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