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 void scsi_eh_wakeup(struct Scsi_Host *shost) 65 { 66 lockdep_assert_held(shost->host_lock); 67 68 if (atomic_read(&shost->host_busy) == shost->host_failed) { 69 trace_scsi_eh_wakeup(shost); 70 wake_up_process(shost->ehandler); 71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 72 "Waking error handler thread\n")); 73 } 74 } 75 76 /** 77 * scsi_schedule_eh - schedule EH for SCSI host 78 * @shost: SCSI host to invoke error handling on. 79 * 80 * Schedule SCSI EH without scmd. 81 */ 82 void scsi_schedule_eh(struct Scsi_Host *shost) 83 { 84 unsigned long flags; 85 86 spin_lock_irqsave(shost->host_lock, flags); 87 88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 90 shost->host_eh_scheduled++; 91 scsi_eh_wakeup(shost); 92 } 93 94 spin_unlock_irqrestore(shost->host_lock, flags); 95 } 96 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 97 98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 99 { 100 if (!shost->last_reset || shost->eh_deadline == -1) 101 return 0; 102 103 /* 104 * 32bit accesses are guaranteed to be atomic 105 * (on all supported architectures), so instead 106 * of using a spinlock we can as well double check 107 * if eh_deadline has been set to 'off' during the 108 * time_before call. 109 */ 110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 111 shost->eh_deadline > -1) 112 return 0; 113 114 return 1; 115 } 116 117 /** 118 * scmd_eh_abort_handler - Handle command aborts 119 * @work: command to be aborted. 120 */ 121 void 122 scmd_eh_abort_handler(struct work_struct *work) 123 { 124 struct scsi_cmnd *scmd = 125 container_of(work, struct scsi_cmnd, abort_work.work); 126 struct scsi_device *sdev = scmd->device; 127 int rtn; 128 129 if (scsi_host_eh_past_deadline(sdev->host)) { 130 SCSI_LOG_ERROR_RECOVERY(3, 131 scmd_printk(KERN_INFO, scmd, 132 "eh timeout, not aborting\n")); 133 } else { 134 SCSI_LOG_ERROR_RECOVERY(3, 135 scmd_printk(KERN_INFO, scmd, 136 "aborting command\n")); 137 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); 138 if (rtn == SUCCESS) { 139 set_host_byte(scmd, DID_TIME_OUT); 140 if (scsi_host_eh_past_deadline(sdev->host)) { 141 SCSI_LOG_ERROR_RECOVERY(3, 142 scmd_printk(KERN_INFO, scmd, 143 "eh timeout, not retrying " 144 "aborted command\n")); 145 } else if (!scsi_noretry_cmd(scmd) && 146 (++scmd->retries <= scmd->allowed)) { 147 SCSI_LOG_ERROR_RECOVERY(3, 148 scmd_printk(KERN_WARNING, scmd, 149 "retry aborted command\n")); 150 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 151 return; 152 } else { 153 SCSI_LOG_ERROR_RECOVERY(3, 154 scmd_printk(KERN_WARNING, scmd, 155 "finish aborted command\n")); 156 scsi_finish_command(scmd); 157 return; 158 } 159 } else { 160 SCSI_LOG_ERROR_RECOVERY(3, 161 scmd_printk(KERN_INFO, scmd, 162 "cmd abort %s\n", 163 (rtn == FAST_IO_FAIL) ? 164 "not send" : "failed")); 165 } 166 } 167 168 scsi_eh_scmd_add(scmd); 169 } 170 171 /** 172 * scsi_abort_command - schedule a command abort 173 * @scmd: scmd to abort. 174 * 175 * We only need to abort commands after a command timeout 176 */ 177 static int 178 scsi_abort_command(struct scsi_cmnd *scmd) 179 { 180 struct scsi_device *sdev = scmd->device; 181 struct Scsi_Host *shost = sdev->host; 182 unsigned long flags; 183 184 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 185 /* 186 * Retry after abort failed, escalate to next level. 187 */ 188 SCSI_LOG_ERROR_RECOVERY(3, 189 scmd_printk(KERN_INFO, scmd, 190 "previous abort failed\n")); 191 BUG_ON(delayed_work_pending(&scmd->abort_work)); 192 return FAILED; 193 } 194 195 spin_lock_irqsave(shost->host_lock, flags); 196 if (shost->eh_deadline != -1 && !shost->last_reset) 197 shost->last_reset = jiffies; 198 spin_unlock_irqrestore(shost->host_lock, flags); 199 200 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 201 SCSI_LOG_ERROR_RECOVERY(3, 202 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 203 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 204 return SUCCESS; 205 } 206 207 /** 208 * scsi_eh_reset - call into ->eh_action to reset internal counters 209 * @scmd: scmd to run eh on. 210 * 211 * The scsi driver might be carrying internal state about the 212 * devices, so we need to call into the driver to reset the 213 * internal state once the error handler is started. 214 */ 215 static void scsi_eh_reset(struct scsi_cmnd *scmd) 216 { 217 if (!blk_rq_is_passthrough(scmd->request)) { 218 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 219 if (sdrv->eh_reset) 220 sdrv->eh_reset(scmd); 221 } 222 } 223 224 static void scsi_eh_inc_host_failed(struct rcu_head *head) 225 { 226 struct Scsi_Host *shost = container_of(head, typeof(*shost), rcu); 227 unsigned long flags; 228 229 spin_lock_irqsave(shost->host_lock, flags); 230 shost->host_failed++; 231 scsi_eh_wakeup(shost); 232 spin_unlock_irqrestore(shost->host_lock, flags); 233 } 234 235 /** 236 * scsi_eh_scmd_add - add scsi cmd to error handling. 237 * @scmd: scmd to run eh on. 238 */ 239 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 240 { 241 struct Scsi_Host *shost = scmd->device->host; 242 unsigned long flags; 243 int ret; 244 245 WARN_ON_ONCE(!shost->ehandler); 246 247 spin_lock_irqsave(shost->host_lock, flags); 248 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 249 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 250 WARN_ON_ONCE(ret); 251 } 252 if (shost->eh_deadline != -1 && !shost->last_reset) 253 shost->last_reset = jiffies; 254 255 scsi_eh_reset(scmd); 256 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 257 spin_unlock_irqrestore(shost->host_lock, flags); 258 /* 259 * Ensure that all tasks observe the host state change before the 260 * host_failed change. 261 */ 262 call_rcu(&shost->rcu, scsi_eh_inc_host_failed); 263 } 264 265 /** 266 * scsi_times_out - Timeout function for normal scsi commands. 267 * @req: request that is timing out. 268 * 269 * Notes: 270 * We do not need to lock this. There is the potential for a race 271 * only in that the normal completion handling might run, but if the 272 * normal completion function determines that the timer has already 273 * fired, then it mustn't do anything. 274 */ 275 enum blk_eh_timer_return scsi_times_out(struct request *req) 276 { 277 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 278 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; 279 struct Scsi_Host *host = scmd->device->host; 280 281 trace_scsi_dispatch_cmd_timeout(scmd); 282 scsi_log_completion(scmd, TIMEOUT_ERROR); 283 284 if (host->eh_deadline != -1 && !host->last_reset) 285 host->last_reset = jiffies; 286 287 if (host->hostt->eh_timed_out) 288 rtn = host->hostt->eh_timed_out(scmd); 289 290 if (rtn == BLK_EH_NOT_HANDLED) { 291 if (scsi_abort_command(scmd) != SUCCESS) { 292 set_host_byte(scmd, DID_TIME_OUT); 293 scsi_eh_scmd_add(scmd); 294 } 295 } 296 297 return rtn; 298 } 299 300 /** 301 * scsi_block_when_processing_errors - Prevent cmds from being queued. 302 * @sdev: Device on which we are performing recovery. 303 * 304 * Description: 305 * We block until the host is out of error recovery, and then check to 306 * see whether the host or the device is offline. 307 * 308 * Return value: 309 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 310 */ 311 int scsi_block_when_processing_errors(struct scsi_device *sdev) 312 { 313 int online; 314 315 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 316 317 online = scsi_device_online(sdev); 318 319 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev, 320 "%s: rtn: %d\n", __func__, online)); 321 322 return online; 323 } 324 EXPORT_SYMBOL(scsi_block_when_processing_errors); 325 326 #ifdef CONFIG_SCSI_LOGGING 327 /** 328 * scsi_eh_prt_fail_stats - Log info on failures. 329 * @shost: scsi host being recovered. 330 * @work_q: Queue of scsi cmds to process. 331 */ 332 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 333 struct list_head *work_q) 334 { 335 struct scsi_cmnd *scmd; 336 struct scsi_device *sdev; 337 int total_failures = 0; 338 int cmd_failed = 0; 339 int cmd_cancel = 0; 340 int devices_failed = 0; 341 342 shost_for_each_device(sdev, shost) { 343 list_for_each_entry(scmd, work_q, eh_entry) { 344 if (scmd->device == sdev) { 345 ++total_failures; 346 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 347 ++cmd_cancel; 348 else 349 ++cmd_failed; 350 } 351 } 352 353 if (cmd_cancel || cmd_failed) { 354 SCSI_LOG_ERROR_RECOVERY(3, 355 shost_printk(KERN_INFO, shost, 356 "%s: cmds failed: %d, cancel: %d\n", 357 __func__, cmd_failed, 358 cmd_cancel)); 359 cmd_cancel = 0; 360 cmd_failed = 0; 361 ++devices_failed; 362 } 363 } 364 365 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 366 "Total of %d commands on %d" 367 " devices require eh work\n", 368 total_failures, devices_failed)); 369 } 370 #endif 371 372 /** 373 * scsi_report_lun_change - Set flag on all *other* devices on the same target 374 * to indicate that a UNIT ATTENTION is expected. 375 * @sdev: Device reporting the UNIT ATTENTION 376 */ 377 static void scsi_report_lun_change(struct scsi_device *sdev) 378 { 379 sdev->sdev_target->expecting_lun_change = 1; 380 } 381 382 /** 383 * scsi_report_sense - Examine scsi sense information and log messages for 384 * certain conditions, also issue uevents for some of them. 385 * @sdev: Device reporting the sense code 386 * @sshdr: sshdr to be examined 387 */ 388 static void scsi_report_sense(struct scsi_device *sdev, 389 struct scsi_sense_hdr *sshdr) 390 { 391 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 392 393 if (sshdr->sense_key == UNIT_ATTENTION) { 394 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 395 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 396 sdev_printk(KERN_WARNING, sdev, 397 "Inquiry data has changed"); 398 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 399 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 400 scsi_report_lun_change(sdev); 401 sdev_printk(KERN_WARNING, sdev, 402 "Warning! Received an indication that the " 403 "LUN assignments on this target have " 404 "changed. The Linux SCSI layer does not " 405 "automatically remap LUN assignments.\n"); 406 } else if (sshdr->asc == 0x3f) 407 sdev_printk(KERN_WARNING, sdev, 408 "Warning! Received an indication that the " 409 "operating parameters on this target have " 410 "changed. The Linux SCSI layer does not " 411 "automatically adjust these parameters.\n"); 412 413 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 414 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 415 sdev_printk(KERN_WARNING, sdev, 416 "Warning! Received an indication that the " 417 "LUN reached a thin provisioning soft " 418 "threshold.\n"); 419 } 420 421 if (sshdr->asc == 0x29) { 422 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 423 sdev_printk(KERN_WARNING, sdev, 424 "Power-on or device reset occurred\n"); 425 } 426 427 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 428 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 429 sdev_printk(KERN_WARNING, sdev, 430 "Mode parameters changed"); 431 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 432 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 433 sdev_printk(KERN_WARNING, sdev, 434 "Asymmetric access state changed"); 435 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 436 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 437 sdev_printk(KERN_WARNING, sdev, 438 "Capacity data has changed"); 439 } else if (sshdr->asc == 0x2a) 440 sdev_printk(KERN_WARNING, sdev, 441 "Parameters changed"); 442 } 443 444 if (evt_type != SDEV_EVT_MAXBITS) { 445 set_bit(evt_type, sdev->pending_events); 446 schedule_work(&sdev->event_work); 447 } 448 } 449 450 /** 451 * scsi_check_sense - Examine scsi cmd sense 452 * @scmd: Cmd to have sense checked. 453 * 454 * Return value: 455 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 456 * 457 * Notes: 458 * When a deferred error is detected the current command has 459 * not been executed and needs retrying. 460 */ 461 int scsi_check_sense(struct scsi_cmnd *scmd) 462 { 463 struct scsi_device *sdev = scmd->device; 464 struct scsi_sense_hdr sshdr; 465 466 if (! scsi_command_normalize_sense(scmd, &sshdr)) 467 return FAILED; /* no valid sense data */ 468 469 scsi_report_sense(sdev, &sshdr); 470 471 if (scsi_sense_is_deferred(&sshdr)) 472 return NEEDS_RETRY; 473 474 if (sdev->handler && sdev->handler->check_sense) { 475 int rc; 476 477 rc = sdev->handler->check_sense(sdev, &sshdr); 478 if (rc != SCSI_RETURN_NOT_HANDLED) 479 return rc; 480 /* handler does not care. Drop down to default handling */ 481 } 482 483 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 484 /* 485 * nasty: for mid-layer issued TURs, we need to return the 486 * actual sense data without any recovery attempt. For eh 487 * issued ones, we need to try to recover and interpret 488 */ 489 return SUCCESS; 490 491 /* 492 * Previous logic looked for FILEMARK, EOM or ILI which are 493 * mainly associated with tapes and returned SUCCESS. 494 */ 495 if (sshdr.response_code == 0x70) { 496 /* fixed format */ 497 if (scmd->sense_buffer[2] & 0xe0) 498 return SUCCESS; 499 } else { 500 /* 501 * descriptor format: look for "stream commands sense data 502 * descriptor" (see SSC-3). Assume single sense data 503 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 504 */ 505 if ((sshdr.additional_length > 3) && 506 (scmd->sense_buffer[8] == 0x4) && 507 (scmd->sense_buffer[11] & 0xe0)) 508 return SUCCESS; 509 } 510 511 switch (sshdr.sense_key) { 512 case NO_SENSE: 513 return SUCCESS; 514 case RECOVERED_ERROR: 515 return /* soft_error */ SUCCESS; 516 517 case ABORTED_COMMAND: 518 if (sshdr.asc == 0x10) /* DIF */ 519 return SUCCESS; 520 521 return NEEDS_RETRY; 522 case NOT_READY: 523 case UNIT_ATTENTION: 524 /* 525 * if we are expecting a cc/ua because of a bus reset that we 526 * performed, treat this just as a retry. otherwise this is 527 * information that we should pass up to the upper-level driver 528 * so that we can deal with it there. 529 */ 530 if (scmd->device->expecting_cc_ua) { 531 /* 532 * Because some device does not queue unit 533 * attentions correctly, we carefully check 534 * additional sense code and qualifier so as 535 * not to squash media change unit attention. 536 */ 537 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 538 scmd->device->expecting_cc_ua = 0; 539 return NEEDS_RETRY; 540 } 541 } 542 /* 543 * we might also expect a cc/ua if another LUN on the target 544 * reported a UA with an ASC/ASCQ of 3F 0E - 545 * REPORTED LUNS DATA HAS CHANGED. 546 */ 547 if (scmd->device->sdev_target->expecting_lun_change && 548 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 549 return NEEDS_RETRY; 550 /* 551 * if the device is in the process of becoming ready, we 552 * should retry. 553 */ 554 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 555 return NEEDS_RETRY; 556 /* 557 * if the device is not started, we need to wake 558 * the error handler to start the motor 559 */ 560 if (scmd->device->allow_restart && 561 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 562 return FAILED; 563 /* 564 * Pass the UA upwards for a determination in the completion 565 * functions. 566 */ 567 return SUCCESS; 568 569 /* these are not supported */ 570 case DATA_PROTECT: 571 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 572 /* Thin provisioning hard threshold reached */ 573 set_host_byte(scmd, DID_ALLOC_FAILURE); 574 return SUCCESS; 575 } 576 /* FALLTHROUGH */ 577 case COPY_ABORTED: 578 case VOLUME_OVERFLOW: 579 case MISCOMPARE: 580 case BLANK_CHECK: 581 set_host_byte(scmd, DID_TARGET_FAILURE); 582 return SUCCESS; 583 584 case MEDIUM_ERROR: 585 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 586 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 587 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 588 set_host_byte(scmd, DID_MEDIUM_ERROR); 589 return SUCCESS; 590 } 591 return NEEDS_RETRY; 592 593 case HARDWARE_ERROR: 594 if (scmd->device->retry_hwerror) 595 return ADD_TO_MLQUEUE; 596 else 597 set_host_byte(scmd, DID_TARGET_FAILURE); 598 /* FALLTHROUGH */ 599 600 case ILLEGAL_REQUEST: 601 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 602 sshdr.asc == 0x21 || /* Logical block address out of range */ 603 sshdr.asc == 0x22 || /* Invalid function */ 604 sshdr.asc == 0x24 || /* Invalid field in cdb */ 605 sshdr.asc == 0x26 || /* Parameter value invalid */ 606 sshdr.asc == 0x27) { /* Write protected */ 607 set_host_byte(scmd, DID_TARGET_FAILURE); 608 } 609 return SUCCESS; 610 611 default: 612 return SUCCESS; 613 } 614 } 615 EXPORT_SYMBOL_GPL(scsi_check_sense); 616 617 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 618 { 619 struct scsi_host_template *sht = sdev->host->hostt; 620 struct scsi_device *tmp_sdev; 621 622 if (!sht->track_queue_depth || 623 sdev->queue_depth >= sdev->max_queue_depth) 624 return; 625 626 if (time_before(jiffies, 627 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 628 return; 629 630 if (time_before(jiffies, 631 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 632 return; 633 634 /* 635 * Walk all devices of a target and do 636 * ramp up on them. 637 */ 638 shost_for_each_device(tmp_sdev, sdev->host) { 639 if (tmp_sdev->channel != sdev->channel || 640 tmp_sdev->id != sdev->id || 641 tmp_sdev->queue_depth == sdev->max_queue_depth) 642 continue; 643 644 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 645 sdev->last_queue_ramp_up = jiffies; 646 } 647 } 648 649 static void scsi_handle_queue_full(struct scsi_device *sdev) 650 { 651 struct scsi_host_template *sht = sdev->host->hostt; 652 struct scsi_device *tmp_sdev; 653 654 if (!sht->track_queue_depth) 655 return; 656 657 shost_for_each_device(tmp_sdev, sdev->host) { 658 if (tmp_sdev->channel != sdev->channel || 659 tmp_sdev->id != sdev->id) 660 continue; 661 /* 662 * We do not know the number of commands that were at 663 * the device when we got the queue full so we start 664 * from the highest possible value and work our way down. 665 */ 666 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 667 } 668 } 669 670 /** 671 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 672 * @scmd: SCSI cmd to examine. 673 * 674 * Notes: 675 * This is *only* called when we are examining the status of commands 676 * queued during error recovery. the main difference here is that we 677 * don't allow for the possibility of retries here, and we are a lot 678 * more restrictive about what we consider acceptable. 679 */ 680 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 681 { 682 /* 683 * first check the host byte, to see if there is anything in there 684 * that would indicate what we need to do. 685 */ 686 if (host_byte(scmd->result) == DID_RESET) { 687 /* 688 * rats. we are already in the error handler, so we now 689 * get to try and figure out what to do next. if the sense 690 * is valid, we have a pretty good idea of what to do. 691 * if not, we mark it as FAILED. 692 */ 693 return scsi_check_sense(scmd); 694 } 695 if (host_byte(scmd->result) != DID_OK) 696 return FAILED; 697 698 /* 699 * next, check the message byte. 700 */ 701 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 702 return FAILED; 703 704 /* 705 * now, check the status byte to see if this indicates 706 * anything special. 707 */ 708 switch (status_byte(scmd->result)) { 709 case GOOD: 710 scsi_handle_queue_ramp_up(scmd->device); 711 /* FALLTHROUGH */ 712 case COMMAND_TERMINATED: 713 return SUCCESS; 714 case CHECK_CONDITION: 715 return scsi_check_sense(scmd); 716 case CONDITION_GOOD: 717 case INTERMEDIATE_GOOD: 718 case INTERMEDIATE_C_GOOD: 719 /* 720 * who knows? FIXME(eric) 721 */ 722 return SUCCESS; 723 case RESERVATION_CONFLICT: 724 if (scmd->cmnd[0] == TEST_UNIT_READY) 725 /* it is a success, we probed the device and 726 * found it */ 727 return SUCCESS; 728 /* otherwise, we failed to send the command */ 729 return FAILED; 730 case QUEUE_FULL: 731 scsi_handle_queue_full(scmd->device); 732 /* fall through */ 733 case BUSY: 734 return NEEDS_RETRY; 735 default: 736 return FAILED; 737 } 738 return FAILED; 739 } 740 741 /** 742 * scsi_eh_done - Completion function for error handling. 743 * @scmd: Cmd that is done. 744 */ 745 static void scsi_eh_done(struct scsi_cmnd *scmd) 746 { 747 struct completion *eh_action; 748 749 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 750 "%s result: %x\n", __func__, scmd->result)); 751 752 eh_action = scmd->device->host->eh_action; 753 if (eh_action) 754 complete(eh_action); 755 } 756 757 /** 758 * scsi_try_host_reset - ask host adapter to reset itself 759 * @scmd: SCSI cmd to send host reset. 760 */ 761 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 762 { 763 unsigned long flags; 764 int rtn; 765 struct Scsi_Host *host = scmd->device->host; 766 struct scsi_host_template *hostt = host->hostt; 767 768 SCSI_LOG_ERROR_RECOVERY(3, 769 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 770 771 if (!hostt->eh_host_reset_handler) 772 return FAILED; 773 774 rtn = hostt->eh_host_reset_handler(scmd); 775 776 if (rtn == SUCCESS) { 777 if (!hostt->skip_settle_delay) 778 ssleep(HOST_RESET_SETTLE_TIME); 779 spin_lock_irqsave(host->host_lock, flags); 780 scsi_report_bus_reset(host, scmd_channel(scmd)); 781 spin_unlock_irqrestore(host->host_lock, flags); 782 } 783 784 return rtn; 785 } 786 787 /** 788 * scsi_try_bus_reset - ask host to perform a bus reset 789 * @scmd: SCSI cmd to send bus reset. 790 */ 791 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 792 { 793 unsigned long flags; 794 int rtn; 795 struct Scsi_Host *host = scmd->device->host; 796 struct scsi_host_template *hostt = host->hostt; 797 798 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 799 "%s: Snd Bus RST\n", __func__)); 800 801 if (!hostt->eh_bus_reset_handler) 802 return FAILED; 803 804 rtn = hostt->eh_bus_reset_handler(scmd); 805 806 if (rtn == SUCCESS) { 807 if (!hostt->skip_settle_delay) 808 ssleep(BUS_RESET_SETTLE_TIME); 809 spin_lock_irqsave(host->host_lock, flags); 810 scsi_report_bus_reset(host, scmd_channel(scmd)); 811 spin_unlock_irqrestore(host->host_lock, flags); 812 } 813 814 return rtn; 815 } 816 817 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 818 { 819 sdev->was_reset = 1; 820 sdev->expecting_cc_ua = 1; 821 } 822 823 /** 824 * scsi_try_target_reset - Ask host to perform a target reset 825 * @scmd: SCSI cmd used to send a target reset 826 * 827 * Notes: 828 * There is no timeout for this operation. if this operation is 829 * unreliable for a given host, then the host itself needs to put a 830 * timer on it, and set the host back to a consistent state prior to 831 * returning. 832 */ 833 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 834 { 835 unsigned long flags; 836 int rtn; 837 struct Scsi_Host *host = scmd->device->host; 838 struct scsi_host_template *hostt = host->hostt; 839 840 if (!hostt->eh_target_reset_handler) 841 return FAILED; 842 843 rtn = hostt->eh_target_reset_handler(scmd); 844 if (rtn == SUCCESS) { 845 spin_lock_irqsave(host->host_lock, flags); 846 __starget_for_each_device(scsi_target(scmd->device), NULL, 847 __scsi_report_device_reset); 848 spin_unlock_irqrestore(host->host_lock, flags); 849 } 850 851 return rtn; 852 } 853 854 /** 855 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 856 * @scmd: SCSI cmd used to send BDR 857 * 858 * Notes: 859 * There is no timeout for this operation. if this operation is 860 * unreliable for a given host, then the host itself needs to put a 861 * timer on it, and set the host back to a consistent state prior to 862 * returning. 863 */ 864 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 865 { 866 int rtn; 867 struct scsi_host_template *hostt = scmd->device->host->hostt; 868 869 if (!hostt->eh_device_reset_handler) 870 return FAILED; 871 872 rtn = hostt->eh_device_reset_handler(scmd); 873 if (rtn == SUCCESS) 874 __scsi_report_device_reset(scmd->device, NULL); 875 return rtn; 876 } 877 878 /** 879 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 880 * @hostt: SCSI driver host template 881 * @scmd: SCSI cmd used to send a target reset 882 * 883 * Return value: 884 * SUCCESS, FAILED, or FAST_IO_FAIL 885 * 886 * Notes: 887 * SUCCESS does not necessarily indicate that the command 888 * has been aborted; it only indicates that the LLDDs 889 * has cleared all references to that command. 890 * LLDDs should return FAILED only if an abort was required 891 * but could not be executed. LLDDs should return FAST_IO_FAIL 892 * if the device is temporarily unavailable (eg due to a 893 * link down on FibreChannel) 894 */ 895 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, 896 struct scsi_cmnd *scmd) 897 { 898 if (!hostt->eh_abort_handler) 899 return FAILED; 900 901 return hostt->eh_abort_handler(scmd); 902 } 903 904 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 905 { 906 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 907 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 908 if (scsi_try_target_reset(scmd) != SUCCESS) 909 if (scsi_try_bus_reset(scmd) != SUCCESS) 910 scsi_try_host_reset(scmd); 911 } 912 913 /** 914 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 915 * @scmd: SCSI command structure to hijack 916 * @ses: structure to save restore information 917 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 918 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 919 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 920 * 921 * This function is used to save a scsi command information before re-execution 922 * as part of the error recovery process. If @sense_bytes is 0 the command 923 * sent must be one that does not transfer any data. If @sense_bytes != 0 924 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 925 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 926 */ 927 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 928 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 929 { 930 struct scsi_device *sdev = scmd->device; 931 932 /* 933 * We need saved copies of a number of fields - this is because 934 * error handling may need to overwrite these with different values 935 * to run different commands, and once error handling is complete, 936 * we will need to restore these values prior to running the actual 937 * command. 938 */ 939 ses->cmd_len = scmd->cmd_len; 940 ses->cmnd = scmd->cmnd; 941 ses->data_direction = scmd->sc_data_direction; 942 ses->sdb = scmd->sdb; 943 ses->next_rq = scmd->request->next_rq; 944 ses->result = scmd->result; 945 ses->underflow = scmd->underflow; 946 ses->prot_op = scmd->prot_op; 947 ses->eh_eflags = scmd->eh_eflags; 948 949 scmd->prot_op = SCSI_PROT_NORMAL; 950 scmd->eh_eflags = 0; 951 scmd->cmnd = ses->eh_cmnd; 952 memset(scmd->cmnd, 0, BLK_MAX_CDB); 953 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 954 scmd->request->next_rq = NULL; 955 scmd->result = 0; 956 957 if (sense_bytes) { 958 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 959 sense_bytes); 960 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 961 scmd->sdb.length); 962 scmd->sdb.table.sgl = &ses->sense_sgl; 963 scmd->sc_data_direction = DMA_FROM_DEVICE; 964 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 965 scmd->cmnd[0] = REQUEST_SENSE; 966 scmd->cmnd[4] = scmd->sdb.length; 967 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 968 } else { 969 scmd->sc_data_direction = DMA_NONE; 970 if (cmnd) { 971 BUG_ON(cmnd_size > BLK_MAX_CDB); 972 memcpy(scmd->cmnd, cmnd, cmnd_size); 973 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 974 } 975 } 976 977 scmd->underflow = 0; 978 979 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 980 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 981 (sdev->lun << 5 & 0xe0); 982 983 /* 984 * Zero the sense buffer. The scsi spec mandates that any 985 * untransferred sense data should be interpreted as being zero. 986 */ 987 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 988 } 989 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 990 991 /** 992 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 993 * @scmd: SCSI command structure to restore 994 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 995 * 996 * Undo any damage done by above scsi_eh_prep_cmnd(). 997 */ 998 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 999 { 1000 /* 1001 * Restore original data 1002 */ 1003 scmd->cmd_len = ses->cmd_len; 1004 scmd->cmnd = ses->cmnd; 1005 scmd->sc_data_direction = ses->data_direction; 1006 scmd->sdb = ses->sdb; 1007 scmd->request->next_rq = ses->next_rq; 1008 scmd->result = ses->result; 1009 scmd->underflow = ses->underflow; 1010 scmd->prot_op = ses->prot_op; 1011 scmd->eh_eflags = ses->eh_eflags; 1012 } 1013 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1014 1015 /** 1016 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1017 * @scmd: SCSI command structure to hijack 1018 * @cmnd: CDB to send 1019 * @cmnd_size: size in bytes of @cmnd 1020 * @timeout: timeout for this request 1021 * @sense_bytes: size of sense data to copy or 0 1022 * 1023 * This function is used to send a scsi command down to a target device 1024 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1025 * 1026 * Return value: 1027 * SUCCESS or FAILED or NEEDS_RETRY 1028 */ 1029 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 1030 int cmnd_size, int timeout, unsigned sense_bytes) 1031 { 1032 struct scsi_device *sdev = scmd->device; 1033 struct Scsi_Host *shost = sdev->host; 1034 DECLARE_COMPLETION_ONSTACK(done); 1035 unsigned long timeleft = timeout; 1036 struct scsi_eh_save ses; 1037 const unsigned long stall_for = msecs_to_jiffies(100); 1038 int rtn; 1039 1040 retry: 1041 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1042 shost->eh_action = &done; 1043 1044 scsi_log_send(scmd); 1045 scmd->scsi_done = scsi_eh_done; 1046 rtn = shost->hostt->queuecommand(shost, scmd); 1047 if (rtn) { 1048 if (timeleft > stall_for) { 1049 scsi_eh_restore_cmnd(scmd, &ses); 1050 timeleft -= stall_for; 1051 msleep(jiffies_to_msecs(stall_for)); 1052 goto retry; 1053 } 1054 /* signal not to enter either branch of the if () below */ 1055 timeleft = 0; 1056 rtn = FAILED; 1057 } else { 1058 timeleft = wait_for_completion_timeout(&done, timeout); 1059 rtn = SUCCESS; 1060 } 1061 1062 shost->eh_action = NULL; 1063 1064 scsi_log_completion(scmd, rtn); 1065 1066 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1067 "%s timeleft: %ld\n", 1068 __func__, timeleft)); 1069 1070 /* 1071 * If there is time left scsi_eh_done got called, and we will examine 1072 * the actual status codes to see whether the command actually did 1073 * complete normally, else if we have a zero return and no time left, 1074 * the command must still be pending, so abort it and return FAILED. 1075 * If we never actually managed to issue the command, because 1076 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1077 * value above (so don't execute either branch of the if) 1078 */ 1079 if (timeleft) { 1080 rtn = scsi_eh_completed_normally(scmd); 1081 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1082 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1083 1084 switch (rtn) { 1085 case SUCCESS: 1086 case NEEDS_RETRY: 1087 case FAILED: 1088 break; 1089 case ADD_TO_MLQUEUE: 1090 rtn = NEEDS_RETRY; 1091 break; 1092 default: 1093 rtn = FAILED; 1094 break; 1095 } 1096 } else if (rtn != FAILED) { 1097 scsi_abort_eh_cmnd(scmd); 1098 rtn = FAILED; 1099 } 1100 1101 scsi_eh_restore_cmnd(scmd, &ses); 1102 1103 return rtn; 1104 } 1105 1106 /** 1107 * scsi_request_sense - Request sense data from a particular target. 1108 * @scmd: SCSI cmd for request sense. 1109 * 1110 * Notes: 1111 * Some hosts automatically obtain this information, others require 1112 * that we obtain it on our own. This function will *not* return until 1113 * the command either times out, or it completes. 1114 */ 1115 static int scsi_request_sense(struct scsi_cmnd *scmd) 1116 { 1117 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1118 } 1119 1120 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) 1121 { 1122 if (!blk_rq_is_passthrough(scmd->request)) { 1123 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1124 if (sdrv->eh_action) 1125 rtn = sdrv->eh_action(scmd, rtn); 1126 } 1127 return rtn; 1128 } 1129 1130 /** 1131 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1132 * @scmd: Original SCSI cmd that eh has finished. 1133 * @done_q: Queue for processed commands. 1134 * 1135 * Notes: 1136 * We don't want to use the normal command completion while we are are 1137 * still handling errors - it may cause other commands to be queued, 1138 * and that would disturb what we are doing. Thus we really want to 1139 * keep a list of pending commands for final completion, and once we 1140 * are ready to leave error handling we handle completion for real. 1141 */ 1142 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1143 { 1144 list_move_tail(&scmd->eh_entry, done_q); 1145 } 1146 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1147 1148 /** 1149 * scsi_eh_get_sense - Get device sense data. 1150 * @work_q: Queue of commands to process. 1151 * @done_q: Queue of processed commands. 1152 * 1153 * Description: 1154 * See if we need to request sense information. if so, then get it 1155 * now, so we have a better idea of what to do. 1156 * 1157 * Notes: 1158 * This has the unfortunate side effect that if a shost adapter does 1159 * not automatically request sense information, we end up shutting 1160 * it down before we request it. 1161 * 1162 * All drivers should request sense information internally these days, 1163 * so for now all I have to say is tough noogies if you end up in here. 1164 * 1165 * XXX: Long term this code should go away, but that needs an audit of 1166 * all LLDDs first. 1167 */ 1168 int scsi_eh_get_sense(struct list_head *work_q, 1169 struct list_head *done_q) 1170 { 1171 struct scsi_cmnd *scmd, *next; 1172 struct Scsi_Host *shost; 1173 int rtn; 1174 1175 /* 1176 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1177 * should not get sense. 1178 */ 1179 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1180 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1181 SCSI_SENSE_VALID(scmd)) 1182 continue; 1183 1184 shost = scmd->device->host; 1185 if (scsi_host_eh_past_deadline(shost)) { 1186 SCSI_LOG_ERROR_RECOVERY(3, 1187 scmd_printk(KERN_INFO, scmd, 1188 "%s: skip request sense, past eh deadline\n", 1189 current->comm)); 1190 break; 1191 } 1192 if (status_byte(scmd->result) != CHECK_CONDITION) 1193 /* 1194 * don't request sense if there's no check condition 1195 * status because the error we're processing isn't one 1196 * that has a sense code (and some devices get 1197 * confused by sense requests out of the blue) 1198 */ 1199 continue; 1200 1201 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1202 "%s: requesting sense\n", 1203 current->comm)); 1204 rtn = scsi_request_sense(scmd); 1205 if (rtn != SUCCESS) 1206 continue; 1207 1208 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1209 "sense requested, result %x\n", scmd->result)); 1210 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1211 1212 rtn = scsi_decide_disposition(scmd); 1213 1214 /* 1215 * if the result was normal, then just pass it along to the 1216 * upper level. 1217 */ 1218 if (rtn == SUCCESS) 1219 /* we don't want this command reissued, just 1220 * finished with the sense data, so set 1221 * retries to the max allowed to ensure it 1222 * won't get reissued */ 1223 scmd->retries = scmd->allowed; 1224 else if (rtn != NEEDS_RETRY) 1225 continue; 1226 1227 scsi_eh_finish_cmd(scmd, done_q); 1228 } 1229 1230 return list_empty(work_q); 1231 } 1232 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1233 1234 /** 1235 * scsi_eh_tur - Send TUR to device. 1236 * @scmd: &scsi_cmnd to send TUR 1237 * 1238 * Return value: 1239 * 0 - Device is ready. 1 - Device NOT ready. 1240 */ 1241 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1242 { 1243 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1244 int retry_cnt = 1, rtn; 1245 1246 retry_tur: 1247 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1248 scmd->device->eh_timeout, 0); 1249 1250 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1251 "%s return: %x\n", __func__, rtn)); 1252 1253 switch (rtn) { 1254 case NEEDS_RETRY: 1255 if (retry_cnt--) 1256 goto retry_tur; 1257 /*FALLTHRU*/ 1258 case SUCCESS: 1259 return 0; 1260 default: 1261 return 1; 1262 } 1263 } 1264 1265 /** 1266 * scsi_eh_test_devices - check if devices are responding from error recovery. 1267 * @cmd_list: scsi commands in error recovery. 1268 * @work_q: queue for commands which still need more error recovery 1269 * @done_q: queue for commands which are finished 1270 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1271 * 1272 * Decription: 1273 * Tests if devices are in a working state. Commands to devices now in 1274 * a working state are sent to the done_q while commands to devices which 1275 * are still failing to respond are returned to the work_q for more 1276 * processing. 1277 **/ 1278 static int scsi_eh_test_devices(struct list_head *cmd_list, 1279 struct list_head *work_q, 1280 struct list_head *done_q, int try_stu) 1281 { 1282 struct scsi_cmnd *scmd, *next; 1283 struct scsi_device *sdev; 1284 int finish_cmds; 1285 1286 while (!list_empty(cmd_list)) { 1287 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1288 sdev = scmd->device; 1289 1290 if (!try_stu) { 1291 if (scsi_host_eh_past_deadline(sdev->host)) { 1292 /* Push items back onto work_q */ 1293 list_splice_init(cmd_list, work_q); 1294 SCSI_LOG_ERROR_RECOVERY(3, 1295 sdev_printk(KERN_INFO, sdev, 1296 "%s: skip test device, past eh deadline", 1297 current->comm)); 1298 break; 1299 } 1300 } 1301 1302 finish_cmds = !scsi_device_online(scmd->device) || 1303 (try_stu && !scsi_eh_try_stu(scmd) && 1304 !scsi_eh_tur(scmd)) || 1305 !scsi_eh_tur(scmd); 1306 1307 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1308 if (scmd->device == sdev) { 1309 if (finish_cmds && 1310 (try_stu || 1311 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1312 scsi_eh_finish_cmd(scmd, done_q); 1313 else 1314 list_move_tail(&scmd->eh_entry, work_q); 1315 } 1316 } 1317 return list_empty(work_q); 1318 } 1319 1320 /** 1321 * scsi_eh_try_stu - Send START_UNIT to device. 1322 * @scmd: &scsi_cmnd to send START_UNIT 1323 * 1324 * Return value: 1325 * 0 - Device is ready. 1 - Device NOT ready. 1326 */ 1327 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1328 { 1329 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1330 1331 if (scmd->device->allow_restart) { 1332 int i, rtn = NEEDS_RETRY; 1333 1334 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1335 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1336 1337 if (rtn == SUCCESS) 1338 return 0; 1339 } 1340 1341 return 1; 1342 } 1343 1344 /** 1345 * scsi_eh_stu - send START_UNIT if needed 1346 * @shost: &scsi host being recovered. 1347 * @work_q: &list_head for pending commands. 1348 * @done_q: &list_head for processed commands. 1349 * 1350 * Notes: 1351 * If commands are failing due to not ready, initializing command required, 1352 * try revalidating the device, which will end up sending a start unit. 1353 */ 1354 static int scsi_eh_stu(struct Scsi_Host *shost, 1355 struct list_head *work_q, 1356 struct list_head *done_q) 1357 { 1358 struct scsi_cmnd *scmd, *stu_scmd, *next; 1359 struct scsi_device *sdev; 1360 1361 shost_for_each_device(sdev, shost) { 1362 if (scsi_host_eh_past_deadline(shost)) { 1363 SCSI_LOG_ERROR_RECOVERY(3, 1364 sdev_printk(KERN_INFO, sdev, 1365 "%s: skip START_UNIT, past eh deadline\n", 1366 current->comm)); 1367 break; 1368 } 1369 stu_scmd = NULL; 1370 list_for_each_entry(scmd, work_q, eh_entry) 1371 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1372 scsi_check_sense(scmd) == FAILED ) { 1373 stu_scmd = scmd; 1374 break; 1375 } 1376 1377 if (!stu_scmd) 1378 continue; 1379 1380 SCSI_LOG_ERROR_RECOVERY(3, 1381 sdev_printk(KERN_INFO, sdev, 1382 "%s: Sending START_UNIT\n", 1383 current->comm)); 1384 1385 if (!scsi_eh_try_stu(stu_scmd)) { 1386 if (!scsi_device_online(sdev) || 1387 !scsi_eh_tur(stu_scmd)) { 1388 list_for_each_entry_safe(scmd, next, 1389 work_q, eh_entry) { 1390 if (scmd->device == sdev && 1391 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1392 scsi_eh_finish_cmd(scmd, done_q); 1393 } 1394 } 1395 } else { 1396 SCSI_LOG_ERROR_RECOVERY(3, 1397 sdev_printk(KERN_INFO, sdev, 1398 "%s: START_UNIT failed\n", 1399 current->comm)); 1400 } 1401 } 1402 1403 return list_empty(work_q); 1404 } 1405 1406 1407 /** 1408 * scsi_eh_bus_device_reset - send bdr if needed 1409 * @shost: scsi host being recovered. 1410 * @work_q: &list_head for pending commands. 1411 * @done_q: &list_head for processed commands. 1412 * 1413 * Notes: 1414 * Try a bus device reset. Still, look to see whether we have multiple 1415 * devices that are jammed or not - if we have multiple devices, it 1416 * makes no sense to try bus_device_reset - we really would need to try 1417 * a bus_reset instead. 1418 */ 1419 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1420 struct list_head *work_q, 1421 struct list_head *done_q) 1422 { 1423 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1424 struct scsi_device *sdev; 1425 int rtn; 1426 1427 shost_for_each_device(sdev, shost) { 1428 if (scsi_host_eh_past_deadline(shost)) { 1429 SCSI_LOG_ERROR_RECOVERY(3, 1430 sdev_printk(KERN_INFO, sdev, 1431 "%s: skip BDR, past eh deadline\n", 1432 current->comm)); 1433 break; 1434 } 1435 bdr_scmd = NULL; 1436 list_for_each_entry(scmd, work_q, eh_entry) 1437 if (scmd->device == sdev) { 1438 bdr_scmd = scmd; 1439 break; 1440 } 1441 1442 if (!bdr_scmd) 1443 continue; 1444 1445 SCSI_LOG_ERROR_RECOVERY(3, 1446 sdev_printk(KERN_INFO, sdev, 1447 "%s: Sending BDR\n", current->comm)); 1448 rtn = scsi_try_bus_device_reset(bdr_scmd); 1449 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1450 if (!scsi_device_online(sdev) || 1451 rtn == FAST_IO_FAIL || 1452 !scsi_eh_tur(bdr_scmd)) { 1453 list_for_each_entry_safe(scmd, next, 1454 work_q, eh_entry) { 1455 if (scmd->device == sdev && 1456 scsi_eh_action(scmd, rtn) != FAILED) 1457 scsi_eh_finish_cmd(scmd, 1458 done_q); 1459 } 1460 } 1461 } else { 1462 SCSI_LOG_ERROR_RECOVERY(3, 1463 sdev_printk(KERN_INFO, sdev, 1464 "%s: BDR failed\n", current->comm)); 1465 } 1466 } 1467 1468 return list_empty(work_q); 1469 } 1470 1471 /** 1472 * scsi_eh_target_reset - send target reset if needed 1473 * @shost: scsi host being recovered. 1474 * @work_q: &list_head for pending commands. 1475 * @done_q: &list_head for processed commands. 1476 * 1477 * Notes: 1478 * Try a target reset. 1479 */ 1480 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1481 struct list_head *work_q, 1482 struct list_head *done_q) 1483 { 1484 LIST_HEAD(tmp_list); 1485 LIST_HEAD(check_list); 1486 1487 list_splice_init(work_q, &tmp_list); 1488 1489 while (!list_empty(&tmp_list)) { 1490 struct scsi_cmnd *next, *scmd; 1491 int rtn; 1492 unsigned int id; 1493 1494 if (scsi_host_eh_past_deadline(shost)) { 1495 /* push back on work queue for further processing */ 1496 list_splice_init(&check_list, work_q); 1497 list_splice_init(&tmp_list, work_q); 1498 SCSI_LOG_ERROR_RECOVERY(3, 1499 shost_printk(KERN_INFO, shost, 1500 "%s: Skip target reset, past eh deadline\n", 1501 current->comm)); 1502 return list_empty(work_q); 1503 } 1504 1505 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1506 id = scmd_id(scmd); 1507 1508 SCSI_LOG_ERROR_RECOVERY(3, 1509 shost_printk(KERN_INFO, shost, 1510 "%s: Sending target reset to target %d\n", 1511 current->comm, id)); 1512 rtn = scsi_try_target_reset(scmd); 1513 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1514 SCSI_LOG_ERROR_RECOVERY(3, 1515 shost_printk(KERN_INFO, shost, 1516 "%s: Target reset failed" 1517 " target: %d\n", 1518 current->comm, id)); 1519 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1520 if (scmd_id(scmd) != id) 1521 continue; 1522 1523 if (rtn == SUCCESS) 1524 list_move_tail(&scmd->eh_entry, &check_list); 1525 else if (rtn == FAST_IO_FAIL) 1526 scsi_eh_finish_cmd(scmd, done_q); 1527 else 1528 /* push back on work queue for further processing */ 1529 list_move(&scmd->eh_entry, work_q); 1530 } 1531 } 1532 1533 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1534 } 1535 1536 /** 1537 * scsi_eh_bus_reset - send a bus reset 1538 * @shost: &scsi host being recovered. 1539 * @work_q: &list_head for pending commands. 1540 * @done_q: &list_head for processed commands. 1541 */ 1542 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1543 struct list_head *work_q, 1544 struct list_head *done_q) 1545 { 1546 struct scsi_cmnd *scmd, *chan_scmd, *next; 1547 LIST_HEAD(check_list); 1548 unsigned int channel; 1549 int rtn; 1550 1551 /* 1552 * we really want to loop over the various channels, and do this on 1553 * a channel by channel basis. we should also check to see if any 1554 * of the failed commands are on soft_reset devices, and if so, skip 1555 * the reset. 1556 */ 1557 1558 for (channel = 0; channel <= shost->max_channel; channel++) { 1559 if (scsi_host_eh_past_deadline(shost)) { 1560 list_splice_init(&check_list, work_q); 1561 SCSI_LOG_ERROR_RECOVERY(3, 1562 shost_printk(KERN_INFO, shost, 1563 "%s: skip BRST, past eh deadline\n", 1564 current->comm)); 1565 return list_empty(work_q); 1566 } 1567 1568 chan_scmd = NULL; 1569 list_for_each_entry(scmd, work_q, eh_entry) { 1570 if (channel == scmd_channel(scmd)) { 1571 chan_scmd = scmd; 1572 break; 1573 /* 1574 * FIXME add back in some support for 1575 * soft_reset devices. 1576 */ 1577 } 1578 } 1579 1580 if (!chan_scmd) 1581 continue; 1582 SCSI_LOG_ERROR_RECOVERY(3, 1583 shost_printk(KERN_INFO, shost, 1584 "%s: Sending BRST chan: %d\n", 1585 current->comm, channel)); 1586 rtn = scsi_try_bus_reset(chan_scmd); 1587 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1588 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1589 if (channel == scmd_channel(scmd)) { 1590 if (rtn == FAST_IO_FAIL) 1591 scsi_eh_finish_cmd(scmd, 1592 done_q); 1593 else 1594 list_move_tail(&scmd->eh_entry, 1595 &check_list); 1596 } 1597 } 1598 } else { 1599 SCSI_LOG_ERROR_RECOVERY(3, 1600 shost_printk(KERN_INFO, shost, 1601 "%s: BRST failed chan: %d\n", 1602 current->comm, channel)); 1603 } 1604 } 1605 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1606 } 1607 1608 /** 1609 * scsi_eh_host_reset - send a host reset 1610 * @shost: host to be reset. 1611 * @work_q: &list_head for pending commands. 1612 * @done_q: &list_head for processed commands. 1613 */ 1614 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1615 struct list_head *work_q, 1616 struct list_head *done_q) 1617 { 1618 struct scsi_cmnd *scmd, *next; 1619 LIST_HEAD(check_list); 1620 int rtn; 1621 1622 if (!list_empty(work_q)) { 1623 scmd = list_entry(work_q->next, 1624 struct scsi_cmnd, eh_entry); 1625 1626 SCSI_LOG_ERROR_RECOVERY(3, 1627 shost_printk(KERN_INFO, shost, 1628 "%s: Sending HRST\n", 1629 current->comm)); 1630 1631 rtn = scsi_try_host_reset(scmd); 1632 if (rtn == SUCCESS) { 1633 list_splice_init(work_q, &check_list); 1634 } else if (rtn == FAST_IO_FAIL) { 1635 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1636 scsi_eh_finish_cmd(scmd, done_q); 1637 } 1638 } else { 1639 SCSI_LOG_ERROR_RECOVERY(3, 1640 shost_printk(KERN_INFO, shost, 1641 "%s: HRST failed\n", 1642 current->comm)); 1643 } 1644 } 1645 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1646 } 1647 1648 /** 1649 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1650 * @work_q: &list_head for pending commands. 1651 * @done_q: &list_head for processed commands. 1652 */ 1653 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1654 struct list_head *done_q) 1655 { 1656 struct scsi_cmnd *scmd, *next; 1657 struct scsi_device *sdev; 1658 1659 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1660 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1661 "not ready after error recovery\n"); 1662 sdev = scmd->device; 1663 1664 mutex_lock(&sdev->state_mutex); 1665 scsi_device_set_state(sdev, SDEV_OFFLINE); 1666 mutex_unlock(&sdev->state_mutex); 1667 1668 scsi_eh_finish_cmd(scmd, done_q); 1669 } 1670 return; 1671 } 1672 1673 /** 1674 * scsi_noretry_cmd - determine if command should be failed fast 1675 * @scmd: SCSI cmd to examine. 1676 */ 1677 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1678 { 1679 switch (host_byte(scmd->result)) { 1680 case DID_OK: 1681 break; 1682 case DID_TIME_OUT: 1683 goto check_type; 1684 case DID_BUS_BUSY: 1685 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1686 case DID_PARITY: 1687 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1688 case DID_ERROR: 1689 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1690 status_byte(scmd->result) == RESERVATION_CONFLICT) 1691 return 0; 1692 /* fall through */ 1693 case DID_SOFT_ERROR: 1694 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1695 } 1696 1697 if (status_byte(scmd->result) != CHECK_CONDITION) 1698 return 0; 1699 1700 check_type: 1701 /* 1702 * assume caller has checked sense and determined 1703 * the check condition was retryable. 1704 */ 1705 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1706 blk_rq_is_passthrough(scmd->request)) 1707 return 1; 1708 else 1709 return 0; 1710 } 1711 1712 /** 1713 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1714 * @scmd: SCSI cmd to examine. 1715 * 1716 * Notes: 1717 * This is *only* called when we are examining the status after sending 1718 * out the actual data command. any commands that are queued for error 1719 * recovery (e.g. test_unit_ready) do *not* come through here. 1720 * 1721 * When this routine returns failed, it means the error handler thread 1722 * is woken. In cases where the error code indicates an error that 1723 * doesn't require the error handler read (i.e. we don't need to 1724 * abort/reset), this function should return SUCCESS. 1725 */ 1726 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1727 { 1728 int rtn; 1729 1730 /* 1731 * if the device is offline, then we clearly just pass the result back 1732 * up to the top level. 1733 */ 1734 if (!scsi_device_online(scmd->device)) { 1735 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1736 "%s: device offline - report as SUCCESS\n", __func__)); 1737 return SUCCESS; 1738 } 1739 1740 /* 1741 * first check the host byte, to see if there is anything in there 1742 * that would indicate what we need to do. 1743 */ 1744 switch (host_byte(scmd->result)) { 1745 case DID_PASSTHROUGH: 1746 /* 1747 * no matter what, pass this through to the upper layer. 1748 * nuke this special code so that it looks like we are saying 1749 * did_ok. 1750 */ 1751 scmd->result &= 0xff00ffff; 1752 return SUCCESS; 1753 case DID_OK: 1754 /* 1755 * looks good. drop through, and check the next byte. 1756 */ 1757 break; 1758 case DID_ABORT: 1759 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1760 set_host_byte(scmd, DID_TIME_OUT); 1761 return SUCCESS; 1762 } 1763 /* FALLTHROUGH */ 1764 case DID_NO_CONNECT: 1765 case DID_BAD_TARGET: 1766 /* 1767 * note - this means that we just report the status back 1768 * to the top level driver, not that we actually think 1769 * that it indicates SUCCESS. 1770 */ 1771 return SUCCESS; 1772 case DID_SOFT_ERROR: 1773 /* 1774 * when the low level driver returns did_soft_error, 1775 * it is responsible for keeping an internal retry counter 1776 * in order to avoid endless loops (db) 1777 */ 1778 goto maybe_retry; 1779 case DID_IMM_RETRY: 1780 return NEEDS_RETRY; 1781 1782 case DID_REQUEUE: 1783 return ADD_TO_MLQUEUE; 1784 case DID_TRANSPORT_DISRUPTED: 1785 /* 1786 * LLD/transport was disrupted during processing of the IO. 1787 * The transport class is now blocked/blocking, 1788 * and the transport will decide what to do with the IO 1789 * based on its timers and recovery capablilities if 1790 * there are enough retries. 1791 */ 1792 goto maybe_retry; 1793 case DID_TRANSPORT_FAILFAST: 1794 /* 1795 * The transport decided to failfast the IO (most likely 1796 * the fast io fail tmo fired), so send IO directly upwards. 1797 */ 1798 return SUCCESS; 1799 case DID_ERROR: 1800 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1801 status_byte(scmd->result) == RESERVATION_CONFLICT) 1802 /* 1803 * execute reservation conflict processing code 1804 * lower down 1805 */ 1806 break; 1807 /* fallthrough */ 1808 case DID_BUS_BUSY: 1809 case DID_PARITY: 1810 goto maybe_retry; 1811 case DID_TIME_OUT: 1812 /* 1813 * when we scan the bus, we get timeout messages for 1814 * these commands if there is no device available. 1815 * other hosts report did_no_connect for the same thing. 1816 */ 1817 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1818 scmd->cmnd[0] == INQUIRY)) { 1819 return SUCCESS; 1820 } else { 1821 return FAILED; 1822 } 1823 case DID_RESET: 1824 return SUCCESS; 1825 default: 1826 return FAILED; 1827 } 1828 1829 /* 1830 * next, check the message byte. 1831 */ 1832 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1833 return FAILED; 1834 1835 /* 1836 * check the status byte to see if this indicates anything special. 1837 */ 1838 switch (status_byte(scmd->result)) { 1839 case QUEUE_FULL: 1840 scsi_handle_queue_full(scmd->device); 1841 /* 1842 * the case of trying to send too many commands to a 1843 * tagged queueing device. 1844 */ 1845 /* FALLTHROUGH */ 1846 case BUSY: 1847 /* 1848 * device can't talk to us at the moment. Should only 1849 * occur (SAM-3) when the task queue is empty, so will cause 1850 * the empty queue handling to trigger a stall in the 1851 * device. 1852 */ 1853 return ADD_TO_MLQUEUE; 1854 case GOOD: 1855 if (scmd->cmnd[0] == REPORT_LUNS) 1856 scmd->device->sdev_target->expecting_lun_change = 0; 1857 scsi_handle_queue_ramp_up(scmd->device); 1858 /* FALLTHROUGH */ 1859 case COMMAND_TERMINATED: 1860 return SUCCESS; 1861 case TASK_ABORTED: 1862 goto maybe_retry; 1863 case CHECK_CONDITION: 1864 rtn = scsi_check_sense(scmd); 1865 if (rtn == NEEDS_RETRY) 1866 goto maybe_retry; 1867 /* if rtn == FAILED, we have no sense information; 1868 * returning FAILED will wake the error handler thread 1869 * to collect the sense and redo the decide 1870 * disposition */ 1871 return rtn; 1872 case CONDITION_GOOD: 1873 case INTERMEDIATE_GOOD: 1874 case INTERMEDIATE_C_GOOD: 1875 case ACA_ACTIVE: 1876 /* 1877 * who knows? FIXME(eric) 1878 */ 1879 return SUCCESS; 1880 1881 case RESERVATION_CONFLICT: 1882 sdev_printk(KERN_INFO, scmd->device, 1883 "reservation conflict\n"); 1884 set_host_byte(scmd, DID_NEXUS_FAILURE); 1885 return SUCCESS; /* causes immediate i/o error */ 1886 default: 1887 return FAILED; 1888 } 1889 return FAILED; 1890 1891 maybe_retry: 1892 1893 /* we requeue for retry because the error was retryable, and 1894 * the request was not marked fast fail. Note that above, 1895 * even if the request is marked fast fail, we still requeue 1896 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1897 if ((++scmd->retries) <= scmd->allowed 1898 && !scsi_noretry_cmd(scmd)) { 1899 return NEEDS_RETRY; 1900 } else { 1901 /* 1902 * no more retries - report this one back to upper level. 1903 */ 1904 return SUCCESS; 1905 } 1906 } 1907 1908 static void eh_lock_door_done(struct request *req, blk_status_t status) 1909 { 1910 __blk_put_request(req->q, req); 1911 } 1912 1913 /** 1914 * scsi_eh_lock_door - Prevent medium removal for the specified device 1915 * @sdev: SCSI device to prevent medium removal 1916 * 1917 * Locking: 1918 * We must be called from process context. 1919 * 1920 * Notes: 1921 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1922 * head of the devices request queue, and continue. 1923 */ 1924 static void scsi_eh_lock_door(struct scsi_device *sdev) 1925 { 1926 struct request *req; 1927 struct scsi_request *rq; 1928 1929 /* 1930 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a 1931 * request becomes available 1932 */ 1933 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL); 1934 if (IS_ERR(req)) 1935 return; 1936 rq = scsi_req(req); 1937 1938 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 1939 rq->cmd[1] = 0; 1940 rq->cmd[2] = 0; 1941 rq->cmd[3] = 0; 1942 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 1943 rq->cmd[5] = 0; 1944 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 1945 1946 req->rq_flags |= RQF_QUIET; 1947 req->timeout = 10 * HZ; 1948 rq->retries = 5; 1949 1950 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); 1951 } 1952 1953 /** 1954 * scsi_restart_operations - restart io operations to the specified host. 1955 * @shost: Host we are restarting. 1956 * 1957 * Notes: 1958 * When we entered the error handler, we blocked all further i/o to 1959 * this device. we need to 'reverse' this process. 1960 */ 1961 static void scsi_restart_operations(struct Scsi_Host *shost) 1962 { 1963 struct scsi_device *sdev; 1964 unsigned long flags; 1965 1966 /* 1967 * If the door was locked, we need to insert a door lock request 1968 * onto the head of the SCSI request queue for the device. There 1969 * is no point trying to lock the door of an off-line device. 1970 */ 1971 shost_for_each_device(sdev, shost) { 1972 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 1973 scsi_eh_lock_door(sdev); 1974 sdev->was_reset = 0; 1975 } 1976 } 1977 1978 /* 1979 * next free up anything directly waiting upon the host. this 1980 * will be requests for character device operations, and also for 1981 * ioctls to queued block devices. 1982 */ 1983 SCSI_LOG_ERROR_RECOVERY(3, 1984 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 1985 1986 spin_lock_irqsave(shost->host_lock, flags); 1987 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1988 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1989 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1990 spin_unlock_irqrestore(shost->host_lock, flags); 1991 1992 wake_up(&shost->host_wait); 1993 1994 /* 1995 * finally we need to re-initiate requests that may be pending. we will 1996 * have had everything blocked while error handling is taking place, and 1997 * now that error recovery is done, we will need to ensure that these 1998 * requests are started. 1999 */ 2000 scsi_run_host_queues(shost); 2001 2002 /* 2003 * if eh is active and host_eh_scheduled is pending we need to re-run 2004 * recovery. we do this check after scsi_run_host_queues() to allow 2005 * everything pent up since the last eh run a chance to make forward 2006 * progress before we sync again. Either we'll immediately re-run 2007 * recovery or scsi_device_unbusy() will wake us again when these 2008 * pending commands complete. 2009 */ 2010 spin_lock_irqsave(shost->host_lock, flags); 2011 if (shost->host_eh_scheduled) 2012 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2013 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2014 spin_unlock_irqrestore(shost->host_lock, flags); 2015 } 2016 2017 /** 2018 * scsi_eh_ready_devs - check device ready state and recover if not. 2019 * @shost: host to be recovered. 2020 * @work_q: &list_head for pending commands. 2021 * @done_q: &list_head for processed commands. 2022 */ 2023 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2024 struct list_head *work_q, 2025 struct list_head *done_q) 2026 { 2027 if (!scsi_eh_stu(shost, work_q, done_q)) 2028 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2029 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2030 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2031 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2032 scsi_eh_offline_sdevs(work_q, 2033 done_q); 2034 } 2035 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2036 2037 /** 2038 * scsi_eh_flush_done_q - finish processed commands or retry them. 2039 * @done_q: list_head of processed commands. 2040 */ 2041 void scsi_eh_flush_done_q(struct list_head *done_q) 2042 { 2043 struct scsi_cmnd *scmd, *next; 2044 2045 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2046 list_del_init(&scmd->eh_entry); 2047 if (scsi_device_online(scmd->device) && 2048 !scsi_noretry_cmd(scmd) && 2049 (++scmd->retries <= scmd->allowed)) { 2050 SCSI_LOG_ERROR_RECOVERY(3, 2051 scmd_printk(KERN_INFO, scmd, 2052 "%s: flush retry cmd\n", 2053 current->comm)); 2054 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2055 } else { 2056 /* 2057 * If just we got sense for the device (called 2058 * scsi_eh_get_sense), scmd->result is already 2059 * set, do not set DRIVER_TIMEOUT. 2060 */ 2061 if (!scmd->result) 2062 scmd->result |= (DRIVER_TIMEOUT << 24); 2063 SCSI_LOG_ERROR_RECOVERY(3, 2064 scmd_printk(KERN_INFO, scmd, 2065 "%s: flush finish cmd\n", 2066 current->comm)); 2067 scsi_finish_command(scmd); 2068 } 2069 } 2070 } 2071 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2072 2073 /** 2074 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2075 * @shost: Host to unjam. 2076 * 2077 * Notes: 2078 * When we come in here, we *know* that all commands on the bus have 2079 * either completed, failed or timed out. we also know that no further 2080 * commands are being sent to the host, so things are relatively quiet 2081 * and we have freedom to fiddle with things as we wish. 2082 * 2083 * This is only the *default* implementation. it is possible for 2084 * individual drivers to supply their own version of this function, and 2085 * if the maintainer wishes to do this, it is strongly suggested that 2086 * this function be taken as a template and modified. this function 2087 * was designed to correctly handle problems for about 95% of the 2088 * different cases out there, and it should always provide at least a 2089 * reasonable amount of error recovery. 2090 * 2091 * Any command marked 'failed' or 'timeout' must eventually have 2092 * scsi_finish_cmd() called for it. we do all of the retry stuff 2093 * here, so when we restart the host after we return it should have an 2094 * empty queue. 2095 */ 2096 static void scsi_unjam_host(struct Scsi_Host *shost) 2097 { 2098 unsigned long flags; 2099 LIST_HEAD(eh_work_q); 2100 LIST_HEAD(eh_done_q); 2101 2102 spin_lock_irqsave(shost->host_lock, flags); 2103 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2104 spin_unlock_irqrestore(shost->host_lock, flags); 2105 2106 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2107 2108 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2109 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2110 2111 spin_lock_irqsave(shost->host_lock, flags); 2112 if (shost->eh_deadline != -1) 2113 shost->last_reset = 0; 2114 spin_unlock_irqrestore(shost->host_lock, flags); 2115 scsi_eh_flush_done_q(&eh_done_q); 2116 } 2117 2118 /** 2119 * scsi_error_handler - SCSI error handler thread 2120 * @data: Host for which we are running. 2121 * 2122 * Notes: 2123 * This is the main error handling loop. This is run as a kernel thread 2124 * for every SCSI host and handles all error handling activity. 2125 */ 2126 int scsi_error_handler(void *data) 2127 { 2128 struct Scsi_Host *shost = data; 2129 2130 /* 2131 * We use TASK_INTERRUPTIBLE so that the thread is not 2132 * counted against the load average as a running process. 2133 * We never actually get interrupted because kthread_run 2134 * disables signal delivery for the created thread. 2135 */ 2136 while (true) { 2137 /* 2138 * The sequence in kthread_stop() sets the stop flag first 2139 * then wakes the process. To avoid missed wakeups, the task 2140 * should always be in a non running state before the stop 2141 * flag is checked 2142 */ 2143 set_current_state(TASK_INTERRUPTIBLE); 2144 if (kthread_should_stop()) 2145 break; 2146 2147 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2148 shost->host_failed != atomic_read(&shost->host_busy)) { 2149 SCSI_LOG_ERROR_RECOVERY(1, 2150 shost_printk(KERN_INFO, shost, 2151 "scsi_eh_%d: sleeping\n", 2152 shost->host_no)); 2153 schedule(); 2154 continue; 2155 } 2156 2157 __set_current_state(TASK_RUNNING); 2158 SCSI_LOG_ERROR_RECOVERY(1, 2159 shost_printk(KERN_INFO, shost, 2160 "scsi_eh_%d: waking up %d/%d/%d\n", 2161 shost->host_no, shost->host_eh_scheduled, 2162 shost->host_failed, 2163 atomic_read(&shost->host_busy))); 2164 2165 /* 2166 * We have a host that is failing for some reason. Figure out 2167 * what we need to do to get it up and online again (if we can). 2168 * If we fail, we end up taking the thing offline. 2169 */ 2170 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2171 SCSI_LOG_ERROR_RECOVERY(1, 2172 shost_printk(KERN_ERR, shost, 2173 "scsi_eh_%d: unable to autoresume\n", 2174 shost->host_no)); 2175 continue; 2176 } 2177 2178 if (shost->transportt->eh_strategy_handler) 2179 shost->transportt->eh_strategy_handler(shost); 2180 else 2181 scsi_unjam_host(shost); 2182 2183 /* All scmds have been handled */ 2184 shost->host_failed = 0; 2185 2186 /* 2187 * Note - if the above fails completely, the action is to take 2188 * individual devices offline and flush the queue of any 2189 * outstanding requests that may have been pending. When we 2190 * restart, we restart any I/O to any other devices on the bus 2191 * which are still online. 2192 */ 2193 scsi_restart_operations(shost); 2194 if (!shost->eh_noresume) 2195 scsi_autopm_put_host(shost); 2196 } 2197 __set_current_state(TASK_RUNNING); 2198 2199 SCSI_LOG_ERROR_RECOVERY(1, 2200 shost_printk(KERN_INFO, shost, 2201 "Error handler scsi_eh_%d exiting\n", 2202 shost->host_no)); 2203 shost->ehandler = NULL; 2204 return 0; 2205 } 2206 2207 /* 2208 * Function: scsi_report_bus_reset() 2209 * 2210 * Purpose: Utility function used by low-level drivers to report that 2211 * they have observed a bus reset on the bus being handled. 2212 * 2213 * Arguments: shost - Host in question 2214 * channel - channel on which reset was observed. 2215 * 2216 * Returns: Nothing 2217 * 2218 * Lock status: Host lock must be held. 2219 * 2220 * Notes: This only needs to be called if the reset is one which 2221 * originates from an unknown location. Resets originated 2222 * by the mid-level itself don't need to call this, but there 2223 * should be no harm. 2224 * 2225 * The main purpose of this is to make sure that a CHECK_CONDITION 2226 * is properly treated. 2227 */ 2228 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2229 { 2230 struct scsi_device *sdev; 2231 2232 __shost_for_each_device(sdev, shost) { 2233 if (channel == sdev_channel(sdev)) 2234 __scsi_report_device_reset(sdev, NULL); 2235 } 2236 } 2237 EXPORT_SYMBOL(scsi_report_bus_reset); 2238 2239 /* 2240 * Function: scsi_report_device_reset() 2241 * 2242 * Purpose: Utility function used by low-level drivers to report that 2243 * they have observed a device reset on the device being handled. 2244 * 2245 * Arguments: shost - Host in question 2246 * channel - channel on which reset was observed 2247 * target - target on which reset was observed 2248 * 2249 * Returns: Nothing 2250 * 2251 * Lock status: Host lock must be held 2252 * 2253 * Notes: This only needs to be called if the reset is one which 2254 * originates from an unknown location. Resets originated 2255 * by the mid-level itself don't need to call this, but there 2256 * should be no harm. 2257 * 2258 * The main purpose of this is to make sure that a CHECK_CONDITION 2259 * is properly treated. 2260 */ 2261 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2262 { 2263 struct scsi_device *sdev; 2264 2265 __shost_for_each_device(sdev, shost) { 2266 if (channel == sdev_channel(sdev) && 2267 target == sdev_id(sdev)) 2268 __scsi_report_device_reset(sdev, NULL); 2269 } 2270 } 2271 EXPORT_SYMBOL(scsi_report_device_reset); 2272 2273 static void 2274 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2275 { 2276 } 2277 2278 /** 2279 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2280 * @dev: scsi_device to operate on 2281 * @arg: reset type (see sg.h) 2282 */ 2283 int 2284 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2285 { 2286 struct scsi_cmnd *scmd; 2287 struct Scsi_Host *shost = dev->host; 2288 struct request *rq; 2289 unsigned long flags; 2290 int error = 0, rtn, val; 2291 2292 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2293 return -EACCES; 2294 2295 error = get_user(val, arg); 2296 if (error) 2297 return error; 2298 2299 if (scsi_autopm_get_host(shost) < 0) 2300 return -EIO; 2301 2302 error = -EIO; 2303 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2304 shost->hostt->cmd_size, GFP_KERNEL); 2305 if (!rq) 2306 goto out_put_autopm_host; 2307 blk_rq_init(NULL, rq); 2308 2309 scmd = (struct scsi_cmnd *)(rq + 1); 2310 scsi_init_command(dev, scmd); 2311 scmd->request = rq; 2312 scmd->cmnd = scsi_req(rq)->cmd; 2313 2314 scmd->scsi_done = scsi_reset_provider_done_command; 2315 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2316 2317 scmd->cmd_len = 0; 2318 2319 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2320 2321 spin_lock_irqsave(shost->host_lock, flags); 2322 shost->tmf_in_progress = 1; 2323 spin_unlock_irqrestore(shost->host_lock, flags); 2324 2325 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2326 case SG_SCSI_RESET_NOTHING: 2327 rtn = SUCCESS; 2328 break; 2329 case SG_SCSI_RESET_DEVICE: 2330 rtn = scsi_try_bus_device_reset(scmd); 2331 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2332 break; 2333 /* FALLTHROUGH */ 2334 case SG_SCSI_RESET_TARGET: 2335 rtn = scsi_try_target_reset(scmd); 2336 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2337 break; 2338 /* FALLTHROUGH */ 2339 case SG_SCSI_RESET_BUS: 2340 rtn = scsi_try_bus_reset(scmd); 2341 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2342 break; 2343 /* FALLTHROUGH */ 2344 case SG_SCSI_RESET_HOST: 2345 rtn = scsi_try_host_reset(scmd); 2346 if (rtn == SUCCESS) 2347 break; 2348 /* FALLTHROUGH */ 2349 default: 2350 rtn = FAILED; 2351 break; 2352 } 2353 2354 error = (rtn == SUCCESS) ? 0 : -EIO; 2355 2356 spin_lock_irqsave(shost->host_lock, flags); 2357 shost->tmf_in_progress = 0; 2358 spin_unlock_irqrestore(shost->host_lock, flags); 2359 2360 /* 2361 * be sure to wake up anyone who was sleeping or had their queue 2362 * suspended while we performed the TMF. 2363 */ 2364 SCSI_LOG_ERROR_RECOVERY(3, 2365 shost_printk(KERN_INFO, shost, 2366 "waking up host to restart after TMF\n")); 2367 2368 wake_up(&shost->host_wait); 2369 scsi_run_host_queues(shost); 2370 2371 scsi_put_command(scmd); 2372 kfree(rq); 2373 2374 out_put_autopm_host: 2375 scsi_autopm_put_host(shost); 2376 return error; 2377 } 2378 EXPORT_SYMBOL(scsi_ioctl_reset); 2379 2380 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2381 struct scsi_sense_hdr *sshdr) 2382 { 2383 return scsi_normalize_sense(cmd->sense_buffer, 2384 SCSI_SENSE_BUFFERSIZE, sshdr); 2385 } 2386 EXPORT_SYMBOL(scsi_command_normalize_sense); 2387 2388 /** 2389 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2390 * @sense_buffer: byte array of sense data 2391 * @sb_len: number of valid bytes in sense_buffer 2392 * @info_out: pointer to 64 integer where 8 or 4 byte information 2393 * field will be placed if found. 2394 * 2395 * Return value: 2396 * true if information field found, false if not found. 2397 */ 2398 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2399 u64 *info_out) 2400 { 2401 const u8 * ucp; 2402 2403 if (sb_len < 7) 2404 return false; 2405 switch (sense_buffer[0] & 0x7f) { 2406 case 0x70: 2407 case 0x71: 2408 if (sense_buffer[0] & 0x80) { 2409 *info_out = get_unaligned_be32(&sense_buffer[3]); 2410 return true; 2411 } 2412 return false; 2413 case 0x72: 2414 case 0x73: 2415 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2416 0 /* info desc */); 2417 if (ucp && (0xa == ucp[1])) { 2418 *info_out = get_unaligned_be64(&ucp[4]); 2419 return true; 2420 } 2421 return false; 2422 default: 2423 return false; 2424 } 2425 } 2426 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2427