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_common.h> 37 #include <scsi/scsi_transport.h> 38 #include <scsi/scsi_host.h> 39 #include <scsi/scsi_ioctl.h> 40 #include <scsi/scsi_dh.h> 41 #include <scsi/sg.h> 42 43 #include "scsi_priv.h" 44 #include "scsi_logging.h" 45 #include "scsi_transport_api.h" 46 47 #include <trace/events/scsi.h> 48 49 static void scsi_eh_done(struct scsi_cmnd *scmd); 50 51 /* 52 * These should *probably* be handled by the host itself. 53 * Since it is allowed to sleep, it probably should. 54 */ 55 #define BUS_RESET_SETTLE_TIME (10) 56 #define HOST_RESET_SETTLE_TIME (10) 57 58 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 59 static int scsi_try_to_abort_cmd(struct scsi_host_template *, 60 struct scsi_cmnd *); 61 62 /* called with shost->host_lock held */ 63 void scsi_eh_wakeup(struct Scsi_Host *shost) 64 { 65 if (atomic_read(&shost->host_busy) == shost->host_failed) { 66 trace_scsi_eh_wakeup(shost); 67 wake_up_process(shost->ehandler); 68 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 69 "Waking error handler thread\n")); 70 } 71 } 72 73 /** 74 * scsi_schedule_eh - schedule EH for SCSI host 75 * @shost: SCSI host to invoke error handling on. 76 * 77 * Schedule SCSI EH without scmd. 78 */ 79 void scsi_schedule_eh(struct Scsi_Host *shost) 80 { 81 unsigned long flags; 82 83 spin_lock_irqsave(shost->host_lock, flags); 84 85 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 86 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 87 shost->host_eh_scheduled++; 88 scsi_eh_wakeup(shost); 89 } 90 91 spin_unlock_irqrestore(shost->host_lock, flags); 92 } 93 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 94 95 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 96 { 97 if (!shost->last_reset || shost->eh_deadline == -1) 98 return 0; 99 100 /* 101 * 32bit accesses are guaranteed to be atomic 102 * (on all supported architectures), so instead 103 * of using a spinlock we can as well double check 104 * if eh_deadline has been set to 'off' during the 105 * time_before call. 106 */ 107 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 108 shost->eh_deadline > -1) 109 return 0; 110 111 return 1; 112 } 113 114 /** 115 * scmd_eh_abort_handler - Handle command aborts 116 * @work: command to be aborted. 117 */ 118 void 119 scmd_eh_abort_handler(struct work_struct *work) 120 { 121 struct scsi_cmnd *scmd = 122 container_of(work, struct scsi_cmnd, abort_work.work); 123 struct scsi_device *sdev = scmd->device; 124 int rtn; 125 126 if (scsi_host_eh_past_deadline(sdev->host)) { 127 SCSI_LOG_ERROR_RECOVERY(3, 128 scmd_printk(KERN_INFO, scmd, 129 "eh timeout, not aborting\n")); 130 } else { 131 SCSI_LOG_ERROR_RECOVERY(3, 132 scmd_printk(KERN_INFO, scmd, 133 "aborting command\n")); 134 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); 135 if (rtn == SUCCESS) { 136 set_host_byte(scmd, DID_TIME_OUT); 137 if (scsi_host_eh_past_deadline(sdev->host)) { 138 SCSI_LOG_ERROR_RECOVERY(3, 139 scmd_printk(KERN_INFO, scmd, 140 "eh timeout, not retrying " 141 "aborted command\n")); 142 } else if (!scsi_noretry_cmd(scmd) && 143 (++scmd->retries <= scmd->allowed)) { 144 SCSI_LOG_ERROR_RECOVERY(3, 145 scmd_printk(KERN_WARNING, scmd, 146 "retry aborted command\n")); 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 "finish aborted command\n")); 153 scsi_finish_command(scmd); 154 return; 155 } 156 } else { 157 SCSI_LOG_ERROR_RECOVERY(3, 158 scmd_printk(KERN_INFO, scmd, 159 "cmd abort %s\n", 160 (rtn == FAST_IO_FAIL) ? 161 "not send" : "failed")); 162 } 163 } 164 165 if (!scsi_eh_scmd_add(scmd, 0)) { 166 SCSI_LOG_ERROR_RECOVERY(3, 167 scmd_printk(KERN_WARNING, scmd, 168 "terminate aborted command\n")); 169 set_host_byte(scmd, DID_TIME_OUT); 170 scsi_finish_command(scmd); 171 } 172 } 173 174 /** 175 * scsi_abort_command - schedule a command abort 176 * @scmd: scmd to abort. 177 * 178 * We only need to abort commands after a command timeout 179 */ 180 static int 181 scsi_abort_command(struct scsi_cmnd *scmd) 182 { 183 struct scsi_device *sdev = scmd->device; 184 struct Scsi_Host *shost = sdev->host; 185 unsigned long flags; 186 187 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 188 /* 189 * Retry after abort failed, escalate to next level. 190 */ 191 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED; 192 SCSI_LOG_ERROR_RECOVERY(3, 193 scmd_printk(KERN_INFO, scmd, 194 "previous abort failed\n")); 195 BUG_ON(delayed_work_pending(&scmd->abort_work)); 196 return FAILED; 197 } 198 199 /* 200 * Do not try a command abort if 201 * SCSI EH has already started. 202 */ 203 spin_lock_irqsave(shost->host_lock, flags); 204 if (scsi_host_in_recovery(shost)) { 205 spin_unlock_irqrestore(shost->host_lock, flags); 206 SCSI_LOG_ERROR_RECOVERY(3, 207 scmd_printk(KERN_INFO, scmd, 208 "not aborting, host in recovery\n")); 209 return FAILED; 210 } 211 212 if (shost->eh_deadline != -1 && !shost->last_reset) 213 shost->last_reset = jiffies; 214 spin_unlock_irqrestore(shost->host_lock, flags); 215 216 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 217 SCSI_LOG_ERROR_RECOVERY(3, 218 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 219 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 220 return SUCCESS; 221 } 222 223 /** 224 * scsi_eh_scmd_add - add scsi cmd to error handling. 225 * @scmd: scmd to run eh on. 226 * @eh_flag: optional SCSI_EH flag. 227 * 228 * Return value: 229 * 0 on failure. 230 */ 231 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) 232 { 233 struct Scsi_Host *shost = scmd->device->host; 234 unsigned long flags; 235 int ret = 0; 236 237 if (!shost->ehandler) 238 return 0; 239 240 spin_lock_irqsave(shost->host_lock, flags); 241 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 242 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) 243 goto out_unlock; 244 245 if (shost->eh_deadline != -1 && !shost->last_reset) 246 shost->last_reset = jiffies; 247 248 ret = 1; 249 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 250 eh_flag &= ~SCSI_EH_CANCEL_CMD; 251 scmd->eh_eflags |= eh_flag; 252 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 253 shost->host_failed++; 254 scsi_eh_wakeup(shost); 255 out_unlock: 256 spin_unlock_irqrestore(shost->host_lock, flags); 257 return ret; 258 } 259 260 /** 261 * scsi_times_out - Timeout function for normal scsi commands. 262 * @req: request that is timing out. 263 * 264 * Notes: 265 * We do not need to lock this. There is the potential for a race 266 * only in that the normal completion handling might run, but if the 267 * normal completion function determines that the timer has already 268 * fired, then it mustn't do anything. 269 */ 270 enum blk_eh_timer_return scsi_times_out(struct request *req) 271 { 272 struct scsi_cmnd *scmd = req->special; 273 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; 274 struct Scsi_Host *host = scmd->device->host; 275 276 trace_scsi_dispatch_cmd_timeout(scmd); 277 scsi_log_completion(scmd, TIMEOUT_ERROR); 278 279 if (host->eh_deadline != -1 && !host->last_reset) 280 host->last_reset = jiffies; 281 282 if (host->hostt->eh_timed_out) 283 rtn = host->hostt->eh_timed_out(scmd); 284 285 if (rtn == BLK_EH_NOT_HANDLED) { 286 if (!host->hostt->no_async_abort && 287 scsi_abort_command(scmd) == SUCCESS) 288 return BLK_EH_NOT_HANDLED; 289 290 set_host_byte(scmd, DID_TIME_OUT); 291 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)) 292 rtn = BLK_EH_HANDLED; 293 } 294 295 return rtn; 296 } 297 298 /** 299 * scsi_block_when_processing_errors - Prevent cmds from being queued. 300 * @sdev: Device on which we are performing recovery. 301 * 302 * Description: 303 * We block until the host is out of error recovery, and then check to 304 * see whether the host or the device is offline. 305 * 306 * Return value: 307 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 308 */ 309 int scsi_block_when_processing_errors(struct scsi_device *sdev) 310 { 311 int online; 312 313 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 314 315 online = scsi_device_online(sdev); 316 317 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev, 318 "%s: rtn: %d\n", __func__, online)); 319 320 return online; 321 } 322 EXPORT_SYMBOL(scsi_block_when_processing_errors); 323 324 #ifdef CONFIG_SCSI_LOGGING 325 /** 326 * scsi_eh_prt_fail_stats - Log info on failures. 327 * @shost: scsi host being recovered. 328 * @work_q: Queue of scsi cmds to process. 329 */ 330 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 331 struct list_head *work_q) 332 { 333 struct scsi_cmnd *scmd; 334 struct scsi_device *sdev; 335 int total_failures = 0; 336 int cmd_failed = 0; 337 int cmd_cancel = 0; 338 int devices_failed = 0; 339 340 shost_for_each_device(sdev, shost) { 341 list_for_each_entry(scmd, work_q, eh_entry) { 342 if (scmd->device == sdev) { 343 ++total_failures; 344 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 345 ++cmd_cancel; 346 else 347 ++cmd_failed; 348 } 349 } 350 351 if (cmd_cancel || cmd_failed) { 352 SCSI_LOG_ERROR_RECOVERY(3, 353 shost_printk(KERN_INFO, shost, 354 "%s: cmds failed: %d, cancel: %d\n", 355 __func__, cmd_failed, 356 cmd_cancel)); 357 cmd_cancel = 0; 358 cmd_failed = 0; 359 ++devices_failed; 360 } 361 } 362 363 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 364 "Total of %d commands on %d" 365 " devices require eh work\n", 366 total_failures, devices_failed)); 367 } 368 #endif 369 370 /** 371 * scsi_report_lun_change - Set flag on all *other* devices on the same target 372 * to indicate that a UNIT ATTENTION is expected. 373 * @sdev: Device reporting the UNIT ATTENTION 374 */ 375 static void scsi_report_lun_change(struct scsi_device *sdev) 376 { 377 sdev->sdev_target->expecting_lun_change = 1; 378 } 379 380 /** 381 * scsi_report_sense - Examine scsi sense information and log messages for 382 * certain conditions, also issue uevents for some of them. 383 * @sdev: Device reporting the sense code 384 * @sshdr: sshdr to be examined 385 */ 386 static void scsi_report_sense(struct scsi_device *sdev, 387 struct scsi_sense_hdr *sshdr) 388 { 389 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 390 391 if (sshdr->sense_key == UNIT_ATTENTION) { 392 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 393 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 394 sdev_printk(KERN_WARNING, sdev, 395 "Inquiry data has changed"); 396 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 397 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 398 scsi_report_lun_change(sdev); 399 sdev_printk(KERN_WARNING, sdev, 400 "Warning! Received an indication that the " 401 "LUN assignments on this target have " 402 "changed. The Linux SCSI layer does not " 403 "automatically remap LUN assignments.\n"); 404 } else if (sshdr->asc == 0x3f) 405 sdev_printk(KERN_WARNING, sdev, 406 "Warning! Received an indication that the " 407 "operating parameters on this target have " 408 "changed. The Linux SCSI layer does not " 409 "automatically adjust these parameters.\n"); 410 411 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 412 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 413 sdev_printk(KERN_WARNING, sdev, 414 "Warning! Received an indication that the " 415 "LUN reached a thin provisioning soft " 416 "threshold.\n"); 417 } 418 419 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 420 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 421 sdev_printk(KERN_WARNING, sdev, 422 "Mode parameters changed"); 423 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 424 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 425 sdev_printk(KERN_WARNING, sdev, 426 "Asymmetric access state 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 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 scsi_report_sense(sdev, &sshdr); 462 463 if (scsi_sense_is_deferred(&sshdr)) 464 return NEEDS_RETRY; 465 466 if (sdev->handler && sdev->handler->check_sense) { 467 int rc; 468 469 rc = sdev->handler->check_sense(sdev, &sshdr); 470 if (rc != SCSI_RETURN_NOT_HANDLED) 471 return rc; 472 /* handler does not care. Drop down to default handling */ 473 } 474 475 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 476 /* 477 * nasty: for mid-layer issued TURs, we need to return the 478 * actual sense data without any recovery attempt. For eh 479 * issued ones, we need to try to recover and interpret 480 */ 481 return SUCCESS; 482 483 /* 484 * Previous logic looked for FILEMARK, EOM or ILI which are 485 * mainly associated with tapes and returned SUCCESS. 486 */ 487 if (sshdr.response_code == 0x70) { 488 /* fixed format */ 489 if (scmd->sense_buffer[2] & 0xe0) 490 return SUCCESS; 491 } else { 492 /* 493 * descriptor format: look for "stream commands sense data 494 * descriptor" (see SSC-3). Assume single sense data 495 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 496 */ 497 if ((sshdr.additional_length > 3) && 498 (scmd->sense_buffer[8] == 0x4) && 499 (scmd->sense_buffer[11] & 0xe0)) 500 return SUCCESS; 501 } 502 503 switch (sshdr.sense_key) { 504 case NO_SENSE: 505 return SUCCESS; 506 case RECOVERED_ERROR: 507 return /* soft_error */ SUCCESS; 508 509 case ABORTED_COMMAND: 510 if (sshdr.asc == 0x10) /* DIF */ 511 return SUCCESS; 512 513 return NEEDS_RETRY; 514 case NOT_READY: 515 case UNIT_ATTENTION: 516 /* 517 * if we are expecting a cc/ua because of a bus reset that we 518 * performed, treat this just as a retry. otherwise this is 519 * information that we should pass up to the upper-level driver 520 * so that we can deal with it there. 521 */ 522 if (scmd->device->expecting_cc_ua) { 523 /* 524 * Because some device does not queue unit 525 * attentions correctly, we carefully check 526 * additional sense code and qualifier so as 527 * not to squash media change unit attention. 528 */ 529 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 530 scmd->device->expecting_cc_ua = 0; 531 return NEEDS_RETRY; 532 } 533 } 534 /* 535 * we might also expect a cc/ua if another LUN on the target 536 * reported a UA with an ASC/ASCQ of 3F 0E - 537 * REPORTED LUNS DATA HAS CHANGED. 538 */ 539 if (scmd->device->sdev_target->expecting_lun_change && 540 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 541 return NEEDS_RETRY; 542 /* 543 * if the device is in the process of becoming ready, we 544 * should retry. 545 */ 546 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 547 return NEEDS_RETRY; 548 /* 549 * if the device is not started, we need to wake 550 * the error handler to start the motor 551 */ 552 if (scmd->device->allow_restart && 553 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 554 return FAILED; 555 /* 556 * Pass the UA upwards for a determination in the completion 557 * functions. 558 */ 559 return SUCCESS; 560 561 /* these are not supported */ 562 case DATA_PROTECT: 563 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 564 /* Thin provisioning hard threshold reached */ 565 set_host_byte(scmd, DID_ALLOC_FAILURE); 566 return SUCCESS; 567 } 568 case COPY_ABORTED: 569 case VOLUME_OVERFLOW: 570 case MISCOMPARE: 571 case BLANK_CHECK: 572 set_host_byte(scmd, DID_TARGET_FAILURE); 573 return SUCCESS; 574 575 case MEDIUM_ERROR: 576 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 577 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 578 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 579 set_host_byte(scmd, DID_MEDIUM_ERROR); 580 return SUCCESS; 581 } 582 return NEEDS_RETRY; 583 584 case HARDWARE_ERROR: 585 if (scmd->device->retry_hwerror) 586 return ADD_TO_MLQUEUE; 587 else 588 set_host_byte(scmd, DID_TARGET_FAILURE); 589 590 case ILLEGAL_REQUEST: 591 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 592 sshdr.asc == 0x21 || /* Logical block address out of range */ 593 sshdr.asc == 0x24 || /* Invalid field in cdb */ 594 sshdr.asc == 0x26) { /* Parameter value invalid */ 595 set_host_byte(scmd, DID_TARGET_FAILURE); 596 } 597 return SUCCESS; 598 599 default: 600 return SUCCESS; 601 } 602 } 603 EXPORT_SYMBOL_GPL(scsi_check_sense); 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->track_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 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 633 sdev->last_queue_ramp_up = jiffies; 634 } 635 } 636 637 static void scsi_handle_queue_full(struct scsi_device *sdev) 638 { 639 struct scsi_host_template *sht = sdev->host->hostt; 640 struct scsi_device *tmp_sdev; 641 642 if (!sht->track_queue_depth) 643 return; 644 645 shost_for_each_device(tmp_sdev, sdev->host) { 646 if (tmp_sdev->channel != sdev->channel || 647 tmp_sdev->id != sdev->id) 648 continue; 649 /* 650 * We do not know the number of commands that were at 651 * the device when we got the queue full so we start 652 * from the highest possible value and work our way down. 653 */ 654 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 655 } 656 } 657 658 /** 659 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 660 * @scmd: SCSI cmd to examine. 661 * 662 * Notes: 663 * This is *only* called when we are examining the status of commands 664 * queued during error recovery. the main difference here is that we 665 * don't allow for the possibility of retries here, and we are a lot 666 * more restrictive about what we consider acceptable. 667 */ 668 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 669 { 670 /* 671 * first check the host byte, to see if there is anything in there 672 * that would indicate what we need to do. 673 */ 674 if (host_byte(scmd->result) == DID_RESET) { 675 /* 676 * rats. we are already in the error handler, so we now 677 * get to try and figure out what to do next. if the sense 678 * is valid, we have a pretty good idea of what to do. 679 * if not, we mark it as FAILED. 680 */ 681 return scsi_check_sense(scmd); 682 } 683 if (host_byte(scmd->result) != DID_OK) 684 return FAILED; 685 686 /* 687 * next, check the message byte. 688 */ 689 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 690 return FAILED; 691 692 /* 693 * now, check the status byte to see if this indicates 694 * anything special. 695 */ 696 switch (status_byte(scmd->result)) { 697 case GOOD: 698 scsi_handle_queue_ramp_up(scmd->device); 699 case COMMAND_TERMINATED: 700 return SUCCESS; 701 case CHECK_CONDITION: 702 return scsi_check_sense(scmd); 703 case CONDITION_GOOD: 704 case INTERMEDIATE_GOOD: 705 case INTERMEDIATE_C_GOOD: 706 /* 707 * who knows? FIXME(eric) 708 */ 709 return SUCCESS; 710 case RESERVATION_CONFLICT: 711 if (scmd->cmnd[0] == TEST_UNIT_READY) 712 /* it is a success, we probed the device and 713 * found it */ 714 return SUCCESS; 715 /* otherwise, we failed to send the command */ 716 return FAILED; 717 case QUEUE_FULL: 718 scsi_handle_queue_full(scmd->device); 719 /* fall through */ 720 case BUSY: 721 return NEEDS_RETRY; 722 default: 723 return FAILED; 724 } 725 return FAILED; 726 } 727 728 /** 729 * scsi_eh_done - Completion function for error handling. 730 * @scmd: Cmd that is done. 731 */ 732 static void scsi_eh_done(struct scsi_cmnd *scmd) 733 { 734 struct completion *eh_action; 735 736 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 737 "%s result: %x\n", __func__, scmd->result)); 738 739 eh_action = scmd->device->host->eh_action; 740 if (eh_action) 741 complete(eh_action); 742 } 743 744 /** 745 * scsi_try_host_reset - ask host adapter to reset itself 746 * @scmd: SCSI cmd to send host reset. 747 */ 748 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 749 { 750 unsigned long flags; 751 int rtn; 752 struct Scsi_Host *host = scmd->device->host; 753 struct scsi_host_template *hostt = host->hostt; 754 755 SCSI_LOG_ERROR_RECOVERY(3, 756 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 757 758 if (!hostt->eh_host_reset_handler) 759 return FAILED; 760 761 rtn = hostt->eh_host_reset_handler(scmd); 762 763 if (rtn == SUCCESS) { 764 if (!hostt->skip_settle_delay) 765 ssleep(HOST_RESET_SETTLE_TIME); 766 spin_lock_irqsave(host->host_lock, flags); 767 scsi_report_bus_reset(host, scmd_channel(scmd)); 768 spin_unlock_irqrestore(host->host_lock, flags); 769 } 770 771 return rtn; 772 } 773 774 /** 775 * scsi_try_bus_reset - ask host to perform a bus reset 776 * @scmd: SCSI cmd to send bus reset. 777 */ 778 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 779 { 780 unsigned long flags; 781 int rtn; 782 struct Scsi_Host *host = scmd->device->host; 783 struct scsi_host_template *hostt = host->hostt; 784 785 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 786 "%s: Snd Bus RST\n", __func__)); 787 788 if (!hostt->eh_bus_reset_handler) 789 return FAILED; 790 791 rtn = hostt->eh_bus_reset_handler(scmd); 792 793 if (rtn == SUCCESS) { 794 if (!hostt->skip_settle_delay) 795 ssleep(BUS_RESET_SETTLE_TIME); 796 spin_lock_irqsave(host->host_lock, flags); 797 scsi_report_bus_reset(host, scmd_channel(scmd)); 798 spin_unlock_irqrestore(host->host_lock, flags); 799 } 800 801 return rtn; 802 } 803 804 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 805 { 806 sdev->was_reset = 1; 807 sdev->expecting_cc_ua = 1; 808 } 809 810 /** 811 * scsi_try_target_reset - Ask host to perform a target reset 812 * @scmd: SCSI cmd used to send a target reset 813 * 814 * Notes: 815 * There is no timeout for this operation. if this operation is 816 * unreliable for a given host, then the host itself needs to put a 817 * timer on it, and set the host back to a consistent state prior to 818 * returning. 819 */ 820 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 821 { 822 unsigned long flags; 823 int rtn; 824 struct Scsi_Host *host = scmd->device->host; 825 struct scsi_host_template *hostt = host->hostt; 826 827 if (!hostt->eh_target_reset_handler) 828 return FAILED; 829 830 rtn = hostt->eh_target_reset_handler(scmd); 831 if (rtn == SUCCESS) { 832 spin_lock_irqsave(host->host_lock, flags); 833 __starget_for_each_device(scsi_target(scmd->device), NULL, 834 __scsi_report_device_reset); 835 spin_unlock_irqrestore(host->host_lock, flags); 836 } 837 838 return rtn; 839 } 840 841 /** 842 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 843 * @scmd: SCSI cmd used to send BDR 844 * 845 * Notes: 846 * There is no timeout for this operation. if this operation is 847 * unreliable for a given host, then the host itself needs to put a 848 * timer on it, and set the host back to a consistent state prior to 849 * returning. 850 */ 851 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 852 { 853 int rtn; 854 struct scsi_host_template *hostt = scmd->device->host->hostt; 855 856 if (!hostt->eh_device_reset_handler) 857 return FAILED; 858 859 rtn = hostt->eh_device_reset_handler(scmd); 860 if (rtn == SUCCESS) 861 __scsi_report_device_reset(scmd->device, NULL); 862 return rtn; 863 } 864 865 /** 866 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 867 * @hostt: SCSI driver host template 868 * @scmd: SCSI cmd used to send a target reset 869 * 870 * Return value: 871 * SUCCESS, FAILED, or FAST_IO_FAIL 872 * 873 * Notes: 874 * SUCCESS does not necessarily indicate that the command 875 * has been aborted; it only indicates that the LLDDs 876 * has cleared all references to that command. 877 * LLDDs should return FAILED only if an abort was required 878 * but could not be executed. LLDDs should return FAST_IO_FAIL 879 * if the device is temporarily unavailable (eg due to a 880 * link down on FibreChannel) 881 */ 882 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, 883 struct scsi_cmnd *scmd) 884 { 885 if (!hostt->eh_abort_handler) 886 return FAILED; 887 888 return hostt->eh_abort_handler(scmd); 889 } 890 891 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 892 { 893 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 894 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 895 if (scsi_try_target_reset(scmd) != SUCCESS) 896 if (scsi_try_bus_reset(scmd) != SUCCESS) 897 scsi_try_host_reset(scmd); 898 } 899 900 /** 901 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 902 * @scmd: SCSI command structure to hijack 903 * @ses: structure to save restore information 904 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 905 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 906 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 907 * 908 * This function is used to save a scsi command information before re-execution 909 * as part of the error recovery process. If @sense_bytes is 0 the command 910 * sent must be one that does not transfer any data. If @sense_bytes != 0 911 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 912 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 913 */ 914 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 915 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 916 { 917 struct scsi_device *sdev = scmd->device; 918 919 /* 920 * We need saved copies of a number of fields - this is because 921 * error handling may need to overwrite these with different values 922 * to run different commands, and once error handling is complete, 923 * we will need to restore these values prior to running the actual 924 * command. 925 */ 926 ses->cmd_len = scmd->cmd_len; 927 ses->cmnd = scmd->cmnd; 928 ses->data_direction = scmd->sc_data_direction; 929 ses->sdb = scmd->sdb; 930 ses->next_rq = scmd->request->next_rq; 931 ses->result = scmd->result; 932 ses->underflow = scmd->underflow; 933 ses->prot_op = scmd->prot_op; 934 935 scmd->prot_op = SCSI_PROT_NORMAL; 936 scmd->eh_eflags = 0; 937 scmd->cmnd = ses->eh_cmnd; 938 memset(scmd->cmnd, 0, BLK_MAX_CDB); 939 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 940 scmd->request->next_rq = NULL; 941 scmd->result = 0; 942 943 if (sense_bytes) { 944 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 945 sense_bytes); 946 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 947 scmd->sdb.length); 948 scmd->sdb.table.sgl = &ses->sense_sgl; 949 scmd->sc_data_direction = DMA_FROM_DEVICE; 950 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 951 scmd->cmnd[0] = REQUEST_SENSE; 952 scmd->cmnd[4] = scmd->sdb.length; 953 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 954 } else { 955 scmd->sc_data_direction = DMA_NONE; 956 if (cmnd) { 957 BUG_ON(cmnd_size > BLK_MAX_CDB); 958 memcpy(scmd->cmnd, cmnd, cmnd_size); 959 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 960 } 961 } 962 963 scmd->underflow = 0; 964 965 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 966 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 967 (sdev->lun << 5 & 0xe0); 968 969 /* 970 * Zero the sense buffer. The scsi spec mandates that any 971 * untransferred sense data should be interpreted as being zero. 972 */ 973 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 974 } 975 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 976 977 /** 978 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 979 * @scmd: SCSI command structure to restore 980 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 981 * 982 * Undo any damage done by above scsi_eh_prep_cmnd(). 983 */ 984 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 985 { 986 /* 987 * Restore original data 988 */ 989 scmd->cmd_len = ses->cmd_len; 990 scmd->cmnd = ses->cmnd; 991 scmd->sc_data_direction = ses->data_direction; 992 scmd->sdb = ses->sdb; 993 scmd->request->next_rq = ses->next_rq; 994 scmd->result = ses->result; 995 scmd->underflow = ses->underflow; 996 scmd->prot_op = ses->prot_op; 997 } 998 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 999 1000 /** 1001 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1002 * @scmd: SCSI command structure to hijack 1003 * @cmnd: CDB to send 1004 * @cmnd_size: size in bytes of @cmnd 1005 * @timeout: timeout for this request 1006 * @sense_bytes: size of sense data to copy or 0 1007 * 1008 * This function is used to send a scsi command down to a target device 1009 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1010 * 1011 * Return value: 1012 * SUCCESS or FAILED or NEEDS_RETRY 1013 */ 1014 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 1015 int cmnd_size, int timeout, unsigned sense_bytes) 1016 { 1017 struct scsi_device *sdev = scmd->device; 1018 struct Scsi_Host *shost = sdev->host; 1019 DECLARE_COMPLETION_ONSTACK(done); 1020 unsigned long timeleft = timeout; 1021 struct scsi_eh_save ses; 1022 const unsigned long stall_for = msecs_to_jiffies(100); 1023 int rtn; 1024 1025 retry: 1026 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1027 shost->eh_action = &done; 1028 1029 scsi_log_send(scmd); 1030 scmd->scsi_done = scsi_eh_done; 1031 rtn = shost->hostt->queuecommand(shost, scmd); 1032 if (rtn) { 1033 if (timeleft > stall_for) { 1034 scsi_eh_restore_cmnd(scmd, &ses); 1035 timeleft -= stall_for; 1036 msleep(jiffies_to_msecs(stall_for)); 1037 goto retry; 1038 } 1039 /* signal not to enter either branch of the if () below */ 1040 timeleft = 0; 1041 rtn = FAILED; 1042 } else { 1043 timeleft = wait_for_completion_timeout(&done, timeout); 1044 rtn = SUCCESS; 1045 } 1046 1047 shost->eh_action = NULL; 1048 1049 scsi_log_completion(scmd, rtn); 1050 1051 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1052 "%s timeleft: %ld\n", 1053 __func__, timeleft)); 1054 1055 /* 1056 * If there is time left scsi_eh_done got called, and we will examine 1057 * the actual status codes to see whether the command actually did 1058 * complete normally, else if we have a zero return and no time left, 1059 * the command must still be pending, so abort it and return FAILED. 1060 * If we never actually managed to issue the command, because 1061 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1062 * value above (so don't execute either branch of the if) 1063 */ 1064 if (timeleft) { 1065 rtn = scsi_eh_completed_normally(scmd); 1066 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1067 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1068 1069 switch (rtn) { 1070 case SUCCESS: 1071 case NEEDS_RETRY: 1072 case FAILED: 1073 break; 1074 case ADD_TO_MLQUEUE: 1075 rtn = NEEDS_RETRY; 1076 break; 1077 default: 1078 rtn = FAILED; 1079 break; 1080 } 1081 } else if (rtn != FAILED) { 1082 scsi_abort_eh_cmnd(scmd); 1083 rtn = FAILED; 1084 } 1085 1086 scsi_eh_restore_cmnd(scmd, &ses); 1087 1088 return rtn; 1089 } 1090 1091 /** 1092 * scsi_request_sense - Request sense data from a particular target. 1093 * @scmd: SCSI cmd for request sense. 1094 * 1095 * Notes: 1096 * Some hosts automatically obtain this information, others require 1097 * that we obtain it on our own. This function will *not* return until 1098 * the command either times out, or it completes. 1099 */ 1100 static int scsi_request_sense(struct scsi_cmnd *scmd) 1101 { 1102 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1103 } 1104 1105 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) 1106 { 1107 if (!blk_rq_is_passthrough(scmd->request)) { 1108 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1109 if (sdrv->eh_action) 1110 rtn = sdrv->eh_action(scmd, rtn); 1111 } 1112 return rtn; 1113 } 1114 1115 /** 1116 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1117 * @scmd: Original SCSI cmd that eh has finished. 1118 * @done_q: Queue for processed commands. 1119 * 1120 * Notes: 1121 * We don't want to use the normal command completion while we are are 1122 * still handling errors - it may cause other commands to be queued, 1123 * and that would disturb what we are doing. Thus we really want to 1124 * keep a list of pending commands for final completion, and once we 1125 * are ready to leave error handling we handle completion for real. 1126 */ 1127 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1128 { 1129 scmd->eh_eflags = 0; 1130 list_move_tail(&scmd->eh_entry, done_q); 1131 } 1132 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1133 1134 /** 1135 * scsi_eh_get_sense - Get device sense data. 1136 * @work_q: Queue of commands to process. 1137 * @done_q: Queue of processed commands. 1138 * 1139 * Description: 1140 * See if we need to request sense information. if so, then get it 1141 * now, so we have a better idea of what to do. 1142 * 1143 * Notes: 1144 * This has the unfortunate side effect that if a shost adapter does 1145 * not automatically request sense information, we end up shutting 1146 * it down before we request it. 1147 * 1148 * All drivers should request sense information internally these days, 1149 * so for now all I have to say is tough noogies if you end up in here. 1150 * 1151 * XXX: Long term this code should go away, but that needs an audit of 1152 * all LLDDs first. 1153 */ 1154 int scsi_eh_get_sense(struct list_head *work_q, 1155 struct list_head *done_q) 1156 { 1157 struct scsi_cmnd *scmd, *next; 1158 struct Scsi_Host *shost; 1159 int rtn; 1160 1161 /* 1162 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1163 * should not get sense. 1164 */ 1165 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1166 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 1167 (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1168 SCSI_SENSE_VALID(scmd)) 1169 continue; 1170 1171 shost = scmd->device->host; 1172 if (scsi_host_eh_past_deadline(shost)) { 1173 SCSI_LOG_ERROR_RECOVERY(3, 1174 scmd_printk(KERN_INFO, scmd, 1175 "%s: skip request sense, past eh deadline\n", 1176 current->comm)); 1177 break; 1178 } 1179 if (status_byte(scmd->result) != CHECK_CONDITION) 1180 /* 1181 * don't request sense if there's no check condition 1182 * status because the error we're processing isn't one 1183 * that has a sense code (and some devices get 1184 * confused by sense requests out of the blue) 1185 */ 1186 continue; 1187 1188 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1189 "%s: requesting sense\n", 1190 current->comm)); 1191 rtn = scsi_request_sense(scmd); 1192 if (rtn != SUCCESS) 1193 continue; 1194 1195 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1196 "sense requested, result %x\n", scmd->result)); 1197 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1198 1199 rtn = scsi_decide_disposition(scmd); 1200 1201 /* 1202 * if the result was normal, then just pass it along to the 1203 * upper level. 1204 */ 1205 if (rtn == SUCCESS) 1206 /* we don't want this command reissued, just 1207 * finished with the sense data, so set 1208 * retries to the max allowed to ensure it 1209 * won't get reissued */ 1210 scmd->retries = scmd->allowed; 1211 else if (rtn != NEEDS_RETRY) 1212 continue; 1213 1214 scsi_eh_finish_cmd(scmd, done_q); 1215 } 1216 1217 return list_empty(work_q); 1218 } 1219 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1220 1221 /** 1222 * scsi_eh_tur - Send TUR to device. 1223 * @scmd: &scsi_cmnd to send TUR 1224 * 1225 * Return value: 1226 * 0 - Device is ready. 1 - Device NOT ready. 1227 */ 1228 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1229 { 1230 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1231 int retry_cnt = 1, rtn; 1232 1233 retry_tur: 1234 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1235 scmd->device->eh_timeout, 0); 1236 1237 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1238 "%s return: %x\n", __func__, rtn)); 1239 1240 switch (rtn) { 1241 case NEEDS_RETRY: 1242 if (retry_cnt--) 1243 goto retry_tur; 1244 /*FALLTHRU*/ 1245 case SUCCESS: 1246 return 0; 1247 default: 1248 return 1; 1249 } 1250 } 1251 1252 /** 1253 * scsi_eh_test_devices - check if devices are responding from error recovery. 1254 * @cmd_list: scsi commands in error recovery. 1255 * @work_q: queue for commands which still need more error recovery 1256 * @done_q: queue for commands which are finished 1257 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1258 * 1259 * Decription: 1260 * Tests if devices are in a working state. Commands to devices now in 1261 * a working state are sent to the done_q while commands to devices which 1262 * are still failing to respond are returned to the work_q for more 1263 * processing. 1264 **/ 1265 static int scsi_eh_test_devices(struct list_head *cmd_list, 1266 struct list_head *work_q, 1267 struct list_head *done_q, int try_stu) 1268 { 1269 struct scsi_cmnd *scmd, *next; 1270 struct scsi_device *sdev; 1271 int finish_cmds; 1272 1273 while (!list_empty(cmd_list)) { 1274 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1275 sdev = scmd->device; 1276 1277 if (!try_stu) { 1278 if (scsi_host_eh_past_deadline(sdev->host)) { 1279 /* Push items back onto work_q */ 1280 list_splice_init(cmd_list, work_q); 1281 SCSI_LOG_ERROR_RECOVERY(3, 1282 sdev_printk(KERN_INFO, sdev, 1283 "%s: skip test device, past eh deadline", 1284 current->comm)); 1285 break; 1286 } 1287 } 1288 1289 finish_cmds = !scsi_device_online(scmd->device) || 1290 (try_stu && !scsi_eh_try_stu(scmd) && 1291 !scsi_eh_tur(scmd)) || 1292 !scsi_eh_tur(scmd); 1293 1294 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1295 if (scmd->device == sdev) { 1296 if (finish_cmds && 1297 (try_stu || 1298 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1299 scsi_eh_finish_cmd(scmd, done_q); 1300 else 1301 list_move_tail(&scmd->eh_entry, work_q); 1302 } 1303 } 1304 return list_empty(work_q); 1305 } 1306 1307 1308 /** 1309 * scsi_eh_abort_cmds - abort pending commands. 1310 * @work_q: &list_head for pending commands. 1311 * @done_q: &list_head for processed commands. 1312 * 1313 * Decription: 1314 * Try and see whether or not it makes sense to try and abort the 1315 * running command. This only works out to be the case if we have one 1316 * command that has timed out. If the command simply failed, it makes 1317 * no sense to try and abort the command, since as far as the shost 1318 * adapter is concerned, it isn't running. 1319 */ 1320 static int scsi_eh_abort_cmds(struct list_head *work_q, 1321 struct list_head *done_q) 1322 { 1323 struct scsi_cmnd *scmd, *next; 1324 LIST_HEAD(check_list); 1325 int rtn; 1326 struct Scsi_Host *shost; 1327 1328 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1329 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 1330 continue; 1331 shost = scmd->device->host; 1332 if (scsi_host_eh_past_deadline(shost)) { 1333 list_splice_init(&check_list, work_q); 1334 SCSI_LOG_ERROR_RECOVERY(3, 1335 scmd_printk(KERN_INFO, scmd, 1336 "%s: skip aborting cmd, past eh deadline\n", 1337 current->comm)); 1338 return list_empty(work_q); 1339 } 1340 SCSI_LOG_ERROR_RECOVERY(3, 1341 scmd_printk(KERN_INFO, scmd, 1342 "%s: aborting cmd\n", current->comm)); 1343 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); 1344 if (rtn == FAILED) { 1345 SCSI_LOG_ERROR_RECOVERY(3, 1346 scmd_printk(KERN_INFO, scmd, 1347 "%s: aborting cmd failed\n", 1348 current->comm)); 1349 list_splice_init(&check_list, work_q); 1350 return list_empty(work_q); 1351 } 1352 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 1353 if (rtn == FAST_IO_FAIL) 1354 scsi_eh_finish_cmd(scmd, done_q); 1355 else 1356 list_move_tail(&scmd->eh_entry, &check_list); 1357 } 1358 1359 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1360 } 1361 1362 /** 1363 * scsi_eh_try_stu - Send START_UNIT to device. 1364 * @scmd: &scsi_cmnd to send START_UNIT 1365 * 1366 * Return value: 1367 * 0 - Device is ready. 1 - Device NOT ready. 1368 */ 1369 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1370 { 1371 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1372 1373 if (scmd->device->allow_restart) { 1374 int i, rtn = NEEDS_RETRY; 1375 1376 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1377 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1378 1379 if (rtn == SUCCESS) 1380 return 0; 1381 } 1382 1383 return 1; 1384 } 1385 1386 /** 1387 * scsi_eh_stu - send START_UNIT if needed 1388 * @shost: &scsi host being recovered. 1389 * @work_q: &list_head for pending commands. 1390 * @done_q: &list_head for processed commands. 1391 * 1392 * Notes: 1393 * If commands are failing due to not ready, initializing command required, 1394 * try revalidating the device, which will end up sending a start unit. 1395 */ 1396 static int scsi_eh_stu(struct Scsi_Host *shost, 1397 struct list_head *work_q, 1398 struct list_head *done_q) 1399 { 1400 struct scsi_cmnd *scmd, *stu_scmd, *next; 1401 struct scsi_device *sdev; 1402 1403 shost_for_each_device(sdev, shost) { 1404 if (scsi_host_eh_past_deadline(shost)) { 1405 SCSI_LOG_ERROR_RECOVERY(3, 1406 sdev_printk(KERN_INFO, sdev, 1407 "%s: skip START_UNIT, past eh deadline\n", 1408 current->comm)); 1409 break; 1410 } 1411 stu_scmd = NULL; 1412 list_for_each_entry(scmd, work_q, eh_entry) 1413 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1414 scsi_check_sense(scmd) == FAILED ) { 1415 stu_scmd = scmd; 1416 break; 1417 } 1418 1419 if (!stu_scmd) 1420 continue; 1421 1422 SCSI_LOG_ERROR_RECOVERY(3, 1423 sdev_printk(KERN_INFO, sdev, 1424 "%s: Sending START_UNIT\n", 1425 current->comm)); 1426 1427 if (!scsi_eh_try_stu(stu_scmd)) { 1428 if (!scsi_device_online(sdev) || 1429 !scsi_eh_tur(stu_scmd)) { 1430 list_for_each_entry_safe(scmd, next, 1431 work_q, eh_entry) { 1432 if (scmd->device == sdev && 1433 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1434 scsi_eh_finish_cmd(scmd, done_q); 1435 } 1436 } 1437 } else { 1438 SCSI_LOG_ERROR_RECOVERY(3, 1439 sdev_printk(KERN_INFO, sdev, 1440 "%s: START_UNIT failed\n", 1441 current->comm)); 1442 } 1443 } 1444 1445 return list_empty(work_q); 1446 } 1447 1448 1449 /** 1450 * scsi_eh_bus_device_reset - send bdr if needed 1451 * @shost: scsi host being recovered. 1452 * @work_q: &list_head for pending commands. 1453 * @done_q: &list_head for processed commands. 1454 * 1455 * Notes: 1456 * Try a bus device reset. Still, look to see whether we have multiple 1457 * devices that are jammed or not - if we have multiple devices, it 1458 * makes no sense to try bus_device_reset - we really would need to try 1459 * a bus_reset instead. 1460 */ 1461 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1462 struct list_head *work_q, 1463 struct list_head *done_q) 1464 { 1465 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1466 struct scsi_device *sdev; 1467 int rtn; 1468 1469 shost_for_each_device(sdev, shost) { 1470 if (scsi_host_eh_past_deadline(shost)) { 1471 SCSI_LOG_ERROR_RECOVERY(3, 1472 sdev_printk(KERN_INFO, sdev, 1473 "%s: skip BDR, past eh deadline\n", 1474 current->comm)); 1475 break; 1476 } 1477 bdr_scmd = NULL; 1478 list_for_each_entry(scmd, work_q, eh_entry) 1479 if (scmd->device == sdev) { 1480 bdr_scmd = scmd; 1481 break; 1482 } 1483 1484 if (!bdr_scmd) 1485 continue; 1486 1487 SCSI_LOG_ERROR_RECOVERY(3, 1488 sdev_printk(KERN_INFO, sdev, 1489 "%s: Sending BDR\n", current->comm)); 1490 rtn = scsi_try_bus_device_reset(bdr_scmd); 1491 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1492 if (!scsi_device_online(sdev) || 1493 rtn == FAST_IO_FAIL || 1494 !scsi_eh_tur(bdr_scmd)) { 1495 list_for_each_entry_safe(scmd, next, 1496 work_q, eh_entry) { 1497 if (scmd->device == sdev && 1498 scsi_eh_action(scmd, rtn) != FAILED) 1499 scsi_eh_finish_cmd(scmd, 1500 done_q); 1501 } 1502 } 1503 } else { 1504 SCSI_LOG_ERROR_RECOVERY(3, 1505 sdev_printk(KERN_INFO, sdev, 1506 "%s: BDR failed\n", current->comm)); 1507 } 1508 } 1509 1510 return list_empty(work_q); 1511 } 1512 1513 /** 1514 * scsi_eh_target_reset - send target reset if needed 1515 * @shost: scsi host being recovered. 1516 * @work_q: &list_head for pending commands. 1517 * @done_q: &list_head for processed commands. 1518 * 1519 * Notes: 1520 * Try a target reset. 1521 */ 1522 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1523 struct list_head *work_q, 1524 struct list_head *done_q) 1525 { 1526 LIST_HEAD(tmp_list); 1527 LIST_HEAD(check_list); 1528 1529 list_splice_init(work_q, &tmp_list); 1530 1531 while (!list_empty(&tmp_list)) { 1532 struct scsi_cmnd *next, *scmd; 1533 int rtn; 1534 unsigned int id; 1535 1536 if (scsi_host_eh_past_deadline(shost)) { 1537 /* push back on work queue for further processing */ 1538 list_splice_init(&check_list, work_q); 1539 list_splice_init(&tmp_list, work_q); 1540 SCSI_LOG_ERROR_RECOVERY(3, 1541 shost_printk(KERN_INFO, shost, 1542 "%s: Skip target reset, past eh deadline\n", 1543 current->comm)); 1544 return list_empty(work_q); 1545 } 1546 1547 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1548 id = scmd_id(scmd); 1549 1550 SCSI_LOG_ERROR_RECOVERY(3, 1551 shost_printk(KERN_INFO, shost, 1552 "%s: Sending target reset to target %d\n", 1553 current->comm, id)); 1554 rtn = scsi_try_target_reset(scmd); 1555 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1556 SCSI_LOG_ERROR_RECOVERY(3, 1557 shost_printk(KERN_INFO, shost, 1558 "%s: Target reset failed" 1559 " target: %d\n", 1560 current->comm, id)); 1561 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1562 if (scmd_id(scmd) != id) 1563 continue; 1564 1565 if (rtn == SUCCESS) 1566 list_move_tail(&scmd->eh_entry, &check_list); 1567 else if (rtn == FAST_IO_FAIL) 1568 scsi_eh_finish_cmd(scmd, done_q); 1569 else 1570 /* push back on work queue for further processing */ 1571 list_move(&scmd->eh_entry, work_q); 1572 } 1573 } 1574 1575 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1576 } 1577 1578 /** 1579 * scsi_eh_bus_reset - send a bus reset 1580 * @shost: &scsi host being recovered. 1581 * @work_q: &list_head for pending commands. 1582 * @done_q: &list_head for processed commands. 1583 */ 1584 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1585 struct list_head *work_q, 1586 struct list_head *done_q) 1587 { 1588 struct scsi_cmnd *scmd, *chan_scmd, *next; 1589 LIST_HEAD(check_list); 1590 unsigned int channel; 1591 int rtn; 1592 1593 /* 1594 * we really want to loop over the various channels, and do this on 1595 * a channel by channel basis. we should also check to see if any 1596 * of the failed commands are on soft_reset devices, and if so, skip 1597 * the reset. 1598 */ 1599 1600 for (channel = 0; channel <= shost->max_channel; channel++) { 1601 if (scsi_host_eh_past_deadline(shost)) { 1602 list_splice_init(&check_list, work_q); 1603 SCSI_LOG_ERROR_RECOVERY(3, 1604 shost_printk(KERN_INFO, shost, 1605 "%s: skip BRST, past eh deadline\n", 1606 current->comm)); 1607 return list_empty(work_q); 1608 } 1609 1610 chan_scmd = NULL; 1611 list_for_each_entry(scmd, work_q, eh_entry) { 1612 if (channel == scmd_channel(scmd)) { 1613 chan_scmd = scmd; 1614 break; 1615 /* 1616 * FIXME add back in some support for 1617 * soft_reset devices. 1618 */ 1619 } 1620 } 1621 1622 if (!chan_scmd) 1623 continue; 1624 SCSI_LOG_ERROR_RECOVERY(3, 1625 shost_printk(KERN_INFO, shost, 1626 "%s: Sending BRST chan: %d\n", 1627 current->comm, channel)); 1628 rtn = scsi_try_bus_reset(chan_scmd); 1629 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1630 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1631 if (channel == scmd_channel(scmd)) { 1632 if (rtn == FAST_IO_FAIL) 1633 scsi_eh_finish_cmd(scmd, 1634 done_q); 1635 else 1636 list_move_tail(&scmd->eh_entry, 1637 &check_list); 1638 } 1639 } 1640 } else { 1641 SCSI_LOG_ERROR_RECOVERY(3, 1642 shost_printk(KERN_INFO, shost, 1643 "%s: BRST failed chan: %d\n", 1644 current->comm, channel)); 1645 } 1646 } 1647 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1648 } 1649 1650 /** 1651 * scsi_eh_host_reset - send a host reset 1652 * @shost: host to be reset. 1653 * @work_q: &list_head for pending commands. 1654 * @done_q: &list_head for processed commands. 1655 */ 1656 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1657 struct list_head *work_q, 1658 struct list_head *done_q) 1659 { 1660 struct scsi_cmnd *scmd, *next; 1661 LIST_HEAD(check_list); 1662 int rtn; 1663 1664 if (!list_empty(work_q)) { 1665 scmd = list_entry(work_q->next, 1666 struct scsi_cmnd, eh_entry); 1667 1668 SCSI_LOG_ERROR_RECOVERY(3, 1669 shost_printk(KERN_INFO, shost, 1670 "%s: Sending HRST\n", 1671 current->comm)); 1672 1673 rtn = scsi_try_host_reset(scmd); 1674 if (rtn == SUCCESS) { 1675 list_splice_init(work_q, &check_list); 1676 } else if (rtn == FAST_IO_FAIL) { 1677 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1678 scsi_eh_finish_cmd(scmd, done_q); 1679 } 1680 } else { 1681 SCSI_LOG_ERROR_RECOVERY(3, 1682 shost_printk(KERN_INFO, shost, 1683 "%s: HRST failed\n", 1684 current->comm)); 1685 } 1686 } 1687 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1688 } 1689 1690 /** 1691 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1692 * @work_q: &list_head for pending commands. 1693 * @done_q: &list_head for processed commands. 1694 */ 1695 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1696 struct list_head *done_q) 1697 { 1698 struct scsi_cmnd *scmd, *next; 1699 1700 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1701 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1702 "not ready after error recovery\n"); 1703 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1704 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1705 /* 1706 * FIXME: Handle lost cmds. 1707 */ 1708 } 1709 scsi_eh_finish_cmd(scmd, done_q); 1710 } 1711 return; 1712 } 1713 1714 /** 1715 * scsi_noretry_cmd - determine if command should be failed fast 1716 * @scmd: SCSI cmd to examine. 1717 */ 1718 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1719 { 1720 switch (host_byte(scmd->result)) { 1721 case DID_OK: 1722 break; 1723 case DID_TIME_OUT: 1724 goto check_type; 1725 case DID_BUS_BUSY: 1726 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1727 case DID_PARITY: 1728 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1729 case DID_ERROR: 1730 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1731 status_byte(scmd->result) == RESERVATION_CONFLICT) 1732 return 0; 1733 /* fall through */ 1734 case DID_SOFT_ERROR: 1735 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1736 } 1737 1738 if (status_byte(scmd->result) != CHECK_CONDITION) 1739 return 0; 1740 1741 check_type: 1742 /* 1743 * assume caller has checked sense and determined 1744 * the check condition was retryable. 1745 */ 1746 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1747 blk_rq_is_passthrough(scmd->request)) 1748 return 1; 1749 else 1750 return 0; 1751 } 1752 1753 /** 1754 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1755 * @scmd: SCSI cmd to examine. 1756 * 1757 * Notes: 1758 * This is *only* called when we are examining the status after sending 1759 * out the actual data command. any commands that are queued for error 1760 * recovery (e.g. test_unit_ready) do *not* come through here. 1761 * 1762 * When this routine returns failed, it means the error handler thread 1763 * is woken. In cases where the error code indicates an error that 1764 * doesn't require the error handler read (i.e. we don't need to 1765 * abort/reset), this function should return SUCCESS. 1766 */ 1767 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1768 { 1769 int rtn; 1770 1771 /* 1772 * if the device is offline, then we clearly just pass the result back 1773 * up to the top level. 1774 */ 1775 if (!scsi_device_online(scmd->device)) { 1776 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1777 "%s: device offline - report as SUCCESS\n", __func__)); 1778 return SUCCESS; 1779 } 1780 1781 /* 1782 * first check the host byte, to see if there is anything in there 1783 * that would indicate what we need to do. 1784 */ 1785 switch (host_byte(scmd->result)) { 1786 case DID_PASSTHROUGH: 1787 /* 1788 * no matter what, pass this through to the upper layer. 1789 * nuke this special code so that it looks like we are saying 1790 * did_ok. 1791 */ 1792 scmd->result &= 0xff00ffff; 1793 return SUCCESS; 1794 case DID_OK: 1795 /* 1796 * looks good. drop through, and check the next byte. 1797 */ 1798 break; 1799 case DID_ABORT: 1800 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1801 set_host_byte(scmd, DID_TIME_OUT); 1802 return SUCCESS; 1803 } 1804 case DID_NO_CONNECT: 1805 case DID_BAD_TARGET: 1806 /* 1807 * note - this means that we just report the status back 1808 * to the top level driver, not that we actually think 1809 * that it indicates SUCCESS. 1810 */ 1811 return SUCCESS; 1812 /* 1813 * when the low level driver returns did_soft_error, 1814 * it is responsible for keeping an internal retry counter 1815 * in order to avoid endless loops (db) 1816 * 1817 * actually this is a bug in this function here. we should 1818 * be mindful of the maximum number of retries specified 1819 * and not get stuck in a loop. 1820 */ 1821 case DID_SOFT_ERROR: 1822 goto maybe_retry; 1823 case DID_IMM_RETRY: 1824 return NEEDS_RETRY; 1825 1826 case DID_REQUEUE: 1827 return ADD_TO_MLQUEUE; 1828 case DID_TRANSPORT_DISRUPTED: 1829 /* 1830 * LLD/transport was disrupted during processing of the IO. 1831 * The transport class is now blocked/blocking, 1832 * and the transport will decide what to do with the IO 1833 * based on its timers and recovery capablilities if 1834 * there are enough retries. 1835 */ 1836 goto maybe_retry; 1837 case DID_TRANSPORT_FAILFAST: 1838 /* 1839 * The transport decided to failfast the IO (most likely 1840 * the fast io fail tmo fired), so send IO directly upwards. 1841 */ 1842 return SUCCESS; 1843 case DID_ERROR: 1844 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1845 status_byte(scmd->result) == RESERVATION_CONFLICT) 1846 /* 1847 * execute reservation conflict processing code 1848 * lower down 1849 */ 1850 break; 1851 /* fallthrough */ 1852 case DID_BUS_BUSY: 1853 case DID_PARITY: 1854 goto maybe_retry; 1855 case DID_TIME_OUT: 1856 /* 1857 * when we scan the bus, we get timeout messages for 1858 * these commands if there is no device available. 1859 * other hosts report did_no_connect for the same thing. 1860 */ 1861 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1862 scmd->cmnd[0] == INQUIRY)) { 1863 return SUCCESS; 1864 } else { 1865 return FAILED; 1866 } 1867 case DID_RESET: 1868 return SUCCESS; 1869 default: 1870 return FAILED; 1871 } 1872 1873 /* 1874 * next, check the message byte. 1875 */ 1876 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1877 return FAILED; 1878 1879 /* 1880 * check the status byte to see if this indicates anything special. 1881 */ 1882 switch (status_byte(scmd->result)) { 1883 case QUEUE_FULL: 1884 scsi_handle_queue_full(scmd->device); 1885 /* 1886 * the case of trying to send too many commands to a 1887 * tagged queueing device. 1888 */ 1889 case BUSY: 1890 /* 1891 * device can't talk to us at the moment. Should only 1892 * occur (SAM-3) when the task queue is empty, so will cause 1893 * the empty queue handling to trigger a stall in the 1894 * device. 1895 */ 1896 return ADD_TO_MLQUEUE; 1897 case GOOD: 1898 if (scmd->cmnd[0] == REPORT_LUNS) 1899 scmd->device->sdev_target->expecting_lun_change = 0; 1900 scsi_handle_queue_ramp_up(scmd->device); 1901 case COMMAND_TERMINATED: 1902 return SUCCESS; 1903 case TASK_ABORTED: 1904 goto maybe_retry; 1905 case CHECK_CONDITION: 1906 rtn = scsi_check_sense(scmd); 1907 if (rtn == NEEDS_RETRY) 1908 goto maybe_retry; 1909 /* if rtn == FAILED, we have no sense information; 1910 * returning FAILED will wake the error handler thread 1911 * to collect the sense and redo the decide 1912 * disposition */ 1913 return rtn; 1914 case CONDITION_GOOD: 1915 case INTERMEDIATE_GOOD: 1916 case INTERMEDIATE_C_GOOD: 1917 case ACA_ACTIVE: 1918 /* 1919 * who knows? FIXME(eric) 1920 */ 1921 return SUCCESS; 1922 1923 case RESERVATION_CONFLICT: 1924 sdev_printk(KERN_INFO, scmd->device, 1925 "reservation conflict\n"); 1926 set_host_byte(scmd, DID_NEXUS_FAILURE); 1927 return SUCCESS; /* causes immediate i/o error */ 1928 default: 1929 return FAILED; 1930 } 1931 return FAILED; 1932 1933 maybe_retry: 1934 1935 /* we requeue for retry because the error was retryable, and 1936 * the request was not marked fast fail. Note that above, 1937 * even if the request is marked fast fail, we still requeue 1938 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1939 if ((++scmd->retries) <= scmd->allowed 1940 && !scsi_noretry_cmd(scmd)) { 1941 return NEEDS_RETRY; 1942 } else { 1943 /* 1944 * no more retries - report this one back to upper level. 1945 */ 1946 return SUCCESS; 1947 } 1948 } 1949 1950 static void eh_lock_door_done(struct request *req, int uptodate) 1951 { 1952 __blk_put_request(req->q, req); 1953 } 1954 1955 /** 1956 * scsi_eh_lock_door - Prevent medium removal for the specified device 1957 * @sdev: SCSI device to prevent medium removal 1958 * 1959 * Locking: 1960 * We must be called from process context. 1961 * 1962 * Notes: 1963 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1964 * head of the devices request queue, and continue. 1965 */ 1966 static void scsi_eh_lock_door(struct scsi_device *sdev) 1967 { 1968 struct request *req; 1969 struct scsi_request *rq; 1970 1971 /* 1972 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a 1973 * request becomes available 1974 */ 1975 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL); 1976 if (IS_ERR(req)) 1977 return; 1978 rq = scsi_req(req); 1979 scsi_req_init(req); 1980 1981 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 1982 rq->cmd[1] = 0; 1983 rq->cmd[2] = 0; 1984 rq->cmd[3] = 0; 1985 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 1986 rq->cmd[5] = 0; 1987 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 1988 1989 req->rq_flags |= RQF_QUIET; 1990 req->timeout = 10 * HZ; 1991 req->retries = 5; 1992 1993 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); 1994 } 1995 1996 /** 1997 * scsi_restart_operations - restart io operations to the specified host. 1998 * @shost: Host we are restarting. 1999 * 2000 * Notes: 2001 * When we entered the error handler, we blocked all further i/o to 2002 * this device. we need to 'reverse' this process. 2003 */ 2004 static void scsi_restart_operations(struct Scsi_Host *shost) 2005 { 2006 struct scsi_device *sdev; 2007 unsigned long flags; 2008 2009 /* 2010 * If the door was locked, we need to insert a door lock request 2011 * onto the head of the SCSI request queue for the device. There 2012 * is no point trying to lock the door of an off-line device. 2013 */ 2014 shost_for_each_device(sdev, shost) { 2015 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2016 scsi_eh_lock_door(sdev); 2017 sdev->was_reset = 0; 2018 } 2019 } 2020 2021 /* 2022 * next free up anything directly waiting upon the host. this 2023 * will be requests for character device operations, and also for 2024 * ioctls to queued block devices. 2025 */ 2026 SCSI_LOG_ERROR_RECOVERY(3, 2027 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2028 2029 spin_lock_irqsave(shost->host_lock, flags); 2030 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2031 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2032 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2033 spin_unlock_irqrestore(shost->host_lock, flags); 2034 2035 wake_up(&shost->host_wait); 2036 2037 /* 2038 * finally we need to re-initiate requests that may be pending. we will 2039 * have had everything blocked while error handling is taking place, and 2040 * now that error recovery is done, we will need to ensure that these 2041 * requests are started. 2042 */ 2043 scsi_run_host_queues(shost); 2044 2045 /* 2046 * if eh is active and host_eh_scheduled is pending we need to re-run 2047 * recovery. we do this check after scsi_run_host_queues() to allow 2048 * everything pent up since the last eh run a chance to make forward 2049 * progress before we sync again. Either we'll immediately re-run 2050 * recovery or scsi_device_unbusy() will wake us again when these 2051 * pending commands complete. 2052 */ 2053 spin_lock_irqsave(shost->host_lock, flags); 2054 if (shost->host_eh_scheduled) 2055 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2056 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2057 spin_unlock_irqrestore(shost->host_lock, flags); 2058 } 2059 2060 /** 2061 * scsi_eh_ready_devs - check device ready state and recover if not. 2062 * @shost: host to be recovered. 2063 * @work_q: &list_head for pending commands. 2064 * @done_q: &list_head for processed commands. 2065 */ 2066 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2067 struct list_head *work_q, 2068 struct list_head *done_q) 2069 { 2070 if (!scsi_eh_stu(shost, work_q, done_q)) 2071 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2072 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2073 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2074 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2075 scsi_eh_offline_sdevs(work_q, 2076 done_q); 2077 } 2078 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2079 2080 /** 2081 * scsi_eh_flush_done_q - finish processed commands or retry them. 2082 * @done_q: list_head of processed commands. 2083 */ 2084 void scsi_eh_flush_done_q(struct list_head *done_q) 2085 { 2086 struct scsi_cmnd *scmd, *next; 2087 2088 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2089 list_del_init(&scmd->eh_entry); 2090 if (scsi_device_online(scmd->device) && 2091 !scsi_noretry_cmd(scmd) && 2092 (++scmd->retries <= scmd->allowed)) { 2093 SCSI_LOG_ERROR_RECOVERY(3, 2094 scmd_printk(KERN_INFO, scmd, 2095 "%s: flush retry cmd\n", 2096 current->comm)); 2097 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2098 } else { 2099 /* 2100 * If just we got sense for the device (called 2101 * scsi_eh_get_sense), scmd->result is already 2102 * set, do not set DRIVER_TIMEOUT. 2103 */ 2104 if (!scmd->result) 2105 scmd->result |= (DRIVER_TIMEOUT << 24); 2106 SCSI_LOG_ERROR_RECOVERY(3, 2107 scmd_printk(KERN_INFO, scmd, 2108 "%s: flush finish cmd\n", 2109 current->comm)); 2110 scsi_finish_command(scmd); 2111 } 2112 } 2113 } 2114 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2115 2116 /** 2117 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2118 * @shost: Host to unjam. 2119 * 2120 * Notes: 2121 * When we come in here, we *know* that all commands on the bus have 2122 * either completed, failed or timed out. we also know that no further 2123 * commands are being sent to the host, so things are relatively quiet 2124 * and we have freedom to fiddle with things as we wish. 2125 * 2126 * This is only the *default* implementation. it is possible for 2127 * individual drivers to supply their own version of this function, and 2128 * if the maintainer wishes to do this, it is strongly suggested that 2129 * this function be taken as a template and modified. this function 2130 * was designed to correctly handle problems for about 95% of the 2131 * different cases out there, and it should always provide at least a 2132 * reasonable amount of error recovery. 2133 * 2134 * Any command marked 'failed' or 'timeout' must eventually have 2135 * scsi_finish_cmd() called for it. we do all of the retry stuff 2136 * here, so when we restart the host after we return it should have an 2137 * empty queue. 2138 */ 2139 static void scsi_unjam_host(struct Scsi_Host *shost) 2140 { 2141 unsigned long flags; 2142 LIST_HEAD(eh_work_q); 2143 LIST_HEAD(eh_done_q); 2144 2145 spin_lock_irqsave(shost->host_lock, flags); 2146 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2147 spin_unlock_irqrestore(shost->host_lock, flags); 2148 2149 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2150 2151 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2152 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 2153 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2154 2155 spin_lock_irqsave(shost->host_lock, flags); 2156 if (shost->eh_deadline != -1) 2157 shost->last_reset = 0; 2158 spin_unlock_irqrestore(shost->host_lock, flags); 2159 scsi_eh_flush_done_q(&eh_done_q); 2160 } 2161 2162 /** 2163 * scsi_error_handler - SCSI error handler thread 2164 * @data: Host for which we are running. 2165 * 2166 * Notes: 2167 * This is the main error handling loop. This is run as a kernel thread 2168 * for every SCSI host and handles all error handling activity. 2169 */ 2170 int scsi_error_handler(void *data) 2171 { 2172 struct Scsi_Host *shost = data; 2173 2174 /* 2175 * We use TASK_INTERRUPTIBLE so that the thread is not 2176 * counted against the load average as a running process. 2177 * We never actually get interrupted because kthread_run 2178 * disables signal delivery for the created thread. 2179 */ 2180 while (true) { 2181 /* 2182 * The sequence in kthread_stop() sets the stop flag first 2183 * then wakes the process. To avoid missed wakeups, the task 2184 * should always be in a non running state before the stop 2185 * flag is checked 2186 */ 2187 set_current_state(TASK_INTERRUPTIBLE); 2188 if (kthread_should_stop()) 2189 break; 2190 2191 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2192 shost->host_failed != atomic_read(&shost->host_busy)) { 2193 SCSI_LOG_ERROR_RECOVERY(1, 2194 shost_printk(KERN_INFO, shost, 2195 "scsi_eh_%d: sleeping\n", 2196 shost->host_no)); 2197 schedule(); 2198 continue; 2199 } 2200 2201 __set_current_state(TASK_RUNNING); 2202 SCSI_LOG_ERROR_RECOVERY(1, 2203 shost_printk(KERN_INFO, shost, 2204 "scsi_eh_%d: waking up %d/%d/%d\n", 2205 shost->host_no, shost->host_eh_scheduled, 2206 shost->host_failed, 2207 atomic_read(&shost->host_busy))); 2208 2209 /* 2210 * We have a host that is failing for some reason. Figure out 2211 * what we need to do to get it up and online again (if we can). 2212 * If we fail, we end up taking the thing offline. 2213 */ 2214 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2215 SCSI_LOG_ERROR_RECOVERY(1, 2216 shost_printk(KERN_ERR, shost, 2217 "scsi_eh_%d: unable to autoresume\n", 2218 shost->host_no)); 2219 continue; 2220 } 2221 2222 if (shost->transportt->eh_strategy_handler) 2223 shost->transportt->eh_strategy_handler(shost); 2224 else 2225 scsi_unjam_host(shost); 2226 2227 /* All scmds have been handled */ 2228 shost->host_failed = 0; 2229 2230 /* 2231 * Note - if the above fails completely, the action is to take 2232 * individual devices offline and flush the queue of any 2233 * outstanding requests that may have been pending. When we 2234 * restart, we restart any I/O to any other devices on the bus 2235 * which are still online. 2236 */ 2237 scsi_restart_operations(shost); 2238 if (!shost->eh_noresume) 2239 scsi_autopm_put_host(shost); 2240 } 2241 __set_current_state(TASK_RUNNING); 2242 2243 SCSI_LOG_ERROR_RECOVERY(1, 2244 shost_printk(KERN_INFO, shost, 2245 "Error handler scsi_eh_%d exiting\n", 2246 shost->host_no)); 2247 shost->ehandler = NULL; 2248 return 0; 2249 } 2250 2251 /* 2252 * Function: scsi_report_bus_reset() 2253 * 2254 * Purpose: Utility function used by low-level drivers to report that 2255 * they have observed a bus reset on the bus being handled. 2256 * 2257 * Arguments: shost - Host in question 2258 * channel - channel on which reset was observed. 2259 * 2260 * Returns: Nothing 2261 * 2262 * Lock status: Host lock must be held. 2263 * 2264 * Notes: This only needs to be called if the reset is one which 2265 * originates from an unknown location. Resets originated 2266 * by the mid-level itself don't need to call this, but there 2267 * should be no harm. 2268 * 2269 * The main purpose of this is to make sure that a CHECK_CONDITION 2270 * is properly treated. 2271 */ 2272 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2273 { 2274 struct scsi_device *sdev; 2275 2276 __shost_for_each_device(sdev, shost) { 2277 if (channel == sdev_channel(sdev)) 2278 __scsi_report_device_reset(sdev, NULL); 2279 } 2280 } 2281 EXPORT_SYMBOL(scsi_report_bus_reset); 2282 2283 /* 2284 * Function: scsi_report_device_reset() 2285 * 2286 * Purpose: Utility function used by low-level drivers to report that 2287 * they have observed a device reset on the device being handled. 2288 * 2289 * Arguments: shost - Host in question 2290 * channel - channel on which reset was observed 2291 * target - target on which reset was observed 2292 * 2293 * Returns: Nothing 2294 * 2295 * Lock status: Host lock must be held 2296 * 2297 * Notes: This only needs to be called if the reset is one which 2298 * originates from an unknown location. Resets originated 2299 * by the mid-level itself don't need to call this, but there 2300 * should be no harm. 2301 * 2302 * The main purpose of this is to make sure that a CHECK_CONDITION 2303 * is properly treated. 2304 */ 2305 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2306 { 2307 struct scsi_device *sdev; 2308 2309 __shost_for_each_device(sdev, shost) { 2310 if (channel == sdev_channel(sdev) && 2311 target == sdev_id(sdev)) 2312 __scsi_report_device_reset(sdev, NULL); 2313 } 2314 } 2315 EXPORT_SYMBOL(scsi_report_device_reset); 2316 2317 static void 2318 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2319 { 2320 } 2321 2322 /** 2323 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2324 * @dev: scsi_device to operate on 2325 * @arg: reset type (see sg.h) 2326 */ 2327 int 2328 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2329 { 2330 struct scsi_cmnd *scmd; 2331 struct Scsi_Host *shost = dev->host; 2332 struct request *rq; 2333 unsigned long flags; 2334 int error = 0, rtn, val; 2335 2336 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2337 return -EACCES; 2338 2339 error = get_user(val, arg); 2340 if (error) 2341 return error; 2342 2343 if (scsi_autopm_get_host(shost) < 0) 2344 return -EIO; 2345 2346 error = -EIO; 2347 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2348 shost->hostt->cmd_size, GFP_KERNEL); 2349 if (!rq) 2350 goto out_put_autopm_host; 2351 blk_rq_init(NULL, rq); 2352 2353 scmd = (struct scsi_cmnd *)(rq + 1); 2354 scsi_init_command(dev, scmd); 2355 scmd->request = rq; 2356 scmd->cmnd = scsi_req(rq)->cmd; 2357 2358 scmd->scsi_done = scsi_reset_provider_done_command; 2359 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2360 2361 scmd->cmd_len = 0; 2362 2363 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2364 2365 spin_lock_irqsave(shost->host_lock, flags); 2366 shost->tmf_in_progress = 1; 2367 spin_unlock_irqrestore(shost->host_lock, flags); 2368 2369 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2370 case SG_SCSI_RESET_NOTHING: 2371 rtn = SUCCESS; 2372 break; 2373 case SG_SCSI_RESET_DEVICE: 2374 rtn = scsi_try_bus_device_reset(scmd); 2375 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2376 break; 2377 /* FALLTHROUGH */ 2378 case SG_SCSI_RESET_TARGET: 2379 rtn = scsi_try_target_reset(scmd); 2380 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2381 break; 2382 /* FALLTHROUGH */ 2383 case SG_SCSI_RESET_BUS: 2384 rtn = scsi_try_bus_reset(scmd); 2385 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2386 break; 2387 /* FALLTHROUGH */ 2388 case SG_SCSI_RESET_HOST: 2389 rtn = scsi_try_host_reset(scmd); 2390 if (rtn == SUCCESS) 2391 break; 2392 default: 2393 /* FALLTHROUGH */ 2394 rtn = FAILED; 2395 break; 2396 } 2397 2398 error = (rtn == SUCCESS) ? 0 : -EIO; 2399 2400 spin_lock_irqsave(shost->host_lock, flags); 2401 shost->tmf_in_progress = 0; 2402 spin_unlock_irqrestore(shost->host_lock, flags); 2403 2404 /* 2405 * be sure to wake up anyone who was sleeping or had their queue 2406 * suspended while we performed the TMF. 2407 */ 2408 SCSI_LOG_ERROR_RECOVERY(3, 2409 shost_printk(KERN_INFO, shost, 2410 "waking up host to restart after TMF\n")); 2411 2412 wake_up(&shost->host_wait); 2413 scsi_run_host_queues(shost); 2414 2415 scsi_put_command(scmd); 2416 kfree(rq); 2417 2418 out_put_autopm_host: 2419 scsi_autopm_put_host(shost); 2420 return error; 2421 } 2422 EXPORT_SYMBOL(scsi_ioctl_reset); 2423 2424 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2425 struct scsi_sense_hdr *sshdr) 2426 { 2427 return scsi_normalize_sense(cmd->sense_buffer, 2428 SCSI_SENSE_BUFFERSIZE, sshdr); 2429 } 2430 EXPORT_SYMBOL(scsi_command_normalize_sense); 2431 2432 /** 2433 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2434 * @sense_buffer: byte array of sense data 2435 * @sb_len: number of valid bytes in sense_buffer 2436 * @info_out: pointer to 64 integer where 8 or 4 byte information 2437 * field will be placed if found. 2438 * 2439 * Return value: 2440 * 1 if information field found, 0 if not found. 2441 */ 2442 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 2443 u64 * info_out) 2444 { 2445 int j; 2446 const u8 * ucp; 2447 u64 ull; 2448 2449 if (sb_len < 7) 2450 return 0; 2451 switch (sense_buffer[0] & 0x7f) { 2452 case 0x70: 2453 case 0x71: 2454 if (sense_buffer[0] & 0x80) { 2455 *info_out = (sense_buffer[3] << 24) + 2456 (sense_buffer[4] << 16) + 2457 (sense_buffer[5] << 8) + sense_buffer[6]; 2458 return 1; 2459 } else 2460 return 0; 2461 case 0x72: 2462 case 0x73: 2463 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2464 0 /* info desc */); 2465 if (ucp && (0xa == ucp[1])) { 2466 ull = 0; 2467 for (j = 0; j < 8; ++j) { 2468 if (j > 0) 2469 ull <<= 8; 2470 ull |= ucp[4 + j]; 2471 } 2472 *info_out = ull; 2473 return 1; 2474 } else 2475 return 0; 2476 default: 2477 return 0; 2478 } 2479 } 2480 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2481