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