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->transportt->eh_timed_out) 283 rtn = host->transportt->eh_timed_out(scmd); 284 else if (host->hostt->eh_timed_out) 285 rtn = host->hostt->eh_timed_out(scmd); 286 287 if (rtn == BLK_EH_NOT_HANDLED) { 288 if (!host->hostt->no_async_abort && 289 scsi_abort_command(scmd) == SUCCESS) 290 return BLK_EH_NOT_HANDLED; 291 292 set_host_byte(scmd, DID_TIME_OUT); 293 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)) 294 rtn = BLK_EH_HANDLED; 295 } 296 297 return rtn; 298 } 299 300 /** 301 * scsi_block_when_processing_errors - Prevent cmds from being queued. 302 * @sdev: Device on which we are performing recovery. 303 * 304 * Description: 305 * We block until the host is out of error recovery, and then check to 306 * see whether the host or the device is offline. 307 * 308 * Return value: 309 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 310 */ 311 int scsi_block_when_processing_errors(struct scsi_device *sdev) 312 { 313 int online; 314 315 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 316 317 online = scsi_device_online(sdev); 318 319 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev, 320 "%s: rtn: %d\n", __func__, online)); 321 322 return online; 323 } 324 EXPORT_SYMBOL(scsi_block_when_processing_errors); 325 326 #ifdef CONFIG_SCSI_LOGGING 327 /** 328 * scsi_eh_prt_fail_stats - Log info on failures. 329 * @shost: scsi host being recovered. 330 * @work_q: Queue of scsi cmds to process. 331 */ 332 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 333 struct list_head *work_q) 334 { 335 struct scsi_cmnd *scmd; 336 struct scsi_device *sdev; 337 int total_failures = 0; 338 int cmd_failed = 0; 339 int cmd_cancel = 0; 340 int devices_failed = 0; 341 342 shost_for_each_device(sdev, shost) { 343 list_for_each_entry(scmd, work_q, eh_entry) { 344 if (scmd->device == sdev) { 345 ++total_failures; 346 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 347 ++cmd_cancel; 348 else 349 ++cmd_failed; 350 } 351 } 352 353 if (cmd_cancel || cmd_failed) { 354 SCSI_LOG_ERROR_RECOVERY(3, 355 shost_printk(KERN_INFO, shost, 356 "%s: cmds failed: %d, cancel: %d\n", 357 __func__, cmd_failed, 358 cmd_cancel)); 359 cmd_cancel = 0; 360 cmd_failed = 0; 361 ++devices_failed; 362 } 363 } 364 365 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 366 "Total of %d commands on %d" 367 " devices require eh work\n", 368 total_failures, devices_failed)); 369 } 370 #endif 371 372 /** 373 * scsi_report_lun_change - Set flag on all *other* devices on the same target 374 * to indicate that a UNIT ATTENTION is expected. 375 * @sdev: Device reporting the UNIT ATTENTION 376 */ 377 static void scsi_report_lun_change(struct scsi_device *sdev) 378 { 379 sdev->sdev_target->expecting_lun_change = 1; 380 } 381 382 /** 383 * scsi_report_sense - Examine scsi sense information and log messages for 384 * certain conditions, also issue uevents for some of them. 385 * @sdev: Device reporting the sense code 386 * @sshdr: sshdr to be examined 387 */ 388 static void scsi_report_sense(struct scsi_device *sdev, 389 struct scsi_sense_hdr *sshdr) 390 { 391 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 392 393 if (sshdr->sense_key == UNIT_ATTENTION) { 394 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 395 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 396 sdev_printk(KERN_WARNING, sdev, 397 "Inquiry data has changed"); 398 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 399 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 400 scsi_report_lun_change(sdev); 401 sdev_printk(KERN_WARNING, sdev, 402 "Warning! Received an indication that the " 403 "LUN assignments on this target have " 404 "changed. The Linux SCSI layer does not " 405 "automatically remap LUN assignments.\n"); 406 } else if (sshdr->asc == 0x3f) 407 sdev_printk(KERN_WARNING, sdev, 408 "Warning! Received an indication that the " 409 "operating parameters on this target have " 410 "changed. The Linux SCSI layer does not " 411 "automatically adjust these parameters.\n"); 412 413 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 414 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 415 sdev_printk(KERN_WARNING, sdev, 416 "Warning! Received an indication that the " 417 "LUN reached a thin provisioning soft " 418 "threshold.\n"); 419 } 420 421 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 422 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 423 sdev_printk(KERN_WARNING, sdev, 424 "Mode parameters changed"); 425 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 426 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 427 sdev_printk(KERN_WARNING, sdev, 428 "Asymmetric access state changed"); 429 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 430 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 431 sdev_printk(KERN_WARNING, sdev, 432 "Capacity data has changed"); 433 } else if (sshdr->asc == 0x2a) 434 sdev_printk(KERN_WARNING, sdev, 435 "Parameters changed"); 436 } 437 438 if (evt_type != SDEV_EVT_MAXBITS) { 439 set_bit(evt_type, sdev->pending_events); 440 schedule_work(&sdev->event_work); 441 } 442 } 443 444 /** 445 * scsi_check_sense - Examine scsi cmd sense 446 * @scmd: Cmd to have sense checked. 447 * 448 * Return value: 449 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 450 * 451 * Notes: 452 * When a deferred error is detected the current command has 453 * not been executed and needs retrying. 454 */ 455 int scsi_check_sense(struct scsi_cmnd *scmd) 456 { 457 struct scsi_device *sdev = scmd->device; 458 struct scsi_sense_hdr sshdr; 459 460 if (! scsi_command_normalize_sense(scmd, &sshdr)) 461 return FAILED; /* no valid sense data */ 462 463 scsi_report_sense(sdev, &sshdr); 464 465 if (scsi_sense_is_deferred(&sshdr)) 466 return NEEDS_RETRY; 467 468 if (sdev->handler && sdev->handler->check_sense) { 469 int rc; 470 471 rc = sdev->handler->check_sense(sdev, &sshdr); 472 if (rc != SCSI_RETURN_NOT_HANDLED) 473 return rc; 474 /* handler does not care. Drop down to default handling */ 475 } 476 477 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 478 /* 479 * nasty: for mid-layer issued TURs, we need to return the 480 * actual sense data without any recovery attempt. For eh 481 * issued ones, we need to try to recover and interpret 482 */ 483 return SUCCESS; 484 485 /* 486 * Previous logic looked for FILEMARK, EOM or ILI which are 487 * mainly associated with tapes and returned SUCCESS. 488 */ 489 if (sshdr.response_code == 0x70) { 490 /* fixed format */ 491 if (scmd->sense_buffer[2] & 0xe0) 492 return SUCCESS; 493 } else { 494 /* 495 * descriptor format: look for "stream commands sense data 496 * descriptor" (see SSC-3). Assume single sense data 497 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 498 */ 499 if ((sshdr.additional_length > 3) && 500 (scmd->sense_buffer[8] == 0x4) && 501 (scmd->sense_buffer[11] & 0xe0)) 502 return SUCCESS; 503 } 504 505 switch (sshdr.sense_key) { 506 case NO_SENSE: 507 return SUCCESS; 508 case RECOVERED_ERROR: 509 return /* soft_error */ SUCCESS; 510 511 case ABORTED_COMMAND: 512 if (sshdr.asc == 0x10) /* DIF */ 513 return SUCCESS; 514 515 return NEEDS_RETRY; 516 case NOT_READY: 517 case UNIT_ATTENTION: 518 /* 519 * if we are expecting a cc/ua because of a bus reset that we 520 * performed, treat this just as a retry. otherwise this is 521 * information that we should pass up to the upper-level driver 522 * so that we can deal with it there. 523 */ 524 if (scmd->device->expecting_cc_ua) { 525 /* 526 * Because some device does not queue unit 527 * attentions correctly, we carefully check 528 * additional sense code and qualifier so as 529 * not to squash media change unit attention. 530 */ 531 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 532 scmd->device->expecting_cc_ua = 0; 533 return NEEDS_RETRY; 534 } 535 } 536 /* 537 * we might also expect a cc/ua if another LUN on the target 538 * reported a UA with an ASC/ASCQ of 3F 0E - 539 * REPORTED LUNS DATA HAS CHANGED. 540 */ 541 if (scmd->device->sdev_target->expecting_lun_change && 542 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 543 return NEEDS_RETRY; 544 /* 545 * if the device is in the process of becoming ready, we 546 * should retry. 547 */ 548 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 549 return NEEDS_RETRY; 550 /* 551 * if the device is not started, we need to wake 552 * the error handler to start the motor 553 */ 554 if (scmd->device->allow_restart && 555 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 556 return FAILED; 557 /* 558 * Pass the UA upwards for a determination in the completion 559 * functions. 560 */ 561 return SUCCESS; 562 563 /* these are not supported */ 564 case DATA_PROTECT: 565 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 566 /* Thin provisioning hard threshold reached */ 567 set_host_byte(scmd, DID_ALLOC_FAILURE); 568 return SUCCESS; 569 } 570 case COPY_ABORTED: 571 case VOLUME_OVERFLOW: 572 case MISCOMPARE: 573 case BLANK_CHECK: 574 set_host_byte(scmd, DID_TARGET_FAILURE); 575 return SUCCESS; 576 577 case MEDIUM_ERROR: 578 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 579 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 580 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 581 set_host_byte(scmd, DID_MEDIUM_ERROR); 582 return SUCCESS; 583 } 584 return NEEDS_RETRY; 585 586 case HARDWARE_ERROR: 587 if (scmd->device->retry_hwerror) 588 return ADD_TO_MLQUEUE; 589 else 590 set_host_byte(scmd, DID_TARGET_FAILURE); 591 592 case ILLEGAL_REQUEST: 593 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 594 sshdr.asc == 0x21 || /* Logical block address out of range */ 595 sshdr.asc == 0x24 || /* Invalid field in cdb */ 596 sshdr.asc == 0x26) { /* Parameter value invalid */ 597 set_host_byte(scmd, DID_TARGET_FAILURE); 598 } 599 return SUCCESS; 600 601 default: 602 return SUCCESS; 603 } 604 } 605 EXPORT_SYMBOL_GPL(scsi_check_sense); 606 607 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 608 { 609 struct scsi_host_template *sht = sdev->host->hostt; 610 struct scsi_device *tmp_sdev; 611 612 if (!sht->track_queue_depth || 613 sdev->queue_depth >= sdev->max_queue_depth) 614 return; 615 616 if (time_before(jiffies, 617 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 618 return; 619 620 if (time_before(jiffies, 621 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 622 return; 623 624 /* 625 * Walk all devices of a target and do 626 * ramp up on them. 627 */ 628 shost_for_each_device(tmp_sdev, sdev->host) { 629 if (tmp_sdev->channel != sdev->channel || 630 tmp_sdev->id != sdev->id || 631 tmp_sdev->queue_depth == sdev->max_queue_depth) 632 continue; 633 634 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 635 sdev->last_queue_ramp_up = jiffies; 636 } 637 } 638 639 static void scsi_handle_queue_full(struct scsi_device *sdev) 640 { 641 struct scsi_host_template *sht = sdev->host->hostt; 642 struct scsi_device *tmp_sdev; 643 644 if (!sht->track_queue_depth) 645 return; 646 647 shost_for_each_device(tmp_sdev, sdev->host) { 648 if (tmp_sdev->channel != sdev->channel || 649 tmp_sdev->id != sdev->id) 650 continue; 651 /* 652 * We do not know the number of commands that were at 653 * the device when we got the queue full so we start 654 * from the highest possible value and work our way down. 655 */ 656 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 657 } 658 } 659 660 /** 661 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 662 * @scmd: SCSI cmd to examine. 663 * 664 * Notes: 665 * This is *only* called when we are examining the status of commands 666 * queued during error recovery. the main difference here is that we 667 * don't allow for the possibility of retries here, and we are a lot 668 * more restrictive about what we consider acceptable. 669 */ 670 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 671 { 672 /* 673 * first check the host byte, to see if there is anything in there 674 * that would indicate what we need to do. 675 */ 676 if (host_byte(scmd->result) == DID_RESET) { 677 /* 678 * rats. we are already in the error handler, so we now 679 * get to try and figure out what to do next. if the sense 680 * is valid, we have a pretty good idea of what to do. 681 * if not, we mark it as FAILED. 682 */ 683 return scsi_check_sense(scmd); 684 } 685 if (host_byte(scmd->result) != DID_OK) 686 return FAILED; 687 688 /* 689 * next, check the message byte. 690 */ 691 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 692 return FAILED; 693 694 /* 695 * now, check the status byte to see if this indicates 696 * anything special. 697 */ 698 switch (status_byte(scmd->result)) { 699 case GOOD: 700 scsi_handle_queue_ramp_up(scmd->device); 701 case COMMAND_TERMINATED: 702 return SUCCESS; 703 case CHECK_CONDITION: 704 return scsi_check_sense(scmd); 705 case CONDITION_GOOD: 706 case INTERMEDIATE_GOOD: 707 case INTERMEDIATE_C_GOOD: 708 /* 709 * who knows? FIXME(eric) 710 */ 711 return SUCCESS; 712 case RESERVATION_CONFLICT: 713 if (scmd->cmnd[0] == TEST_UNIT_READY) 714 /* it is a success, we probed the device and 715 * found it */ 716 return SUCCESS; 717 /* otherwise, we failed to send the command */ 718 return FAILED; 719 case QUEUE_FULL: 720 scsi_handle_queue_full(scmd->device); 721 /* fall through */ 722 case BUSY: 723 return NEEDS_RETRY; 724 default: 725 return FAILED; 726 } 727 return FAILED; 728 } 729 730 /** 731 * scsi_eh_done - Completion function for error handling. 732 * @scmd: Cmd that is done. 733 */ 734 static void scsi_eh_done(struct scsi_cmnd *scmd) 735 { 736 struct completion *eh_action; 737 738 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 739 "%s result: %x\n", __func__, scmd->result)); 740 741 eh_action = scmd->device->host->eh_action; 742 if (eh_action) 743 complete(eh_action); 744 } 745 746 /** 747 * scsi_try_host_reset - ask host adapter to reset itself 748 * @scmd: SCSI cmd to send host reset. 749 */ 750 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 751 { 752 unsigned long flags; 753 int rtn; 754 struct Scsi_Host *host = scmd->device->host; 755 struct scsi_host_template *hostt = host->hostt; 756 757 SCSI_LOG_ERROR_RECOVERY(3, 758 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 759 760 if (!hostt->eh_host_reset_handler) 761 return FAILED; 762 763 rtn = hostt->eh_host_reset_handler(scmd); 764 765 if (rtn == SUCCESS) { 766 if (!hostt->skip_settle_delay) 767 ssleep(HOST_RESET_SETTLE_TIME); 768 spin_lock_irqsave(host->host_lock, flags); 769 scsi_report_bus_reset(host, scmd_channel(scmd)); 770 spin_unlock_irqrestore(host->host_lock, flags); 771 } 772 773 return rtn; 774 } 775 776 /** 777 * scsi_try_bus_reset - ask host to perform a bus reset 778 * @scmd: SCSI cmd to send bus reset. 779 */ 780 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 781 { 782 unsigned long flags; 783 int rtn; 784 struct Scsi_Host *host = scmd->device->host; 785 struct scsi_host_template *hostt = host->hostt; 786 787 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 788 "%s: Snd Bus RST\n", __func__)); 789 790 if (!hostt->eh_bus_reset_handler) 791 return FAILED; 792 793 rtn = hostt->eh_bus_reset_handler(scmd); 794 795 if (rtn == SUCCESS) { 796 if (!hostt->skip_settle_delay) 797 ssleep(BUS_RESET_SETTLE_TIME); 798 spin_lock_irqsave(host->host_lock, flags); 799 scsi_report_bus_reset(host, scmd_channel(scmd)); 800 spin_unlock_irqrestore(host->host_lock, flags); 801 } 802 803 return rtn; 804 } 805 806 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 807 { 808 sdev->was_reset = 1; 809 sdev->expecting_cc_ua = 1; 810 } 811 812 /** 813 * scsi_try_target_reset - Ask host to perform a target reset 814 * @scmd: SCSI cmd used to send a target reset 815 * 816 * Notes: 817 * There is no timeout for this operation. if this operation is 818 * unreliable for a given host, then the host itself needs to put a 819 * timer on it, and set the host back to a consistent state prior to 820 * returning. 821 */ 822 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 823 { 824 unsigned long flags; 825 int rtn; 826 struct Scsi_Host *host = scmd->device->host; 827 struct scsi_host_template *hostt = host->hostt; 828 829 if (!hostt->eh_target_reset_handler) 830 return FAILED; 831 832 rtn = hostt->eh_target_reset_handler(scmd); 833 if (rtn == SUCCESS) { 834 spin_lock_irqsave(host->host_lock, flags); 835 __starget_for_each_device(scsi_target(scmd->device), NULL, 836 __scsi_report_device_reset); 837 spin_unlock_irqrestore(host->host_lock, flags); 838 } 839 840 return rtn; 841 } 842 843 /** 844 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 845 * @scmd: SCSI cmd used to send BDR 846 * 847 * Notes: 848 * There is no timeout for this operation. if this operation is 849 * unreliable for a given host, then the host itself needs to put a 850 * timer on it, and set the host back to a consistent state prior to 851 * returning. 852 */ 853 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 854 { 855 int rtn; 856 struct scsi_host_template *hostt = scmd->device->host->hostt; 857 858 if (!hostt->eh_device_reset_handler) 859 return FAILED; 860 861 rtn = hostt->eh_device_reset_handler(scmd); 862 if (rtn == SUCCESS) 863 __scsi_report_device_reset(scmd->device, NULL); 864 return rtn; 865 } 866 867 /** 868 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 869 * @hostt: SCSI driver host template 870 * @scmd: SCSI cmd used to send a target reset 871 * 872 * Return value: 873 * SUCCESS, FAILED, or FAST_IO_FAIL 874 * 875 * Notes: 876 * SUCCESS does not necessarily indicate that the command 877 * has been aborted; it only indicates that the LLDDs 878 * has cleared all references to that command. 879 * LLDDs should return FAILED only if an abort was required 880 * but could not be executed. LLDDs should return FAST_IO_FAIL 881 * if the device is temporarily unavailable (eg due to a 882 * link down on FibreChannel) 883 */ 884 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, 885 struct scsi_cmnd *scmd) 886 { 887 if (!hostt->eh_abort_handler) 888 return FAILED; 889 890 return hostt->eh_abort_handler(scmd); 891 } 892 893 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 894 { 895 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 896 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 897 if (scsi_try_target_reset(scmd) != SUCCESS) 898 if (scsi_try_bus_reset(scmd) != SUCCESS) 899 scsi_try_host_reset(scmd); 900 } 901 902 /** 903 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 904 * @scmd: SCSI command structure to hijack 905 * @ses: structure to save restore information 906 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 907 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 908 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 909 * 910 * This function is used to save a scsi command information before re-execution 911 * as part of the error recovery process. If @sense_bytes is 0 the command 912 * sent must be one that does not transfer any data. If @sense_bytes != 0 913 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 914 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 915 */ 916 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 917 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 918 { 919 struct scsi_device *sdev = scmd->device; 920 921 /* 922 * We need saved copies of a number of fields - this is because 923 * error handling may need to overwrite these with different values 924 * to run different commands, and once error handling is complete, 925 * we will need to restore these values prior to running the actual 926 * command. 927 */ 928 ses->cmd_len = scmd->cmd_len; 929 ses->cmnd = scmd->cmnd; 930 ses->data_direction = scmd->sc_data_direction; 931 ses->sdb = scmd->sdb; 932 ses->next_rq = scmd->request->next_rq; 933 ses->result = scmd->result; 934 ses->underflow = scmd->underflow; 935 ses->prot_op = scmd->prot_op; 936 937 scmd->prot_op = SCSI_PROT_NORMAL; 938 scmd->eh_eflags = 0; 939 scmd->cmnd = ses->eh_cmnd; 940 memset(scmd->cmnd, 0, BLK_MAX_CDB); 941 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 942 scmd->request->next_rq = NULL; 943 scmd->result = 0; 944 945 if (sense_bytes) { 946 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 947 sense_bytes); 948 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 949 scmd->sdb.length); 950 scmd->sdb.table.sgl = &ses->sense_sgl; 951 scmd->sc_data_direction = DMA_FROM_DEVICE; 952 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 953 scmd->cmnd[0] = REQUEST_SENSE; 954 scmd->cmnd[4] = scmd->sdb.length; 955 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 956 } else { 957 scmd->sc_data_direction = DMA_NONE; 958 if (cmnd) { 959 BUG_ON(cmnd_size > BLK_MAX_CDB); 960 memcpy(scmd->cmnd, cmnd, cmnd_size); 961 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 962 } 963 } 964 965 scmd->underflow = 0; 966 967 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 968 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 969 (sdev->lun << 5 & 0xe0); 970 971 /* 972 * Zero the sense buffer. The scsi spec mandates that any 973 * untransferred sense data should be interpreted as being zero. 974 */ 975 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 976 } 977 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 978 979 /** 980 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 981 * @scmd: SCSI command structure to restore 982 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 983 * 984 * Undo any damage done by above scsi_eh_prep_cmnd(). 985 */ 986 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 987 { 988 /* 989 * Restore original data 990 */ 991 scmd->cmd_len = ses->cmd_len; 992 scmd->cmnd = ses->cmnd; 993 scmd->sc_data_direction = ses->data_direction; 994 scmd->sdb = ses->sdb; 995 scmd->request->next_rq = ses->next_rq; 996 scmd->result = ses->result; 997 scmd->underflow = ses->underflow; 998 scmd->prot_op = ses->prot_op; 999 } 1000 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1001 1002 /** 1003 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1004 * @scmd: SCSI command structure to hijack 1005 * @cmnd: CDB to send 1006 * @cmnd_size: size in bytes of @cmnd 1007 * @timeout: timeout for this request 1008 * @sense_bytes: size of sense data to copy or 0 1009 * 1010 * This function is used to send a scsi command down to a target device 1011 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1012 * 1013 * Return value: 1014 * SUCCESS or FAILED or NEEDS_RETRY 1015 */ 1016 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 1017 int cmnd_size, int timeout, unsigned sense_bytes) 1018 { 1019 struct scsi_device *sdev = scmd->device; 1020 struct Scsi_Host *shost = sdev->host; 1021 DECLARE_COMPLETION_ONSTACK(done); 1022 unsigned long timeleft = timeout; 1023 struct scsi_eh_save ses; 1024 const unsigned long stall_for = msecs_to_jiffies(100); 1025 int rtn; 1026 1027 retry: 1028 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1029 shost->eh_action = &done; 1030 1031 scsi_log_send(scmd); 1032 scmd->scsi_done = scsi_eh_done; 1033 rtn = shost->hostt->queuecommand(shost, scmd); 1034 if (rtn) { 1035 if (timeleft > stall_for) { 1036 scsi_eh_restore_cmnd(scmd, &ses); 1037 timeleft -= stall_for; 1038 msleep(jiffies_to_msecs(stall_for)); 1039 goto retry; 1040 } 1041 /* signal not to enter either branch of the if () below */ 1042 timeleft = 0; 1043 rtn = FAILED; 1044 } else { 1045 timeleft = wait_for_completion_timeout(&done, timeout); 1046 rtn = SUCCESS; 1047 } 1048 1049 shost->eh_action = NULL; 1050 1051 scsi_log_completion(scmd, rtn); 1052 1053 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1054 "%s timeleft: %ld\n", 1055 __func__, timeleft)); 1056 1057 /* 1058 * If there is time left scsi_eh_done got called, and we will examine 1059 * the actual status codes to see whether the command actually did 1060 * complete normally, else if we have a zero return and no time left, 1061 * the command must still be pending, so abort it and return FAILED. 1062 * If we never actually managed to issue the command, because 1063 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1064 * value above (so don't execute either branch of the if) 1065 */ 1066 if (timeleft) { 1067 rtn = scsi_eh_completed_normally(scmd); 1068 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1069 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1070 1071 switch (rtn) { 1072 case SUCCESS: 1073 case NEEDS_RETRY: 1074 case FAILED: 1075 break; 1076 case ADD_TO_MLQUEUE: 1077 rtn = NEEDS_RETRY; 1078 break; 1079 default: 1080 rtn = FAILED; 1081 break; 1082 } 1083 } else if (rtn != FAILED) { 1084 scsi_abort_eh_cmnd(scmd); 1085 rtn = FAILED; 1086 } 1087 1088 scsi_eh_restore_cmnd(scmd, &ses); 1089 1090 return rtn; 1091 } 1092 1093 /** 1094 * scsi_request_sense - Request sense data from a particular target. 1095 * @scmd: SCSI cmd for request sense. 1096 * 1097 * Notes: 1098 * Some hosts automatically obtain this information, others require 1099 * that we obtain it on our own. This function will *not* return until 1100 * the command either times out, or it completes. 1101 */ 1102 static int scsi_request_sense(struct scsi_cmnd *scmd) 1103 { 1104 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1105 } 1106 1107 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) 1108 { 1109 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) { 1110 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1111 if (sdrv->eh_action) 1112 rtn = sdrv->eh_action(scmd, rtn); 1113 } 1114 return rtn; 1115 } 1116 1117 /** 1118 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1119 * @scmd: Original SCSI cmd that eh has finished. 1120 * @done_q: Queue for processed commands. 1121 * 1122 * Notes: 1123 * We don't want to use the normal command completion while we are are 1124 * still handling errors - it may cause other commands to be queued, 1125 * and that would disturb what we are doing. Thus we really want to 1126 * keep a list of pending commands for final completion, and once we 1127 * are ready to leave error handling we handle completion for real. 1128 */ 1129 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1130 { 1131 scmd->eh_eflags = 0; 1132 list_move_tail(&scmd->eh_entry, done_q); 1133 } 1134 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1135 1136 /** 1137 * scsi_eh_get_sense - Get device sense data. 1138 * @work_q: Queue of commands to process. 1139 * @done_q: Queue of processed commands. 1140 * 1141 * Description: 1142 * See if we need to request sense information. if so, then get it 1143 * now, so we have a better idea of what to do. 1144 * 1145 * Notes: 1146 * This has the unfortunate side effect that if a shost adapter does 1147 * not automatically request sense information, we end up shutting 1148 * it down before we request it. 1149 * 1150 * All drivers should request sense information internally these days, 1151 * so for now all I have to say is tough noogies if you end up in here. 1152 * 1153 * XXX: Long term this code should go away, but that needs an audit of 1154 * all LLDDs first. 1155 */ 1156 int scsi_eh_get_sense(struct list_head *work_q, 1157 struct list_head *done_q) 1158 { 1159 struct scsi_cmnd *scmd, *next; 1160 struct Scsi_Host *shost; 1161 int rtn; 1162 1163 /* 1164 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1165 * should not get sense. 1166 */ 1167 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1168 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 1169 (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1170 SCSI_SENSE_VALID(scmd)) 1171 continue; 1172 1173 shost = scmd->device->host; 1174 if (scsi_host_eh_past_deadline(shost)) { 1175 SCSI_LOG_ERROR_RECOVERY(3, 1176 scmd_printk(KERN_INFO, scmd, 1177 "%s: skip request sense, past eh deadline\n", 1178 current->comm)); 1179 break; 1180 } 1181 if (status_byte(scmd->result) != CHECK_CONDITION) 1182 /* 1183 * don't request sense if there's no check condition 1184 * status because the error we're processing isn't one 1185 * that has a sense code (and some devices get 1186 * confused by sense requests out of the blue) 1187 */ 1188 continue; 1189 1190 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1191 "%s: requesting sense\n", 1192 current->comm)); 1193 rtn = scsi_request_sense(scmd); 1194 if (rtn != SUCCESS) 1195 continue; 1196 1197 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1198 "sense requested, result %x\n", scmd->result)); 1199 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1200 1201 rtn = scsi_decide_disposition(scmd); 1202 1203 /* 1204 * if the result was normal, then just pass it along to the 1205 * upper level. 1206 */ 1207 if (rtn == SUCCESS) 1208 /* we don't want this command reissued, just 1209 * finished with the sense data, so set 1210 * retries to the max allowed to ensure it 1211 * won't get reissued */ 1212 scmd->retries = scmd->allowed; 1213 else if (rtn != NEEDS_RETRY) 1214 continue; 1215 1216 scsi_eh_finish_cmd(scmd, done_q); 1217 } 1218 1219 return list_empty(work_q); 1220 } 1221 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1222 1223 /** 1224 * scsi_eh_tur - Send TUR to device. 1225 * @scmd: &scsi_cmnd to send TUR 1226 * 1227 * Return value: 1228 * 0 - Device is ready. 1 - Device NOT ready. 1229 */ 1230 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1231 { 1232 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1233 int retry_cnt = 1, rtn; 1234 1235 retry_tur: 1236 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1237 scmd->device->eh_timeout, 0); 1238 1239 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1240 "%s return: %x\n", __func__, rtn)); 1241 1242 switch (rtn) { 1243 case NEEDS_RETRY: 1244 if (retry_cnt--) 1245 goto retry_tur; 1246 /*FALLTHRU*/ 1247 case SUCCESS: 1248 return 0; 1249 default: 1250 return 1; 1251 } 1252 } 1253 1254 /** 1255 * scsi_eh_test_devices - check if devices are responding from error recovery. 1256 * @cmd_list: scsi commands in error recovery. 1257 * @work_q: queue for commands which still need more error recovery 1258 * @done_q: queue for commands which are finished 1259 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1260 * 1261 * Decription: 1262 * Tests if devices are in a working state. Commands to devices now in 1263 * a working state are sent to the done_q while commands to devices which 1264 * are still failing to respond are returned to the work_q for more 1265 * processing. 1266 **/ 1267 static int scsi_eh_test_devices(struct list_head *cmd_list, 1268 struct list_head *work_q, 1269 struct list_head *done_q, int try_stu) 1270 { 1271 struct scsi_cmnd *scmd, *next; 1272 struct scsi_device *sdev; 1273 int finish_cmds; 1274 1275 while (!list_empty(cmd_list)) { 1276 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1277 sdev = scmd->device; 1278 1279 if (!try_stu) { 1280 if (scsi_host_eh_past_deadline(sdev->host)) { 1281 /* Push items back onto work_q */ 1282 list_splice_init(cmd_list, work_q); 1283 SCSI_LOG_ERROR_RECOVERY(3, 1284 sdev_printk(KERN_INFO, sdev, 1285 "%s: skip test device, past eh deadline", 1286 current->comm)); 1287 break; 1288 } 1289 } 1290 1291 finish_cmds = !scsi_device_online(scmd->device) || 1292 (try_stu && !scsi_eh_try_stu(scmd) && 1293 !scsi_eh_tur(scmd)) || 1294 !scsi_eh_tur(scmd); 1295 1296 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1297 if (scmd->device == sdev) { 1298 if (finish_cmds && 1299 (try_stu || 1300 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1301 scsi_eh_finish_cmd(scmd, done_q); 1302 else 1303 list_move_tail(&scmd->eh_entry, work_q); 1304 } 1305 } 1306 return list_empty(work_q); 1307 } 1308 1309 1310 /** 1311 * scsi_eh_abort_cmds - abort pending commands. 1312 * @work_q: &list_head for pending commands. 1313 * @done_q: &list_head for processed commands. 1314 * 1315 * Decription: 1316 * Try and see whether or not it makes sense to try and abort the 1317 * running command. This only works out to be the case if we have one 1318 * command that has timed out. If the command simply failed, it makes 1319 * no sense to try and abort the command, since as far as the shost 1320 * adapter is concerned, it isn't running. 1321 */ 1322 static int scsi_eh_abort_cmds(struct list_head *work_q, 1323 struct list_head *done_q) 1324 { 1325 struct scsi_cmnd *scmd, *next; 1326 LIST_HEAD(check_list); 1327 int rtn; 1328 struct Scsi_Host *shost; 1329 1330 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1331 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 1332 continue; 1333 shost = scmd->device->host; 1334 if (scsi_host_eh_past_deadline(shost)) { 1335 list_splice_init(&check_list, work_q); 1336 SCSI_LOG_ERROR_RECOVERY(3, 1337 scmd_printk(KERN_INFO, scmd, 1338 "%s: skip aborting cmd, past eh deadline\n", 1339 current->comm)); 1340 return list_empty(work_q); 1341 } 1342 SCSI_LOG_ERROR_RECOVERY(3, 1343 scmd_printk(KERN_INFO, scmd, 1344 "%s: aborting cmd\n", current->comm)); 1345 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); 1346 if (rtn == FAILED) { 1347 SCSI_LOG_ERROR_RECOVERY(3, 1348 scmd_printk(KERN_INFO, scmd, 1349 "%s: aborting cmd failed\n", 1350 current->comm)); 1351 list_splice_init(&check_list, work_q); 1352 return list_empty(work_q); 1353 } 1354 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 1355 if (rtn == FAST_IO_FAIL) 1356 scsi_eh_finish_cmd(scmd, done_q); 1357 else 1358 list_move_tail(&scmd->eh_entry, &check_list); 1359 } 1360 1361 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1362 } 1363 1364 /** 1365 * scsi_eh_try_stu - Send START_UNIT to device. 1366 * @scmd: &scsi_cmnd to send START_UNIT 1367 * 1368 * Return value: 1369 * 0 - Device is ready. 1 - Device NOT ready. 1370 */ 1371 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1372 { 1373 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1374 1375 if (scmd->device->allow_restart) { 1376 int i, rtn = NEEDS_RETRY; 1377 1378 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1379 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1380 1381 if (rtn == SUCCESS) 1382 return 0; 1383 } 1384 1385 return 1; 1386 } 1387 1388 /** 1389 * scsi_eh_stu - send START_UNIT if needed 1390 * @shost: &scsi host being recovered. 1391 * @work_q: &list_head for pending commands. 1392 * @done_q: &list_head for processed commands. 1393 * 1394 * Notes: 1395 * If commands are failing due to not ready, initializing command required, 1396 * try revalidating the device, which will end up sending a start unit. 1397 */ 1398 static int scsi_eh_stu(struct Scsi_Host *shost, 1399 struct list_head *work_q, 1400 struct list_head *done_q) 1401 { 1402 struct scsi_cmnd *scmd, *stu_scmd, *next; 1403 struct scsi_device *sdev; 1404 1405 shost_for_each_device(sdev, shost) { 1406 if (scsi_host_eh_past_deadline(shost)) { 1407 SCSI_LOG_ERROR_RECOVERY(3, 1408 sdev_printk(KERN_INFO, sdev, 1409 "%s: skip START_UNIT, past eh deadline\n", 1410 current->comm)); 1411 break; 1412 } 1413 stu_scmd = NULL; 1414 list_for_each_entry(scmd, work_q, eh_entry) 1415 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1416 scsi_check_sense(scmd) == FAILED ) { 1417 stu_scmd = scmd; 1418 break; 1419 } 1420 1421 if (!stu_scmd) 1422 continue; 1423 1424 SCSI_LOG_ERROR_RECOVERY(3, 1425 sdev_printk(KERN_INFO, sdev, 1426 "%s: Sending START_UNIT\n", 1427 current->comm)); 1428 1429 if (!scsi_eh_try_stu(stu_scmd)) { 1430 if (!scsi_device_online(sdev) || 1431 !scsi_eh_tur(stu_scmd)) { 1432 list_for_each_entry_safe(scmd, next, 1433 work_q, eh_entry) { 1434 if (scmd->device == sdev && 1435 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1436 scsi_eh_finish_cmd(scmd, done_q); 1437 } 1438 } 1439 } else { 1440 SCSI_LOG_ERROR_RECOVERY(3, 1441 sdev_printk(KERN_INFO, sdev, 1442 "%s: START_UNIT failed\n", 1443 current->comm)); 1444 } 1445 } 1446 1447 return list_empty(work_q); 1448 } 1449 1450 1451 /** 1452 * scsi_eh_bus_device_reset - send bdr if needed 1453 * @shost: scsi host being recovered. 1454 * @work_q: &list_head for pending commands. 1455 * @done_q: &list_head for processed commands. 1456 * 1457 * Notes: 1458 * Try a bus device reset. Still, look to see whether we have multiple 1459 * devices that are jammed or not - if we have multiple devices, it 1460 * makes no sense to try bus_device_reset - we really would need to try 1461 * a bus_reset instead. 1462 */ 1463 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1464 struct list_head *work_q, 1465 struct list_head *done_q) 1466 { 1467 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1468 struct scsi_device *sdev; 1469 int rtn; 1470 1471 shost_for_each_device(sdev, shost) { 1472 if (scsi_host_eh_past_deadline(shost)) { 1473 SCSI_LOG_ERROR_RECOVERY(3, 1474 sdev_printk(KERN_INFO, sdev, 1475 "%s: skip BDR, past eh deadline\n", 1476 current->comm)); 1477 break; 1478 } 1479 bdr_scmd = NULL; 1480 list_for_each_entry(scmd, work_q, eh_entry) 1481 if (scmd->device == sdev) { 1482 bdr_scmd = scmd; 1483 break; 1484 } 1485 1486 if (!bdr_scmd) 1487 continue; 1488 1489 SCSI_LOG_ERROR_RECOVERY(3, 1490 sdev_printk(KERN_INFO, sdev, 1491 "%s: Sending BDR\n", current->comm)); 1492 rtn = scsi_try_bus_device_reset(bdr_scmd); 1493 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1494 if (!scsi_device_online(sdev) || 1495 rtn == FAST_IO_FAIL || 1496 !scsi_eh_tur(bdr_scmd)) { 1497 list_for_each_entry_safe(scmd, next, 1498 work_q, eh_entry) { 1499 if (scmd->device == sdev && 1500 scsi_eh_action(scmd, rtn) != FAILED) 1501 scsi_eh_finish_cmd(scmd, 1502 done_q); 1503 } 1504 } 1505 } else { 1506 SCSI_LOG_ERROR_RECOVERY(3, 1507 sdev_printk(KERN_INFO, sdev, 1508 "%s: BDR failed\n", current->comm)); 1509 } 1510 } 1511 1512 return list_empty(work_q); 1513 } 1514 1515 /** 1516 * scsi_eh_target_reset - send target reset if needed 1517 * @shost: scsi host being recovered. 1518 * @work_q: &list_head for pending commands. 1519 * @done_q: &list_head for processed commands. 1520 * 1521 * Notes: 1522 * Try a target reset. 1523 */ 1524 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1525 struct list_head *work_q, 1526 struct list_head *done_q) 1527 { 1528 LIST_HEAD(tmp_list); 1529 LIST_HEAD(check_list); 1530 1531 list_splice_init(work_q, &tmp_list); 1532 1533 while (!list_empty(&tmp_list)) { 1534 struct scsi_cmnd *next, *scmd; 1535 int rtn; 1536 unsigned int id; 1537 1538 if (scsi_host_eh_past_deadline(shost)) { 1539 /* push back on work queue for further processing */ 1540 list_splice_init(&check_list, work_q); 1541 list_splice_init(&tmp_list, work_q); 1542 SCSI_LOG_ERROR_RECOVERY(3, 1543 shost_printk(KERN_INFO, shost, 1544 "%s: Skip target reset, past eh deadline\n", 1545 current->comm)); 1546 return list_empty(work_q); 1547 } 1548 1549 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1550 id = scmd_id(scmd); 1551 1552 SCSI_LOG_ERROR_RECOVERY(3, 1553 shost_printk(KERN_INFO, shost, 1554 "%s: Sending target reset to target %d\n", 1555 current->comm, id)); 1556 rtn = scsi_try_target_reset(scmd); 1557 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1558 SCSI_LOG_ERROR_RECOVERY(3, 1559 shost_printk(KERN_INFO, shost, 1560 "%s: Target reset failed" 1561 " target: %d\n", 1562 current->comm, id)); 1563 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1564 if (scmd_id(scmd) != id) 1565 continue; 1566 1567 if (rtn == SUCCESS) 1568 list_move_tail(&scmd->eh_entry, &check_list); 1569 else if (rtn == FAST_IO_FAIL) 1570 scsi_eh_finish_cmd(scmd, done_q); 1571 else 1572 /* push back on work queue for further processing */ 1573 list_move(&scmd->eh_entry, work_q); 1574 } 1575 } 1576 1577 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1578 } 1579 1580 /** 1581 * scsi_eh_bus_reset - send a bus reset 1582 * @shost: &scsi host being recovered. 1583 * @work_q: &list_head for pending commands. 1584 * @done_q: &list_head for processed commands. 1585 */ 1586 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1587 struct list_head *work_q, 1588 struct list_head *done_q) 1589 { 1590 struct scsi_cmnd *scmd, *chan_scmd, *next; 1591 LIST_HEAD(check_list); 1592 unsigned int channel; 1593 int rtn; 1594 1595 /* 1596 * we really want to loop over the various channels, and do this on 1597 * a channel by channel basis. we should also check to see if any 1598 * of the failed commands are on soft_reset devices, and if so, skip 1599 * the reset. 1600 */ 1601 1602 for (channel = 0; channel <= shost->max_channel; channel++) { 1603 if (scsi_host_eh_past_deadline(shost)) { 1604 list_splice_init(&check_list, work_q); 1605 SCSI_LOG_ERROR_RECOVERY(3, 1606 shost_printk(KERN_INFO, shost, 1607 "%s: skip BRST, past eh deadline\n", 1608 current->comm)); 1609 return list_empty(work_q); 1610 } 1611 1612 chan_scmd = NULL; 1613 list_for_each_entry(scmd, work_q, eh_entry) { 1614 if (channel == scmd_channel(scmd)) { 1615 chan_scmd = scmd; 1616 break; 1617 /* 1618 * FIXME add back in some support for 1619 * soft_reset devices. 1620 */ 1621 } 1622 } 1623 1624 if (!chan_scmd) 1625 continue; 1626 SCSI_LOG_ERROR_RECOVERY(3, 1627 shost_printk(KERN_INFO, shost, 1628 "%s: Sending BRST chan: %d\n", 1629 current->comm, channel)); 1630 rtn = scsi_try_bus_reset(chan_scmd); 1631 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1632 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1633 if (channel == scmd_channel(scmd)) { 1634 if (rtn == FAST_IO_FAIL) 1635 scsi_eh_finish_cmd(scmd, 1636 done_q); 1637 else 1638 list_move_tail(&scmd->eh_entry, 1639 &check_list); 1640 } 1641 } 1642 } else { 1643 SCSI_LOG_ERROR_RECOVERY(3, 1644 shost_printk(KERN_INFO, shost, 1645 "%s: BRST failed chan: %d\n", 1646 current->comm, channel)); 1647 } 1648 } 1649 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1650 } 1651 1652 /** 1653 * scsi_eh_host_reset - send a host reset 1654 * @shost: host to be reset. 1655 * @work_q: &list_head for pending commands. 1656 * @done_q: &list_head for processed commands. 1657 */ 1658 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1659 struct list_head *work_q, 1660 struct list_head *done_q) 1661 { 1662 struct scsi_cmnd *scmd, *next; 1663 LIST_HEAD(check_list); 1664 int rtn; 1665 1666 if (!list_empty(work_q)) { 1667 scmd = list_entry(work_q->next, 1668 struct scsi_cmnd, eh_entry); 1669 1670 SCSI_LOG_ERROR_RECOVERY(3, 1671 shost_printk(KERN_INFO, shost, 1672 "%s: Sending HRST\n", 1673 current->comm)); 1674 1675 rtn = scsi_try_host_reset(scmd); 1676 if (rtn == SUCCESS) { 1677 list_splice_init(work_q, &check_list); 1678 } else if (rtn == FAST_IO_FAIL) { 1679 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1680 scsi_eh_finish_cmd(scmd, done_q); 1681 } 1682 } else { 1683 SCSI_LOG_ERROR_RECOVERY(3, 1684 shost_printk(KERN_INFO, shost, 1685 "%s: HRST failed\n", 1686 current->comm)); 1687 } 1688 } 1689 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1690 } 1691 1692 /** 1693 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1694 * @work_q: &list_head for pending commands. 1695 * @done_q: &list_head for processed commands. 1696 */ 1697 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1698 struct list_head *done_q) 1699 { 1700 struct scsi_cmnd *scmd, *next; 1701 1702 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1703 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1704 "not ready after error recovery\n"); 1705 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1706 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1707 /* 1708 * FIXME: Handle lost cmds. 1709 */ 1710 } 1711 scsi_eh_finish_cmd(scmd, done_q); 1712 } 1713 return; 1714 } 1715 1716 /** 1717 * scsi_noretry_cmd - determine if command should be failed fast 1718 * @scmd: SCSI cmd to examine. 1719 */ 1720 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1721 { 1722 switch (host_byte(scmd->result)) { 1723 case DID_OK: 1724 break; 1725 case DID_TIME_OUT: 1726 goto check_type; 1727 case DID_BUS_BUSY: 1728 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1729 case DID_PARITY: 1730 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1731 case DID_ERROR: 1732 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1733 status_byte(scmd->result) == RESERVATION_CONFLICT) 1734 return 0; 1735 /* fall through */ 1736 case DID_SOFT_ERROR: 1737 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1738 } 1739 1740 if (status_byte(scmd->result) != CHECK_CONDITION) 1741 return 0; 1742 1743 check_type: 1744 /* 1745 * assume caller has checked sense and determined 1746 * the check condition was retryable. 1747 */ 1748 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1749 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC) 1750 return 1; 1751 else 1752 return 0; 1753 } 1754 1755 /** 1756 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1757 * @scmd: SCSI cmd to examine. 1758 * 1759 * Notes: 1760 * This is *only* called when we are examining the status after sending 1761 * out the actual data command. any commands that are queued for error 1762 * recovery (e.g. test_unit_ready) do *not* come through here. 1763 * 1764 * When this routine returns failed, it means the error handler thread 1765 * is woken. In cases where the error code indicates an error that 1766 * doesn't require the error handler read (i.e. we don't need to 1767 * abort/reset), this function should return SUCCESS. 1768 */ 1769 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1770 { 1771 int rtn; 1772 1773 /* 1774 * if the device is offline, then we clearly just pass the result back 1775 * up to the top level. 1776 */ 1777 if (!scsi_device_online(scmd->device)) { 1778 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1779 "%s: device offline - report as SUCCESS\n", __func__)); 1780 return SUCCESS; 1781 } 1782 1783 /* 1784 * first check the host byte, to see if there is anything in there 1785 * that would indicate what we need to do. 1786 */ 1787 switch (host_byte(scmd->result)) { 1788 case DID_PASSTHROUGH: 1789 /* 1790 * no matter what, pass this through to the upper layer. 1791 * nuke this special code so that it looks like we are saying 1792 * did_ok. 1793 */ 1794 scmd->result &= 0xff00ffff; 1795 return SUCCESS; 1796 case DID_OK: 1797 /* 1798 * looks good. drop through, and check the next byte. 1799 */ 1800 break; 1801 case DID_ABORT: 1802 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1803 set_host_byte(scmd, DID_TIME_OUT); 1804 return SUCCESS; 1805 } 1806 case DID_NO_CONNECT: 1807 case DID_BAD_TARGET: 1808 /* 1809 * note - this means that we just report the status back 1810 * to the top level driver, not that we actually think 1811 * that it indicates SUCCESS. 1812 */ 1813 return SUCCESS; 1814 /* 1815 * when the low level driver returns did_soft_error, 1816 * it is responsible for keeping an internal retry counter 1817 * in order to avoid endless loops (db) 1818 * 1819 * actually this is a bug in this function here. we should 1820 * be mindful of the maximum number of retries specified 1821 * and not get stuck in a loop. 1822 */ 1823 case DID_SOFT_ERROR: 1824 goto maybe_retry; 1825 case DID_IMM_RETRY: 1826 return NEEDS_RETRY; 1827 1828 case DID_REQUEUE: 1829 return ADD_TO_MLQUEUE; 1830 case DID_TRANSPORT_DISRUPTED: 1831 /* 1832 * LLD/transport was disrupted during processing of the IO. 1833 * The transport class is now blocked/blocking, 1834 * and the transport will decide what to do with the IO 1835 * based on its timers and recovery capablilities if 1836 * there are enough retries. 1837 */ 1838 goto maybe_retry; 1839 case DID_TRANSPORT_FAILFAST: 1840 /* 1841 * The transport decided to failfast the IO (most likely 1842 * the fast io fail tmo fired), so send IO directly upwards. 1843 */ 1844 return SUCCESS; 1845 case DID_ERROR: 1846 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1847 status_byte(scmd->result) == RESERVATION_CONFLICT) 1848 /* 1849 * execute reservation conflict processing code 1850 * lower down 1851 */ 1852 break; 1853 /* fallthrough */ 1854 case DID_BUS_BUSY: 1855 case DID_PARITY: 1856 goto maybe_retry; 1857 case DID_TIME_OUT: 1858 /* 1859 * when we scan the bus, we get timeout messages for 1860 * these commands if there is no device available. 1861 * other hosts report did_no_connect for the same thing. 1862 */ 1863 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1864 scmd->cmnd[0] == INQUIRY)) { 1865 return SUCCESS; 1866 } else { 1867 return FAILED; 1868 } 1869 case DID_RESET: 1870 return SUCCESS; 1871 default: 1872 return FAILED; 1873 } 1874 1875 /* 1876 * next, check the message byte. 1877 */ 1878 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1879 return FAILED; 1880 1881 /* 1882 * check the status byte to see if this indicates anything special. 1883 */ 1884 switch (status_byte(scmd->result)) { 1885 case QUEUE_FULL: 1886 scsi_handle_queue_full(scmd->device); 1887 /* 1888 * the case of trying to send too many commands to a 1889 * tagged queueing device. 1890 */ 1891 case BUSY: 1892 /* 1893 * device can't talk to us at the moment. Should only 1894 * occur (SAM-3) when the task queue is empty, so will cause 1895 * the empty queue handling to trigger a stall in the 1896 * device. 1897 */ 1898 return ADD_TO_MLQUEUE; 1899 case GOOD: 1900 if (scmd->cmnd[0] == REPORT_LUNS) 1901 scmd->device->sdev_target->expecting_lun_change = 0; 1902 scsi_handle_queue_ramp_up(scmd->device); 1903 case COMMAND_TERMINATED: 1904 return SUCCESS; 1905 case TASK_ABORTED: 1906 goto maybe_retry; 1907 case CHECK_CONDITION: 1908 rtn = scsi_check_sense(scmd); 1909 if (rtn == NEEDS_RETRY) 1910 goto maybe_retry; 1911 /* if rtn == FAILED, we have no sense information; 1912 * returning FAILED will wake the error handler thread 1913 * to collect the sense and redo the decide 1914 * disposition */ 1915 return rtn; 1916 case CONDITION_GOOD: 1917 case INTERMEDIATE_GOOD: 1918 case INTERMEDIATE_C_GOOD: 1919 case ACA_ACTIVE: 1920 /* 1921 * who knows? FIXME(eric) 1922 */ 1923 return SUCCESS; 1924 1925 case RESERVATION_CONFLICT: 1926 sdev_printk(KERN_INFO, scmd->device, 1927 "reservation conflict\n"); 1928 set_host_byte(scmd, DID_NEXUS_FAILURE); 1929 return SUCCESS; /* causes immediate i/o error */ 1930 default: 1931 return FAILED; 1932 } 1933 return FAILED; 1934 1935 maybe_retry: 1936 1937 /* we requeue for retry because the error was retryable, and 1938 * the request was not marked fast fail. Note that above, 1939 * even if the request is marked fast fail, we still requeue 1940 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1941 if ((++scmd->retries) <= scmd->allowed 1942 && !scsi_noretry_cmd(scmd)) { 1943 return NEEDS_RETRY; 1944 } else { 1945 /* 1946 * no more retries - report this one back to upper level. 1947 */ 1948 return SUCCESS; 1949 } 1950 } 1951 1952 static void eh_lock_door_done(struct request *req, int uptodate) 1953 { 1954 __blk_put_request(req->q, req); 1955 } 1956 1957 /** 1958 * scsi_eh_lock_door - Prevent medium removal for the specified device 1959 * @sdev: SCSI device to prevent medium removal 1960 * 1961 * Locking: 1962 * We must be called from process context. 1963 * 1964 * Notes: 1965 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1966 * head of the devices request queue, and continue. 1967 */ 1968 static void scsi_eh_lock_door(struct scsi_device *sdev) 1969 { 1970 struct request *req; 1971 1972 /* 1973 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a 1974 * request becomes available 1975 */ 1976 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL); 1977 if (IS_ERR(req)) 1978 return; 1979 1980 blk_rq_set_block_pc(req); 1981 1982 req->cmd[0] = ALLOW_MEDIUM_REMOVAL; 1983 req->cmd[1] = 0; 1984 req->cmd[2] = 0; 1985 req->cmd[3] = 0; 1986 req->cmd[4] = SCSI_REMOVAL_PREVENT; 1987 req->cmd[5] = 0; 1988 1989 req->cmd_len = COMMAND_SIZE(req->cmd[0]); 1990 1991 req->rq_flags |= RQF_QUIET; 1992 req->timeout = 10 * HZ; 1993 req->retries = 5; 1994 1995 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); 1996 } 1997 1998 /** 1999 * scsi_restart_operations - restart io operations to the specified host. 2000 * @shost: Host we are restarting. 2001 * 2002 * Notes: 2003 * When we entered the error handler, we blocked all further i/o to 2004 * this device. we need to 'reverse' this process. 2005 */ 2006 static void scsi_restart_operations(struct Scsi_Host *shost) 2007 { 2008 struct scsi_device *sdev; 2009 unsigned long flags; 2010 2011 /* 2012 * If the door was locked, we need to insert a door lock request 2013 * onto the head of the SCSI request queue for the device. There 2014 * is no point trying to lock the door of an off-line device. 2015 */ 2016 shost_for_each_device(sdev, shost) { 2017 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2018 scsi_eh_lock_door(sdev); 2019 sdev->was_reset = 0; 2020 } 2021 } 2022 2023 /* 2024 * next free up anything directly waiting upon the host. this 2025 * will be requests for character device operations, and also for 2026 * ioctls to queued block devices. 2027 */ 2028 SCSI_LOG_ERROR_RECOVERY(3, 2029 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2030 2031 spin_lock_irqsave(shost->host_lock, flags); 2032 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2033 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2034 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2035 spin_unlock_irqrestore(shost->host_lock, flags); 2036 2037 wake_up(&shost->host_wait); 2038 2039 /* 2040 * finally we need to re-initiate requests that may be pending. we will 2041 * have had everything blocked while error handling is taking place, and 2042 * now that error recovery is done, we will need to ensure that these 2043 * requests are started. 2044 */ 2045 scsi_run_host_queues(shost); 2046 2047 /* 2048 * if eh is active and host_eh_scheduled is pending we need to re-run 2049 * recovery. we do this check after scsi_run_host_queues() to allow 2050 * everything pent up since the last eh run a chance to make forward 2051 * progress before we sync again. Either we'll immediately re-run 2052 * recovery or scsi_device_unbusy() will wake us again when these 2053 * pending commands complete. 2054 */ 2055 spin_lock_irqsave(shost->host_lock, flags); 2056 if (shost->host_eh_scheduled) 2057 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2058 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2059 spin_unlock_irqrestore(shost->host_lock, flags); 2060 } 2061 2062 /** 2063 * scsi_eh_ready_devs - check device ready state and recover if not. 2064 * @shost: host to be recovered. 2065 * @work_q: &list_head for pending commands. 2066 * @done_q: &list_head for processed commands. 2067 */ 2068 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2069 struct list_head *work_q, 2070 struct list_head *done_q) 2071 { 2072 if (!scsi_eh_stu(shost, work_q, done_q)) 2073 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2074 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2075 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2076 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2077 scsi_eh_offline_sdevs(work_q, 2078 done_q); 2079 } 2080 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2081 2082 /** 2083 * scsi_eh_flush_done_q - finish processed commands or retry them. 2084 * @done_q: list_head of processed commands. 2085 */ 2086 void scsi_eh_flush_done_q(struct list_head *done_q) 2087 { 2088 struct scsi_cmnd *scmd, *next; 2089 2090 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2091 list_del_init(&scmd->eh_entry); 2092 if (scsi_device_online(scmd->device) && 2093 !scsi_noretry_cmd(scmd) && 2094 (++scmd->retries <= scmd->allowed)) { 2095 SCSI_LOG_ERROR_RECOVERY(3, 2096 scmd_printk(KERN_INFO, scmd, 2097 "%s: flush retry cmd\n", 2098 current->comm)); 2099 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2100 } else { 2101 /* 2102 * If just we got sense for the device (called 2103 * scsi_eh_get_sense), scmd->result is already 2104 * set, do not set DRIVER_TIMEOUT. 2105 */ 2106 if (!scmd->result) 2107 scmd->result |= (DRIVER_TIMEOUT << 24); 2108 SCSI_LOG_ERROR_RECOVERY(3, 2109 scmd_printk(KERN_INFO, scmd, 2110 "%s: flush finish cmd\n", 2111 current->comm)); 2112 scsi_finish_command(scmd); 2113 } 2114 } 2115 } 2116 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2117 2118 /** 2119 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2120 * @shost: Host to unjam. 2121 * 2122 * Notes: 2123 * When we come in here, we *know* that all commands on the bus have 2124 * either completed, failed or timed out. we also know that no further 2125 * commands are being sent to the host, so things are relatively quiet 2126 * and we have freedom to fiddle with things as we wish. 2127 * 2128 * This is only the *default* implementation. it is possible for 2129 * individual drivers to supply their own version of this function, and 2130 * if the maintainer wishes to do this, it is strongly suggested that 2131 * this function be taken as a template and modified. this function 2132 * was designed to correctly handle problems for about 95% of the 2133 * different cases out there, and it should always provide at least a 2134 * reasonable amount of error recovery. 2135 * 2136 * Any command marked 'failed' or 'timeout' must eventually have 2137 * scsi_finish_cmd() called for it. we do all of the retry stuff 2138 * here, so when we restart the host after we return it should have an 2139 * empty queue. 2140 */ 2141 static void scsi_unjam_host(struct Scsi_Host *shost) 2142 { 2143 unsigned long flags; 2144 LIST_HEAD(eh_work_q); 2145 LIST_HEAD(eh_done_q); 2146 2147 spin_lock_irqsave(shost->host_lock, flags); 2148 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2149 spin_unlock_irqrestore(shost->host_lock, flags); 2150 2151 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2152 2153 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2154 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 2155 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2156 2157 spin_lock_irqsave(shost->host_lock, flags); 2158 if (shost->eh_deadline != -1) 2159 shost->last_reset = 0; 2160 spin_unlock_irqrestore(shost->host_lock, flags); 2161 scsi_eh_flush_done_q(&eh_done_q); 2162 } 2163 2164 /** 2165 * scsi_error_handler - SCSI error handler thread 2166 * @data: Host for which we are running. 2167 * 2168 * Notes: 2169 * This is the main error handling loop. This is run as a kernel thread 2170 * for every SCSI host and handles all error handling activity. 2171 */ 2172 int scsi_error_handler(void *data) 2173 { 2174 struct Scsi_Host *shost = data; 2175 2176 /* 2177 * We use TASK_INTERRUPTIBLE so that the thread is not 2178 * counted against the load average as a running process. 2179 * We never actually get interrupted because kthread_run 2180 * disables signal delivery for the created thread. 2181 */ 2182 while (true) { 2183 /* 2184 * The sequence in kthread_stop() sets the stop flag first 2185 * then wakes the process. To avoid missed wakeups, the task 2186 * should always be in a non running state before the stop 2187 * flag is checked 2188 */ 2189 set_current_state(TASK_INTERRUPTIBLE); 2190 if (kthread_should_stop()) 2191 break; 2192 2193 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2194 shost->host_failed != atomic_read(&shost->host_busy)) { 2195 SCSI_LOG_ERROR_RECOVERY(1, 2196 shost_printk(KERN_INFO, shost, 2197 "scsi_eh_%d: sleeping\n", 2198 shost->host_no)); 2199 schedule(); 2200 continue; 2201 } 2202 2203 __set_current_state(TASK_RUNNING); 2204 SCSI_LOG_ERROR_RECOVERY(1, 2205 shost_printk(KERN_INFO, shost, 2206 "scsi_eh_%d: waking up %d/%d/%d\n", 2207 shost->host_no, shost->host_eh_scheduled, 2208 shost->host_failed, 2209 atomic_read(&shost->host_busy))); 2210 2211 /* 2212 * We have a host that is failing for some reason. Figure out 2213 * what we need to do to get it up and online again (if we can). 2214 * If we fail, we end up taking the thing offline. 2215 */ 2216 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2217 SCSI_LOG_ERROR_RECOVERY(1, 2218 shost_printk(KERN_ERR, shost, 2219 "scsi_eh_%d: unable to autoresume\n", 2220 shost->host_no)); 2221 continue; 2222 } 2223 2224 if (shost->transportt->eh_strategy_handler) 2225 shost->transportt->eh_strategy_handler(shost); 2226 else 2227 scsi_unjam_host(shost); 2228 2229 /* All scmds have been handled */ 2230 shost->host_failed = 0; 2231 2232 /* 2233 * Note - if the above fails completely, the action is to take 2234 * individual devices offline and flush the queue of any 2235 * outstanding requests that may have been pending. When we 2236 * restart, we restart any I/O to any other devices on the bus 2237 * which are still online. 2238 */ 2239 scsi_restart_operations(shost); 2240 if (!shost->eh_noresume) 2241 scsi_autopm_put_host(shost); 2242 } 2243 __set_current_state(TASK_RUNNING); 2244 2245 SCSI_LOG_ERROR_RECOVERY(1, 2246 shost_printk(KERN_INFO, shost, 2247 "Error handler scsi_eh_%d exiting\n", 2248 shost->host_no)); 2249 shost->ehandler = NULL; 2250 return 0; 2251 } 2252 2253 /* 2254 * Function: scsi_report_bus_reset() 2255 * 2256 * Purpose: Utility function used by low-level drivers to report that 2257 * they have observed a bus reset on the bus being handled. 2258 * 2259 * Arguments: shost - Host in question 2260 * channel - channel on which reset was observed. 2261 * 2262 * Returns: Nothing 2263 * 2264 * Lock status: Host lock must be held. 2265 * 2266 * Notes: This only needs to be called if the reset is one which 2267 * originates from an unknown location. Resets originated 2268 * by the mid-level itself don't need to call this, but there 2269 * should be no harm. 2270 * 2271 * The main purpose of this is to make sure that a CHECK_CONDITION 2272 * is properly treated. 2273 */ 2274 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2275 { 2276 struct scsi_device *sdev; 2277 2278 __shost_for_each_device(sdev, shost) { 2279 if (channel == sdev_channel(sdev)) 2280 __scsi_report_device_reset(sdev, NULL); 2281 } 2282 } 2283 EXPORT_SYMBOL(scsi_report_bus_reset); 2284 2285 /* 2286 * Function: scsi_report_device_reset() 2287 * 2288 * Purpose: Utility function used by low-level drivers to report that 2289 * they have observed a device reset on the device being handled. 2290 * 2291 * Arguments: shost - Host in question 2292 * channel - channel on which reset was observed 2293 * target - target on which reset was observed 2294 * 2295 * Returns: Nothing 2296 * 2297 * Lock status: Host lock must be held 2298 * 2299 * Notes: This only needs to be called if the reset is one which 2300 * originates from an unknown location. Resets originated 2301 * by the mid-level itself don't need to call this, but there 2302 * should be no harm. 2303 * 2304 * The main purpose of this is to make sure that a CHECK_CONDITION 2305 * is properly treated. 2306 */ 2307 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2308 { 2309 struct scsi_device *sdev; 2310 2311 __shost_for_each_device(sdev, shost) { 2312 if (channel == sdev_channel(sdev) && 2313 target == sdev_id(sdev)) 2314 __scsi_report_device_reset(sdev, NULL); 2315 } 2316 } 2317 EXPORT_SYMBOL(scsi_report_device_reset); 2318 2319 static void 2320 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2321 { 2322 } 2323 2324 /** 2325 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2326 * @dev: scsi_device to operate on 2327 * @arg: reset type (see sg.h) 2328 */ 2329 int 2330 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2331 { 2332 struct scsi_cmnd *scmd; 2333 struct Scsi_Host *shost = dev->host; 2334 struct request req; 2335 unsigned long flags; 2336 int error = 0, rtn, val; 2337 2338 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2339 return -EACCES; 2340 2341 error = get_user(val, arg); 2342 if (error) 2343 return error; 2344 2345 if (scsi_autopm_get_host(shost) < 0) 2346 return -EIO; 2347 2348 error = -EIO; 2349 scmd = scsi_get_command(dev, GFP_KERNEL); 2350 if (!scmd) 2351 goto out_put_autopm_host; 2352 2353 blk_rq_init(NULL, &req); 2354 scmd->request = &req; 2355 2356 scmd->cmnd = req.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 2417 out_put_autopm_host: 2418 scsi_autopm_put_host(shost); 2419 return error; 2420 } 2421 EXPORT_SYMBOL(scsi_ioctl_reset); 2422 2423 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2424 struct scsi_sense_hdr *sshdr) 2425 { 2426 return scsi_normalize_sense(cmd->sense_buffer, 2427 SCSI_SENSE_BUFFERSIZE, sshdr); 2428 } 2429 EXPORT_SYMBOL(scsi_command_normalize_sense); 2430 2431 /** 2432 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2433 * @sense_buffer: byte array of sense data 2434 * @sb_len: number of valid bytes in sense_buffer 2435 * @info_out: pointer to 64 integer where 8 or 4 byte information 2436 * field will be placed if found. 2437 * 2438 * Return value: 2439 * 1 if information field found, 0 if not found. 2440 */ 2441 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 2442 u64 * info_out) 2443 { 2444 int j; 2445 const u8 * ucp; 2446 u64 ull; 2447 2448 if (sb_len < 7) 2449 return 0; 2450 switch (sense_buffer[0] & 0x7f) { 2451 case 0x70: 2452 case 0x71: 2453 if (sense_buffer[0] & 0x80) { 2454 *info_out = (sense_buffer[3] << 24) + 2455 (sense_buffer[4] << 16) + 2456 (sense_buffer[5] << 8) + sense_buffer[6]; 2457 return 1; 2458 } else 2459 return 0; 2460 case 0x72: 2461 case 0x73: 2462 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2463 0 /* info desc */); 2464 if (ucp && (0xa == ucp[1])) { 2465 ull = 0; 2466 for (j = 0; j < 8; ++j) { 2467 if (j > 0) 2468 ull <<= 8; 2469 ull |= ucp[4 + j]; 2470 } 2471 *info_out = ull; 2472 return 1; 2473 } else 2474 return 0; 2475 default: 2476 return 0; 2477 } 2478 } 2479 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2480