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 #include <asm/unaligned.h> 50 51 static void scsi_eh_done(struct scsi_cmnd *scmd); 52 53 /* 54 * These should *probably* be handled by the host itself. 55 * Since it is allowed to sleep, it probably should. 56 */ 57 #define BUS_RESET_SETTLE_TIME (10) 58 #define HOST_RESET_SETTLE_TIME (10) 59 60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 61 static int scsi_try_to_abort_cmd(struct scsi_host_template *, 62 struct scsi_cmnd *); 63 64 /* called with shost->host_lock held */ 65 void scsi_eh_wakeup(struct Scsi_Host *shost) 66 { 67 if (atomic_read(&shost->host_busy) == shost->host_failed) { 68 trace_scsi_eh_wakeup(shost); 69 wake_up_process(shost->ehandler); 70 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 71 "Waking error handler thread\n")); 72 } 73 } 74 75 /** 76 * scsi_schedule_eh - schedule EH for SCSI host 77 * @shost: SCSI host to invoke error handling on. 78 * 79 * Schedule SCSI EH without scmd. 80 */ 81 void scsi_schedule_eh(struct Scsi_Host *shost) 82 { 83 unsigned long flags; 84 85 spin_lock_irqsave(shost->host_lock, flags); 86 87 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 88 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 89 shost->host_eh_scheduled++; 90 scsi_eh_wakeup(shost); 91 } 92 93 spin_unlock_irqrestore(shost->host_lock, flags); 94 } 95 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 96 97 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 98 { 99 if (!shost->last_reset || shost->eh_deadline == -1) 100 return 0; 101 102 /* 103 * 32bit accesses are guaranteed to be atomic 104 * (on all supported architectures), so instead 105 * of using a spinlock we can as well double check 106 * if eh_deadline has been set to 'off' during the 107 * time_before call. 108 */ 109 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 110 shost->eh_deadline > -1) 111 return 0; 112 113 return 1; 114 } 115 116 /** 117 * scmd_eh_abort_handler - Handle command aborts 118 * @work: command to be aborted. 119 */ 120 void 121 scmd_eh_abort_handler(struct work_struct *work) 122 { 123 struct scsi_cmnd *scmd = 124 container_of(work, struct scsi_cmnd, abort_work.work); 125 struct scsi_device *sdev = scmd->device; 126 int rtn; 127 128 if (scsi_host_eh_past_deadline(sdev->host)) { 129 SCSI_LOG_ERROR_RECOVERY(3, 130 scmd_printk(KERN_INFO, scmd, 131 "eh timeout, not aborting\n")); 132 } else { 133 SCSI_LOG_ERROR_RECOVERY(3, 134 scmd_printk(KERN_INFO, scmd, 135 "aborting command\n")); 136 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); 137 if (rtn == SUCCESS) { 138 set_host_byte(scmd, DID_TIME_OUT); 139 if (scsi_host_eh_past_deadline(sdev->host)) { 140 SCSI_LOG_ERROR_RECOVERY(3, 141 scmd_printk(KERN_INFO, scmd, 142 "eh timeout, not retrying " 143 "aborted command\n")); 144 } else if (!scsi_noretry_cmd(scmd) && 145 (++scmd->retries <= scmd->allowed)) { 146 SCSI_LOG_ERROR_RECOVERY(3, 147 scmd_printk(KERN_WARNING, scmd, 148 "retry aborted command\n")); 149 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 150 return; 151 } else { 152 SCSI_LOG_ERROR_RECOVERY(3, 153 scmd_printk(KERN_WARNING, scmd, 154 "finish aborted command\n")); 155 scsi_finish_command(scmd); 156 return; 157 } 158 } else { 159 SCSI_LOG_ERROR_RECOVERY(3, 160 scmd_printk(KERN_INFO, scmd, 161 "cmd abort %s\n", 162 (rtn == FAST_IO_FAIL) ? 163 "not send" : "failed")); 164 } 165 } 166 167 scsi_eh_scmd_add(scmd); 168 } 169 170 /** 171 * scsi_abort_command - schedule a command abort 172 * @scmd: scmd to abort. 173 * 174 * We only need to abort commands after a command timeout 175 */ 176 static int 177 scsi_abort_command(struct scsi_cmnd *scmd) 178 { 179 struct scsi_device *sdev = scmd->device; 180 struct Scsi_Host *shost = sdev->host; 181 unsigned long flags; 182 183 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 184 /* 185 * Retry after abort failed, escalate to next level. 186 */ 187 SCSI_LOG_ERROR_RECOVERY(3, 188 scmd_printk(KERN_INFO, scmd, 189 "previous abort failed\n")); 190 BUG_ON(delayed_work_pending(&scmd->abort_work)); 191 return FAILED; 192 } 193 194 spin_lock_irqsave(shost->host_lock, flags); 195 if (shost->eh_deadline != -1 && !shost->last_reset) 196 shost->last_reset = jiffies; 197 spin_unlock_irqrestore(shost->host_lock, flags); 198 199 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 200 SCSI_LOG_ERROR_RECOVERY(3, 201 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 202 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 203 return SUCCESS; 204 } 205 206 /** 207 * scsi_eh_reset - call into ->eh_action to reset internal counters 208 * @scmd: scmd to run eh on. 209 * 210 * The scsi driver might be carrying internal state about the 211 * devices, so we need to call into the driver to reset the 212 * internal state once the error handler is started. 213 */ 214 static void scsi_eh_reset(struct scsi_cmnd *scmd) 215 { 216 if (!blk_rq_is_passthrough(scmd->request)) { 217 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 218 if (sdrv->eh_reset) 219 sdrv->eh_reset(scmd); 220 } 221 } 222 223 /** 224 * scsi_eh_scmd_add - add scsi cmd to error handling. 225 * @scmd: scmd to run eh on. 226 */ 227 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 228 { 229 struct Scsi_Host *shost = scmd->device->host; 230 unsigned long flags; 231 int ret; 232 233 WARN_ON_ONCE(!shost->ehandler); 234 235 spin_lock_irqsave(shost->host_lock, flags); 236 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 237 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 238 WARN_ON_ONCE(ret); 239 } 240 if (shost->eh_deadline != -1 && !shost->last_reset) 241 shost->last_reset = jiffies; 242 243 scsi_eh_reset(scmd); 244 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 245 shost->host_failed++; 246 scsi_eh_wakeup(shost); 247 spin_unlock_irqrestore(shost->host_lock, flags); 248 } 249 250 /** 251 * scsi_times_out - Timeout function for normal scsi commands. 252 * @req: request that is timing out. 253 * 254 * Notes: 255 * We do not need to lock this. There is the potential for a race 256 * only in that the normal completion handling might run, but if the 257 * normal completion function determines that the timer has already 258 * fired, then it mustn't do anything. 259 */ 260 enum blk_eh_timer_return scsi_times_out(struct request *req) 261 { 262 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 263 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; 264 struct Scsi_Host *host = scmd->device->host; 265 266 trace_scsi_dispatch_cmd_timeout(scmd); 267 scsi_log_completion(scmd, TIMEOUT_ERROR); 268 269 if (host->eh_deadline != -1 && !host->last_reset) 270 host->last_reset = jiffies; 271 272 if (host->hostt->eh_timed_out) 273 rtn = host->hostt->eh_timed_out(scmd); 274 275 if (rtn == BLK_EH_NOT_HANDLED) { 276 if (scsi_abort_command(scmd) != SUCCESS) { 277 set_host_byte(scmd, DID_TIME_OUT); 278 scsi_eh_scmd_add(scmd); 279 } 280 } 281 282 return rtn; 283 } 284 285 /** 286 * scsi_block_when_processing_errors - Prevent cmds from being queued. 287 * @sdev: Device on which we are performing recovery. 288 * 289 * Description: 290 * We block until the host is out of error recovery, and then check to 291 * see whether the host or the device is offline. 292 * 293 * Return value: 294 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 295 */ 296 int scsi_block_when_processing_errors(struct scsi_device *sdev) 297 { 298 int online; 299 300 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 301 302 online = scsi_device_online(sdev); 303 304 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev, 305 "%s: rtn: %d\n", __func__, online)); 306 307 return online; 308 } 309 EXPORT_SYMBOL(scsi_block_when_processing_errors); 310 311 #ifdef CONFIG_SCSI_LOGGING 312 /** 313 * scsi_eh_prt_fail_stats - Log info on failures. 314 * @shost: scsi host being recovered. 315 * @work_q: Queue of scsi cmds to process. 316 */ 317 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 318 struct list_head *work_q) 319 { 320 struct scsi_cmnd *scmd; 321 struct scsi_device *sdev; 322 int total_failures = 0; 323 int cmd_failed = 0; 324 int cmd_cancel = 0; 325 int devices_failed = 0; 326 327 shost_for_each_device(sdev, shost) { 328 list_for_each_entry(scmd, work_q, eh_entry) { 329 if (scmd->device == sdev) { 330 ++total_failures; 331 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 332 ++cmd_cancel; 333 else 334 ++cmd_failed; 335 } 336 } 337 338 if (cmd_cancel || cmd_failed) { 339 SCSI_LOG_ERROR_RECOVERY(3, 340 shost_printk(KERN_INFO, shost, 341 "%s: cmds failed: %d, cancel: %d\n", 342 __func__, cmd_failed, 343 cmd_cancel)); 344 cmd_cancel = 0; 345 cmd_failed = 0; 346 ++devices_failed; 347 } 348 } 349 350 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 351 "Total of %d commands on %d" 352 " devices require eh work\n", 353 total_failures, devices_failed)); 354 } 355 #endif 356 357 /** 358 * scsi_report_lun_change - Set flag on all *other* devices on the same target 359 * to indicate that a UNIT ATTENTION is expected. 360 * @sdev: Device reporting the UNIT ATTENTION 361 */ 362 static void scsi_report_lun_change(struct scsi_device *sdev) 363 { 364 sdev->sdev_target->expecting_lun_change = 1; 365 } 366 367 /** 368 * scsi_report_sense - Examine scsi sense information and log messages for 369 * certain conditions, also issue uevents for some of them. 370 * @sdev: Device reporting the sense code 371 * @sshdr: sshdr to be examined 372 */ 373 static void scsi_report_sense(struct scsi_device *sdev, 374 struct scsi_sense_hdr *sshdr) 375 { 376 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 377 378 if (sshdr->sense_key == UNIT_ATTENTION) { 379 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 380 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 381 sdev_printk(KERN_WARNING, sdev, 382 "Inquiry data has changed"); 383 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 384 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 385 scsi_report_lun_change(sdev); 386 sdev_printk(KERN_WARNING, sdev, 387 "Warning! Received an indication that the " 388 "LUN assignments on this target have " 389 "changed. The Linux SCSI layer does not " 390 "automatically remap LUN assignments.\n"); 391 } else if (sshdr->asc == 0x3f) 392 sdev_printk(KERN_WARNING, sdev, 393 "Warning! Received an indication that the " 394 "operating parameters on this target have " 395 "changed. The Linux SCSI layer does not " 396 "automatically adjust these parameters.\n"); 397 398 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 399 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 400 sdev_printk(KERN_WARNING, sdev, 401 "Warning! Received an indication that the " 402 "LUN reached a thin provisioning soft " 403 "threshold.\n"); 404 } 405 406 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 407 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 408 sdev_printk(KERN_WARNING, sdev, 409 "Mode parameters changed"); 410 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 411 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 412 sdev_printk(KERN_WARNING, sdev, 413 "Asymmetric access state changed"); 414 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 415 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 416 sdev_printk(KERN_WARNING, sdev, 417 "Capacity data has changed"); 418 } else if (sshdr->asc == 0x2a) 419 sdev_printk(KERN_WARNING, sdev, 420 "Parameters changed"); 421 } 422 423 if (evt_type != SDEV_EVT_MAXBITS) { 424 set_bit(evt_type, sdev->pending_events); 425 schedule_work(&sdev->event_work); 426 } 427 } 428 429 /** 430 * scsi_check_sense - Examine scsi cmd sense 431 * @scmd: Cmd to have sense checked. 432 * 433 * Return value: 434 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 435 * 436 * Notes: 437 * When a deferred error is detected the current command has 438 * not been executed and needs retrying. 439 */ 440 int scsi_check_sense(struct scsi_cmnd *scmd) 441 { 442 struct scsi_device *sdev = scmd->device; 443 struct scsi_sense_hdr sshdr; 444 445 if (! scsi_command_normalize_sense(scmd, &sshdr)) 446 return FAILED; /* no valid sense data */ 447 448 scsi_report_sense(sdev, &sshdr); 449 450 if (scsi_sense_is_deferred(&sshdr)) 451 return NEEDS_RETRY; 452 453 if (sdev->handler && sdev->handler->check_sense) { 454 int rc; 455 456 rc = sdev->handler->check_sense(sdev, &sshdr); 457 if (rc != SCSI_RETURN_NOT_HANDLED) 458 return rc; 459 /* handler does not care. Drop down to default handling */ 460 } 461 462 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 463 /* 464 * nasty: for mid-layer issued TURs, we need to return the 465 * actual sense data without any recovery attempt. For eh 466 * issued ones, we need to try to recover and interpret 467 */ 468 return SUCCESS; 469 470 /* 471 * Previous logic looked for FILEMARK, EOM or ILI which are 472 * mainly associated with tapes and returned SUCCESS. 473 */ 474 if (sshdr.response_code == 0x70) { 475 /* fixed format */ 476 if (scmd->sense_buffer[2] & 0xe0) 477 return SUCCESS; 478 } else { 479 /* 480 * descriptor format: look for "stream commands sense data 481 * descriptor" (see SSC-3). Assume single sense data 482 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 483 */ 484 if ((sshdr.additional_length > 3) && 485 (scmd->sense_buffer[8] == 0x4) && 486 (scmd->sense_buffer[11] & 0xe0)) 487 return SUCCESS; 488 } 489 490 switch (sshdr.sense_key) { 491 case NO_SENSE: 492 return SUCCESS; 493 case RECOVERED_ERROR: 494 return /* soft_error */ SUCCESS; 495 496 case ABORTED_COMMAND: 497 if (sshdr.asc == 0x10) /* DIF */ 498 return SUCCESS; 499 500 return NEEDS_RETRY; 501 case NOT_READY: 502 case UNIT_ATTENTION: 503 /* 504 * if we are expecting a cc/ua because of a bus reset that we 505 * performed, treat this just as a retry. otherwise this is 506 * information that we should pass up to the upper-level driver 507 * so that we can deal with it there. 508 */ 509 if (scmd->device->expecting_cc_ua) { 510 /* 511 * Because some device does not queue unit 512 * attentions correctly, we carefully check 513 * additional sense code and qualifier so as 514 * not to squash media change unit attention. 515 */ 516 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 517 scmd->device->expecting_cc_ua = 0; 518 return NEEDS_RETRY; 519 } 520 } 521 /* 522 * we might also expect a cc/ua if another LUN on the target 523 * reported a UA with an ASC/ASCQ of 3F 0E - 524 * REPORTED LUNS DATA HAS CHANGED. 525 */ 526 if (scmd->device->sdev_target->expecting_lun_change && 527 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 528 return NEEDS_RETRY; 529 /* 530 * if the device is in the process of becoming ready, we 531 * should retry. 532 */ 533 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 534 return NEEDS_RETRY; 535 /* 536 * if the device is not started, we need to wake 537 * the error handler to start the motor 538 */ 539 if (scmd->device->allow_restart && 540 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 541 return FAILED; 542 /* 543 * Pass the UA upwards for a determination in the completion 544 * functions. 545 */ 546 return SUCCESS; 547 548 /* these are not supported */ 549 case DATA_PROTECT: 550 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 551 /* Thin provisioning hard threshold reached */ 552 set_host_byte(scmd, DID_ALLOC_FAILURE); 553 return SUCCESS; 554 } 555 /* FALLTHROUGH */ 556 case COPY_ABORTED: 557 case VOLUME_OVERFLOW: 558 case MISCOMPARE: 559 case BLANK_CHECK: 560 set_host_byte(scmd, DID_TARGET_FAILURE); 561 return SUCCESS; 562 563 case MEDIUM_ERROR: 564 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 565 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 566 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 567 set_host_byte(scmd, DID_MEDIUM_ERROR); 568 return SUCCESS; 569 } 570 return NEEDS_RETRY; 571 572 case HARDWARE_ERROR: 573 if (scmd->device->retry_hwerror) 574 return ADD_TO_MLQUEUE; 575 else 576 set_host_byte(scmd, DID_TARGET_FAILURE); 577 /* FALLTHROUGH */ 578 579 case ILLEGAL_REQUEST: 580 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 581 sshdr.asc == 0x21 || /* Logical block address out of range */ 582 sshdr.asc == 0x24 || /* Invalid field in cdb */ 583 sshdr.asc == 0x26) { /* Parameter value invalid */ 584 set_host_byte(scmd, DID_TARGET_FAILURE); 585 } 586 return SUCCESS; 587 588 default: 589 return SUCCESS; 590 } 591 } 592 EXPORT_SYMBOL_GPL(scsi_check_sense); 593 594 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 595 { 596 struct scsi_host_template *sht = sdev->host->hostt; 597 struct scsi_device *tmp_sdev; 598 599 if (!sht->track_queue_depth || 600 sdev->queue_depth >= sdev->max_queue_depth) 601 return; 602 603 if (time_before(jiffies, 604 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 605 return; 606 607 if (time_before(jiffies, 608 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 609 return; 610 611 /* 612 * Walk all devices of a target and do 613 * ramp up on them. 614 */ 615 shost_for_each_device(tmp_sdev, sdev->host) { 616 if (tmp_sdev->channel != sdev->channel || 617 tmp_sdev->id != sdev->id || 618 tmp_sdev->queue_depth == sdev->max_queue_depth) 619 continue; 620 621 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 622 sdev->last_queue_ramp_up = jiffies; 623 } 624 } 625 626 static void scsi_handle_queue_full(struct scsi_device *sdev) 627 { 628 struct scsi_host_template *sht = sdev->host->hostt; 629 struct scsi_device *tmp_sdev; 630 631 if (!sht->track_queue_depth) 632 return; 633 634 shost_for_each_device(tmp_sdev, sdev->host) { 635 if (tmp_sdev->channel != sdev->channel || 636 tmp_sdev->id != sdev->id) 637 continue; 638 /* 639 * We do not know the number of commands that were at 640 * the device when we got the queue full so we start 641 * from the highest possible value and work our way down. 642 */ 643 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 644 } 645 } 646 647 /** 648 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 649 * @scmd: SCSI cmd to examine. 650 * 651 * Notes: 652 * This is *only* called when we are examining the status of commands 653 * queued during error recovery. the main difference here is that we 654 * don't allow for the possibility of retries here, and we are a lot 655 * more restrictive about what we consider acceptable. 656 */ 657 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 658 { 659 /* 660 * first check the host byte, to see if there is anything in there 661 * that would indicate what we need to do. 662 */ 663 if (host_byte(scmd->result) == DID_RESET) { 664 /* 665 * rats. we are already in the error handler, so we now 666 * get to try and figure out what to do next. if the sense 667 * is valid, we have a pretty good idea of what to do. 668 * if not, we mark it as FAILED. 669 */ 670 return scsi_check_sense(scmd); 671 } 672 if (host_byte(scmd->result) != DID_OK) 673 return FAILED; 674 675 /* 676 * next, check the message byte. 677 */ 678 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 679 return FAILED; 680 681 /* 682 * now, check the status byte to see if this indicates 683 * anything special. 684 */ 685 switch (status_byte(scmd->result)) { 686 case GOOD: 687 scsi_handle_queue_ramp_up(scmd->device); 688 /* FALLTHROUGH */ 689 case COMMAND_TERMINATED: 690 return SUCCESS; 691 case CHECK_CONDITION: 692 return scsi_check_sense(scmd); 693 case CONDITION_GOOD: 694 case INTERMEDIATE_GOOD: 695 case INTERMEDIATE_C_GOOD: 696 /* 697 * who knows? FIXME(eric) 698 */ 699 return SUCCESS; 700 case RESERVATION_CONFLICT: 701 if (scmd->cmnd[0] == TEST_UNIT_READY) 702 /* it is a success, we probed the device and 703 * found it */ 704 return SUCCESS; 705 /* otherwise, we failed to send the command */ 706 return FAILED; 707 case QUEUE_FULL: 708 scsi_handle_queue_full(scmd->device); 709 /* fall through */ 710 case BUSY: 711 return NEEDS_RETRY; 712 default: 713 return FAILED; 714 } 715 return FAILED; 716 } 717 718 /** 719 * scsi_eh_done - Completion function for error handling. 720 * @scmd: Cmd that is done. 721 */ 722 static void scsi_eh_done(struct scsi_cmnd *scmd) 723 { 724 struct completion *eh_action; 725 726 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 727 "%s result: %x\n", __func__, scmd->result)); 728 729 eh_action = scmd->device->host->eh_action; 730 if (eh_action) 731 complete(eh_action); 732 } 733 734 /** 735 * scsi_try_host_reset - ask host adapter to reset itself 736 * @scmd: SCSI cmd to send host reset. 737 */ 738 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 739 { 740 unsigned long flags; 741 int rtn; 742 struct Scsi_Host *host = scmd->device->host; 743 struct scsi_host_template *hostt = host->hostt; 744 745 SCSI_LOG_ERROR_RECOVERY(3, 746 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 747 748 if (!hostt->eh_host_reset_handler) 749 return FAILED; 750 751 rtn = hostt->eh_host_reset_handler(scmd); 752 753 if (rtn == SUCCESS) { 754 if (!hostt->skip_settle_delay) 755 ssleep(HOST_RESET_SETTLE_TIME); 756 spin_lock_irqsave(host->host_lock, flags); 757 scsi_report_bus_reset(host, scmd_channel(scmd)); 758 spin_unlock_irqrestore(host->host_lock, flags); 759 } 760 761 return rtn; 762 } 763 764 /** 765 * scsi_try_bus_reset - ask host to perform a bus reset 766 * @scmd: SCSI cmd to send bus reset. 767 */ 768 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 769 { 770 unsigned long flags; 771 int rtn; 772 struct Scsi_Host *host = scmd->device->host; 773 struct scsi_host_template *hostt = host->hostt; 774 775 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 776 "%s: Snd Bus RST\n", __func__)); 777 778 if (!hostt->eh_bus_reset_handler) 779 return FAILED; 780 781 rtn = hostt->eh_bus_reset_handler(scmd); 782 783 if (rtn == SUCCESS) { 784 if (!hostt->skip_settle_delay) 785 ssleep(BUS_RESET_SETTLE_TIME); 786 spin_lock_irqsave(host->host_lock, flags); 787 scsi_report_bus_reset(host, scmd_channel(scmd)); 788 spin_unlock_irqrestore(host->host_lock, flags); 789 } 790 791 return rtn; 792 } 793 794 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 795 { 796 sdev->was_reset = 1; 797 sdev->expecting_cc_ua = 1; 798 } 799 800 /** 801 * scsi_try_target_reset - Ask host to perform a target reset 802 * @scmd: SCSI cmd used to send a target reset 803 * 804 * Notes: 805 * There is no timeout for this operation. if this operation is 806 * unreliable for a given host, then the host itself needs to put a 807 * timer on it, and set the host back to a consistent state prior to 808 * returning. 809 */ 810 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 811 { 812 unsigned long flags; 813 int rtn; 814 struct Scsi_Host *host = scmd->device->host; 815 struct scsi_host_template *hostt = host->hostt; 816 817 if (!hostt->eh_target_reset_handler) 818 return FAILED; 819 820 rtn = hostt->eh_target_reset_handler(scmd); 821 if (rtn == SUCCESS) { 822 spin_lock_irqsave(host->host_lock, flags); 823 __starget_for_each_device(scsi_target(scmd->device), NULL, 824 __scsi_report_device_reset); 825 spin_unlock_irqrestore(host->host_lock, flags); 826 } 827 828 return rtn; 829 } 830 831 /** 832 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 833 * @scmd: SCSI cmd used to send BDR 834 * 835 * Notes: 836 * There is no timeout for this operation. if this operation is 837 * unreliable for a given host, then the host itself needs to put a 838 * timer on it, and set the host back to a consistent state prior to 839 * returning. 840 */ 841 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 842 { 843 int rtn; 844 struct scsi_host_template *hostt = scmd->device->host->hostt; 845 846 if (!hostt->eh_device_reset_handler) 847 return FAILED; 848 849 rtn = hostt->eh_device_reset_handler(scmd); 850 if (rtn == SUCCESS) 851 __scsi_report_device_reset(scmd->device, NULL); 852 return rtn; 853 } 854 855 /** 856 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 857 * @hostt: SCSI driver host template 858 * @scmd: SCSI cmd used to send a target reset 859 * 860 * Return value: 861 * SUCCESS, FAILED, or FAST_IO_FAIL 862 * 863 * Notes: 864 * SUCCESS does not necessarily indicate that the command 865 * has been aborted; it only indicates that the LLDDs 866 * has cleared all references to that command. 867 * LLDDs should return FAILED only if an abort was required 868 * but could not be executed. LLDDs should return FAST_IO_FAIL 869 * if the device is temporarily unavailable (eg due to a 870 * link down on FibreChannel) 871 */ 872 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, 873 struct scsi_cmnd *scmd) 874 { 875 if (!hostt->eh_abort_handler) 876 return FAILED; 877 878 return hostt->eh_abort_handler(scmd); 879 } 880 881 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 882 { 883 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 884 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 885 if (scsi_try_target_reset(scmd) != SUCCESS) 886 if (scsi_try_bus_reset(scmd) != SUCCESS) 887 scsi_try_host_reset(scmd); 888 } 889 890 /** 891 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 892 * @scmd: SCSI command structure to hijack 893 * @ses: structure to save restore information 894 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 895 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 896 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 897 * 898 * This function is used to save a scsi command information before re-execution 899 * as part of the error recovery process. If @sense_bytes is 0 the command 900 * sent must be one that does not transfer any data. If @sense_bytes != 0 901 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 902 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 903 */ 904 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 905 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 906 { 907 struct scsi_device *sdev = scmd->device; 908 909 /* 910 * We need saved copies of a number of fields - this is because 911 * error handling may need to overwrite these with different values 912 * to run different commands, and once error handling is complete, 913 * we will need to restore these values prior to running the actual 914 * command. 915 */ 916 ses->cmd_len = scmd->cmd_len; 917 ses->cmnd = scmd->cmnd; 918 ses->data_direction = scmd->sc_data_direction; 919 ses->sdb = scmd->sdb; 920 ses->next_rq = scmd->request->next_rq; 921 ses->result = scmd->result; 922 ses->underflow = scmd->underflow; 923 ses->prot_op = scmd->prot_op; 924 ses->eh_eflags = scmd->eh_eflags; 925 926 scmd->prot_op = SCSI_PROT_NORMAL; 927 scmd->eh_eflags = 0; 928 scmd->cmnd = ses->eh_cmnd; 929 memset(scmd->cmnd, 0, BLK_MAX_CDB); 930 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 931 scmd->request->next_rq = NULL; 932 scmd->result = 0; 933 934 if (sense_bytes) { 935 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 936 sense_bytes); 937 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 938 scmd->sdb.length); 939 scmd->sdb.table.sgl = &ses->sense_sgl; 940 scmd->sc_data_direction = DMA_FROM_DEVICE; 941 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 942 scmd->cmnd[0] = REQUEST_SENSE; 943 scmd->cmnd[4] = scmd->sdb.length; 944 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 945 } else { 946 scmd->sc_data_direction = DMA_NONE; 947 if (cmnd) { 948 BUG_ON(cmnd_size > BLK_MAX_CDB); 949 memcpy(scmd->cmnd, cmnd, cmnd_size); 950 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 951 } 952 } 953 954 scmd->underflow = 0; 955 956 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 957 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 958 (sdev->lun << 5 & 0xe0); 959 960 /* 961 * Zero the sense buffer. The scsi spec mandates that any 962 * untransferred sense data should be interpreted as being zero. 963 */ 964 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 965 } 966 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 967 968 /** 969 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 970 * @scmd: SCSI command structure to restore 971 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 972 * 973 * Undo any damage done by above scsi_eh_prep_cmnd(). 974 */ 975 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 976 { 977 /* 978 * Restore original data 979 */ 980 scmd->cmd_len = ses->cmd_len; 981 scmd->cmnd = ses->cmnd; 982 scmd->sc_data_direction = ses->data_direction; 983 scmd->sdb = ses->sdb; 984 scmd->request->next_rq = ses->next_rq; 985 scmd->result = ses->result; 986 scmd->underflow = ses->underflow; 987 scmd->prot_op = ses->prot_op; 988 scmd->eh_eflags = ses->eh_eflags; 989 } 990 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 991 992 /** 993 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 994 * @scmd: SCSI command structure to hijack 995 * @cmnd: CDB to send 996 * @cmnd_size: size in bytes of @cmnd 997 * @timeout: timeout for this request 998 * @sense_bytes: size of sense data to copy or 0 999 * 1000 * This function is used to send a scsi command down to a target device 1001 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1002 * 1003 * Return value: 1004 * SUCCESS or FAILED or NEEDS_RETRY 1005 */ 1006 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 1007 int cmnd_size, int timeout, unsigned sense_bytes) 1008 { 1009 struct scsi_device *sdev = scmd->device; 1010 struct Scsi_Host *shost = sdev->host; 1011 DECLARE_COMPLETION_ONSTACK(done); 1012 unsigned long timeleft = timeout; 1013 struct scsi_eh_save ses; 1014 const unsigned long stall_for = msecs_to_jiffies(100); 1015 int rtn; 1016 1017 retry: 1018 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1019 shost->eh_action = &done; 1020 1021 scsi_log_send(scmd); 1022 scmd->scsi_done = scsi_eh_done; 1023 rtn = shost->hostt->queuecommand(shost, scmd); 1024 if (rtn) { 1025 if (timeleft > stall_for) { 1026 scsi_eh_restore_cmnd(scmd, &ses); 1027 timeleft -= stall_for; 1028 msleep(jiffies_to_msecs(stall_for)); 1029 goto retry; 1030 } 1031 /* signal not to enter either branch of the if () below */ 1032 timeleft = 0; 1033 rtn = FAILED; 1034 } else { 1035 timeleft = wait_for_completion_timeout(&done, timeout); 1036 rtn = SUCCESS; 1037 } 1038 1039 shost->eh_action = NULL; 1040 1041 scsi_log_completion(scmd, rtn); 1042 1043 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1044 "%s timeleft: %ld\n", 1045 __func__, timeleft)); 1046 1047 /* 1048 * If there is time left scsi_eh_done got called, and we will examine 1049 * the actual status codes to see whether the command actually did 1050 * complete normally, else if we have a zero return and no time left, 1051 * the command must still be pending, so abort it and return FAILED. 1052 * If we never actually managed to issue the command, because 1053 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1054 * value above (so don't execute either branch of the if) 1055 */ 1056 if (timeleft) { 1057 rtn = scsi_eh_completed_normally(scmd); 1058 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1059 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1060 1061 switch (rtn) { 1062 case SUCCESS: 1063 case NEEDS_RETRY: 1064 case FAILED: 1065 break; 1066 case ADD_TO_MLQUEUE: 1067 rtn = NEEDS_RETRY; 1068 break; 1069 default: 1070 rtn = FAILED; 1071 break; 1072 } 1073 } else if (rtn != FAILED) { 1074 scsi_abort_eh_cmnd(scmd); 1075 rtn = FAILED; 1076 } 1077 1078 scsi_eh_restore_cmnd(scmd, &ses); 1079 1080 return rtn; 1081 } 1082 1083 /** 1084 * scsi_request_sense - Request sense data from a particular target. 1085 * @scmd: SCSI cmd for request sense. 1086 * 1087 * Notes: 1088 * Some hosts automatically obtain this information, others require 1089 * that we obtain it on our own. This function will *not* return until 1090 * the command either times out, or it completes. 1091 */ 1092 static int scsi_request_sense(struct scsi_cmnd *scmd) 1093 { 1094 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1095 } 1096 1097 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) 1098 { 1099 if (!blk_rq_is_passthrough(scmd->request)) { 1100 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1101 if (sdrv->eh_action) 1102 rtn = sdrv->eh_action(scmd, rtn); 1103 } 1104 return rtn; 1105 } 1106 1107 /** 1108 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1109 * @scmd: Original SCSI cmd that eh has finished. 1110 * @done_q: Queue for processed commands. 1111 * 1112 * Notes: 1113 * We don't want to use the normal command completion while we are are 1114 * still handling errors - it may cause other commands to be queued, 1115 * and that would disturb what we are doing. Thus we really want to 1116 * keep a list of pending commands for final completion, and once we 1117 * are ready to leave error handling we handle completion for real. 1118 */ 1119 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1120 { 1121 list_move_tail(&scmd->eh_entry, done_q); 1122 } 1123 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1124 1125 /** 1126 * scsi_eh_get_sense - Get device sense data. 1127 * @work_q: Queue of commands to process. 1128 * @done_q: Queue of processed commands. 1129 * 1130 * Description: 1131 * See if we need to request sense information. if so, then get it 1132 * now, so we have a better idea of what to do. 1133 * 1134 * Notes: 1135 * This has the unfortunate side effect that if a shost adapter does 1136 * not automatically request sense information, we end up shutting 1137 * it down before we request it. 1138 * 1139 * All drivers should request sense information internally these days, 1140 * so for now all I have to say is tough noogies if you end up in here. 1141 * 1142 * XXX: Long term this code should go away, but that needs an audit of 1143 * all LLDDs first. 1144 */ 1145 int scsi_eh_get_sense(struct list_head *work_q, 1146 struct list_head *done_q) 1147 { 1148 struct scsi_cmnd *scmd, *next; 1149 struct Scsi_Host *shost; 1150 int rtn; 1151 1152 /* 1153 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1154 * should not get sense. 1155 */ 1156 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1157 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1158 SCSI_SENSE_VALID(scmd)) 1159 continue; 1160 1161 shost = scmd->device->host; 1162 if (scsi_host_eh_past_deadline(shost)) { 1163 SCSI_LOG_ERROR_RECOVERY(3, 1164 scmd_printk(KERN_INFO, scmd, 1165 "%s: skip request sense, past eh deadline\n", 1166 current->comm)); 1167 break; 1168 } 1169 if (status_byte(scmd->result) != CHECK_CONDITION) 1170 /* 1171 * don't request sense if there's no check condition 1172 * status because the error we're processing isn't one 1173 * that has a sense code (and some devices get 1174 * confused by sense requests out of the blue) 1175 */ 1176 continue; 1177 1178 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1179 "%s: requesting sense\n", 1180 current->comm)); 1181 rtn = scsi_request_sense(scmd); 1182 if (rtn != SUCCESS) 1183 continue; 1184 1185 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1186 "sense requested, result %x\n", scmd->result)); 1187 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1188 1189 rtn = scsi_decide_disposition(scmd); 1190 1191 /* 1192 * if the result was normal, then just pass it along to the 1193 * upper level. 1194 */ 1195 if (rtn == SUCCESS) 1196 /* we don't want this command reissued, just 1197 * finished with the sense data, so set 1198 * retries to the max allowed to ensure it 1199 * won't get reissued */ 1200 scmd->retries = scmd->allowed; 1201 else if (rtn != NEEDS_RETRY) 1202 continue; 1203 1204 scsi_eh_finish_cmd(scmd, done_q); 1205 } 1206 1207 return list_empty(work_q); 1208 } 1209 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1210 1211 /** 1212 * scsi_eh_tur - Send TUR to device. 1213 * @scmd: &scsi_cmnd to send TUR 1214 * 1215 * Return value: 1216 * 0 - Device is ready. 1 - Device NOT ready. 1217 */ 1218 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1219 { 1220 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1221 int retry_cnt = 1, rtn; 1222 1223 retry_tur: 1224 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1225 scmd->device->eh_timeout, 0); 1226 1227 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1228 "%s return: %x\n", __func__, rtn)); 1229 1230 switch (rtn) { 1231 case NEEDS_RETRY: 1232 if (retry_cnt--) 1233 goto retry_tur; 1234 /*FALLTHRU*/ 1235 case SUCCESS: 1236 return 0; 1237 default: 1238 return 1; 1239 } 1240 } 1241 1242 /** 1243 * scsi_eh_test_devices - check if devices are responding from error recovery. 1244 * @cmd_list: scsi commands in error recovery. 1245 * @work_q: queue for commands which still need more error recovery 1246 * @done_q: queue for commands which are finished 1247 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1248 * 1249 * Decription: 1250 * Tests if devices are in a working state. Commands to devices now in 1251 * a working state are sent to the done_q while commands to devices which 1252 * are still failing to respond are returned to the work_q for more 1253 * processing. 1254 **/ 1255 static int scsi_eh_test_devices(struct list_head *cmd_list, 1256 struct list_head *work_q, 1257 struct list_head *done_q, int try_stu) 1258 { 1259 struct scsi_cmnd *scmd, *next; 1260 struct scsi_device *sdev; 1261 int finish_cmds; 1262 1263 while (!list_empty(cmd_list)) { 1264 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1265 sdev = scmd->device; 1266 1267 if (!try_stu) { 1268 if (scsi_host_eh_past_deadline(sdev->host)) { 1269 /* Push items back onto work_q */ 1270 list_splice_init(cmd_list, work_q); 1271 SCSI_LOG_ERROR_RECOVERY(3, 1272 sdev_printk(KERN_INFO, sdev, 1273 "%s: skip test device, past eh deadline", 1274 current->comm)); 1275 break; 1276 } 1277 } 1278 1279 finish_cmds = !scsi_device_online(scmd->device) || 1280 (try_stu && !scsi_eh_try_stu(scmd) && 1281 !scsi_eh_tur(scmd)) || 1282 !scsi_eh_tur(scmd); 1283 1284 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1285 if (scmd->device == sdev) { 1286 if (finish_cmds && 1287 (try_stu || 1288 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1289 scsi_eh_finish_cmd(scmd, done_q); 1290 else 1291 list_move_tail(&scmd->eh_entry, work_q); 1292 } 1293 } 1294 return list_empty(work_q); 1295 } 1296 1297 /** 1298 * scsi_eh_try_stu - Send START_UNIT to device. 1299 * @scmd: &scsi_cmnd to send START_UNIT 1300 * 1301 * Return value: 1302 * 0 - Device is ready. 1 - Device NOT ready. 1303 */ 1304 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1305 { 1306 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1307 1308 if (scmd->device->allow_restart) { 1309 int i, rtn = NEEDS_RETRY; 1310 1311 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1312 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1313 1314 if (rtn == SUCCESS) 1315 return 0; 1316 } 1317 1318 return 1; 1319 } 1320 1321 /** 1322 * scsi_eh_stu - send START_UNIT if needed 1323 * @shost: &scsi host being recovered. 1324 * @work_q: &list_head for pending commands. 1325 * @done_q: &list_head for processed commands. 1326 * 1327 * Notes: 1328 * If commands are failing due to not ready, initializing command required, 1329 * try revalidating the device, which will end up sending a start unit. 1330 */ 1331 static int scsi_eh_stu(struct Scsi_Host *shost, 1332 struct list_head *work_q, 1333 struct list_head *done_q) 1334 { 1335 struct scsi_cmnd *scmd, *stu_scmd, *next; 1336 struct scsi_device *sdev; 1337 1338 shost_for_each_device(sdev, shost) { 1339 if (scsi_host_eh_past_deadline(shost)) { 1340 SCSI_LOG_ERROR_RECOVERY(3, 1341 sdev_printk(KERN_INFO, sdev, 1342 "%s: skip START_UNIT, past eh deadline\n", 1343 current->comm)); 1344 break; 1345 } 1346 stu_scmd = NULL; 1347 list_for_each_entry(scmd, work_q, eh_entry) 1348 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1349 scsi_check_sense(scmd) == FAILED ) { 1350 stu_scmd = scmd; 1351 break; 1352 } 1353 1354 if (!stu_scmd) 1355 continue; 1356 1357 SCSI_LOG_ERROR_RECOVERY(3, 1358 sdev_printk(KERN_INFO, sdev, 1359 "%s: Sending START_UNIT\n", 1360 current->comm)); 1361 1362 if (!scsi_eh_try_stu(stu_scmd)) { 1363 if (!scsi_device_online(sdev) || 1364 !scsi_eh_tur(stu_scmd)) { 1365 list_for_each_entry_safe(scmd, next, 1366 work_q, eh_entry) { 1367 if (scmd->device == sdev && 1368 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1369 scsi_eh_finish_cmd(scmd, done_q); 1370 } 1371 } 1372 } else { 1373 SCSI_LOG_ERROR_RECOVERY(3, 1374 sdev_printk(KERN_INFO, sdev, 1375 "%s: START_UNIT failed\n", 1376 current->comm)); 1377 } 1378 } 1379 1380 return list_empty(work_q); 1381 } 1382 1383 1384 /** 1385 * scsi_eh_bus_device_reset - send bdr if needed 1386 * @shost: scsi host being recovered. 1387 * @work_q: &list_head for pending commands. 1388 * @done_q: &list_head for processed commands. 1389 * 1390 * Notes: 1391 * Try a bus device reset. Still, look to see whether we have multiple 1392 * devices that are jammed or not - if we have multiple devices, it 1393 * makes no sense to try bus_device_reset - we really would need to try 1394 * a bus_reset instead. 1395 */ 1396 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1397 struct list_head *work_q, 1398 struct list_head *done_q) 1399 { 1400 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1401 struct scsi_device *sdev; 1402 int rtn; 1403 1404 shost_for_each_device(sdev, shost) { 1405 if (scsi_host_eh_past_deadline(shost)) { 1406 SCSI_LOG_ERROR_RECOVERY(3, 1407 sdev_printk(KERN_INFO, sdev, 1408 "%s: skip BDR, past eh deadline\n", 1409 current->comm)); 1410 break; 1411 } 1412 bdr_scmd = NULL; 1413 list_for_each_entry(scmd, work_q, eh_entry) 1414 if (scmd->device == sdev) { 1415 bdr_scmd = scmd; 1416 break; 1417 } 1418 1419 if (!bdr_scmd) 1420 continue; 1421 1422 SCSI_LOG_ERROR_RECOVERY(3, 1423 sdev_printk(KERN_INFO, sdev, 1424 "%s: Sending BDR\n", current->comm)); 1425 rtn = scsi_try_bus_device_reset(bdr_scmd); 1426 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1427 if (!scsi_device_online(sdev) || 1428 rtn == FAST_IO_FAIL || 1429 !scsi_eh_tur(bdr_scmd)) { 1430 list_for_each_entry_safe(scmd, next, 1431 work_q, eh_entry) { 1432 if (scmd->device == sdev && 1433 scsi_eh_action(scmd, rtn) != FAILED) 1434 scsi_eh_finish_cmd(scmd, 1435 done_q); 1436 } 1437 } 1438 } else { 1439 SCSI_LOG_ERROR_RECOVERY(3, 1440 sdev_printk(KERN_INFO, sdev, 1441 "%s: BDR failed\n", current->comm)); 1442 } 1443 } 1444 1445 return list_empty(work_q); 1446 } 1447 1448 /** 1449 * scsi_eh_target_reset - send target reset if needed 1450 * @shost: scsi host being recovered. 1451 * @work_q: &list_head for pending commands. 1452 * @done_q: &list_head for processed commands. 1453 * 1454 * Notes: 1455 * Try a target reset. 1456 */ 1457 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1458 struct list_head *work_q, 1459 struct list_head *done_q) 1460 { 1461 LIST_HEAD(tmp_list); 1462 LIST_HEAD(check_list); 1463 1464 list_splice_init(work_q, &tmp_list); 1465 1466 while (!list_empty(&tmp_list)) { 1467 struct scsi_cmnd *next, *scmd; 1468 int rtn; 1469 unsigned int id; 1470 1471 if (scsi_host_eh_past_deadline(shost)) { 1472 /* push back on work queue for further processing */ 1473 list_splice_init(&check_list, work_q); 1474 list_splice_init(&tmp_list, work_q); 1475 SCSI_LOG_ERROR_RECOVERY(3, 1476 shost_printk(KERN_INFO, shost, 1477 "%s: Skip target reset, past eh deadline\n", 1478 current->comm)); 1479 return list_empty(work_q); 1480 } 1481 1482 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1483 id = scmd_id(scmd); 1484 1485 SCSI_LOG_ERROR_RECOVERY(3, 1486 shost_printk(KERN_INFO, shost, 1487 "%s: Sending target reset to target %d\n", 1488 current->comm, id)); 1489 rtn = scsi_try_target_reset(scmd); 1490 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1491 SCSI_LOG_ERROR_RECOVERY(3, 1492 shost_printk(KERN_INFO, shost, 1493 "%s: Target reset failed" 1494 " target: %d\n", 1495 current->comm, id)); 1496 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1497 if (scmd_id(scmd) != id) 1498 continue; 1499 1500 if (rtn == SUCCESS) 1501 list_move_tail(&scmd->eh_entry, &check_list); 1502 else if (rtn == FAST_IO_FAIL) 1503 scsi_eh_finish_cmd(scmd, done_q); 1504 else 1505 /* push back on work queue for further processing */ 1506 list_move(&scmd->eh_entry, work_q); 1507 } 1508 } 1509 1510 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1511 } 1512 1513 /** 1514 * scsi_eh_bus_reset - send a bus reset 1515 * @shost: &scsi host being recovered. 1516 * @work_q: &list_head for pending commands. 1517 * @done_q: &list_head for processed commands. 1518 */ 1519 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1520 struct list_head *work_q, 1521 struct list_head *done_q) 1522 { 1523 struct scsi_cmnd *scmd, *chan_scmd, *next; 1524 LIST_HEAD(check_list); 1525 unsigned int channel; 1526 int rtn; 1527 1528 /* 1529 * we really want to loop over the various channels, and do this on 1530 * a channel by channel basis. we should also check to see if any 1531 * of the failed commands are on soft_reset devices, and if so, skip 1532 * the reset. 1533 */ 1534 1535 for (channel = 0; channel <= shost->max_channel; channel++) { 1536 if (scsi_host_eh_past_deadline(shost)) { 1537 list_splice_init(&check_list, work_q); 1538 SCSI_LOG_ERROR_RECOVERY(3, 1539 shost_printk(KERN_INFO, shost, 1540 "%s: skip BRST, past eh deadline\n", 1541 current->comm)); 1542 return list_empty(work_q); 1543 } 1544 1545 chan_scmd = NULL; 1546 list_for_each_entry(scmd, work_q, eh_entry) { 1547 if (channel == scmd_channel(scmd)) { 1548 chan_scmd = scmd; 1549 break; 1550 /* 1551 * FIXME add back in some support for 1552 * soft_reset devices. 1553 */ 1554 } 1555 } 1556 1557 if (!chan_scmd) 1558 continue; 1559 SCSI_LOG_ERROR_RECOVERY(3, 1560 shost_printk(KERN_INFO, shost, 1561 "%s: Sending BRST chan: %d\n", 1562 current->comm, channel)); 1563 rtn = scsi_try_bus_reset(chan_scmd); 1564 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1565 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1566 if (channel == scmd_channel(scmd)) { 1567 if (rtn == FAST_IO_FAIL) 1568 scsi_eh_finish_cmd(scmd, 1569 done_q); 1570 else 1571 list_move_tail(&scmd->eh_entry, 1572 &check_list); 1573 } 1574 } 1575 } else { 1576 SCSI_LOG_ERROR_RECOVERY(3, 1577 shost_printk(KERN_INFO, shost, 1578 "%s: BRST failed chan: %d\n", 1579 current->comm, channel)); 1580 } 1581 } 1582 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1583 } 1584 1585 /** 1586 * scsi_eh_host_reset - send a host reset 1587 * @shost: host to be reset. 1588 * @work_q: &list_head for pending commands. 1589 * @done_q: &list_head for processed commands. 1590 */ 1591 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1592 struct list_head *work_q, 1593 struct list_head *done_q) 1594 { 1595 struct scsi_cmnd *scmd, *next; 1596 LIST_HEAD(check_list); 1597 int rtn; 1598 1599 if (!list_empty(work_q)) { 1600 scmd = list_entry(work_q->next, 1601 struct scsi_cmnd, eh_entry); 1602 1603 SCSI_LOG_ERROR_RECOVERY(3, 1604 shost_printk(KERN_INFO, shost, 1605 "%s: Sending HRST\n", 1606 current->comm)); 1607 1608 rtn = scsi_try_host_reset(scmd); 1609 if (rtn == SUCCESS) { 1610 list_splice_init(work_q, &check_list); 1611 } else if (rtn == FAST_IO_FAIL) { 1612 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1613 scsi_eh_finish_cmd(scmd, done_q); 1614 } 1615 } else { 1616 SCSI_LOG_ERROR_RECOVERY(3, 1617 shost_printk(KERN_INFO, shost, 1618 "%s: HRST failed\n", 1619 current->comm)); 1620 } 1621 } 1622 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1623 } 1624 1625 /** 1626 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1627 * @work_q: &list_head for pending commands. 1628 * @done_q: &list_head for processed commands. 1629 */ 1630 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1631 struct list_head *done_q) 1632 { 1633 struct scsi_cmnd *scmd, *next; 1634 struct scsi_device *sdev; 1635 1636 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1637 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1638 "not ready after error recovery\n"); 1639 sdev = scmd->device; 1640 1641 mutex_lock(&sdev->state_mutex); 1642 scsi_device_set_state(sdev, SDEV_OFFLINE); 1643 mutex_unlock(&sdev->state_mutex); 1644 1645 scsi_eh_finish_cmd(scmd, done_q); 1646 } 1647 return; 1648 } 1649 1650 /** 1651 * scsi_noretry_cmd - determine if command should be failed fast 1652 * @scmd: SCSI cmd to examine. 1653 */ 1654 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1655 { 1656 switch (host_byte(scmd->result)) { 1657 case DID_OK: 1658 break; 1659 case DID_TIME_OUT: 1660 goto check_type; 1661 case DID_BUS_BUSY: 1662 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1663 case DID_PARITY: 1664 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1665 case DID_ERROR: 1666 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1667 status_byte(scmd->result) == RESERVATION_CONFLICT) 1668 return 0; 1669 /* fall through */ 1670 case DID_SOFT_ERROR: 1671 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1672 } 1673 1674 if (status_byte(scmd->result) != CHECK_CONDITION) 1675 return 0; 1676 1677 check_type: 1678 /* 1679 * assume caller has checked sense and determined 1680 * the check condition was retryable. 1681 */ 1682 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1683 blk_rq_is_passthrough(scmd->request)) 1684 return 1; 1685 else 1686 return 0; 1687 } 1688 1689 /** 1690 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1691 * @scmd: SCSI cmd to examine. 1692 * 1693 * Notes: 1694 * This is *only* called when we are examining the status after sending 1695 * out the actual data command. any commands that are queued for error 1696 * recovery (e.g. test_unit_ready) do *not* come through here. 1697 * 1698 * When this routine returns failed, it means the error handler thread 1699 * is woken. In cases where the error code indicates an error that 1700 * doesn't require the error handler read (i.e. we don't need to 1701 * abort/reset), this function should return SUCCESS. 1702 */ 1703 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1704 { 1705 int rtn; 1706 1707 /* 1708 * if the device is offline, then we clearly just pass the result back 1709 * up to the top level. 1710 */ 1711 if (!scsi_device_online(scmd->device)) { 1712 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1713 "%s: device offline - report as SUCCESS\n", __func__)); 1714 return SUCCESS; 1715 } 1716 1717 /* 1718 * first check the host byte, to see if there is anything in there 1719 * that would indicate what we need to do. 1720 */ 1721 switch (host_byte(scmd->result)) { 1722 case DID_PASSTHROUGH: 1723 /* 1724 * no matter what, pass this through to the upper layer. 1725 * nuke this special code so that it looks like we are saying 1726 * did_ok. 1727 */ 1728 scmd->result &= 0xff00ffff; 1729 return SUCCESS; 1730 case DID_OK: 1731 /* 1732 * looks good. drop through, and check the next byte. 1733 */ 1734 break; 1735 case DID_ABORT: 1736 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1737 set_host_byte(scmd, DID_TIME_OUT); 1738 return SUCCESS; 1739 } 1740 /* FALLTHROUGH */ 1741 case DID_NO_CONNECT: 1742 case DID_BAD_TARGET: 1743 /* 1744 * note - this means that we just report the status back 1745 * to the top level driver, not that we actually think 1746 * that it indicates SUCCESS. 1747 */ 1748 return SUCCESS; 1749 /* 1750 * when the low level driver returns did_soft_error, 1751 * it is responsible for keeping an internal retry counter 1752 * in order to avoid endless loops (db) 1753 * 1754 * actually this is a bug in this function here. we should 1755 * be mindful of the maximum number of retries specified 1756 * and not get stuck in a loop. 1757 */ 1758 case DID_SOFT_ERROR: 1759 goto maybe_retry; 1760 case DID_IMM_RETRY: 1761 return NEEDS_RETRY; 1762 1763 case DID_REQUEUE: 1764 return ADD_TO_MLQUEUE; 1765 case DID_TRANSPORT_DISRUPTED: 1766 /* 1767 * LLD/transport was disrupted during processing of the IO. 1768 * The transport class is now blocked/blocking, 1769 * and the transport will decide what to do with the IO 1770 * based on its timers and recovery capablilities if 1771 * there are enough retries. 1772 */ 1773 goto maybe_retry; 1774 case DID_TRANSPORT_FAILFAST: 1775 /* 1776 * The transport decided to failfast the IO (most likely 1777 * the fast io fail tmo fired), so send IO directly upwards. 1778 */ 1779 return SUCCESS; 1780 case DID_ERROR: 1781 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1782 status_byte(scmd->result) == RESERVATION_CONFLICT) 1783 /* 1784 * execute reservation conflict processing code 1785 * lower down 1786 */ 1787 break; 1788 /* fallthrough */ 1789 case DID_BUS_BUSY: 1790 case DID_PARITY: 1791 goto maybe_retry; 1792 case DID_TIME_OUT: 1793 /* 1794 * when we scan the bus, we get timeout messages for 1795 * these commands if there is no device available. 1796 * other hosts report did_no_connect for the same thing. 1797 */ 1798 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1799 scmd->cmnd[0] == INQUIRY)) { 1800 return SUCCESS; 1801 } else { 1802 return FAILED; 1803 } 1804 case DID_RESET: 1805 return SUCCESS; 1806 default: 1807 return FAILED; 1808 } 1809 1810 /* 1811 * next, check the message byte. 1812 */ 1813 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1814 return FAILED; 1815 1816 /* 1817 * check the status byte to see if this indicates anything special. 1818 */ 1819 switch (status_byte(scmd->result)) { 1820 case QUEUE_FULL: 1821 scsi_handle_queue_full(scmd->device); 1822 /* 1823 * the case of trying to send too many commands to a 1824 * tagged queueing device. 1825 */ 1826 /* FALLTHROUGH */ 1827 case BUSY: 1828 /* 1829 * device can't talk to us at the moment. Should only 1830 * occur (SAM-3) when the task queue is empty, so will cause 1831 * the empty queue handling to trigger a stall in the 1832 * device. 1833 */ 1834 return ADD_TO_MLQUEUE; 1835 case GOOD: 1836 if (scmd->cmnd[0] == REPORT_LUNS) 1837 scmd->device->sdev_target->expecting_lun_change = 0; 1838 scsi_handle_queue_ramp_up(scmd->device); 1839 /* FALLTHROUGH */ 1840 case COMMAND_TERMINATED: 1841 return SUCCESS; 1842 case TASK_ABORTED: 1843 goto maybe_retry; 1844 case CHECK_CONDITION: 1845 rtn = scsi_check_sense(scmd); 1846 if (rtn == NEEDS_RETRY) 1847 goto maybe_retry; 1848 /* if rtn == FAILED, we have no sense information; 1849 * returning FAILED will wake the error handler thread 1850 * to collect the sense and redo the decide 1851 * disposition */ 1852 return rtn; 1853 case CONDITION_GOOD: 1854 case INTERMEDIATE_GOOD: 1855 case INTERMEDIATE_C_GOOD: 1856 case ACA_ACTIVE: 1857 /* 1858 * who knows? FIXME(eric) 1859 */ 1860 return SUCCESS; 1861 1862 case RESERVATION_CONFLICT: 1863 sdev_printk(KERN_INFO, scmd->device, 1864 "reservation conflict\n"); 1865 set_host_byte(scmd, DID_NEXUS_FAILURE); 1866 return SUCCESS; /* causes immediate i/o error */ 1867 default: 1868 return FAILED; 1869 } 1870 return FAILED; 1871 1872 maybe_retry: 1873 1874 /* we requeue for retry because the error was retryable, and 1875 * the request was not marked fast fail. Note that above, 1876 * even if the request is marked fast fail, we still requeue 1877 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1878 if ((++scmd->retries) <= scmd->allowed 1879 && !scsi_noretry_cmd(scmd)) { 1880 return NEEDS_RETRY; 1881 } else { 1882 /* 1883 * no more retries - report this one back to upper level. 1884 */ 1885 return SUCCESS; 1886 } 1887 } 1888 1889 static void eh_lock_door_done(struct request *req, blk_status_t status) 1890 { 1891 __blk_put_request(req->q, req); 1892 } 1893 1894 /** 1895 * scsi_eh_lock_door - Prevent medium removal for the specified device 1896 * @sdev: SCSI device to prevent medium removal 1897 * 1898 * Locking: 1899 * We must be called from process context. 1900 * 1901 * Notes: 1902 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1903 * head of the devices request queue, and continue. 1904 */ 1905 static void scsi_eh_lock_door(struct scsi_device *sdev) 1906 { 1907 struct request *req; 1908 struct scsi_request *rq; 1909 1910 /* 1911 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a 1912 * request becomes available 1913 */ 1914 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL); 1915 if (IS_ERR(req)) 1916 return; 1917 rq = scsi_req(req); 1918 1919 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 1920 rq->cmd[1] = 0; 1921 rq->cmd[2] = 0; 1922 rq->cmd[3] = 0; 1923 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 1924 rq->cmd[5] = 0; 1925 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 1926 1927 req->rq_flags |= RQF_QUIET; 1928 req->timeout = 10 * HZ; 1929 rq->retries = 5; 1930 1931 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); 1932 } 1933 1934 /** 1935 * scsi_restart_operations - restart io operations to the specified host. 1936 * @shost: Host we are restarting. 1937 * 1938 * Notes: 1939 * When we entered the error handler, we blocked all further i/o to 1940 * this device. we need to 'reverse' this process. 1941 */ 1942 static void scsi_restart_operations(struct Scsi_Host *shost) 1943 { 1944 struct scsi_device *sdev; 1945 unsigned long flags; 1946 1947 /* 1948 * If the door was locked, we need to insert a door lock request 1949 * onto the head of the SCSI request queue for the device. There 1950 * is no point trying to lock the door of an off-line device. 1951 */ 1952 shost_for_each_device(sdev, shost) { 1953 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 1954 scsi_eh_lock_door(sdev); 1955 sdev->was_reset = 0; 1956 } 1957 } 1958 1959 /* 1960 * next free up anything directly waiting upon the host. this 1961 * will be requests for character device operations, and also for 1962 * ioctls to queued block devices. 1963 */ 1964 SCSI_LOG_ERROR_RECOVERY(3, 1965 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 1966 1967 spin_lock_irqsave(shost->host_lock, flags); 1968 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1969 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1970 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1971 spin_unlock_irqrestore(shost->host_lock, flags); 1972 1973 wake_up(&shost->host_wait); 1974 1975 /* 1976 * finally we need to re-initiate requests that may be pending. we will 1977 * have had everything blocked while error handling is taking place, and 1978 * now that error recovery is done, we will need to ensure that these 1979 * requests are started. 1980 */ 1981 scsi_run_host_queues(shost); 1982 1983 /* 1984 * if eh is active and host_eh_scheduled is pending we need to re-run 1985 * recovery. we do this check after scsi_run_host_queues() to allow 1986 * everything pent up since the last eh run a chance to make forward 1987 * progress before we sync again. Either we'll immediately re-run 1988 * recovery or scsi_device_unbusy() will wake us again when these 1989 * pending commands complete. 1990 */ 1991 spin_lock_irqsave(shost->host_lock, flags); 1992 if (shost->host_eh_scheduled) 1993 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 1994 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 1995 spin_unlock_irqrestore(shost->host_lock, flags); 1996 } 1997 1998 /** 1999 * scsi_eh_ready_devs - check device ready state and recover if not. 2000 * @shost: host to be recovered. 2001 * @work_q: &list_head for pending commands. 2002 * @done_q: &list_head for processed commands. 2003 */ 2004 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2005 struct list_head *work_q, 2006 struct list_head *done_q) 2007 { 2008 if (!scsi_eh_stu(shost, work_q, done_q)) 2009 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2010 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2011 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2012 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2013 scsi_eh_offline_sdevs(work_q, 2014 done_q); 2015 } 2016 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2017 2018 /** 2019 * scsi_eh_flush_done_q - finish processed commands or retry them. 2020 * @done_q: list_head of processed commands. 2021 */ 2022 void scsi_eh_flush_done_q(struct list_head *done_q) 2023 { 2024 struct scsi_cmnd *scmd, *next; 2025 2026 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2027 list_del_init(&scmd->eh_entry); 2028 if (scsi_device_online(scmd->device) && 2029 !scsi_noretry_cmd(scmd) && 2030 (++scmd->retries <= scmd->allowed)) { 2031 SCSI_LOG_ERROR_RECOVERY(3, 2032 scmd_printk(KERN_INFO, scmd, 2033 "%s: flush retry cmd\n", 2034 current->comm)); 2035 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2036 } else { 2037 /* 2038 * If just we got sense for the device (called 2039 * scsi_eh_get_sense), scmd->result is already 2040 * set, do not set DRIVER_TIMEOUT. 2041 */ 2042 if (!scmd->result) 2043 scmd->result |= (DRIVER_TIMEOUT << 24); 2044 SCSI_LOG_ERROR_RECOVERY(3, 2045 scmd_printk(KERN_INFO, scmd, 2046 "%s: flush finish cmd\n", 2047 current->comm)); 2048 scsi_finish_command(scmd); 2049 } 2050 } 2051 } 2052 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2053 2054 /** 2055 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2056 * @shost: Host to unjam. 2057 * 2058 * Notes: 2059 * When we come in here, we *know* that all commands on the bus have 2060 * either completed, failed or timed out. we also know that no further 2061 * commands are being sent to the host, so things are relatively quiet 2062 * and we have freedom to fiddle with things as we wish. 2063 * 2064 * This is only the *default* implementation. it is possible for 2065 * individual drivers to supply their own version of this function, and 2066 * if the maintainer wishes to do this, it is strongly suggested that 2067 * this function be taken as a template and modified. this function 2068 * was designed to correctly handle problems for about 95% of the 2069 * different cases out there, and it should always provide at least a 2070 * reasonable amount of error recovery. 2071 * 2072 * Any command marked 'failed' or 'timeout' must eventually have 2073 * scsi_finish_cmd() called for it. we do all of the retry stuff 2074 * here, so when we restart the host after we return it should have an 2075 * empty queue. 2076 */ 2077 static void scsi_unjam_host(struct Scsi_Host *shost) 2078 { 2079 unsigned long flags; 2080 LIST_HEAD(eh_work_q); 2081 LIST_HEAD(eh_done_q); 2082 2083 spin_lock_irqsave(shost->host_lock, flags); 2084 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2085 spin_unlock_irqrestore(shost->host_lock, flags); 2086 2087 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2088 2089 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2090 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2091 2092 spin_lock_irqsave(shost->host_lock, flags); 2093 if (shost->eh_deadline != -1) 2094 shost->last_reset = 0; 2095 spin_unlock_irqrestore(shost->host_lock, flags); 2096 scsi_eh_flush_done_q(&eh_done_q); 2097 } 2098 2099 /** 2100 * scsi_error_handler - SCSI error handler thread 2101 * @data: Host for which we are running. 2102 * 2103 * Notes: 2104 * This is the main error handling loop. This is run as a kernel thread 2105 * for every SCSI host and handles all error handling activity. 2106 */ 2107 int scsi_error_handler(void *data) 2108 { 2109 struct Scsi_Host *shost = data; 2110 2111 /* 2112 * We use TASK_INTERRUPTIBLE so that the thread is not 2113 * counted against the load average as a running process. 2114 * We never actually get interrupted because kthread_run 2115 * disables signal delivery for the created thread. 2116 */ 2117 while (true) { 2118 /* 2119 * The sequence in kthread_stop() sets the stop flag first 2120 * then wakes the process. To avoid missed wakeups, the task 2121 * should always be in a non running state before the stop 2122 * flag is checked 2123 */ 2124 set_current_state(TASK_INTERRUPTIBLE); 2125 if (kthread_should_stop()) 2126 break; 2127 2128 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2129 shost->host_failed != atomic_read(&shost->host_busy)) { 2130 SCSI_LOG_ERROR_RECOVERY(1, 2131 shost_printk(KERN_INFO, shost, 2132 "scsi_eh_%d: sleeping\n", 2133 shost->host_no)); 2134 schedule(); 2135 continue; 2136 } 2137 2138 __set_current_state(TASK_RUNNING); 2139 SCSI_LOG_ERROR_RECOVERY(1, 2140 shost_printk(KERN_INFO, shost, 2141 "scsi_eh_%d: waking up %d/%d/%d\n", 2142 shost->host_no, shost->host_eh_scheduled, 2143 shost->host_failed, 2144 atomic_read(&shost->host_busy))); 2145 2146 /* 2147 * We have a host that is failing for some reason. Figure out 2148 * what we need to do to get it up and online again (if we can). 2149 * If we fail, we end up taking the thing offline. 2150 */ 2151 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2152 SCSI_LOG_ERROR_RECOVERY(1, 2153 shost_printk(KERN_ERR, shost, 2154 "scsi_eh_%d: unable to autoresume\n", 2155 shost->host_no)); 2156 continue; 2157 } 2158 2159 if (shost->transportt->eh_strategy_handler) 2160 shost->transportt->eh_strategy_handler(shost); 2161 else 2162 scsi_unjam_host(shost); 2163 2164 /* All scmds have been handled */ 2165 shost->host_failed = 0; 2166 2167 /* 2168 * Note - if the above fails completely, the action is to take 2169 * individual devices offline and flush the queue of any 2170 * outstanding requests that may have been pending. When we 2171 * restart, we restart any I/O to any other devices on the bus 2172 * which are still online. 2173 */ 2174 scsi_restart_operations(shost); 2175 if (!shost->eh_noresume) 2176 scsi_autopm_put_host(shost); 2177 } 2178 __set_current_state(TASK_RUNNING); 2179 2180 SCSI_LOG_ERROR_RECOVERY(1, 2181 shost_printk(KERN_INFO, shost, 2182 "Error handler scsi_eh_%d exiting\n", 2183 shost->host_no)); 2184 shost->ehandler = NULL; 2185 return 0; 2186 } 2187 2188 /* 2189 * Function: scsi_report_bus_reset() 2190 * 2191 * Purpose: Utility function used by low-level drivers to report that 2192 * they have observed a bus reset on the bus being handled. 2193 * 2194 * Arguments: shost - Host in question 2195 * channel - channel on which reset was observed. 2196 * 2197 * Returns: Nothing 2198 * 2199 * Lock status: Host lock must be held. 2200 * 2201 * Notes: This only needs to be called if the reset is one which 2202 * originates from an unknown location. Resets originated 2203 * by the mid-level itself don't need to call this, but there 2204 * should be no harm. 2205 * 2206 * The main purpose of this is to make sure that a CHECK_CONDITION 2207 * is properly treated. 2208 */ 2209 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2210 { 2211 struct scsi_device *sdev; 2212 2213 __shost_for_each_device(sdev, shost) { 2214 if (channel == sdev_channel(sdev)) 2215 __scsi_report_device_reset(sdev, NULL); 2216 } 2217 } 2218 EXPORT_SYMBOL(scsi_report_bus_reset); 2219 2220 /* 2221 * Function: scsi_report_device_reset() 2222 * 2223 * Purpose: Utility function used by low-level drivers to report that 2224 * they have observed a device reset on the device being handled. 2225 * 2226 * Arguments: shost - Host in question 2227 * channel - channel on which reset was observed 2228 * target - target on which reset was observed 2229 * 2230 * Returns: Nothing 2231 * 2232 * Lock status: Host lock must be held 2233 * 2234 * Notes: This only needs to be called if the reset is one which 2235 * originates from an unknown location. Resets originated 2236 * by the mid-level itself don't need to call this, but there 2237 * should be no harm. 2238 * 2239 * The main purpose of this is to make sure that a CHECK_CONDITION 2240 * is properly treated. 2241 */ 2242 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2243 { 2244 struct scsi_device *sdev; 2245 2246 __shost_for_each_device(sdev, shost) { 2247 if (channel == sdev_channel(sdev) && 2248 target == sdev_id(sdev)) 2249 __scsi_report_device_reset(sdev, NULL); 2250 } 2251 } 2252 EXPORT_SYMBOL(scsi_report_device_reset); 2253 2254 static void 2255 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2256 { 2257 } 2258 2259 /** 2260 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2261 * @dev: scsi_device to operate on 2262 * @arg: reset type (see sg.h) 2263 */ 2264 int 2265 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2266 { 2267 struct scsi_cmnd *scmd; 2268 struct Scsi_Host *shost = dev->host; 2269 struct request *rq; 2270 unsigned long flags; 2271 int error = 0, rtn, val; 2272 2273 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2274 return -EACCES; 2275 2276 error = get_user(val, arg); 2277 if (error) 2278 return error; 2279 2280 if (scsi_autopm_get_host(shost) < 0) 2281 return -EIO; 2282 2283 error = -EIO; 2284 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2285 shost->hostt->cmd_size, GFP_KERNEL); 2286 if (!rq) 2287 goto out_put_autopm_host; 2288 blk_rq_init(NULL, rq); 2289 2290 scmd = (struct scsi_cmnd *)(rq + 1); 2291 scsi_init_command(dev, scmd); 2292 scmd->request = rq; 2293 scmd->cmnd = scsi_req(rq)->cmd; 2294 2295 scmd->scsi_done = scsi_reset_provider_done_command; 2296 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2297 2298 scmd->cmd_len = 0; 2299 2300 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2301 2302 spin_lock_irqsave(shost->host_lock, flags); 2303 shost->tmf_in_progress = 1; 2304 spin_unlock_irqrestore(shost->host_lock, flags); 2305 2306 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2307 case SG_SCSI_RESET_NOTHING: 2308 rtn = SUCCESS; 2309 break; 2310 case SG_SCSI_RESET_DEVICE: 2311 rtn = scsi_try_bus_device_reset(scmd); 2312 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2313 break; 2314 /* FALLTHROUGH */ 2315 case SG_SCSI_RESET_TARGET: 2316 rtn = scsi_try_target_reset(scmd); 2317 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2318 break; 2319 /* FALLTHROUGH */ 2320 case SG_SCSI_RESET_BUS: 2321 rtn = scsi_try_bus_reset(scmd); 2322 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2323 break; 2324 /* FALLTHROUGH */ 2325 case SG_SCSI_RESET_HOST: 2326 rtn = scsi_try_host_reset(scmd); 2327 if (rtn == SUCCESS) 2328 break; 2329 /* FALLTHROUGH */ 2330 default: 2331 rtn = FAILED; 2332 break; 2333 } 2334 2335 error = (rtn == SUCCESS) ? 0 : -EIO; 2336 2337 spin_lock_irqsave(shost->host_lock, flags); 2338 shost->tmf_in_progress = 0; 2339 spin_unlock_irqrestore(shost->host_lock, flags); 2340 2341 /* 2342 * be sure to wake up anyone who was sleeping or had their queue 2343 * suspended while we performed the TMF. 2344 */ 2345 SCSI_LOG_ERROR_RECOVERY(3, 2346 shost_printk(KERN_INFO, shost, 2347 "waking up host to restart after TMF\n")); 2348 2349 wake_up(&shost->host_wait); 2350 scsi_run_host_queues(shost); 2351 2352 scsi_put_command(scmd); 2353 kfree(rq); 2354 2355 out_put_autopm_host: 2356 scsi_autopm_put_host(shost); 2357 return error; 2358 } 2359 EXPORT_SYMBOL(scsi_ioctl_reset); 2360 2361 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2362 struct scsi_sense_hdr *sshdr) 2363 { 2364 return scsi_normalize_sense(cmd->sense_buffer, 2365 SCSI_SENSE_BUFFERSIZE, sshdr); 2366 } 2367 EXPORT_SYMBOL(scsi_command_normalize_sense); 2368 2369 /** 2370 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2371 * @sense_buffer: byte array of sense data 2372 * @sb_len: number of valid bytes in sense_buffer 2373 * @info_out: pointer to 64 integer where 8 or 4 byte information 2374 * field will be placed if found. 2375 * 2376 * Return value: 2377 * true if information field found, false if not found. 2378 */ 2379 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2380 u64 *info_out) 2381 { 2382 const u8 * ucp; 2383 2384 if (sb_len < 7) 2385 return false; 2386 switch (sense_buffer[0] & 0x7f) { 2387 case 0x70: 2388 case 0x71: 2389 if (sense_buffer[0] & 0x80) { 2390 *info_out = get_unaligned_be32(&sense_buffer[3]); 2391 return true; 2392 } 2393 return false; 2394 case 0x72: 2395 case 0x73: 2396 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2397 0 /* info desc */); 2398 if (ucp && (0xa == ucp[1])) { 2399 *info_out = get_unaligned_be64(&ucp[4]); 2400 return true; 2401 } 2402 return false; 2403 default: 2404 return false; 2405 } 2406 } 2407 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2408