so set
1208                          * retries to the max allowed to ensure it
1209                          * won't get reissued */
1210                         scmd->retries = scmd->allowed;
1211                 else if (rtn != NEEDS_RETRY)
1212                         continue;
1213
1214                 scsi_eh_finish_cmd(scmd, done_q);
1215         }
1216
1217         return list_empty(work_q);
1218 }
1219 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1220
1221 /**
1222  * scsi_eh_tur - Send TUR to device.
1223  * @scmd:       &scsi_cmnd to send TUR
1224  *
1225  * Return value:
1226  *    0 - Device is ready. 1 - Device NOT ready.
1227  */
1228 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1229 {
1230         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1231         int retry_cnt = 1, rtn;
1232
1233 retry_tur:
1234         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1235                                 scmd->device->eh_timeout, 0);
1236
1237         SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1238                 "%s: scmd %p rtn %x\n", __func__, scmd, rtn));
1239
1240         switch (rtn) {
1241         case NEEDS_RETRY:
1242                 if (retry_cnt--)
1243                         goto retry_tur;
1244                 /*FALLTHRU*/
1245         case SUCCESS:
1246                 return 0;
1247         default:
1248                 return 1;
1249         }
1250 }
1251
1252 /**
1253  * scsi_eh_test_devices - check if devices are responding from error recovery.
1254  * @cmd_list:   scsi commands in error recovery.
1255  * @work_q:     queue for commands which still need more error recovery
1256  * @done_q:     queue for commands which are finished
1257  * @try_stu:    boolean on if a STU command should be tried in addition to TUR.
1258  *
1259  * Decription:
1260  *    Tests if devices are in a working state.  Commands to devices now in
1261  *    a working state are sent to the done_q while commands to devices which
1262  *    are still failing to respond are returned to the work_q for more
1263  *    processing.
1264  **/
1265 static int scsi_eh_test_devices(struct list_head *cmd_list,
1266                                 struct list_head *work_q,
1267                                 struct list_head *done_q, int try_stu)
1268 {
1269         struct scsi_cmnd *scmd, *next;
1270         struct scsi_device *sdev;
1271         int finish_cmds;
1272
1273         while (!list_empty(cmd_list)) {
1274                 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1275                 sdev = scmd->device;
1276
1277                 if (!try_stu) {
1278                         if (scsi_host_eh_past_deadline(sdev->host)) {
1279                                 /* Push items back onto work_q */
1280                                 list_splice_init(cmd_list, work_q);
1281                                 SCSI_LOG_ERROR_RECOVERY(3,
1282                                         sdev_printk(KERN_INFO, sdev,
1283                                                     "%s: skip test device, past eh deadline",
1284                                                     current->comm));
1285                                 break;
1286                         }
1287                 }
1288
1289                 finish_cmds = !scsi_device_online(scmd->device) ||
1290                         (try_stu && !scsi_eh_try_stu(scmd) &&
1291                          !scsi_eh_tur(scmd)) ||
1292                         !scsi_eh_tur(scmd);
1293
1294                 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1295                         if (scmd->device == sdev) {
1296                                 if (finish_cmds &&
1297                                     (try_stu ||
1298                                      scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1299                                         scsi_eh_finish_cmd(scmd, done_q);
1300                                 else
1301                                         list_move_tail(&scmd->eh_entry, work_q);
1302                         }
1303         }
1304         return list_empty(work_q);
1305 }
1306
1307
1308 /**
1309  * scsi_eh_abort_cmds - abort pending commands.
1310  * @work_q:     &list_head for pending commands.
1311  * @done_q:     &list_head for processed commands.
1312  *
1313  * Decription:
1314  *    Try and see whether or not it makes sense to try and abort the
1315  *    running command.  This only works out to be the case if we have one
1316  *    command that has timed out.  If the command simply failed, it makes
1317  *    no sense to try and abort the command, since as far as the shost
1318  *    adapter is concerned, it isn't running.
1319  */
1320 static int scsi_eh_abort_cmds(struct list_head *work_q,
1321                               struct list_head *done_q)
1322 {
1323         struct scsi_cmnd *scmd, *next;
1324         LIST_HEAD(check_list);
1325         int rtn;
1326         struct Scsi_Host *shost;
1327
1328         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1329                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1330                         continue;
1331                 shost = scmd->device->host;
1332                 if (scsi_host_eh_past_deadline(shost)) {
1333                         list_splice_init(&check_list, work_q);
1334                         SCSI_LOG_ERROR_RECOVERY(3,
1335                                 scmd_printk(KERN_INFO, scmd,
1336                                             "%s: skip aborting cmd, past eh deadline\n",
1337                                             current->comm));
1338                         return list_empty(work_q);
1339                 }
1340                 SCSI_LOG_ERROR_RECOVERY(3,
1341                         scmd_printk(KERN_INFO, scmd,
1342                                      "%s: aborting cmd\n", current->comm));
1343                 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1344                 if (rtn == FAILED) {
1345                         SCSI_LOG_ERROR_RECOVERY(3,
1346                                 scmd_printk(KERN_INFO, scmd,
1347                                             "%s: aborting cmd failed\n",
1348                                              current->comm));
1349                         list_splice_init(&check_list, work_q);
1350                         return list_empty(work_q);
1351                 }
1352                 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1353                 if (rtn == FAST_IO_FAIL)
1354                         scsi_eh_finish_cmd(scmd, done_q);
1355                 else
1356                         list_move_tail(&scmd->eh_entry, &check_list);
1357         }
1358
1359         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1360 }
1361
1362 /**
1363  * scsi_eh_try_stu - Send START_UNIT to device.
1364  * @scmd:       &scsi_cmnd to send START_UNIT
1365  *
1366  * Return value:
1367  *    0 - Device is ready. 1 - Device NOT ready.
1368  */
1369 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1370 {
1371         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1372
1373         if (scmd->device->allow_restart) {
1374                 int i, rtn = NEEDS_RETRY;
1375
1376                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1377                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1378
1379                 if (rtn == SUCCESS)
1380                         return 0;
1381         }
1382
1383         return 1;
1384 }
1385
1386  /**
1387  * scsi_eh_stu - send START_UNIT if needed
1388  * @shost:      &scsi host being recovered.
1389  * @work_q:     &list_head for pending commands.
1390  * @done_q:     &list_head for processed commands.
1391  *
1392  * Notes:
1393  *    If commands are failing due to not ready, initializing command required,
1394  *      try revalidating the device, which will end up sending a start unit.
1395  */
1396 static int scsi_eh_stu(struct Scsi_Host *shost,
1397                               struct list_head *work_q,
1398                               struct list_head *done_q)
1399 {
1400         struct scsi_cmnd *scmd, *stu_scmd, *next;
1401         struct scsi_device *sdev;
1402
1403         shost_for_each_device(sdev, shost) {
1404                 if (scsi_host_eh_past_deadline(shost)) {
1405                         SCSI_LOG_ERROR_RECOVERY(3,
1406                                 sdev_printk(KERN_INFO, sdev,
1407                                             "%s: skip START_UNIT, past eh deadline\n",
1408                                             current->comm));
1409                         break;
1410                 }
1411                 stu_scmd = NULL;
1412                 list_for_each_entry(scmd, work_q, eh_entry)
1413                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1414                             scsi_check_sense(scmd) == FAILED ) {
1415                                 stu_scmd = scmd;
1416                                 break;
1417                         }
1418
1419                 if (!stu_scmd)
1420                         continue;
1421
1422                 SCSI_LOG_ERROR_RECOVERY(3,
1423                         sdev_printk(KERN_INFO, sdev,
1424                                      "%s: Sending START_UNIT\n",
1425                                     current->comm));
1426
1427                 if (!scsi_eh_try_stu(stu_scmd)) {
1428                         if (!scsi_device_online(sdev) ||
1429                             !scsi_eh_tur(stu_scmd)) {
1430                                 list_for_each_entry_safe(scmd, next,
1431                                                           work_q, eh_entry) {
1432                                         if (scmd->device == sdev &&
1433                                             scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1434                                                 scsi_eh_finish_cmd(scmd, done_q);
1435                                 }
1436                         }
1437                 } else {
1438                         SCSI_LOG_ERROR_RECOVERY(3,
1439                                 sdev_printk(KERN_INFO, sdev,
1440                                             "%s: START_UNIT failed\n",
1441                                             current->comm));
1442                 }
1443         }
1444
1445         return list_empty(work_q);
1446 }
1447
1448
1449 /**
1450  * scsi_eh_bus_device_reset - send bdr if needed
1451  * @shost:      scsi host being recovered.
1452  * @work_q:     &list_head for pending commands.
1453  * @done_q:     &list_head for processed commands.
1454  *
1455  * Notes:
1456  *    Try a bus device reset.  Still, look to see whether we have multiple
1457  *    devices that are jammed or not - if we have multiple devices, it
1458  *    makes no sense to try bus_device_reset - we really would need to try
1459  *    a bus_reset instead.
1460  */
1461 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1462                                     struct list_head *work_q,
1463                                     struct list_head *done_q)
1464 {
1465         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1466         struct scsi_device *sdev;
1467         int rtn;
1468
1469         shost_for_each_device(sdev, shost) {
1470                 if (scsi_host_eh_past_deadline(shost)) {
1471                         SCSI_LOG_ERROR_RECOVERY(3,
1472                                 sdev_printk(KERN_INFO, sdev,
1473                                             "%s: skip BDR, past eh deadline\n",
1474                                              current->comm));
1475                         break;
1476                 }
1477                 bdr_scmd = NULL;
1478                 list_for_each_entry(scmd, work_q, eh_entry)
1479                         if (scmd->device == sdev) {
1480                                 bdr_scmd = scmd;
1481                                 break;
1482                         }
1483
1484                 if (!bdr_scmd)
1485                         continue;
1486
1487                 SCSI_LOG_ERROR_RECOVERY(3,
1488                         sdev_printk(KERN_INFO, sdev,
1489                                      "%s: Sending BDR\n", current->comm));
1490                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1491                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1492                         if (!scsi_device_online(sdev) ||
1493                             rtn == FAST_IO_FAIL ||
1494                             !scsi_eh_tur(bdr_scmd)) {
1495                                 list_for_each_entry_safe(scmd, next,
1496                                                          work_q, eh_entry) {
1497                                         if (scmd->device == sdev &&
1498                                             scsi_eh_action(scmd, rtn) != FAILED)
1499                                                 scsi_eh_finish_cmd(scmd,
1500                                                                    done_q);
1501                                 }
1502                         }
1503                 } else {
1504                         SCSI_LOG_ERROR_RECOVERY(3,
1505                                 sdev_printk(KERN_INFO, sdev,
1506                                             "%s: BDR failed\n", current->comm));
1507                 }
1508         }
1509
1510         return list_empty(work_q);
1511 }
1512
1513 /**
1514  * scsi_eh_target_reset - send target reset if needed
1515  * @shost:      scsi host being recovered.
1516  * @work_q:     &list_head for pending commands.
1517  * @done_q:     &list_head for processed commands.
1518  *
1519  * Notes:
1520  *    Try a target reset.
1521  */
1522 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1523                                 struct list_head *work_q,
1524                                 struct list_head *done_q)
1525 {
1526         LIST_HEAD(tmp_list);
1527         LIST_HEAD(check_list);
1528
1529         list_splice_init(work_q, &tmp_list);
1530
1531         while (!list_empty(&tmp_list)) {
1532                 struct scsi_cmnd *next, *scmd;
1533                 int rtn;
1534                 unsigned int id;
1535
1536                 if (scsi_host_eh_past_deadline(shost)) {
1537                         /* push back on work queue for further processing */
1538                         list_splice_init(&check_list, work_q);
1539                         list_splice_init(&tmp_list, work_q);
1540                         SCSI_LOG_ERROR_RECOVERY(3,
1541                                 shost_printk(KERN_INFO, shost,
1542                                             "%s: Skip target reset, past eh deadline\n",
1543                                              current->comm));
1544                         return list_empty(work_q);
1545                 }
1546
1547                 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1548                 id = scmd_id(scmd);
1549
1550                 SCSI_LOG_ERROR_RECOVERY(3,
1551                         shost_printk(KERN_INFO, shost,
1552                                      "%s: Sending target reset to target %d\n",
1553                                      current->comm, id));
1554                 rtn = scsi_try_target_reset(scmd);
1555                 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1556                         SCSI_LOG_ERROR_RECOVERY(3,
1557                                 shost_printk(KERN_INFO, shost,
1558                                              "%s: Target reset failed"
1559                                              " target: %d\n",
1560                                              current->comm, id));
1561                 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1562                         if (scmd_id(scmd) != id)
1563                                 continue;
1564
1565                         if (rtn == SUCCESS)
1566                                 list_move_tail(&scmd->eh_entry, &check_list);
1567                         else if (rtn == FAST_IO_FAIL)
1568                                 scsi_eh_finish_cmd(scmd, done_q);
1569                         else
1570                                 /* push back on work queue for further processing */
1571                                 list_move(&scmd->eh_entry, work_q);
1572                 }
1573         }
1574
1575         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1576 }
1577
1578 /**
1579  * scsi_eh_bus_reset - send a bus reset
1580  * @shost:      &scsi host being recovered.
1581  * @work_q:     &list_head for pending commands.
1582  * @done_q:     &list_head for processed commands.
1583  */
1584 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1585                              struct list_head *work_q,
1586                              struct list_head *done_q)
1587 {
1588         struct scsi_cmnd *scmd, *chan_scmd, *next;
1589         LIST_HEAD(check_list);
1590         unsigned int channel;
1591         int rtn;
1592
1593         /*
1594          * we really want to loop over the various channels, and do this on
1595          * a channel by channel basis.  we should also check to see if any
1596          * of the failed commands are on soft_reset devices, and if so, skip
1597          * the reset.
1598          */
1599
1600         for (channel = 0; channel <= shost->max_channel; channel++) {
1601                 if (scsi_host_eh_past_deadline(shost)) {
1602                         list_splice_init(&check_list, work_q);
1603                         SCSI_LOG_ERROR_RECOVERY(3,
1604                                 shost_printk(KERN_INFO, shost,
1605                                             "%s: skip BRST, past eh deadline\n",
1606                                              current->comm));
1607                         return list_empty(work_q);
1608                 }
1609
1610                 chan_scmd = NULL;
1611                 list_for_each_entry(scmd, work_q, eh_entry) {
1612                         if (channel == scmd_channel(scmd)) {
1613                                 chan_scmd = scmd;
1614                                 break;
1615                                 /*
1616                                  * FIXME add back in some support for
1617                                  * soft_reset devices.
1618                                  */
1619                         }
1620                 }
1621
1622                 if (!chan_scmd)
1623                         continue;
1624                 SCSI_LOG_ERROR_RECOVERY(3,
1625                         shost_printk(KERN_INFO, shost,
1626                                      "%s: Sending BRST chan: %d\n",
1627                                      current->comm, channel));
1628                 rtn = scsi_try_bus_reset(chan_scmd);
1629                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1630                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1631                                 if (channel == scmd_channel(scmd)) {
1632                                         if (rtn == FAST_IO_FAIL)
1633                                                 scsi_eh_finish_cmd(scmd,
1634                                                                    done_q);
1635                                         else
1636                                                 list_move_tail(&scmd->eh_entry,
1637                                                                &check_list);
1638                                 }
1639                         }
1640                 } else {
1641                         SCSI_LOG_ERROR_RECOVERY(3,
1642                                 shost_printk(KERN_INFO, shost,
1643                                              "%s: BRST failed chan: %d\n",
1644                                              current->comm, channel));
1645                 }
1646         }
1647         return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1648 }
1649
1650 /**
1651  * scsi_eh_host_reset - send a host reset
1652  * @shost:      host to be reset.
1653  * @work_q:     &list_head for pending commands.
1654  * @done_q:     &list_head for processed commands.
1655  */
1656 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1657                               struct list_head *work_q,
1658                               struct list_head *done_q)
1659 {
1660         struct scsi_cmnd *scmd, *next;
1661         LIST_HEAD(check_list);
1662         int rtn;
1663
1664         if (!list_empty(work_q)) {
1665                 scmd = list_entry(work_q->next,
1666                                   struct scsi_cmnd, eh_entry);
1667
1668                 SCSI_LOG_ERROR_RECOVERY(3,
1669                         shost_printk(KERN_INFO, shost,
1670                                      "%s: Sending HRST\n",
1671                                      current->comm));
1672
1673                 rtn = scsi_try_host_reset(scmd);
1674                 if (rtn == SUCCESS) {
1675                         list_splice_init(work_q, &check_list);
1676                 } else if (rtn == FAST_IO_FAIL) {
1677                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1678                                         scsi_eh_finish_cmd(scmd, done_q);
1679                         }
1680                 } else {
1681                         SCSI_LOG_ERROR_RECOVERY(3,
1682                                 shost_printk(KERN_INFO, shost,
1683                                              "%s: HRST failed\n",
1684                                              current->comm));
1685                 }
1686         }
1687         return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1688 }
1689
1690 /**
1691  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1692  * @work_q:     &list_head for pending commands.
1693  * @done_q:     &list_head for processed commands.
1694  */
1695 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1696                                   struct list_head *done_q)
1697 {
1698         struct scsi_cmnd *scmd, *next;
1699
1700         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1701                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1702                             "not ready after error recovery\n");
1703                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1704                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1705                         /*
1706                          * FIXME: Handle lost cmds.
1707                          */
1708                 }
1709                 scsi_eh_finish_cmd(scmd, done_q);
1710         }
1711         return;
1712 }
1713
1714 /**
1715  * scsi_noretry_cmd - determine if command should be failed fast
1716  * @scmd:       SCSI cmd to examine.
1717  */
1718 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1719 {
1720         switch (host_byte(scmd->result)) {
1721         case DID_OK:
1722                 break;
1723         case DID_TIME_OUT:
1724                 goto check_type;
1725         case DID_BUS_BUSY:
1726                 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1727         case DID_PARITY:
1728                 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1729         case DID_ERROR:
1730                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1731                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1732                         return 0;
1733                 /* fall through */
1734         case DID_SOFT_ERROR:
1735                 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1736         }
1737
1738         if (status_byte(scmd->result) != CHECK_CONDITION)
1739                 return 0;
1740
1741 check_type:
1742         /*
1743          * assume caller has checked sense and determined
1744          * the check condition was retryable.
1745          */
1746         if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1747             scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1748                 return 1;
1749         else
1750                 return 0;
1751 }
1752
1753 /**
1754  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1755  * @scmd:       SCSI cmd to examine.
1756  *
1757  * Notes:
1758  *    This is *only* called when we are examining the status after sending
1759  *    out the actual data command.  any commands that are queued for error
1760  *    recovery (e.g. test_unit_ready) do *not* come through here.
1761  *
1762  *    When this routine returns failed, it means the error handler thread
1763  *    is woken.  In cases where the error code indicates an error that
1764  *    doesn't require the error handler read (i.e. we don't need to
1765  *    abort/reset), this function should return SUCCESS.
1766  */
1767 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1768 {
1769         int rtn;
1770
1771         /*
1772          * if the device is offline, then we clearly just pass the result back
1773          * up to the top level.
1774          */
1775         if (!scsi_device_online(scmd->device)) {
1776                 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1777                         "%s: device offline - report as SUCCESS\n", __func__));
1778                 return SUCCESS;
1779         }
1780
1781         /*
1782          * first check the host byte, to see if there is anything in there
1783          * that would indicate what we need to do.
1784          */
1785         switch (host_byte(scmd->result)) {
1786         case DID_PASSTHROUGH:
1787                 /*
1788                  * no matter what, pass this through to the upper layer.
1789                  * nuke this special code so that it looks like we are saying
1790                  * did_ok.
1791                  */
1792                 scmd->result &= 0xff00ffff;
1793                 return SUCCESS;
1794         case DID_OK:
1795                 /*
1796                  * looks good.  drop through, and check the next byte.
1797                  */
1798                 break;
1799         case DID_ABORT:
1800                 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1801                         set_host_byte(scmd, DID_TIME_OUT);
1802                         return SUCCESS;
1803                 }
1804         case DID_NO_CONNECT:
1805         case DID_BAD_TARGET:
1806                 /*
1807                  * note - this means that we just report the status back
1808                  * to the top level driver, not that we actually think
1809                  * that it indicates SUCCESS.
1810                  */
1811                 return SUCCESS;
1812                 /*
1813                  * when the low level driver returns did_soft_error,
1814                  * it is responsible for keeping an internal retry counter
1815                  * in order to avoid endless loops (db)
1816                  *
1817                  * actually this is a bug in this function here.  we should
1818                  * be mindful of the maximum number of retries specified
1819                  * and not get stuck in a loop.
1820                  */
1821         case DID_SOFT_ERROR:
1822                 goto maybe_retry;
1823         case DID_IMM_RETRY:
1824                 return NEEDS_RETRY;
1825
1826         case DID_REQUEUE:
1827                 return ADD_TO_MLQUEUE;
1828         case DID_TRANSPORT_DISRUPTED:
1829                 /*
1830                  * LLD/transport was disrupted during processing of the IO.
1831                  * The transport class is now blocked/blocking,
1832                  * and the transport will decide what to do with the IO
1833                  * based on its timers and recovery capablilities if
1834                  * there are enough retries.
1835                  */
1836                 goto maybe_retry;
1837         case DID_TRANSPORT_FAILFAST:
1838                 /*
1839                  * The transport decided to failfast the IO (most likely
1840                  * the fast io fail tmo fired), so send IO directly upwards.
1841                  */
1842                 return SUCCESS;
1843         case DID_ERROR:
1844                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1845                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1846                         /*
1847                          * execute reservation conflict processing code
1848                          * lower down
1849                          */
1850                         break;
1851                 /* fallthrough */
1852         case DID_BUS_BUSY:
1853         case DID_PARITY:
1854                 goto maybe_retry;
1855         case DID_TIME_OUT:
1856                 /*
1857                  * when we scan the bus, we get timeout messages for
1858                  * these commands if there is no device available.
1859                  * other hosts report did_no_connect for the same thing.
1860                  */
1861                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1862                      scmd->cmnd[0] == INQUIRY)) {
1863                         return SUCCESS;
1864                 } else {
1865                         return FAILED;
1866                 }
1867         case DID_RESET:
1868                 return SUCCESS;
1869         default:
1870                 return FAILED;
1871         }
1872
1873         /*
1874          * next, check the message byte.
1875          */
1876         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1877                 return FAILED;
1878
1879         /*
1880          * check the status byte to see if this indicates anything special.
1881          */
1882         switch (status_byte(scmd->result)) {
1883         case QUEUE_FULL:
1884                 scsi_handle_queue_full(scmd->device);
1885                 /*
1886                  * the case of trying to send too many commands to a
1887                  * tagged queueing device.
1888                  */
1889         case BUSY:
1890                 /*
1891                  * device can't talk to us at the moment.  Should only
1892                  * occur (SAM-3) when the task queue is empty, so will cause
1893                  * the empty queue handling to trigger a stall in the
1894                  * device.
1895                  */
1896                 return ADD_TO_MLQUEUE;
1897         case GOOD:
1898                 if (scmd->cmnd[0] == REPORT_LUNS)
1899                         scmd->device->sdev_target->expecting_lun_change = 0;
1900                 scsi_handle_queue_ramp_up(scmd->device);
1901         case COMMAND_TERMINATED:
1902                 return SUCCESS;
1903         case TASK_ABORTED:
1904                 goto maybe_retry;
1905         case CHECK_CONDITION:
1906                 rtn = scsi_check_sense(scmd);
1907                 if (rtn == NEEDS_RETRY)
1908                         goto maybe_retry;
1909                 /* if rtn == FAILED, we have no sense information;
1910                  * returning FAILED will wake the error handler thread
1911                  * to collect the sense and redo the decide
1912                  * disposition */
1913                 return rtn;
1914         case CONDITION_GOOD:
1915         case INTERMEDIATE_GOOD:
1916         case INTERMEDIATE_C_GOOD:
1917         case ACA_ACTIVE:
1918                 /*
1919                  * who knows?  FIXME(eric)
1920                  */
1921                 return SUCCESS;
1922
1923         case RESERVATION_CONFLICT:
1924                 sdev_printk(KERN_INFO, scmd->device,
1925                             "reservation conflict\n");
1926                 set_host_byte(scmd, DID_NEXUS_FAILURE);
1927                 return SUCCESS; /* causes immediate i/o error */
1928         default:
1929                 return FAILED;
1930         }
1931         return FAILED;
1932
1933       maybe_retry:
1934
1935         /* we requeue for retry because the error was retryable, and
1936          * the request was not marked fast fail.  Note that above,
1937          * even if the request is marked fast fail, we still requeue
1938          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1939         if ((++scmd->retries) <= scmd->allowed
1940             && !scsi_noretry_cmd(scmd)) {
1941                 return NEEDS_RETRY;
1942         } else {
1943                 /*
1944                  * no more retries - report this one back to upper level.
1945                  */
1946                 return SUCCESS;
1947         }
1948 }
1949
1950 static void eh_lock_door_done(struct request *req, int uptodate)
1951 {
1952         __blk_put_request(req->q, req);
1953 }
1954
1955 /**
1956  * scsi_eh_lock_door - Prevent medium removal for the specified device
1957  * @sdev:       SCSI device to prevent medium removal
1958  *
1959  * Locking:
1960  *      We must be called from process context.
1961  *
1962  * Notes:
1963  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1964  *      head of the devices request queue, and continue.
1965  */
1966 static void scsi_eh_lock_door(struct scsi_device *sdev)
1967 {
1968         struct request *req;
1969
1970         /*
1971          * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1972          * request becomes available
1973          */
1974         req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1975         if (IS_ERR(req))
1976                 return;
1977
1978         blk_rq_set_block_pc(req);
1979
1980         req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1981         req->cmd[1] = 0;
1982         req->cmd[2] = 0;
1983         req->cmd[3] = 0;
1984         req->cmd[4] = SCSI_REMOVAL_PREVENT;
1985         req->cmd[5] = 0;
1986
1987         req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1988
1989         req->cmd_flags |= REQ_QUIET;
1990         req->timeout = 10 * HZ;
1991         req->retries = 5;
1992
1993         blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1994 }
1995
1996 /**
1997  * scsi_restart_operations - restart io operations to the specified host.
1998  * @shost:      Host we are restarting.
1999  *
2000  * Notes:
2001  *    When we entered the error handler, we blocked all further i/o to
2002  *    this device.  we need to 'reverse' this process.
2003  */
2004 static void scsi_restart_operations(struct Scsi_Host *shost)
2005 {
2006         struct scsi_device *sdev;
2007         unsigned long flags;
2008
2009         /*
2010          * If the door was locked, we need to insert a door lock request
2011          * onto the head of the SCSI request queue for the device.  There
2012          * is no point trying to lock the door of an off-line device.
2013          */
2014         shost_for_each_device(sdev, shost) {
2015                 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2016                         scsi_eh_lock_door(sdev);
2017                         sdev->was_reset = 0;
2018                 }
2019         }
2020
2021         /*
2022          * next free up anything directly waiting upon the host.  this
2023          * will be requests for character device operations, and also for
2024          * ioctls to queued block devices.
2025          */
2026         SCSI_LOG_ERROR_RECOVERY(3,
2027                 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2028
2029         spin_lock_irqsave(shost->host_lock, flags);
2030         if (scsi_host_set_state(shost, SHOST_RUNNING))
2031                 if (scsi_host_set_state(shost, SHOST_CANCEL))
2032                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2033         spin_unlock_irqrestore(shost->host_lock, flags);
2034
2035         wake_up(&shost->host_wait);
2036
2037         /*
2038          * finally we need to re-initiate requests that may be pending.  we will
2039          * have had everything blocked while error handling is taking place, and
2040          * now that error recovery is done, we will need to ensure that these
2041          * requests are started.
2042          */
2043         scsi_run_host_queues(shost);
2044
2045         /*
2046          * if eh is active and host_eh_scheduled is pending we need to re-run
2047          * recovery.  we do this check after scsi_run_host_queues() to allow
2048          * everything pent up since the last eh run a chance to make forward
2049          * progress before we sync again.  Either we'll immediately re-run
2050          * recovery or scsi_device_unbusy() will wake us again when these
2051          * pending commands complete.
2052          */
2053         spin_lock_irqsave(shost->host_lock, flags);
2054         if (shost->host_eh_scheduled)
2055                 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2056                         WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2057         spin_unlock_irqrestore(shost->host_lock, flags);
2058 }
2059
2060 /**
2061  * scsi_eh_ready_devs - check device ready state and recover if not.
2062  * @shost:      host to be recovered.
2063  * @work_q:     &list_head for pending commands.
2064  * @done_q:     &list_head for processed commands.
2065  */
2066 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2067                         struct list_head *work_q,
2068                         struct list_head *done_q)
2069 {
2070         if (!scsi_eh_stu(shost, work_q, done_q))
2071                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2072                         if (!scsi_eh_target_reset(shost, work_q, done_q))
2073                                 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2074                                         if (!scsi_eh_host_reset(shost, work_q, done_q))
2075                                                 scsi_eh_offline_sdevs(work_q,
2076                                                                       done_q);
2077 }
2078 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2079
2080 /**
2081  * scsi_eh_flush_done_q - finish processed commands or retry them.
2082  * @done_q:     list_head of processed commands.
2083  */
2084 void scsi_eh_flush_done_q(struct list_head *done_q)
2085 {
2086         struct scsi_cmnd *scmd, *next;
2087
2088         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2089                 list_del_init(&scmd->eh_entry);
2090                 if (scsi_device_online(scmd->device) &&
2091                     !scsi_noretry_cmd(scmd) &&
2092                     (++scmd->retries <= scmd->allowed)) {
2093                         SCSI_LOG_ERROR_RECOVERY(3,
2094                                 scmd_printk(KERN_INFO, scmd,
2095                                              "%s: flush retry cmd: %p\n",
2096                                              current->comm, scmd));
2097                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2098                 } else {
2099                         /*
2100                          * If just we got sense for the device (called
2101                          * scsi_eh_get_sense), scmd->result is already
2102                          * set, do not set DRIVER_TIMEOUT.
2103                          */
2104                         if (!scmd->result)
2105                                 scmd->result |= (DRIVER_TIMEOUT << 24);
2106                         SCSI_LOG_ERROR_RECOVERY(3,
2107                                 scmd_printk(KERN_INFO, scmd,
2108                                              "%s: flush finish cmd: %p\n",
2109                                              current->comm, scmd));
2110                         scsi_finish_command(scmd);
2111                 }
2112         }
2113 }
2114 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2115
2116 /**
2117  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2118  * @shost:      Host to unjam.
2119  *
2120  * Notes:
2121  *    When we come in here, we *know* that all commands on the bus have
2122  *    either completed, failed or timed out.  we also know that no further
2123  *    commands are being sent to the host, so things are relatively quiet
2124  *    and we have freedom to fiddle with things as we wish.
2125  *
2126  *    This is only the *default* implementation.  it is possible for
2127  *    individual drivers to supply their own version of this function, and
2128  *    if the maintainer wishes to do this, it is strongly suggested that
2129  *    this function be taken as a template and modified.  this function
2130  *    was designed to correctly handle problems for about 95% of the
2131  *    different cases out there, and it should always provide at least a
2132  *    reasonable amount of error recovery.
2133  *
2134  *    Any command marked 'failed' or 'timeout' must eventually have
2135  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2136  *    here, so when we restart the host after we return it should have an
2137  *    empty queue.
2138  */
2139 static void scsi_unjam_host(struct Scsi_Host *shost)
2140 {
2141         unsigned long flags;
2142         LIST_HEAD(eh_work_q);
2143         LIST_HEAD(eh_done_q);
2144
2145         spin_lock_irqsave(shost->host_lock, flags);
2146         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2147         spin_unlock_irqrestore(shost->host_lock, flags);
2148
2149         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2150
2151         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2152                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2153                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2154
2155         spin_lock_irqsave(shost->host_lock, flags);
2156         if (shost->eh_deadline != -1)
2157                 shost->last_reset = 0;
2158         spin_unlock_irqrestore(shost->host_lock, flags);
2159         scsi_eh_flush_done_q(&eh_done_q);
2160 }
2161
2162 /**
2163  * scsi_error_handler - SCSI error handler thread
2164  * @data:       Host for which we are running.
2165  *
2166  * Notes:
2167  *    This is the main error handling loop.  This is run as a kernel thread
2168  *    for every SCSI host and handles all error handling activity.
2169  */
2170 int scsi_error_handler(void *data)
2171 {
2172         struct Scsi_Host *shost = data;
2173
2174         /*
2175          * We use TASK_INTERRUPTIBLE so that the thread is not
2176          * counted against the load average as a running process.
2177          * We never actually get interrupted because kthread_run
2178          * disables signal delivery for the created thread.
2179          */
2180         while (!kthread_should_stop()) {
2181                 set_current_state(TASK_INTERRUPTIBLE);
2182                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2183                     shost->host_failed != atomic_read(&shost->host_busy)) {
2184                         SCSI_LOG_ERROR_RECOVERY(1,
2185                                 shost_printk(KERN_INFO, shost,
2186                                              "scsi_eh_%d: sleeping\n",
2187                                              shost->host_no));
2188                         schedule();
2189                         continue;
2190                 }
2191
2192                 __set_current_state(TASK_RUNNING);
2193                 SCSI_LOG_ERROR_RECOVERY(1,
2194                         shost_printk(KERN_INFO, shost,
2195                                      "scsi_eh_%d: waking up %d/%d/%d\n",
2196                                      shost->host_no, shost->host_eh_scheduled,
2197                                      shost->host_failed,
2198                                      atomic_read(&shost->host_busy)));
2199
2200                 /*
2201                  * We have a host that is failing for some reason.  Figure out
2202                  * what we need to do to get it up and online again (if we can).
2203                  * If we fail, we end up taking the thing offline.
2204                  */
2205                 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2206                         SCSI_LOG_ERROR_RECOVERY(1,
2207                                 shost_printk(KERN_ERR, shost,
2208                                              "scsi_eh_%d: unable to autoresume\n",
2209                                              shost->host_no));
2210                         continue;
2211                 }
2212
2213                 if (shost->transportt->eh_strategy_handler)
2214                         shost->transportt->eh_strategy_handler(shost);
2215                 else
2216                         scsi_unjam_host(shost);
2217
2218                 /*
2219                  * Note - if the above fails completely, the action is to take
2220                  * individual devices offline and flush the queue of any
2221                  * outstanding requests that may have been pending.  When we
2222                  * restart, we restart any I/O to any other devices on the bus
2223                  * which are still online.
2224                  */
2225                 scsi_restart_operations(shost);
2226                 if (!shost->eh_noresume)
2227                         scsi_autopm_put_host(shost);
2228         }
2229         __set_current_state(TASK_RUNNING);
2230
2231         SCSI_LOG_ERROR_RECOVERY(1,
2232                 shost_printk(KERN_INFO, shost,
2233                              "Error handler scsi_eh_%d exiting\n",
2234                              shost->host_no));
2235         shost->ehandler = NULL;
2236         return 0;
2237 }
2238
2239 /*
2240  * Function:    scsi_report_bus_reset()
2241  *
2242  * Purpose:     Utility function used by low-level drivers to report that
2243  *              they have observed a bus reset on the bus being handled.
2244  *
2245  * Arguments:   shost       - Host in question
2246  *              channel     - channel on which reset was observed.
2247  *
2248  * Returns:     Nothing
2249  *
2250  * Lock status: Host lock must be held.
2251  *
2252  * Notes:       This only needs to be called if the reset is one which
2253  *              originates from an unknown location.  Resets originated
2254  *              by the mid-level itself don't need to call this, but there
2255  *              should be no harm.
2256  *
2257  *              The main purpose of this is to make sure that a CHECK_CONDITION
2258  *              is properly treated.
2259  */
2260 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2261 {
2262         struct scsi_device *sdev;
2263
2264         __shost_for_each_device(sdev, shost) {
2265                 if (channel == sdev_channel(sdev))
2266                         __scsi_report_device_reset(sdev, NULL);
2267         }
2268 }
2269 EXPORT_SYMBOL(scsi_report_bus_reset);
2270
2271 /*
2272  * Function:    scsi_report_device_reset()
2273  *
2274  * Purpose:     Utility function used by low-level drivers to report that
2275  *              they have observed a device reset on the device being handled.
2276  *
2277  * Arguments:   shost       - Host in question
2278  *              channel     - channel on which reset was observed
2279  *              target      - target on which reset was observed
2280  *
2281  * Returns:     Nothing
2282  *
2283  * Lock status: Host lock must be held
2284  *
2285  * Notes:       This only needs to be called if the reset is one which
2286  *              originates from an unknown location.  Resets originated
2287  *              by the mid-level itself don't need to call this, but there
2288  *              should be no harm.
2289  *
2290  *              The main purpose of this is to make sure that a CHECK_CONDITION
2291  *              is properly treated.
2292  */
2293 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2294 {
2295         struct scsi_device *sdev;
2296
2297         __shost_for_each_device(sdev, shost) {
2298                 if (channel == sdev_channel(sdev) &&
2299                     target == sdev_id(sdev))
2300                         __scsi_report_device_reset(sdev, NULL);
2301         }
2302 }
2303 EXPORT_SYMBOL(scsi_report_device_reset);
2304
2305 static void
2306 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2307 {
2308 }
2309
2310 /**
2311  * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2312  * @dev:        scsi_device to operate on
2313  * @arg:        reset type (see sg.h)
2314  */
2315 int
2316 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2317 {
2318         struct scsi_cmnd *scmd;
2319         struct Scsi_Host *shost = dev->host;
2320         struct request req;
2321         unsigned long flags;
2322         int error = 0, rtn, val;
2323
2324         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2325                 return -EACCES;
2326
2327         error = get_user(val, arg);
2328         if (error)
2329                 return error;
2330
2331         if (scsi_autopm_get_host(shost) < 0)
2332                 return -EIO;
2333
2334         error = -EIO;
2335         scmd = scsi_get_command(dev, GFP_KERNEL);
2336         if (!scmd)
2337                 goto out_put_autopm_host;
2338
2339         blk_rq_init(NULL, &req);
2340         scmd->request = &req;
2341
2342         scmd->cmnd = req.cmd;
2343
2344         scmd->scsi_done         = scsi_reset_provider_done_command;
2345         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2346
2347         scmd->cmd_len                   = 0;
2348
2349         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
2350
2351         spin_lock_irqsave(shost->host_lock, flags);
2352         shost->tmf_in_progress = 1;
2353         spin_unlock_irqrestore(shost->host_lock, flags);
2354
2355         switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2356         case SG_SCSI_RESET_NOTHING:
2357                 rtn = SUCCESS;
2358                 break;
2359         case SG_SCSI_RESET_DEVICE:
2360                 rtn = scsi_try_bus_device_reset(scmd);
2361                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2362                         break;
2363                 /* FALLTHROUGH */
2364         case SG_SCSI_RESET_TARGET:
2365                 rtn = scsi_try_target_reset(scmd);
2366                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2367                         break;
2368                 /* FALLTHROUGH */
2369         case SG_SCSI_RESET_BUS:
2370                 rtn = scsi_try_bus_reset(scmd);
2371                 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2372                         break;
2373                 /* FALLTHROUGH */
2374         case SG_SCSI_RESET_HOST:
2375                 rtn = scsi_try_host_reset(scmd);
2376                 if (rtn == SUCCESS)
2377                         break;
2378         default:
2379                 /* FALLTHROUGH */
2380                 rtn = FAILED;
2381                 break;
2382         }
2383
2384         error = (rtn == SUCCESS) ? 0 : -EIO;
2385
2386         spin_lock_irqsave(shost->host_lock, flags);
2387         shost->tmf_in_progress = 0;
2388         spin_unlock_irqrestore(shost->host_lock, flags);
2389
2390         /*
2391          * be sure to wake up anyone who was sleeping or had their queue
2392          * suspended while we performed the TMF.
2393          */
2394         SCSI_LOG_ERROR_RECOVERY(3,
2395                 shost_printk(KERN_INFO, shost,
2396                              "waking up host to restart after TMF\n"));
2397
2398         wake_up(&shost->host_wait);
2399         scsi_run_host_queues(shost);
2400
2401         scsi_put_command(scmd);
2402
2403 out_put_autopm_host:
2404         scsi_autopm_put_host(shost);
2405         return error;
2406 }
2407 EXPORT_SYMBOL(scsi_ioctl_reset);
2408
2409 /**
2410  * scsi_normalize_sense - normalize main elements from either fixed or
2411  *                      descriptor sense data format into a common format.
2412  *
2413  * @sense_buffer:       byte array containing sense data returned by device
2414  * @sb_len:             number of valid bytes in sense_buffer
2415  * @sshdr:              pointer to instance of structure that common
2416  *                      elements are written to.
2417  *
2418  * Notes:
2419  *      The "main elements" from sense data are: response_code, sense_key,
2420  *      asc, ascq and additional_length (only for descriptor format).
2421  *
2422  *      Typically this function can be called after a device has
2423  *      responded to a SCSI command with the CHECK_CONDITION status.
2424  *
2425  * Return value:
2426  *      true if valid sense data information found, else false;
2427  */
2428 bool scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2429                           struct scsi_sense_hdr *sshdr)
2430 {
2431         if (!sense_buffer || !sb_len)
2432                 return false;
2433
2434         memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2435
2436         sshdr->response_code = (sense_buffer[0] & 0x7f);
2437
2438         if (!scsi_sense_valid(sshdr))
2439                 return false;
2440
2441         if (sshdr->response_code >= 0x72) {
2442                 /*
2443                  * descriptor format
2444                  */
2445                 if (sb_len > 1)
2446                         sshdr->sense_key = (sense_buffer[1] & 0xf);
2447                 if (sb_len > 2)
2448                         sshdr->asc = sense_buffer[2];
2449                 if (sb_len > 3)
2450                         sshdr->ascq = sense_buffer[3];
2451                 if (sb_len > 7)
2452                         sshdr->additional_length = sense_buffer[7];
2453         } else {
2454                 /*
2455                  * fixed format
2456                  */
2457                 if (sb_len > 2)
2458                         sshdr->sense_key = (sense_buffer[2] & 0xf);
2459                 if (sb_len > 7) {
2460                         sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2461                                          sb_len : (sense_buffer[7] + 8);
2462                         if (sb_len > 12)
2463                                 sshdr->asc = sense_buffer[12];
2464                         if (sb_len > 13)
2465                                 sshdr->ascq = sense_buffer[13];
2466                 }
2467         }
2468
2469         return true;
2470 }
2471 EXPORT_SYMBOL(scsi_normalize_sense);
2472
2473 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2474                                   struct scsi_sense_hdr *sshdr)
2475 {
2476         return scsi_normalize_sense(cmd->sense_buffer,
2477                         SCSI_SENSE_BUFFERSIZE, sshdr);
2478 }
2479 EXPORT_SYMBOL(scsi_command_normalize_sense);
2480
2481 /**
2482  * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2483  * @sense_buffer:       byte array of descriptor format sense data
2484  * @sb_len:             number of valid bytes in sense_buffer
2485  * @desc_type:          value of descriptor type to find
2486  *                      (e.g. 0 -> information)
2487  *
2488  * Notes:
2489  *      only valid when sense data is in descriptor format
2490  *
2491  * Return value:
2492  *      pointer to start of (first) descriptor if found else NULL
2493  */
2494 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2495                                 int desc_type)
2496 {
2497         int add_sen_len, add_len, desc_len, k;
2498         const u8 * descp;
2499
2500         if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2501                 return NULL;
2502         if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2503                 return NULL;
2504         add_sen_len = (add_sen_len < (sb_len - 8)) ?
2505                         add_sen_len : (sb_len - 8);
2506         descp = &sense_buffer[8];
2507         for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2508                 descp += desc_len;
2509                 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2510                 desc_len = add_len + 2;
2511                 if (descp[0] == desc_type)
2512                         return descp;
2513                 if (add_len < 0) // short descriptor ??
2514                         break;
2515         }
2516         return NULL;
2517 }
2518 EXPORT_SYMBOL(scsi_sense_desc_find);
2519
2520 /**
2521  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2522  * @sense_buffer:       byte array of sense data
2523  * @sb_len:             number of valid bytes in sense_buffer
2524  * @info_out:           pointer to 64 integer where 8 or 4 byte information
2525  *                      field will be placed if found.
2526  *
2527  * Return value:
2528  *      1 if information field found, 0 if not found.
2529  */
2530 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2531                             u64 * info_out)
2532 {
2533         int j;
2534         const u8 * ucp;
2535         u64 ull;
2536
2537         if (sb_len < 7)
2538                 return 0;
2539         switch (sense_buffer[0] & 0x7f) {
2540         case 0x70:
2541         case 0x71:
2542                 if (sense_buffer[0] & 0x80) {
2543                         *info_out = (sense_buffer[3] << 24) +
2544                                     (sense_buffer[4] << 16) +
2545                                     (sense_buffer[5] << 8) + sense_buffer[6];
2546                         return 1;
2547                 } else
2548                         return 0;
2549         case 0x72:
2550         case 0x73:
2551                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2552                                            0 /* info desc */);
2553                 if (ucp && (0xa == ucp[1])) {
2554                         ull = 0;
2555                         for (j = 0; j < 8; ++j) {
2556                                 if (j > 0)
2557                                         ull <<= 8;
2558                                 ull |= ucp[4 + j];
2559                         }
2560                         *info_out = ull;
2561                         return 1;
2562                 } else
2563                         return 0;
2564         default:
2565                 return 0;
2566         }
2567 }
2568 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2569
2570 /**
2571  * scsi_build_sense_buffer - build sense data in a buffer
2572  * @desc:       Sense format (non zero == descriptor format,
2573  *              0 == fixed format)
2574  * @buf:        Where to build sense data
2575  * @key:        Sense key
2576  * @asc:        Additional sense code
2577  * @ascq:       Additional sense code qualifier
2578  *
2579  **/
2580 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2581 {
2582         if (desc) {
2583                 buf[0] = 0x72;  /* descriptor, current */
2584                 buf[1] = key;
2585                 buf[2] = asc;
2586                 buf[3] = ascq;
2587                 buf[7] = 0;
2588         } else {
2589                 buf[0] = 0x70;  /* fixed, current */
2590                 buf[2] = key;
2591                 buf[7] = 0xa;
2592                 buf[12] = asc;
2593                 buf[13] = ascq;
2594         }
2595 }
2596 EXPORT_SYMBOL(scsi_build_sense_buffer);