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 == 0x29) { 407 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 408 sdev_printk(KERN_WARNING, sdev, 409 "Power-on or device reset occurred\n"); 410 } 411 412 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 413 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 414 sdev_printk(KERN_WARNING, sdev, 415 "Mode parameters changed"); 416 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 417 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 418 sdev_printk(KERN_WARNING, sdev, 419 "Asymmetric access state changed"); 420 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 421 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 422 sdev_printk(KERN_WARNING, sdev, 423 "Capacity data has changed"); 424 } else if (sshdr->asc == 0x2a) 425 sdev_printk(KERN_WARNING, sdev, 426 "Parameters changed"); 427 } 428 429 if (evt_type != SDEV_EVT_MAXBITS) { 430 set_bit(evt_type, sdev->pending_events); 431 schedule_work(&sdev->event_work); 432 } 433 } 434 435 /** 436 * scsi_check_sense - Examine scsi cmd sense 437 * @scmd: Cmd to have sense checked. 438 * 439 * Return value: 440 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 441 * 442 * Notes: 443 * When a deferred error is detected the current command has 444 * not been executed and needs retrying. 445 */ 446 int scsi_check_sense(struct scsi_cmnd *scmd) 447 { 448 struct scsi_device *sdev = scmd->device; 449 struct scsi_sense_hdr sshdr; 450 451 if (! scsi_command_normalize_sense(scmd, &sshdr)) 452 return FAILED; /* no valid sense data */ 453 454 scsi_report_sense(sdev, &sshdr); 455 456 if (scsi_sense_is_deferred(&sshdr)) 457 return NEEDS_RETRY; 458 459 if (sdev->handler && sdev->handler->check_sense) { 460 int rc; 461 462 rc = sdev->handler->check_sense(sdev, &sshdr); 463 if (rc != SCSI_RETURN_NOT_HANDLED) 464 return rc; 465 /* handler does not care. Drop down to default handling */ 466 } 467 468 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 469 /* 470 * nasty: for mid-layer issued TURs, we need to return the 471 * actual sense data without any recovery attempt. For eh 472 * issued ones, we need to try to recover and interpret 473 */ 474 return SUCCESS; 475 476 /* 477 * Previous logic looked for FILEMARK, EOM or ILI which are 478 * mainly associated with tapes and returned SUCCESS. 479 */ 480 if (sshdr.response_code == 0x70) { 481 /* fixed format */ 482 if (scmd->sense_buffer[2] & 0xe0) 483 return SUCCESS; 484 } else { 485 /* 486 * descriptor format: look for "stream commands sense data 487 * descriptor" (see SSC-3). Assume single sense data 488 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 489 */ 490 if ((sshdr.additional_length > 3) && 491 (scmd->sense_buffer[8] == 0x4) && 492 (scmd->sense_buffer[11] & 0xe0)) 493 return SUCCESS; 494 } 495 496 switch (sshdr.sense_key) { 497 case NO_SENSE: 498 return SUCCESS; 499 case RECOVERED_ERROR: 500 return /* soft_error */ SUCCESS; 501 502 case ABORTED_COMMAND: 503 if (sshdr.asc == 0x10) /* DIF */ 504 return SUCCESS; 505 506 return NEEDS_RETRY; 507 case NOT_READY: 508 case UNIT_ATTENTION: 509 /* 510 * if we are expecting a cc/ua because of a bus reset that we 511 * performed, treat this just as a retry. otherwise this is 512 * information that we should pass up to the upper-level driver 513 * so that we can deal with it there. 514 */ 515 if (scmd->device->expecting_cc_ua) { 516 /* 517 * Because some device does not queue unit 518 * attentions correctly, we carefully check 519 * additional sense code and qualifier so as 520 * not to squash media change unit attention. 521 */ 522 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 523 scmd->device->expecting_cc_ua = 0; 524 return NEEDS_RETRY; 525 } 526 } 527 /* 528 * we might also expect a cc/ua if another LUN on the target 529 * reported a UA with an ASC/ASCQ of 3F 0E - 530 * REPORTED LUNS DATA HAS CHANGED. 531 */ 532 if (scmd->device->sdev_target->expecting_lun_change && 533 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 534 return NEEDS_RETRY; 535 /* 536 * if the device is in the process of becoming ready, we 537 * should retry. 538 */ 539 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 540 return NEEDS_RETRY; 541 /* 542 * if the device is not started, we need to wake 543 * the error handler to start the motor 544 */ 545 if (scmd->device->allow_restart && 546 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 547 return FAILED; 548 /* 549 * Pass the UA upwards for a determination in the completion 550 * functions. 551 */ 552 return SUCCESS; 553 554 /* these are not supported */ 555 case DATA_PROTECT: 556 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 557 /* Thin provisioning hard threshold reached */ 558 set_host_byte(scmd, DID_ALLOC_FAILURE); 559 return SUCCESS; 560 } 561 /* FALLTHROUGH */ 562 case COPY_ABORTED: 563 case VOLUME_OVERFLOW: 564 case MISCOMPARE: 565 case BLANK_CHECK: 566 set_host_byte(scmd, DID_TARGET_FAILURE); 567 return SUCCESS; 568 569 case MEDIUM_ERROR: 570 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 571 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 572 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 573 set_host_byte(scmd, DID_MEDIUM_ERROR); 574 return SUCCESS; 575 } 576 return NEEDS_RETRY; 577 578 case HARDWARE_ERROR: 579 if (scmd->device->retry_hwerror) 580 return ADD_TO_MLQUEUE; 581 else 582 set_host_byte(scmd, DID_TARGET_FAILURE); 583 /* FALLTHROUGH */ 584 585 case ILLEGAL_REQUEST: 586 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 587 sshdr.asc == 0x21 || /* Logical block address out of range */ 588 sshdr.asc == 0x22 || /* Invalid function */ 589 sshdr.asc == 0x24 || /* Invalid field in cdb */ 590 sshdr.asc == 0x26 || /* Parameter value invalid */ 591 sshdr.asc == 0x27) { /* Write protected */ 592 set_host_byte(scmd, DID_TARGET_FAILURE); 593 } 594 return SUCCESS; 595 596 default: 597 return SUCCESS; 598 } 599 } 600 EXPORT_SYMBOL_GPL(scsi_check_sense); 601 602 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 603 { 604 struct scsi_host_template *sht = sdev->host->hostt; 605 struct scsi_device *tmp_sdev; 606 607 if (!sht->track_queue_depth || 608 sdev->queue_depth >= sdev->max_queue_depth) 609 return; 610 611 if (time_before(jiffies, 612 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 613 return; 614 615 if (time_before(jiffies, 616 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 617 return; 618 619 /* 620 * Walk all devices of a target and do 621 * ramp up on them. 622 */ 623 shost_for_each_device(tmp_sdev, sdev->host) { 624 if (tmp_sdev->channel != sdev->channel || 625 tmp_sdev->id != sdev->id || 626 tmp_sdev->queue_depth == sdev->max_queue_depth) 627 continue; 628 629 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 630 sdev->last_queue_ramp_up = jiffies; 631 } 632 } 633 634 static void scsi_handle_queue_full(struct scsi_device *sdev) 635 { 636 struct scsi_host_template *sht = sdev->host->hostt; 637 struct scsi_device *tmp_sdev; 638 639 if (!sht->track_queue_depth) 640 return; 641 642 shost_for_each_device(tmp_sdev, sdev->host) { 643 if (tmp_sdev->channel != sdev->channel || 644 tmp_sdev->id != sdev->id) 645 continue; 646 /* 647 * We do not know the number of commands that were at 648 * the device when we got the queue full so we start 649 * from the highest possible value and work our way down. 650 */ 651 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 652 } 653 } 654 655 /** 656 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 657 * @scmd: SCSI cmd to examine. 658 * 659 * Notes: 660 * This is *only* called when we are examining the status of commands 661 * queued during error recovery. the main difference here is that we 662 * don't allow for the possibility of retries here, and we are a lot 663 * more restrictive about what we consider acceptable. 664 */ 665 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 666 { 667 /* 668 * first check the host byte, to see if there is anything in there 669 * that would indicate what we need to do. 670 */ 671 if (host_byte(scmd->result) == DID_RESET) { 672 /* 673 * rats. we are already in the error handler, so we now 674 * get to try and figure out what to do next. if the sense 675 * is valid, we have a pretty good idea of what to do. 676 * if not, we mark it as FAILED. 677 */ 678 return scsi_check_sense(scmd); 679 } 680 if (host_byte(scmd->result) != DID_OK) 681 return FAILED; 682 683 /* 684 * next, check the message byte. 685 */ 686 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 687 return FAILED; 688 689 /* 690 * now, check the status byte to see if this indicates 691 * anything special. 692 */ 693 switch (status_byte(scmd->result)) { 694 case GOOD: 695 scsi_handle_queue_ramp_up(scmd->device); 696 /* FALLTHROUGH */ 697 case COMMAND_TERMINATED: 698 return SUCCESS; 699 case CHECK_CONDITION: 700 return scsi_check_sense(scmd); 701 case CONDITION_GOOD: 702 case INTERMEDIATE_GOOD: 703 case INTERMEDIATE_C_GOOD: 704 /* 705 * who knows? FIXME(eric) 706 */ 707 return SUCCESS; 708 case RESERVATION_CONFLICT: 709 if (scmd->cmnd[0] == TEST_UNIT_READY) 710 /* it is a success, we probed the device and 711 * found it */ 712 return SUCCESS; 713 /* otherwise, we failed to send the command */ 714 return FAILED; 715 case QUEUE_FULL: 716 scsi_handle_queue_full(scmd->device); 717 /* fall through */ 718 case BUSY: 719 return NEEDS_RETRY; 720 default: 721 return FAILED; 722 } 723 return FAILED; 724 } 725 726 /** 727 * scsi_eh_done - Completion function for error handling. 728 * @scmd: Cmd that is done. 729 */ 730 static void scsi_eh_done(struct scsi_cmnd *scmd) 731 { 732 struct completion *eh_action; 733 734 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 735 "%s result: %x\n", __func__, scmd->result)); 736 737 eh_action = scmd->device->host->eh_action; 738 if (eh_action) 739 complete(eh_action); 740 } 741 742 /** 743 * scsi_try_host_reset - ask host adapter to reset itself 744 * @scmd: SCSI cmd to send host reset. 745 */ 746 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 747 { 748 unsigned long flags; 749 int rtn; 750 struct Scsi_Host *host = scmd->device->host; 751 struct scsi_host_template *hostt = host->hostt; 752 753 SCSI_LOG_ERROR_RECOVERY(3, 754 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 755 756 if (!hostt->eh_host_reset_handler) 757 return FAILED; 758 759 rtn = hostt->eh_host_reset_handler(scmd); 760 761 if (rtn == SUCCESS) { 762 if (!hostt->skip_settle_delay) 763 ssleep(HOST_RESET_SETTLE_TIME); 764 spin_lock_irqsave(host->host_lock, flags); 765 scsi_report_bus_reset(host, scmd_channel(scmd)); 766 spin_unlock_irqrestore(host->host_lock, flags); 767 } 768 769 return rtn; 770 } 771 772 /** 773 * scsi_try_bus_reset - ask host to perform a bus reset 774 * @scmd: SCSI cmd to send bus reset. 775 */ 776 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 777 { 778 unsigned long flags; 779 int rtn; 780 struct Scsi_Host *host = scmd->device->host; 781 struct scsi_host_template *hostt = host->hostt; 782 783 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 784 "%s: Snd Bus RST\n", __func__)); 785 786 if (!hostt->eh_bus_reset_handler) 787 return FAILED; 788 789 rtn = hostt->eh_bus_reset_handler(scmd); 790 791 if (rtn == SUCCESS) { 792 if (!hostt->skip_settle_delay) 793 ssleep(BUS_RESET_SETTLE_TIME); 794 spin_lock_irqsave(host->host_lock, flags); 795 scsi_report_bus_reset(host, scmd_channel(scmd)); 796 spin_unlock_irqrestore(host->host_lock, flags); 797 } 798 799 return rtn; 800 } 801 802 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 803 { 804 sdev->was_reset = 1; 805 sdev->expecting_cc_ua = 1; 806 } 807 808 /** 809 * scsi_try_target_reset - Ask host to perform a target reset 810 * @scmd: SCSI cmd used to send a target reset 811 * 812 * Notes: 813 * There is no timeout for this operation. if this operation is 814 * unreliable for a given host, then the host itself needs to put a 815 * timer on it, and set the host back to a consistent state prior to 816 * returning. 817 */ 818 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 819 { 820 unsigned long flags; 821 int rtn; 822 struct Scsi_Host *host = scmd->device->host; 823 struct scsi_host_template *hostt = host->hostt; 824 825 if (!hostt->eh_target_reset_handler) 826 return FAILED; 827 828 rtn = hostt->eh_target_reset_handler(scmd); 829 if (rtn == SUCCESS) { 830 spin_lock_irqsave(host->host_lock, flags); 831 __starget_for_each_device(scsi_target(scmd->device), NULL, 832 __scsi_report_device_reset); 833 spin_unlock_irqrestore(host->host_lock, flags); 834 } 835 836 return rtn; 837 } 838 839 /** 840 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 841 * @scmd: SCSI cmd used to send BDR 842 * 843 * Notes: 844 * There is no timeout for this operation. if this operation is 845 * unreliable for a given host, then the host itself needs to put a 846 * timer on it, and set the host back to a consistent state prior to 847 * returning. 848 */ 849 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 850 { 851 int rtn; 852 struct scsi_host_template *hostt = scmd->device->host->hostt; 853 854 if (!hostt->eh_device_reset_handler) 855 return FAILED; 856 857 rtn = hostt->eh_device_reset_handler(scmd); 858 if (rtn == SUCCESS) 859 __scsi_report_device_reset(scmd->device, NULL); 860 return rtn; 861 } 862 863 /** 864 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 865 * @hostt: SCSI driver host template 866 * @scmd: SCSI cmd used to send a target reset 867 * 868 * Return value: 869 * SUCCESS, FAILED, or FAST_IO_FAIL 870 * 871 * Notes: 872 * SUCCESS does not necessarily indicate that the command 873 * has been aborted; it only indicates that the LLDDs 874 * has cleared all references to that command. 875 * LLDDs should return FAILED only if an abort was required 876 * but could not be executed. LLDDs should return FAST_IO_FAIL 877 * if the device is temporarily unavailable (eg due to a 878 * link down on FibreChannel) 879 */ 880 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, 881 struct scsi_cmnd *scmd) 882 { 883 if (!hostt->eh_abort_handler) 884 return FAILED; 885 886 return hostt->eh_abort_handler(scmd); 887 } 888 889 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 890 { 891 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 892 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 893 if (scsi_try_target_reset(scmd) != SUCCESS) 894 if (scsi_try_bus_reset(scmd) != SUCCESS) 895 scsi_try_host_reset(scmd); 896 } 897 898 /** 899 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 900 * @scmd: SCSI command structure to hijack 901 * @ses: structure to save restore information 902 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 903 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 904 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 905 * 906 * This function is used to save a scsi command information before re-execution 907 * as part of the error recovery process. If @sense_bytes is 0 the command 908 * sent must be one that does not transfer any data. If @sense_bytes != 0 909 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 910 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 911 */ 912 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 913 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 914 { 915 struct scsi_device *sdev = scmd->device; 916 917 /* 918 * We need saved copies of a number of fields - this is because 919 * error handling may need to overwrite these with different values 920 * to run different commands, and once error handling is complete, 921 * we will need to restore these values prior to running the actual 922 * command. 923 */ 924 ses->cmd_len = scmd->cmd_len; 925 ses->cmnd = scmd->cmnd; 926 ses->data_direction = scmd->sc_data_direction; 927 ses->sdb = scmd->sdb; 928 ses->next_rq = scmd->request->next_rq; 929 ses->result = scmd->result; 930 ses->underflow = scmd->underflow; 931 ses->prot_op = scmd->prot_op; 932 ses->eh_eflags = scmd->eh_eflags; 933 934 scmd->prot_op = SCSI_PROT_NORMAL; 935 scmd->eh_eflags = 0; 936 scmd->cmnd = ses->eh_cmnd; 937 memset(scmd->cmnd, 0, BLK_MAX_CDB); 938 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 939 scmd->request->next_rq = NULL; 940 scmd->result = 0; 941 942 if (sense_bytes) { 943 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 944 sense_bytes); 945 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 946 scmd->sdb.length); 947 scmd->sdb.table.sgl = &ses->sense_sgl; 948 scmd->sc_data_direction = DMA_FROM_DEVICE; 949 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 950 scmd->cmnd[0] = REQUEST_SENSE; 951 scmd->cmnd[4] = scmd->sdb.length; 952 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 953 } else { 954 scmd->sc_data_direction = DMA_NONE; 955 if (cmnd) { 956 BUG_ON(cmnd_size > BLK_MAX_CDB); 957 memcpy(scmd->cmnd, cmnd, cmnd_size); 958 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 959 } 960 } 961 962 scmd->underflow = 0; 963 964 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 965 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 966 (sdev->lun << 5 & 0xe0); 967 968 /* 969 * Zero the sense buffer. The scsi spec mandates that any 970 * untransferred sense data should be interpreted as being zero. 971 */ 972 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 973 } 974 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 975 976 /** 977 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 978 * @scmd: SCSI command structure to restore 979 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 980 * 981 * Undo any damage done by above scsi_eh_prep_cmnd(). 982 */ 983 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 984 { 985 /* 986 * Restore original data 987 */ 988 scmd->cmd_len = ses->cmd_len; 989 scmd->cmnd = ses->cmnd; 990 scmd->sc_data_direction = ses->data_direction; 991 scmd->sdb = ses->sdb; 992 scmd->request->next_rq = ses->next_rq; 993 scmd->result = ses->result; 994 scmd->underflow = ses->underflow; 995 scmd->prot_op = ses->prot_op; 996 scmd->eh_eflags = ses->eh_eflags; 997 } 998 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 999 1000 /** 1001 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1002 * @scmd: SCSI command structure to hijack 1003 * @cmnd: CDB to send 1004 * @cmnd_size: size in bytes of @cmnd 1005 * @timeout: timeout for this request 1006 * @sense_bytes: size of sense data to copy or 0 1007 * 1008 * This function is used to send a scsi command down to a target device 1009 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1010 * 1011 * Return value: 1012 * SUCCESS or FAILED or NEEDS_RETRY 1013 */ 1014 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 1015 int cmnd_size, int timeout, unsigned sense_bytes) 1016 { 1017 struct scsi_device *sdev = scmd->device; 1018 struct Scsi_Host *shost = sdev->host; 1019 DECLARE_COMPLETION_ONSTACK(done); 1020 unsigned long timeleft = timeout; 1021 struct scsi_eh_save ses; 1022 const unsigned long stall_for = msecs_to_jiffies(100); 1023 int rtn; 1024 1025 retry: 1026 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1027 shost->eh_action = &done; 1028 1029 scsi_log_send(scmd); 1030 scmd->scsi_done = scsi_eh_done; 1031 rtn = shost->hostt->queuecommand(shost, scmd); 1032 if (rtn) { 1033 if (timeleft > stall_for) { 1034 scsi_eh_restore_cmnd(scmd, &ses); 1035 timeleft -= stall_for; 1036 msleep(jiffies_to_msecs(stall_for)); 1037 goto retry; 1038 } 1039 /* signal not to enter either branch of the if () below */ 1040 timeleft = 0; 1041 rtn = FAILED; 1042 } else { 1043 timeleft = wait_for_completion_timeout(&done, timeout); 1044 rtn = SUCCESS; 1045 } 1046 1047 shost->eh_action = NULL; 1048 1049 scsi_log_completion(scmd, rtn); 1050 1051 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1052 "%s timeleft: %ld\n", 1053 __func__, timeleft)); 1054 1055 /* 1056 * If there is time left scsi_eh_done got called, and we will examine 1057 * the actual status codes to see whether the command actually did 1058 * complete normally, else if we have a zero return and no time left, 1059 * the command must still be pending, so abort it and return FAILED. 1060 * If we never actually managed to issue the command, because 1061 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1062 * value above (so don't execute either branch of the if) 1063 */ 1064 if (timeleft) { 1065 rtn = scsi_eh_completed_normally(scmd); 1066 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1067 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1068 1069 switch (rtn) { 1070 case SUCCESS: 1071 case NEEDS_RETRY: 1072 case FAILED: 1073 break; 1074 case ADD_TO_MLQUEUE: 1075 rtn = NEEDS_RETRY; 1076 break; 1077 default: 1078 rtn = FAILED; 1079 break; 1080 } 1081 } else if (rtn != FAILED) { 1082 scsi_abort_eh_cmnd(scmd); 1083 rtn = FAILED; 1084 } 1085 1086 scsi_eh_restore_cmnd(scmd, &ses); 1087 1088 return rtn; 1089 } 1090 1091 /** 1092 * scsi_request_sense - Request sense data from a particular target. 1093 * @scmd: SCSI cmd for request sense. 1094 * 1095 * Notes: 1096 * Some hosts automatically obtain this information, others require 1097 * that we obtain it on our own. This function will *not* return until 1098 * the command either times out, or it completes. 1099 */ 1100 static int scsi_request_sense(struct scsi_cmnd *scmd) 1101 { 1102 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1103 } 1104 1105 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) 1106 { 1107 if (!blk_rq_is_passthrough(scmd->request)) { 1108 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1109 if (sdrv->eh_action) 1110 rtn = sdrv->eh_action(scmd, rtn); 1111 } 1112 return rtn; 1113 } 1114 1115 /** 1116 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1117 * @scmd: Original SCSI cmd that eh has finished. 1118 * @done_q: Queue for processed commands. 1119 * 1120 * Notes: 1121 * We don't want to use the normal command completion while we are are 1122 * still handling errors - it may cause other commands to be queued, 1123 * and that would disturb what we are doing. Thus we really want to 1124 * keep a list of pending commands for final completion, and once we 1125 * are ready to leave error handling we handle completion for real. 1126 */ 1127 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1128 { 1129 list_move_tail(&scmd->eh_entry, done_q); 1130 } 1131 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1132 1133 /** 1134 * scsi_eh_get_sense - Get device sense data. 1135 * @work_q: Queue of commands to process. 1136 * @done_q: Queue of processed commands. 1137 * 1138 * Description: 1139 * See if we need to request sense information. if so, then get it 1140 * now, so we have a better idea of what to do. 1141 * 1142 * Notes: 1143 * This has the unfortunate side effect that if a shost adapter does 1144 * not automatically request sense information, we end up shutting 1145 * it down before we request it. 1146 * 1147 * All drivers should request sense information internally these days, 1148 * so for now all I have to say is tough noogies if you end up in here. 1149 * 1150 * XXX: Long term this code should go away, but that needs an audit of 1151 * all LLDDs first. 1152 */ 1153 int scsi_eh_get_sense(struct list_head *work_q, 1154 struct list_head *done_q) 1155 { 1156 struct scsi_cmnd *scmd, *next; 1157 struct Scsi_Host *shost; 1158 int rtn; 1159 1160 /* 1161 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1162 * should not get sense. 1163 */ 1164 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1165 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1166 SCSI_SENSE_VALID(scmd)) 1167 continue; 1168 1169 shost = scmd->device->host; 1170 if (scsi_host_eh_past_deadline(shost)) { 1171 SCSI_LOG_ERROR_RECOVERY(3, 1172 scmd_printk(KERN_INFO, scmd, 1173 "%s: skip request sense, past eh deadline\n", 1174 current->comm)); 1175 break; 1176 } 1177 if (status_byte(scmd->result) != CHECK_CONDITION) 1178 /* 1179 * don't request sense if there's no check condition 1180 * status because the error we're processing isn't one 1181 * that has a sense code (and some devices get 1182 * confused by sense requests out of the blue) 1183 */ 1184 continue; 1185 1186 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1187 "%s: requesting sense\n", 1188 current->comm)); 1189 rtn = scsi_request_sense(scmd); 1190 if (rtn != SUCCESS) 1191 continue; 1192 1193 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1194 "sense requested, result %x\n", scmd->result)); 1195 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1196 1197 rtn = scsi_decide_disposition(scmd); 1198 1199 /* 1200 * if the result was normal, then just pass it along to the 1201 * upper level. 1202 */ 1203 if (rtn == SUCCESS) 1204 /* we don't want this command reissued, just 1205 * finished with the sense data, so set 1206 * retries to the max allowed to ensure it 1207 * won't get reissued */ 1208 scmd->retries = scmd->allowed; 1209 else if (rtn != NEEDS_RETRY) 1210 continue; 1211 1212 scsi_eh_finish_cmd(scmd, done_q); 1213 } 1214 1215 return list_empty(work_q); 1216 } 1217 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1218 1219 /** 1220 * scsi_eh_tur - Send TUR to device. 1221 * @scmd: &scsi_cmnd to send TUR 1222 * 1223 * Return value: 1224 * 0 - Device is ready. 1 - Device NOT ready. 1225 */ 1226 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1227 { 1228 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1229 int retry_cnt = 1, rtn; 1230 1231 retry_tur: 1232 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1233 scmd->device->eh_timeout, 0); 1234 1235 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1236 "%s return: %x\n", __func__, rtn)); 1237 1238 switch (rtn) { 1239 case NEEDS_RETRY: 1240 if (retry_cnt--) 1241 goto retry_tur; 1242 /*FALLTHRU*/ 1243 case SUCCESS: 1244 return 0; 1245 default: 1246 return 1; 1247 } 1248 } 1249 1250 /** 1251 * scsi_eh_test_devices - check if devices are responding from error recovery. 1252 * @cmd_list: scsi commands in error recovery. 1253 * @work_q: queue for commands which still need more error recovery 1254 * @done_q: queue for commands which are finished 1255 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1256 * 1257 * Decription: 1258 * Tests if devices are in a working state. Commands to devices now in 1259 * a working state are sent to the done_q while commands to devices which 1260 * are still failing to respond are returned to the work_q for more 1261 * processing. 1262 **/ 1263 static int scsi_eh_test_devices(struct list_head *cmd_list, 1264 struct list_head *work_q, 1265 struct list_head *done_q, int try_stu) 1266 { 1267 struct scsi_cmnd *scmd, *next; 1268 struct scsi_device *sdev; 1269 int finish_cmds; 1270 1271 while (!list_empty(cmd_list)) { 1272 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1273 sdev = scmd->device; 1274 1275 if (!try_stu) { 1276 if (scsi_host_eh_past_deadline(sdev->host)) { 1277 /* Push items back onto work_q */ 1278 list_splice_init(cmd_list, work_q); 1279 SCSI_LOG_ERROR_RECOVERY(3, 1280 sdev_printk(KERN_INFO, sdev, 1281 "%s: skip test device, past eh deadline", 1282 current->comm)); 1283 break; 1284 } 1285 } 1286 1287 finish_cmds = !scsi_device_online(scmd->device) || 1288 (try_stu && !scsi_eh_try_stu(scmd) && 1289 !scsi_eh_tur(scmd)) || 1290 !scsi_eh_tur(scmd); 1291 1292 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1293 if (scmd->device == sdev) { 1294 if (finish_cmds && 1295 (try_stu || 1296 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1297 scsi_eh_finish_cmd(scmd, done_q); 1298 else 1299 list_move_tail(&scmd->eh_entry, work_q); 1300 } 1301 } 1302 return list_empty(work_q); 1303 } 1304 1305 /** 1306 * scsi_eh_try_stu - Send START_UNIT to device. 1307 * @scmd: &scsi_cmnd to send START_UNIT 1308 * 1309 * Return value: 1310 * 0 - Device is ready. 1 - Device NOT ready. 1311 */ 1312 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1313 { 1314 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1315 1316 if (scmd->device->allow_restart) { 1317 int i, rtn = NEEDS_RETRY; 1318 1319 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1320 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1321 1322 if (rtn == SUCCESS) 1323 return 0; 1324 } 1325 1326 return 1; 1327 } 1328 1329 /** 1330 * scsi_eh_stu - send START_UNIT if needed 1331 * @shost: &scsi host being recovered. 1332 * @work_q: &list_head for pending commands. 1333 * @done_q: &list_head for processed commands. 1334 * 1335 * Notes: 1336 * If commands are failing due to not ready, initializing command required, 1337 * try revalidating the device, which will end up sending a start unit. 1338 */ 1339 static int scsi_eh_stu(struct Scsi_Host *shost, 1340 struct list_head *work_q, 1341 struct list_head *done_q) 1342 { 1343 struct scsi_cmnd *scmd, *stu_scmd, *next; 1344 struct scsi_device *sdev; 1345 1346 shost_for_each_device(sdev, shost) { 1347 if (scsi_host_eh_past_deadline(shost)) { 1348 SCSI_LOG_ERROR_RECOVERY(3, 1349 sdev_printk(KERN_INFO, sdev, 1350 "%s: skip START_UNIT, past eh deadline\n", 1351 current->comm)); 1352 break; 1353 } 1354 stu_scmd = NULL; 1355 list_for_each_entry(scmd, work_q, eh_entry) 1356 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1357 scsi_check_sense(scmd) == FAILED ) { 1358 stu_scmd = scmd; 1359 break; 1360 } 1361 1362 if (!stu_scmd) 1363 continue; 1364 1365 SCSI_LOG_ERROR_RECOVERY(3, 1366 sdev_printk(KERN_INFO, sdev, 1367 "%s: Sending START_UNIT\n", 1368 current->comm)); 1369 1370 if (!scsi_eh_try_stu(stu_scmd)) { 1371 if (!scsi_device_online(sdev) || 1372 !scsi_eh_tur(stu_scmd)) { 1373 list_for_each_entry_safe(scmd, next, 1374 work_q, eh_entry) { 1375 if (scmd->device == sdev && 1376 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1377 scsi_eh_finish_cmd(scmd, done_q); 1378 } 1379 } 1380 } else { 1381 SCSI_LOG_ERROR_RECOVERY(3, 1382 sdev_printk(KERN_INFO, sdev, 1383 "%s: START_UNIT failed\n", 1384 current->comm)); 1385 } 1386 } 1387 1388 return list_empty(work_q); 1389 } 1390 1391 1392 /** 1393 * scsi_eh_bus_device_reset - send bdr if needed 1394 * @shost: scsi host being recovered. 1395 * @work_q: &list_head for pending commands. 1396 * @done_q: &list_head for processed commands. 1397 * 1398 * Notes: 1399 * Try a bus device reset. Still, look to see whether we have multiple 1400 * devices that are jammed or not - if we have multiple devices, it 1401 * makes no sense to try bus_device_reset - we really would need to try 1402 * a bus_reset instead. 1403 */ 1404 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1405 struct list_head *work_q, 1406 struct list_head *done_q) 1407 { 1408 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1409 struct scsi_device *sdev; 1410 int rtn; 1411 1412 shost_for_each_device(sdev, shost) { 1413 if (scsi_host_eh_past_deadline(shost)) { 1414 SCSI_LOG_ERROR_RECOVERY(3, 1415 sdev_printk(KERN_INFO, sdev, 1416 "%s: skip BDR, past eh deadline\n", 1417 current->comm)); 1418 break; 1419 } 1420 bdr_scmd = NULL; 1421 list_for_each_entry(scmd, work_q, eh_entry) 1422 if (scmd->device == sdev) { 1423 bdr_scmd = scmd; 1424 break; 1425 } 1426 1427 if (!bdr_scmd) 1428 continue; 1429 1430 SCSI_LOG_ERROR_RECOVERY(3, 1431 sdev_printk(KERN_INFO, sdev, 1432 "%s: Sending BDR\n", current->comm)); 1433 rtn = scsi_try_bus_device_reset(bdr_scmd); 1434 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1435 if (!scsi_device_online(sdev) || 1436 rtn == FAST_IO_FAIL || 1437 !scsi_eh_tur(bdr_scmd)) { 1438 list_for_each_entry_safe(scmd, next, 1439 work_q, eh_entry) { 1440 if (scmd->device == sdev && 1441 scsi_eh_action(scmd, rtn) != FAILED) 1442 scsi_eh_finish_cmd(scmd, 1443 done_q); 1444 } 1445 } 1446 } else { 1447 SCSI_LOG_ERROR_RECOVERY(3, 1448 sdev_printk(KERN_INFO, sdev, 1449 "%s: BDR failed\n", current->comm)); 1450 } 1451 } 1452 1453 return list_empty(work_q); 1454 } 1455 1456 /** 1457 * scsi_eh_target_reset - send target reset if needed 1458 * @shost: scsi host being recovered. 1459 * @work_q: &list_head for pending commands. 1460 * @done_q: &list_head for processed commands. 1461 * 1462 * Notes: 1463 * Try a target reset. 1464 */ 1465 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1466 struct list_head *work_q, 1467 struct list_head *done_q) 1468 { 1469 LIST_HEAD(tmp_list); 1470 LIST_HEAD(check_list); 1471 1472 list_splice_init(work_q, &tmp_list); 1473 1474 while (!list_empty(&tmp_list)) { 1475 struct scsi_cmnd *next, *scmd; 1476 int rtn; 1477 unsigned int id; 1478 1479 if (scsi_host_eh_past_deadline(shost)) { 1480 /* push back on work queue for further processing */ 1481 list_splice_init(&check_list, work_q); 1482 list_splice_init(&tmp_list, work_q); 1483 SCSI_LOG_ERROR_RECOVERY(3, 1484 shost_printk(KERN_INFO, shost, 1485 "%s: Skip target reset, past eh deadline\n", 1486 current->comm)); 1487 return list_empty(work_q); 1488 } 1489 1490 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1491 id = scmd_id(scmd); 1492 1493 SCSI_LOG_ERROR_RECOVERY(3, 1494 shost_printk(KERN_INFO, shost, 1495 "%s: Sending target reset to target %d\n", 1496 current->comm, id)); 1497 rtn = scsi_try_target_reset(scmd); 1498 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1499 SCSI_LOG_ERROR_RECOVERY(3, 1500 shost_printk(KERN_INFO, shost, 1501 "%s: Target reset failed" 1502 " target: %d\n", 1503 current->comm, id)); 1504 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1505 if (scmd_id(scmd) != id) 1506 continue; 1507 1508 if (rtn == SUCCESS) 1509 list_move_tail(&scmd->eh_entry, &check_list); 1510 else if (rtn == FAST_IO_FAIL) 1511 scsi_eh_finish_cmd(scmd, done_q); 1512 else 1513 /* push back on work queue for further processing */ 1514 list_move(&scmd->eh_entry, work_q); 1515 } 1516 } 1517 1518 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1519 } 1520 1521 /** 1522 * scsi_eh_bus_reset - send a bus reset 1523 * @shost: &scsi host being recovered. 1524 * @work_q: &list_head for pending commands. 1525 * @done_q: &list_head for processed commands. 1526 */ 1527 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1528 struct list_head *work_q, 1529 struct list_head *done_q) 1530 { 1531 struct scsi_cmnd *scmd, *chan_scmd, *next; 1532 LIST_HEAD(check_list); 1533 unsigned int channel; 1534 int rtn; 1535 1536 /* 1537 * we really want to loop over the various channels, and do this on 1538 * a channel by channel basis. we should also check to see if any 1539 * of the failed commands are on soft_reset devices, and if so, skip 1540 * the reset. 1541 */ 1542 1543 for (channel = 0; channel <= shost->max_channel; channel++) { 1544 if (scsi_host_eh_past_deadline(shost)) { 1545 list_splice_init(&check_list, work_q); 1546 SCSI_LOG_ERROR_RECOVERY(3, 1547 shost_printk(KERN_INFO, shost, 1548 "%s: skip BRST, past eh deadline\n", 1549 current->comm)); 1550 return list_empty(work_q); 1551 } 1552 1553 chan_scmd = NULL; 1554 list_for_each_entry(scmd, work_q, eh_entry) { 1555 if (channel == scmd_channel(scmd)) { 1556 chan_scmd = scmd; 1557 break; 1558 /* 1559 * FIXME add back in some support for 1560 * soft_reset devices. 1561 */ 1562 } 1563 } 1564 1565 if (!chan_scmd) 1566 continue; 1567 SCSI_LOG_ERROR_RECOVERY(3, 1568 shost_printk(KERN_INFO, shost, 1569 "%s: Sending BRST chan: %d\n", 1570 current->comm, channel)); 1571 rtn = scsi_try_bus_reset(chan_scmd); 1572 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1573 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1574 if (channel == scmd_channel(scmd)) { 1575 if (rtn == FAST_IO_FAIL) 1576 scsi_eh_finish_cmd(scmd, 1577 done_q); 1578 else 1579 list_move_tail(&scmd->eh_entry, 1580 &check_list); 1581 } 1582 } 1583 } else { 1584 SCSI_LOG_ERROR_RECOVERY(3, 1585 shost_printk(KERN_INFO, shost, 1586 "%s: BRST failed chan: %d\n", 1587 current->comm, channel)); 1588 } 1589 } 1590 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1591 } 1592 1593 /** 1594 * scsi_eh_host_reset - send a host reset 1595 * @shost: host to be reset. 1596 * @work_q: &list_head for pending commands. 1597 * @done_q: &list_head for processed commands. 1598 */ 1599 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1600 struct list_head *work_q, 1601 struct list_head *done_q) 1602 { 1603 struct scsi_cmnd *scmd, *next; 1604 LIST_HEAD(check_list); 1605 int rtn; 1606 1607 if (!list_empty(work_q)) { 1608 scmd = list_entry(work_q->next, 1609 struct scsi_cmnd, eh_entry); 1610 1611 SCSI_LOG_ERROR_RECOVERY(3, 1612 shost_printk(KERN_INFO, shost, 1613 "%s: Sending HRST\n", 1614 current->comm)); 1615 1616 rtn = scsi_try_host_reset(scmd); 1617 if (rtn == SUCCESS) { 1618 list_splice_init(work_q, &check_list); 1619 } else if (rtn == FAST_IO_FAIL) { 1620 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1621 scsi_eh_finish_cmd(scmd, done_q); 1622 } 1623 } else { 1624 SCSI_LOG_ERROR_RECOVERY(3, 1625 shost_printk(KERN_INFO, shost, 1626 "%s: HRST failed\n", 1627 current->comm)); 1628 } 1629 } 1630 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1631 } 1632 1633 /** 1634 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1635 * @work_q: &list_head for pending commands. 1636 * @done_q: &list_head for processed commands. 1637 */ 1638 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1639 struct list_head *done_q) 1640 { 1641 struct scsi_cmnd *scmd, *next; 1642 struct scsi_device *sdev; 1643 1644 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1645 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1646 "not ready after error recovery\n"); 1647 sdev = scmd->device; 1648 1649 mutex_lock(&sdev->state_mutex); 1650 scsi_device_set_state(sdev, SDEV_OFFLINE); 1651 mutex_unlock(&sdev->state_mutex); 1652 1653 scsi_eh_finish_cmd(scmd, done_q); 1654 } 1655 return; 1656 } 1657 1658 /** 1659 * scsi_noretry_cmd - determine if command should be failed fast 1660 * @scmd: SCSI cmd to examine. 1661 */ 1662 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1663 { 1664 switch (host_byte(scmd->result)) { 1665 case DID_OK: 1666 break; 1667 case DID_TIME_OUT: 1668 goto check_type; 1669 case DID_BUS_BUSY: 1670 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1671 case DID_PARITY: 1672 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1673 case DID_ERROR: 1674 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1675 status_byte(scmd->result) == RESERVATION_CONFLICT) 1676 return 0; 1677 /* fall through */ 1678 case DID_SOFT_ERROR: 1679 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1680 } 1681 1682 if (status_byte(scmd->result) != CHECK_CONDITION) 1683 return 0; 1684 1685 check_type: 1686 /* 1687 * assume caller has checked sense and determined 1688 * the check condition was retryable. 1689 */ 1690 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1691 blk_rq_is_passthrough(scmd->request)) 1692 return 1; 1693 else 1694 return 0; 1695 } 1696 1697 /** 1698 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1699 * @scmd: SCSI cmd to examine. 1700 * 1701 * Notes: 1702 * This is *only* called when we are examining the status after sending 1703 * out the actual data command. any commands that are queued for error 1704 * recovery (e.g. test_unit_ready) do *not* come through here. 1705 * 1706 * When this routine returns failed, it means the error handler thread 1707 * is woken. In cases where the error code indicates an error that 1708 * doesn't require the error handler read (i.e. we don't need to 1709 * abort/reset), this function should return SUCCESS. 1710 */ 1711 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1712 { 1713 int rtn; 1714 1715 /* 1716 * if the device is offline, then we clearly just pass the result back 1717 * up to the top level. 1718 */ 1719 if (!scsi_device_online(scmd->device)) { 1720 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1721 "%s: device offline - report as SUCCESS\n", __func__)); 1722 return SUCCESS; 1723 } 1724 1725 /* 1726 * first check the host byte, to see if there is anything in there 1727 * that would indicate what we need to do. 1728 */ 1729 switch (host_byte(scmd->result)) { 1730 case DID_PASSTHROUGH: 1731 /* 1732 * no matter what, pass this through to the upper layer. 1733 * nuke this special code so that it looks like we are saying 1734 * did_ok. 1735 */ 1736 scmd->result &= 0xff00ffff; 1737 return SUCCESS; 1738 case DID_OK: 1739 /* 1740 * looks good. drop through, and check the next byte. 1741 */ 1742 break; 1743 case DID_ABORT: 1744 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1745 set_host_byte(scmd, DID_TIME_OUT); 1746 return SUCCESS; 1747 } 1748 /* FALLTHROUGH */ 1749 case DID_NO_CONNECT: 1750 case DID_BAD_TARGET: 1751 /* 1752 * note - this means that we just report the status back 1753 * to the top level driver, not that we actually think 1754 * that it indicates SUCCESS. 1755 */ 1756 return SUCCESS; 1757 case DID_SOFT_ERROR: 1758 /* 1759 * when the low level driver returns did_soft_error, 1760 * it is responsible for keeping an internal retry counter 1761 * in order to avoid endless loops (db) 1762 */ 1763 goto maybe_retry; 1764 case DID_IMM_RETRY: 1765 return NEEDS_RETRY; 1766 1767 case DID_REQUEUE: 1768 return ADD_TO_MLQUEUE; 1769 case DID_TRANSPORT_DISRUPTED: 1770 /* 1771 * LLD/transport was disrupted during processing of the IO. 1772 * The transport class is now blocked/blocking, 1773 * and the transport will decide what to do with the IO 1774 * based on its timers and recovery capablilities if 1775 * there are enough retries. 1776 */ 1777 goto maybe_retry; 1778 case DID_TRANSPORT_FAILFAST: 1779 /* 1780 * The transport decided to failfast the IO (most likely 1781 * the fast io fail tmo fired), so send IO directly upwards. 1782 */ 1783 return SUCCESS; 1784 case DID_ERROR: 1785 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1786 status_byte(scmd->result) == RESERVATION_CONFLICT) 1787 /* 1788 * execute reservation conflict processing code 1789 * lower down 1790 */ 1791 break; 1792 /* fallthrough */ 1793 case DID_BUS_BUSY: 1794 case DID_PARITY: 1795 goto maybe_retry; 1796 case DID_TIME_OUT: 1797 /* 1798 * when we scan the bus, we get timeout messages for 1799 * these commands if there is no device available. 1800 * other hosts report did_no_connect for the same thing. 1801 */ 1802 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1803 scmd->cmnd[0] == INQUIRY)) { 1804 return SUCCESS; 1805 } else { 1806 return FAILED; 1807 } 1808 case DID_RESET: 1809 return SUCCESS; 1810 default: 1811 return FAILED; 1812 } 1813 1814 /* 1815 * next, check the message byte. 1816 */ 1817 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1818 return FAILED; 1819 1820 /* 1821 * check the status byte to see if this indicates anything special. 1822 */ 1823 switch (status_byte(scmd->result)) { 1824 case QUEUE_FULL: 1825 scsi_handle_queue_full(scmd->device); 1826 /* 1827 * the case of trying to send too many commands to a 1828 * tagged queueing device. 1829 */ 1830 /* FALLTHROUGH */ 1831 case BUSY: 1832 /* 1833 * device can't talk to us at the moment. Should only 1834 * occur (SAM-3) when the task queue is empty, so will cause 1835 * the empty queue handling to trigger a stall in the 1836 * device. 1837 */ 1838 return ADD_TO_MLQUEUE; 1839 case GOOD: 1840 if (scmd->cmnd[0] == REPORT_LUNS) 1841 scmd->device->sdev_target->expecting_lun_change = 0; 1842 scsi_handle_queue_ramp_up(scmd->device); 1843 /* FALLTHROUGH */ 1844 case COMMAND_TERMINATED: 1845 return SUCCESS; 1846 case TASK_ABORTED: 1847 goto maybe_retry; 1848 case CHECK_CONDITION: 1849 rtn = scsi_check_sense(scmd); 1850 if (rtn == NEEDS_RETRY) 1851 goto maybe_retry; 1852 /* if rtn == FAILED, we have no sense information; 1853 * returning FAILED will wake the error handler thread 1854 * to collect the sense and redo the decide 1855 * disposition */ 1856 return rtn; 1857 case CONDITION_GOOD: 1858 case INTERMEDIATE_GOOD: 1859 case INTERMEDIATE_C_GOOD: 1860 case ACA_ACTIVE: 1861 /* 1862 * who knows? FIXME(eric) 1863 */ 1864 return SUCCESS; 1865 1866 case RESERVATION_CONFLICT: 1867 sdev_printk(KERN_INFO, scmd->device, 1868 "reservation conflict\n"); 1869 set_host_byte(scmd, DID_NEXUS_FAILURE); 1870 return SUCCESS; /* causes immediate i/o error */ 1871 default: 1872 return FAILED; 1873 } 1874 return FAILED; 1875 1876 maybe_retry: 1877 1878 /* we requeue for retry because the error was retryable, and 1879 * the request was not marked fast fail. Note that above, 1880 * even if the request is marked fast fail, we still requeue 1881 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1882 if ((++scmd->retries) <= scmd->allowed 1883 && !scsi_noretry_cmd(scmd)) { 1884 return NEEDS_RETRY; 1885 } else { 1886 /* 1887 * no more retries - report this one back to upper level. 1888 */ 1889 return SUCCESS; 1890 } 1891 } 1892 1893 static void eh_lock_door_done(struct request *req, blk_status_t status) 1894 { 1895 __blk_put_request(req->q, req); 1896 } 1897 1898 /** 1899 * scsi_eh_lock_door - Prevent medium removal for the specified device 1900 * @sdev: SCSI device to prevent medium removal 1901 * 1902 * Locking: 1903 * We must be called from process context. 1904 * 1905 * Notes: 1906 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1907 * head of the devices request queue, and continue. 1908 */ 1909 static void scsi_eh_lock_door(struct scsi_device *sdev) 1910 { 1911 struct request *req; 1912 struct scsi_request *rq; 1913 1914 /* 1915 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a 1916 * request becomes available 1917 */ 1918 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL); 1919 if (IS_ERR(req)) 1920 return; 1921 rq = scsi_req(req); 1922 1923 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 1924 rq->cmd[1] = 0; 1925 rq->cmd[2] = 0; 1926 rq->cmd[3] = 0; 1927 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 1928 rq->cmd[5] = 0; 1929 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 1930 1931 req->rq_flags |= RQF_QUIET; 1932 req->timeout = 10 * HZ; 1933 rq->retries = 5; 1934 1935 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); 1936 } 1937 1938 /** 1939 * scsi_restart_operations - restart io operations to the specified host. 1940 * @shost: Host we are restarting. 1941 * 1942 * Notes: 1943 * When we entered the error handler, we blocked all further i/o to 1944 * this device. we need to 'reverse' this process. 1945 */ 1946 static void scsi_restart_operations(struct Scsi_Host *shost) 1947 { 1948 struct scsi_device *sdev; 1949 unsigned long flags; 1950 1951 /* 1952 * If the door was locked, we need to insert a door lock request 1953 * onto the head of the SCSI request queue for the device. There 1954 * is no point trying to lock the door of an off-line device. 1955 */ 1956 shost_for_each_device(sdev, shost) { 1957 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 1958 scsi_eh_lock_door(sdev); 1959 sdev->was_reset = 0; 1960 } 1961 } 1962 1963 /* 1964 * next free up anything directly waiting upon the host. this 1965 * will be requests for character device operations, and also for 1966 * ioctls to queued block devices. 1967 */ 1968 SCSI_LOG_ERROR_RECOVERY(3, 1969 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 1970 1971 spin_lock_irqsave(shost->host_lock, flags); 1972 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1973 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1974 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1975 spin_unlock_irqrestore(shost->host_lock, flags); 1976 1977 wake_up(&shost->host_wait); 1978 1979 /* 1980 * finally we need to re-initiate requests that may be pending. we will 1981 * have had everything blocked while error handling is taking place, and 1982 * now that error recovery is done, we will need to ensure that these 1983 * requests are started. 1984 */ 1985 scsi_run_host_queues(shost); 1986 1987 /* 1988 * if eh is active and host_eh_scheduled is pending we need to re-run 1989 * recovery. we do this check after scsi_run_host_queues() to allow 1990 * everything pent up since the last eh run a chance to make forward 1991 * progress before we sync again. Either we'll immediately re-run 1992 * recovery or scsi_device_unbusy() will wake us again when these 1993 * pending commands complete. 1994 */ 1995 spin_lock_irqsave(shost->host_lock, flags); 1996 if (shost->host_eh_scheduled) 1997 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 1998 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 1999 spin_unlock_irqrestore(shost->host_lock, flags); 2000 } 2001 2002 /** 2003 * scsi_eh_ready_devs - check device ready state and recover if not. 2004 * @shost: host to be recovered. 2005 * @work_q: &list_head for pending commands. 2006 * @done_q: &list_head for processed commands. 2007 */ 2008 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2009 struct list_head *work_q, 2010 struct list_head *done_q) 2011 { 2012 if (!scsi_eh_stu(shost, work_q, done_q)) 2013 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2014 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2015 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2016 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2017 scsi_eh_offline_sdevs(work_q, 2018 done_q); 2019 } 2020 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2021 2022 /** 2023 * scsi_eh_flush_done_q - finish processed commands or retry them. 2024 * @done_q: list_head of processed commands. 2025 */ 2026 void scsi_eh_flush_done_q(struct list_head *done_q) 2027 { 2028 struct scsi_cmnd *scmd, *next; 2029 2030 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2031 list_del_init(&scmd->eh_entry); 2032 if (scsi_device_online(scmd->device) && 2033 !scsi_noretry_cmd(scmd) && 2034 (++scmd->retries <= scmd->allowed)) { 2035 SCSI_LOG_ERROR_RECOVERY(3, 2036 scmd_printk(KERN_INFO, scmd, 2037 "%s: flush retry cmd\n", 2038 current->comm)); 2039 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2040 } else { 2041 /* 2042 * If just we got sense for the device (called 2043 * scsi_eh_get_sense), scmd->result is already 2044 * set, do not set DRIVER_TIMEOUT. 2045 */ 2046 if (!scmd->result) 2047 scmd->result |= (DRIVER_TIMEOUT << 24); 2048 SCSI_LOG_ERROR_RECOVERY(3, 2049 scmd_printk(KERN_INFO, scmd, 2050 "%s: flush finish cmd\n", 2051 current->comm)); 2052 scsi_finish_command(scmd); 2053 } 2054 } 2055 } 2056 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2057 2058 /** 2059 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2060 * @shost: Host to unjam. 2061 * 2062 * Notes: 2063 * When we come in here, we *know* that all commands on the bus have 2064 * either completed, failed or timed out. we also know that no further 2065 * commands are being sent to the host, so things are relatively quiet 2066 * and we have freedom to fiddle with things as we wish. 2067 * 2068 * This is only the *default* implementation. it is possible for 2069 * individual drivers to supply their own version of this function, and 2070 * if the maintainer wishes to do this, it is strongly suggested that 2071 * this function be taken as a template and modified. this function 2072 * was designed to correctly handle problems for about 95% of the 2073 * different cases out there, and it should always provide at least a 2074 * reasonable amount of error recovery. 2075 * 2076 * Any command marked 'failed' or 'timeout' must eventually have 2077 * scsi_finish_cmd() called for it. we do all of the retry stuff 2078 * here, so when we restart the host after we return it should have an 2079 * empty queue. 2080 */ 2081 static void scsi_unjam_host(struct Scsi_Host *shost) 2082 { 2083 unsigned long flags; 2084 LIST_HEAD(eh_work_q); 2085 LIST_HEAD(eh_done_q); 2086 2087 spin_lock_irqsave(shost->host_lock, flags); 2088 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2089 spin_unlock_irqrestore(shost->host_lock, flags); 2090 2091 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2092 2093 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2094 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2095 2096 spin_lock_irqsave(shost->host_lock, flags); 2097 if (shost->eh_deadline != -1) 2098 shost->last_reset = 0; 2099 spin_unlock_irqrestore(shost->host_lock, flags); 2100 scsi_eh_flush_done_q(&eh_done_q); 2101 } 2102 2103 /** 2104 * scsi_error_handler - SCSI error handler thread 2105 * @data: Host for which we are running. 2106 * 2107 * Notes: 2108 * This is the main error handling loop. This is run as a kernel thread 2109 * for every SCSI host and handles all error handling activity. 2110 */ 2111 int scsi_error_handler(void *data) 2112 { 2113 struct Scsi_Host *shost = data; 2114 2115 /* 2116 * We use TASK_INTERRUPTIBLE so that the thread is not 2117 * counted against the load average as a running process. 2118 * We never actually get interrupted because kthread_run 2119 * disables signal delivery for the created thread. 2120 */ 2121 while (true) { 2122 /* 2123 * The sequence in kthread_stop() sets the stop flag first 2124 * then wakes the process. To avoid missed wakeups, the task 2125 * should always be in a non running state before the stop 2126 * flag is checked 2127 */ 2128 set_current_state(TASK_INTERRUPTIBLE); 2129 if (kthread_should_stop()) 2130 break; 2131 2132 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2133 shost->host_failed != atomic_read(&shost->host_busy)) { 2134 SCSI_LOG_ERROR_RECOVERY(1, 2135 shost_printk(KERN_INFO, shost, 2136 "scsi_eh_%d: sleeping\n", 2137 shost->host_no)); 2138 schedule(); 2139 continue; 2140 } 2141 2142 __set_current_state(TASK_RUNNING); 2143 SCSI_LOG_ERROR_RECOVERY(1, 2144 shost_printk(KERN_INFO, shost, 2145 "scsi_eh_%d: waking up %d/%d/%d\n", 2146 shost->host_no, shost->host_eh_scheduled, 2147 shost->host_failed, 2148 atomic_read(&shost->host_busy))); 2149 2150 /* 2151 * We have a host that is failing for some reason. Figure out 2152 * what we need to do to get it up and online again (if we can). 2153 * If we fail, we end up taking the thing offline. 2154 */ 2155 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2156 SCSI_LOG_ERROR_RECOVERY(1, 2157 shost_printk(KERN_ERR, shost, 2158 "scsi_eh_%d: unable to autoresume\n", 2159 shost->host_no)); 2160 continue; 2161 } 2162 2163 if (shost->transportt->eh_strategy_handler) 2164 shost->transportt->eh_strategy_handler(shost); 2165 else 2166 scsi_unjam_host(shost); 2167 2168 /* All scmds have been handled */ 2169 shost->host_failed = 0; 2170 2171 /* 2172 * Note - if the above fails completely, the action is to take 2173 * individual devices offline and flush the queue of any 2174 * outstanding requests that may have been pending. When we 2175 * restart, we restart any I/O to any other devices on the bus 2176 * which are still online. 2177 */ 2178 scsi_restart_operations(shost); 2179 if (!shost->eh_noresume) 2180 scsi_autopm_put_host(shost); 2181 } 2182 __set_current_state(TASK_RUNNING); 2183 2184 SCSI_LOG_ERROR_RECOVERY(1, 2185 shost_printk(KERN_INFO, shost, 2186 "Error handler scsi_eh_%d exiting\n", 2187 shost->host_no)); 2188 shost->ehandler = NULL; 2189 return 0; 2190 } 2191 2192 /* 2193 * Function: scsi_report_bus_reset() 2194 * 2195 * Purpose: Utility function used by low-level drivers to report that 2196 * they have observed a bus reset on the bus being handled. 2197 * 2198 * Arguments: shost - Host in question 2199 * channel - channel on which reset was observed. 2200 * 2201 * Returns: Nothing 2202 * 2203 * Lock status: Host lock must be held. 2204 * 2205 * Notes: This only needs to be called if the reset is one which 2206 * originates from an unknown location. Resets originated 2207 * by the mid-level itself don't need to call this, but there 2208 * should be no harm. 2209 * 2210 * The main purpose of this is to make sure that a CHECK_CONDITION 2211 * is properly treated. 2212 */ 2213 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2214 { 2215 struct scsi_device *sdev; 2216 2217 __shost_for_each_device(sdev, shost) { 2218 if (channel == sdev_channel(sdev)) 2219 __scsi_report_device_reset(sdev, NULL); 2220 } 2221 } 2222 EXPORT_SYMBOL(scsi_report_bus_reset); 2223 2224 /* 2225 * Function: scsi_report_device_reset() 2226 * 2227 * Purpose: Utility function used by low-level drivers to report that 2228 * they have observed a device reset on the device being handled. 2229 * 2230 * Arguments: shost - Host in question 2231 * channel - channel on which reset was observed 2232 * target - target on which reset was observed 2233 * 2234 * Returns: Nothing 2235 * 2236 * Lock status: Host lock must be held 2237 * 2238 * Notes: This only needs to be called if the reset is one which 2239 * originates from an unknown location. Resets originated 2240 * by the mid-level itself don't need to call this, but there 2241 * should be no harm. 2242 * 2243 * The main purpose of this is to make sure that a CHECK_CONDITION 2244 * is properly treated. 2245 */ 2246 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2247 { 2248 struct scsi_device *sdev; 2249 2250 __shost_for_each_device(sdev, shost) { 2251 if (channel == sdev_channel(sdev) && 2252 target == sdev_id(sdev)) 2253 __scsi_report_device_reset(sdev, NULL); 2254 } 2255 } 2256 EXPORT_SYMBOL(scsi_report_device_reset); 2257 2258 static void 2259 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2260 { 2261 } 2262 2263 /** 2264 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2265 * @dev: scsi_device to operate on 2266 * @arg: reset type (see sg.h) 2267 */ 2268 int 2269 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2270 { 2271 struct scsi_cmnd *scmd; 2272 struct Scsi_Host *shost = dev->host; 2273 struct request *rq; 2274 unsigned long flags; 2275 int error = 0, rtn, val; 2276 2277 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2278 return -EACCES; 2279 2280 error = get_user(val, arg); 2281 if (error) 2282 return error; 2283 2284 if (scsi_autopm_get_host(shost) < 0) 2285 return -EIO; 2286 2287 error = -EIO; 2288 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2289 shost->hostt->cmd_size, GFP_KERNEL); 2290 if (!rq) 2291 goto out_put_autopm_host; 2292 blk_rq_init(NULL, rq); 2293 2294 scmd = (struct scsi_cmnd *)(rq + 1); 2295 scsi_init_command(dev, scmd); 2296 scmd->request = rq; 2297 scmd->cmnd = scsi_req(rq)->cmd; 2298 2299 scmd->scsi_done = scsi_reset_provider_done_command; 2300 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2301 2302 scmd->cmd_len = 0; 2303 2304 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2305 2306 spin_lock_irqsave(shost->host_lock, flags); 2307 shost->tmf_in_progress = 1; 2308 spin_unlock_irqrestore(shost->host_lock, flags); 2309 2310 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2311 case SG_SCSI_RESET_NOTHING: 2312 rtn = SUCCESS; 2313 break; 2314 case SG_SCSI_RESET_DEVICE: 2315 rtn = scsi_try_bus_device_reset(scmd); 2316 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2317 break; 2318 /* FALLTHROUGH */ 2319 case SG_SCSI_RESET_TARGET: 2320 rtn = scsi_try_target_reset(scmd); 2321 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2322 break; 2323 /* FALLTHROUGH */ 2324 case SG_SCSI_RESET_BUS: 2325 rtn = scsi_try_bus_reset(scmd); 2326 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2327 break; 2328 /* FALLTHROUGH */ 2329 case SG_SCSI_RESET_HOST: 2330 rtn = scsi_try_host_reset(scmd); 2331 if (rtn == SUCCESS) 2332 break; 2333 /* FALLTHROUGH */ 2334 default: 2335 rtn = FAILED; 2336 break; 2337 } 2338 2339 error = (rtn == SUCCESS) ? 0 : -EIO; 2340 2341 spin_lock_irqsave(shost->host_lock, flags); 2342 shost->tmf_in_progress = 0; 2343 spin_unlock_irqrestore(shost->host_lock, flags); 2344 2345 /* 2346 * be sure to wake up anyone who was sleeping or had their queue 2347 * suspended while we performed the TMF. 2348 */ 2349 SCSI_LOG_ERROR_RECOVERY(3, 2350 shost_printk(KERN_INFO, shost, 2351 "waking up host to restart after TMF\n")); 2352 2353 wake_up(&shost->host_wait); 2354 scsi_run_host_queues(shost); 2355 2356 scsi_put_command(scmd); 2357 kfree(rq); 2358 2359 out_put_autopm_host: 2360 scsi_autopm_put_host(shost); 2361 return error; 2362 } 2363 EXPORT_SYMBOL(scsi_ioctl_reset); 2364 2365 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2366 struct scsi_sense_hdr *sshdr) 2367 { 2368 return scsi_normalize_sense(cmd->sense_buffer, 2369 SCSI_SENSE_BUFFERSIZE, sshdr); 2370 } 2371 EXPORT_SYMBOL(scsi_command_normalize_sense); 2372 2373 /** 2374 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2375 * @sense_buffer: byte array of sense data 2376 * @sb_len: number of valid bytes in sense_buffer 2377 * @info_out: pointer to 64 integer where 8 or 4 byte information 2378 * field will be placed if found. 2379 * 2380 * Return value: 2381 * true if information field found, false if not found. 2382 */ 2383 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2384 u64 *info_out) 2385 { 2386 const u8 * ucp; 2387 2388 if (sb_len < 7) 2389 return false; 2390 switch (sense_buffer[0] & 0x7f) { 2391 case 0x70: 2392 case 0x71: 2393 if (sense_buffer[0] & 0x80) { 2394 *info_out = get_unaligned_be32(&sense_buffer[3]); 2395 return true; 2396 } 2397 return false; 2398 case 0x72: 2399 case 0x73: 2400 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2401 0 /* info desc */); 2402 if (ucp && (0xa == ucp[1])) { 2403 *info_out = get_unaligned_be64(&ucp[4]); 2404 return true; 2405 } 2406 return false; 2407 default: 2408 return false; 2409 } 2410 } 2411 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2412