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