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