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