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