1 /* 2 * scsi_error.c Copyright (C) 1997 Eric Youngdale 3 * 4 * SCSI error/timeout handling 5 * Initial versions: Eric Youngdale. Based upon conversations with 6 * Leonard Zubkoff and David Miller at Linux Expo, 7 * ideas originating from all over the place. 8 * 9 * Restructured scsi_unjam_host and associated functions. 10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com) 11 * 12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and 13 * minor cleanups. 14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com) 15 */ 16 17 #include <linux/module.h> 18 #include <linux/sched.h> 19 #include <linux/gfp.h> 20 #include <linux/timer.h> 21 #include <linux/string.h> 22 #include <linux/kernel.h> 23 #include <linux/freezer.h> 24 #include <linux/kthread.h> 25 #include <linux/interrupt.h> 26 #include <linux/blkdev.h> 27 #include <linux/delay.h> 28 #include <linux/jiffies.h> 29 30 #include <scsi/scsi.h> 31 #include <scsi/scsi_cmnd.h> 32 #include <scsi/scsi_dbg.h> 33 #include <scsi/scsi_device.h> 34 #include <scsi/scsi_driver.h> 35 #include <scsi/scsi_eh.h> 36 #include <scsi/scsi_common.h> 37 #include <scsi/scsi_transport.h> 38 #include <scsi/scsi_host.h> 39 #include <scsi/scsi_ioctl.h> 40 #include <scsi/scsi_dh.h> 41 #include <scsi/sg.h> 42 43 #include "scsi_priv.h" 44 #include "scsi_logging.h" 45 #include "scsi_transport_api.h" 46 47 #include <trace/events/scsi.h> 48 49 #include <asm/unaligned.h> 50 51 static void scsi_eh_done(struct scsi_cmnd *scmd); 52 53 /* 54 * These should *probably* be handled by the host itself. 55 * Since it is allowed to sleep, it probably should. 56 */ 57 #define BUS_RESET_SETTLE_TIME (10) 58 #define HOST_RESET_SETTLE_TIME (10) 59 60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 61 static int scsi_try_to_abort_cmd(struct scsi_host_template *, 62 struct scsi_cmnd *); 63 64 /* called with shost->host_lock held */ 65 void scsi_eh_wakeup(struct Scsi_Host *shost) 66 { 67 if (atomic_read(&shost->host_busy) == shost->host_failed) { 68 trace_scsi_eh_wakeup(shost); 69 wake_up_process(shost->ehandler); 70 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 71 "Waking error handler thread\n")); 72 } 73 } 74 75 /** 76 * scsi_schedule_eh - schedule EH for SCSI host 77 * @shost: SCSI host to invoke error handling on. 78 * 79 * Schedule SCSI EH without scmd. 80 */ 81 void scsi_schedule_eh(struct Scsi_Host *shost) 82 { 83 unsigned long flags; 84 85 spin_lock_irqsave(shost->host_lock, flags); 86 87 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 88 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 89 shost->host_eh_scheduled++; 90 scsi_eh_wakeup(shost); 91 } 92 93 spin_unlock_irqrestore(shost->host_lock, flags); 94 } 95 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 96 97 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 98 { 99 if (!shost->last_reset || shost->eh_deadline == -1) 100 return 0; 101 102 /* 103 * 32bit accesses are guaranteed to be atomic 104 * (on all supported architectures), so instead 105 * of using a spinlock we can as well double check 106 * if eh_deadline has been set to 'off' during the 107 * time_before call. 108 */ 109 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 110 shost->eh_deadline > -1) 111 return 0; 112 113 return 1; 114 } 115 116 /** 117 * scmd_eh_abort_handler - Handle command aborts 118 * @work: command to be aborted. 119 */ 120 void 121 scmd_eh_abort_handler(struct work_struct *work) 122 { 123 struct scsi_cmnd *scmd = 124 container_of(work, struct scsi_cmnd, abort_work.work); 125 struct scsi_device *sdev = scmd->device; 126 int rtn; 127 128 if (scsi_host_eh_past_deadline(sdev->host)) { 129 SCSI_LOG_ERROR_RECOVERY(3, 130 scmd_printk(KERN_INFO, scmd, 131 "eh timeout, not aborting\n")); 132 } else { 133 SCSI_LOG_ERROR_RECOVERY(3, 134 scmd_printk(KERN_INFO, scmd, 135 "aborting command\n")); 136 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); 137 if (rtn == SUCCESS) { 138 set_host_byte(scmd, DID_TIME_OUT); 139 if (scsi_host_eh_past_deadline(sdev->host)) { 140 SCSI_LOG_ERROR_RECOVERY(3, 141 scmd_printk(KERN_INFO, scmd, 142 "eh timeout, not retrying " 143 "aborted command\n")); 144 } else if (!scsi_noretry_cmd(scmd) && 145 (++scmd->retries <= scmd->allowed)) { 146 SCSI_LOG_ERROR_RECOVERY(3, 147 scmd_printk(KERN_WARNING, scmd, 148 "retry aborted command\n")); 149 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 150 return; 151 } else { 152 SCSI_LOG_ERROR_RECOVERY(3, 153 scmd_printk(KERN_WARNING, scmd, 154 "finish aborted command\n")); 155 scsi_finish_command(scmd); 156 return; 157 } 158 } else { 159 SCSI_LOG_ERROR_RECOVERY(3, 160 scmd_printk(KERN_INFO, scmd, 161 "cmd abort %s\n", 162 (rtn == FAST_IO_FAIL) ? 163 "not send" : "failed")); 164 } 165 } 166 167 scsi_eh_scmd_add(scmd); 168 } 169 170 /** 171 * scsi_abort_command - schedule a command abort 172 * @scmd: scmd to abort. 173 * 174 * We only need to abort commands after a command timeout 175 */ 176 static int 177 scsi_abort_command(struct scsi_cmnd *scmd) 178 { 179 struct scsi_device *sdev = scmd->device; 180 struct Scsi_Host *shost = sdev->host; 181 unsigned long flags; 182 183 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 184 /* 185 * Retry after abort failed, escalate to next level. 186 */ 187 SCSI_LOG_ERROR_RECOVERY(3, 188 scmd_printk(KERN_INFO, scmd, 189 "previous abort failed\n")); 190 BUG_ON(delayed_work_pending(&scmd->abort_work)); 191 return FAILED; 192 } 193 194 spin_lock_irqsave(shost->host_lock, flags); 195 if (shost->eh_deadline != -1 && !shost->last_reset) 196 shost->last_reset = jiffies; 197 spin_unlock_irqrestore(shost->host_lock, flags); 198 199 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 200 SCSI_LOG_ERROR_RECOVERY(3, 201 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 202 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 203 return SUCCESS; 204 } 205 206 /** 207 * scsi_eh_reset - call into ->eh_action to reset internal counters 208 * @scmd: scmd to run eh on. 209 * 210 * The scsi driver might be carrying internal state about the 211 * devices, so we need to call into the driver to reset the 212 * internal state once the error handler is started. 213 */ 214 static void scsi_eh_reset(struct scsi_cmnd *scmd) 215 { 216 if (!blk_rq_is_passthrough(scmd->request)) { 217 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 218 if (sdrv->eh_reset) 219 sdrv->eh_reset(scmd); 220 } 221 } 222 223 /** 224 * scsi_eh_scmd_add - add scsi cmd to error handling. 225 * @scmd: scmd to run eh on. 226 */ 227 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 228 { 229 struct Scsi_Host *shost = scmd->device->host; 230 unsigned long flags; 231 int ret; 232 233 WARN_ON_ONCE(!shost->ehandler); 234 235 spin_lock_irqsave(shost->host_lock, flags); 236 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 237 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 238 WARN_ON_ONCE(ret); 239 } 240 if (shost->eh_deadline != -1 && !shost->last_reset) 241 shost->last_reset = jiffies; 242 243 scsi_eh_reset(scmd); 244 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 245 shost->host_failed++; 246 scsi_eh_wakeup(shost); 247 spin_unlock_irqrestore(shost->host_lock, flags); 248 } 249 250 /** 251 * scsi_times_out - Timeout function for normal scsi commands. 252 * @req: request that is timing out. 253 * 254 * Notes: 255 * We do not need to lock this. There is the potential for a race 256 * only in that the normal completion handling might run, but if the 257 * normal completion function determines that the timer has already 258 * fired, then it mustn't do anything. 259 */ 260 enum blk_eh_timer_return scsi_times_out(struct request *req) 261 { 262 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 263 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; 264 struct Scsi_Host *host = scmd->device->host; 265 266 trace_scsi_dispatch_cmd_timeout(scmd); 267 scsi_log_completion(scmd, TIMEOUT_ERROR); 268 269 if (host->eh_deadline != -1 && !host->last_reset) 270 host->last_reset = jiffies; 271 272 if (host->hostt->eh_timed_out) 273 rtn = host->hostt->eh_timed_out(scmd); 274 275 if (rtn == BLK_EH_NOT_HANDLED) { 276 if (scsi_abort_command(scmd) != SUCCESS) { 277 set_host_byte(scmd, DID_TIME_OUT); 278 scsi_eh_scmd_add(scmd); 279 } 280 } 281 282 return rtn; 283 } 284 285 /** 286 * scsi_block_when_processing_errors - Prevent cmds from being queued. 287 * @sdev: Device on which we are performing recovery. 288 * 289 * Description: 290 * We block until the host is out of error recovery, and then check to 291 * see whether the host or the device is offline. 292 * 293 * Return value: 294 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 295 */ 296 int scsi_block_when_processing_errors(struct scsi_device *sdev) 297 { 298 int online; 299 300 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 301 302 online = scsi_device_online(sdev); 303 304 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev, 305 "%s: rtn: %d\n", __func__, online)); 306 307 return online; 308 } 309 EXPORT_SYMBOL(scsi_block_when_processing_errors); 310 311 #ifdef CONFIG_SCSI_LOGGING 312 /** 313 * scsi_eh_prt_fail_stats - Log info on failures. 314 * @shost: scsi host being recovered. 315 * @work_q: Queue of scsi cmds to process. 316 */ 317 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 318 struct list_head *work_q) 319 { 320 struct scsi_cmnd *scmd; 321 struct scsi_device *sdev; 322 int total_failures = 0; 323 int cmd_failed = 0; 324 int cmd_cancel = 0; 325 int devices_failed = 0; 326 327 shost_for_each_device(sdev, shost) { 328 list_for_each_entry(scmd, work_q, eh_entry) { 329 if (scmd->device == sdev) { 330 ++total_failures; 331 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 332 ++cmd_cancel; 333 else 334 ++cmd_failed; 335 } 336 } 337 338 if (cmd_cancel || cmd_failed) { 339 SCSI_LOG_ERROR_RECOVERY(3, 340 shost_printk(KERN_INFO, shost, 341 "%s: cmds failed: %d, cancel: %d\n", 342 __func__, cmd_failed, 343 cmd_cancel)); 344 cmd_cancel = 0; 345 cmd_failed = 0; 346 ++devices_failed; 347 } 348 } 349 350 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 351 "Total of %d commands on %d" 352 " devices require eh work\n", 353 total_failures, devices_failed)); 354 } 355 #endif 356 357 /** 358 * scsi_report_lun_change - Set flag on all *other* devices on the same target 359 * to indicate that a UNIT ATTENTION is expected. 360 * @sdev: Device reporting the UNIT ATTENTION 361 */ 362 static void scsi_report_lun_change(struct scsi_device *sdev) 363 { 364 sdev->sdev_target->expecting_lun_change = 1; 365 } 366 367 /** 368 * scsi_report_sense - Examine scsi sense information and log messages for 369 * certain conditions, also issue uevents for some of them. 370 * @sdev: Device reporting the sense code 371 * @sshdr: sshdr to be examined 372 */ 373 static void scsi_report_sense(struct scsi_device *sdev, 374 struct scsi_sense_hdr *sshdr) 375 { 376 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 377 378 if (sshdr->sense_key == UNIT_ATTENTION) { 379 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 380 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 381 sdev_printk(KERN_WARNING, sdev, 382 "Inquiry data has changed"); 383 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 384 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 385 scsi_report_lun_change(sdev); 386 sdev_printk(KERN_WARNING, sdev, 387 "Warning! Received an indication that the " 388 "LUN assignments on this target have " 389 "changed. The Linux SCSI layer does not " 390 "automatically remap LUN assignments.\n"); 391 } else if (sshdr->asc == 0x3f) 392 sdev_printk(KERN_WARNING, sdev, 393 "Warning! Received an indication that the " 394 "operating parameters on this target have " 395 "changed. The Linux SCSI layer does not " 396 "automatically adjust these parameters.\n"); 397 398 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 399 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 400 sdev_printk(KERN_WARNING, sdev, 401 "Warning! Received an indication that the " 402 "LUN reached a thin provisioning soft " 403 "threshold.\n"); 404 } 405 406 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 407 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 408 sdev_printk(KERN_WARNING, sdev, 409 "Mode parameters changed"); 410 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 411 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 412 sdev_printk(KERN_WARNING, sdev, 413 "Asymmetric access state changed"); 414 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 415 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 416 sdev_printk(KERN_WARNING, sdev, 417 "Capacity data has changed"); 418 } else if (sshdr->asc == 0x2a) 419 sdev_printk(KERN_WARNING, sdev, 420 "Parameters changed"); 421 } 422 423 if (evt_type != SDEV_EVT_MAXBITS) { 424 set_bit(evt_type, sdev->pending_events); 425 schedule_work(&sdev->event_work); 426 } 427 } 428 429 /** 430 * scsi_check_sense - Examine scsi cmd sense 431 * @scmd: Cmd to have sense checked. 432 * 433 * Return value: 434 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 435 * 436 * Notes: 437 * When a deferred error is detected the current command has 438 * not been executed and needs retrying. 439 */ 440 int scsi_check_sense(struct scsi_cmnd *scmd) 441 { 442 struct scsi_device *sdev = scmd->device; 443 struct scsi_sense_hdr sshdr; 444 445 if (! scsi_command_normalize_sense(scmd, &sshdr)) 446 return FAILED; /* no valid sense data */ 447 448 scsi_report_sense(sdev, &sshdr); 449 450 if (scsi_sense_is_deferred(&sshdr)) 451 return NEEDS_RETRY; 452 453 if (sdev->handler && sdev->handler->check_sense) { 454 int rc; 455 456 rc = sdev->handler->check_sense(sdev, &sshdr); 457 if (rc != SCSI_RETURN_NOT_HANDLED) 458 return rc; 459 /* handler does not care. Drop down to default handling */ 460 } 461 462 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 463 /* 464 * nasty: for mid-layer issued TURs, we need to return the 465 * actual sense data without any recovery attempt. For eh 466 * issued ones, we need to try to recover and interpret 467 */ 468 return SUCCESS; 469 470 /* 471 * Previous logic looked for FILEMARK, EOM or ILI which are 472 * mainly associated with tapes and returned SUCCESS. 473 */ 474 if (sshdr.response_code == 0x70) { 475 /* fixed format */ 476 if (scmd->sense_buffer[2] & 0xe0) 477 return SUCCESS; 478 } else { 479 /* 480 * descriptor format: look for "stream commands sense data 481 * descriptor" (see SSC-3). Assume single sense data 482 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 483 */ 484 if ((sshdr.additional_length > 3) && 485 (scmd->sense_buffer[8] == 0x4) && 486 (scmd->sense_buffer[11] & 0xe0)) 487 return SUCCESS; 488 } 489 490 switch (sshdr.sense_key) { 491 case NO_SENSE: 492 return SUCCESS; 493 case RECOVERED_ERROR: 494 return /* soft_error */ SUCCESS; 495 496 case ABORTED_COMMAND: 497 if (sshdr.asc == 0x10) /* DIF */ 498 return SUCCESS; 499 500 return NEEDS_RETRY; 501 case NOT_READY: 502 case UNIT_ATTENTION: 503 /* 504 * if we are expecting a cc/ua because of a bus reset that we 505 * performed, treat this just as a retry. otherwise this is 506 * information that we should pass up to the upper-level driver 507 * so that we can deal with it there. 508 */ 509 if (scmd->device->expecting_cc_ua) { 510 /* 511 * Because some device does not queue unit 512 * attentions correctly, we carefully check 513 * additional sense code and qualifier so as 514 * not to squash media change unit attention. 515 */ 516 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 517 scmd->device->expecting_cc_ua = 0; 518 return NEEDS_RETRY; 519 } 520 } 521 /* 522 * we might also expect a cc/ua if another LUN on the target 523 * reported a UA with an ASC/ASCQ of 3F 0E - 524 * REPORTED LUNS DATA HAS CHANGED. 525 */ 526 if (scmd->device->sdev_target->expecting_lun_change && 527 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 528 return NEEDS_RETRY; 529 /* 530 * if the device is in the process of becoming ready, we 531 * should retry. 532 */ 533 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 534 return NEEDS_RETRY; 535 /* 536 * if the device is not started, we need to wake 537 * the error handler to start the motor 538 */ 539 if (scmd->device->allow_restart && 540 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 541 return FAILED; 542 /* 543 * Pass the UA upwards for a determination in the completion 544 * functions. 545 */ 546 return SUCCESS; 547 548 /* these are not supported */ 549 case DATA_PROTECT: 550 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 551 /* Thin provisioning hard threshold reached */ 552 set_host_byte(scmd, DID_ALLOC_FAILURE); 553 return SUCCESS; 554 } 555 /* FALLTHROUGH */ 556 case COPY_ABORTED: 557 case VOLUME_OVERFLOW: 558 case MISCOMPARE: 559 case BLANK_CHECK: 560 set_host_byte(scmd, DID_TARGET_FAILURE); 561 return SUCCESS; 562 563 case MEDIUM_ERROR: 564 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 565 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 566 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 567 set_host_byte(scmd, DID_MEDIUM_ERROR); 568 return SUCCESS; 569 } 570 return NEEDS_RETRY; 571 572 case HARDWARE_ERROR: 573 if (scmd->device->retry_hwerror) 574 return ADD_TO_MLQUEUE; 575 else 576 set_host_byte(scmd, DID_TARGET_FAILURE); 577 /* FALLTHROUGH */ 578 579 case ILLEGAL_REQUEST: 580 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 581 sshdr.asc == 0x21 || /* Logical block address out of range */ 582 sshdr.asc == 0x24 || /* Invalid field in cdb */ 583 sshdr.asc == 0x26 || /* Parameter value invalid */ 584 sshdr.asc == 0x27) { /* Write protected */ 585 set_host_byte(scmd, DID_TARGET_FAILURE); 586 } 587 return SUCCESS; 588 589 default: 590 return SUCCESS; 591 } 592 } 593 EXPORT_SYMBOL_GPL(scsi_check_sense); 594 595 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 596 { 597 struct scsi_host_template *sht = sdev->host->hostt; 598 struct scsi_device *tmp_sdev; 599 600 if (!sht->track_queue_depth || 601 sdev->queue_depth >= sdev->max_queue_depth) 602 return; 603 604 if (time_before(jiffies, 605 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 606 return; 607 608 if (time_before(jiffies, 609 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 610 return; 611 612 /* 613 * Walk all devices of a target and do 614 * ramp up on them. 615 */ 616 shost_for_each_device(tmp_sdev, sdev->host) { 617 if (tmp_sdev->channel != sdev->channel || 618 tmp_sdev->id != sdev->id || 619 tmp_sdev->queue_depth == sdev->max_queue_depth) 620 continue; 621 622 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 623 sdev->last_queue_ramp_up = jiffies; 624 } 625 } 626 627 static void scsi_handle_queue_full(struct scsi_device *sdev) 628 { 629 struct scsi_host_template *sht = sdev->host->hostt; 630 struct scsi_device *tmp_sdev; 631 632 if (!sht->track_queue_depth) 633 return; 634 635 shost_for_each_device(tmp_sdev, sdev->host) { 636 if (tmp_sdev->channel != sdev->channel || 637 tmp_sdev->id != sdev->id) 638 continue; 639 /* 640 * We do not know the number of commands that were at 641 * the device when we got the queue full so we start 642 * from the highest possible value and work our way down. 643 */ 644 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 645 } 646 } 647 648 /** 649 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 650 * @scmd: SCSI cmd to examine. 651 * 652 * Notes: 653 * This is *only* called when we are examining the status of commands 654 * queued during error recovery. the main difference here is that we 655 * don't allow for the possibility of retries here, and we are a lot 656 * more restrictive about what we consider acceptable. 657 */ 658 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 659 { 660 /* 661 * first check the host byte, to see if there is anything in there 662 * that would indicate what we need to do. 663 */ 664 if (host_byte(scmd->result) == DID_RESET) { 665 /* 666 * rats. we are already in the error handler, so we now 667 * get to try and figure out what to do next. if the sense 668 * is valid, we have a pretty good idea of what to do. 669 * if not, we mark it as FAILED. 670 */ 671 return scsi_check_sense(scmd); 672 } 673 if (host_byte(scmd->result) != DID_OK) 674 return FAILED; 675 676 /* 677 * next, check the message byte. 678 */ 679 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 680 return FAILED; 681 682 /* 683 * now, check the status byte to see if this indicates 684 * anything special. 685 */ 686 switch (status_byte(scmd->result)) { 687 case GOOD: 688 scsi_handle_queue_ramp_up(scmd->device); 689 /* FALLTHROUGH */ 690 case COMMAND_TERMINATED: 691 return SUCCESS; 692 case CHECK_CONDITION: 693 return scsi_check_sense(scmd); 694 case CONDITION_GOOD: 695 case INTERMEDIATE_GOOD: 696 case INTERMEDIATE_C_GOOD: 697 /* 698 * who knows? FIXME(eric) 699 */ 700 return SUCCESS; 701 case RESERVATION_CONFLICT: 702 if (scmd->cmnd[0] == TEST_UNIT_READY) 703 /* it is a success, we probed the device and 704 * found it */ 705 return SUCCESS; 706 /* otherwise, we failed to send the command */ 707 return FAILED; 708 case QUEUE_FULL: 709 scsi_handle_queue_full(scmd->device); 710 /* fall through */ 711 case BUSY: 712 return NEEDS_RETRY; 713 default: 714 return FAILED; 715 } 716 return FAILED; 717 } 718 719 /** 720 * scsi_eh_done - Completion function for error handling. 721 * @scmd: Cmd that is done. 722 */ 723 static void scsi_eh_done(struct scsi_cmnd *scmd) 724 { 725 struct completion *eh_action; 726 727 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 728 "%s result: %x\n", __func__, scmd->result)); 729 730 eh_action = scmd->device->host->eh_action; 731 if (eh_action) 732 complete(eh_action); 733 } 734 735 /** 736 * scsi_try_host_reset - ask host adapter to reset itself 737 * @scmd: SCSI cmd to send host reset. 738 */ 739 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 740 { 741 unsigned long flags; 742 int rtn; 743 struct Scsi_Host *host = scmd->device->host; 744 struct scsi_host_template *hostt = host->hostt; 745 746 SCSI_LOG_ERROR_RECOVERY(3, 747 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 748 749 if (!hostt->eh_host_reset_handler) 750 return FAILED; 751 752 rtn = hostt->eh_host_reset_handler(scmd); 753 754 if (rtn == SUCCESS) { 755 if (!hostt->skip_settle_delay) 756 ssleep(HOST_RESET_SETTLE_TIME); 757 spin_lock_irqsave(host->host_lock, flags); 758 scsi_report_bus_reset(host, scmd_channel(scmd)); 759 spin_unlock_irqrestore(host->host_lock, flags); 760 } 761 762 return rtn; 763 } 764 765 /** 766 * scsi_try_bus_reset - ask host to perform a bus reset 767 * @scmd: SCSI cmd to send bus reset. 768 */ 769 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 770 { 771 unsigned long flags; 772 int rtn; 773 struct Scsi_Host *host = scmd->device->host; 774 struct scsi_host_template *hostt = host->hostt; 775 776 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 777 "%s: Snd Bus RST\n", __func__)); 778 779 if (!hostt->eh_bus_reset_handler) 780 return FAILED; 781 782 rtn = hostt->eh_bus_reset_handler(scmd); 783 784 if (rtn == SUCCESS) { 785 if (!hostt->skip_settle_delay) 786 ssleep(BUS_RESET_SETTLE_TIME); 787 spin_lock_irqsave(host->host_lock, flags); 788 scsi_report_bus_reset(host, scmd_channel(scmd)); 789 spin_unlock_irqrestore(host->host_lock, flags); 790 } 791 792 return rtn; 793 } 794 795 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 796 { 797 sdev->was_reset = 1; 798 sdev->expecting_cc_ua = 1; 799 } 800 801 /** 802 * scsi_try_target_reset - Ask host to perform a target reset 803 * @scmd: SCSI cmd used to send a target reset 804 * 805 * Notes: 806 * There is no timeout for this operation. if this operation is 807 * unreliable for a given host, then the host itself needs to put a 808 * timer on it, and set the host back to a consistent state prior to 809 * returning. 810 */ 811 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 812 { 813 unsigned long flags; 814 int rtn; 815 struct Scsi_Host *host = scmd->device->host; 816 struct scsi_host_template *hostt = host->hostt; 817 818 if (!hostt->eh_target_reset_handler) 819 return FAILED; 820 821 rtn = hostt->eh_target_reset_handler(scmd); 822 if (rtn == SUCCESS) { 823 spin_lock_irqsave(host->host_lock, flags); 824 __starget_for_each_device(scsi_target(scmd->device), NULL, 825 __scsi_report_device_reset); 826 spin_unlock_irqrestore(host->host_lock, flags); 827 } 828 829 return rtn; 830 } 831 832 /** 833 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 834 * @scmd: SCSI cmd used to send BDR 835 * 836 * Notes: 837 * There is no timeout for this operation. if this operation is 838 * unreliable for a given host, then the host itself needs to put a 839 * timer on it, and set the host back to a consistent state prior to 840 * returning. 841 */ 842 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 843 { 844 int rtn; 845 struct scsi_host_template *hostt = scmd->device->host->hostt; 846 847 if (!hostt->eh_device_reset_handler) 848 return FAILED; 849 850 rtn = hostt->eh_device_reset_handler(scmd); 851 if (rtn == SUCCESS) 852 __scsi_report_device_reset(scmd->device, NULL); 853 return rtn; 854 } 855 856 /** 857 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 858 * @hostt: SCSI driver host template 859 * @scmd: SCSI cmd used to send a target reset 860 * 861 * Return value: 862 * SUCCESS, FAILED, or FAST_IO_FAIL 863 * 864 * Notes: 865 * SUCCESS does not necessarily indicate that the command 866 * has been aborted; it only indicates that the LLDDs 867 * has cleared all references to that command. 868 * LLDDs should return FAILED only if an abort was required 869 * but could not be executed. LLDDs should return FAST_IO_FAIL 870 * if the device is temporarily unavailable (eg due to a 871 * link down on FibreChannel) 872 */ 873 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, 874 struct scsi_cmnd *scmd) 875 { 876 if (!hostt->eh_abort_handler) 877 return FAILED; 878 879 return hostt->eh_abort_handler(scmd); 880 } 881 882 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 883 { 884 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 885 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 886 if (scsi_try_target_reset(scmd) != SUCCESS) 887 if (scsi_try_bus_reset(scmd) != SUCCESS) 888 scsi_try_host_reset(scmd); 889 } 890 891 /** 892 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 893 * @scmd: SCSI command structure to hijack 894 * @ses: structure to save restore information 895 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 896 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 897 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 898 * 899 * This function is used to save a scsi command information before re-execution 900 * as part of the error recovery process. If @sense_bytes is 0 the command 901 * sent must be one that does not transfer any data. If @sense_bytes != 0 902 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 903 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 904 */ 905 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 906 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 907 { 908 struct scsi_device *sdev = scmd->device; 909 910 /* 911 * We need saved copies of a number of fields - this is because 912 * error handling may need to overwrite these with different values 913 * to run different commands, and once error handling is complete, 914 * we will need to restore these values prior to running the actual 915 * command. 916 */ 917 ses->cmd_len = scmd->cmd_len; 918 ses->cmnd = scmd->cmnd; 919 ses->data_direction = scmd->sc_data_direction; 920 ses->sdb = scmd->sdb; 921 ses->next_rq = scmd->request->next_rq; 922 ses->result = scmd->result; 923 ses->underflow = scmd->underflow; 924 ses->prot_op = scmd->prot_op; 925 ses->eh_eflags = scmd->eh_eflags; 926 927 scmd->prot_op = SCSI_PROT_NORMAL; 928 scmd->eh_eflags = 0; 929 scmd->cmnd = ses->eh_cmnd; 930 memset(scmd->cmnd, 0, BLK_MAX_CDB); 931 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 932 scmd->request->next_rq = NULL; 933 scmd->result = 0; 934 935 if (sense_bytes) { 936 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 937 sense_bytes); 938 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 939 scmd->sdb.length); 940 scmd->sdb.table.sgl = &ses->sense_sgl; 941 scmd->sc_data_direction = DMA_FROM_DEVICE; 942 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 943 scmd->cmnd[0] = REQUEST_SENSE; 944 scmd->cmnd[4] = scmd->sdb.length; 945 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 946 } else { 947 scmd->sc_data_direction = DMA_NONE; 948 if (cmnd) { 949 BUG_ON(cmnd_size > BLK_MAX_CDB); 950 memcpy(scmd->cmnd, cmnd, cmnd_size); 951 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 952 } 953 } 954 955 scmd->underflow = 0; 956 957 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 958 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 959 (sdev->lun << 5 & 0xe0); 960 961 /* 962 * Zero the sense buffer. The scsi spec mandates that any 963 * untransferred sense data should be interpreted as being zero. 964 */ 965 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 966 } 967 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 968 969 /** 970 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 971 * @scmd: SCSI command structure to restore 972 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 973 * 974 * Undo any damage done by above scsi_eh_prep_cmnd(). 975 */ 976 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 977 { 978 /* 979 * Restore original data 980 */ 981 scmd->cmd_len = ses->cmd_len; 982 scmd->cmnd = ses->cmnd; 983 scmd->sc_data_direction = ses->data_direction; 984 scmd->sdb = ses->sdb; 985 scmd->request->next_rq = ses->next_rq; 986 scmd->result = ses->result; 987 scmd->underflow = ses->underflow; 988 scmd->prot_op = ses->prot_op; 989 scmd->eh_eflags = ses->eh_eflags; 990 } 991 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 992 993 /** 994 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 995 * @scmd: SCSI command structure to hijack 996 * @cmnd: CDB to send 997 * @cmnd_size: size in bytes of @cmnd 998 * @timeout: timeout for this request 999 * @sense_bytes: size of sense data to copy or 0 1000 * 1001 * This function is used to send a scsi command down to a target device 1002 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1003 * 1004 * Return value: 1005 * SUCCESS or FAILED or NEEDS_RETRY 1006 */ 1007 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 1008 int cmnd_size, int timeout, unsigned sense_bytes) 1009 { 1010 struct scsi_device *sdev = scmd->device; 1011 struct Scsi_Host *shost = sdev->host; 1012 DECLARE_COMPLETION_ONSTACK(done); 1013 unsigned long timeleft = timeout; 1014 struct scsi_eh_save ses; 1015 const unsigned long stall_for = msecs_to_jiffies(100); 1016 int rtn; 1017 1018 retry: 1019 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1020 shost->eh_action = &done; 1021 1022 scsi_log_send(scmd); 1023 scmd->scsi_done = scsi_eh_done; 1024 rtn = shost->hostt->queuecommand(shost, scmd); 1025 if (rtn) { 1026 if (timeleft > stall_for) { 1027 scsi_eh_restore_cmnd(scmd, &ses); 1028 timeleft -= stall_for; 1029 msleep(jiffies_to_msecs(stall_for)); 1030 goto retry; 1031 } 1032 /* signal not to enter either branch of the if () below */ 1033 timeleft = 0; 1034 rtn = FAILED; 1035 } else { 1036 timeleft = wait_for_completion_timeout(&done, timeout); 1037 rtn = SUCCESS; 1038 } 1039 1040 shost->eh_action = NULL; 1041 1042 scsi_log_completion(scmd, rtn); 1043 1044 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1045 "%s timeleft: %ld\n", 1046 __func__, timeleft)); 1047 1048 /* 1049 * If there is time left scsi_eh_done got called, and we will examine 1050 * the actual status codes to see whether the command actually did 1051 * complete normally, else if we have a zero return and no time left, 1052 * the command must still be pending, so abort it and return FAILED. 1053 * If we never actually managed to issue the command, because 1054 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1055 * value above (so don't execute either branch of the if) 1056 */ 1057 if (timeleft) { 1058 rtn = scsi_eh_completed_normally(scmd); 1059 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1060 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1061 1062 switch (rtn) { 1063 case SUCCESS: 1064 case NEEDS_RETRY: 1065 case FAILED: 1066 break; 1067 case ADD_TO_MLQUEUE: 1068 rtn = NEEDS_RETRY; 1069 break; 1070 default: 1071 rtn = FAILED; 1072 break; 1073 } 1074 } else if (rtn != FAILED) { 1075 scsi_abort_eh_cmnd(scmd); 1076 rtn = FAILED; 1077 } 1078 1079 scsi_eh_restore_cmnd(scmd, &ses); 1080 1081 return rtn; 1082 } 1083 1084 /** 1085 * scsi_request_sense - Request sense data from a particular target. 1086 * @scmd: SCSI cmd for request sense. 1087 * 1088 * Notes: 1089 * Some hosts automatically obtain this information, others require 1090 * that we obtain it on our own. This function will *not* return until 1091 * the command either times out, or it completes. 1092 */ 1093 static int scsi_request_sense(struct scsi_cmnd *scmd) 1094 { 1095 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1096 } 1097 1098 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) 1099 { 1100 if (!blk_rq_is_passthrough(scmd->request)) { 1101 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1102 if (sdrv->eh_action) 1103 rtn = sdrv->eh_action(scmd, rtn); 1104 } 1105 return rtn; 1106 } 1107 1108 /** 1109 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1110 * @scmd: Original SCSI cmd that eh has finished. 1111 * @done_q: Queue for processed commands. 1112 * 1113 * Notes: 1114 * We don't want to use the normal command completion while we are are 1115 * still handling errors - it may cause other commands to be queued, 1116 * and that would disturb what we are doing. Thus we really want to 1117 * keep a list of pending commands for final completion, and once we 1118 * are ready to leave error handling we handle completion for real. 1119 */ 1120 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1121 { 1122 list_move_tail(&scmd->eh_entry, done_q); 1123 } 1124 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1125 1126 /** 1127 * scsi_eh_get_sense - Get device sense data. 1128 * @work_q: Queue of commands to process. 1129 * @done_q: Queue of processed commands. 1130 * 1131 * Description: 1132 * See if we need to request sense information. if so, then get it 1133 * now, so we have a better idea of what to do. 1134 * 1135 * Notes: 1136 * This has the unfortunate side effect that if a shost adapter does 1137 * not automatically request sense information, we end up shutting 1138 * it down before we request it. 1139 * 1140 * All drivers should request sense information internally these days, 1141 * so for now all I have to say is tough noogies if you end up in here. 1142 * 1143 * XXX: Long term this code should go away, but that needs an audit of 1144 * all LLDDs first. 1145 */ 1146 int scsi_eh_get_sense(struct list_head *work_q, 1147 struct list_head *done_q) 1148 { 1149 struct scsi_cmnd *scmd, *next; 1150 struct Scsi_Host *shost; 1151 int rtn; 1152 1153 /* 1154 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1155 * should not get sense. 1156 */ 1157 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1158 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1159 SCSI_SENSE_VALID(scmd)) 1160 continue; 1161 1162 shost = scmd->device->host; 1163 if (scsi_host_eh_past_deadline(shost)) { 1164 SCSI_LOG_ERROR_RECOVERY(3, 1165 scmd_printk(KERN_INFO, scmd, 1166 "%s: skip request sense, past eh deadline\n", 1167 current->comm)); 1168 break; 1169 } 1170 if (status_byte(scmd->result) != CHECK_CONDITION) 1171 /* 1172 * don't request sense if there's no check condition 1173 * status because the error we're processing isn't one 1174 * that has a sense code (and some devices get 1175 * confused by sense requests out of the blue) 1176 */ 1177 continue; 1178 1179 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1180 "%s: requesting sense\n", 1181 current->comm)); 1182 rtn = scsi_request_sense(scmd); 1183 if (rtn != SUCCESS) 1184 continue; 1185 1186 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1187 "sense requested, result %x\n", scmd->result)); 1188 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1189 1190 rtn = scsi_decide_disposition(scmd); 1191 1192 /* 1193 * if the result was normal, then just pass it along to the 1194 * upper level. 1195 */ 1196 if (rtn == SUCCESS) 1197 /* we don't want this command reissued, just 1198 * finished with the sense data, so set 1199 * retries to the max allowed to ensure it 1200 * won't get reissued */ 1201 scmd->retries = scmd->allowed; 1202 else if (rtn != NEEDS_RETRY) 1203 continue; 1204 1205 scsi_eh_finish_cmd(scmd, done_q); 1206 } 1207 1208 return list_empty(work_q); 1209 } 1210 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1211 1212 /** 1213 * scsi_eh_tur - Send TUR to device. 1214 * @scmd: &scsi_cmnd to send TUR 1215 * 1216 * Return value: 1217 * 0 - Device is ready. 1 - Device NOT ready. 1218 */ 1219 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1220 { 1221 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1222 int retry_cnt = 1, rtn; 1223 1224 retry_tur: 1225 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1226 scmd->device->eh_timeout, 0); 1227 1228 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1229 "%s return: %x\n", __func__, rtn)); 1230 1231 switch (rtn) { 1232 case NEEDS_RETRY: 1233 if (retry_cnt--) 1234 goto retry_tur; 1235 /*FALLTHRU*/ 1236 case SUCCESS: 1237 return 0; 1238 default: 1239 return 1; 1240 } 1241 } 1242 1243 /** 1244 * scsi_eh_test_devices - check if devices are responding from error recovery. 1245 * @cmd_list: scsi commands in error recovery. 1246 * @work_q: queue for commands which still need more error recovery 1247 * @done_q: queue for commands which are finished 1248 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1249 * 1250 * Decription: 1251 * Tests if devices are in a working state. Commands to devices now in 1252 * a working state are sent to the done_q while commands to devices which 1253 * are still failing to respond are returned to the work_q for more 1254 * processing. 1255 **/ 1256 static int scsi_eh_test_devices(struct list_head *cmd_list, 1257 struct list_head *work_q, 1258 struct list_head *done_q, int try_stu) 1259 { 1260 struct scsi_cmnd *scmd, *next; 1261 struct scsi_device *sdev; 1262 int finish_cmds; 1263 1264 while (!list_empty(cmd_list)) { 1265 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1266 sdev = scmd->device; 1267 1268 if (!try_stu) { 1269 if (scsi_host_eh_past_deadline(sdev->host)) { 1270 /* Push items back onto work_q */ 1271 list_splice_init(cmd_list, work_q); 1272 SCSI_LOG_ERROR_RECOVERY(3, 1273 sdev_printk(KERN_INFO, sdev, 1274 "%s: skip test device, past eh deadline", 1275 current->comm)); 1276 break; 1277 } 1278 } 1279 1280 finish_cmds = !scsi_device_online(scmd->device) || 1281 (try_stu && !scsi_eh_try_stu(scmd) && 1282 !scsi_eh_tur(scmd)) || 1283 !scsi_eh_tur(scmd); 1284 1285 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1286 if (scmd->device == sdev) { 1287 if (finish_cmds && 1288 (try_stu || 1289 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1290 scsi_eh_finish_cmd(scmd, done_q); 1291 else 1292 list_move_tail(&scmd->eh_entry, work_q); 1293 } 1294 } 1295 return list_empty(work_q); 1296 } 1297 1298 /** 1299 * scsi_eh_try_stu - Send START_UNIT to device. 1300 * @scmd: &scsi_cmnd to send START_UNIT 1301 * 1302 * Return value: 1303 * 0 - Device is ready. 1 - Device NOT ready. 1304 */ 1305 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1306 { 1307 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1308 1309 if (scmd->device->allow_restart) { 1310 int i, rtn = NEEDS_RETRY; 1311 1312 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1313 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1314 1315 if (rtn == SUCCESS) 1316 return 0; 1317 } 1318 1319 return 1; 1320 } 1321 1322 /** 1323 * scsi_eh_stu - send START_UNIT if needed 1324 * @shost: &scsi host being recovered. 1325 * @work_q: &list_head for pending commands. 1326 * @done_q: &list_head for processed commands. 1327 * 1328 * Notes: 1329 * If commands are failing due to not ready, initializing command required, 1330 * try revalidating the device, which will end up sending a start unit. 1331 */ 1332 static int scsi_eh_stu(struct Scsi_Host *shost, 1333 struct list_head *work_q, 1334 struct list_head *done_q) 1335 { 1336 struct scsi_cmnd *scmd, *stu_scmd, *next; 1337 struct scsi_device *sdev; 1338 1339 shost_for_each_device(sdev, shost) { 1340 if (scsi_host_eh_past_deadline(shost)) { 1341 SCSI_LOG_ERROR_RECOVERY(3, 1342 sdev_printk(KERN_INFO, sdev, 1343 "%s: skip START_UNIT, past eh deadline\n", 1344 current->comm)); 1345 break; 1346 } 1347 stu_scmd = NULL; 1348 list_for_each_entry(scmd, work_q, eh_entry) 1349 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1350 scsi_check_sense(scmd) == FAILED ) { 1351 stu_scmd = scmd; 1352 break; 1353 } 1354 1355 if (!stu_scmd) 1356 continue; 1357 1358 SCSI_LOG_ERROR_RECOVERY(3, 1359 sdev_printk(KERN_INFO, sdev, 1360 "%s: Sending START_UNIT\n", 1361 current->comm)); 1362 1363 if (!scsi_eh_try_stu(stu_scmd)) { 1364 if (!scsi_device_online(sdev) || 1365 !scsi_eh_tur(stu_scmd)) { 1366 list_for_each_entry_safe(scmd, next, 1367 work_q, eh_entry) { 1368 if (scmd->device == sdev && 1369 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1370 scsi_eh_finish_cmd(scmd, done_q); 1371 } 1372 } 1373 } else { 1374 SCSI_LOG_ERROR_RECOVERY(3, 1375 sdev_printk(KERN_INFO, sdev, 1376 "%s: START_UNIT failed\n", 1377 current->comm)); 1378 } 1379 } 1380 1381 return list_empty(work_q); 1382 } 1383 1384 1385 /** 1386 * scsi_eh_bus_device_reset - send bdr if needed 1387 * @shost: scsi host being recovered. 1388 * @work_q: &list_head for pending commands. 1389 * @done_q: &list_head for processed commands. 1390 * 1391 * Notes: 1392 * Try a bus device reset. Still, look to see whether we have multiple 1393 * devices that are jammed or not - if we have multiple devices, it 1394 * makes no sense to try bus_device_reset - we really would need to try 1395 * a bus_reset instead. 1396 */ 1397 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1398 struct list_head *work_q, 1399 struct list_head *done_q) 1400 { 1401 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1402 struct scsi_device *sdev; 1403 int rtn; 1404 1405 shost_for_each_device(sdev, shost) { 1406 if (scsi_host_eh_past_deadline(shost)) { 1407 SCSI_LOG_ERROR_RECOVERY(3, 1408 sdev_printk(KERN_INFO, sdev, 1409 "%s: skip BDR, past eh deadline\n", 1410 current->comm)); 1411 break; 1412 } 1413 bdr_scmd = NULL; 1414 list_for_each_entry(scmd, work_q, eh_entry) 1415 if (scmd->device == sdev) { 1416 bdr_scmd = scmd; 1417 break; 1418 } 1419 1420 if (!bdr_scmd) 1421 continue; 1422 1423 SCSI_LOG_ERROR_RECOVERY(3, 1424 sdev_printk(KERN_INFO, sdev, 1425 "%s: Sending BDR\n", current->comm)); 1426 rtn = scsi_try_bus_device_reset(bdr_scmd); 1427 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1428 if (!scsi_device_online(sdev) || 1429 rtn == FAST_IO_FAIL || 1430 !scsi_eh_tur(bdr_scmd)) { 1431 list_for_each_entry_safe(scmd, next, 1432 work_q, eh_entry) { 1433 if (scmd->device == sdev && 1434 scsi_eh_action(scmd, rtn) != FAILED) 1435 scsi_eh_finish_cmd(scmd, 1436 done_q); 1437 } 1438 } 1439 } else { 1440 SCSI_LOG_ERROR_RECOVERY(3, 1441 sdev_printk(KERN_INFO, sdev, 1442 "%s: BDR failed\n", current->comm)); 1443 } 1444 } 1445 1446 return list_empty(work_q); 1447 } 1448 1449 /** 1450 * scsi_eh_target_reset - send target reset if needed 1451 * @shost: scsi host being recovered. 1452 * @work_q: &list_head for pending commands. 1453 * @done_q: &list_head for processed commands. 1454 * 1455 * Notes: 1456 * Try a target reset. 1457 */ 1458 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1459 struct list_head *work_q, 1460 struct list_head *done_q) 1461 { 1462 LIST_HEAD(tmp_list); 1463 LIST_HEAD(check_list); 1464 1465 list_splice_init(work_q, &tmp_list); 1466 1467 while (!list_empty(&tmp_list)) { 1468 struct scsi_cmnd *next, *scmd; 1469 int rtn; 1470 unsigned int id; 1471 1472 if (scsi_host_eh_past_deadline(shost)) { 1473 /* push back on work queue for further processing */ 1474 list_splice_init(&check_list, work_q); 1475 list_splice_init(&tmp_list, work_q); 1476 SCSI_LOG_ERROR_RECOVERY(3, 1477 shost_printk(KERN_INFO, shost, 1478 "%s: Skip target reset, past eh deadline\n", 1479 current->comm)); 1480 return list_empty(work_q); 1481 } 1482 1483 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1484 id = scmd_id(scmd); 1485 1486 SCSI_LOG_ERROR_RECOVERY(3, 1487 shost_printk(KERN_INFO, shost, 1488 "%s: Sending target reset to target %d\n", 1489 current->comm, id)); 1490 rtn = scsi_try_target_reset(scmd); 1491 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1492 SCSI_LOG_ERROR_RECOVERY(3, 1493 shost_printk(KERN_INFO, shost, 1494 "%s: Target reset failed" 1495 " target: %d\n", 1496 current->comm, id)); 1497 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1498 if (scmd_id(scmd) != id) 1499 continue; 1500 1501 if (rtn == SUCCESS) 1502 list_move_tail(&scmd->eh_entry, &check_list); 1503 else if (rtn == FAST_IO_FAIL) 1504 scsi_eh_finish_cmd(scmd, done_q); 1505 else 1506 /* push back on work queue for further processing */ 1507 list_move(&scmd->eh_entry, work_q); 1508 } 1509 } 1510 1511 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1512 } 1513 1514 /** 1515 * scsi_eh_bus_reset - send a bus reset 1516 * @shost: &scsi host being recovered. 1517 * @work_q: &list_head for pending commands. 1518 * @done_q: &list_head for processed commands. 1519 */ 1520 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1521 struct list_head *work_q, 1522 struct list_head *done_q) 1523 { 1524 struct scsi_cmnd *scmd, *chan_scmd, *next; 1525 LIST_HEAD(check_list); 1526 unsigned int channel; 1527 int rtn; 1528 1529 /* 1530 * we really want to loop over the various channels, and do this on 1531 * a channel by channel basis. we should also check to see if any 1532 * of the failed commands are on soft_reset devices, and if so, skip 1533 * the reset. 1534 */ 1535 1536 for (channel = 0; channel <= shost->max_channel; channel++) { 1537 if (scsi_host_eh_past_deadline(shost)) { 1538 list_splice_init(&check_list, work_q); 1539 SCSI_LOG_ERROR_RECOVERY(3, 1540 shost_printk(KERN_INFO, shost, 1541 "%s: skip BRST, past eh deadline\n", 1542 current->comm)); 1543 return list_empty(work_q); 1544 } 1545 1546 chan_scmd = NULL; 1547 list_for_each_entry(scmd, work_q, eh_entry) { 1548 if (channel == scmd_channel(scmd)) { 1549 chan_scmd = scmd; 1550 break; 1551 /* 1552 * FIXME add back in some support for 1553 * soft_reset devices. 1554 */ 1555 } 1556 } 1557 1558 if (!chan_scmd) 1559 continue; 1560 SCSI_LOG_ERROR_RECOVERY(3, 1561 shost_printk(KERN_INFO, shost, 1562 "%s: Sending BRST chan: %d\n", 1563 current->comm, channel)); 1564 rtn = scsi_try_bus_reset(chan_scmd); 1565 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1566 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1567 if (channel == scmd_channel(scmd)) { 1568 if (rtn == FAST_IO_FAIL) 1569 scsi_eh_finish_cmd(scmd, 1570 done_q); 1571 else 1572 list_move_tail(&scmd->eh_entry, 1573 &check_list); 1574 } 1575 } 1576 } else { 1577 SCSI_LOG_ERROR_RECOVERY(3, 1578 shost_printk(KERN_INFO, shost, 1579 "%s: BRST failed chan: %d\n", 1580 current->comm, channel)); 1581 } 1582 } 1583 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1584 } 1585 1586 /** 1587 * scsi_eh_host_reset - send a host reset 1588 * @shost: host to be reset. 1589 * @work_q: &list_head for pending commands. 1590 * @done_q: &list_head for processed commands. 1591 */ 1592 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1593 struct list_head *work_q, 1594 struct list_head *done_q) 1595 { 1596 struct scsi_cmnd *scmd, *next; 1597 LIST_HEAD(check_list); 1598 int rtn; 1599 1600 if (!list_empty(work_q)) { 1601 scmd = list_entry(work_q->next, 1602 struct scsi_cmnd, eh_entry); 1603 1604 SCSI_LOG_ERROR_RECOVERY(3, 1605 shost_printk(KERN_INFO, shost, 1606 "%s: Sending HRST\n", 1607 current->comm)); 1608 1609 rtn = scsi_try_host_reset(scmd); 1610 if (rtn == SUCCESS) { 1611 list_splice_init(work_q, &check_list); 1612 } else if (rtn == FAST_IO_FAIL) { 1613 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1614 scsi_eh_finish_cmd(scmd, done_q); 1615 } 1616 } else { 1617 SCSI_LOG_ERROR_RECOVERY(3, 1618 shost_printk(KERN_INFO, shost, 1619 "%s: HRST failed\n", 1620 current->comm)); 1621 } 1622 } 1623 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1624 } 1625 1626 /** 1627 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1628 * @work_q: &list_head for pending commands. 1629 * @done_q: &list_head for processed commands. 1630 */ 1631 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1632 struct list_head *done_q) 1633 { 1634 struct scsi_cmnd *scmd, *next; 1635 struct scsi_device *sdev; 1636 1637 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1638 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1639 "not ready after error recovery\n"); 1640 sdev = scmd->device; 1641 1642 mutex_lock(&sdev->state_mutex); 1643 scsi_device_set_state(sdev, SDEV_OFFLINE); 1644 mutex_unlock(&sdev->state_mutex); 1645 1646 scsi_eh_finish_cmd(scmd, done_q); 1647 } 1648 return; 1649 } 1650 1651 /** 1652 * scsi_noretry_cmd - determine if command should be failed fast 1653 * @scmd: SCSI cmd to examine. 1654 */ 1655 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1656 { 1657 switch (host_byte(scmd->result)) { 1658 case DID_OK: 1659 break; 1660 case DID_TIME_OUT: 1661 goto check_type; 1662 case DID_BUS_BUSY: 1663 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1664 case DID_PARITY: 1665 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1666 case DID_ERROR: 1667 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1668 status_byte(scmd->result) == RESERVATION_CONFLICT) 1669 return 0; 1670 /* fall through */ 1671 case DID_SOFT_ERROR: 1672 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1673 } 1674 1675 if (status_byte(scmd->result) != CHECK_CONDITION) 1676 return 0; 1677 1678 check_type: 1679 /* 1680 * assume caller has checked sense and determined 1681 * the check condition was retryable. 1682 */ 1683 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1684 blk_rq_is_passthrough(scmd->request)) 1685 return 1; 1686 else 1687 return 0; 1688 } 1689 1690 /** 1691 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1692 * @scmd: SCSI cmd to examine. 1693 * 1694 * Notes: 1695 * This is *only* called when we are examining the status after sending 1696 * out the actual data command. any commands that are queued for error 1697 * recovery (e.g. test_unit_ready) do *not* come through here. 1698 * 1699 * When this routine returns failed, it means the error handler thread 1700 * is woken. In cases where the error code indicates an error that 1701 * doesn't require the error handler read (i.e. we don't need to 1702 * abort/reset), this function should return SUCCESS. 1703 */ 1704 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1705 { 1706 int rtn; 1707 1708 /* 1709 * if the device is offline, then we clearly just pass the result back 1710 * up to the top level. 1711 */ 1712 if (!scsi_device_online(scmd->device)) { 1713 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1714 "%s: device offline - report as SUCCESS\n", __func__)); 1715 return SUCCESS; 1716 } 1717 1718 /* 1719 * first check the host byte, to see if there is anything in there 1720 * that would indicate what we need to do. 1721 */ 1722 switch (host_byte(scmd->result)) { 1723 case DID_PASSTHROUGH: 1724 /* 1725 * no matter what, pass this through to the upper layer. 1726 * nuke this special code so that it looks like we are saying 1727 * did_ok. 1728 */ 1729 scmd->result &= 0xff00ffff; 1730 return SUCCESS; 1731 case DID_OK: 1732 /* 1733 * looks good. drop through, and check the next byte. 1734 */ 1735 break; 1736 case DID_ABORT: 1737 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1738 set_host_byte(scmd, DID_TIME_OUT); 1739 return SUCCESS; 1740 } 1741 /* FALLTHROUGH */ 1742 case DID_NO_CONNECT: 1743 case DID_BAD_TARGET: 1744 /* 1745 * note - this means that we just report the status back 1746 * to the top level driver, not that we actually think 1747 * that it indicates SUCCESS. 1748 */ 1749 return SUCCESS; 1750 /* 1751 * when the low level driver returns did_soft_error, 1752 * it is responsible for keeping an internal retry counter 1753 * in order to avoid endless loops (db) 1754 * 1755 * actually this is a bug in this function here. we should 1756 * be mindful of the maximum number of retries specified 1757 * and not get stuck in a loop. 1758 */ 1759 case DID_SOFT_ERROR: 1760 goto maybe_retry; 1761 case DID_IMM_RETRY: 1762 return NEEDS_RETRY; 1763 1764 case DID_REQUEUE: 1765 return ADD_TO_MLQUEUE; 1766 case DID_TRANSPORT_DISRUPTED: 1767 /* 1768 * LLD/transport was disrupted during processing of the IO. 1769 * The transport class is now blocked/blocking, 1770 * and the transport will decide what to do with the IO 1771 * based on its timers and recovery capablilities if 1772 * there are enough retries. 1773 */ 1774 goto maybe_retry; 1775 case DID_TRANSPORT_FAILFAST: 1776 /* 1777 * The transport decided to failfast the IO (most likely 1778 * the fast io fail tmo fired), so send IO directly upwards. 1779 */ 1780 return SUCCESS; 1781 case DID_ERROR: 1782 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1783 status_byte(scmd->result) == RESERVATION_CONFLICT) 1784 /* 1785 * execute reservation conflict processing code 1786 * lower down 1787 */ 1788 break; 1789 /* fallthrough */ 1790 case DID_BUS_BUSY: 1791 case DID_PARITY: 1792 goto maybe_retry; 1793 case DID_TIME_OUT: 1794 /* 1795 * when we scan the bus, we get timeout messages for 1796 * these commands if there is no device available. 1797 * other hosts report did_no_connect for the same thing. 1798 */ 1799 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1800 scmd->cmnd[0] == INQUIRY)) { 1801 return SUCCESS; 1802 } else { 1803 return FAILED; 1804 } 1805 case DID_RESET: 1806 return SUCCESS; 1807 default: 1808 return FAILED; 1809 } 1810 1811 /* 1812 * next, check the message byte. 1813 */ 1814 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1815 return FAILED; 1816 1817 /* 1818 * check the status byte to see if this indicates anything special. 1819 */ 1820 switch (status_byte(scmd->result)) { 1821 case QUEUE_FULL: 1822 scsi_handle_queue_full(scmd->device); 1823 /* 1824 * the case of trying to send too many commands to a 1825 * tagged queueing device. 1826 */ 1827 /* FALLTHROUGH */ 1828 case BUSY: 1829 /* 1830 * device can't talk to us at the moment. Should only 1831 * occur (SAM-3) when the task queue is empty, so will cause 1832 * the empty queue handling to trigger a stall in the 1833 * device. 1834 */ 1835 return ADD_TO_MLQUEUE; 1836 case GOOD: 1837 if (scmd->cmnd[0] == REPORT_LUNS) 1838 scmd->device->sdev_target->expecting_lun_change = 0; 1839 scsi_handle_queue_ramp_up(scmd->device); 1840 /* FALLTHROUGH */ 1841 case COMMAND_TERMINATED: 1842 return SUCCESS; 1843 case TASK_ABORTED: 1844 goto maybe_retry; 1845 case CHECK_CONDITION: 1846 rtn = scsi_check_sense(scmd); 1847 if (rtn == NEEDS_RETRY) 1848 goto maybe_retry; 1849 /* if rtn == FAILED, we have no sense information; 1850 * returning FAILED will wake the error handler thread 1851 * to collect the sense and redo the decide 1852 * disposition */ 1853 return rtn; 1854 case CONDITION_GOOD: 1855 case INTERMEDIATE_GOOD: 1856 case INTERMEDIATE_C_GOOD: 1857 case ACA_ACTIVE: 1858 /* 1859 * who knows? FIXME(eric) 1860 */ 1861 return SUCCESS; 1862 1863 case RESERVATION_CONFLICT: 1864 sdev_printk(KERN_INFO, scmd->device, 1865 "reservation conflict\n"); 1866 set_host_byte(scmd, DID_NEXUS_FAILURE); 1867 return SUCCESS; /* causes immediate i/o error */ 1868 default: 1869 return FAILED; 1870 } 1871 return FAILED; 1872 1873 maybe_retry: 1874 1875 /* we requeue for retry because the error was retryable, and 1876 * the request was not marked fast fail. Note that above, 1877 * even if the request is marked fast fail, we still requeue 1878 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1879 if ((++scmd->retries) <= scmd->allowed 1880 && !scsi_noretry_cmd(scmd)) { 1881 return NEEDS_RETRY; 1882 } else { 1883 /* 1884 * no more retries - report this one back to upper level. 1885 */ 1886 return SUCCESS; 1887 } 1888 } 1889 1890 static void eh_lock_door_done(struct request *req, blk_status_t status) 1891 { 1892 __blk_put_request(req->q, req); 1893 } 1894 1895 /** 1896 * scsi_eh_lock_door - Prevent medium removal for the specified device 1897 * @sdev: SCSI device to prevent medium removal 1898 * 1899 * Locking: 1900 * We must be called from process context. 1901 * 1902 * Notes: 1903 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1904 * head of the devices request queue, and continue. 1905 */ 1906 static void scsi_eh_lock_door(struct scsi_device *sdev) 1907 { 1908 struct request *req; 1909 struct scsi_request *rq; 1910 1911 /* 1912 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a 1913 * request becomes available 1914 */ 1915 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL); 1916 if (IS_ERR(req)) 1917 return; 1918 rq = scsi_req(req); 1919 1920 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 1921 rq->cmd[1] = 0; 1922 rq->cmd[2] = 0; 1923 rq->cmd[3] = 0; 1924 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 1925 rq->cmd[5] = 0; 1926 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 1927 1928 req->rq_flags |= RQF_QUIET; 1929 req->timeout = 10 * HZ; 1930 rq->retries = 5; 1931 1932 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); 1933 } 1934 1935 /** 1936 * scsi_restart_operations - restart io operations to the specified host. 1937 * @shost: Host we are restarting. 1938 * 1939 * Notes: 1940 * When we entered the error handler, we blocked all further i/o to 1941 * this device. we need to 'reverse' this process. 1942 */ 1943 static void scsi_restart_operations(struct Scsi_Host *shost) 1944 { 1945 struct scsi_device *sdev; 1946 unsigned long flags; 1947 1948 /* 1949 * If the door was locked, we need to insert a door lock request 1950 * onto the head of the SCSI request queue for the device. There 1951 * is no point trying to lock the door of an off-line device. 1952 */ 1953 shost_for_each_device(sdev, shost) { 1954 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 1955 scsi_eh_lock_door(sdev); 1956 sdev->was_reset = 0; 1957 } 1958 } 1959 1960 /* 1961 * next free up anything directly waiting upon the host. this 1962 * will be requests for character device operations, and also for 1963 * ioctls to queued block devices. 1964 */ 1965 SCSI_LOG_ERROR_RECOVERY(3, 1966 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 1967 1968 spin_lock_irqsave(shost->host_lock, flags); 1969 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1970 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1971 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1972 spin_unlock_irqrestore(shost->host_lock, flags); 1973 1974 wake_up(&shost->host_wait); 1975 1976 /* 1977 * finally we need to re-initiate requests that may be pending. we will 1978 * have had everything blocked while error handling is taking place, and 1979 * now that error recovery is done, we will need to ensure that these 1980 * requests are started. 1981 */ 1982 scsi_run_host_queues(shost); 1983 1984 /* 1985 * if eh is active and host_eh_scheduled is pending we need to re-run 1986 * recovery. we do this check after scsi_run_host_queues() to allow 1987 * everything pent up since the last eh run a chance to make forward 1988 * progress before we sync again. Either we'll immediately re-run 1989 * recovery or scsi_device_unbusy() will wake us again when these 1990 * pending commands complete. 1991 */ 1992 spin_lock_irqsave(shost->host_lock, flags); 1993 if (shost->host_eh_scheduled) 1994 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 1995 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 1996 spin_unlock_irqrestore(shost->host_lock, flags); 1997 } 1998 1999 /** 2000 * scsi_eh_ready_devs - check device ready state and recover if not. 2001 * @shost: host to be recovered. 2002 * @work_q: &list_head for pending commands. 2003 * @done_q: &list_head for processed commands. 2004 */ 2005 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2006 struct list_head *work_q, 2007 struct list_head *done_q) 2008 { 2009 if (!scsi_eh_stu(shost, work_q, done_q)) 2010 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2011 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2012 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2013 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2014 scsi_eh_offline_sdevs(work_q, 2015 done_q); 2016 } 2017 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2018 2019 /** 2020 * scsi_eh_flush_done_q - finish processed commands or retry them. 2021 * @done_q: list_head of processed commands. 2022 */ 2023 void scsi_eh_flush_done_q(struct list_head *done_q) 2024 { 2025 struct scsi_cmnd *scmd, *next; 2026 2027 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2028 list_del_init(&scmd->eh_entry); 2029 if (scsi_device_online(scmd->device) && 2030 !scsi_noretry_cmd(scmd) && 2031 (++scmd->retries <= scmd->allowed)) { 2032 SCSI_LOG_ERROR_RECOVERY(3, 2033 scmd_printk(KERN_INFO, scmd, 2034 "%s: flush retry cmd\n", 2035 current->comm)); 2036 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2037 } else { 2038 /* 2039 * If just we got sense for the device (called 2040 * scsi_eh_get_sense), scmd->result is already 2041 * set, do not set DRIVER_TIMEOUT. 2042 */ 2043 if (!scmd->result) 2044 scmd->result |= (DRIVER_TIMEOUT << 24); 2045 SCSI_LOG_ERROR_RECOVERY(3, 2046 scmd_printk(KERN_INFO, scmd, 2047 "%s: flush finish cmd\n", 2048 current->comm)); 2049 scsi_finish_command(scmd); 2050 } 2051 } 2052 } 2053 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2054 2055 /** 2056 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2057 * @shost: Host to unjam. 2058 * 2059 * Notes: 2060 * When we come in here, we *know* that all commands on the bus have 2061 * either completed, failed or timed out. we also know that no further 2062 * commands are being sent to the host, so things are relatively quiet 2063 * and we have freedom to fiddle with things as we wish. 2064 * 2065 * This is only the *default* implementation. it is possible for 2066 * individual drivers to supply their own version of this function, and 2067 * if the maintainer wishes to do this, it is strongly suggested that 2068 * this function be taken as a template and modified. this function 2069 * was designed to correctly handle problems for about 95% of the 2070 * different cases out there, and it should always provide at least a 2071 * reasonable amount of error recovery. 2072 * 2073 * Any command marked 'failed' or 'timeout' must eventually have 2074 * scsi_finish_cmd() called for it. we do all of the retry stuff 2075 * here, so when we restart the host after we return it should have an 2076 * empty queue. 2077 */ 2078 static void scsi_unjam_host(struct Scsi_Host *shost) 2079 { 2080 unsigned long flags; 2081 LIST_HEAD(eh_work_q); 2082 LIST_HEAD(eh_done_q); 2083 2084 spin_lock_irqsave(shost->host_lock, flags); 2085 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2086 spin_unlock_irqrestore(shost->host_lock, flags); 2087 2088 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2089 2090 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2091 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2092 2093 spin_lock_irqsave(shost->host_lock, flags); 2094 if (shost->eh_deadline != -1) 2095 shost->last_reset = 0; 2096 spin_unlock_irqrestore(shost->host_lock, flags); 2097 scsi_eh_flush_done_q(&eh_done_q); 2098 } 2099 2100 /** 2101 * scsi_error_handler - SCSI error handler thread 2102 * @data: Host for which we are running. 2103 * 2104 * Notes: 2105 * This is the main error handling loop. This is run as a kernel thread 2106 * for every SCSI host and handles all error handling activity. 2107 */ 2108 int scsi_error_handler(void *data) 2109 { 2110 struct Scsi_Host *shost = data; 2111 2112 /* 2113 * We use TASK_INTERRUPTIBLE so that the thread is not 2114 * counted against the load average as a running process. 2115 * We never actually get interrupted because kthread_run 2116 * disables signal delivery for the created thread. 2117 */ 2118 while (true) { 2119 /* 2120 * The sequence in kthread_stop() sets the stop flag first 2121 * then wakes the process. To avoid missed wakeups, the task 2122 * should always be in a non running state before the stop 2123 * flag is checked 2124 */ 2125 set_current_state(TASK_INTERRUPTIBLE); 2126 if (kthread_should_stop()) 2127 break; 2128 2129 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2130 shost->host_failed != atomic_read(&shost->host_busy)) { 2131 SCSI_LOG_ERROR_RECOVERY(1, 2132 shost_printk(KERN_INFO, shost, 2133 "scsi_eh_%d: sleeping\n", 2134 shost->host_no)); 2135 schedule(); 2136 continue; 2137 } 2138 2139 __set_current_state(TASK_RUNNING); 2140 SCSI_LOG_ERROR_RECOVERY(1, 2141 shost_printk(KERN_INFO, shost, 2142 "scsi_eh_%d: waking up %d/%d/%d\n", 2143 shost->host_no, shost->host_eh_scheduled, 2144 shost->host_failed, 2145 atomic_read(&shost->host_busy))); 2146 2147 /* 2148 * We have a host that is failing for some reason. Figure out 2149 * what we need to do to get it up and online again (if we can). 2150 * If we fail, we end up taking the thing offline. 2151 */ 2152 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2153 SCSI_LOG_ERROR_RECOVERY(1, 2154 shost_printk(KERN_ERR, shost, 2155 "scsi_eh_%d: unable to autoresume\n", 2156 shost->host_no)); 2157 continue; 2158 } 2159 2160 if (shost->transportt->eh_strategy_handler) 2161 shost->transportt->eh_strategy_handler(shost); 2162 else 2163 scsi_unjam_host(shost); 2164 2165 /* All scmds have been handled */ 2166 shost->host_failed = 0; 2167 2168 /* 2169 * Note - if the above fails completely, the action is to take 2170 * individual devices offline and flush the queue of any 2171 * outstanding requests that may have been pending. When we 2172 * restart, we restart any I/O to any other devices on the bus 2173 * which are still online. 2174 */ 2175 scsi_restart_operations(shost); 2176 if (!shost->eh_noresume) 2177 scsi_autopm_put_host(shost); 2178 } 2179 __set_current_state(TASK_RUNNING); 2180 2181 SCSI_LOG_ERROR_RECOVERY(1, 2182 shost_printk(KERN_INFO, shost, 2183 "Error handler scsi_eh_%d exiting\n", 2184 shost->host_no)); 2185 shost->ehandler = NULL; 2186 return 0; 2187 } 2188 2189 /* 2190 * Function: scsi_report_bus_reset() 2191 * 2192 * Purpose: Utility function used by low-level drivers to report that 2193 * they have observed a bus reset on the bus being handled. 2194 * 2195 * Arguments: shost - Host in question 2196 * channel - channel on which reset was observed. 2197 * 2198 * Returns: Nothing 2199 * 2200 * Lock status: Host lock must be held. 2201 * 2202 * Notes: This only needs to be called if the reset is one which 2203 * originates from an unknown location. Resets originated 2204 * by the mid-level itself don't need to call this, but there 2205 * should be no harm. 2206 * 2207 * The main purpose of this is to make sure that a CHECK_CONDITION 2208 * is properly treated. 2209 */ 2210 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2211 { 2212 struct scsi_device *sdev; 2213 2214 __shost_for_each_device(sdev, shost) { 2215 if (channel == sdev_channel(sdev)) 2216 __scsi_report_device_reset(sdev, NULL); 2217 } 2218 } 2219 EXPORT_SYMBOL(scsi_report_bus_reset); 2220 2221 /* 2222 * Function: scsi_report_device_reset() 2223 * 2224 * Purpose: Utility function used by low-level drivers to report that 2225 * they have observed a device reset on the device being handled. 2226 * 2227 * Arguments: shost - Host in question 2228 * channel - channel on which reset was observed 2229 * target - target on which reset was observed 2230 * 2231 * Returns: Nothing 2232 * 2233 * Lock status: Host lock must be held 2234 * 2235 * Notes: This only needs to be called if the reset is one which 2236 * originates from an unknown location. Resets originated 2237 * by the mid-level itself don't need to call this, but there 2238 * should be no harm. 2239 * 2240 * The main purpose of this is to make sure that a CHECK_CONDITION 2241 * is properly treated. 2242 */ 2243 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2244 { 2245 struct scsi_device *sdev; 2246 2247 __shost_for_each_device(sdev, shost) { 2248 if (channel == sdev_channel(sdev) && 2249 target == sdev_id(sdev)) 2250 __scsi_report_device_reset(sdev, NULL); 2251 } 2252 } 2253 EXPORT_SYMBOL(scsi_report_device_reset); 2254 2255 static void 2256 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2257 { 2258 } 2259 2260 /** 2261 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2262 * @dev: scsi_device to operate on 2263 * @arg: reset type (see sg.h) 2264 */ 2265 int 2266 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2267 { 2268 struct scsi_cmnd *scmd; 2269 struct Scsi_Host *shost = dev->host; 2270 struct request *rq; 2271 unsigned long flags; 2272 int error = 0, rtn, val; 2273 2274 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2275 return -EACCES; 2276 2277 error = get_user(val, arg); 2278 if (error) 2279 return error; 2280 2281 if (scsi_autopm_get_host(shost) < 0) 2282 return -EIO; 2283 2284 error = -EIO; 2285 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2286 shost->hostt->cmd_size, GFP_KERNEL); 2287 if (!rq) 2288 goto out_put_autopm_host; 2289 blk_rq_init(NULL, rq); 2290 2291 scmd = (struct scsi_cmnd *)(rq + 1); 2292 scsi_init_command(dev, scmd); 2293 scmd->request = rq; 2294 scmd->cmnd = scsi_req(rq)->cmd; 2295 2296 scmd->scsi_done = scsi_reset_provider_done_command; 2297 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2298 2299 scmd->cmd_len = 0; 2300 2301 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2302 2303 spin_lock_irqsave(shost->host_lock, flags); 2304 shost->tmf_in_progress = 1; 2305 spin_unlock_irqrestore(shost->host_lock, flags); 2306 2307 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2308 case SG_SCSI_RESET_NOTHING: 2309 rtn = SUCCESS; 2310 break; 2311 case SG_SCSI_RESET_DEVICE: 2312 rtn = scsi_try_bus_device_reset(scmd); 2313 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2314 break; 2315 /* FALLTHROUGH */ 2316 case SG_SCSI_RESET_TARGET: 2317 rtn = scsi_try_target_reset(scmd); 2318 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2319 break; 2320 /* FALLTHROUGH */ 2321 case SG_SCSI_RESET_BUS: 2322 rtn = scsi_try_bus_reset(scmd); 2323 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2324 break; 2325 /* FALLTHROUGH */ 2326 case SG_SCSI_RESET_HOST: 2327 rtn = scsi_try_host_reset(scmd); 2328 if (rtn == SUCCESS) 2329 break; 2330 /* FALLTHROUGH */ 2331 default: 2332 rtn = FAILED; 2333 break; 2334 } 2335 2336 error = (rtn == SUCCESS) ? 0 : -EIO; 2337 2338 spin_lock_irqsave(shost->host_lock, flags); 2339 shost->tmf_in_progress = 0; 2340 spin_unlock_irqrestore(shost->host_lock, flags); 2341 2342 /* 2343 * be sure to wake up anyone who was sleeping or had their queue 2344 * suspended while we performed the TMF. 2345 */ 2346 SCSI_LOG_ERROR_RECOVERY(3, 2347 shost_printk(KERN_INFO, shost, 2348 "waking up host to restart after TMF\n")); 2349 2350 wake_up(&shost->host_wait); 2351 scsi_run_host_queues(shost); 2352 2353 scsi_put_command(scmd); 2354 kfree(rq); 2355 2356 out_put_autopm_host: 2357 scsi_autopm_put_host(shost); 2358 return error; 2359 } 2360 EXPORT_SYMBOL(scsi_ioctl_reset); 2361 2362 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2363 struct scsi_sense_hdr *sshdr) 2364 { 2365 return scsi_normalize_sense(cmd->sense_buffer, 2366 SCSI_SENSE_BUFFERSIZE, sshdr); 2367 } 2368 EXPORT_SYMBOL(scsi_command_normalize_sense); 2369 2370 /** 2371 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2372 * @sense_buffer: byte array of sense data 2373 * @sb_len: number of valid bytes in sense_buffer 2374 * @info_out: pointer to 64 integer where 8 or 4 byte information 2375 * field will be placed if found. 2376 * 2377 * Return value: 2378 * true if information field found, false if not found. 2379 */ 2380 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2381 u64 *info_out) 2382 { 2383 const u8 * ucp; 2384 2385 if (sb_len < 7) 2386 return false; 2387 switch (sense_buffer[0] & 0x7f) { 2388 case 0x70: 2389 case 0x71: 2390 if (sense_buffer[0] & 0x80) { 2391 *info_out = get_unaligned_be32(&sense_buffer[3]); 2392 return true; 2393 } 2394 return false; 2395 case 0x72: 2396 case 0x73: 2397 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2398 0 /* info desc */); 2399 if (ucp && (0xa == ucp[1])) { 2400 *info_out = get_unaligned_be64(&ucp[4]); 2401 return true; 2402 } 2403 return false; 2404 default: 2405 return false; 2406 } 2407 } 2408 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2409