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