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